s

*
4S
s
G
*
s
&
s
a
&
*
&
*
s
6
€
s
*
6
*
*
;
*
€
ry

*
s
€i
&
I
*
I
#
*
*
s
i*
*
s
tr
s
*
*
4
*
#
&
&
#
*
*
r&
s

&
R
*
s
*
*
#
s
#
&
s
*
s
#
&
*
*
s
&
s
*
a
*'
+
s
*
s
w
R
6
*
&
m
s
s
s
*
q
&
*
&
s
s
*
*
s
\
 M. Mitchell
Wafdrop

A TOUCHSTONE BOOK
pueusnro
sySrA4oN
* Scnusrrn
NewYonr LoNooN TonoNro SvonevToxyo
SrNcnponr
 7K
TOUCHSTONE
Simon & SchusterBuilding
RockefdllerCenter
1230Avenueof the Americas
New York, New York 10020

Copyright @ l99Z by M. Mitchell Waldrop

All rights reserved

including the right of reprod-uction
in whole or in Part in anYtorm'
First TouchstoneEdition l99l
TOUCHSTONE and colophon are registeredhademarks
o[ Simon & SchusterInc'

DesignedbY Deirdre C' Amthor

Manufactur-edIn the United Statesof America

)579108642

Data
' of CongressCataloging-in-Publication
Library
W.ldtoP, M. Mitchell'
Complexity:the emergingscienceat thg edgeof order and
chaos/ M. Mitchell waldroP'
p. cm'
Includesbibliographicalreferencesand index' -
l. Sctce-Philosiphv. 2. Complexi$ (Philosophv)
I. Title.
Q17t.w2581992
501-dc20 92-28757
CIP

ISBN:0-671-76789-5
(PBK)
ISBN:0-671-87214-6
To A. E. F.
 Contents

VsroNsoF THEWHors 9

l. The lrish Ideaof a Hero l5

2. The Revoltof the Old Turks
I A >L
3. Secrets
of theold one Kd,fitn rn 99
4. "YouGuysReallyBelieveThat?"
Arlbna 136
5. Masterof theGame ttll*l t44
6. Lifeat theEdgeof Chaos L*4"" 198
7. PeasantsUnderGlassrifh;, ltJl^rl 241
8. Waitingfor Carnot l(rr$-o, 27:.
9. Work in Progress
324

BBr,rocnapHy
360
AcruowLnncuenrs
36'
INpnx
367

i
 Visionsof the Whole

This is a bookaboutthe scienceof eomplexity-a subiectthat'sstill so new

and so wide-rangingthat nobodyknowsquite how to defineit, or even
whereits boundarieslie. But then,that'sthewholepoint. If the field seems
poorlydefinedat the moment,it's because complexityresearch is trying to
grapplewith questionsthat defy all the conventionalcategories'For ex-
ample:

. Why did the SovietUnion's forg-yearhegemonyovereasternEurope

collapsewithin a few monthsin 1989?And why did the SovietUnion
itselfcomeapartlessthat two yearslater?Why wasthe collapseof com-
munismso fastand socomplete? It surelyhadsomethingto do with two
men namedGorbachevand Yeltsin. And yet eventhey seemedto k
sweptup in eventsthat werefar beyondtheir control.Was theresome
globaldynamicat work that hanscends individualpersonalities?
. Why did thestockmarketcrashmorethan 500pointson a singleMonday
in October1987?A lot of the blamegoesto computerized trading.But
the computershad beenaroundfor years.ls thereany reasonwhy the
crashcameon that particularMonday?
. Why do ancientspecies and ecosystems oftenremainstablein the fossil
recordfor millionsof years-and then eitherdie out or transformthem-
selvesinto somethingnew in a geologicalinstant?Perhaps the dinosaurs
gotwipedoutby an asteroidimpact.But thereweren'tthat manyasteroids.
What elsewasgoingon?
. Why do rural familiesin a nationsuchasBangladesh still producean
10 COMPLEX'TY

averageof sevenchildren apiece, even when birth control is made freely

availflle-and even when the villagersseemperfectlywell awareof how
they're being hurt by the country's immense overpopulationand stagnant
development?Why do they continue in a course of behavior that's so
obviouslydisastrous?
. How did a primordial soup of amino acidsand other simple molecules
manageto turn itself into the first living cell some four billion yearsago?
There;sno way the moleculescould have iust fallen togetherat random;
asthe creationists arefond ofpointing out, the oddsagainstthat happening
are ludicrous. So was the creationof life a miracle?or wasthere some-
thing elsegoing on in that primordialsoupthat we still don't understand?
. wht did individual cells begin to form alliancessome 600 million years
ago, therebygiving rise to multicellular organismssuch as seaweed,iel-
lyfish, insects,and eventuallyhumans?For that matter,why do humans
s end so much time and effort organizingthemselvesinto families, tribes,
communities, nations, and societiesof all types?If evolution (or free-
market capitalism) is really iust a matter of the survival of the fiftest, then
why should it ever produce anything other than ruthless competition
among individuals? In a world where nice guys all too often 6nish last,
why should there be any such thing as trust or cooperation?And why,
in spiteofeverything,do trustand cooperationnot only existbut flourish?
. How can Darwinian natural selectionaccountfor such wonderfully in-
tricate structures as the eye or the kidney? Is the incredibly precise or-
ganization that we find in living creaturesreally iust the result of random
evolutionary accidents?Or has something more been going on for the
pastfour billion years,somethingthat Darwin didn't know about?
. What is life, anyrvay?Is it nothing more than a particularly complicated
kind of carbon chemistry?or is it something more subtle?And what are
we to make of creations such as computer viruses?Are they iust pesky
imitationsof life-or in somefundamentalsenseare they really alive?
. What is a mind? How doesa three-poundlump of ordinarymatter, the
brain, give rise to such ineffable qualities as feeling, thought, PurPose'
and awareness?
. And perhapsmost fundamentally,why is there somethingrather than
nothing? The universe startedout from the formless miasma of the Big
Bang. l,nd ever since then it's been governedby an inexorabletendency
towa"rddisorder, dissolution, and decay, as describedby the second law
of thermodynamics.Yet the universehas also managedto bring forth
structureon everyscale:galaxies,stars,planets,bacteria,plants,animals,
and brains. How? Is the cosmic compulsionfor disordermatchedby an
 Visionsof the Whole fi
equally powerful compulsion for order, structure, and organization?And
ifso, how can both processes be going on at once?

At first glance,about the only thing that thesequestionshavein common

is that they all have the same answer: "Nobody knows." Some of them
don't even seemlike scientific issuesat all. And yet, when you look a little
closer, they actually have quite a lot in common. For example, every one
of these questions refers to a systemthat is complex, in the sensethat a
great many independent agentsare interacting with each other in a great
manyways.Think of the quadrillionsof chemicallyreactingproteins,lipids,
and nucleic acidsthat makeup a living cell, or the billionsof interconnected
neuronsthat make up the brain, or the millions of mutually interdependent
individualswho make up a human society.
In every case,moreover,the very richnessof theseinteractionsallows
the system as a whole to undergo sfontaneousself-organization.Thus,
people trying to satisfytheir material needsunconsciouslyorganize them-
selvesinto an economythrough myriad individual actsof buying and selling;
it happenswithout anyone being in chargeor consciouslyplanning it. The
genesin a developing embryo organize themselvesin one way to make a
liver cell and in anotherway to make a musclecell. Flying birds adaptto
the actionsof their neighbors,unconsciouslyorganizingthemselvesinto a
flock. organisms constantlyadaptto each other through evolution, thereby
organizing themselvesinto an exquisitely tuned ecosystem.Atoms search
for a minimum energy stateby forming chemical bonds with each other,
thereby organizingthemselvesinto structuresknown asmolecules.In every
case'groupsof agentsseekingmutual accommodationand self-consistency
somehow manage to transcendthemselves,acquiring collective properties
such as life, thought, and purpose that they might never have possessed
individually.
Furthermore, these complex, self-organizingsystems are adaptite, in
that they don't just passivelyrespondto eventsthe way a rock might rolr
around in an earthquake. They actively try to turn whateverhappens to
their advantage.Thus, the human brain constantlyorganizes,ndi.org"-
nizes its billions of neural connectionsso as to learn from experience
(sometimes,anyway).Speciesevolvefor bettersurvival in a chaniing en-
vironment-and so do corporationsand indushies. And the marketplace
respondsto changingtastesand lifestyles,immigration, technologicalde-
velopments,shiftsin the price of raw materials,and a hostof othei factors.
Finally, every one of thesecomplex, self-organizing,adaptivesystems
possesses a kind of dynamism that makesthem qualitatively different from
12 COMPLEXITY
whicharemerely
staticobiectssuchas computerchipsor snowflakes,
complicated. Complex systemsare more spontaneous,more disorderly,
more alive than that. At the sametime, however,their peculiar dynamism
is also a far cry from the weirdly unpredictablegyrationsknown as chaos.
In the past two decades,chaostheory has shakenscienceto its foundations
with the realizationthat very simple dynamical rules can give rise to ex-
traordinarily intricate behavior; witness the endlesslydetailed beauty of
ftactals,or the foaming turbulence of a river. And yet chaosby itself doesn't
explain the structure,the coherence,the self-organizingcohesiveness of
complex systems.
Instead, all thesecomplex systemshave somehowacquired the ability to
bring order and chaosinto a specialkind ofbalance.This balancepoint-
often called the edgeof chaos-is were the components of a system never
quite lock into place, and yet neverquite dissolveinto turbulence,either.
ihe edge of chaos is where life has enough stability to sustain itself and
enough creativityto deservethe name of life. The edgeof chaosis where
n.* id.r, and innovative genogpesare forever nibbling awayat the edges
of the status quo, and where even the most entrenched old guard will
eventually be overthrown. The edgeof chaosis where centuriesof slavery
and segregationsuddenlygive way to the civil rights movementof the 1950s
and 1960s;where seventyyearsof Sovietcommunism suddenlygive way
to political turmoil and ferment;where eonsof evolutionarystabilitysud-
denly give way to wholesalespeciestransformation.The edge of chaos is
the constantlyshifting battle zone betweenstagnationand anarchy,the one
place where a complex systemcan be spontaneous,adaptive,and alive.

Complexig, adaptation,upheavalsat the edgeof chaos-these common

themesare so striking that a growing number of scientistsare convinced
that there is more here than iust a seriesof nice analogies.The movement's
nerve center is a think tank known as the Santa Fe Institute, which was
founded in the mid-1980sand which was originally housedin a rented
convent in the midst of SantaFb'sart colony along canyon Road.(seminars
wereheld in what usedto be the chapel.)The researchers who gatherthere
are an eclecticbunch, ranging from pony-tailedgraduatestudentsto Nobel
laureatessuch as Murray Gell-Mann and Philip Andersonin physicsand
KennethArrow in economics.But they all sharethe visionof an underlying
unity, a common theoretical framework for complexity that would illu-
minate natureand humankind alike. They believethat they have in hand
the mathematical tools to createsuch a framework, drawing from the past
 Visionsof the Whole 13
twenty yearsof intellectual ferment in such fields as neural networks, ecol-
ogy, artificial intelligence, and chaos theory. They believe that their ap-
plication of these ideas is allowing them to understandthe spontaneous,
self-organizingdynamicsof the world in a way that no one everhasbefore-
with the potential for immense impact on the conduct of economics,
business,and even politics, They believe that they are forging the first
rigorous alternative to the kind of linear, reductionist thinking that has
dominated sciencesince the time of Newton-and that hasnow gone about
as far as it can go in addressingthe problems of our modern world. They
believethey are creating, in the wordsof SantaFe Institute founder George
Cowan, "the sciencesof the twenty-firstcentury."
This is their storv.

I
*a ar *sa* s * s s 6 * s & ' { s * * & & * & I 6 * & & S { b4 *

Sitting alone at his table by the bar, Brian Arthur staredout the front
windJwof thetavernanddid hisbestto ignoretheyoungurbanprofessionals
driftingin to getan earlystarton HappyHour. outside,in the concrete
of the financialdishict,the $pical SanFrancisco fogwastuming
""rryor,,
into a typical San Francisco drizzle.That was 6ne by him. On this late
aft rnoonof March 17, 1987, he wasn'tin the mood to be impressed with
brassfittings,ferns,and stained glass.He wasn'tin a mood to celebrate
Saint Patrick'sDay. And he mostdefinitelywasn'tin a mood to carouse
withersatz Irishmenwearing bitsof greenon theirpinstripes. He iustwanted
to silently sip his beer in frustratedrage. Stanford University Professor
William brian Arthur, nativesonof Belfas!Northern lreland, was at rock
bottom.
And the dayhad startedsowell.
That wasthe ironyof it all. When he'dsetout for Berkeley thatmorning,
he'dactuallybeenlookingforwardto thetrip asa kindof triumphal reunion:
local boy makesgood.He'd reallylovedhis yearsin Berkeley, back in the
early1970s.Perchedon the hillsidesnorth of Oakland,iust across the bay
from SanFrancisco, it wasa pushy,vital,alivekindof place full of ethnics
and streetpeopleand outrageous ideas.Berkeleywaswherehe'dgottel his
Ph.D. from the Universityof California,wherehe'd met and marrieda
tall blondedoctoralstudentin statisticsnamedSusanPeterson, wherehe'd
spenthis first"postdoc"yealin the economics department. Berkeley,of all
the placeshe'd lived and workedeversince,wasthe placehe wantedto
comehometo.
16 COMPLEXITY
Well now he wascominghome,sortof. The eventibelf wouldn'tbe a
big deal:iust lunch with the chairmanof the Berkeleyeconomicsdepart-
ment and one of his formerprofesorsthere.But it wasthe first time he'd
comebackto his old departmentin years,andcertainlythe firsttime he'd
everdone so feelinglike an academicequal. He wascomingbackwith
twelveyeanof experience workingall overtheglobeanda maiorreputation
asa scholarof humanfertilityin the Third World. He wascomingbackas
the occupantof an endowedchair of economicsat Stanfold-the sort of
thingthat rarelygetshandedout to anyoneunderagefifty' At ageforty-one,
Arthur wascomingbackassomeonewho had madeit in academia.And
who knew?The folksat Berkeleymightevenstarttalkingabouta iob offer.
Oh yes,he'd reallybeenhigh on himselfthat morning.So why hadn't
he, yearsago, iust stuckto the mainsheaminsteadof trying to invent a
whole new approachto economics? Why hadn't he playedit safeinstead
of trying to get in stepwith somenebulous,half-imaginaryscientificrev-
olution?
Because he couldn'tgetit out of his head,that'swhy.Because he could
seeit almosteverywherehe looked.The scientists barely seemed to rec-
ognizeit themselves, mostof the time. But after three hundred years of
dissecting everythinginto moleculesandatomsand nuclei andquarks,they
finally seemedto be turningthat process insideout. Insteadof lookingfor
the simplestpiecespossible,theywerestartingto look at how thosepieces
go togetherinto complexwholes.
He could seeit happeningin biology,wherepeoplehad spentthe past
twentyyearslayingbarethe molecularmechanisms of DNA, andproteins,
and all the other components of the cell. Now they werealsobeginning
to grapplewith the essentialmystery:how can severalquadrillionsuch
moleculesorganizethemselves into an entity that moves'that responds,
that reproduces, that is alive?
He could seeit happeningin the brain sciences, whereneuroscientisb,
psychologists, computerscientists, andartificialintelligence were
researchers
strugglingto comprehendthe essence of mind: How do thosebillions of
denselyinterconnected nervecells insideour skullsgive rise to feeling,
thought,purpose,and awareness?
He could evenseeit happeningin physics,wherethe physicistswere
still trying to cometo termswith the mathematical theoryof chaos,the
inhicate beaug of fractals,and the weird inner workingsof solidsand
liquids.Therewasprofoundmysteryhere:Why is it that simpleparticles
obeyingsimpleruleswill sometimes engagein the mostastonishing,un-
The lrish ldea of a Hero 17
predictablebehavior? And why is it thatsimpleparticleswill spontaneously
organizethemselves into complexshuctures like stars,galaxies,snowflakes,
and hurricanes-almostas if they were ohying a hidden yeamingfor
organizationand order?
The signsweroeverywhere. Arthur couldn't quite put the feelinginto
words.Nobodycould,sofarashe couldtell. But somehow, he couldsense
that all thesequestions werereallythe scmequestion.Somehow, the old
categories of science werebeginningto dissolve. Somehow, a new,unified
sciencewasout therewaitingto be born. It would be a rigorousscience,
Arthur was convinced,iust as "hard" as physicseverwas,and iust as
thoroughlygroundedin naturallaw. But insteadof beinga questfor the
ultimateparticles,it would be aboutfux, change,and the forming and
dissolvingof pattems.lnsteadof ignoringeverythingthat wasn'tuniform
and predictable,it would havea placefor individualityand the accidents
of history. Insteadof being about simplicity,it would be about-well,
complexity.
And that wasprecisely whereArthur'snew economics camein. Con-
ventionaleconomics, the kind he'd beentaughtin school,wasaboutas
far from thisvisionof complexigasyoucouldimagine.Theoretical econ-
omistsendlesslytalkedaboutthe stabilityof the marketplace, and the bal-
ance of supply and demand. They transcribedthe concept into
mathematicalequationsand provedtheoremsaboutit. They accepted the
gospelaccordingto Adam Smith as the foundationfor a kind of state
religion. But when it cameto instabilityand changein the economy-
well, they seemedto find the very ideadisturbing,somethingthey'djust
assoonnot talkabout.
But Arthur had embraced instability.Look out the window,he'd told
his colleagues. Like it or not, the marketplace isn'tstable.The worldisn't
stable.It'sfull of evolution,upheaval, andsurprise. Economics hadto take
that fermentinto account.And now he believedhe'dfoundthe wayto do
that, usinga principleknownas"increasingreturns"-or in the King)ames
translation,"To them that hath shallbe given."Why had high-techcom-
paniesscrambled to locatein theSiliconValleyareaaroundStanfordinstead
of in Ann Arbor or Berkeley? Because a lot of olderhigh-techcompanies
werealreadythere.Them that hasgets.Why did the VHS video system
run awaywith the market,eventhoughBetawastechnicallya little bit
better?Because a fewmorepeoplehappened to buy VHS systems earlyon,
which led to moreVHS moviesin the videostores,which led to still more
peoplebuyingVHS players, andsoon. Them thathasgets.
1B COMPLEXITY
The examplescould be multiplied endlessly.Arthur had convinced him-
self that increasingreturns pointed the way to the future for economics, a
future in which he and his colleagueswould work alongsidethe physicists
and the biologists to understand the messiness,the upheaval, and the
spontaneousself-organizationof the world. He'd convinced himself that
increasing returns could be the foundation for a new and very different
kind of economic science.
Unfortunately, however, he hadn't had much luck convincing any-
body else. Outside of his immediate circle at Stanford, most economists
thought his ideaswere-strange. |ournal editors were telling him that this
increasing-returnsstuff "wasn't economics." In seminars, a good fraction
of the audience reactedwith outrage:how dare he suggestthat the economy
was not in equilibrium! Arthur found the vehemencebaffing. But clearly
he neededallies, people who could open their minds and hear what he
wastrying to tell them. And that, asmuch asany desirefor a homecoming,
was the reasonhe'd gone to Berkeley.
So there they had all been, sitting down to sandwichesat the faculty
club. Tom Rothenberg, one of his former professors,had askedthe inev-
itable question: "So, Brian, what are you working on thesedays?"Arthur
had given him the two-word answerlust to get started:"Increasingreturns.'l
And the economicsdepartmentchairman, Al Fishlow, had staredat him
with a kind of deadpanlook,
"$ul-\rys know increasingreturnsdon't exist."
"Besides,"jumped in Rothenbergwith a grin, "if they did, we'd haveto
outlaw them!"
And then they'd laughed. Not unkindly. It was iust an insider'sioke.
Arthur knew lt was a ioke. It was trivial. Yet that one sound had somehow
shatteredhis whole bubble of anticipation. He'd satthere, shuck speechless.
Here were two of the economistshe respectedmost, and they iust-couldn't
listen. SuddenlyArthur had felt naive. Stupid. Like someonewho didn't
know enough not to believein increasingreturns.Somehow,it had been
the last straw
He'd barelypaid attentionduring the restof the lunch. After it wasover
and everyone had said their polite good-byes,he'd climbed into his faded
old Volvo and driven back over the Bay Bridge into San Francisco. He'd
taken the first exit he could, onto the Embarcadero.He'd stoppedat the
first bar he found. And he'd come in here to sit amidst the ferns and to
give some seriousthought to getting out of economicsentirely.
The trish ldea of a Hero 19
Somewhere aroundthe bottomof his secondbeer,Arthur realizedthat
the placewasbeginningto getseriouslynoisy.The yuppieswerearriving
in forceto celebratethe patronsaintof lreland.Well, maybeit wastime
to go home.This certainlywasn'taccomplishing anything.He gotup and
walkedout to his car;the foggydrizzlewasstill comingdown.
Homewasin PaloAlto, thirty-fivemilessouthof thecity in thesuburban
flatsaroundStanford.It wassunsetwhenhe finallypulledintothedriveway.
He must havemadesomenoise.His wife, Susan,openedthe front door
and watchedhim ashe waswalkingacross,the lawn:a slim, prematurely
grayman who doubtless lookedaboutasfed up and bedraggled ashe felt.
"Well," she said,standingtherein the doorway,"how did it go in
Berkeley? Did theylike yourideas?"
"lt wasthe pits," saidArthur. "Nobodytherebelievesin increasing
returns."
SusanArthur had seenher husbandreturningfrom the academicwars
before."Well," she said, trying to find somethingcomfortingto say,"l
guessit wouldn'tbe a revolution,wouldit, if everybody believedin it at
the start?"
Arthur lookedat her,struckspeechless for thesecondtime thatday.And
thenhe iustcouldn'thelp it. He startedto laugh.

The Education of a Scientist

When you'regrowingup Catholicin Belfast,saysBrianArthur,speaking

in the soft, high cadencesof that city, a certainrebelliousness setsin
naturally.It wasn'tthat he everfelt oppressed, exactly.His fatherwasa
bankmanager andhis familywassolidlymiddleclass.The only sectarian
incidentthat everinvolvedhim personally cameoneafternoonashe was
walkinghomein his parochialschooluniform:a bunchof Protestant boys
startedpeltinghim with bitsof brickandstone,andonepieceof brickhit
him in the forehead. (He couldhardlyseefor the bloodpouringinto his
eyes*but he damnwell threwthat brickback.)Nor did he reallyfeelthat
the Protestantsweredevils;his motherwasa Protestant who convertedto
Catholicism whenshemarried.He neverevenfelt especially political.He
tendedto be much moreinterested in ideasand philosophy.
No, the rebelliousness is justsomethingyoupickup fromthe air. "The
culturedoesn'tequipyou to lead,but to undermine,"he says.Look at
whomthe Irish admire:WolfeTone,RobertEmmet,DanielO'Connell,
PadraicPearse. "All the Irishheroeswererevolutionaries.The highestpeak
20 COMPLEXITY

of heroismis to leadan absolutelyhopeless revolution,and then givethe

greatest speechof yourlife from the dock-the nightbeforeyou'rehanged.
"In Ireland,"he says,"an appealto authori$ neverworks."
In an odd sort of way,Arthur adds,that sheakof Irish rebelliousness is
whatgot him startedin his own academiccareer.CatholicBelfasttended
to be rathercontemptuous of intellectuals.So, of course,he becameone'
In fact, he can rememberwantingto be a "scientist"asearlyasagefour,
longbeforehe knewwhata scientistwas.The idealustseemed deliciously
exoticand mysterious. And yet,havinggoftenthat ideain his head,young
Brianwasnothingif not determined. At schoolhe plungedinto engineering
andphysicsandhard-edged mathematics assoonashe could.And in 1966
he had takenfirst-class honorsin electricalengineeringat Queen'sUni-
versityin Belfast. "Oh, I suppose you'llendup a weeprofessor somewhere,"
saidhis mother,who wasin factveryproud;no onein her generation of
the family had everevenattendeda university.
Laterin 1966that samedetermination had led him acrossthe Irish Sea
to Englandand the Universityof Lancaster,wherehe startedgraduate
studiesin a highly mathematical form of engineering knownasoperations
research*basically, a setof techniques for calculatingsuchthingsashow
to organizea factoryto get the mostoutputfor the leastinput, or how to
keepa fighterjet undercontrolwhen it is buffetedby unexpected forces.
"At the time, Britishindustrywasin terribleshape,"saysArthur. "l thought
that maybethroughsciencewe could reorganize it and sortit out."
And in 1967,afterthe professorsat Lancaster had provedinsufferably
stuffyand condescending-"\fsll,rt says Arthur, doing his bestimitation
"it's
of boredBritishsnobbery, niceto havean Irishman in the department;
it addsa littlecolour"-he leftfor Americaand the University of Michigan
in Ann Arbor. "From the momentI setfoot here, I felt right at home,"
he says."This wasthesixties.The people were oPen' the culture wasoPen,
the scientificeducationwassecondto none.In the United States,anything
seemed possible."
The one thing that wasn'tpossiblein Ann Arbor, unfortunately,was
readyaccessto the mountainsand the sea,both of which Arthur loved.
So he arranged to finishhis Ph.D. workat Berkeley startingin the fall of
1969.And to supporthimselfin the summerbeforehand he appliedfor a
iob with McKinseyandCompany, oneof thetop management consulting
companies in the world. l
That wasa pieceof incrediblygoodfortune'Arthur didn't realizeuntil
lateriusthowluckyhewas;peoplewereclamoringto behiredat McKinsey.
But it turnedout thatthecompanylikedhisoperations research background
The lrish ldea of a Hero 21

and the fact that he knew German. They neededsomeoneto work out of
the Diisseldorf office. Was he interested?
Was he? Arthur had the time of his life. The last time he'd been in
Germany he'd worked at a blue-collar summer iob at 75 cents an hour.
Now here he was, twenty-threeyearsold, advisingthe board of directors
of BASF on what to do with an oil and gasdivision or a fertilizer division
worth hundredsof millions of dollars. "l learnedthat operatingat the top
was just as easyas operatingat the bottom," he laughs.
But it was more than iust an ego trip, Basically,McKinsey was selling
modern American managementtechniques(a conceptthat didn't sound
asfunny in 1969asit would havefifteen yearslater). "Companiesin Europe
at that time typicallyhad hundredsof suMivisions,"he says."They didn't
even know what they owned." Arthur discoveredthat he had a real taste
for wading into messyproblems like this and coming to grips with them
firsthand. "McKinsey was genuinely 6rst-rate," he says. "They weren't
selling theories and they weren't selling fads. Their approach was to ab-
solutelyrevel in the complexig, to live with it and breatheit. The McKinsey
team would staywith a company for five or six months or more' studying
a very complicated set of arrangements,until somehow certain patterns
became clear. We'd all sit around on the edgeof our desksand someone
'This
would say, must be happening becauseof that,' and someone else
'Then
would say, thot must be so.'Then we'd go out and check it. And
'Well, you're almost right, but you
maybe the local executive would say,
forgotaboutsuchand such.' Sowe'dspendmonthsclarifyingand clarifying,
until the issueswere all worked out and the answerspokefor itself."
It didn't takevery long for Arthur to realizethat, when it came to real-
world complexities,the elegantequationsand the fancy mathematicshe'd
spentso much time on in schoolwere no more than tools-and limited
tools at that. The crucial skill was insight, the ability to see connections.
And that fact, ironically,waswhat led him into economics.He remembers
the occasionvividly. It was shortly before he wasdue to leavefor Berkeley'
He and his American boss, GeorgeThucher,were driving one evening
through West Germany'sRuhr Valley, the country'sindustrialheartland.
And asthey went, Thucherstartedtalkingaboutthe historyof eachcompany
they passed-who had ownedwhat for a hundredyears'and how the whole
thing had built up in an absolutelyorganic,historicalway. For Arthur it
was a revelation."l realizedall of a suddenthat this was economics."If
he everwantedto understandthis messyworld that fascinated him so much,
if he ever wantedto make a real differencein people'slives, then he was
going to have to learn economics.
22 COMPLEXITY
So Arthur headedto Berkeleyafter that 6rst summer on an intellectual
high. And in total innocence,he announcedthat economicswaswhat he
would study.
Actually, he had no intention of completely shifting fields at this late
date. He'd alreadyfinishedmost of his requirementsfor a Ph.D. in oper-
ationsresearchat Michigan; the only remaininghurdle wasto completea
dissertation,the large piece of original researchwith which a Ph.D. can-
didate supposedlydemonshatesthat he or she has masteredthe craft. But
he had more than enoughtime to do that the Universityof California was
insisting that he hang around Berkeleyfor another three yearsto fulfill its
residency requirements. So Arthur was welcome to spend his extra time
taking all the economics courseshe could.
He did. "But after the McKinsey experience,I wasvery disappointed,"
he says."This was nothing like the historical drama I'd been so fascinated
with in the Ruhr Valley." In the lecturehallsof Berkeley,economicsseemed
to be a branch of pure mathematics."Neoclassical"economics,as the
fundamental theory was known, had reduced the rich complexity of the
world to a narrow set of abstractprinciplesthat could be written on a few
pages.Whole textbookswere practically solid with equations.The brightest
young economistsseemedto be devoting their careersto proving theorem
after theorem after theorem-whether or not thosetheoremshad much to
do with the world. "This extraordinaryemphasison mathematicssurprised
me," saysArthur. "To me, coming from applied mathematics,a theorem
was a statementabout an everlastingmathematicaltruth-not the dressing
up of a trivial observationin a lot of formalism."
He couldn't help but feel that the theory was iust too neat by half. It
wasn't the mathematical rigor he objectedto. He loved mathematics.After
all those yearsof studyingelectricalengineeringand operationsresearch,
moreover,he'd acquired considerablymore backgroundin mathematics
than most of his economics classmates.No, what bothered him was the
weird unreality of it all. The mathematicaleconomistshad been so suc-
cessfulat turning their discipline into ersatzphysicsthat they had leached
their theoriesclean of all human frailty and passion.Their theoriesde-
scribedthe human animal as a kind of elementaryparticle:"economic
man," a godlike being whosereasoningis alwaysperfect,and whosegoals
are alwayspursuedwith serenelypredictableself-interest.And iust as phys-
icists could predict how a particle will respondto any given set of forces,
economistscould predict how economic man will respondto any given
economic situation:he (or it) will iust optimize his "utility function."
Neoclassicaleconomicslikewisedescribeda societywhere the economy
The lrish tdea of a Hero 23

is poisedforever in perfect equilibrium, where supply alwaysexactly equals

demand, where the stock market is never iolted by surgesand crashes,
where no companyevergetsbig enough to dominate the market, and where
the magic of a perfectly free market makeseverythingturn out for the best.
It wasa vision that reminded Arthur of nothing so much asthe eighteenth-
century Enlightenment, when philosopherssaw the cosmosas a kind of
vast clockwork device kept in perfect running order by the laws o[ Sir Isaac
Newton. The only differencewasthat the economistsseemedto seehuman
society as a perfectly oiled machine governed by the Invisible Hand of
Adam Smith.
He iust couldn't buy it. Granted, the free market wasa wonderful thing,
and Adam Smith had been a brilliant man' In fairness,moreover' neo-
classicaltheorists had embroidered the basic model with all sorts of elab-
orations to cover things like uncertain$ about the future, or the transfer
of property from one generation to the next. They had adapted it to fit
taxation, monopolies, international trade, employment, finance, monetary
policy-everything economiststhought about. But none of that changed
any of the fundamental assumptions'The theory still didn't describethe
messiness and the irrationalityof the human world that Arthur had seen
in the valley of the Ruhr-or, for that matter, that he could see every day
on the streetsofBerkeley.
Arthur didn't exactly keep his opinions to himself. "I think I annoyed
severalof my professorsby showinga greatdealof impatiencewith theorems,
and by wanting to know about the real economy," he says.He also knew
he washardly alone in thoseopinions:he could hear the grumbling in the
hallways of any economics meeting he went to.
And yet, there wasalso a part of Arthur that found the neoclassicaltheory
breathtakinglybeautiful. As an intellectual tour de force it ranked right up
therewith the physicsof Newton or Einstein.It had the kind of hard-edged
clarity and precisionthat the mathematicianin him couldn't help respond-
ing to. Moreover, he could see why a previous generationof economists
had welcomed it so enthusiastically.He'd heard horror storiesabout what
economicswas like when they were coming of age.Back in the 1930s,the
English economist]ohn MaynardKeyneshad remarkedthat you could put
five economistsin a room and you'd get six differentopinions.And from
all reports,he was being kind. The economistsof the 1930sand 1940s
were long on insight, but they were often a trifle weak on logic. And even
when they weren't, you'd still find that they came to very different con-
clusionson the sameproblem: it turns out they were arguing from different,
unstatedassumptions.So thesemaior wars would be fought out between
24 COMPLEXITY
differentfactionsovergovernment policiesor theoriesof thebusiness cycle.
The generationof economisbwho craftedthe mathematical theoryin the
1940sand 1950swerethe YoungTurksoftheir day,a packofbrashupstarb
determinedto cleanout the stablesand makeeconomicsinto a scienceas
rigorousand aspreciseasphysics.And theyhad comeremarkably close;
the YoungTurkswho had achievedit*Kenneth Arrow of Stanford,Paul
Samuelsonof MIT GerardDebreuof Berkeley,Tialling Koopmans,and
Lionel McKenzieof Rochesteq amongothers-had deservedly goneon to
becomethe GrandOld Men, the new establishment,
Besides, if you weregoingto do economics at all-and Arthur wasstill
determined to do economics-what other theory wereyou goingto use?
Marxism?Well, this was Berkeley,and Karl Marx hadhis follow-
certainly
of
ers.But Arthur wasn'tone them: sofar ashe was concerned, thisbusiness
of classstruggleproceeding predictable
in scientifically stageswas iustplain
silly. No, asthe gambleroncesaid,the gamemaybe crooked, it's thebut
only gamein town. So he kepton with his courses,determinedto master
the theoretical toolshe couldn'tquitebelievein.

AU this time, of course,Arthur had beenworkingon his Ph.D. disser-

tationfor operations research. And his adviser,mathematician StuartDrey-
fus, hadprovedto be both an excellentteacheranda kindredspirit.Arthur
remembers stoppingby Dreyfus'sofficeto introducehimselfshortlyafter
he arrivedat Berkeleyin 1969.He meta long-haired bead-wearinggraduate
studentcoming out. "l'm looking for Professor Dreyfus," said Arthur.
"Could you tell me when he'sdue back?"
"I'm Dreyfus,"saidthe "student,"who wasin fact aboutforty.
Dreyfusreinforcedall the lessons that Arthur had learnedat McKinseS
and providedan ongoingantidoteto the economicsclasses' "He believed
in gettingto the heartof a problem,"saysArthur. "lnsteadof solving
incrediblycomplicated equations, he taughtme to keepsimplifyingthe
problemuntil you found somethingyou could dealwith. Look for what
madea problemtick. Look for the keyhctor, the keyingredient,the key
solution."Dreyfuswould not let him getawaywith fancymathematics for
its own sake.
Arthur tookDreyfus's lessons to heart."It wasboth goodandbad,"he
saysa bit sadly.Lateron, his ideas on increasingreturnsmight havegone
down betterwith traditional economists if he'd hiddenthem in a thicket
of mathematicalformalism. In fact, colleagues urgedhim to do so. He
"I
wouldn't. wanted to sayit as plainly and assimply asI could,"he says.
The lrish ldea of a Hero 25

In 1970 Arthur went back to Diisseldorffor a secondsummerwith

McKinseyandCompany,andfoundit to be iustasenthralling asthe first.
Sometimeshe wondersif he shouldhavekeptup his contactsthereand
becomea big-timeinternational consultant afterhe graduated. He could
haveaffordeda very luxuriouslifesgle,
But he didn't. Insteadhe found himselfbeingdrawnto an economics
specialtythat focusedon a problemevenmessierthan industrialEurope:
Third World populationgrowth.
Of course,it didn't hurt that this special$gavehim the opportuni$to
go backand forth for studyat the East-West PopulationInstitutein Hon-
olulu, wherehe could keepa surfboardreadyfor actionon the beaches.
But he was quite seriousabout it. This was the early 1970s,and the
population problemwasloominglarge.Stanford biologistPaulEhrlichhad
fust written his apocalyptic best-sellerThe Population Bomb.The Third
World wasfull of newlyindependent formercoloniesstrugglingto achieve
somekind of economicviability. And economists were full of theories
abouthow to help them. The standardadviceat the time tendedto place
a heavyrelianceon economicdeterminism: to achieveits "optimum" pop-
ulation,all a countryhad to do wasgiveits peoplethe right economic
incentivesto control their reproduction,and they would automatically
follow their own rationalself-interest.In particular,manyeconomists were
arguingthat when and if a country became a modern industrialstate-
organizedalong Westernlines, of course-its citizenrywould naturally
undergoa "demographic transition,"automatically loweringtheirbirthrates
to matchthosethatprevailed in Europeancountries.
Arthur, however,wasconvincedthat he had a betterapproach,or at
leasta moresophisticated one:analyzepopulationcontrolin termsof "time-
delayed"controltheory,thesubiectof hisPh.D,dissertation. "The problem
wasone of timing," he says. a "lf government manages to cut backon
birthstoday,it will affectschoolsizesin about l0 years,the laborforcein
20years,the sizeofthe nextgeneration in about30 years, andthe number
of retireesin about60 years.MathematicallXthis is verymuch like trying
to controla spaceprobefar out in the solarsystem,whereyour commands
takehours to reachit, or like trying to control the temperatureof your
showerwhenthere'sa half-minutedelaybetweenadiustingthe tapand the
hot waterreachingyou. Ifyou don't takethat delayinto accountproperly,
you cangetscalded."
ln 1977,Arthur includedhis populationanalysisasthe final chapterin
his dissertation:an equation-filledtomeentitledDynamicProgramming as
Appliedto Time-Delayed ControlTheory. "lt wasverymuchan engineering
26 COMPLEXITY

approachto the populationproblem," he says,looking back on it ruefully.

"lt was all fust numbers." Despiteall his experiencewith McKinsey and
Dreyfus, and despiteall his impatiencewith overmathematizedeconomics,
he was still feeling the same impulse that had led him into operations
researchin the first place:let's use scienceand mathematicsto help run
socieg rationally."Most peoplein developmenteconomicshavethis kind
of attitude," he says."They're the missionaries of this century,But instead
of bringing Christianity to the heathen,they'retrying to bring economic
developmentto the Third World."
What brought him back to reality with a lolt was going to work for a
small New York think tank known as the PopulationCouncil. He arrived
in 1974,after he had completedhis doctorateand spenta yearasa "postdoc"
researcherin the Berkeleyeconomicsdepartment. Physically,the Popula-
tion Council was about as far from the Third World as you could geL it
was set up in a Park Avenue skyscraperunder the chairmanship of |ohn D.
RockefellerIII. But it did fund seriousresearchinto conhaception,family
planning, and economicdevelopment.And mostimportant,from Arthur's
point of view, it had a policy of getting its researchers awayfrom their desks
and out into the field as much as possible'
"Brian," the director would ask, "how much do you know about pop-
ulation and developmentin Bangladesh?"
"Very little."
"How would you like to 6nd out?"
Bangladeshwas a watershedfor Arthur. He went there in 1975 with
demographer Geoffrey McNicoll, an Australian who had been a fellow
graduate student at Berkeley and who had been responsiblefor bringing
Arthur to the Population Council in the first place. They arrived in the
first plane permitted to land in the aftermath of a coup; they could still
hear machine guns firing as they touched down. Then they proceededinto
the countryside, where they acted like investigativereporters:"We talked
to headmen in the villages,women in the villages'everyone.We inter-
viewed and interviewed to understandhow the rural society worked." In
particular, they tried to find out why rural families were still producing an
averageof seven children apiece, even when modern birth control was
madefreely available-and even when the villagers seemedperfectly well
awareof the country's immense overpopulationand stagnantdevelopment.
"What we found was that the terrible predicament of Bangladeshwas
the outcome of a network of individual and group interestsat the village
level," saysArthur. Since children could go to'work at an early age, it
was a net benefit to any individual family to have as many children as
The lrish tdeaof a Hero 27
possible.Since a defenseless widow's relativesand neighborsmight very
well come in and take everythingshe possessed, it was in a young wife's
interest to have as many sonsas possibleas quickly as possible,so that she
would have grown sons to protect her in her old age. And so it went:
"Patriarchs,women who were trying to hold onto their husbands,irrigation
communities-all these interests combined to produce children and to
stagnatedevelopment. "
After six weeks in Bangladesh,Arthur and McNicoll returned to the
United Statesto digestthe information they had and to do further research
in the anthropology and sociologyiournals. One of Arthur's first stopswas
Berkeley,where he dropped by the economics department in search of a
reference.While he wasthere,he remembers,he happenedto flip through
a list of the latestcourseofferings.They were pretty much the samecourses
he had taken himself not so long ago. "But I had this very strangeimpres-
sion, as if I'd been offcenter a bit, that economicshad changedin the year
I'd been away. And then it dawned on me: eeonomicshadn't changed' I
had." After Bangladesh,all those neoclassicaltheorems that he'd worked
so hard to learn seemedso-irrelevant. "SuddenlyI felt 100percentlighter,
like a great weight had been lifted from me. I didn't have to believe this
anymorel I felt it as a greatfreedom."
Arthur and McNicoll's eighty-pagereport, publishedin 1978, became
something of a classicin social science-and was immediately banned in
Bangladesh.(Much to the chagrin of the elite in Dacca, the capital, the
authors had pointed out that the governmenthad essentiallyno control of
anything outsidethe capital;the countrysidewas essentiallybeing run by
local feudal godfathers.)But in any case, saysArthur, other missions for
the Population Council in Syria and Kuwait only reinforcedthe lesson:the
quantitative engineering approach-the idea that human beings will re-
spondto abstracteconomic incentiveslike machines-was highly limited
at best.Economics,as any historianor anthropologistcould havetold him
instantly, was hopelesslyintertwined with politics and culture. Perhapsthe
lessonwas obvious, saysArthur, "But I had to learn it the hard way."
That insight likewise led him to abandonany hope of finding a general,
deterministic theory of human fertility. Instead he began to conceive of
fertility as part of a self-consistentpattern of folkways, myths, and social
mores-a pattern,moreovei that wasdifferentfor eachculture."You could
measuresomethinglike income or childbearingin one counhy, and find
that another country had the samelevelsof one, and totally different levels
of the other. It would be a different paftern." Everything interlocked, and
no piece of the puzzle could be consideredin isolation from the others:
2B COMPLEXITY

"The numberof childreninteractedwith the waytheir societywasorga-

nized, and the way their societywasorganizedhad a lot to do with the
numberof childrentheyhad."
Patterns.Once he had made the leap, Arthur found that there was
somethingabout the conceptthat resonated. He had beenfascinatedby
patterns all his life. Given a choice he would always takethe window,seat
on airplanes, so he could look out on the ever-changing panorama below'
He wouldgenerally see the same elements everlvhere he went: rock, earth,
ice, clouds,and so on. But these elements would be organized into char-
acteristicpatternsthat might go on for half an hour. "So I asked myself the
question,why doesthat geological pattern exist? Why is there a certain
textureof rockformations and meandering rivers,and then half an hour
laterthere'sa totallydifferentpattern?"
Now, however,he beganto seepatternseverywhere he went' ln 1977,
for example,he left the PopulationCouncil for a U.S.-Soviet think tank
knownasIIASA: the InternationalInstitute for Applied Systems Analysis.
Createdby Brezhnevand Nixon asa symbol of d6tente, it was housed in
MariaTheresa's magnificent eighteenth-century "hunting lodge" in Lax-
enburg,a smallvillageaboutten milesoutsideof Vienna.It wasalso,as
Arthui quicklydetermined,within readydrivingdistancefromtheskislopes
of the TyroleanAlps.
"Whatstruckme," he says,"wasthatif youwentintooneof theseAlpine
villages,it would havetheseornate,Tlroleanroofsand balustrades and
balconies, with characteristic pitchesto theroofs,characteristic gables, and
characteristic shutterson the windows.But ratherthan thinking that this
wasa nice jigsawpuzzlepicture,I realizedthattherewasnot a singlepart
of the villagethat wasn'ttherefor a purpose,and interconnected with the
otherparts.The pitchesof the roofshad to do with whatwould keepthe
rightamountof snowon the rooffor insulationin the winter.The'degree
oiouerh".,gofthe gablesbeyondthebalconies hadto do with keepingsnow
from falling on thi balconies. So I used to amuse myselflookingat the
villages, thinking that this part has this purpose, that parthasthatpurpose,
andtheywere all interconnected."
Whai alsostruckhim, he says,wasthat just a$ossthe Italianborderin
the DolomiteAlps, the villagesweresuddenlynot Tyroleanat all. It was
no onething thai you couldpoint to. It wasiustthat myriadvariantdetails
addedup toa totallydifferentwhole.And yet the ltalianvillagersand the
Austrianvillagerswerecopingwith essentially thesameproblemof snowfall.
'ithe two cultureshad arrivedat mutuallyself-
"Over time,'; h" ,"yr,
consistent patterns thataredifferent."
The lrish ldeaof a Hero 29
Epiphanyon the Beach

Everyone has a researchstyle, saysArthur. If you think of a research

problem as being like a medievalwalled city, then a lot of peoplewill attack
it head on, like a battering ram. They will storm the gatesand try to smash
through the defenseswith sheerintellectualpowerand brilliance.
But Arthur has neverfelt that the batteringram approachwashis strength.
"I like to take my time as I think," he says."So I iust camp oubide the
city. I wait. And I think. Until one day-maybe after I've turned to a
completelydifferent problem-the drawbridgecomesdown and the de-
'We
fenderssay, surrender.'Theanswerto the problemcomesall at once'"
In the caseof what he later came to call increasingreturnseconomics,
he had been campedfor quite a long time. McKinsey. Bangladesh.His
generaldisillusionmentwith standardeconomics.Patterns.None of it was
quite the answer.But he can vividly rememberwhen the drawbridgebegan
to open.
It was in April 1979. His wife, Susan,was in a stateof exhaustionafter
finishing her Ph.D. in statistics,and Arthur had arrangedfor an eight-week
sabbaticalfrom IIASA so that they could take a much-neededrest together
in Honolulu. For himself, he made it a partial working vacation' From
nine in the morning until three in the afternoon he would go over to the
East-WestPopulation Institute to work on a researchpaper while Susan
continued to sleep-literally fifteen hours a day.Then in the late afternoon
they would drive up to Hauula beach on the north side of Oahu: a tiny,
almost desertedstrip of sand where they could body-surf and lie around
drinking beer, eating cheese,and reading. It was here, onelazy afternoon
shortly after they arrived, that Arthur had opened up the book he had
broughtalongfor just sucha moment:HoraceFreelandfudson'sTheEighth
Day of Creation, a 600-pagehistory of molecularbiology.
"l was enthralled," he recalls.He read how famesWatsonand Francis
Crick had discoveredthe double-helixstructureof DNA in 1952.He read
how the geneticcode had been broken in the 1950sand 1960s.He read
how scientistshad slowly decipheredthe intricatelyconvolutedstructures
of proteinsand enzymes.And as a lifetime laboratoryklutz-"I've done
miserablyin everylaboratoryI've beenin"-he readaboutthe painstaking
experimentsthat brought this scienceto life: the questionsthat made this
or that experimentnecessarS the monthsspentin planningeachexperiment
and assemblingthe apparatus,and then the triumph or deiection when the
answerwas in hand. "|udson had the ability to bring the drama of science
alive."
30 COMPLEXITY
But what really galvanizedhim wasthe realizationthat here was a whoie
messyworld-the interior of a living cell-that wasat leastas complicated
as the messyhuman world. And yet it wasa science."I realizedthat I had
been terribly unsophisticatedabout biology," he says."When you're trained
the way I was, in mathematicsand engineeringand economics, you tend
to view scienceas somethingthat only applieswhen you can usetheorems
and mathematics.But when it came to looking out the window at the
domain of life, of organisms,of nature, I had this view that, somehow,
sciencestopsshort." How do you write down a mathematicalequationfor
a tree or a paramecium?You can't. "My vaguenotion wasthat biochemishy
and molecularbiology were just a bunch of classifications of this molecule
or that. They didn't really help you understandanything."
Wrong. On every page, fudson was proving to him that biology was as
much a scienceas physicshad everbeen-that this messy,organic, non-
mechanisticworld wasin fact governedby a handful of principles that were
asdeepand profoundasNewton's'lawsof motion' In everyliving cell there
residesa long, helical DNA molecule: a chain of chemically encoded
instructions, genes, that together constitute a blueprint for the cell. The
genetic blueprints may be wildly different from one organism to the next.
But in both, the geneswill use essentiallythe same geneticcode' That
code will be decipheredby the same molecular code-breakingmachinery.
And that blueprint will be turned into proteinsand membranesand other
cellular structuresin the same molecular worlshops.
To Arthur, thinking of all the myriad forms of life on Earth, this was a
revelation.At a molecularlevel, every living cell was astonishinglyalike.
The basic mechanismswere universal. And yet a tiny, almost undetectable
mutation in the geneticblueprint might be enoughto producean enormous
change in the organism as a whole. A fbw molecular shifts here and there
might be enough to make the difference between brown eyesand blue,
between a gymnast and a sumo wrestler,betweengood health and sickle-
cell anemia. A few more molecularshifts,accumulatingover millions of
years through natural selection, might make the difference between a
human and a chimpanzee,betweena fig tree and a cactus,betweenan
amoebaand a whale. In the biological world, Arthur realized,small chance
eventsare magnified, exploited, built upon. One tiny accidentcan change
everything. Life develops.It hasa history.Maybe, he thought, maybethat's
why this biological world seemsso spontaneous,organic, and-well, alive'
Come to think of it, maybethat was alsowhy the economists'imaginary
world of perfect equilibrium had alwaysstruck him as static, machinelike,
and dead. Nothing much could everhappenthere;tiny chance imbalances
The lrish ldea of a Hero 31

in the marketweresupposedto die awayasquickly asthey occurred.Arthur

couldn't imagineanythinglesslike the real economy,wherenew products,
technologies,and marketswere constantly arising and old ones were con-
stantly dying off. The real economy was not a machine but a kind of living
system,with all the spontaneityand complexity that fudson was showing
him in the world of molecular biology. Arthur had no idea yet how to use
that insight. But it fired his imagination.
He kept reading: there was more. "Of all the drama in the book,l' says
Arthur, "what appealedto me most was the work of ]acob and Monod."
Working at the Institut Pasteurin Paris in the early 1960s,the French
biologistsFrancoisfacob and JacquesMonod had discoveredthat a small
fraction of the thousandsof genesarrayedalong the DNA molecule can
function as tiny switches.Turn one of theseswitcheson-by exposingthe
cell to a certain hormone, for example-and the newly activatedgenewill
sendout a chemical signalto its fellow genes.This signalwill then travel
up and down the length of the DNA molecule and trip other genetic
switches,flipping some of them on and some of them off. These genes,
in turn, start sendingout chemical signalsof their own (or stop sending
them out). And as a result, still more genetic switcheswill be tripped in a
mounting cascade,until the cell's collection of genessetdesdown into a
new and stablepattern.
For biologiststhe implications of this discoverywere enormous(so much
so that facob and Monod later sharedthe Nobel Prize for it). It meant that
the DNA residingin a cell'snucleuswasnot iust a blueprintfor the cell-
a catalog of how to make this protein or that protein. DNA was actually
the foreman in charge of construction. In effect, DNA was a kind of
molecular-scale computerthat directedhow the cell wasto build itselfand
repair itself and interactwith the outsideworld. Furthermore,Jacoband
Monod's discoverysolvedthe long standingmysteryof how one fertilized
egg cell could divide and differentiate itself into muscle cells, brain cells,
liver cells, and all the other kinds of cells that make up a newborn baby.
Each different type of cell correspondedto a different pattern of activated
genes.
To Arthur, the combination of d€i) vu and excitement when he read
this was overwhelming. Here it was again: patterns. An entire sprawling
set of self-consistentpatterns that formed and evolved and changed in
responseto the outside world. It reminded him of nothing so much as a
kaleidoscope,where a handful of beadswill lock in to one pattern and hold
it-until a slow turn of the barrel causesthem to suddenlycascadeinto a
new configuration. A handful ofpieces and an infinity ofpossible pafterns.
32 COMPLEXITY
Somehow,in a wayhe couldn'tquiteexpress, thisseemed to be the essence
of life.
When Arthur finishedfudson'sbook he went prowling through the
Universigof Hawaiibookstore, snatching up everybookhe could6nd on
molecularbiology.Backon the beach,he devoured themall' "l wascap-
tured,"he says,"obsessed." By the time he refurnedto IIASA in fune he
wasmovingon pureintellectual adrenaline. He stillhadno clearideahow
to applyall this to the economy.But he couldfeelthat the essential clues
werethere. He continuedto pour throughbiologytextsall that summer.
And in September, at the suggestion of a physicistcolleague at IIASA, he
starteddelvinginto the moderntheoriesof condensed matter-the inner
workingsof liquidsandsolids.
He wasasastonished ashehadbeenatHauulabeach.He hadn'tthought
that physicswas anything like biology.In fact, it wcsn'tlike biology;the
atomsandmolecules thatthe physicists usuallystudiedweremuch,much
simplerthan proteins and DNA. And yet,whenyou lookedat thosesimple
atomsandmolecules interacting in massive numbers,you sawall the same
phenomena: tiny initialdifferences producing enormously differenteffects.
Simpledynamics producing astonishingly complex behaviors' A handful
of piecesfallinginto a near-infini$ of possible patterns.Somehow, at some
verydeeplevelthat Arthur didn't know how to define, the phenomena of
physics andbiologywerethe same.
On the otherhand,therewasoneveryimportantdifferenceat a practical
level:the systems thatphysicists studiedweresimpleenoughthattheycould
analyzethem with rigorousmathematics. Suddenly,Arthur beganto feel
right at home. If he'd had any lingeringdoubtsbefore,he knew now he
wasdealingwithscience. "Thesewerenot iustfuzzynotions,"he says.
He found that he wasmostimpressed with the writingsof the Belgian
physicistIlya Prigogine.Prigogine,ashe laterdiscovered, wasconsidered
ty -r.ry other physiciststo be an insufferableself-promoter who often
exaggerated the significanceof what he had accomplished. Nonetheless,
he wasan undeniably compelling writer.And perhaps not coincidentally,
his workin the fieldof "nonequilibrium thermodynamics" hadconvinced
the swedishAcademyof Sciences to awardhim the NobelPrizein 1977.
Basically,Prigoginewasaddressing the question,Why is thereorderand
structurein the world?Wheredoesit comefrom?
This turns out to be a much tougherquestionthan it might sound,
especiallywhen you considerthe world'sgeneraltendencytowarddecay.
Iron rusts.Fallenlogsrot. Bathwatercoolsto the temperature of its sur-
roundings.Natureseemsto be lessinterestedin creatingstructuresthan
The lrish ldea of a Hero 33

in tearing structuresapart and mixing things up into a kind of average.

Indeed, the processof disorder and decay seemsinexorable-so much so
that nineteenth-century physicistscodified it as the second law of ther-
modynamics, which can be paraphrasedas"You can't unscramblean egg."
Left to themselves,saysthe secondlaw, atoms will mix and randomize
themselvesas much as possible.That's why iron rusts: atoms in the iron
are forever hying to mingle with oxygenin the air to form iron oxide. And
that's why bathwater cools: fast-moving molecules on the surface of the
watercollide with slower-movingmoleculesin the aiq and graduallytransfer
their energy.
Yet for all of that, we do seeplenty of order and shucture around. Fallern
logs rot-but trees also grow. So how do you reconcile this growth of
structurewith the secondlaw of thermodynamics?
The answer,as Prigogine and othen realized back in the 1960s,lies in
that innocuous-soundingphrase, "Left to themselves. ." In the real
world, atoms and molecules are almost never left to themselves,not com-
pletely; they are almost alwaysexposedto a certain amount of energy and
material flowing in from the outside.And if that flow of energyand material
is strong enough, then the steadydegradationdemandedby the secondlaw
can be partially reversed.Over a limited region, in fact, a system can
spontaneouslyorganizeitself into a whole seriesof complex structures.
The most familiar example is probably a pot of soup sitting on the
stovetop.If the gas is off, then nothing happens.fust as the secondlaw
predicts, the soup will sit there at room temperature,in equilibrium with
its surroundings.If the gasis turned on with a very tiny flame, then still
nothing much happens. The system is no longer in equilibrium-heat
energy is rising up through the soup from the bottom of the pot-but the
difference isn't large enough to really disturb anything. But now turn the
flame up just a little bit higher, moving the systemiust a little fartherfrom
equilibrium. Suddenly,the increasedflux of heat energyturns the soup
unstable.Tiny, random motions of the soup moleculesno longer average
out to zero; some of the motions start to grow. Portions of the fluid begin
to rise. Other portions begin to fall. Very quickly, the soupbeginsto organize
its motions on a large scale:looking down on the surfaceyou can see a
hexagonalpattern of convectioncells, with fluid rising in the middle of
each cell and falling along the sides.The soup has acquired order and
structure.In a word, it has begun to simmer.
Such self-organizingstructuresare ubiquitous in nature, said Prigogine.
A laser is a self-organizingsystemin which particlesof light, photons, can
spontaneouslygroup themselvesinto a singlepowerful beam that has every
34 COMPLEXITY
systempowered
photon moving in lockstep.A hurricaneis a self-organizing
by the steadystreamof energycoming in from the sun, which drivesthe
winds and drawsrainwaterfrom the oceans.A living cell-although much
too complicatedto analyzemathematically-is a self-organizing systemthat
survives by taking in energy in the form of food and excreting energy in
the form of heat and waste.
In fact, wrote Prigoginein one article, it's conceivablethat the economy
is a self-organizingsystem, in which market shuctures are spontaneously
organizedby such things asthe demand for labor and the demand for goods
and services.
Arthur sat up immediately when he readthose words. "The economy is
a self-organizing system."That wasitl That waspreciselywhat he had been
thinking ever since he'd read The Eighth Doy of Creation, although he
hadn't known how to articulateit. Prigogine's principleof self-organization,
the spontaneous dynamics of living systems-now Arthur could finally see
how to relate all of it to economic systems.
ln hindsight it was all so obvious. In mathematicalterms, Prigogine's
central point was that self-organizationdependsupon self-reinforcement:a
tendency for small effectsto become magnified when conditions are right,
instead of dying away. It was precisely the same messagethat had been
implicit in |acob and Monod's work on DNA. And suddenly,saysArthur,
,,1
recognized it as what in engineering we would have called positive
feedbact." Tiny molecular motions grow into convectioncells. Mild trop-
ical winds grow into a hurricane. Seedsand embryosgrow into fully de-
velopedliving creatures.Positivefeedbackseemedto be the sine qua non
of change,of surprise,of life itself.
And yet, positive feedbackis preciselywhat conventionaleconomics
didn't have, Arthur realized.Quite the opposite.Neoclassicaltheory as-
sumes that the economy is entirely dominated by negdtivefeedback the
tendency of small effectsto die away. In fact, he can remember listening
with some puzzlement as his economics professorsback in Berkeley had
hammered on the point. Of course, they didn't call it negativefeed-
"*"y
back. The dying-awaytendency was implicit in the economic doctrine of
,,diminishing returns": the idea that the second candy bar doesn't taste
nearly asgood asthe first one, that twice the fertilizer doesn'tproduce twice
the yield, that the more you do of anything, the lessuseful, Iessprofitable,
or lessenioyble the last little bit becomes.But Arthur could seethat the
net effect was the same:iust as negativefeedbackkeepssmall perturbations
from running awayand tearingthingsapartin physicalsystems, diminishing
returns ..rr*. that no one firm or product can ever grow big enough to
The lrish ldeaof a Hero 35
dominate the marketplace.When peopleget tired of candybars,they switch
to applesor whatever.When all the besthydroelectricdam siteshave been
used,the utility companiesstartbuilding coal-firedplants.When enough
fertilizer is enough, farmersquit applying it. Indeed, negativefeedback/
diminishing returnsis what underliesthe whole neoclassical vision of har-
mony, stabilig, and equilibrium in the economy.
But evenbackin Berkeley,Arthur the engineeringstudentcouldn't help
but wonder: What happensif you have positivefeedbackin the economy?
Or in the economics jargon, what happensif you have increasingreturns?
"Don't worry about it," his teachershad reassuredhim. "Increasing-
returns situations are extremely rare, and they don't last very long." And
since Arthur didn't have any particularexamplein mind, he had shut up
about it and gone on to other things.
But now, readingPrigogine,it all came flooding back to him. Positive
feedback, increasing returns-maybe these things did happen in the real
economy. Maybe they explainedthe liveliness,the complexity, the richness
he saw in the real-worldeconomyall around him.
Maybe so. The more he thought about it, in fact, the more Arthur came
to realize what an immense differenceincreasingreturnswould make to
economics. Thke efficiency, for example. Neoclassicaltheory would have
us believe that a free market will alwayswinnow out the best and most
efficient technologies.And, in fact, the market doesn't do too badly. But
then, Arthur wondered,what are we to make of the standardQWBRTY
keyboardlayout, the one used on virtually every typewriter and computer
keyboard in the Western world? (The name QWERTY is spelled out by
the 6rst six lettersalong the top row.) Is this the most efhcient way to arrange
the keyson a typewriter keyboard?Not by a long shot. An engineernamed
ChristopherScholesdesignedthe QWERfi layoutin 1873specificallyto
slow typistsdown; the typewriting machinesof the day tended to jam if the
typist went too fast. But then the Remington Sewing Machine Company
mass-produceda typewriter using the QWERTY keyboard, which meant
that lots of typists began to learn the system, which meant that other
typewriter companiesbeganto offer the QWERTY keyboard,which meant
that still more gpists began to learn it, et cetera, et cetera. To them that
hath shall be given, thought Arthur-increasing returns. And now that
QWERTY is a standardused by millions of people, it's essentiallylocked
in forever.
Or considerthe Beta versusVHS competitionin the mid-1970s.Even
in 1979 it was clear that the VHS videotapeformat was well on its way to
cornering the market, despitethe fact that many expertshad originally rated
36 COMPLEXITY
it slightly inferior to Betatechnologically.How could this havehappened?
Becausethe VHS vendorswerelucky enoughto gain a slightly biggermarket
sharein the beginning,which gavethem an enormousadvantagein spite
of the technologicaldifferences:the video storeshated having to stock
everythingin two differentformats,and consumershatedthe ideaof being
stuck with obsoleteVCRs. So everyonehad a big incentive to go with the
market leader. That pushed up VHS's market share even more, and the
small initial differencegrew rapidly.Once again, increasingreturns.
Or take this endlesslyfascinatingbusinessof pattems. Pure neoclassical
theory tells us that high-tech firms will tend to distribute themselvesevenly
acrossthe landscape:there'sno reasonfor any ofthem to preferone location
overanother.But in real life, of course,they fock to placeslike California's
Silicon Valley and Boston'sRoute 128 to be near other high-tech firms'
Them that has gets-and the world acquires structure. In fact, Arthur
suddenly realized, that's why you get patternsin any system:a rich mixture
of positive and negativefeedbacls can't help producing patterns. lmagine
spilling a little water onto the surfaceof a highly polished tray, he says;it
beadsup into a complex paftern of droplets. And it does so becausetwo
countervailing forces are at work. There is gravi$, which tries to spread
out the water to make a very thin, flat film acrossthe whole surface.That's
negativefeedback.And there is surfacetension, the attraction of one water
molecule to another, which tries to pull the liquid together into compact
globules.That's positivefeedback.It's the mix of the two forcesthat produces
the complex pattern of beads. Moreover, that pattern is unique' Try the
experiment again and you'll get a completely different arrangement of
droplets. Tiny accidentsof history-infinitesimal dust motes and invisible
irrelularities in the surfaceof the tray-get magnified by the positive feed-
back into maior differencesin the outcome.
Indeed, thought Arthur, that probably explainswhy history, in Winston
churchill's phrase,is iust one damn thing afteranother.Increasingreturns
can take a trivial happenstance-who bumped into whom in the hallway,
where the wagon train happenedto stop for the night, where trading posts
happenedto be set up, where ltalian shoemakershappenedto emigrate-
anJmagnify it into somethinghistoricallyirreversible.Did a certainyoung
actressbecome a superstaron the basisof pure talent? Hardly: the luck of
being in a.singlehit movie sent her careerinto hyperdriveon name rec-
og.,ilion alone, while her equally talentedcontemporaries went nowhere.
nid gritish colonists flock to cold, stormy, rocky shoresof Massachusetts
Bay becauseNew England had the best land for farms?No: They came
becauseMassachusettsBay was where the Pilgrims got off the boat, and
The trish ldea of a Hero 37

the Pilgrims got off the boat there becausethe Mayflowergot lost looking
for Virginia. Them that has gets-and once the colony was established,
there was no turning back. Nobody was about to pick up Boston and move
it someplaceelse.
Increasingreturns, lock-in, unpredictability,tiny eventsthat have im-
mense historical consequenss5-"fhs5e properties of increasing-returns
economics shockedme at first," saysArthur. "But when I recognizedthat
each properg had a counterpart in the nonlinear physicsI was reading, I
got very excited.Insteadof being shocked,I becamefascinated."Econo-
mists had actually been talking about such things for generations,he
learned. But their effortshad alwaysbeen isolatedand scattered.He felt as
though he were recognizingfor the first time that all theseproblemswere
the sameproblem. "l found myselfwalking into Aladdin's cave," he says,
"picking up one treasureafter another."
By the autumn, everythinghad fallen into place.On November5, 1979,
he poured it all out. At the top of one page of his notebook he wrote the
words "Economics Old and New," and under them listedtwo columns:

Old Economics New Economics

Decreasingreturns . Much use of increasingreturns

Basedon l9th-century physics . Basedon biology (structure,

(equilibrium, stability,deter- pattern, self-organization,life
ministic dynamics) cycle)

Peopleidentical . Focuson individual life; people

separateand different

If only there were no externali- . Externalitiesand differencesbe-

ties and all had equal abilities, come driving force. No Nir-
we'd reach Nirvana vana. Systemconstantly
unfolding.

Elements are quantities and . Elements are patternsand pos-

prices sibilities

No real dynamicsin the sense . Economy is constantly on thq

that everythingis at equilib- edgeof time. It rushesforward,
rium structuresconstantly coalescing,
decaying,changing.
3B COMPLEXITY
sim-
Seessubjectasstructurally Seessubjectasinherentlycom-
ple plex

Economicsassoftphysics ashigh-complexity
Economics
science

And so it went, for three pages.It was his manifesto for a whole new
kind of economics, After all those years,he says,"I finally had a point of
view. A vision. A solution." It was a vision much like that of the Greek
philosopherHeraclitus,who observedthat you can neverstepinto the same
river twice. In Arthur's new economics,the economicworld would be part
of the human world. lt would alwaysbe the same, but it would never be
the same. It would be fluid, ever-changing,and alive.

What's the Point?

To say that Arthur was bubbling over with enthusiasmfor his vision
would be an understatement.But it didn't take him too long to realize that
his enthusiasmwas lessthan infectious, especiallyto other economists."I
thought that if you did somethingdifferent and important-and I did think
increasingreturns made senseof a lot of phenomenain economicsand
gavea direction that was badly needed-people would hoist me on their
shouldersand carry me in triumph. But that was iust incredibly naive."
Beforethe month of Novemberwasout he found himself walkingin the
park near IIASA'SHapsburgpalace,excitedlyexplainingincreasingreturns
to a visiting Norwegian economist, Victor Norman. And he was suddenly
taken aback to realize that Norman, a distinguished international trade
theorist, was looking at him in bafflement What was the point of all this?
He heardmuch the samereactionwhen he beganto givetalksand seminars
on increasingreturnsin 1980.About half his audiencewould typicallybe
very interested, while the other half ranged from puzzled to skeptical to
hostile. What was the point? And what doesany of this increasing-returns
stuff have to do with real economics?
it? The
Questions like that left Arthur at a loss. How could they not see
point wasthat you have to look at the world as it is, not as some elegant
theory says it ought to be. The whole thing reminded him of medical
practice in the Renaissance,when doctors of medicine were learned in
matters of theory and rarely deigned to touch a real patient. Health was
simply a matter of equilibrium back then: If you were a sanguinePerson'
The lrish ldeaof a Hero 39
or a choleric person,or whatever,you merely neededto have your fluids
brought back into balance."But what we know from 300 yearsworth of
medicine, going frorir Harvey'sdiscoveryof the circulation of the blood
on through molecular biology, is that the human organismis profoundly
complicated. And that means that we now listen to a doctor who puts a
stethoscope to a patient'schestand looksat eachindividual case."Indeed,
it was only when medical researchersstartedpaying attention to the teal
complications of the body that they were able to devise proceduresand
drugs that actually had a chance of doing some good.
He saw the increasing-refurnsapproach as a step down that same path
for economics."The important thing is to observethe actualliving economy
out there," he says."lt's path-dependent,it's complicated,it's evolving,it's
open, and it's organic."
Very quickly, however,it becameapparentthat what was really getting
his critics riled up was this conceptof the economylocking itself in to an
unpredictableoutcome. If the world can organizeitself into many, many
possiblepatterns, they asked,and if the pattern it finally choosesis a his-
torical accident, then how can you predict anything? And if you can't
predict anything, then how can what you're doing be called science?
Arthur had to admit that was a good question. Economistshad long ago
gotten the idea that their field had to be as "scientific" as physics,meaning
that everything had to be mathematically predictable. And it was quite
some time before even he got it through his head that physicsisn't the only
kind of science. Was Darwin "unscientific" becausehe couldn't predict
what specieswill evolvein the nextmillion years? Are geologists unscientific
because they can't predict preciselywhere the next earthquake will come,
or where the next mountain ranee will rise?Are astronomers unscientific
becausethey can't predict preciselywhere the next starwill be born?
Not at all. Predictionsare nice, if you can make them. But the
"tt.n". f /
of sciencelies in explanation, laying bare the fundamental mechanismsof f f
nature.That'swhat biologists,geologists, and astronomers do in their fields.{
And that's what he was trying to do for increasing returns.
Not surprisingly,argumenb like that didn't convinceanyonewho didn't
want to be convinced. On one occasionat IIASA in February1982, for
example, as Arthur was answering questions from the audience after a
lecture on increasingreturns, a visiting U.S. economistgot up and de-
manded rather angrily, "fust give me one exampleof a technology that we
are locked in to that isn't superiorto its rivals!"
Arthur glancedat the lecture hall clock becausehe was running out of
time, and almost without thinking said, "Oh! The clock."
40 COMPLEXITY
The clock?Well, he explained,all our clocbloday havehandsthat move
"clockwise."But under his theory,you'd expectthere might be fossiltech-
nologies,buried deep in history, that might havebeen just as good as the
ones that prevailed. It's iust that by chance they didn't get going. "For all
I know, at some stagein history there may have been clocks with hands
that went backward. They might have been as common as the ones we
have now."
His questionerwas unimpressed.Another distinguishedU.S. economist
then got up and snapped,"I don't seethat it's locked in anyway.I wear a
digital watch."
To Arthur, that was missingthe point. But time was uP for that day. And
besides, it was iust a coniecture. About three weeks later, however, he
receiveda postcardfrom his IIASA colleaguefamesVaupel, who had been
vacationing in Florence. The postcard showed the Florence Cathedral
clock, which had been designedby Paolo Uccello in l44?'-and which
ran backward.(lt alsodisplayedall 24 hours.)On the fip side,Vaupelhad
simply written, "Congratulations!"
Arthur loved the Uccello clock so much that he made a transparencyo[
it so he could show it in overheadproiectors in all his future lectures on
lock-in. It alwaysproduced a reaction. Once, in fact, he was showing the
clock hansparencyduring a talk at Stanford when an economics graduate
student leaped up, flipped the'transparency over so that everything was
reversed,and saidtriumphantly,"You see!This is a hoax!The clockactually
goesclockwisel" Fortunately, however, Arthur had been doing a little re-
searchinto clocks in the meantime, and he had another transparencyof a
backwardclock with a Latin inscription. He put this transparencyup, and
said, "Unless you assumethis is mirror writing done by Leonardoda Vinci,
you have to accept that these clocks go backways."
Actually, by that point Arthur wasable to give his audiencesany number
of lock-in examples.There were-Ug!g5tlqq;!!g and Q-WERI*Y,of
course.But there wasalso the strangecaseof the-iqlejgalc@bgglign
."engine-Jnthe 1890s,Arthur discovered, when the automotiveindustry
wasstill in its infancy,gasg\ggwasconsidered the least-promisingpower
source.Itschiefrival,SlqLyas welldeveloped, gasoline
familiar,andsafe;
wasexpensiv.,,roiry,-Eif.rously explosive,hard to obtain in the right
grade,and requireda new kind of enginecontainingcomplicated new
parts.Casolineengineswere alsoinherentlylessfuel-efficient, If things
hadbeendifferentandif steamengineshadbenefitedfrom thesamenine$
yearsof development lavishedon gasolineengines,we mightnowbe living
The lrish ldea of a Hero 41

with considerably lessair pollution and considerably lessdependence on

foreignoil.
But gasolinedid win out-largely, Arthur found,because of a seriesof
historicalaccidents.In 1895,for example,a horseless-carriage competition
sponsoredby the Chicago Times-Heraldwaswon by a gasoline-powered
Duryea-one of only two carsto finishout of sixstarters. This mayhave
beenthe inspirationfor RansomOlds's1896 patent of a gasoline engine
that he subsequently mass-produced in the "Curved-Dash Olds." This
allowedgasolinepowerto overcomeits slow start. Then in l9l4 there
wasan outbreakof hoof-and-mouth diseasein North America,leading
to the withdrawalof horsetroughs*which werethe only placeswhere
steamcarscouldfill up with water.By thetimethe Stanleybrothers,makers
of the StanleySteamer,wereableto developa condenser and boiler sys-
tem that did not need to be refilled everythirty or forty miles, it was
too late. The steamcar neverrecovered. Casolinepowerquicklybecame
Iockedin.
And then therewasthe caseof.g.l.upget' When the UnitedStates
embarkedon its civilian nuclearpowerprogmmin 1956,a number of
designswereproposed: reactorscooledby gas,by ordinary"light" water,
by a moreexoticfluid knownas"heavy"water,andevenby liquid sodium.
Eachdesignhad it technicaladvantages and disadvantages; indeed,with
a perspective of thirtyyears,manyengineers believethatahigh-temperature,
gas-cooled. designwouldhavebeeninherentlysaferandmoreefficientthan
the others,and may haveforestalledmostof the public anxietyand op-
positionto nuclearpower.But as it happened,the technicalarguments
were almostirreleventto the final choice.When the Sovietslaunched
Sputnikin Octoberof 1957,the Eisenhower administration wassuddenly
eagerto get some reactor up and running-any reactor. And at the time,
the only reactorthat was anywhere nearbeing ready was a highly compact,
high-powered version of the light-waterreactor,which had been developed
by the Navy as a pg;ucr-Dla4!*&Lltj-IggLeel-CUbggegg." So the Navy's
design was hurriediy scaled up to commercialsize and placed into operation.
That led to further technical development of the light-water design, and
by the mid-1960s, it had essentiallydisplaced all the others in the United
States.
Arthur recallsusingthe light-water reactorexamplein 1984duringa
talk at the KennedySchoolof Governmentat Harvard."I wassayingthat
here'sa simplemodelthat showsthe economycan lock in to an inferior
outcome,asit appears to havedonewith thelight-waterreactor.Whereupon
42 COMPLEXITY
'Well,
a certain very distinguishedeconomiststoodup and shouted, under
perfectcapitalmarkets,it couldn't happen!'He gavea lot of technicalities,
but basically,if you wheel up a lot of extra assumptions,then perfect
capitalismwould restorethe Adam Smith world."
Well, maybehe was right. But six months later, Arthur gavethe same
talk in Moscow. Whereupon a member of the SupremeSovietwho hap-
penedto be in the audiencegot up and said,"What you'redescribingmay
happenin Westerneconomies.But with perfectsocialistplanningthis can't
happen.We would arrive at the correctoutcome."
Of course, so long as QWERTY, steam cars, and light-water reactors
were iust isolatedexamples,critics could alwaysdismisslock-in and in-
creasingreturnsas somethingrare and pathological.Surely,they said, the
normal economy isn't that messyand unpredictable. And at first Arthur
suspectedthat they might be right; most of the time the market is fairly
stable. It was only much later, as he was preparinga lecture on increasing
returns for a group ofpostgraduatestudents,that he suddenlyrealizedwhy
the critics were alsowrong. Increasingreturnsisn't an isolatedphenomenon
at all: the principle applies to everything in high technology.
[,ook at a softwareproduct like Microsoft'sSindows, he says.The com-
pany spent $50 million in research,nd d.u"iiliffint-to get ihe first copy
out the door. The second copy cost it-what, $10 in materials?It's the
same story in electronics, computers, pharmaceuticals,even aerospace.
(Cost for the first 82 bomber: $21 billion. Cost per copy: $500 million.)
High technology could almost be defined as "congealedknowledge," says
Arthur. "The marginal cost is next to zilch, which meansthat every copy
you produce makes the product cheaper and cheaper." More than that,
every copy offers a chance for learning: getting the yield up on micropro-
cessorchips, and so on. So there's a tremendous reward for increasing
production-in short, the systemis governedby increasingreturns.
Among high-techcustomers,meanwhile,there'san equallylargereward
for flocking to a standard."lf I'm an airline buying a Boeing iet," says
Arthur, "l want to make sure I buy a lot of them so that my pilots don't
have to switch." By the sametoken, if you're an office manager,you try
to buy all the samekind of personalcomputer so that everyonein the office
can run the samesoftware.The resultis that high technologies very quickly
tend to lock in to a relativelyfew standards:IBM and Macintosh in the
personalcomputerworld, or Boeing, McDonnell Douglas,and Lockheed
in commercialpassenger aircraft.
Now comparethat with standardbulk commoditiessuch as grain, fer-
tilizer, or cement, where most of the know-howwas acquiredgenerations
The lrish ldea of a Hero 43

ago.Todaythe real costsare for labor,land, and raw materials,areaswhere

diminishing returnscan set in easily.(Producingmore grain, for example,
may require that farmers start to open up lessproductive land.) So these
tend to be stable, mature industriesthat are describedreasonablywell by
standardneoclassicaleconomics."ln that sense,increasingreturns isn't
displacing the standardtheory at all," saysArthur, "Ifs helping complete
the standardtheory. It iust appliesin a differentdomain'"
What this meansin practicalterms, he adds,is that U.S' policy-makers
ought to be very careful about their economicassumptionsregarding,say,
tradepolicy vis-)-visfapan. "If you're using standardtheory you can get it
very badly wrong," he says.Severalyearsago, for example,he was at a
conferencewhere the British economistChristopherReeman got up and
declared that fapan's successin consumer electronicsand other hightech
marketswas inevitable.fust look at the counhy's low cost of capital, said
Freeman,along with its canny investmentbanks,its powerful cartels,and
its compellingneedto exploittechnologyin the absenceof oil and mineral
resources.
'Let's
"Well, I was the next speaker,"saysArthur. "So I said, imagine
that Thailand or Indonesia had taken off and fapan was still languishing.
Conventional economistswould then be pointing to all the same reasons
to explain Japan'sbackwardness.The low cost of capital means a low rate
of return on capital-so there's no reasonto invest. cartels are known to
be inefficient. Collective decision-makingmeansmolasses-slow decision-
making. Banksare not set up to take risks.And economies are hobbled if
they lack oil and mineral resources. So how could the fapanese economy
possiblyhave developed?'"
Since the )apaneseeconomyquite obviouslydid develoP'saysArthur,
he arguedfor a differentexplanation:"l saidthatJapanese companiesweren't
successfulbecausethey had some magical qualities that U.S. and Euro-
pean companiesdidn't have. They were successfulbecauseincreasingre-
turns make high+ech marketsunstable,lucrative, and possibleto corner-
and becausefapan understoodthis better and earlier than other coun-
tries. The fapaneseare very quick at learning from other nations. And
they are very good at targetingmarkets,going in with huge volume, and
taking advantageof the dynamics of increasingreturns to lock in their
advantage."
He still believesthat, saysArthur. And by the sametoken, he suspects
that one of the main reasonsthe United Stateshashad such a big problem
with "competitiveness"is that governmentpolicy-makersand businessex-
ecutivesalike were very slow to recognizethe,winner-take-all"gtgp*-o-f
4 COMPLEXITY

hJsb:tffb-aa*gts. All through the 1970sand well into the 1980s,Arthur

points out, the federal governmentfollowed a "hands-off" policy basedon
a conventional economic wisdom, which did not recognizethe importance
of nurturing an early advantagebeforethe other side locksin the market.
As a result, high-techindustrieswere treatedexactlythe sameas low-tech,
bulk-commodityindustries.Any "industrialpolicy" that might havegiven
a boost to infant industrieswas ridiculed as an assaulton the free market.
Free and open hade on everythingremaineda national goal. And firms
were discouragedfrom cooperatingby antitrust regulationsdrawn up in an
era when the world was dominated by bulk commodities. That approach
has begun to change a bit in the 1990s,saysArthur. But only a bit. So
he, for one, argues that it is high time to rethink the conventional wis-
dom in light of increasingreturns."If we want to continue manufacturing
our wealth from knowledge,"he says,"we needto accommodatethe new
rules."

Meanwhile, even as he was collectingdozensof real-worldexamplesof

increasingreturns, Arthur waslooking for a wayto analyzethe phenomenon
in rigorous mathematicalterms. "l'm certainly not againstmathematicsper
se," he says."I'm a heavy-duty user. I'm just againstmathematics when
it's misapplied,when it becomesformalism for its own sake." Used cor-
rectly, he says,mathematicscan give your ideasa tremendousclarity. It's
like an engineerwho gets an idea for a device and then builds a work-
ing model. The equationscan tell you which parts of your theory work
and which don't. They can tell you which conceptsare necessaryand
which aren't. "When you mathematizesomethingyou distill its essence,"
he says.
Besides,saysArthur, he knew that if he didn't come up with a rigorous
mathematicalanalysisof increasingreturns, the wider economicscom-
munity would never regard his theory as anything more than a collection
ofanecdotes. Look at what had happenedin every previous effort to intro-
duce the concept.Back in 1891, the greatEnglish economistAlfred Mar-
shall actually devotedquite a bit of spaceto the increasingreturns in his
Principlesof Economics-the book in which he also introducedthe concept
of diminishing returns. "Marshall thought very deeply on increasingre-
turns," saysArthur. "But he didn't havethe mathematicaltoolsto do much
with it in an analyticalway. In particular,he says,Marshall recognized
even then that increasingreturns could lead to multiple possibleoutcomes
The lrish ldea of a Hero 45

in the economy,which meant that the fundamentalproblem for economists

was to understandpreciselyhow one solution rather than another came to
be selected.And economistsever since have gotten hung up on the same
,,wherever there is more than one equilibrium point possible,,the
point.
Lutcom. was deemedto be indeterminate," he says."End of story.There
wasno theory ofhowan equilibrium point cameto be selected.And without
that, economists couldn't bring themselvesto incorporate increasing re-
turns."
Somethingsimilar happenedin the 1920s,w-hena number of European
economists iried to use increasing-returnsconceptsto explain why cities
grew and concentrated the way they did, and why different cities (and
lifferent countries)would specializein, say,shoesor chocolatesor fine
violins. The basicconceptswerecorrect,saysArthur' But againthe math-
ematicaltools just weren't there. "ln the face of indeterminacy,"he says,
"economicscame to a halt."
So Arthur sharpenedhis pencils and went to work. What he wanted was
a mathematical framework that incorporated dynamics-that showed ex-
plicitly, step by step,how the marketplacechoseamong the multiple possible
-"In
Lutcomes. the real world, outcomesdon't iust happen," he says'"They
build up gradually as small chance eventsbecome magnified by positive
feedbaclsj'What he finally cameup with in 1981,aftermany consultations
with friends and colleagues, was a set of abstractequations based on a
sophisticatedtheory of nonlinear, random processes.The equations were
aciually quite general, he says,and applied to essentiallyany kind of in-
creasing-returnssituation. Conceptually, however,they worked something
like this: Supposeyou are buying a car. (At the time, lots of peopleat IIASA
werebuying Volkswagensand Fiats.)And suppose,for the sake'ofsimplicity,
that there are iust two models to choosefrom. Call them A and B' Now,
you ve read the brochureson both cars,saysArthur, but they'repret$
similar,and you still aren'tsurewhich to buy. So whatdo you do?Like
anysensible person,you startaskingyourfriends.And thenit sohappens,
purely by chance,that the first two or threepeopleyou talk to saythat
they'vebeendrivingcarA. Theytell youthatit worksfine. Soyou decide
to buy one,too.
But notice,saysArthur, thereis nowonemoreA-typedriverin theworld:
you. And that meansthat the nextpersonto comealongaskingaboutcars
is iust a little more likely to encounteran A-typedriver. So that person
will be iust a little bit more likely to choosecar A than you were,With
enoughlucky breakslike this, car A can cometo dominatethe market'
46 COMPLEXITY
On the other hand, he says,supposethe breals had gone the other way.
Then you might have chosento go with model B, and then car B would
have gotten the edgeand come to dominate.
In fact, saysArthur, under some conditionsyou can even show math-
ematicallythat with a few lucky breakseither way in the beginning, this
kind of processcan produce any outcome at all. The car salesmight even-
tually come to lock in at a ratio of 40 percent A to 60 percent B, or 89
percent A to I I percent B, or anything else. And it all works purely by
chance."Showing how chanceevenhwork to selectone equilibrium point
from many possiblein random processes was the most challengingthing
I've ever don€," saysArthur. But by 1981, working in collaborationwith
his IIASA colleaguesYuri Ermoliev and Yuri Kaniovski of the Skorokhod
School in Kiev-"two of the best probability theoristsin the world"-he
had it. The three of them published the first of their severalpaperson the
subiect in the Soviet iournal Kibemetika in 1983. "No*," saysArthur,
"economistscould not only follow the entireprocessby which oneoutcome
emerged,they could see mathematicallyhow different setsof historical
accidentscould causeradicallydifferent outcomesto emerge."
And most important, he says,increasingreturns was no longer, in the
words of the great Austrian economistJosephSchumpeter,"a chaosthat is
not under analyticalconhol.'l

ViolatingSacredCround

In 1982,Arthursuddenly
foundIIASAto bea farlesshospitableplace
thanit hadbeen,courtesy
of the rapidlychillingColdWar.The Reagan
administration, eagerto avoid any further taint by associationwith the Evil
Empire, had abruptly pulled the United Statesout of the organization.
Arthur was sorry to go. He'd greatlyenioyedworking with his Soviet col-
leagues,and how could you beatan office in a Hapsburgpalace?But things
worked out well enough, as it happens. As a stopgap,Arthur took up a
one-year visiting professorshipat Stanford, where his reputation for de-
mography seemedto stand him in good stead.And shortly before his year
there drew to a close he got a call from the dean: "What would it take to
keepyou here?"
"Well," Afthur replied, securein the knowledgethat he alreadyhad a
fistful of iob offersfrom the World Bank, the London School of Economics,
and Princeton, "l seethere'sthis endowedchair coming open. ."
 47
The lrish ldea of a Hero
prestigiou-s'
The dean was shocked. Endowed professorshipsare very
ln
Thef are generally only awardedto the most distinguishedresearchen.
chairs!"
effect, they are sinecuresfor life. "We don't negotiatewith endowed
she declared.
would
"l wasn't negotiating,"said Arthur' "You iust askedme what it
take to keep me here."
becamethe
So they gaveit to him. In 1983,at agethir$-seven,Arthur
of
Professor Population Studies and Econom-
Dea., and iirginia Morrison
in academia!" he laughs' He was one of the
ics. "My first permanent iob
youngestendowed professors in Stanford's history'
to be gof
It was a moment to savor-which in retrospect,turned out "
to have many such momenb for a long while'
thing. He wasn't destined
demog-
Honieue, much his fellow economistsmay have liked his work in
his ideas on increasing'returns eco-
raphy, many of them seemedto find
nomics outrageous.
en-
To be fair, he says,many of them were also quite receptive' even
critics had almost always
thusiastic.But it was true that his most virulent
that
been Americans. And being at Stanfordbrought him face to face with
no sweat whatever' I could
fact. "I could talk about theie ideasin Caracas,
these
talk about them in Vienna, no sweat. But whenever I talked about
pay. People got angry at the
ideas in the United States,there was hell to
very notion that anything like this could happen'"
Arthur found the Am-ericans'hostility both mystifying and disturbing.
some of it he put down to their well-known fondnessfor mathematics.
of
After all, if you spendyour careerproving theoremsabout the existence
market .q,rilib.i,r*, and the uniquenessof market equilibrium, and_the
when
efficiency of market equilibrium, you aren't likely to be -v9ryhappy
,o-.o.t" along and tells you that there'ssomethingfishyabout market
"o*es
As the economistfohn R. Hicks had written in 1939' when
equilibrium.
he looked aghastat the implications of increasingreturns, "The threatened
wreckageis that of the greaterpart of economic theory'"
But Arthur also sensedthat the hostility went deeperthan that. American
economists are famous for being far more passionatelydevoted to free-
market principles than almost anyone else in the world. At the time, in
fact, the R.ajan administration was busily cutting taxes, iunking federal
regulations,"f,rivatizing" federalservices,and generallytreatingfree-market
..!itrlir* as a kind of statereligion. And the reasonfor that passion,as
Aithur slowly came to realize,was that the free-marketideal had become
bound up wiih American idealsof individual rightsand individual liberty:
4B CoMPLEXITY
both are groundedin the notion that societyworksbestwhen peopleare
left aloneto do whattheywant.
"Every democraticsocietyhasto solvea certainproblem,"saysArthur:
"If you let peopledo their own thing, how do you assurethe common
good?In Germany,thatproblemissolvedby everybody watchingeverybody
elseout the windows.Peoplewill comeright up to you andsay,'Put a cap
on that baby!'"
In England,they havethis notion of a bodyof wisepeopleat the top
lookingafterthings."Oh, yes,we'vehadthisRoyalCommission, chaired
by Lord So-and-So. We'vetakenall yourinterests intoaccount,andthere'll
be a nuclearreactorin your backyardtomorrow."
But in the United States,the idealis maximumindividualfreedom-
or, asArthur putsit, "lettingeverybody be their own )ohnWayneand run
aroundwith guns." Howevermuch that idealis compromised in practice,
it still holdsmythic power.
But increasingreturnscut to the heartof that myth. If small chance
eventscanlockyouin to anyof several possibleoutcomes, thentheoutcome
that'sactuallyselectedmaynotbethe best.And that meansthat maximum
individualfreedom-and the freemarket-might nof producethe bestof
all possibleworlds.So by advocatingincreasingreturns,Arthur wasin-
nocentlytreadinginto a minefield.
Well, he hadto admitthathe'dhadfair warning.
It wasin 1980,he recalls.He had beeninvitedto givea seriesof talks
on economicdemography at the Academyof Sciences in Budapest. And
one evening,at the barof the Budapest Intercontinental Hotel,he found
himselfchattingwith academician MariaAugusztinovics. Standingthere
with a scotchin one hand and a cigarettein the other,shewasa most
formidablelady.Not only hadshemarried,in succession, mostof the top
economistsin Hungarybut shewasa very perceptiveeconomistherself.
Moreover,shewasan influentialpolitician,with a posthigh in the Hun-
gariangovernment.Shewasrumoredto eatbureaucrats for breakfast.
Ar-
thur sawno reasonto doubtit,
What areyou workingon thesedays?sheasked.Arthur enthusiastically
launchedinto a discourse aboutincreasing returns."lt explainsso many
problems,"he concluded,"all theseprocesses and patterns."
Augusztinovics, who knewexactlywhatthephilosophical stakeswerefor
Westerneconomists, simplyIookedat him with a kind of pig. "They will
crucifyyou," shesaid.
"Shewasright,"saysArthur."The yearsfrom 1982through1987were
dreadful.That'swhenmy hair turnedgray."
The lrish ldeaof a Hero 49
.,If
Arthur hasto admit that he broughta lot of that agonyon himse]f.
I hadbeenthekind of personwhoformsinsideallegiances in theprofession,
he says' "But I'm not
then the wholething might havegonesmoother,"
an insiderby nature.I'm iustnot a ioiner'"
With thai lrish streakof rebelliousness, he wasalsonot in a mood to
dressup his ideasin a lot of iargonandphonyanalysis iustto-makethem
palatabieto the mainstream.'Ri-td that'swhatled him to makea critical
first
Lctical blunder:in the summerof 1983,when he waspreparinghis
paperon increasingreturnsfor ofFcialpublication,he wrote the thing in
moreor lessplain English.
,,I *", conui.rced that I wasontosomething crucialin economics," he
l,So I decidedthat I shouldwrite it at a very intelligiblelevel'
.*pi"inr.
whereit couldbe understood evenby undergraduates. I thoughtthatfancy
mathematics would iust get in the wayof the argument.I alsothought,
'Gee, I've publishedhe"uily tnathematical papersbefore'I don't needto
'proveanything.' "
Wrorrg.tf h-ehadn'tknownit before,he says,he learnedit soonenough'
Theoreticaleconomists usetheir mathematical prowess the waythe great
stagsof the forest use their antlers:to do battle with one anotherand to
.rtiUtlrh dominance. A stagwho doesn't use his antlers is nothing.It was
fortunatethat Arthur circulated his manuscript informally that autumnas
an IIASA working paper. The official, published version wasn't to seethe
light
"The of dayfor anothersix Years.
mostprestigious u.S. iournal,TheAmericanEconomicReview, sent
the paperbackin early1984with a letterfrom the editor saying, in essence'
..No way!,'The sentthe paperbacksaying
euar:terlylournalof Economics
that its ,.ui"*.r, could find no technical fault, but that'they iust didn't
think the workwasworthanythin g. The AmericanEconomicReview, under
a neweditorthistime, tentatively accepted the paper on its second submittal,
bouncedit around internallyfor two and a half yearswhile demanding
innumerablerewrites,andthenreiectedit again.AndTheEconomiclournal
in Britain simplysaid,"Nol" (Aftersomefourteenrewrites,the paperwas
finally accepted by The Economicloumalandpublishedin March 1989as
,.CompetiniTechnologies, IncreasingReturns,and Lock-Inby Historical
Events.")
Arthur wasleft in helpless rage.Martin Luthercouldnail his ninety-
fivethesesto the churchdoorof Wittenbergto be readby oneand all. But
in modernacademia,thereare no churchdoors;an ideathat hasn'tbeen
publishedin an established iournal doesn'tofficiallyexist.And what he
iound doublyfrustrating,ironically,wasthe factthatthe ideaof increasing
50 COMPLEXITY
returnswasfinally beginningto catchon. It wasbecomingsomethingof
a movementin economics-andso long as his paperwasin limbo, he
couldn'ttakepartin it.
Thkethe economichistorians,for example-the peoplewho did em-
pirical studieson the historyof technologies, the origin of industries,and
thedevelopment of realeconomies. Stanfordhada first-class groupof them,
andtheyhadbeenamongArthur'searliestandmostenthusiastic supporters.
For years,they hadsufferedfrom the factthat neoclassical theory,ifreally
takenseriously, saysthat historyis irrelevant.An economyin perfectequi-
librium exisboutsideof history;the marketplace will converge to the best
of all possibleworldsno matterwhat historicalaccidentsintervene.And
while very few economisbtook it quite that seriously,a lot of economics
departments aroundthe countrywerethinkingof scrapping their required
courses in economichistory.Sothe historianslikeillock-in.They likedthe
ideathat small eventscould havelargeconsequences. They sawArthur's
ideasaboutincreasingreturnsasprovidingthem with a rationalefor their
existence.
No onewasa moreeffectiveadvocate of thatpoint of viewthanArthur's
StanfordcolleaguePaul David, who had independently publishedsome
thoughnaboutincreasingreturnsand economichistorybackin the mid-
1970s.But from Arthur'spoint of view,evenDavid'ssupportbackfired.At
the nationalmeetingof theAmericanEconomicsAssociation in late 1984,
Davidparticipatedin a paneldiscussion on "What Is the Useof History?"
and usedthe QWEMY keyboardexampleto explainlock-in and path
dependence to 600economists at once.The talk createda sensation. Even
the hard-coremathematicaleconomists wereimpressed: here wasa theo-
retical reasonfor thinking that history wasimportant..Even the Boston
Clobewroteaboutit. And Arthur wassoonhearingpeopleaskhim, "Oh,
you'refrom Stanford.Haveyou heardaboutPaulDavid'sworkon lock-in
and pathdependence?"
"lt wassimplydreadful,"Arthur recalls."I felt I hadsomething to say,
and I couldn'tsayit-and the ideasweregeftingcreditedto otherpeople.
It appearedthat I wasfollowingratherthan leading.I felt lilceI wasin
somedoomedfairy story."
The Berkeleydebaclewith Fishlowand Rothenberg in March 1987was
arguablyhis lowestmoment-but not by much. He beganto havenight-
mares."About threetimesa weekI'd havethis dreamof a planetaking
off-and I wasnot on it. I felt I wasdefinitelygeftingleft behind." He
seriously beganto think of abandoning economics and devotinghimself
The lrish ldea of a Hero
51
careerseemed
full time again to his demographicresearch.His academic
to be turning to ashes.
All that kept him going was stubbornness'"I lust pushed' and-pushed'
had to give
and pushed,';h. r"1o. "ilust kept believingthat the system
somewhere."
much
Actually, he was right. And as it happens, he didn't have too
longer to wait.
l*t*i**@***6s &$$4SS8SSS 9S *S**A** aatc{t

About a month after his ill-starredtrip to Berkeley,as Brian Arthur was

walkingacrossthe StanfordUniversitycampuson a sunnyCaliforniaday
in April 1987, he wasstartledto seea bicyclepull up in front of him
bearinga distinguished figurein sportscoat,tie, andbatteredwhitebicycle
helmet."Brian," saidKennethArrow,"I wasiustgoingto call you."
Arrow. Arthur wasinstantlyon the alert. It wasn'tthat he wasafraidof
Arrow, exactly.True, Arrow had preftymuch inventedthe kind of hyper-
mathematical economicsthat he wasrebellingagainst.But he knewArrow
to be an affable,open-mindedman who lovednothingbetterthan a good
debate,and who could still be your ftiend aftertearingyour argumentsto
shreds.No, it wasjust that-well, talkingto Arrow waslike talkingto the
pope.Arrow wasarguablythe finestliving economistin the world. He had
won the NobelPrizein economicsmorethana decadebefore.At agesixg-
five he still possessed
a lightning-fastintellectand a legendaryimpatience
with sloppyreasoning.He could changethe wholetoneof a seminariust
by walkinginto the room:The speaker wouldstartwalkingon eggs.People
in the audiencewould quit iokingaroundand straighten up. Everyone
would focus intently on the subjectat hand. They would frame their
questions and commentsvery,verycarefully.NobodXbut nobody,wanted
to look dumb in front of Ken Arrow.
"Oh, hi," saidArthur.
Arrow,clearlyin a hurry,quicklyexplained that'hewashelpingto put
togethera meetingof economists andphysicists at a smallinstitutein New
Mexico. It would be held towardthe end of the summer.he said.The
The Revoltof the Old Turks 53

plan was that he would invite ten economistsand that Phil Anderson, the
"So could
co.,derrsed-*atterphysicist,would invite ten physicalscientists.
you come and give a paper on mode-locking?" he asked'
"Certainly I iould,; Arthur heard himself say' ModeJocking? What the
devil was rnodeJocking?could Arrow be talking about his work on lock-
in and increasingreturns?Did Arrow even know about his work on in-
creasingreturns?"lfmm-whsre is this institute?"
,,lt'sin SantaFe, up in the foothillsof the Rockies,"Arrow said,climbing
back on his bicycle. with a quick good-byeand a promiseto send more
information later, he pedaled off, his white helmet making him visible
down Stanford'spalm-shadedwalkwaysfor quite a long distance'
Arthur staredafter him, hying to figure out what in the world he had
just committedhimself to. He didn't know which surprisedhim more: that
physicistswould want to talk to economists-or that Arrow would want to
talk to him.

A few weekslater, in May of 1987, Arthur got a telephonecall from

a soft-spokenman who introduced himself as George cowan, from the
Santa Fe Institute. Cowan thanked him for agreeingto come to the eco-
nomics meeting that fall. He and his colleaguestook this meeting very
seriously indeed, he explained. The institute was a small, private organi-
zation set up by the physicistMurray Gell-Mann and othersto study aspects
of complex systemi by which they meant everythingfrom condensed-
matter ihysics to society as a whole-anything with lots of strongly inter-
acting paris, The institutehad no faculty or students.But it was interested
in building as wide a network of researchers as possible.And economics
was very much on its agenda.
But what he was really calling for, added Cowan, was that Ken Arrow
had suggestedthat the institute invite Arthur to be a visiting fellow that
f"ll. T[is meant that Arthur could come out several weeks before the
economics meeting and then stay for severalweeksafterward, so that he
would have the time to talk and work with other researchersin residence
at the institute. Would he be interested?
"Certainly," saidArthur. Six weeksin SantaFe in the autumn, with all
expensespaid-why not? Besides,he had to admit that he was impressed
by the academicfirepower.After Arrow and Anderson, Cell-Mann was
the third Nobel laureate in a row that he'd heard of in connection with
this Santa Fe Institute. Gell-Mann was the fellow who'd invented the idea
of "quarks," the little thingies that are supposedto run around inside of
54 COMPLEXITY
protonsand neutrons.Arthur still had no clearideaof what this guy Cowan
meant by "complex systems."But the whole thing wasbeginningto sound
just crazy enough to be interesting.
"Oh, by the way," Arthur said, "l'm afraid no one's mention.d yout
name to me before.What do you do there?"
There wasa pauseand a kind ofcough at the other end ofthe line. "l'm
the president,"said Cowan.

George

Actually, Arthur wasn't the only one baffed by the Santa Fe lnstitute.
The first encounter was alwaysa bit of a shock to everybody.The place
violated stereotypes wholesale.Here was an outfit founded by aging aca-
demicsrich with privilege,fame, and Nobel Prizes-the very peopleyou'd
expectto be smugly content with the statusquo. And yet they were using
it as a platform to foment a self-proclaimedscientificrevolution.
Here wasan institutepopulatedlargelyby hard-corephysicistsand com-
puter iocksfrom Los Alamos, the original Shangri [,a of nuclear weaponry.
And yet the hallways were full of excited talk about the new sciencesof
"complexity": a kind of Grand Unified Holism that would run the gamut
from evolutionarybiology to fuzzy subjectslike economics,politics, and
history-not to mention helping us all to build a more sustainableand
peacefulworld.
In short, here was a total paradox.If you tried to imagine the SantaFe
Institutehappeningin the business world, you'dhaveto imaginethe director
of corporate research for IBM going off to starta little New Age Karmic
counselingservicein his garage-and then talkingthe chairmenof Xerox,
ChaseManhattan, and GM into ioining him.
What makes it even more remarkableis that the entrepreneurin this
picture-George A. Cowan, the former head of researchat [.os Alamos-
wasabout as un-New Age as anyonecould imagine.At agesixty-seven he
was a retiring, soft-spokenman who managedto look a bit like Mother
Teresain a golf shirt and unbuttoned sweater.He was not noted for his
charisma;in any given group he was usuallythe fellow standingoff to one
side, listening. And he was certainly not known for his soaringrhetoric.
Anyone who askedhim why he had organizedthe institutewas liable to
get a precise,high-mindeddiscussionof the shapeof sciencein the twenty-
first century and the needto takehold of scientificopportunities-the sort
The Revolt of the Old Turks 55

of recitation that might do well as an earnestguest editorial in Science

magazine.
only slowly, in fact, would it begin to dawn on the listener that cowan,
in his own cerebralway, was a fervent and determinedman indeed. He
didn't seethe SantaFe Instituteasa paradoxat all. He sawit as embodying
a purposefar more important than GeorgeA. Cowan, Los Alamos, or any
of the other accidentsof its creation-and for that matter, far more im-
portant than the institute itself. If things didn't work out this time, he often
said, then somebodywould iust have to do it all over again twenty years
down the road. To cowan, the SantaFb Institute wasa mission. To Cowan,
it wasa chancefor scienceas a whole to achievea kind of redemptionand
rebirth.

There was a time, distant as it seemsnow, when it wasperfectlypossible

for an idealistic young scientistto devotehimself to the creation of nuclear
weaponsfor the sakeof a betterworld. And GeorgeCowan has never found
causeto regret that devotion. "I've had secondthoughts my whole life,"
..But regretson moral grounds?No. without nuclearweaPonswe
he says.
mighi have bee., on an even more ruinous road to destruction through
biological and chemical weapons.I suspectthat the history of the past 6f$
years has been a lot kinder to human beings than if the 1940s hadn't
occurred."
Indeed, he says,in those days, work on nuclear weaponswas almost a
moral imperative.During the war, of course,Cowan and his fellow scientists
saw themselvesin a desperaterace againstthe Nazis, who still had some
of the best physicistsin the world, and who were thought-wrongly, as it
turned out-to be way aheadon bomb designs."We knew that if we didn't
get cracking, then Hitler would get the bomb," saysCowan, "and that
would be the end."
He actuallyfound himself sweptup in the bomb effort beforethe Man-
hattan Project even existed.In the fall of 1941, when he wasa twenty-one-
year-oldwith a fresh undergraduatedegreein chemistry from the Worcester
PolytechnicInstitutein his hometownof Worcester,Massachusetts, he had
gone to work on the cyclotron proiect at Princeton, where physicists were
studyingthe newly discovered process of nuclear fissionand its effectson
an isotope known as uranium-Z35. His intention had been to start taking
graduatecoursesin physicson the side. But that intention got put on
indefinite hold as of December7, 1941, when the laboratorysuddenly
56 COMPLEXITY
wenton a seven-day workweek.Eventhen it wasfearedthat the Germans
wereworkingon an atomicbomb,he says,and the physicists werefrantic
to find out if sucha thing wasevenpossible."The measurements we were
makingwereabsolutelyessential to decidingwhetheryou could achievea
chain reactionin uranium," saysCowan.The answer,it turnedout, was
yes.And the federalgovernment suddenlyfounditselfto be much in need
of Mr. Cowan'sservices."That particularmix of chemishyand nuclear
physicsmademe an experton a numberof thingsthat wereneededin the
bombproject."
From 1942until the end of the war he workedout of the Metallurgy
lab at the Universityof Chicago,wherethe ItalianphysicistEnrico Fermi
wasleadingthe effortto build the 6rstatomic"pile"-4 stackof uranium
and graphiteblocksthat could demonstrate a conhollablechain reaction.
Asa veryjuniormemberof thatteam,Cowanbecame something of a jack-
of-all-trades,castinguranium metal, machiningthe graphiteblocksthat
wouldcontrolthe pile'sreactionrate,andanythingelsethatneededdoing.
But by the time Fermi'satomicpile successfully wentcriticalin December
1942,Cowanfound that his experience therehad madehim one of the
ManhattanProiect'sexpertson the chemistryof radioactiveelements.So
the project managersstartedsendinghim off to placeslike Oak Ridge,
Tennessee, wherehe helpedtheengineers atthehastilyconstructed nuclear
facilitytherefigureout exactlyhowmuchplutoniumtheywereproducing.
"I was single,so they transferredme all aroundthe qoUnhy,"he says.
lryr ;tr; waslikely
to be sent off to help fix it." Indeed,Cowanwasone of the very select
groupof peoplewhowereallowedto travelbackandforthbetweendifferent
componentsof the project,which waskepttightly compartmentalized for
"I
securi$reasons. don'tknowwhytheytrustedme," he laughs. drank "l
as much as anyoneelse." He still hasa souvenirof that period:a letter
ftom the Chicagopersonnelofficeto his local draft boardin Worcester,
statingthat Mr. Cowanpossessed skillsthat wereuniquelyusefulto the
war effort,that he had beengranteda defermentby the presidenthimself,
and would theypleasequit trying to reclassify him l-A?
After the war,the scientists'desperateraceagainstHitler wastransformed
into an anxiety-ridden raceagainstthe Russians. It wasa decidedlynasty
time,saysCowan.Stalin'sseizure of Eastern Europe,theBerlinblockade,
and then Korea-the Cold War seemedall too closeto becominga very
hotwarindeed.The Sovietswereknownto beworkingon theirownnuclear
capabilities. It seemedthatthe only wayto maintainthe precarious balance
of power-and not incidentally,to defendthe causeof democracyand
The Revoltof the Old Turks 57
human freedom-was to continue improving the nuclearweaponson our
side.That senseof urgencyis what led Cowan to return to Los Alamos in
in
|uly of 1949, having spentthe previousthree yearsgettinghis doctorate
physicalchemistryat CarnegieTech in Pittsburgh. It wasn't an automatic
choice. In fact he madeit only after considerable thought and soul search-
ing. But the decisionwas reinforcedalmost immediately.
A week or two after he arrived, Cowan recalls,the director of radiochem-
istry researchdropped by and, in a hush-hush, oblique way, askedhim if
his new laboratory was totally free of radioactive contamination. When
Cowan said yes, he and his facility were immediately commandeeredfor
a crash-priority, top-secretanalysisiob. The air samplesarrived that very
night. He wasn't told where they came from, but he could guessthat they
had been obtained from somewhetenear the bordersof Russia.And once
he and his colleagueshad detectedthe telltalesignsof radioactivefallout,
there was no getting around iL the Sovietshad explodedan atomic bomb
of their own.
"So they eventuallyput me on this panel in Washington,"saysCowan.
"Very covert." Cryptically known as the Bethe Panel-its first chairman
was Cornell University physicist Hans Bethe-it was actually a group of
atomic scientish convened to track Soviet nuclear weaponsdevelopment'
Cowan was thirg yearsold. HighJevel governmentofficials believedat the
start that the fallout detectedby the chemistscouldn't possiblymean what
it obviously did mean. The officials knew that it would be years before
Stalin had an atomic bomb; the Sovietsmust have had a reactor blow up.
"But the nice thing about radiochemistryis that you can tell exactlywhat
happened,"saysCowan. The distributionof radioactiveisotopesproduced
in a reactoris very different from the distributioncoming out of a bomb
explosion."It took a lot of debateto convincethem." But in the end, the
older and wiser officialshad no choice but to acceptthe hard evidence.
The Soviet bomb was dubbed "Joe-I" in honor of JosephStalin, and the
nuclear arms race was on.
So no, saysCowan, he has no apologiesto make for his work on nuclear
weapons.But he doeshave one very large regretabout thoseyears:his sense
that the scientific community collectively abdicatedresponsibilityfor what
it had done.
Oh, not immediately,of course,and not completely.In 1945a number
of the Manhattan Proiectscientistsat Chicagocirculateda petition urging
that the bomb be demonstratedon an uninhabited island instead of on
]apan itself. Then after the bombsweredroppedon Hiroshima and Nagasaki
and the war was over, many of the scientistson the proiect startedforming
58 COMPLEXITY
political activist groupslobbying for the strictestpossiblecontrol ofnuclear
weaponry-civilian contral, not military. Those yearssawthe founding of
The Bulletin of the AtomicScientists,a magazinefor dealingwith the social
and political consequencesof this new form of power, and the formation
of activist organizationssuch as the Federationof Atomic Scientists(now
the Federationof American Scientists),of which Cowan was a member.
"The people from the Manhattan Proiect who went to Washington were
listened to very carefully," saysCowan. "ln the 1940s, after the bomb,
physical scientistswere looked to as miracle workers.They had a lot to do
with drafting the McMahon bill that createdthe Atomic Energy Com-
missionand put atomic energyunder civilian control.
"But that effort wasn't as completely supportedby scientistsas it might
havebeen," saysCowan. And afterthe McMahon bill passedin fuly 1946,
the scientists'activism largely died away.It was probably inevitable, he
says.The culture of sciencedoesnot mix well with the culture of politics.
"scientists who go to Washington as scientistsgenerallyleave screaming,"
he says."lt's totally alien to them. They want policy to be made on the
basisof logic and scientificfacts,and that'sprobablyiust a will-o'-the-wisp."
But for whatever reason, the researcherswent happily back to their labs,
leaving war to the generalsand politics to the politicians. And in so doing,
saysCowan, they blew a chance for accessand influence that they may
never have again.
Cowan doesn't exempt himself from this indictment, although he ac-
tually remained more involved than most. ln 1954,for example,he became
presidentof an associationof Los Alamos scientistswho met with Atomic
EnergyCommissionchairmanLewis Straussat the heightof the McCarthy
uproar, when the senatorfrom Wisconsin seemedto be convincing everyone
that the counhy was riddled with communisb. Cowan and his colleagues
protestedthe anticommunist witch hunts and made the case for greater
freedom of information and lesssecrecyat the lab. They alsotried-without
much success-to defend former Manhattan Proiect director f. Robert
Oppenheimer, who was even then being strippedof his securi$ clearance
on the grounds that he might have associatedwith some pepplewho might
have attendeda Communist party meetingback in the 1930s'
With his ongoing service on the Bethe Panel, meanwhile-a task he
continued for some three decades-Qeq,'2n had come to realize what a
disturbingly simpleminded place Washington could be. In the aftermath
of World War II, he says,the United Stateshad emergedfrom its prewar
isolationism with a clear understandingthat, yes, military power was ex-
tremely important. But having learned that lesson, all too many officials
The Revoltof the Old Turks 59
'You 'em
seemedoblivious to anything else' "Their view was, gotta grab
by the balls."'he says."l felt then that power is a symphonyorchestra,
and too many peoplecould only play bull fiddle."
In fact, Cowan had the distressing sensethat the Sovietsunderstoodthe
intricate harmoniesof power much better than Washingtondid. "They
seemedto pay a great deal of attention to power's intellectual appeals,to
its emotionaland ideologicalaspects.And at the time, I thoughtthey were
paying a greatdeal of attention to its scientific aspects.It turns out that we
thought they wereten feettall and they weren't.But I wasthinking in terms
of the contrastbetweenthe Russianapproachand ours. They were playing
it as though it were a big chessgame, with lots of moves. We played it as
a more one-dimensionalsort of game."
Even at the time, saysCowan, he wondered if this was yet another area
wherescientistswerefailing in their responsibility. "l felt, althoughI didn't
have it spelled out in my own mind as well as I do now, that scientistswere
in a position to take a more generalview of the nature of the postwarworld-"
is
But the fact that they didn't. And more to the point, fre didn't. There
wasn'ttime. After the Soviets fired foe-l in August 1949, [,os Alamos had
gone full-speed-aheadon designsfor a much more powerful thermonuclear
weapon: the hydrogenbomb. And then, after the first H-bomb was tested
in the fall of 1952, the lab continued full-speed-aheedin the effort to make
the things smaller, lighter, more reliable, and easierto handle. Playedout
againstthe backdropof Korea and the continuing confrontation in Europe,
saysCowan, "There was a tremendousfeeling that nuclear weaponswere
going to tip the balance of power one way or another. This was a tremen-
dously important mission."
On top of that, Cowan wasbeing drawn more and more into management
responsibilityat Los Alamos,which didn't leavehim much time for science.
As a team leader, he was reduced to doing his own experimentson week-
ends. "So I've had a very undistinguishedscientificcareer,"he says,with
a trace of sadness.
The issuesof power and responsibilitycontinued to haunt him, however.
And in 1982, after Cowan had steppeddown as head of researchat Los
Alamos and accepteda seaton the White House ScienceCouncil, they
returned full force-even as he began to see the possibility of a second
chance.

If nothing else,Cowan'smeetingswith the White HouseScienceCouncil

were a vivid reminder of iust why all those researchershad so eagerlygone
60 COMPLEXITY

back to their labs in 1946. There he would sit with his fellow council
members:a bunch of augustscientistsgatheredaround some conference
table in the New Executive Office Building in Washington.Then the
president'sscienceadviser-George (fay)KeyworthII, who had been named
to the post the year before while he was a young division leader working
under Cowan at Los Alamos-would lay out a seriesof issuesfor their
comment. And Cowan would haveto admit to himselfthat he didn't have
a clue as to what to sayabout them.
"The AIDS thing was still quiet back then," he says,1'butthere was a
senseof suddenalarm. It was coming up every meeting. And frankly, I
was very puzzledhow to respond."Was it a public health issue?Was it a
moral issue?What? The answerwasn't so obvious at the time.
"Another issuewas manned spaceflight versusunmanned spaceprobes.
We were told that Congresswasn't going to vote a dime for the unmanned
spaceprogramwithout the manned component.But I had no idea if that
wastrue or not, It wasa political issuemuch more than a scientificissue'"
Then there was PresidentReagan's"Star Wars" StrategicDefenseIni-
tiative, the vision of a space-basedshield to protectthe country againsta
massivenuclearmissileattack.Wasit technicallyfeasible? Could it be built
without bankrupting the country? And even if it could be, was it wise?
Wouldn't it destabilizethe balanceof powerand spin the world into another
ruinous arms race?
And what about nuclea1power?How did you balancethe risk of a reactor
meltdownand the difficulty of disposingof nuclearwastesagainstthe virtual
certaintyof greenhousewarming due to the burning of fossilfuels?
And so it went. Cowan found the experiencedistressing."These were
very provocativelessonsin the interlinked aspectsof science, policy, eco-
nomics, the environment, even religion and morality," saysCowan. Yet
he felt incapableof giving relevantadvice. Nor did the other academic
typeson the ScienceCouncil seemto be doing much befter.How could
they?Theseissuesdemandedexpertiseover a broadrange.Yet as scientists
and as administrators,most of them had spent their entire lives being
specialists.The corporateculture of sciencedemandedit.
"The royal road to a Nobel Prize has generallybeen through the reduc-
tionist approach," he says-dissecting the world into the smallestand sim-
plest pieces you can. "You look for the solution of some more or less
idealized set of problems, somewhat divorced from the real world, and
constrainedsufficientlyso that you can find a solution," he says."And that
leadsto more and more fragmentationof science'Whereasthe real world
demands-though I hatethe word-a more holisticapproach. " Everything
The Revoltof the Old Turks 61
affects everything else, and you have to understand that whole web of
connections.
Even more distressingwas his sensethat things were only getting worse
for the younger generationof scientisb. fudging from what he'd seenof the
ones coming through Los Alamos, they were impressivelybright and en-
ergetic-but conditioned by a culture that was enforcing more and more
intellectual fragmentationall the time. Institutionally (asopposedto polit-
ically), universitiesare incrediblyconservative places.YoungPh.D.'s don't
dare break the mold. They have to spendthe better part of a decadein the
desperatepursuit of tenure in an existing department, which means that
they had better be doing researchthat the department'stenure committee
will recognize.Otherwise,they'regoingto hearsomethinglike, "Joe,you've
been working hard over there with the biologists.But how doesthat show
you're a leaderover here in physics?"Older researchers, meanwhile,have
to spend all their waking hours in the desperatepursuit of grantsto pay for
their research,which means that they had better tailor their proiectsto fit
into categoriesthat the funding agencieswill recognize.Otherwise, they're
going to hear something like, "foe, this is a great idea-too bad it's not
our department." And everybodyhas to get papersacceptedfor publication
in establishedscholarly iournals-which are almost invariably going to
restrict themselvesto papersin a recognizedspecialty.
After a few yearsof this, saysCowan, the enforcedtunnel vision becomes
so instinctive that people don't even notice it anymore. In his experience,
the closer any of his tos Alamos researcherswere to the academic world,
the harder it was to get them to participate in team efforts. "I've wrestled
with it for thirty years," he sighs.
As he thought about it, however,he beganto feel that the most distressing
thing of all was what this fragmentation processhad done to science as a
whole. The traditional disciplineshad becomesoentrenchedand so isolated
from one another that they seemedto be stranglingthemselves.There were
rich scientific opportunitieseverywhereyou looked, and too many scientists
seemedto be ignoring them.
If you wanted an example, Cowan thought, iust look at the kind of
opportunitiesopeningup in-well, he didn't really havea good name for
it. But if what he'd seenaround Los Alamoswasany indication, something
big was brewing. More and more over the pastdecade,he'd begun to sense
that the old reductionist approacheswere reaching a dead end, and that
even some of the hard-core physical scientistswere getting fed up with
mathematical abstractionsthat ignored the real complexitiesof the world.
They seemedto be half-consciouslygroping for a new approach-and in
62 COMPLEXITY
the process,he thought, they werecutting acrossthe traditionalboundaries
in a way they hadn't done in years.Maybe centuries.
One of their inspirations,ironically enough, seemedto be molecular
biology. That's not the sort of thing that mostpeoplewould expecta weapons
laboratoryto be interestedin. But in fact, saysCowan, physicistshave been
deeply involvedwith molecularbiology from the beginning.Many of the
pioneersin the 6eld had actuallystartedout as physicists; one of their big
motivations to switch was a slim volume entitled What ls Life?, a seriesof
provocativespeculationsabout the physicaland chemical basisof life pub-
lished in 1944 by the Austrian physicist Erwin Schr<idinger,a coinventor
of quantum mechanics.(Having fled from Hitler, Schrddingerspent the
war safelyensconcedin Dublin.)One of thosewho wasinfluencedby the
book was Francis Crick, who deducedthe molecular shucture of DNA
along with JamesWatson in 1953-using data obtainedfrom x-ray crys-
tallography,a kind of submicroscopicimaging techniquedevelopedby phys
icists decades earlier. Crick, in fact, had originally hained as an
experimentalphysicist.GeorgeGamow, a Hungariantheoreticalphysicist
who wasone of the original proponentsof the Big Bangtheoryof the origin
of the universe, becameintensely interestedin the structure of the genetic
code in the early 1950sand helped inspire still more physicistsinto the
field. "The first really perceptive lecture I heard on the subject was by
Gamow," saysCowan.
Molecular biologyhad fascinatedhim eversince,he says,especiallyafter
the discoveryof recombinantDNA technologyin the early 1970sgave
biologiststhe power to analyzeand manipulate life-forms almost molecule
by molecule. So when he became head of researchat the laboratory in
1978,he had quickly thrown his supportbehind a maior researchinitiative
in the field-officially to study radiation damagein cells, but actually to
get Los Alamos involved in molecular biology on a broaderfront. [t was a
particularlygood time to do so, he recalls.Under director Harold Agnew,
Los Alamos had nearly doubled its size in the 1970s,and had opened itself
up to much more nonclassifiedbasic and applied research.Cowan's em-
phasison molecular biology fit right in. And that program, in turn, had
had a tremendousimpaet on people'sthinking at the laboratory' Especially
his.
"Almost by definition,ll he says,"the physicalsciencesare fields char-
acterizedby conceptual eleganceand analytical simplicity. So you make a
virtue of that and avoid the other stuff." Indeed, physicistsare notorious
for curling their lips at "soft" scienceslike sociologyor psychology,which
try to grapple with real-world complexity. But then here came molecular
The Revoltof the Old Turks 63
biology, which describedincredibly complicated living systemsthat were
nonethelessgovernedby deep principles. "Once you're in a partnership
with biology," saysCowan, "you give up that elegance,you give up that
simplicity.You'remessy.And from thereit's somuch easierto startdiffusing
into economicsand social issues.Once you're partially immersed, you
might as well startswimming."
But at the sametime, saysCowan, scientists werealsobeginningto think
about more and more complex systemssimply becausethey could think
about them. When you're stuck with solving mathematical equations by
paper and pencil, how many variables can you handle without bogging
down?Three?Four?But when you haveenoughcomputerpowet you can
handle as many variablesas you like. And by the early 1980s,computers
were everywhere.Personalcomputerswere booming. Scientistswere load-
ing up their desktopswith high-poweredgraphicsworkstations.And the big
corporateand national labsweresproutingsupercomputerslike mushrooms.
Suddenly,hairy equationswith zillions of variablesdidn't look quite so
hairy anymore. Nor did it seem quite so impossibleto drink from the
firehoseof data. Columns of figures and miles of data tapescould be
transformed into color-coded maps of crop yields or of oil-bearing strata
lying under miles of rock. "Computers," saysCowan with considerable
understatement,"are great bookkeepingmachines."
But they could also be much more than that. Properly programmed,
computerscould become entire, self-containedworlds, which scientists
could explorein waysthat vastlyenrichedtheir understandingof the real
world. In fact, computersimulationhad becomeso powerfulby the 1980s
that somepeoplewere beginning to talk about it asa "third form of science,"
standing halfway between theory and experiment. A computer simulation
of a thunderstorm, for example, would be like a theory becausenothing
would existinsidethe computerbut the equationsdescribingsunlight,wind,
and water vapor. But the simulation would also be like an experiment,
becausethose equationsare far too complicatedto solveby hand. So the
scientistswatching the simulated thunderstorm on their computer screens
would see their equations unfold in patternsthey might never have pre-
dicted. Even very simple equations can sometimesproduce astonishing
behavior. The mathematicsof a thunderstormactually describeshow each
puff of air pusheson its neighbors,how eachbit of watervaporcondenses
and evaporates,and other such small-scalematters; there's nothing that
explicitly talks about "a rising column of air with rain freezing into hail-
stones" or "a cold, rainy downdraft bursting from the boftom of the cloud
and spreadingalong the ground." But when the computer integratesthose
 U COMPLEXITY

equations overmilesof spaceandhoursof time, thatis exactlythebehavior

theyproduce.Furthermore, thatveryfactallowsthescientists to experiment
with their computermodelsin waysthat they could neverdo in the real
world. What really causestheseupdraftsand downdraf*?How do they
changewhenI varythetemperature andhumidity?Which factorsarereally
importantto the dynamicsof this storm,and which aren't?And are the
samefactorsequallyimportantin otherstorms?
By the beginningof the 1980s,saysCowan,suchnumericalexperiments
had becomealmostcommonplace. The behaviorof a new aircraftdesign
in flight, the turbulentfow of interstellargasinto the mawof a blackhole,
the formationof galaxies in the aftermathof the Big Bang-at leastamong
physicalscientists,he says,the whole idea of computersimulationwas
becomingmore and more accepted."So you could beginto think about
tacklingverycomplexsystems."
But the fascinationwith complexitywent still deeperthan that, says
Cowan.In partbecause of their computersimulations,andin partbecause
of new mathematicalinsighb,physicists hadbegunto realizeby the early
1980sthat a lot of messy,complicatedsystemscould be describedby a
powerfultheoryknownas"rtonlineardynamics."And in the process, they
hadbeenforcedto faceup to a disconcerting fact the wholereallycan be
greaterthan the sum of its parts.
Now,for mostpeoplethatfactsoundsprettyobvious.It wasdisconcerting
for the physicisbonly because theyhad spentthe past300yearshavinga
d[],e, love affair with linear systems-in which the whole is preciselyequalto
rz,.gv$*l the sum of itr p"rtr}-ffi they had had pleng of ieasonto feel this
way.If a systemis preciselyequalto the sum of its parts,then eachcom-
ponentis freeto do its own thing regardless of what'shappeningelsewhere.
And that tendsto makethe mathematics relativelyeasyto analyze.(The
name"linear" refersto the factthat if you plot suchan equationon graph
paper,the plot is a straightline.) Besides, an awful lot ofnaturedo€sseem
to work that way.Soundis a linear system,whieh is why we can hearan
oboeplayingoverits stringaccompaniment andrecognize themboth' The
soundwavesintermingleand yet retaintheir separate identities.Light is
alsoa linear system,which is why you can still seethe WdlklDon'tWalk
signacrossthe streetevenon a sunnyday:the light raysbouncingfrom the
siln to youreyesarenot smashed to thegroundby sunlightsheamingdown
from above.The various light rays operateindependently, passingright
througheachotherasif nothing were there.ln some ways even theeconomy
is a linear system,in the sense that small economic agents can act inde-
pendently.When someone buys a newspaPer at the corner drugstore, for
The Revoltof the Old Turks 65
example,it hasno effecton yourdecisionto buy a tubeof toothpaste at
the supermarket.
However,it'salsotruethata lot of natureis not linear-includingmost
of what'sreallyinterestingin the world. Our brainscertainlyaren'tlinearl
eventhoughthe soundof an oboeand the soundof a stringsectionmayl
be independentwhen they enteryour ear,the emotionalimpactof bothll
soundstogethermaybe very much greaterthan eitherone alone.(This isll
whatkeepssymphonyorchestras in business.) Nor is the economyreally
linear.Millionsof individualdecisions to buy or not to buy canreinforce
eachother, creatinga boom or a recession. And that economicclimate
can then feedbackto shapethe very buyingdecisionsthat producedit.
Indeed,exceptfor the very simplestphysicalsystems, virtually everything
and everybodyin the world is caughtup in a vast, nonlinearweb of
incentivesand constrainband connections.The slightestchangein one
placecauseshemorseverywhere else,We ean'thelp but disturbthe uni-
verse,as S. Eliot almostsaid.The wholeis almostalwaysequalto a
T.
gooddealmorethanthe sumof its parb. And the mathematical expression
of thatproperty-to the extentthatsuchsystems canbedescribed by math-
ematicsat all-is a nonlinearequation: one whose graph is curvy.
Nonlinearequationsarenotoriouslydifficult to solveby hand,which is
why scientiststried to avoidthem for so long. But that is preciselywhere
computers camein. Assoonasscientists startedplayingwiththesemachines
backin the 1950sand 1960s,theyrealizedthata computercouldn'tcare
lessaboutlinearversusnonlinear.It wouldiustgrindout thesolutioneither
way.And astheystartedto takeadvantage of that fact,applyingthat com-
puter powerto more and morekindsof nonlinearequations,they began
to find strange,
wonderfulbehaviors thattheirexperience with linearsystems
had neverpreparedthem for.
The passage of a waterwavedowna shallowcanal,for example,turned
out to haveprofoundconnections to certainsubtledynamics in quantum
field theory:bothwereexamples of isolated,self-sustainingpulsesof energy
calledsolitons.The CreatRedSpoton fupitermaybeanothersuchsoliton.
A swirlinghurricanebiggerthan Earth, it hassustaineditselffor at least
400 years.
The self-organizing systems championed sovociferously by the physicist
Ilya Prigoginewerealsogovemedby nonlineardynamics;indeed,the self-
organized motionin a simmeringpot of soupturnedout to be governed
by dynamicsvery similar to the nonlinearformationof other kinds of
patterns,suchasthe stripesof.a zebraor the spobon a butterfly'swings.
But moststartlingof all wasthenonlinearphenomenon knownaschaos.
 66 COMPLEXITY
In the everyday worldof humanaffairs,no oneis surprised to learnthata
tiny event overherecan havean enormouseffectoverthere'For want of
a nail, the shoewaslost,et cetera.But whenthe physicists startedpaying
seriousattentionto nonlinearsystems in their own domain,they beganto
realizeiust how profounda principlethis reallywas.The equationsthat
governedthe flow'of wind and moisturelookedsimpleenough,for ex-
ample-until researchers realizedthat the flap of a butterfly'swings in
Texascouldchangethe courseof a hurricanein Haiti a weeklater.Or that
a fap of that butterfly'swingsa millimeterto the left might havedeflected
the hurricanein a totallydifferentdirection.In exampleafterexample,
the message wasthe same:everythingis connected,and often with in-
crediblesensitivity. Tiny perturbations won'talwaysremaintiny. Underthe
right circumstances, the uncertain$
slightest can growuntil the system's
futurebecomes utterly unpredictable-s1, ip a word,chaotic.
Conversely,researchers began to realize that even somevery simple
systems couldproduceastonishingly rich patterns of behavior.All that was
requiredwasa little bit of nonlinearity. The drip-drip-drip of waterfrom a
leakyfaucet,for example,could be as maddeningly regular as a metro-
nome-so long asthe leakwasslowenough.But if you ignored the leak
for a while and let the flow rateincreaseeverso slightln then the drops
would soonstartto alternatebetweenlargeand small:DRIP-drip-DRIP-
drip. If you ignoredit a while longerand let the fow increasestill more,
the dropswould soonstartto comein sequences of 4-and then 8, 16,
and soforth. Eventually,the sequence wouldbecomesocomplexthat the
dropswouldseemto comeat random-again,chaos.Moreover, thissame
patternof ever-increasing complexi$couldbeseenin thepopulationswings
of fruit flies,or in theturbulentflowof fuids,or in anynumberof domains.
It wasno wonderthe physicists weredisconcerted. They had certainly
knownthat thereweresomefunny thingsgoingon with quantumme-
chanicsandblackholesand such.But in the 100yearssincethe time of
Newton, they and their predecessors had gottenusedto thinking of the
everyday worldasa fundamentally tidy andpredictable placeobeyingwell-
understoodlaws.Now it wasasif they had spentthe pastthreecenturies
living on a tiny desertislandand ignoringwhatwasall aroundthem. "The
momentyou departfrom the linearapproximation," saysCowan,"you're
navigating on a very broad ocean."
I LosAlamos,as it happened,wasnearlyan idealenvironmentfor non-
linearresearch. Not only had the laboratory beena leaderin advanced
I computing since the 1950s, saysCowan, but the researchers therehadbeen
grapiling with nonlinear problems from the day the place wasfounded.
I
The Revolt of the Old Turks 67

High-energy particlephysics, fuid dynamics, fusionenergyresearch, ther-f

monuclearblastwaves,you nameit. By the early1970s,in fact, it wasl
clearthata goodmanyof thesenonlinearproblemswerethesameproblemsJ'
deepdown, in the senseof havingthe samemathematical structure.So1
peoplecould obviouslysavethemselves a lot of effort if they would fus\
startworkingon thoseproblemstogether.The result,with the enthusiastic ,'
supportof the LosAlamostheorygroup,wasa vigorousProgramfor non-
linear sciencewithin the theory?iuiiion, and eventuallya Centerfor I
NonlinearSystems operating entirelyon its own.
And yet, as intriguingas molecularbiologyand computersimulation
and nonlinearsciencewereseparately, Cowanhad a suspicion that they
were only the beginning. It was more a gut feeling than anything else.But
hesensed thatthere was an underlying unity here, one thatwould ultimately
encompass not iust physicsand chemistry,but biology,informationpro-
cessing,economics,political science,and everyother aspectof human
affairslWhat he had in mind wasa conceptof scholarship thatwasalmostI
medieval.If thisunig were real,he thought, it would be a way of knowingI
the world that madelittle distinction between biological sciences andphys I
icalsciences-orbetween eitherof thosesciences and historyor philosophy. J
Once, saysCowan, "The whole intellectualfabric wasseamless'"And
maybeit couldbe thatwayagain.
To Cowan, it seemedlike an incredibleopportuni$. So why weren't
scientists out in the universities iumpingon it? Well, to a certainextent
they were,here and there. But this reallybroadview he waslookingfor
seemedto be falling throughthe cracks.By its very nature,it lay outside
the purviewof anyoneacademic department. True,universities werefull
of "interdisciplinaryresearchinstitutes."But so far as Cowancould tell,
theseinstituteswererarelymuch more than a bunch of peoplewho oc-
casionallyshareda commonofficespace.Professors andgraduatestudents
still had to give their loyaltyto their home departments, which held the
powerto grantdegrees, tenute,andpromotions.leftto themselves, thought
Cowan,the universitiesweren'tgoingto pick up on complexityresearch
for a generationat least.
Unfortunately,Los Alamosdidn't seemlikely to pick up on it, either.
And that wastoo bad. Ordinarily,a weaponslaboratoryis a much betterl
environmentfor this kind of broad,multidisciplinaryresearchthan thel
universities are.That fact is somethingthat visitingacademics alwaysfind
startling.But it goesright backto the laboratory's founding,saysCowan.
The ManhattanProjectstartedwith a specificresearch challenge-building
the bomb-and broughttogetherscientists from everyrelevantspecialtyto
l=* COMPLEXITY
tackle that challenge as a team. Cranted, it was a pretty remarkableteam.
RobertOppenheimer,Enrico Fermi, NielsBohr, Johnvon Neumann, Hans
Bethe, Richard Feyqman, EugeneWigner-one observerat the time called
them the greatestgathering of intellects since ancient Athens. But that's
been the laboratory's approach to researchever since. The big iob for
managementwasto makesurethat the right specialists weretalkingto each
"l
other. sometimes felt like a marriage broker," saysCowan.
The only problem was that Cowan's grand synthesisiust wasn't part of
the laboratory'sbasic mission. Indeed, it was about as far from nuclear
weaponsdevelopmentas you could get. And thingsthat weren'tpart of the
laboratory's mission had essentiallyzero chance of getting funded there'
The laboratory would certainly go on doing bits and piecesof complexity
researchthe way it already had, thought Cowan. But it would never do
much more than that.
No, he thought, there was really only one way. Cowan beganto imagine
a new, independentinstitute. Ideally, that institute would combine the best
of both worlds: it would be a place having the broad charter of a university,
while retaining Los Alamos'ability to mingle the separatedisciplines'It
would almost certainly have to be physicallyseparatefrom Los Alamos, he
knew. But if possible, it should be close enough to share some of the
laboratory'spersonneland computer power. Presumablythat meant Santa
Fe, which was only thirty-five miles away and which was the nearestcity
of any size. But whereveryou put it, he thought, this institute ought to be
a place where you could takevery good scientists-people who would really
know what they were talking about in their own fields-and offer them a
much broader curriculum than they usually get' It ought to be a place
where senior researcherscould work on speculativeideas without being
laughed at by their colleagues, and where the brightest young scientists
could come and work alongsideworld-classfigures who would give them
credibility back home.
It ought to be a place, in short, that could educatethe kind of scientist
that had proved all too rare after World War II: "a kind of twenty-6rst-
century Renaissanceman," saysCowan, "starting in sciencebut able to
deal with the real messyworld, which is not elegant,which sciencedoesn't
really deal with."
Naive? Of course. But Cowan thought it iust might work, if only he
could entice people with the vision of an incredible scientific challenge.
As he put the question to himself, "What kind of sciencehdd to be taught
to brilliant scientistsin the 1980sand 1990s?"
Sd, who might be willing to listen?And, not incidentally,who might
The Revoltof the Old Turks 69
have the clout to make this thing work? As a trial run one day when he
wasin Washington, Cowan tried explaining the institute ideato the science
adviser,fay Keyworth, and his fellow scienceboardmember David Packard,
cofounderof Hewlett-Packard. Amazingly enough, they didn't laugh' In
fact, they were both quite encouraging.So in the springof 1983, Cowan
decided to take the idea to his weeklylunch companions,the Los Alamos
senior fellows.
They loved it.

The Fellows

From the outside,it would be easyenoughto dismissthe seniorfellows

as a collection of old geezerswho had been put out to pastureat a ridic-
ulously high salary. From the outside, that's iust about what they looked
like. The groupwascomposedof abouthalf a dozenlongtimeLos Alamites
who, like Cowan,.had done yeomanserviceat the lab and who had been
rewardedwith researchpositionsfree from any administrativechores or
other bureaucraticbusywork.Their only duties as a grouPwere to meet
for lunch in the cafeteria once a week and to occasionally advise the
laboratory director on various policy issues.
But, in fact, the fellows were a remarkablyfrisky group, the kind of guys
whose responseto their new statuswas to say, "Thank God I can finally
get some real work done." And since many of them had had heavy ad-
ministrativeresponsibilityat one time or another,they were not shy about
telling the laboratory director exactly what he ought to be doing, whether
he wanted to hear from them or not. So when Cowan laid out his institute
idea for them, looking for advice and maybe allies, he got both.
PeteCarruthers, for example, immediately resonatedwith Cowan's sense
that somethingnew wasin the air-and with his sensethat the opportunity
wasgoing begging.Under a rumpled and cynical exterior,Carrutherswas
passionately enthusiasticabout "complex" systems-"the next maior thrust
in science,"he declared.He had reasonto be. Brought in from Cornell
University to head the Los Alamos theory division in l97J-at the rec-
ommendation of a searchcommittee chaired by Cowan-he had managed
to hire nearly 100 new researchers and start half a dozen new research
groups even as the laboratory'sbudget for such things was going down'
Among other things, he had insistedon hiring a handful of young wild
men back in 1974to work on what wasthen an obscuresubfieldof nonlinear
dynamics. ("What am I supposedto pay them with?" asked his depu$
70 COMPLEXITY
director,Mike Simmons. "Find the money somewhere,"saidCarruthers.)
And it was under Carruthersthat that subFeldhad blossomed,making Los
Alamos into a world center for what was soon being called chaos theory.
So if Cowan wanted to build on that foundation, Carrutherswas ready to
help.
Another seniorfellow,astrophysicist StirlingColgate,declaredhis fervent
support for a different reason:"We neededanything that could organize
and reinforce the intellectual capability in the state," he says.Los Alamos,
despiteall its efforts to open up to the outside world, was still a scientific
enclavesitting up on its mesa in splendid isolation. In his ten yearsas
presidentof the New Mexico Instituteof Mining and Technologydown in
Socorro, 200 miles to the south, Colgate had learned all too well that the
restof the stateremainedbeautiful, but backward.All the billions of federal
dollarsthat had pouredinto the regionsincethe 1940shad had depressingly
little impacton its schoolsand industrialbase.Its universities weremediocre
at best. And largely becauseof that, high+ech entrepreneurslooking to
relocatefrom overcrowdedCalifornia would routinely fly right over the Rio
Grande valley en route to Austin and points east.
Along with Carruthers, Colgate had recently tried to get New Mexico
to dramatically upgrade its universig system.And they had quickly given
it up as hopeless:the statewas iust too poor. So Cowan's institute looked
like a last, best hope. "Anything that could raisethe intellectual extremum
of our environmentwas not only in our personalinterest,but in the lab-
oratory'sinterest,and most of all in the national interest," declaresColgate.
Senior fellowNickMetropolis liked the ideabecauseof Cowan'semphasis
on computation. And he had good reasonto: Mehopolis was pretty much
Mr. Computer at Los Alamos. It washe who had supervisedthe construc-
tion of the laboratory'sfirst computer back in the late 1940s,basing it on
a design pioneered by the legendary Hungarian mathematician fohn von
Neumann of Princeton'sInstitute for AdvancedStudy,who wasa consultant
and frequentvisitor at Los Alamos. (The machine wasdubbedthe Math-
ematical Analyzer, Numerator, lntegrator, And Computer: MANIAC.) It
was Metropolis, along with the Polish mathematician StanislausUlam,
who had pioneeredthe art of computersimulation.And in no smallmeasure
it was Metropolis who was responsiblefor Los Alamos' now having some
of the biggestand fastestsupercomputerson the planet.
And yet, Metropolis felt that the laboratory was not being sufficiently
innovative even in this arena. Along with Gian-Carlo Rota, a mathema-
tician from MIT who was a Los Alamos visiting fellow and who often came
The Revolt of the old Turks /1

for an extendedstay,Metropolispointedout to the assembled fellowsthat

computational science wasundergoing iustasmuch ferment as biology and
nonlinearsciences. Therewererevolutionary changes going on in hardware
designalone,he said.The existingone-step-at-a-time computers hadgotten
aboutasfastastheywereevergoingto get, anddesigners werebeginning
to investigate newkindsof computers thatcoulddo hundreds,or thousands,
or evenmillionsof computational stepsin parallel.It wasa goodthing,
too: anyonewho seriouslywantedto tacklethe kind of complexproblems
Cowanwastalkingaboutwasprobablygoingto needsucha machine.
But computationalsciencewent much further than that. Rota,in par-
ticular,thoughtof it asextendingall the wayto the studyof the mind-
basedon the idea that thinking and informationprocessing werefunda-
mentallythe samething. Also knownascognitivescience.. this wasa hot
areaand gettinghofter.When done properly,it combinedthe talentsof
neuroscientists studyingthe detailedwiring of the brain, cognitivepsy-
chologists studyingthesecond-by-second Process ofhighJevelthinkingand
reasoning, artificialintelligenceresearchers trying to modelthosethinking
processes in a computer-even linguistsstudyingthe structureof human
languages andanthropologists studyinghumanculture.
Now that, Rota and Metropolis told Cowan, wesan interdisciplinary
topic worthyof his institute.
AnothervisitorwasDavid Pines,who had startedsittingin on the dis-
cussionsat Metropolis's invitationin themidsummer of 1983.A theoretical
physicistfrom the Universityof lllinois, Pines was editor of the iournal
Reyiews of MoilernPhysics andchairman of the advisory boardfor the l,os
AlamosTheoryDivision.He also turned out to be someone who resonated
stronglywith Cowan'sideaof a grand synthesis in science. After all, much
of hisown research, startingwith his Ph.D. dissertation in 1950, hadbeen
focusedon innovativewaysof understanding "collective" behavior in sys-
temsof manyparticles; examples rangedfromthevibrationmodesof certain
massive atomicnucleito the quantumfow of liquid helium.And Pines
had beenknownto speculate aloudthat a similaranalysismight leadto a
betterunderstanding of collectivehuman behaviorin organizations and
societies. "So I had an intellectualpredisposition to the idea,"he says.
Pineswaslikewise an enthusiast for Cowan'svisionof a newinstitute.He'd
hadquitea bit of experience alongthoselineshimself,havingbeenfound-
ing directorof lllinois' Centerfor AdvancedStudy,anda longtimeregular
at the AspenCenterfor Physicsin Colorado.Go for it, he told Cowan;he
could hardly wait to get goingon this one, "l alwaysfind it greatfun to
72 COMPLEXITY
bring togethervery ablescientists to talk aboutsomethingquite new," says,
Pines."It can be as much fun to startan institutionas to write a good
scientificpaper."
And so it went. The fellowshad a greattime with the instituteidea,to
thepoint of occasionally gettinga bit giddy.Therewastheday,for example,
whentheyall gotveryexcitedaboutthe thoughtthatdreymightbefounding
"the New Athens"-a centerfor intellectualipQuiryon a Parwith the ci$-
statethat gaveus Socrates, Plato,and Arist6tle.On a morepracticallevel,
theydebatedinnumerable questions. How big shouldthe placebe?How
manystudents shouldit have-or",#ouldit haveany?How closelyshould
it be tied to Los Alamos?Shouldit havea permanent faculty,or should
peoplerotatethroughand"thengo backto their own institutions? And
gradually,beforeth.ey'fullyrealizedit, this hypothetical institutebegan to
becomemore-andmorerealin their minds.
The only''froblem, unfortunately,wasthat everybodyhad something
ditrqrerltinmind. "Everyweek,"sighsCowan,"We'dgobackto firstbase,
prfilgo roundand roundagain."
The most seriousbone of contentionwasalsothe mostfundamental:
What shouldthe institutebe about?
On one sidewereMetropolisand Rota,who felt the placeshouldfocus
exclusivelyon computationalscience.A grand"synthesis" wasnice, they
argued.But if nobodyherecouldquitedefineit, howcouldtheyeverhope
to get somebodyout there to drop $400 million on it? That's about what
you would needto endowa facility on the scaleof, say,the Rockefeller
Institutein New York. Of course,it wasn'tgoingto be easyto raisethat
kind of money in any case.But at leastif you focusedon information
processing and cognitivescience,you would covera lot of what George
wastalkingabout,andyou mightconceivably getan endowmentfrom one
of these new teen-age computer zillionaires'
On the othersidewereCarruthers, Pines,andmostof theothers.Com-
puterswere nice, they felt. And Metropolis and Rotacertainlyhada point
aboutthe money. But damn it, another computer researchcenter?Was
thatreallygoingto setanybody on fire? The instituteought to besomething
morethan that-even if theycouldn't figure out precisely what.And that
was just the problem.As senior fellow Darragh Nagle poinb out, "We
didn't articulatethe alternative very well." Everyone felt that Cowan was
right, that something new was brewing out there. But no one could do
much betterthan vague talk about "new ways of thinking."
Cowanhimselfkepta low profileon this issue.He knewwherefrewas
coming ftom: he privatelythought of the placeasan "institute on the art
The Revolt of the Old Turks /3

of survival."And to him thatmeanta programasbrbadasit possibly could

Ee]ffiiTtee of shingsasit possiblycould be. At the same time, however,
he wasconvincedthat gettinga consensus on the directionof the institute
wasfar moreimportantthan money or any of the restof the details.If this
institutewereiusta one-manshow, he felt, thenit wasn'tgoinganywhere.
After thirg yearsasan administrator, he was convincedthat the only way
to makesomethinglike this happen was to get a lot of peopleexcitedaboutl
it. ,,Youhaveto persuade very good people that this is an importantthing I
to do," he says."And by the way, I'm not talking abouta democracy'I'm I
talkingaboutthe top one-halfof one percent' An elite. But onceyou do I
that,tlen the moneyis-well, not easy,but a smaller partof theproblem'" I
It wassomethingof a slow-motiondebate,since everyone wasworking
morethan full time on their variousresearch proiects. (Cowan, in partic-
ular, wasimmersedin an experimentto detectsolarneutrinos, which are
near-invisibleparticlesemittedfrom the coreof the sun.) But that couldn't
lastforever.On August17, 1983,Cowancalledthefellowstogetherin one
of the laboratory administrationbuilding'sfourth-floorconference rooms
and suggested it wastime to getserious.Somefriendsof his weretalking
aboutoffering50or 100acresof landasa campusfor the institute,he told
them. But at a minimum, theywouldwantto knowwhatthe institutewas
goingto be about.
No go. The fellowswereamicably,but firmly, dividedinto two camps.
They endedthat meetingno closerto a resolutionthanbefore-which was
probablyiust aswell, sincethe couplewho hadpromisedCowanthe land
got divorceda few monthslaterand had to rescindthe offer.But Cowan
had to wonderif this wasevergoingto go anywhere.

Murray

It wasMurraywho reallybrokethelogiam.Professor MurrayGell-Mann

of Caltech,the fifty-five-year-oldenfant tenibleof particle physics.
Gell-Mann had called up Cowan about a week before the August 17
meeting,saying that Pines had told him about the institute idea' Gell-
Mann thought it was fantastic.He'd been wanting to do something like
thisall his life, he said. He wanted to tackleproblems like the riseand fall
of ancientcivilizations and the long-term of
sustainability our own civili-
zation-problems that would transcendthe disciplinaryboundariesin a
big way.He'dhadno success whatsoever gettinganythingstartedat Caltech.
So could he the institutediscussions the next time he was in Los
 74 COMPLEXITY
Alamos?(Gell-Mann had beena consultantto the laboratorysince.lhe
1950s,and camequiteoften.)
Cowancouldn'tbelievehis luck "By all means,comeon by!" If ever
thereweresomeonewho belongedto that top one-halfof one percent,it
wasMurrayGell-Mann.Bornandraisedin NewYorkCity,hisdark-rimmed
glasses and whitecrew-cuthair givinghim the lookof a cherubicHenry
Kissinger,Gell-Mannwasbrash,brilliant,charming,andincessantly ver-
bal-not to mentionbeingself-confident to thepoint of arrogance. In fact,
morethan one personfound him insufferable. He had spenta lifetime
being the smartestkid in class.At Caltech,wherethe late, irrepressible
physicistRichardRynman had entitledhis best-selling memoirsSurely
You'reloking,Mn Feynman!,it wassaidthat Cell-Mannwouldhaveto
call his own memoirsWell, You'reRight Again,Munay! On thoserare
occasions when he didn't get his own way,Gell-Mann could alsobe re-
markably childish:colleagues hadobserved hislowerlip extending outward
'
in what lookedsuspiciously like a pout.
But for all of that, Murray Gell-Mann wasclearlyone of the maior
figuresof twentieth-century science.,.'When he arrivedon the sceneasa
youngPh.D. in the early1950s,the subatomic worldseemed a senseless
mess-a hodgepodge of pi particles, sigma particles,
rho particles,and on
and on throughan endlesslist,of Creekalphabetical names assigned at
random.But twodecades laterylargely because of concepts thatGell-Mann
had pioneered,physicists weredrawingup GrandUnified Theoriesof all
the interparticleforcesandwereconfidentlyclassifying that hodgepodge of
particlesasvariouscombinations of "quark5"-simplesubatomic building
blocksthat Gell-Mannhad namedaftera made-upwordin )amesfoyce's
'.!For generation,"saysa theoreticalphysicistwho has
FinnegansWake. a
knownhim for Meng years,"Murray definedthe centroidof the research
effortin particlephysics.What Murraywasthinkingaboutwaswhatevery-
one else shouldbe thinking about.He knewwherethe truth lay, and he
led peopleto it." .?
On the faceof it, this thirty-yearpreoccupation with the inner reaches
of protonsandneuhonsmadeGell-Mannan oddrecruittoCowan'svision
of scientificholism;it's hardto imagineanythingmorereductionist. But,
in fact,Gell-Mann's interests werelegion.Hewasdrivenbyan omnivorous
I curiosity.He had beenknownto turn to strangers sittingnextto him on
I' an airplane and grill them about their life for
stories hours.He had first
co-e to science througha loveof naturalhistory,whichhestarted learning
at age6ve when his older brothertook him on naturewalla throughthe
Manhattanparks."We thoughtof New Yorkasa hemlockforestthat had
The Revoltof the Old Turks 75

beenoverlogged,"he says.Ever since,he had beenan ardentbird-watcher

and conservationist.As chairman of the committee on World Environment
and Resourcesat the fohn D. and Catherine T. MacArthur Foundation,
he had helped found a washington environmental think tank known as
the World ResourcesInstitute, and he was deeply involved in efforts to
preservetropical forests.
Gell-Mann likewise had a lifelong fascinationwith psychology,archeol-
ogy, and linguistics. (He originally enrolled as a physicsmaior at Yale-only
to satisfyhislather, who fearedhe would starveif he maioredin archeology.)
When mentioning a foreign scientisthe pronouncesthe name with a lov-
ingly precise accent-in any of severaldozen languages'One colleague
,.-.mbers mentioning that he would soon be visiting his sisterin lreland'
"What's her name?"askedGell-Mann.
"Gillespie."
"What doesit mean?"
"Well, in Caelic I think it means'servantof a bishop''
Gell-Mann thought fE a moment. "No-in medieval scots-Gaelic it
meansmore like'religious follower of a bishop.' "
And if anyoneaf .LosAlamos didn't alreadyknow it, Cell-Mann could
use that verbal ability with immensely persuasiveeffect. "Murray can im-
provise,on thespot, an inspirationalspeechthat may not be Churchillian,"
saysCarruthers, "but the clarity and brilliance of it are overwhelming." As
soonaste joined the institutediscussions, his argumentsfor a broad-based
institute gave the maiority of the fellows something to rally around, and
ther4dletropolis-Rota concept of a computer-focused institute quickly lost
rldtud..
Gell-Mann got his realchanceto shineiust afterChristmas1983.Thking
advantageof the fact that Gell-Mann, Rota, and Pines loved to spend the
holidays in New Mexico-in fact, Gell-Mann had iust finished building
a house in Santa Fb-Cowan called yet another meeting of the fellows to
try to get this institute moving.
Gell-Mann pulled out all the stops.These narrow conceptionsweren't
grand enough, he told the fellows. "We had to set ourselvesa really big
task.And that wasto tackle the great,emergingsynthesesin science-ones
that involve many, many disciplines." Darwin's theory of biological evo-
lution had been iust such a grandsynthesisin the nineteenthcentury, he
said. It combined evidence from biology, which revealed that different
speciesof plantsand animals wereclearly related;from the emergingscience
of geology, which showed that the earth was incredibly ancient and that
the past afforded immense vistas of time; and from paleontology, which
76 COMPLEXITY
proved that the plants and animals who dwelled in that immense past had
been very different from those alive today. More recently, he said, there
had been the grand synthesisknown asthe Big Bang theory, which detailed
how all the matter in all the starsand galaxieshad come into being in an
unimaginablyvastcosmic explosionsome fiftebn billion yearsago.
"l said I felt that what we should look for were greatsynthesesthat were
emergingtoday,that werehighly interdisciplinary,"saysCell-Mann. Some
were already well on their way: Molecular biology. Nonlinear science.
Cognitive science.But surelytherewereother emerging$f,ltffifffiEre,
IEffiil,EfthG new institute should seekthem out.
By all means, he added, choosetopics that could be helped along by
thesehuge, big, rapid computersthat peoplewere talking about-not only
becausewe can use the machines for modeling, but also becausethese
machinesthemselveswere examplesof complexsystems.Nick and Gian-
Carlo were perfectly correct:computersmight very well turn out to be part
of such a synthesis.But don't put blinderson beforeyou start. If you're
going to do this at all, he concluded,do it right.
To his listenersit was spellbinding stuff. "I had said it before," saysGell-
Mann, "but perhapsnot so convincingly."
Gell-Mann's rhetoric pretty much carried the day. Here, in compelling
terms,was the vision that Cowan and the maiorityof the fellowshad been
trying to articulate for nearly a year. After that it became more or less
unanimous: The fellows would try to build an institute with the broadest
possiblecharter. And if Gell-Mann was willing to go out and knock the
potential donors dead-as apparentlyhe was-then maybe it was time to
move.
With that settled,however,the group then had to deal with a lessexalted
question: Exacdy who was going to do the work? Who was going to make
this institute happen?
Everyonelooked in the obviousdirection.
Actually, this was about the last job that Cowan himself wanted. Yes,
the institute had been his idea. He believedin it. He thought it ought to
be done. He thought it had to be done. But, damn it, he'd been an
administrator practically all his adult life. He wastired of ih tired of always
scramblingfor funds, tired of telling his friendsthat he'd haveto cut their
budgets,tired of trying to sneakin his own scientific work on the weekends.
He was sixty-three years old and he had notebookscrammed with ideas
he'd neverhad time to work on. Searchingfor solarneutrinos,investigating
an extremely rare and intriguing form of radioactivity known as double
The Revoltof the Old Turks //

betadecay-this wasthe kind of sciencehe hadalwayswantedto do. And

now this waswhathe wasgoingto do.
So, of course,when Pinesnominatedhim to spearhead the effort'-he
said,"Yes."Cowanhad already given it some thought, sincePine,shad
talkedto him aboutthe nominationbeforehand. And what had finally
persuaded him wasthe samething that had always lured him into man'
agement positions at LosAlamos:"Management wasstuffthatother people
cJuld do-but I alwaysfelt that maybetheyweredoingit wrong."Besides,
nobodyelsewasexactlyfrothingat the mouth to stepforward'
okrv he told the group.He waswilling to be the Little RedHen and
get everythingdone,at leastuntil they couldconvincesomeoneelseto
Itep in. But iirstonething:in the meantimehe wantedMurrayout front
doingthe talking.
"frh.n you'relookingfor funds,"saysCowan,"peoplewantto hearhow
you'regoingto solvethe energycrisistomorrow.But wewerestartingmuch
*or. i'oaJrtly. I thought ii wasgoing to be yearsbeforewe produced
anythingterriblyuseful,otherthan a newwayof lookingat the world. So
'Here'sprofessorsoandso,who'sgivingup his preoccu-
whatyou sayis,
pationwith quarks in orderto work on somethinga liftle more relatedto
yourdaily aren'tquitecertainof whatyou'retalkingabout.
"on".r.rr.'They
But theylisten."
The iellowsagreed.Cowanwould be the institute'spresidentand man
on the spot.Gell-Mannwouldbe the chairmanof the board'

Ceorge

Reticenceaside, Cowan was actually well suited to be the man on the

spot.He had contactseverywhere.Of course,he could hardly haveavoided
li. New Mexico has such a comparatively tiny population that any Los
Alamos administrator quickly gets to know all the powers that be. But it
helps if that Los Alamos administrator also happensto have made himself
a millionaire severaltimes over.
cowan usually won't bring up that subiecthimself, and seemsalmost
embarrassed when asked."Anybody who tells me there'sanythingdifficult
about that-well, I iust don't agree."
It happenedback in the early 1960s,he explains."Los Alamos was the
ideal exampleof a kind of socialisteconomy:There wasno privateproper$.
People were assignedhousing according to their rank and importance.
78 COMPLEXITY

funior people were assignedto what were essentiallyshacks.They looked

like army barracks."
"Well, I wastrying to hire people-in thosedaysit wasusually a man-
but it wasn't easy.There would be immediate friction with his wife about
having to live in those shacks.So we persuadedthe government to make
real estateavailable. But the banls wouldn't lend to a government instal-
lation. So we said to ourselves,'We'll start our own savingsand loan.' I
remember telling my wife that we were probably going to lose our invest-
ment. She said, 'Okay.'But we didn't! The savingsand loan turned out to
be very profitable, so we decidedto start a bank, the Los Alamos National
Bank. It was an immediatesuccess."
"All it took," he says, "was a good lawyer and a couple of friendly
senators. "
Cowan had already foreseenthe need for seed money for the institute
back in the summer of 1983, and he'd gone for help to an old friend of
his: Art Spiegel of the SpiegelCatalog fortune. He and Spiegel had been
membersof the group that founded the SantaFe Opera, and he knew that
Spiegeland his wife were principal fund-raisersfor the New Mexico Sym-
phony Orchestra,Spiegel,for his part, had no clear ideawhat Cowan was
talking about with this institute thing. But it soundedto him like a great
idea, if only as a much-neededresponseto the growing fapaneseleadership
in high technology. So he beganhelping Cowan to canvasthe assortedrich
people in Santa Fe, of whom there are many.
By the spring of 1984, Spiegelhad been able to raisea bit ofcash from
Mountain Bell and one of the more prosperouslocal savingsand loans
(which has since gone broke). It wasn't a lot. But, then, Cowan didn't
consider fund-raising his top priori$ yet, either. He felt it was more im-
portantto lay somegroundwork.Around Easter1984,for example,Cowan
laid out $300 of his own moneyfor a lunch for communig leadersin Santa
*We
Fe. felt it was politically desirableto let them know what we had in
mind and to invite their interestand support. We didn't pursue it very
strongly. We just didn't want them to read about it in the papersthat a
bunch of eggheadsfrom Los Alamos were suddenlyappearingin Santa Fe
to do somethingthey didn't know about."
This lunch didn't bring in any money, either. But it wasgood practice.
Gell-Mann came and gave a speech.The crowd loved it: a Nobel Prize
winner!
Meanwhile, there wasthe matter of incorporation:if you're going to start
askingpeople for money, you really ought to have somethingbesidesyour
own personal checking account to put it into. So Cowan and Nick Me-
 79
The Revolt of the Old Turks
tropoliswentto |ackCampbell,an old ftiendwho hadoncebeengovernor
ofihe stateand who *", no* headofa veryprosperous law firm in Santa
f.. C"*pb"ll wasenthusiastic. He'd wanted to do something like this all
the universities in New Mexico were
the time he wasgovernor'he said;
problems. Campbell agreed to provide
too damn isolateJfromreal-world
his firm'sservices probonoto drawup the incorporation papers and-b{1ws'
He also advisedco*rn on how to persuade the IRS that the fledgling
i"rtit"" reallydid deservea not-for-profit status'(The IRS is notoriously
,l.fii."f about such things;Cowan had to fy to Dallasand make the
argumentin Person.)
in lvt"y 19b4,the SantaFe Institutewasincorporated. lt didn't havea
locationor a staff It had essentially zeromoney. In fact, it wasn'tmuch
more than a post office box and a telephone number that rang in'Art
Spiegel'soffi". i., Albuquerque.And it didn't even have the right name:
"S"rrt Felnstitute" *as alreadycopyrightedby a therapy service' soCowan
(The Rio
andthe fellowshadhadto settlefor "The Rio GrandeInstitute."
Grandeflowsa few mileswestof town.) But it existed'
However,therewasstill that naggingquestionof content.Gell-Mann's
visionaryrhetoricwasall verywell. Gell-Mannwasa verysmartguy. But
nobody*as goingto plunk downseveralhundredmillion dollarsuntil they
he"rdprecisiy w-hatihe institutewasgoingto do-or, for thatmatter,until
theyhad soml evidencethat it wasgoingto work. "Herb, how do we get
this thing started?"cowan askedLos Alamosfellow Herb Andersonthat
spring. fell, saidAnderson,his favoriteformulawasto bring a bunch of
-good
u.ry peopletogetherin a workshop,and haveeachone talk about
whateverwasnearestand dearestto his heart.You could getthe coverage
of all the differentdisciplinesby the kind of peopleyou invited,he said.
And if therewerereallya convergence betweenthe disciPlines, you would
seeit startto emergefrom the debate'
'Fine, you startdevelopingthat,' " saysCowan,and that's
"So I said,
what he did." Shortlythereafter,Pinesvolunteeredto put thervorkshops
together-he hadbeenthinkingalongmuchthesamelines-and Anderson
happilyturnedit overto him'

At Princeton,PhilipAnderson gotthenotefromPineson |une29, 1984:

Would he like to attenda workshopthat fall on "EmergingSyntheses"
in
science?
80 COMPLEXITY
Hmm. Maybe. Andersonwas skeptical,to saythe least.He'd heard
rumors about this outfit. Gell-Mann had been talking up the institute
everywhere hewent,andsofarasAndersoncouldtell, theplacewasshaping
up asa cushyretirementhomefor agingNobellaureates from Caltech-
completewith megabuck endowments andlotsof scientificglitzandglamor.
Well, Andersoncould matchcredentials with Murray Gell-Mann any
day of the week,thank you. He'd won his own Nobel Prizein 1977for
his work in condensed-matter physics,and for thirg yearshe'd been as
much a cenhoid in that field as Gell-Mann had in his. But personaliy,
Andersondespised glitz and glamor.He didn't evenlike wo*ing on fash-
ionableproblems. Whenever hefeltothertheorists crowdingin on a subject
he wasworkingon, his instinctwasto moveto somethingelse.
He particularlyfound it insufferablethe way so manyyoung hotshots
wentaroundwearingtheirspecialty likea badgeof academic rank-whether
they'daccomplished anythingor not:"[ook at me, I'm a particlephysicist!
l,ook at me, I'm a cosmologistl"And he wasouhagedat the wayCongress
lavishedmoneyon shiny new telescopes and fantasticallyexpensive new
accelerators while smaller-scale projects-and,in Anderson's opinion,
morescientificallyproductiveprojects-werestarving.He hadalreadyspent
morethan his shareof time in front of congressional committees denounc-
ing the particlephysicists'recentlyannouncedplansfor a multibillion-
dollarSuperconducting Supercollider.
Besides,hethought,thisSantaFebunchsounded likea packof amateurs.
What did Murray Gell-Mannknow aboutputtingtogetheran interdisci-
plinary institute?He'd neverworkedon an interdisciplinary proiectin his
life. Pineshad at leastspentsometime workingwith astrophysicists and
trying to applysolid-statephysicsto the structureof neutronstars.Indeed,
he and Andersonwereworkingon that little problemtogether.But what
aboutthe restof them?Andersonhadspentmostof hisown research career
at Bell Labs,an interdisciolinaryenvironment if evertherewasone.And
h.ril.*6ffikfi"tr r""p.
is litteredwith the corpsesof fancynew institutesthat failedmiserably;if
theydidn't gettakenoverby crackpots, theygenerallyfustsankinto high-
mindedstagnation. In fact,Andersonhada close-upviewof asadexaqgl_e-
righttherein Princeton: theaugustInstitutefor Advanced Study,thehome
of Oppenheimer, Einstein,andvonffiings very
well, Iike math. But asan interdisciplinary institutehe consideredit an
abjectfailure, a collectionof very bright peoplewho eachdid their own
thing and barelytalkedto one another.Andersonhad seena lot of good
scientistsgo in thereand neverlive up to their promise.
The Revolt of the Old Turks 81

And yet, Andersonwasintriguedwith this santaFe Institutein spiteof

himself. Reversingthe tide of reductionism-now that washis kind of
language.He had personallybeen fightinga guerillawar againstreduc-
tionismfor decades.
What firstincitedhim to action,he recalls,wasreadinga lectureback
in 1965by the particlephysicistVictor Weisskopf. In it, weisskopfseemed
,tundamental"science-that is, particlephysicsand some
to imply that
partsoi cosmology-wassomehowdifferentfrom and betterthan more
applieddisciplineisuchascondensed-matter physics. Deeplyannoyed, and
,.rttting asonly an insulted condensed-matter physicist can be, Anderson
had im;ediatjy pr.par.d a lectureof his own in rebuthl. ln 1972he had
publishedit asan article in Science magazineentided"More Is Different-"
And he had beenpushingthe argument at everyopportunity sincethen.
To beginwith, he says, he is the 6rst to admit that there is a "philo-
sophicallycorrect"form ofsggllggiq namely, a belief thatthe universe
is governedby naturallaw. TEe uast maiori$ of working scientistsaccept
thit assertionwholeheartedly, says Anderson. Indeed, it's hard to imagine
howscience couldexistif theydidn't.To believein naturallawisto believe
that the universeis ultimatelycomprehensible-thatthe sameforcesthat
determinethe destinyof a galaxycan alsodeterminethe fall of an apple
hereon Earth;that the sameatomsthat refractthe light passingthrougha
diamondcan alsoform the shrffof a living cell; that the sameelectrons,
n€utrons,and protonsthat emerged from the Big Bangcannow giverise
to the humanbrain,mind, andsoul.To believein naturallawis to believe
in the unity of natureat the deepest possiblelevel.
Howevel saysAnderson,thisbeliefdoesnof imply thatthe fundamental\\
Iawsand the fundamentalparticlesare the only thingsworth studying- l\
and that everything elsecouldbe predicted if you only hada big enoughI J
computer. A lot of scientisb certainly do seem to think that way,he says.
Backin 1912,the physicistwho discovered the positron-the antimafter
versionof the electron-declared,"The restis chemistry!"More recendy,
Murray Cell-Mann himselfhadbeenknownto dismisscondensed-matter
theoryas"dirt physics."But thatwasprecisely the kind ofarrogance that
Andeisonfoundsoinfuriating.As he wrotein his 1972article,"The ability
to reduceeverythingto simplefundamentallawsdoesnot imply the abili$
to startfrom thoselawsand reconstruct the universe.In fact,the morethe
elementary particle physiciststell us about the natureof the fundamental
Iaws,the less relevance they seem to have to the veryrealproblemsof the
restof science,much lesssociety."
This everything-else-is-chemistry nonsense breaksaparton the twin
 82 COMPLEXITY
shoalsof scaleandcomplexity,he explains. Thkewater,for example.There's
nothingverycomplicated abouta watermolecule:it's iustonebig oxygen
atom with two little hydrogenatomsstuckto it like Mickey Mouseears.
Ifs behavioris governed by well-understoodequations of atomicphysics.
But now put a few zillion of thosemoleculestogetherin the samepot.
$uddenlyyou'vegot a substance that shimmersand gurglesand sloshes.
l'hosezillionsof molecules havecollectivelyacquireda property,
liquidity,
thatnoneofthem possesses alone.In fact,unlessyouknowpreciselywhere
afldhow to lookfor it, there'snothingin thosewell-understood equations
of atomicphysicsthatevenhintsat sucha property.The liquidity is "emer-
gent."
In muchthesameway,saysAnderson,emergent properties
oftenproduce
emergent behaviors. Cool those liquid water molecules down a bit, for
example,and at J2'F they will suddenlyquit tumbling over one another
at random. Insteadthey will undergoa :p@rnsition," locking them-
selvesinto the orderly crystalline array known as iiffirf you were to go
the other direction and heat the liquid, thosesametumbling water mol-
ecules will suddenly fy apart and undergo a phase transition into water
vapor. Neither phasetransition would have any meaning for one molecule
alone.
And so it goes, saysAnderson. Weather is an emergent proper$: take
your water vaporout over the Gulf of Mexico and let it interactwith sunlight
and wind, and it can organizeitself into an emergentshuctureknown as
a hurricane. Life is an emergentproperty,the product of DNA molecules
and protein moleculesand myriad other kinds of molecules,all obeying
the laws of chemistry. The mind is an emergentproperg, the product of
severalbillion neurons obeying the biological laws of the living cell. In
fact, as Anderson pointed out in the 1972 paper, you can think of the
ffis*forming a kind of "At eachl.u.ljl
.h:ierarcby: ggp&dlL
-entirely
new propertiesappear.[And] at each stage,entirely new laws,
concepts,and generalizations are necessary,requiringinspirationand crea-
tivity to just as great a degreeas in the previous one. Psychologyis not
appliedbiology, nor is biology appliedchemistry."
No one reading that 1972 article or talking to its author could have any
doubt where his sympathieslay. To Anderson, emergencein all iU infinite
varietywas the most compelling mysteryin science.Next to that, quarks
iust seemed so-boring. That's why he had gone into condensed-matter
physicsin the first place: It was a wonderlandof emergentphenomena.
(The Nobel Prize he receivedin 1977 honored his theoretical explanation
The Revottof the Old Turks 83

of a subtlephasetransitionin which certainmetalswentfrom beingcon-

ductorsof electricityto beinginsulators.) And that'salsowhy condensed-
matterphysicswasneverquiteenoughto containhim' By the time Pines's
invitationreachedhim in fune 1984,Andersonwasbusilyapplyingtech-
niqueshe'd developedin physicsto understanding the three-dimensional
structureof protein molecules,and to analyzing behaviorof neural
the
networla-airaysof simpleprocessors that try to do computationin much
thesamewaythatnetworkolneuronsdo in thebrain' He hadevengrappled
with oneof the ultimatemysteries, suggesting a modelof how the firstlife-
forms on Earth might have arisenfrom simple chemicalcompounds
throughcollectiveself-organization.
Solf this SantaFe outfit *"r for real,thoughtAnderson,he wasready
to listen.If it wasfor real.
A few weeksafter gettingPines'sinvitation,he had his chanceto find
out. That summer,asit hafpens,he wasservingaschairmanof the board
for the AspenCenterfor Physics, a summerretreatfor theoreticalphysicists
located,"ror, , broad meadowfrom the Aspenlnstitute.Andersonhad
alreadyplannedto meetwith Pinesthereto discuss somecalculations about
the innardsof neutron stars.So at their first encounter in Pines's officehe
poinL 'Okay, Dave,is this thing faky o1is it for-real?"He
got right to the
f.,.*-.*a"tly whatPineswasgoingto say-"lt's for real"-but he wanted
to hearhow the answersounded.
Pinesdid his bestto makeit soundgood.He badlywantedAndersonin
on this. For all of Anderson's skepticism, he had a breadthof interestand
insightthat was at leastthe equal of Gell-Mann's. He wouldserveas a
muJh-needed counterbalance and, not incidentalln his NobelPrizewould
givethe institutean additional quantum leap in credibility'
So PinesassuredAndersonthat, yes,the institutereallywasgoing to
look at the intersectionsbetweendisciplines,not iust look at a few fash-
ionabletopics.And no, it wasnot goingto be a front for Murray Gell-
Mann. Nor, fo, that matter,was'it goingto be iust an appendage of Los
Alamos-which Anderson, Pines knew, would have nothing to do with.
Cowan was playinga lead role. Pines was playing a lead role' And if
Andersonwould comeaboard,he, Pines, would see to it that he played a
lead role. In fact-did Andersonhave any speakers to suggest for these
workshops?
Thatdld it. As soonasAnderson heardhimselfmullingovernamesand
topics,he knewhe washooked.The opportuni$to makehis presence felt
wasiusttoo tempting."lt wasthe sense thatI could have some infuence
 U COMPLEXITY
on the institute," he says."If it was really going to happen, I was eager
to
be there trying to contribute to the way things went, to avoid the miltakes
of the past, to have it happen more or lessright."
The discussionsabout the worlshopsand thi institute continued through-
out the summer, since Gell-Mann and carruthers were also in Aspen. And
as soon as Anderson got back to princeton at the end of the summer he
iotted down three or four pagesof suggestionsfor how to organize the
institute so as to avoid the pitfalls. (The main point Don't have separate
departments!)
And he made reservationsto travel to Santa Fe in the autumn.

"What am / doing here?,,

Puttingtogether
theworkshops
provedto bea trickybusiness.
Actually,
it hadn't beentoo difficult to find funding. cell-Mann had usedhis contacts
to wangle$25,000from the CarnegieFoundation.IBM kickedin $t0.000.
And cowan had gottenanother$25,000from the MacArthur Foundation.
(Gell-Mann, who was on the MacArthur board, had felt it was improper
to ask himself.)
Much tougher, however,wasthe issueof whom to invite. "The question
was,"sayscowan, ''could you getpeopleto talk to one anotherand mutually
stimulate one another about what washappeningat the boundariesbehveen
disciplines?And could we developa community that would actually nurture
this kind of thing?" It wasall too easyto imaginesuch a meetingdissolving
into mutual incomprehension, with everyone talking right past each
other-if they didn't walk out first in utter boredom. The only way to guard
againstthat was to invite peoplewith the right quality of mind.
"we didn't want the reclusivetypes,the oneswho shut themselvesoff
to write their book in some office," sayscowan. "we neededcommuni-
cation,we neededexcitement,we neededmutual intellectualstimulation."
In particular, he says,they neededpeople who had demonstratedreal
expertiseand creativity in an establisheddiscipline, but who were also open
to new ideas.That turned out to be a depressinglyrare combination, even
among (or especially among) the most prestigiousscientisb. Gell-Mann
suggesteda number of people who might do. "He has great taste about
intellectual strengths,"sayscowan. "And he knows everybody."Herb An-
derson suggestedsome others, as did Pines and phil Anderson. "phil has
a hell of a lot of common sense,"saysCowan. "He comesdown hard on
people he feels are fakers." Finding a mix that covereda broad enough
The Revoltof the old Turks 85

rangetook a summer'sworth of cross-country telephonecallsand brain-

stor-ming,saysCowan.But in the end, he feels that what they cameuP
.,an astonishinglist of goodpeople,"rangingfrom physicists to
with wa-s
archeologists to clinical psychologists.
Of corlrse,neitherCowannor anyoneelsehad the slightestideawhat
would happenwhen all thesepeoplegot together'
Actualiy, thereturnedout to be no wayto get all of them togetherin
'scheduling
anycase. confictsforcedPinesto splitthe workshops overtwo
separate weekends, october6-7 andNovember I0-lI, 1984. But Cowan
,"-.*b"r, that for a while, at least,eventhis huncated group had trouble
geftingstarted.Gell-Mannhad led off the October 6 with
session a forty-
Eu"-ril.,ute talk, "The Conceptof the Instituls"-s55sn1i3lly an enlarged
versionof his "emergingsyntheses" exhortationto the fellowsthe previous
Christmas.And then therehad followedan extended discussion of how to
turn that conceptinto a realscientificagenda anda real institute'"There
wasa little ,prriingaround,"saysCowan.It wasn'tentirelyobviousat first
how to find commonground'
For example,Universi$ of Chicagoneuroscientist fackCowan(no re-
lation) arguedthat it washigh time for molecularbiologistsand neuro-
scientistsio startpayingmor! attentionto theoreticalconsiderations, asa j

media"teobjectionsthat cells and biomoleculesare too much the product

of random evolution for theory to do much good. But ]ack Cowan had
heard that argumentbefore, and he stoodhis ground' As an example,he
pointedto the visual hallucinationscausedby peyoteor LSD. Thesecome
in a variety of patterns,including lattices,spirals,and funnels, he said'
And every one of them could be explained as linear wavesof electrical
activity marching acrossthe visual cortex of the brain. Might it be possible,
he suggested,that these wavescould be modeled with the kind of mathe-
matical field theories used by physicists?
Douglas Schwartz of the School for American Research,the Santa Fe-
based archeology center that was hosting the workshop, argued that ar-
cheologywas a subiectthat was especiallyripe for interactionswith other
disciplines. Researchers in the field wereconfrontedwith threefundamental
mysteries,he said.First,when did nonhumanprimatesfirstbeginto acquire
the essenceof humanity, including complexlanguageand culture?Did it
happen nearly a million yearsago, with the rise of Homo etectus?or only
a few tens of ihousandsof yearsago, as the Neanderthalsgaveway to fully
modern humans, Homo sapienssapiens? And either way, what causedthe
86 COMPLEXITY
change? Millionsof species havegottenalongjustfine withoutbrainsas
largeasours.Why wasour species different?
Second,saidSchwartz,whydid agricultureandfixedsettlements replace
nomadichunting and gathering? And third, whatforceshiggeredthe de-
velopmentof culturalcomplexity,includingspecialization of crafts,the
riseof elites,andtheemergence of powerbasedon factorssuchaseconomics
and religion?
Noneof thesemysteries hadanyrealansweryet,saidSchwartz,although
the archeological recordleft by the riseandfall of the Anasazicivilizations
in the AmericanSouthwest offereda wonderfulfield laboratoryfor inves-
tigatingthe lasttwo. The only hopeof findingsomeanswers,he felt, lay
in achievingmuch morecooperation betweenarcheologists and otherspe-
cialiststhan any of them have been usedto. Field researchers needed
increased input from physicists,chemists,geologists, andpaleontologists to
help to reconshuctthe upsanddownsof climatesandecosystems in those
ancienttimes.And morethan that, he said,theyneededinput from his-
torians,economists, sociologists,and anthropologists to help them under-
standwhat motivationsmayhavedriventhesepeoples.
That kind of talk certainlyresonated with RobertMcCormackAdams,
a Universityof Chicagoarcheologist who hadbeenswornin asthesecretary
of the SmithsonianInstitutiononly weeksbefore.For at leastthe past
decade,he said,he had beengettingmoreand moreimpatientwith an-
thropologists' gradualistapproachto the evolutionof'ililiiiTiffi"When
hewentoutdi cultures
undergoing oscillations
andgpheaval. Increasingly,he said,he was
beginningl"hroti" f-r.f-
@ fi;frGFT;;;l ng'.t'o'.Er.i.ni iit.,'
of cultural alternatives
at differentmomentsin response "i per-
to different
ceptionsof environment.
This self-organizationthemewasalsotakenup in a quitedifferentform
by Sfrep-l.Ablfraun-of the Institutefor AdvancedStudy,a hventy-five-
year-oldwunderkindfrom Englandwho wastrying to investigate the phe-
nomenonof complexityat the most fundamentallevel. Indeed,he was
alreadynegotiating with the Universityof Illinoisto founda Centerfor
ComplexSystems Research there.Wheneveryou lookat verycomplicated
systemsin physicsor biology,he said,you generally6nd that the basic
componenband the basiclawsare quite simple;the complexityarises
becauseyou havea greatmany of thesesimplecomponentsinteracting
simultaneously. The complexityis actuallyin theorganization-themyriad
possiblewaysthat the componenbof the systemcan interact.
The Revolt of the Old Turks 87

Recently,Wolframsaid,he andmanyothertheoristshadbegunto study

complexig using cellular automata,which are essentiallyprogramsfor
generatingpatternson a computerscreenaccordingto rulesspecifiedby
ih. progrr*rner. Cellular automatahavethe virtue of beingpreciselyde-
6n.i, * that they can be analyzedin detail. And yet they are still rich
enoughfor verysimplerulesto generate patternsof startlingdynamismand
comfrexity.The challengefor theorisb,he said,is to formulateuniversal
lawsthatiescribewhenandhow suchcomplexities emergein nature.And
while the answerwasn'tin yet, he remainedoptimistic'
In the meantime,he added,whateverelseyou do with this institute,
makesurethat everyresearcher in the placeis equippedwith a state-of-
the-artcomputer.Computersarethe essential tool in complexityresearch.
And so iiwent. How shouldyou organizethe institute?RobertWilson,
the foundingdirectorof the FermiNationalAccelerator Laboratoryoutside
of Chicago,said it was crucial for the instituteto keepclosetabswith
experimenters; too much theoryand you could end up gazinginto your
navel.LouisBranscomb, chiefscientistof IBM, stronglyendorsed the idea
of an institutewithout departmental walls, wherepeoplecould talk and
interactcreatively."It's importantto havepeoplewho stealideas!"he said.
By lunchtimeon the 6rstday,saysCowan,the participants werebegin-
ning to warm to their task.As luck would haveit, SantaFe wasshowing
off with one of its characteristically marvelousautumndays;peoplewent
through the buffet line and carried their platesoutsideto continuetalking
and aiguing out on the American School grounds.(The schoolis located
on an estateonceowned by an eccentric heiresswho hadburied220dogs
there.) "Theybegan to realize thatsomething was going on, andtheyopened
up," saysCowan.By the second day, Sunday, he adds, "it becamea very
excitingthing."And by the time the participantsheaded homeon Monday
morning,it wasclear to everyone thatthere reallycould be a coreof science
here.
Carruthers,for one,spenttheweekendin heaven."Herewasa collection
of manyof the mostcreativepeoplein the wholeworld' in many6elds,"
he says."And they turned out to have a lot to sayto each other' They
basicallyhad the sameworldview,in the sensethattheyall seemedto feel
that 'emergingsyntheses' reallymeanta reshucturingof science-that the
overlappingthemesof differentpartsof sciencewould be put togetherin
a new way.I can rememberdiscussions with fackCowan[Stanfordpop-
ulationbiologist],MarcFeldman, various mathematicians, all of uscoming
from a differentresearch culture, and discovering that our problemshad
enormous overlap,in bothtechnique and structure. Now, some of thismay
88 COMPLEXITY
be that the humanmind only worksin certainways.[But]thoseworkshops
tumedall of usinto huebelievers. I won'tquitecall it a religiousexperience,
but it wascloseenough."
For Ed Knapp,a LosAlamitewho hadgoneoff to Washingtonto serve
a term as directorof the NationalScienceFoundation,and who had sat
in on someof the earlyinstifutediscussions, it wasoverwhelmingto find
himselfamongso many accomplished peopleat once. At one point he
cameup to Carruthersand said,"Hey, whatam I doinghere?"
And the Smithsonian's BobAdamshadmuchthe samereaction."It was
a wonderfularrayof papers,"he says."When thingsare in the air, and
you are beginningto makelittle linkagesanway,and then you go out to
something likethatsymposium in SantaFe,andsuddenly therearegropings
in neurobiology andcosmology andecosystem theoryandwhatever-fesus,
you wantto be in on it."
The secondworkshop,held a month later with a whole new crew of
participants,turnedout to be iust aseffectiveasthe first. EvenAnderson
wasimpressed. "You couldn'thelpbut be enthusiastic," he says.The event
removedany lastvestiges of doubtin his mind: this outfit reallywasgoing
to be differentfrom all the other advancedresearchinstituteshe knew
about."It wasgoingto be much moreinterdisciplinary," he says."They
reallyweregoing to focuson the spacesbetweenthe fields." Moreover,
therereallywassomethingthere."lt wasn'tclearthatall thesethingswould
be on the agenda,but it wasclearthat manyof them could be."
More than that, however,the workshops gavesomemuchneededclarity
to whatCowan'svisionof a unifiedsciencemight actuallybe about.As
Gell-Mannrecalls,"We hadfantasticamountsof similarities. Therewere
a huge number of common features in thethingsthat were presented
among
variousfields.You had to look carefully,but onceyou got pastthe jargon
of all thesethings,it wasthere,"
In particular,the foundingworkshops madeit clearthat everytopic of
interesthadat its hearta systemcomposed of many,many'!gg[" These
agentsmight be moleculesor neuronsor speciesor consumersor even
corporations.But whatevertheir nature,the agentswereconstantlyorga-
nizingand reorganizing themselves into largerstructures throughthe clash
of mutual accommodation and mutual rivalry.Thus, moleculeswould
form cells,neuronswould form brains,species wouldform ecosystems,
consumers and corporationq wouldform economies, and so on. At each
level,new emergent structures wouldform and engage in new emergent
behaviors. Complexity, in otherwords,wasreallya science of emergence.
The Revoltof the Old Turks 89
And the challenge that Cowan had been trying to articulate was to find
the fundamentallaws of emergence.
By no coincidence, it was also about this time that the new, unified
science acquired a name: the sciencesof P-Cgtg. "lt seemeda much
better canopy for everything we were dolngE?i--any other phrasewe were
'emerging
using, including syntheses,'" saysCowan. "It embracedevery-
thing I was interestedin, and probably everything that anyone else at the
institute was interestedin."
So after the hvo founding workshops,Cowan and companywere on their
way. All they needed was for that fabled donor to step forward and give
them the money.

lohn

Fifteenmonthslater,theywerestill waiting.l,ookingbackon thatperiod,

Cowanmaintainsthat he wasstill confidentthat the moneywould follow
the excitement. "It wasan incubationperiod,"he says."l hadthe feeling
thingsweremovingratherrapidly."But othersin the groupwerebiting
their fingernailsto the elbow."We had a growingsenseof urgency,"says
Pines."If we didn't keepa certainmomentum,thenwe weregoingto lose
support. "
Granted,the time hadn'tbeentotallyunproductive. In manyways,in
fact, thosefifteenmonthshadgoneratherwell. Cowan and his colleagues
had come up with enough money to run a few worlshops.They had
hammeredout an infinite number of organizational details'They had
persuaded Mike Simmons,PeteCarruthers' formerright-handman at the
Los Alamostheorydivision,to comein part-timeasa vice-president and
therebytakea lot of the administrative problemsoff Cowan'sshoulders.
And they had evengottenbackthe nametheywanted.After morethan a
yearof existenceas "The Rio GrandeInstitute,"which the fellowshad
acceptedonly out of necessity, they had beenapproached by a local firm
wantingthat name.So they said,"Sure-if you can get us the namewe
want." The firm hadaccordinglyboughtout the name"SantaFeInstitute"
from the moribundtherapybusiness thatownedit, andthetradewasmade.
Perhapsmostimportant,however,Cowanand his grouphad finessed a
potentiallyexplosivesituationwith Gell-Mann. Gell-Manncontinuedto
bea superblyinspirational Moreover,he haddrawnon his contacts
speaker.
to recruita numberof newmembersfor theinstituteboard."I alwaysexpect
90 COMPLEXITY
'No. I'm busy,'" saysGell-Mann."But theyalmostalways
that they'll say,
'Oh, my
say, God,yes!WhencanI come?I lovethisidea.I'vebeenwaiting
for this all my life!' "
And yetaschairmanof theboard-the fund-raiser in chief*Gell-Mann
wassimply not gettinganythingdone. The politestwayto sayit wasthat
he wasnot a natural-bomadministrator. Cowanwasexasperated: "Murray
wasalwayssomewhere else."Gell-Mannhadhis fingersin a dozenpies,
not all of them locatedin SantaFe. Paperwork waspiling up on his desk,
hewasnot returningphonecalls,andpeopleweregoingcrazy.The situation
wasonly resolvedto everyone's satisfactionwith an executivemeetingat
Pines'housein Aspenin fuly 1985;Cell-Mannagreedto stepdown as
headofthe boardoftrustees andinstead be headofa newScienceBoard,
wherehe could happilyplot the intellectualagendaof the institute.The
new chairmanof the boardof husteeswouldbe Ed Knapp,who had iust
finishedhis term at the NationalScienceFoundation.
But for all of that,the hoped-for $100-million angelhadnot yet mater-
ialized,despiteany numberof feelersput out by Cowanand others.The
maiorfoundations weren'texactlyeagerto pourmoneyintoa faky-sounding
idea like this when the esbablished research programsdesperately needed
their help to survivethe Reagan budget cuts."We were goingto solveall
of the outstanding problems of the modernworld,"says Carruthers."A lot
of peoplejustlaughed."
Meanwhile,the federalfunding agencieswerea giant questionmark.
Althoughtheycertainlyweren'tgoingto put up $100million or anything
like it, Eric Bloch, Knapp'ssuccessor at the NationalScienceFoundation,
seemed to besympathetic to theideaof givingtheinstitutesomedesperately
neededseedmoneyat, say,the $l million level.SowasCowan'sold friend
Alvin Tiivelpiece,who wasnow headof researchat the Departmentof
Energy.Blochhadevensuggested thepossibilityof ioint fundingfrom both
agencies.The problemwasthat nothing wasgoingto happenuntil the
institutecouldput togethera formalproposal andgetit approved-aProcess
that could easilytakea coupleof years,considering that everyonewasstill
workingparttime. And until then,Cowanbarelyhadoperatingfunds.The
SantaFe Instituteseemedto be floundering.

So it wasthat the boardof trusteesmeetingon March 9, 1986,was

largelydevotedto brainstormingfor namesof peoplewho mightgivethem
money.Lots of ideaswerebattedaround.In fact, it wasonly towardthe
The Revoltof the Old Turks 91
end that Bob Adams, sitting at the far end of the conferencetable near the
back of the room, rather diffidently put up his hand.
By the way, he said, he'd recently been up in New York at a meeting of
the boardof the RussellSageFoundation,which givesawaya lot of money
for social science-typeresearch.And while he was there he'd talked to a
friend of his, fohn Reed,the new chief executiveofficer of CiticorP. Now,
Reed was a pretty interestingguy, said Adams. He had iust turned forty-
seven, which made him one of the youngestCEOs in the country. He'd
grown up in Argentina and Brazil, where his father had worked as an
ixecutive for Armour and Company. He had a bachelor'sdegreein liberal
arts from Washington and fefferson Universi$, another bachelor'sdegree
in metallurgy from MIT, and a master'sdegreein businessfrom the Sloan
School at MIT. He wasvery knowledgeableabout science,and he genuinely
seemedto enjoy kicking around ideaswith the academictypesat the Russell
Sageboard.
Any*"y, said Adams, during one of the coffee breaks he'd told Reed
about the institute, as bestas he could explain it, and Reedhad been very
interested.He certainly didn't have $100 million to give away.But he was
wondering if the institute might help him understandthe world economy.
When it came to world financial markets, Reed had decided that profes-
sional economistswereoffwith the fairies. under Reed'spredecessor, walter
Wriston, Citicorp had iust taken a bath in the Third World debt crisis.The
bank had lost $l billion in profits in one year,and was still sitting on $13
billion of loans that might never be paid back. And not only had the in-
house economistsnot predicted it, their advice had made matters worse.
So Reed thought that a whole new approach to economics might be
necessary,said Adams, and he had askedhim to 6nd out if the Santa Fe
Institute might be interestedin taking a crack at the problem. Reed had
said he'd even be willing to come out to Santa Fe himself and talk about
it. What do you think?
When Adams was finished, saysPines, "I thought about the proposition
'That's great
for about six microsecondsand said, a idea!' " Cowan wasn't
far behind. "Get him out here," he said, "and I'll find some money to pay
for it." Gell-Mann and the others chimed in with their approval. So far as
any of them could tell, it was about twenty yearstoo early to be tackling
anything nearly as complex as economics-"lt almost set the boundary
condition for difficulty," saysCowan. "It irwolved human behavior'" But
what the hell. At the rate they were going, they were in no position to say
no to anybody. It was worth a shot.
92 COMPLEXITY

Yes, Dave, Phil Anderson told Pines over the phone. Yes, he was inter-
estedin economics. It was a bit of a hobby, in fact. And yes, this meeting
with Reedsoundedinteresting.But no, Dave, I can't come. I'm too busy.
But, Phil, said Pines, who knew that Anderson hated to travel, if you
work it right you can ride out on Reed'sprivate plane. You can bring your
wife, and you both can havethe fun of takinga privatejet. It's incredible.
Those iets go right to your destination. It cuts six hours off the door-to-
door time. It'll give you a chance to get to know fohn, and discussthe
programwith him. You can . . .
All right, said Anderson.All right. I'll come.
And so late in the afternoonof Wednesday,August 6, 1986, Anderson
and his wife, foyce, climbed aboardthe Citicorp Gulfstreamiet and rocketed
off toward Santa Fe. Well, Anderson had to admit it was fast. It was also
freezing.The Citicorp jet flew at about i0,000 feet,well abovecommercial
airspace,and its heatersdidn't seemquite able to handle the chill. foyce
Anderson huddled in back under a blanket while Anderson himself sat up
front talking economicswith Reedand threeof his assistants: Byron Knief,
Eugenia Singer, and Victor Menezes.Also along was Carl Kaysenof MIT
an economistwho had once been head of the Instifute for AdvancedStudy,
and who now served on the boards of both the Russell SageFoundation
and the SantaFe Institute.
Anderson found Reed to be pretty much as Adams had describedhim:
smart, direct, and articulate.Around New York he had a fiercereputation
for firing people en masse.But in person he struck Anderson as easygoing
and unpretentious-the kind of CEO who likesto chat with one leg draped
over an arm of his seat.He clearlywasn'tintimidatedby Nobel laureates.
In fact, he said he'd been looking forward to this meeting, for exactly the
same reasonhe enjoyed the meetingsof the RussellSageboard and all the
other academicboardshe was on. "That sort of thing is fun for me," he
says."It gives me an opportunity to talk to an academic-intellectualgroup
of folks who tend to look at the world quite differently from my dayto-day
iob. I think I benefit from seeing it both ways." In this particular case,
Reed recalls having had a great time thinking about how to explain his
admittedly biasedview of the world economyto a set of scholars."lt was
obviously different from the way one would explain it to a bunch of
bankers. "
For Anderson, the trip to SantaFe provedto be a marvelousbull session
The Revoltof the Old Turks 93

on physics,economics,and the vagariesof global capital ffows' He also

in particular,was not about to be left
fourrd'th"t o.re of Reed'sassistants,
out of the conversation.Shivering under severallayersof sweaters,Eugenia
Singer was coming along to talk about a survey she'd done for Reed on
econometric models: the big computer simulations of the world economy
usedby the FederalReserveBank, the Bank offapan, and others.Anderson
liked her immediately.
Singer, as it happens,was not shivering iust becauseof the cabin tem-
she laughs.
-I-Iere "l was terrified of what fohn had gotten me intol"
peratuie.
she was with nothing more than a master'sdegreein mathematical
statistics,and essentiallyno recentexperienceworking in that area. "And
based on that, fohn had me running out there and talking to all these
Nobel Prize-winning physicistslI didn't feel up to that technical level, to
put it mildly.
"It was the only time I'd ever tried to say no to an assignmenttrom
'Ah, Eugenia,
fohn," shesays."But he'd said,in a very casual,offhandway,
you'll do fine. You know more about it than they do.' " So she'd come'
And Reed was right.

The encounter, iointly chaired by Adams and Cowan, startedat 8 e'u'

the next morning at Rancho Encantado, a kind of dude ranch about ten
miles north of SantaFe. Only a dozenpeoplewere present-among them
Cowan'sold friend ferry Geist, chairman of the Public ServiceCompany
of New Mexico and the man who had put up the money for this meeting'
The event was not really intended to be a scientific interchange. It was a
show-and-tell, with each side trying to persuadethe other side to do what
it very much wanted to do anyhow.
Reed, armed with a fisdul of overheadtransparencies,went first. Basi-
cally, he said, his problem was that he was up to his eyeballsin a world
economic systemthat defied economic analysis.The existing neoclassical
theory and the computer models basedon it simply did not give him the
kind of information he neededto make real-time decisionsin the face of
risk and uncertainty. Some of thesecomputer modelswere incredibly elab-
orate. One, which Singer would be talking about in more detail later,
covered the whole world in 4500 equationsand 6000 variables.And yet
none of the models really dealt with social and political factors, which were
often the most important variablesof all. Most of them assumedthat the
modelers would put in interest rates, currency exchangerates, and other
 94 COMPLEXITY
such variablesby hand*even though theseare preciselythe quantitiesthat
a banker wants to predict. And virtually all of them tended to assumethat
the world was never very far from staticeconomicequilibrium, when in
fact the world was constantlybeing shakenby economic shocksand up-
heavals.In short, the big econometricmodels often left Reed and his
colleagueswith little more to go on than gut instinct-with results that
might be imagined.
A casein point was the most recentworld economic upheaval,which
was symbolizedby Presidentcarter's 1979appointmentof paul Volker to
head the FederalReserveBoard. The story ofthat upheavalactually began
in the 1940s,explainedReed, at a time when governmentsaround the
world found themselves strugglingto copewith the economicconsequences
of two World Wars and a Great Depressionin between.Their efforts,which
culminatedin the Brettonwoods agreements of 1944,led to a widespread
recognition that the world economy had becomefar more interconnected
than ever before. Under the new regime, nations shifted away from iso-
lationismand protectionismasinstrumentsof nationalpolicy;instead,they
agreedto operatethrough international institutionssuch asthe world Bank,
the International Monetary Fund, and the General Agreementson Thriffs
and Trade. And it worked, said Reed. In financial terms, at least, the world
remained remarkably stablefor a quarter of a century.
But then came the 1970s.The oil shocksof 1977and 1979, the Nixon
administration'sdecisionto let the price of the dollar foat on the world
currencymarket,risingunemployment,rampant"stagflation"-the system
cobbled together at Bretton Woods began to unravel, said Reed. Money
beganflowing around the world at an ever-increasingrate.And Third World
countries that had once been starving for investment capital now began
borrowing heavily to build their own ggon6miss-helped along by U.S.
and European companies that were moving their production offshore to
minimize costs.
Following the advice of their in-houseeconomists,said Reed,Citicorp
and many other international banks had happily lent billions of dollars to
thesedevelopingcountries. No one had really believedit when Paul Volker
came to the Fed vowing to reign in inflation no matter what it took, even
if it meant raising interest ratesthrough the roof and causinga recession.
In fact, the banksand their economistshad failed to appreciatethe similar
words being voiced in ministerial officesall over the world. No democracy
could toleratethat kind of pain. Could it? And so, saidReed,Citicorp and
the other banks had continued loaning money to the developing nations
The Revolt of the Old Turks 95

throughoutthe early1980s-rightup until 1982,whenfirstMexico,and

then Argentina,Brazil,Venezuela, the Philippines, and manyothersre-
vealedthat the worldwide recession triggered by the anti-inflationfight
wouldmakeit impossible to meet their loan payments.
SincebecomingCEO in 1984,saidReed,he'd spentthe bulk of his
time cleaningup this mess.It had alreadycostCitibankseveralbillion
dollars-so far-and hadcausedworldwidebankinglosses of roughly$300
billion.
Sowhatkind of alternative washe lookingfor?Well, Reeddidn'texpect
that any new economictheorywould be ableto predictthe appointment
ofa specificpersonsuchasPaulVolker.But a theorythatwasbetterattuned
to socialand politicalrealitiesmight havepredictedthe appointment of
someone /ifteVolker-who, afterall, wasiustdoingthepoliticallynecessary
job of inflationcontrolsuperblywell.
More important,he said,a bettertheorymight havehelpedthe banks
appreciate the significance of Volker'sactionsas they werehappening.
"Anything we could do that would enhanceour understanding and tease
out a betterappreciationfor the dynamicsof the economyin which we
livewouldbewellworthhaving,"hesaid.And fromwhathe'dheardabout
modernphysicsand chaostheory,the physicists hadsomeideasthat might
apply.Could the SantaFe Institutehelp?
The SantaFe contingentwasfascinated; to mostof them this wasbrand
new stuff. They were equallyintriguedwith EugeniaSinger'sdetailed
reviewof globalcomputermodels.TheseincludedProiectLink (the one
with 6000variables), the FederalReserve Multi-CountryModel,theWorld
Bank Global Development Model, the Whalley Tiade Model, and the
Global Optimization Model. None of them couldquitefill the bill, she
concluded, especiallywhen it came to dealing with changeand upheaval.
So, again,could Santa Fe help?
Well, maybe.A lot of the afternoon wasgivenoverto theinstitute's side
of the showandtell. Andersontalkedabout mathematical models of emer-
gent,collectivebehavior,Otherstalkedabouttheuseof advanced computer
graphics to convertmountainsof datainto vivid and cor,nprehensible pat-
terns;the useof artificialintelligencetechniques to modelagentsthatcould
adapt,evolve,and learn with experience; and the possibleuseof chaos
theoryto analyzeandpredictthe gyrationsof stock-market prices,weather
records, andothersuchrandom-seeming phenomena' in theend,not
And
surprisingly,the consensus on both sidesof the iable wasthat, yes' an
economicsprogramwasworth a try. As Andersonrecalls,"We all said
96 COMPLEX'TY

there'sa possibleintellectualagendahere.What wasmissingin modern

equilibriumeconomicsthat permitsthe kind of upheavals that fohn was
talkingabout?"
However,the SantaFe contingentalsoplayedit very cagey.As dearly
as Cowan and companywantedto seesomeof Citicorp'smoney,they
alsowantedto makeit very clearto Reedthat they couldn'tpromisehim
a miracle.Yes,theyhad someideasthat mightbe helpful.But this was
a high-riskenterprise that might not leadanywhere. The lastthing the
fedgling instituteneededwasa lot of infated expectations and hype;it
wouldbe suicideif theyseemed to be promisingsomething theycouldn't
deliver.
Reedsaidhe understood completely. "My viewwasthat I didn't think
we were going to get something hard and concrete,"he recalls.He iust
wanted some new ideas. So he promised not to put a time limit on the
product,or evento definea deliverable product.IfSantaFeiustgotstarted
on the iob and madevisibleprogress from yearto year,that would be
enough.
"That fueledmy enthusiasm for doingthis thing," saysAnderson.The
nextthing to do, they agreed,wasto hold anothermeeting-an extended
workshopwhereyou would havea significantnumberof economists and
physicalscientists sittingdown togetherto thrashout the issuesand seta
real agenda.If Reedwaspreparedto kick in a few thousanddollarsto
supportan effortalongthoselines,the SantaFeInstitutewouldbeprepared
to undertake it.
Sothe dealwasdone.The nextmorning.Reedhadthe EastCoastcrew
routedout of bed at 5 l.tr,t. and piled into limousinesfor the SantaFe
airport.He wantedto be backin New Yorkassoonaspossibleto put in a
full day'swork.

Ken

No, Dave,saidAnderson.I don't havetime to organizethis new eco-

nomicsworkshop.
But, Phil, saidPinesoverthe phone,you saida lot of interestingthings
whenwemetwith Reed.And thisnewworlshopisgoingto bean incredible
and we'll geta top-notch
opportunity.You'll invite the physicalscientists,
economistto invite the economics half.
No.
The Revolt of the Old Turks 97

Look, saidPines,I knowthis is one morething you'retakingon, but I

think you'll find it really interesting.Think aboutit. Talk to foyceabout
it. And if you sayyes,I'll help.Youwon'tbe out thereall alone.
All right, Andersonsighed.All right, Dave.I'll do it.
Having saidyes,Andersonwasat a losshow to proceed.He'd never
organized anythingquitelike this. Who had?Well, the firstthing to do,
obviously,wasto find someoneto headup the economics half of the
meeting.He did know at leastone economist:famesTobin of Yale,who
hadbeena fewyears aheadof him at UniversityHighSchoolin Champaign-
Urbana,Illinois-and who, as it happens,had won a Nobel Prizern
economics.fim, he saidoverthe phone,would you be interested
in such
a thing?
No, saidTobinafterhe'dheardAndersonexplainwhathe wanted.He
wasn'tthe right person.But KenArrow out at Stanfordmight be. In fact,
he wouldbe happyto call Arrowfor him if he'dlike.
Tobinhadapparently beenglowingin his description. WhenAnderson
called,Arrow provedto be veryinterested indeed."Ken andI chattedover
the phone quite a bit," saysAnderson."It turnedout that we had very
similarideas. " For all that Arrowwasoneof the foundersof establishment
economics, he had also,like Anderson,remaineda bit of an iconoclast
himself.He knewfull well whatthe drawbacks of the standardtheorywere.
In fact, he could articulatethem betterthan mostof the criticscould.
Occasionallyhe even publishedwhat he called his "dissident"papers,
callingfor new approaches. He'd urgedeconomisbto paymoreaftention
to real human psychology, for example,and mostrecentlyhe had gotten
intriguedwith thepossibility of usingthemathematics of nonlinearscience
and chaostheoryin economics.So if Andersonand the SantaFe crowd
thought they could shike out in new directions-"Well," he says,"it
soundedlike something that couldn'tbe uninteresting."
So Andersonand Arrow eachstarteddrawingup a list of names,using
prettymuchthe samecriteriaashadbeenusedfor thefoundingworlshops:
theywantedpeoplewith superbtechnicalbackgrounds coupledwith open
minds.
Arrow, in particular,felt that he neededpeoplewho had a very strong
commandof the orthodoxview of economics. He didn't mind people
criticizingthe standardmodel,but they'dbetterdamn well understand
what it wasthey werecriticizing.He thoughta bit and wrotedown a few
names.
And thenhe wantedto mix in a fewpeoplewith an empiricalbent.It
98 COMPLEXITY
wouldn'tbe healthyto havea solidphalanxof neoclassical theorists,
he
thought;you neededsomebody to remindyouof thingsthe standardtheory
had troublewith. Let'ssee-maybethe youngfellowhe'd heardgivethat
seminarlast year,the one who'd done all that work in demography and
who wasalwaysgoingon aboutincreasing returns.Nice sfuff.
Arthur,he wroteon the list. BrianArthur.
t f s { b * t &4 * * * * * se ** * * * s s * * $ * * & & *6 **ss**

In the autumn of 1986, evenasPhil Andersonand Ken Arrow weredrawing

up names for the economics meeting, George Cowan was making a deal
with the Archdioceseof SantaFe for a three-yearleaseon the Christo Rey
Convent: a one-story adobeshucture located iust pastthe pricey stretch of
art gallerieslining the crooked liftle lane known as Canyon Road.
It was about time. By that point Cowan and his colleagueshad begun
to hire a small cadre of staffersfor the institute, thanks to operating funds
that were beginning to trickle in from sourcessuch as the MacArthur
Foundation. And thosestaffersdesperatelyneededa spaceto call their own.
Furthermore, what with the economics meeting coming up and several
other workshopsbeing planned, the institute desperatelyneeded a little
office spacewhere it could keep ib academicvisitors happy with desksand
telephones. Cowan decided that the convent was small, but workable-
and came at a price that was too good to passup. So in February 1987,
the institute staff moved in. And within daysthey had filled the tiny space
to overflowing.

Chaos

The crowdingnevergotanybetter.When BrianArthur walkedthrough

the front entrancefor the first time on Monday August 24, 1987, he
practicallyfell overthe receptionist's
desk;it wascrammedinto a sort of
entryalcoveandleft the doorwith only aboutan inch of clearance
to swing
100 COMPLEXITY
open. The corridors were lined with boxesfull of books and papers.The
copier machine was tucked into a closet.One staffer's"office" was in a
hallway.The placewaschaos.And Arthur fell in love with it immediately.
"l couldn't have designeda place that was bettersuitedto my interests
and temperament,"he says.The chaotic convent somehowmanagedto
convey a senseof intellectual ferment in the midst of peace, shelter, and
serenity.When the institute'sdirector of programs,Ginger Richardson,
came out to welcomehim and show him around, she and Arthur walked
over creased}inoleum floorsand lookedat lovingly crafteddoors,polished
mantles, and intricately decoratedceilings. She showedhim the way to the
coffeepot in the Eisenhower-erakitchen; to get there you had to walk
through the mother superior'soffice, where Cowan was now ensconcedas
president.She showedhim the former chapel,which now servedas the
largeconferenceroom; on the far wall, where the altar had once been, a
blackboardfull of equationsand diagramswas washedby the ever-shifting
light from stained-glass windows. She showedhim the row of cramped
little officesfor the visiting scholars;formerlythe nuns'bedroomcubicles,
the rooms were now filled with cheap metal desksand typist chairs, and
had windows looking out onto a sun-drenchedpatio and the Sangrede
Cristo Mountains.
As a first-timevisitor to New Mexico, Arthur wasalreadyin a mood to
be enchanted.The mountains, the clear desertsunlight, the crystalline
desertvistashad all had the sameeffecton him as they had had on gen-
erationsofpaintersand photographers. But he felt at oncethat the convent
had a special magic. "The whole atmospherewas unbelievable," saysAr-
thur. "As I looked at the sort of books that were out on display, the kind
of articles that were lying around, the freedom of the atmosphere, the
informality-l couldn't believethat sucha placeexisted." He wasbeginning
to think that this economicsworkshopmight be very exciting indeed.
Accommodationsbeing what they were, academicvisitorsto the institute
were crammed into the officestwo or three at a time, .with their names
handwritten on piecesofpaper taped up by each door. One ofSce featured
a name that Arthur was very interestedto see:Stuart Kauffman of the
University of Pennsylvania.Arthur had briefly met Kauftnan two years
earlierat a conferencein Brussels,wherehe had beenimmenselyimpressed
by Kauffman'stalk on cells in a developingembryo. The ideawasthat the
cells sendout chemical messengers to triggerthe developmentof other cells
in the embryo in a self-consistentnetwork, thus producing a coherent
organism instead of iust a lump of protoplasm. It was a concept that re-
Secrets of the Old One 101

sonatedstronglywith Arthur's ideason the self-consistent, mutually sup-

portivewebsof interactionsin human societies,and he rememberscoming
back from that conferenceand telling his wife, Susan,"l've iust listened
to the besttalk I've ever heard!"
So as soon as he had gottensettledin his own office, Arthur wandered
down the hall to Kauffman'scubicle: Hello, he said, do you remember
that we met two yearsago . . . ?
Well, no, actually, Kauffman didn't. But come on in! Tanned, curly-
haired, and California casual,the forty-eight-year-oldKauffmanwasnoth-
ing if not affable. But then, so was Arthur; he was in a mood to love
everybodythat morning. The two men found themselveshitting it off
immediately."Stu is an immenselywarm person,"saysArthur, "someone
you feel you have to hug-and I don't go around hugging people. He's
just such a lovablecharacter."
They quickly fell to discussingeconomics,of course.With the meeting
coming up, the subjectwas very much on their mind-and, no, neither
of them had the slightest idea of what to expect. Arthur started to tell
Kauffman a bit about his work on increasingreturns."And that," he laughs,
"was a good excusefor Stuart to corner me and tell me about his latest
ideas. "
It alwayswas. Kauffman, as Arthur quickly learned,wasan immensely
creativeman, like a composerwhosemind wasendlesslyaboil with melody.
He emitted ideasnonstop. He also displayeda very high ratio of talking to
listening.Indeed,that seemedto be the way he thoughtthingsthrough:by
talkingabout his ideasout loud. And talkingaboutthem. And talkingabout
them.
It was a trait that was alreadywell known around the SantaFe lnstitute.
In the course of the previous year, Kauftnan had become the ubiquitous
man there. As the son and heir of a Romanian immigrant who had ac-
cumulateda minor fortune in real estateand insurance,he wasone of the
few scientistswho could affordto sethimself up in a secondhome in Santa
Fe and live there half the year. At any given institute planning session,
Kauffman was to be found spir'--i rg out a steadystream of suggestionsin
a mellifluous, confidentbaritone.At any givenseminarhe could be heard
thinking out loud during the questionand answerperiod about how to
conceptualizethe subjectat hand:"lmagine a networkof light bulbshooked
togetherat random, okay,and . . . " And at any given moment in between,
he could be heard trying out his latestideason anyonewho would listen;
rumor had it that he wasonce overheardexplaining someof the finer points
102 COMPLEXITY
of theoreticalbiology to the copier repair man. Or if visitorsweren't around,
he would soon be explaining the ideas to the nearestavailable colleague
about a hundred times in a row. At length, In detail.
It was enough to drive his best friends away screaming. Worse, it had
given Kauffman a widespreadreputation for possessingan oversizedego
combined with a nagginginsecuri$, even among colleagueswho say in
the next breath that they care about him deeply.It came acrossas a craving
to be told, "Yes, Stuart, that'sa greatidea. You're very smart." But whatever
truth there may havebeen to that perception,it wasalsotrue that Kauffman
couldn't help it. For nearly a quarter of a century he had been a man in
the grip of a vision-a vision that he found so powerful, so compelling,
so overwhelminglybeautiful that he simply could not hold it in.
The closestEnglish word for it is "order." But that word doesn't begin
to capture what he meant by it. To hear Kauffman talk about order was to
hear the languageof mathematics,Iogic, and sciencebeing usedto express
a kind of primal mysticism. For Kauffman, order was an answer to the
mystery of human existence, an explanation for how we could possibly
come to exist as living, thinking creaturesin a universe that seemsto be
governedby accident, chaos, and blind natural law. For Kauffman, order
told us how we could indeed be an accident of nature-and yet be very
much more than lusf an accident.
Yes, Kauffman alwayshastenedto add, Charles Darwin was absolutely
right human beingsand all other living things are undoubtedlythe hein
of four billion yearsof random mutation, random catastrophes,and random
strugglesfor survival;we are not here as the resultof divine intervention,
or even spacealiens. But, he would emphasize,neither was Darwinian
natural selectionthe whole story. Darwin didn't know about self-organi-
zation-matter's incessantattemptsto organize itself into ever more com-
plex shuctures, even in the face of the incessantforces of dissolution
describedby the second law of thermodynamics. Nor did Darwin know
that the forcesoforder and self-organizationapply to the creation ofliving
systemsjust as surely as they do to the formation of snowflakesor the
appearanceof convectioncells in a simmeringpot of soup. So the story of
life is, indeed, the story of accidentand happenstance,declaredKauffman.
But it is also the story of order: a kind of deep, inner creativity that is woven
into the very fabric of nature.
"l love it," he says."I do huly love it. My whole life hasbeenan unfolding
of this story."
Secretsof the Old One 103
Order

Walk down the corridors of almost any scientific institution in the world,
and you won't get far before a glance through an open office door reveals
a posterof Albert Einstein:Einsteinbundled in an overcoat,absentmind-
edly walking through the snowsof Princeton.Einsteingazingsoulfully at
the camera, a fountain pen clipped to the neck of his ratty sweater.Einstein
with a maniacalgrin, stickinghis tongue out at the world. The creatorof
relativig is very nearly the universalscientifichero, the very emblem of
profound thought and free creativespirit.
Back in the early 1950s,in Sacramento,California, Einstein was cer-
tainly a hero to a teenagedboy namedStuartKauffman."l admiredEinstein
enormously," he says."|rJe-3dmi.sd is the wrong word. loved. I loved
his image of theory as the free inventionof the human mind. And I loved
his idea that sciencewasa questfor the secretsof the Old One"-Einstein's
metaphorfor the creatorof the universe.Kauffman especiallyremembers
his first contact with Einstein's ideasin 1954, when he wasfifteen and read
a popular book on the originsof relativityby Einsteinand his collaborator
teopold Infeld. "I was so thrilled that I could understandit, or thought I
could understandit. Somehow,by being powerfully inventive,and free,
Einstein had been able to createa world in his head. I remember thinking
that it was absolutelybeautiful that anyonecould do that. And I remember
crying when he died [in 1955]. It was as if I'd lost an old friend."
Until reading that book, Kauffman had been a good, if not spectacular,
studentearningAs and Bs, Afterward,he wasinfamed by a passionfor-
well, not scienceexactly.He didn't feel that he had to follow in Albert's
footstepsthat closely. But he certainly felt that same fierce desire to peer
into the depths. "When you look at a cubist painting and seethe structure
hidden within it-that's what I wanted." The most immediate manifes-
tation, in fact, was not scientific at all. The teenagedKauftnan developed
an all-consumingdesireto be a playwright,to fathom the forcesof light
and dark within the human soul. His first efiort, a musical written in
collaborationwith his high-schoolEnglish teacher,Fred Todd, was "ab-
solutely atrocious." And yet the thrill of being taken seriously by a real
adult-Todd was tweng-four at the time-was for Kauffman a crucial step
in his intellectual awakening. "lf Fred and I could write a musical when I
wassixteen,evenif it wasn'ta very goodmusical,then why not the world?"
So the Stuart Kauffman who enteredDartrnouth as a freshman in 1957
was every inch a playwright.He even smokeda pipe, becausea friend of
his had told him that if you wanted to be a playwright, you had to smoke
 1M COMPLEXITY
a pipe.And, of course,he continuedto writeplays:threemoreof them
. thatyearwith hisfreshmanroommateandhigh-school buddx Mac Magary.
I nut Kauffman soon began to noticesomething about his plays:the char-
I acters pontificateda greatdeal. "They blabbered about the meaningof life
I and what it means to be a good person-talking about it instead of doing
lit." He began to realizethathe waslessinterested in the plays persethan
J in the ideashis characters were grapplingwith. "I wanted to find my way
to somethinghiddenand powerfulandwonderful-without beingableto
articulatewhat it was.And when I found out that my friend Dick Green
at Harvardwasgoingto maiorin philosophy,I wasterriblyupset.I wished
I couldbe a philosopher. But, of course,I hadto be a playwright. To give
it up meantgivingup an identityI had begunto assumefor myself."
It tookabouta weekofstruggle,he recalls,beforehe reacheda profound
revelation:"l didn't haveto be a playwright-l couldbe a philosopherlSo
I spentthe nextsix yearsstudyingphilosophywith enormouspasion." He
startedout in ethics,of course.He hadwantedto understand the problem
of goodand evil asa playwright,and sowhatelsecould he possiblydo as
a philosopher? Yet he quickly found himselfattractedelsewhere,toward
the philosophyof scienceand the philosophyof mind. "lt seemedto me
theyharboredthe depths,"he says.What is it aboutsciencethat allowsit
to discoverthe natureof the worldTAnd whatis it aboutthe mind that
allowsit to knowthe world?
With that passionto carryhim, Kauffmangraduated third in his class
at Dartmouthin 1961,and then went on to a Marshallscholarship at
Oxfordfrom 1961to 1963.As it happens, he didn'thavelby a verydirect
route."I had eightmonthsbeforeI had to be at Oxford,so I did the only
rationalthing possible:I boughta Volkswagen busand lived in the Alps,
skiing.I hadthe mostprestigious possibleaddress in St. Antonin Austria.
I parkedin the parkinglot of the PostHotel, and I usedtheir reshoomall
winterlong,"
But oncehe arrivedat Oxfordhefelt in hiselement.He canrecallhaving
beenin threeincrediblyexcitingintellectualenvironmenbin his life, and
Oxfordwasthe first. "lt wasthe first time in my life that I wassurrounded
by peoplewho weresmarterthan I was.The Americanstherewereiust
spectacular. The Rhodesscholars,the Marshallscholars.Someof them
are now fairly well known, David Souter,who wasin our groupat Mag-
dalene,is now on the SupremeCourt. And GeorgeF. Will and I usedto
go haveIndianmealsall the time to escape collegecooking."
Kauftnan'spassionfor understanding scienceand the mind led him to
Secretsof the Old One 105

studya curriculumthat OxfordcalledPhilosophy,Psychology, and Phys-

iology.It includednot iusttraditional philosophy, but a much more modern
emphasis on the neuralanatomy the of visualsystem and on more general
modelsof neuralwiringin the brain. In short, it dealt with what science
couldtell us abouthow the mind reallyworks. His tutor in psychology was
namedStuartSutherland, whoturnedout to be another influentialfigure'
Sutherlandhad a penchantfor siftingbehindhis deskand subiectinghis
students to a nonstopvolleyof brainteasers: "Kauftnan!Howcoulda visual
systempossiblymake a discriminationbetweenhvo pointsof light that
projectonto adiacentconesin the retina?"Kauffmandiscovered that he
lovedthis kind of challenge.He found that he had a facility for thinking
up modelson the spotto provide.at leasta plausibleanswer.("Well, the
eye isn't still, but it iiggles.So maybeyou spreadout the sensationover
severalrodsand cones,and . . .") Indeed,he admitsthat this imprompfu
model-making got to be a habit;in onewayor anotherhe'sbeendoingit
eversince.
And yet,he alsohasto acknowledge a certainironyin it all. It wasthis
veryability to makeup modelsthat led him to giveup philosophyin favor
of something much moredownto earth:medicalschool.
"l decidedin a way that provedI wasnevera greatphilosopher,"he
'l'll neverbe as smartas Kant. There'sno
laughs."The argumentwas:
point in beinga philosopher unlessyou areassmartasKant.Therefore,
I should go to medical school.'You'll noticeit's not a syllogism."
Seriously, he says, the fact was that he wasgettingimpatientwith phi-
"lt
losophy. wasn'tthat I didn't love philosophy. It's that I distrusteda
certainfacileness in it. Contemporary philosophers, or at leastthoseof the
1950sand 1960s,took themselves to be examiningconceptsand the im-
plicationsof concepts-not the factsof the world. So you could find out
if your argumentswerecogent,felicitous,coherent,and so on. But you
couldn't find out if you werertght. And in the end I felt dissatisfied with
that." He wantedto delveinto reali$, to know the secrets of the Old One.
"If I had to choose,I'd ratherbe Einsteinthan Wittgenstein."
Furthermore, he distrusted a certainfacileness in himself."l've always
hada conceptualfacility,"he says. its best, the deepest
"At it is partof me,
God'sgreatest gift to me. But at its worst,it's glib. Shallow'Because of
'I'll go to med schooland thosesonsof
that concern,I said to myself,
bitcheswon't let me be glib and swivel-hippedintellectually-because
I'll have to take care of people. They'll force me to learn a lot of
facts.'"
106 COMPLEXITY

They did. But somehow,they couldn'tdeflectKauftnan from his fas-

cinationwith the playof ideas.In fact, they neverreallygot a chanceto.
Sincehe had takenno premedcourseswhaboever,he arrangedto attend
Berkeleystartingin the fall of 1963for a yearof preparatory work before
goingon to medicalschoolacrossthe bayat the Universityof California,
SanFrancisco.So it wasin Berkeleythat he tookhis firstcoursein devel-
opmentalbiology.
He wasthunderstruck."Here wasthis absolutelystunningphenomen-
ology,"he says."Here you startwith a fertilizedegg,and the damn thing
unfolds,and it givesriseto an orderednewbornand adult." Somehow,
thatsingleeggcell manages to divideanddifferentiate into nervecellsand
musclecellsand liver cells-hundredsof differentkinds.And it doesso
with the mostastonishing precision.The shangethingisn'tthatbirth defects
happen,astragicastheyare;the strangething is that mostbabiesareborn
perfectand whole."This still standsasone of the mostbeautifulmysteries
in biology,"saysKauffman."Well, I becameabsolutely enthralledwith the
problemof cellulardifferentiation,and setstraightawayto thinking hard
aboutit."
It wasa goodtime to be doingso:]acoband Monodwerepublishing
theirfirstpaperson geneticcircuitsin 1961through1963.It wasthe work
for which they laterwon the NobelPrize(andwhich BrianArthur wasto
discoversixteenyearslater on the beachat Hauula).So Kauffmansoon
cameacrosstheir work showingthat any cell containsa numberof "reg-
ulatory" genesthat act asswitchesand can turn one anotheron and off.
"That workwasa revelationfor all biologists.If genescanturn oneanother
on and off, then you can h"€€qn.ti.gi:gits.,Somehow, the genomehas
to be somekind of biochemical comDuter. It is thecomDuting behavior-
"e
the orderly behavioFiITE iI effi ffi ; that somlfiisfrns-h dw'one
cell can becomedifferentfrom another."
The questionwashow?,
Actually,saysKauffman,mostresearchers at the time (or for that matter,
now) weren'tterribly botheredby this question.They talkedabout the
"developmental program"of the cell asif the DNA computerwerereally
carryingout ib geneticinstructions in thesamewaythatan IBM mainframe
executes a programwrittenin FOMRAN: stepby stepby step.Moreover,
they seemedto believethat thesegeneticinstructionswereorganizedwith
exquisiteprecision,havingbeenasthoroughlydebugged by naturalselec-
tion asanypieceof computercodeeverdevised by humans.How could
Secretsof the Old One 107

it be otherwise?The slightest error in the genetic proglam could turn a

developingcell cancerousor kill it entirely.That's why hundredsof mo-
leculai geneticists werealreadyhardat work in their laboratories deciphering
the precisebiochemicalmechanisms by which gene A switched on gene
B, and how that switching Processwas affected by activities of genes C, D,
and E. Detail, they felt, was everything.
And yet, the more Kauffman thought about this picture, the more he
found this questionof How? looming large.The genomewasa computer,
all right. But it wasn't anything at all like the machinesthat IBM was
turniig out. In a real cell, he realized,a greatmany regulatorygenescould
be activeat the sametime. So insteadof executingits instructionsstep by
step by step, the way human-built computersdo, the genomic computer
must be executingmost or all of its genetic
-h. instructioruslmu!!ry11 in
parallel. And if ihrt *", the case, r."ron.d, then wffiI61fr6frl*tt \
not whether flris regulatory gen€ activated that regulatory gene in some I
preciselydefined sequence.What matteredwas whether the genome as a I
whole could settledown into a stable,self-consistenlpatternof activegenes. I
At most, the regulatory genesmight be going through a cycle of two or
three or fou, config,rrations-a small number, anyhow; otherwise, the cell
would just thrash around chaotically,with genesswitchingeach other on
and off at random. Of course, the pattern of active genesin a liver cell
would be very different from the patternin a musclecell or a brain cell'.
But maybethat was fust the point, Kauffman thought. The fact that a single\
genome can have many stablepatternsof activation might be what allows I
it to give rise to many different cell types during development' ,
Kruff*an was also troubled by people's tacit assumption that detail was
everything.The biomoleculardetailswere obviouslyimportant, he knew.
But if the genome really had to be organizedand fine-tuned to exquisite
perfectionbeforeit could work at all, then how could it havearisenthrough
ih. r"r,dorn trial and error of evolution? That would be like shuffing an
honestdeck of cardsand then dealingyourselfa bridge hand of thirteen
spades:possible,but not very likely. "It iust didn't feel right," he says'"You
don't want to askthat much of eitherGod or selection.If we had to explain
the order in biology by lots of detailed, incredibly improbablebits of se-
lection, and ad hocery, if everything we see was a hard struggle in the
beginning,we wouldn't be here. There simply was not world enough and
time for chanceto have brought it forth."
There had to be more to it than that, he thought. "Somehow, I wanted
it to be true that the order emergedin the first place, without having to be
built in, without having to be evolved.I intentionallywantedit to be true
108 COMPLEXITY
that the order in a geneticregulatorysystemwasndturuI, that it wasquasi-
inevitable. Somehow, the order would just be there for free. It would be
spontaneous."If that was the case,he reasoned,then this spontaneous,
self-organizingproperg of life would be the flip side of natural selection.
The precisegeneticdetailsof any given organismwould be a product of
random mutationsand natural selectionworking just as Darwin had de-
scribedthem. But the organizationof life itself,the order,would be deeper
and more fundamental. It would arise purely from the structureof the
network, not the details. Order, in fact, would rank as one of the secrets
of the Old One.
"Where that impulsecomesfrom, I don't know," he says."Why should
Stu Kauffman have happenedto come along and wonder that? It's an
absolutelywonderful puzzle. I think it's a bizarre and wonderful thing that
a mind can come afreshto a problemand can aska questionlike that. But
I've felt that way all my life. And all the sciencethat I've done, that I truly
love, is an effort to understandthat vision."
Indeed, for the twenty-four-year-old premed student, the question of
order was like an itch that wouldn't go away.What would it really mean
for genetic order to be there "for free," he wondered?Well, look at the
geneticcircuits you find in real cells. They've obviouslybeen refined by
millions of yearsof evolution. But other than that, is thereanything really
specialabout them? Out of all the zillions of possiblegeneticcircuits, are
they the only ones that can produceorderly,stableconfigurations?If so,
then they would be the analog of the bridge hand with thirteen spades,
and it would truly be a miracle that evolution was ever lucky enough to
produce them. Or are stablenetworksactually as common asthe usual mix
of spades,hearts,diamonds,and clubs?Becauseif thatwas the case,then
it would be easyfor evolution to stumble on a useful one; the networksin
real cellswould just be the onesthat happenedto survivenaturalselection.
The only way to find the answer,Kauffman decided, was to shuffie the
deck, so to speak,deal out a bunch of "utterly typical" genetic circuits,
and seeif they did indeedproducestableconfigurations."So I immediately
started thinking about what would happen if you just took thousandsof
genesand hookedthem togetherat random-what would they do?"
Now here was a problem he knew how to think about: he had studied
neural circuitry ad nauseamat Oxford. Realgeneswereprettycomplicated,
of course. But Jacoband Monod had shown that the rgg&lgg3epgs, at
least,were essentiallyjust switches.And the essenceof a switch is that it
flips back and forth betweentwo states:active or inactive. Kauffman liked
to think of them as light bulbs (on-off) or as a statementin logic (true-
Secrets of the Old One 109

false). But whatever the image, he felt that this on-off behavior captured
the essenceof the regulatorygene. What remainedwas the network of
interactionsbetweengenes.So, as the Berkeleyfree speechmovement was
unfolding down on campus,he spenthis sparetime sittingon the rooftop
of his apartment in Oakland, obsessivelydrawing little diagrams of his
regulatoiy geneshooked up in wiring diagrams, and trying to understand
how they turned each other on and off'
The obsessiondidn't let up, even after he finished his premed coulses
at Berkeleyand startedgoing to medical school in San Franciscofull time.
It wasn't tirat he wasbored in medical school. Quite the opposite:he found
medical schoolvery, very difficult. When his teachersweren'tdemanding
mountainous quantities of memorization, they would go through an infi-
nitely painstakingsystems analysisof thingslike the physiologyof the kidney.
At thai point, moreover,he still had every intention of practicing medicine.
It appealedto the Boy Scout in him: it was a combinationof doing good
and knowing exactly what to do in any given situation, like pitching a tent
in a storm.
No, Kauftnan kept playing with the networksbecausehe almost couldn't
help himself. "l passionatelywanted to be doing this bizarre scienceabout
these random nets." He got a C in pharmacology:"My notes from that
classare all full of diagramsof genetic circuitry," he says'
He found that circuitry terribly confusing at first. He knew a lot about
abstractlogic, but almost nothing about mathematics.And the computer
textbookshe found in the library told him almost nothing that was helpful.
..Automata theory was well establishedby then, and that was all about
logical switching nets. But those bookstold me how to synthesizea system
that would do something,or what the generallimits were of the capaci$
of complex automata. What I was interestedin were the natural laws of
comple* systems.Whence cometh the order?And nobody was thinking
about that at all. Certainly nobody that l knew." So he kept on drawing
reams of diagrams, trying to get an intuitive feel for how these networks
might behave. And whatever mathematics he needed, he invented for
himself as best he could.
He quickly convinced himself that if the network became as densely \
tangled as a plate of spaghetti,so that every gene was controlled by lots of I
other genes,then the systemwould iust thrasharound chaotically.Using I
the lig-ht bulb analogy, it would be like a giant Las vegas-style billboard I
gone
- haywire, so that all the lights twinkled at random. No order here'
Kauftnan likewise convi.,cel himself that if each gene were controlled I
by at most one othergen€, so that the network wasvery sparselyconnected,
I
 110 COMPLEXITY
t
I then its behaviorwould be too simple. It would be like a billboard where
I most of the bulbs just pulsed on and off like mindless shobe lights. That
wasn't the kind of order Kauffman had in mind; he wantedhis geneticlight
bulbs to organize themselvesinto interestingpatternsanalogousto waving
palm hees or dancing flamingos. Besides,he knew that very sparselycon-
nected networks were unrealistic facob and Monod had already demon-
that real genestendedto be controlled by severalother genes.(Today,
fstrated
/' the number is known to be gpically two ro !ep.)
So Kauffman startedto concentrateon networksin between,where the
connectionswere sparse,but not too sparse.To keepthings simple, in fact,
he looked at networks with precisely two inputs p". g.n.. And here he
began to find hints of something special. He already knew thet densely
connected networkswere hypeisensitivein the extreme:if you went in and
flipped the stateof any one genefrom say,on to olff,then you would higger
a whole avalancheof changesthat would cascadeback and forth through
the network indefinitely. Thatt why denselyconnectednetworls tended to
be chaotic. They could never settledown. But in his two-input networks,
Kauffman discoveredthat flipping one gene would typica\y not produce
an ever-expandingwave of change. Most often, the flipped gene would
simply unflip, going back to what it was before. In fact, ro ton! as the two
different patternsof gene activation were not too different, theyLuld tend
to converge. "Things were simplifying," saysKauftnan. 'rI could seethat
light bulbs tended to get into stateswhere they got stuckon or off. " In other
,, yodtr the two-input networla were like a billboard where you could start
s" lights blinking at random, and yet they would alwaysorganizethemselves
/ { into a flamingo or
\ a champagneglass.
order! Stealingwhatevertime he could from his medical counes. Kauff-
man filled his notebookswith more and more of his random two-input
nehvorks,analyzingthe behavior of each one in detail. It was both tan-
talizing and frushating work. The goodnewswasthat the two-input networks
almost alwaysseemedto stabilize very quickly. At most they would cycle
over and over through a handful ofdifferent states.That's exactlywhat you
wantedfor a stablecell. The bad newswasthat he couldn't tell if his models
had anything at all to do with real geneticregulatorynetworks.Realnetworks
in real cells involved tensof thousandsof genes.And yet Kauffman'spencil-
and-papernetworla were alreadygetting out of hand when they contained
only 6ve or six genes. Keeping track of all the possible statesand state
transitionsof a seven-genenetwork meant filling in a matrix containing
128 rows and 14 columns. Doing it for an eight-genenetwork would have
Secretsof the Old One 111

required a matrix twice as big as that, and so on. "The chancesfor making
an error by hand were iust awfully large," saysKauffman' "l kept looking
longingly at my seven-elementnetworks, I just couldn't stand the idea of
having to do eight."
"Aiyway," he rays,"somewherein my sophomoreyearin medicalschool
I couldn't take it anymore. I'd been playing long enough. So I went across
the street to the computer center, and I askedif someonewould help me
'Sure,
programit. They said, but you haveto pay for it'' So I whipped out
my wallet. I was readyto pay for it."
Having decided to take the plunge into computers, Kauffman vowed to
go all out, he would simulate a network with 100 genes-looking back on
it no*, he laughs, it was a good thing he didn't quite know what he was
doing. Think of it this way. One gene by itself can have only two states:
on off. But a network of two genescan have 2 X 7, or four states:on-
"nd
on, on-off, offon, off'off' A network of three genescan have 2 x 2 x 2,
or eight states,and so on. So the number of statesin a network of 100
g..,.J i, 2 multiplied by itself 100 times, which turns out to be almost
lxactly equal to one million hillion trillion: I followedby 30 zeros.That's
an immense spaceof possibilities,saysKauffman. In principle, moreover,
there was,,o ,."ron why his simulatednetworkshouldn't haveiust wandered
around in that spaceat random; after all, he was deliberatelywiring it up
at random. And that would have meant that his idea of cell cycles was
hopeless:the computer would have had to go through roughly one million
triiion trillion transitionsbefore it ever startedretracing its steps.It would
be a cell cycle of sorts,but vastbeyondimagining. "If it takesa microsecond
for the computer to go from one stateto another," saysKauffman, "and if
it had to keep running for somethinglike a million trillion trillion micro-
seconds,you'd have billions of times the history of the universe. I'd have
n.u., *"d. it through medicalschool!"lndeed,the computerchargesalone
would have bankrupted him long before graduation'
Fortunately,however,Kauffman hadn't done that calculation at the time.
So, with the aid of a very helpful programmerat the computer center, he
coded up a simulated two-input network with 100 genesin it, and then
blithely turned in his deck of punch cards at the front desk. The answer
came back ten minutes later, printed out on wide sheetsof fan-fold paper.
And exactly as he had expected,it showedhis network quickly settling into
orderly states,with most of the genesfrozen on or off and the rest cycling
through a handful of configurations.Thesepatternscertainlydidn't look
like famingos or anything recognizable;if his 100-genenetwork had been
 112 COMPLEXITY
a las vegasbillboardwith a hundredlight bulbs,then the orderlystates
would havelookedlike oscillatingblobs.But they werethere, and they
werestable.
"I wasiust unbelia,ablythrilled!" sayslGuftnan. "And I felt then and feel
now that it wasratherprofound.I had found somethingthat no one would
haveintuitedthen." Insteadof wandering througha spaceof one million
hillion trillion states,his hvo-inputnetworkhadquicklymovedto an in6n-
itesimalcomerof thatspaceandstayed there."It settleddownandoscillated
througha cycleof five or six or sevenor, more typicallyit turned out,
aboutten states.That'san amazingamountof orderiI wasjust stunned."
That firstsimulationwasonly the beginning.Kauffmanstill hadno idea
of whysparselyconnectednetworlswereso magical.But they were,and
he felt asthoughthey had givenhim a wholenewwayof thinking about
genesand embryonicdevelopment.using that originalprogramasa tem-
plate and modifyingit as needed,he ran simulationsin endlessvariety.
When and why did this orderlybehavioroccur,he wantedto know.And,
not incidentally,how could he testhis theorywith realdata?
well, he thought, one obviouspredictionof his model wasthat real
geneticnetworkswould haveto be sparsely connected; denselyconnected
networksseemedincapableof settlingdown into stablecycles.He didn't
expectthemto havepreciselytwo inputspergene,like his modelnetworks.
Natureis neverquitethatregular.But from his computersimulations and
his reamsof calculations,he realizedthat the connections only had to be
sparsein a certainstatisticalsense.And when you lookedat the data,by
golly, real nehvorlcseemedto be sparsein exactlythat way.
Sofarsogood.Anothertestof thetheorywasto lookata givenorganism
with a givensetof regulatorygenes,and askhow many typ., it *",
capableof producing.Kauftnanknewhe couldn'tsayanything "-ll
specificalln
ofcourse,sincehe wasdeliberately trying to studythe typicir behaviorof
networks.But he could certainlylook for a statisticalrelationship.His
presumption all alonghadbeenthata oneof his
*"bJe-st4te cycles.$o hebeganto runbiglera simulationsl
trackof how manystatecyclesoccurredasthe sizi of the modelnetwork
. increased.By the time he got up to networksof 400 to 500 genes,he had
/ determined that the number of cyclesscaledroughly as the squareroot of
(\ the number of genesin the network.Meanwhile,he had alsobeenspending
'*every
sparehour in the medical school library, pouring through obscure
. .. references looking for comparable data on real organisms.And when he
finallr plotted it all up, there it was:the number of cell typesin an organism
I ff
did indeedscaleroughly as the squareroot of the number of genesit had.
II I
I ,L '
Secretsof the Old One 113
.,Goddamnit, it workedl"saysKauftnan. It wasthe
And so it went.
mostbeautifulthing he hadeverexperienced. By the endof his sophomore
yearat medicalschool he had run up hundredsupon hundredsof dollars
in computerbills. He paidit all withouta quiver'

In 1966,at thebeginningof his thirdyearof medicalschool,Kauffman

wrote a letter to the neurophysiologist warren Mcculloch of MII ex-
plainingwhat he had donewith his genetic networkmodelsand askingif
McCulloch wasinterested.
It tooka certainchutzpahto writethatletter,Kauffmanadmits.Originally
trainedas an MD himself, Mcculloch wasone of the grandold men of
neurophysiology-notto mentioncomputerscience,artificialintelligence,
,f,.'pnifJophy of mind. For the pasttwo decades,he and a band
"ra
oi toy"t fotto*.ir had beenworkingout the implicationsof an idea first
put iorward in 1943,when he and an eighteen-year-old mathematician
.r"-.d walter Pittshad publisheda paperentitled "A logical calcu]usof
the IdeasImmanentin NervousActivity."In thatpaper,McCullochand
Pittshad claimedthat the brain could be modeledasa networkof logical
operations suchasund, or, not, andsoforth' It hadbeena revolutionary
idea at the time, to put it mildlX and had provedto be immenselyinflu-
.ntirl. Not only wasthe McCulloch-Pittsmodelthe first exampleof what
would now be calleda neuralnetwork,it wasthe first attemptto under-
standmentalactivityasa form of informauonprocessing-aninsightthat
providedthe inspiration for artificialintelligenceandcognitivepsychology
th.i, model was also the first indicationthat a networkof very sim-
"lik..
ple logic gatescould perform exceedingly complex computations-an
i"sighithai wassoon incorporated into the general theory of computing
machines.
Grand old man or not, however,McCulloch seemedto be the only
scientistKauffmancould sharehis work with. "Mcculloch wasthe only
personI knewwho had donea lot of stuffwith neuralnetworks,"he says.
l.And it wasclearthat geneticnetworksand neuralnetworkswerefunda-
mentallythe samething."
Besides,Kauffmanbadly neededa litde outsidesupportby that point.
Medicalschoolwasshapingup to be a decidedly mixedblessing. He was
certainlygettingthe"facts"thathe'dwanted sobadly as a philosoph-y student
at Oxfoidl But theyweren'tgoingdown very well. "l think I chafed inside
at havingto takeotherpeople'sword for whatyou're supposed to do," he
says."What onehadto do in medicalschool was to master the facts,master
114 CAMPLEXIW
the diagnosis,absorbthe pearlsof diagnosticwisdom,and then execute
the appropriateprocedures. And while thereis a ioy in executingthose
procedures, it didn'thavethebeaugthatI wanted.It wasn'tlikesearching
for the secrets of the Old One."
His professors, meanwhile,didn't takekindly to the factthat Kauftnan
wasfindingthatbeautyin hisgeneticnehvorks. "One of the mostprofound
thingsyougothroughin medicalschoolisalmosta hazing,"saysKauftnan.
The round-the-clock duty shifts,the endlessdemands-"The purposeis
to makeit clearto you that the patientcomesfirst. You wiII get up at four
thirty in the morningto do what'snecessary. That part I didn't mind at
all. But thereweresomeof the facultyin medicalschoolwhothoughtthey
stoodasguardians to the houseof medicine.If youdidn't havcthe proper
attitudeto be a doctor,then you could neverbe a realdoctor."
Kauffmanremembers his junior-yearsurgeryprofessor in particular:"He
thoughtmy mindwaselsewhere-and he wasright,"he says,"I remember
him tellingme thathe didn't careif I gotan A on the final,he wouldgive
me a D for the course.I think I got a B on the final and he still gaveme
aD.
"So you haveto picturebeinga medicalstudentand beinggrottyand
not very happy,gettinga D in surgery-it wasan emotionalthing for me.
I hadbeena Marshallscholaranda big success academically, and hereI
wastryingto survivein medicalschool,with my professor of surgerytelling
me whata loserI was."
Aboutthe onlybrightspotin his life, in fact,wasthathe hadjustgotten
marriedto an ltalian-American NewYorkernamedElizabeth Ann Bianchi,
a graduatestudentin art whom he had met duringhis Oxforddayswhile
shewasan undergraduate travelingthroughEurope."I washoldingthe
dooropenfor herandthought,'Gee,that'sa prettygirl.' I'vebeenholding
the door for her eversince."And yet evenshehad to wonderaboutthis
networkstuff."Liz is muchmoreconcrete thanI am," sayKauffman."She
wasprofoundlyinterested in medicine.Shecamewith meto anatomy class
and a bunch of otherthings.But her response to the networkswas'That's
nice-but is it real?'Toher it seemed awfullyphantasmagorical."
It wasin the midst of all that that the replyarrivedfrom McCulloch:
"All Cambridgeexcitedaboutyour work," he wrote.Kauftnan laughsat
the memory."It took me abouta yearto figureout thatwhenWarrensaid
that, it meanthe had readwhat I senthim and thoughtit waskind of
interesting."
But at the time, he wasboth thrilledand astonished at McCulloch's
reply.He hadn't expectedanythinglike it. He wasemboldened to write
 115
Secretsof the Old One
back,explainingthatUC-SanFrancisco encouraged its iunior-yearmedical
,tudent"to go slom.*h.re elsefor threemonthsto getoutsideexperience'
with Mc-
So could he come out to MIT and spendthat time working
Culloch?
Liz
Certainly,McCullochwroteback'And furthermore,Kauftnanand
chouldstaywith him while theywere there'
They acceptedinstantly'Kauffmanwill neverforgethis first meeting
he and
with McCulloch:it wasaboutnineo'clockof a winter'sevening,as
ii, *.r. driving aroundandaround in the dark in a strange neighborhood
"Marsachrrsetts, lost after havingdriven all the
in Cambridge, hopelessly
wayacrossclountry'"And then there was Warren, loomingout of the fog
*,tr, ni, patriarchalbeard,to welcomeus into his home." while his wife,
travelers, McCulloch called
Rook,seiout cheeseandteafor the exhausted
group: "Kauffmanfs
,rf tU"ruin Minsky,guru of MIT's arti6cialintelligence
in town."
McCulloch,a devoutQuaker,provedto be a considerate andfascinating
host. Enigmatic,lyrical, possessed of a mind that wandered freelyovera
vastinteliectual'landscape, he wasendlessly enthusiastic about the quest
bygone s$le' filling his
for the inner workingtof thought.He wrotein a
to saint
scientificarticleswith allusionsto everyonefrom Shakespeare
Bonaventura-andthen giving them titles like "Where Is FancyBred?"'
"Why the Mind Is in the HeaJ," and "Throughthe Den of the Metaphy-
one of
sician.,'He lovedriddlesand wordplay.And he tumed out to be
th. f.* peoplein the world who could outtalk Kauftnan himself.
.,W"rr"n iendedto corneryou into conversations that draggedon," says
Kauffman.Former studentswho had lived with Mcculloch told storiesof
ie"uing the house through the upper bedroom window to avoid being
trappei. McCulloch *ouid h"bit tally follow Kauffrnan into the bathroom
*hii. he wastaking a shower,flip down the toiletseat, and sit there happily
discussing networksand logical functions of various kinds while Kauffman
wastrying to getthe soapout of his ear.
Uort i"*poit"nt, however,McCulloch becamea mentor,guide' and
friendto Kauffrnan-as he did with virtuallyall of his students.Knowing
that Kauffman's goalat MIT wasto run reallybig computersimulations'
so that he coulJ begin to get detailedstatisticalinformationabout his
networks'behavior,McCulloch introducedhim to Minsky and his col-
leagueSeymourPapert,who in turn arrangedfor Kauftnan to do his
sim--ulationson the powerhousecomputersof what was then known as
ProjectMAC: Machine-AidedCognition. McCulloch likewisearranged
for kauffmanto getprogramminghelp from an undergraduate who knew
 116 COMPLEXITY
a lot more about computercode than he did; they endedup running
simulationswith thousands of genes.
Meanwhile,Mcculloch wasintroducingKauftnanto the small but
intenseworld of theoreticalbiology.It wasin the living room of Mc-
culloch'shousethat he met the neurophysiologist Jackcowan, who had
beenMcculloch'sresearch assistantbackin thelatei950sandearly1960s,
and who had just beengiven the mandateto reiuvenatethe theoretical
biologygroupat the universig of chicago. It wasin Mcculloch's office
thatKauffmanmetBrianGoodwinof theUniversity of Susexin England,
who hasbeenone of his closest friendseversince.
"warren was like Fred Todd," saysKauftnan. "warren wasthe first
personto takeme seriouslyasa youngscientistin my own right, not asa
student." Sadly,McCulloch died only a few yearslater, iri 1969. But
Kauftnan still considershimself, in somesmall way, his heir. "warren
literallycatapultedme into the world that I've lived in eversince."
Indeedhe did. Kauffmanhaddecidedbeforehe evercameto MIT that
hewoulddevotehimselfto science oncehegraduated, notmedicalpractice.
It wasthegroupof peoplehe metthroughMccullochwho realllbrought
him into the fold.
"It wasthroughfackcowan, BrianGoodwin,andothersthatI wasinvited
to my first scientificconference in 1967,"he says.The eventwasthe third
in a seriesof conferences on theoreticalbiologyrun by the late British
embryologistconrad waddington."Thoseeonferences werean attempt.
in the mid-late1960s,to bewhatthe instituteis like today,"saysKauftnan.
"lt wasjustwonderful.Fromdrawingbloodandchecking stoolsamples
at
four in the morning-talk about geftingyour handson realityl-I was
flown to northernltaly to the Villa serbellonion lake como, a sitepicked
out by Pliny the Younger.Absolutelygorgeous. And herewerealf these
amazingpeople.fohn MaynardSmith wasthere. ReneThom was iust
inventingcatastrophe theory.Dick Lewontinfrornchicagowasthere.Dick
Levinsfrom chicago.Lewiswolpert from London.Thesearepeoplewho
arestill friends.
"So I gavemy talk aboutorderin thesegeneticnets,and the numbers
of cell types,and so on," he says."And afterwardwe went out for coffee
on the terraceoverlookingthreearmsof the lake.fackcowan walkedout
and askedme if I wantedto be on the facultyat chicago. So I thought
aboutit for exactlya nanosecond and said,'of course!'I didn't askfack
for a yearand a half whatmy salarywouldbe."
Secretsof the Old One 117
Death and Life

Aroundlunchtimeon thatfirstdayof Arthur'sresidence at the SantaFe

Institute,he and Kauffmanwandered down through the adobe art galleries
C"nyon Roadto what wasthen called Babe's, one of Kauffman's
"tong
f"uoit. *atering holes.And almostevery day thereafter for the next two
weeksthey met for lunch againor iust to talk'
They did much of their talkingon their feet.Kauftnan lovedthe open
air evenmore than Arthur did. As a teenagerin the Boy Scoutshe had
gon. on innumerablehiking and campingexpeditionsin the Sierras'In
lollegehe had beenan avid skierand mountainclimber.And nowadays
he stiil wentout hikingwheneverpossible.SoKauffrnanandArthur would
talk asthey strolledalotg CanyonRoad,or astheywalkedup behindthe
conventontoa broadhill, wheretheycouldsit on top andgazeout at an
immenseview of SantaFe andthe mountains'
Therewasan ineffablesadness aboutKauftnan,Arthur beganto realize.
In the midst of his iokes,his wordplay,his omnivorouscuriosity,and his
incessant talkingabouthis ideas,therewouldsometimes comea pause'A
fash of grief. And one eveningshortlyafter he arrivedin SantaFe, as
Arthur nir wife, Susan,werehavingdinner out with the Kauftnans,
"ia
Kauffman told them the story:How he and Liz had come home
Stuart
one Saturday night the previousOctoberto learnthat their thirteen-year-
o1l daughter, M1rit, had beenstruckby a hifand-run driverand wasin
grru. .o"ndition at a localhospital.How thel hadracedto the hospitalwith
ih"ir ron, Ethan. And how theyhadlearnedwhentheygottherethatMerit
had died fifteen minutes before.
Today,more than half a decadeafter the fact, Kauffmancan tell that
storywiihoutbreakingdown.But he couldn'tthatnight.Merit Kauffman
hadbeenverymuch her father'sdaughter."lt waspulverizing,"saysKauff-
,,fustagonizing.There'sno wayof describing it. we wentupstairs
man.
to seeher. There was my daughter's broken body on a table,coolingoff.
The three ofus huddled in bedtogether thatnight,crying.
|ustunbearable.
'Sh.
t a kind of feistiness,but an awareness of people we iust marveled
"d the bestof the four of us'"
at. We all felt that shehad been
..Theysaythattime heals,"he adds."But that'snot quitetrue. It's simply
that the grief eruptslessoften."

As theywalkedalongthe roadsandhillsidesaroundthe convent,Arthur

couldn'thelp but be intriguedby Kauffman'sconceptof orderand self-
 118 COMPLEXITY
organization.The irony of it wasthat when Kauftnan usedthe word "order,"
he wasobviously referring to the samething that Arthur meant by the word
"mg5sin6s5"-namely emergence,the incessanturge of complex
systems
to organizethemselvesinto patterns.But then, maybeit wasn'tso surprising
that Kauftnan was using exactly the oppositeword; he was cominjat the
concept from exactly the oppositedirection. Arthur talked about imessi-
ness" because he had started from the icy, abshact world of economic
equilibrium, in which the laws of the market are supposedto determine
everythingaspreciselyasthe lawsof physics.Kauffman talkedabout "order"
becausehe had started from the messy,contingent world of Darwin, in
which thereareno laws- justaccidentand naturalselection.But by starting
from totally different places,they had arrived at essentiallythe sameplace.
Kauffman, meahwhile, was both intrigued and perplexed by Arthur's
increasing-returnsideas. "I had a hard time understandingwhy this was
new," he says. "Biologists have been dealing with positive feedback for
years." It took him a long time to comprehendjusthow staticand changeless
the neoclassicalworld view really is.
He waseven more intrigued, however,when Arthur startedaskingKauff_
man about another economic problem that had been bugginghim: tech-
nological change. This had alreadybecome a political hot-button issue,to
put it mildly. You could feel the undercurrent of anxieg in almost every
magazineor newspaperyou picked up: Can America Compete?Have we
lost our fabled American ingenuity, the old yankee know-how?Are the
|apanesegoing to wipe us out, industry by industry?
Good questions.The problem, as Arthur explainedto Kauffman, was
that economistsdidn't have any answers-at least, not at the level of fun-
damental theory. The whole dynamic of technological development was
like a black box. "Until about fifteen or twenty years ago,,,he says, ..the
notion was that technologiescame at random out of the blue, fell from
heavenin celestialbooksof blueprintsfor making processsteel,or silicon
chips, or anything like that. And thosethings were made possibleby in-
vsn[615-51v12rtpeople like Thomas Edison who sort of got these ideas in
their bathtubsand addeda pageto their book of blueprints." strictly spear-
ing, in fact, technology wasn't part of economics at all. It was ..exoge-
nous"-delivered magically by noneconomic processes.More recentln
there had been a number of efforts to model technology as being ,.endog-
enous," meaningthat it's producedwithin the economicsystemitself. But
usually that meant regardingtechnologyas the outcome of investmentin
researchand development,almost like a commodity. And while Arthur
Secretsof the Ofd One 119

thought there might be some truth to that, he still didn't think it got to
the heart of the matter.
When you look at economic history, as opposedto economic theory, he
told Kauffman, technology isn't really like a commodity at all. It is much
more like an evolving ecosystem."In particular, innovations rarely happen
in a vacuum. They are usually made possibleby other innovations being
alreadyin place. For example,a laserprinter is basicallya Xerox machine
with a laser and a liule computer circuitry to tell the laser where to etch
on the Xerox drum for printing. So a laser printer is possiblewhen you
havecomputer technology,lasertechnology,and a Xerox reproducingtech-
nology. But it is also only possiblebecausepeople need fancy, high-speed
printing."
In short, technologiesform a highly interconnectedweb-or in Kauff-
man's language,a network. Furthermore, these technologicalwebs are
highly dynamic and unstable.They can grow in a fashionthat is almost
organic, as when laserprintersgive rise to desktoppublishing software,and
desktop publishing opens up a new niche for graphics programs. "Tech-
nology A, B, and C might make possibletechnologyD, and so on," says
Arthur. "So there'd be a networkof possibletechnologies,all interconnected
and growing as more things becamepossible.And thereforethe economy
could becomemore complex."
Moreover, these technological webs can undergo burstsof evolutionary
creativity and massiveextinction events,iust like biological ecosystems.Say
a new technology like the automobile comes in and replacesan older
technology,the horse.Along with the horsego the smithy, the pony express,
the watering troughs, the stables,the people who curried horses, and so
on. The whole subnetworkof technologiesthat dependedupon the horse
suddenly collapsesin wlrat the economist JosephSchumpeteronce called
"a gale of destruction." But along with the car come paved roads, gas
stations, fast-food restaurants,motels, traffic courts and traffic cops, and
traffic lights. A whole new network of goodsand servicesbegins to grow,
each one filling a niche opened up by the goodsand servicesthat came
before it.
Indeed, said Arthur, this processis an excellent example of what he
meant by increasingreturns:once a new technologystartsopening up new
niches for other goodsand services,the people who fill those niches have
every incentive to help that technology grow and prosper.Moreover, this
processis a major driving force behind the phenomenonof lock-in: the
more niches that spring up dependenton a given technology, the harder
120 COMPLEXITY
it is to change that technology-until somethingvery much befter comes
along.
So this idea about technological webs was very much in keeping with
his vision of a new economics, Arthur explained. The problem was that
the mathematics that he had developedwas only good for looking at one
technology at a time. What he really neededwasa networkykind of model
like the ones Kauftnan had developed."So," he askedKauffman, ',could
you do a model where a technologyis 'switchedon'when it is created,
maybe,and...?"
Kauffman listenedto all this thunderstruck.Could he? What Arthur was
describingwas, in a very different language,a problem that Kauftnan had
been working on for a decadeand a half.
Within minutes Kauffman wasoff, explaining to Arthur why the process
of technological change is exactly like the origin of life.

Kauffman 6rst got the idea back in 1969, around the time he arrived at
the theoreticalbiology group in Chicago.
After medicalschool,he says,beingin Chicagofelt like beingin heaven.
Looking back on it, in fact, Chicago was the second of his three most
exciting intellectual environments."lt was an extraordinaryplace full of
exhaordinarily able people," he says."The departrnentin Chicago was the
focus in the United Statesfor the sameset of friends I'd met in ltaly." fack
Cowan wasdoing his groundbreaking work on corticaltissue,writing down
simple equationsthat describedwavesof excitation and inhibition moving
acrosstwo-dimensionalsheetsof neuronsin the brain. JohnMaynardSmith
wasdoing his equallygroundbreaking workon evolutionarydynamics,using
a mathematical technique known as game theory to clarify the nature of
competition and cooperationamong species.Maynard Smith, who was
there on sabbaticalfrom the Universig of Sussex,alsogaveKauftnan some
much-neededhelp on the mathematicalanalysisof networla. "John taught
me to'Do sums,'ashe put it," saysKauftnan, "and I curedhis pneumonia
one day."
Now that he was surroundedby colleaguesand soul mates,Kauftnan
quickly discoveredthat he had not been alone in thinking about the sta-
tistical properties of networks. In 1952, for example, the English neuro-
physiologistRossAshby had speculatedalong much the samelines in his
book Design for a Brain. "He wasaskingquite similar questionsabout what
is the generic behavior of complex networls," saysKauffman. "But that
Secretsof the Old One 121
wascompletelyunknown to me. I got in touch with him assoonas I found
out about it."
At the same time, Kauffman discoveredthat in developing his genetic
networks, he had reinventedsome of the most avant-gardework in physics
and applied mathematics-albeit in a totally new context.The dynamics
of his genetic regulatory networksturned out to be a special caseof what
the physicistswerecalling "nonlineardynamics." From the nonlinearpoint
of view, in fact, it was easy to see why his sparselyconnected networks
could organizethemselvesinto stablecyclesso easily:mathematically,their
behavior was equivalentto the way all the rain falling on the hillsides
around a valley will flow into a lake at the bottom of the valley. In the
spaceof all possiblenetwork behaviors,the stablecycleswere like basins-
or as the physicistsput it, "athactors."
After six yearsof agonizing over thesenetworks,Kauffman was gratified
to think that he was finally beginning to understandthem so well. And
yet-he couldn't help but feel that somethingwas still missing.Thlking
about self-organizationin genetic regulatory networks was all very nice.
But at a molecularlevel, geneticactivig dependson the incrediblycomplex
and sophisticatedmolecules known as RNA and DNA. Where did they
come from?
How did life get startedat all?
Well, according to the standard theory in the biology textbooks, the
origin of life was rather straighdorward. DNA, RNA, proteins, polysac-
charides,and all the other moleculesof life must have arisenbillions of
yearsagoin somewarm little pond, wheresimple molecular building blocks
like amino acids and such had accumulated from the primordial atmos-
phere. Back in 1953, in fact, the Nobel laureatechemistHarold Urey and
his graduate student Stanley Miller showed experimentally that an early
atmosphereof methane, ammonia, and the like could haveproducedthose
building blocks quite spontaneously;all it would have taken was the oc-
casionallightning bolt to provide energyfor chemical reactions.So over
time, went the argument, thosesimple compoundswould have collected
in ponds and lakes, undergoing further chemical reactionsand growing
more and more complex. Eventually,therewould havearisena collection
of moleculesthat included the DNA double helix and/or ib single-strand
cousin, RNA-both of which havethe power to reproducethemslves.And
once there wasself-reproduction,all the restwould then follow from natural
selection. Or so went the standardtheory.
But Kauffman didn't buy it. For one thing, most biologicalmolecules
122 COMPLEXITY
are enormousobjects.To makea singleproteinmolecule,for example,
you might haveto chain togetherseveralhundredamino-acidbuilding
blocksin a preciseorder.That'shardenoughto do in a modernlaboratory,
whereyouhaveaccess to all thelatesttoolsofbiotechnology. Sohowcould
sucha thingformall by itselfin a pond?Lotsof peoplehadtriedto calculate
the oddsof thathappening,andtheir answers alwayscameout prettymuch
the same:if the formationweretruly random,you wouldhaveto wait far
longerthan the lifetimeof the universeto produceevenoneusefulprotein
molecule,much lessall the myriadsof proteinsandsugars andlipidsand
nucleicacidsthat you needto makea fully functioningcell. Evenif you
assumed that all the trillionsof starsin all the millionsof galaxies in the
observable universehad planetslike Earth, with warm oceansand an
atmosphere, the probabilitythat anyof them wouldbring forth life would
still be-infinitesimal. If the originof life hadreallybeena randomevent,
then it had reallybeena miracle.
More specifically,however,Kauffmandidn't buy the standardtheory
because it equatedthe originof life with the appearance of DNA. It iust
didn't seemreasonable to Kauftnan that the origin of life shoulddepend
on somethingsocomplicated. The DNA double helix can reproduce itself,
all right. But its ability to do so dependscriticallyon its ability to uncoil,
unzipitstwostrands, andmakecopiesof itself.In moderncells,moreover,
thatprocess alsodepends on a hostof specializedproteinmolecules,which
servevarioushelperroles.How could all that havehappenedin a pond?
"It wasthesameimpulsethatlaybehindmy question of whetheryoucould
Fnd order in geneticregulatorynetworks,"saysKauffman'"There was
somethingtoo marvelousaboutDNA. I simplydidn't want it to be true
that the origin of life depended on somethingquiteasspecialasthat. The
'What if Cod had hung anothervalence
wayI phrasedit to myselfwas,
bond on nitrogen?fNihogenatomsare abundantin DNA molecules']
Would life be impossible?'And it seemed to me to be an appallingcon-
clusionthat life shouldbe thatdelicately poised."
But then, thoughtKauftnan, who saysthat the criticalthing aboutlife
is DNA? For that matter,who saysthat the origin of life wasa random
event?Maybetherewasanotherwayto geta self-replicating systemstarted,
a way that would have allowed living systems to bootstrap their way into
existence from simple reactions.
Okay,then. Think aboutwhatthatprimordialsoupmusthavebeenlike,
with all thoselittle amino acidsand sugarsand suchbangingaround.
Obviously,you couldn't expectthem to iust fall togetherinto a cell. But
you couldexpectthem to undergoat leastsomerandomreactions with one
Secrets of the Old One 123

another. In fact, it's hard to seewhat could have stoppedthem from doing
so. And while random reactionswouldn't have producedanything very
fancy, you could do the calculationsand show that, on the average,they
would have produceda fair number of smallish moleculeshaving short
chains and branches.
Now, that fact in itself wouldn't have made the origin of life any more
probable. But suppose,thought Kauffman, just supposethat some of these
smallishmoleculesfloatingaround in the primordial soupwereable to act
as"catalysts"-submicroscopic matchmakers.Chemistsseethis sort of thing
all the time: one molecule, the catalyst,grabshvo other moleculesas they
go tumbling by and brings them together,so that they can interact and fuse
very quickly. Then the catalystreleasesthe newly weddedpair, grabsanother
pair, and so on. Chemists also know of a lot of catalystmolecules that act
as chemical axe murderers,sidling up to one moleculeafter another and
slicing them apart. Either way, catalysb are the backboneof the modem
chemical industry. Gasoline, plastics,dyes,pharmaceuticals-almost none
of it would be possiblewithout catalysts.
All right, thought Kauffman, imagine that you had a primordial soup
containing some molecule A that was busily catalyzingthe formation of
another molecule B. The first molecule probably wasn't a very effective
catalyst,since it essentiallyformed at random. But then, it didn't need to
be very effective. Even a feeblecatalystwould have made B-type molecuies
form faster than they would have otherwise.
Now, thought Kauffman, supposethat molecule B itself had a weak
catalytic effect, so that it boostedthe production of some molecule C. And
supposethat C also acted as a catalyst,and so on. If the pot of primordial
soup wasbig enough, he reasoned,and if there were enough different kinds
of molecules in there to start with, then somewheredown the line you
might very well havefound a molecule Z that closedthe loop and catalyzed
the creation of A. But now you would have had more A around, which
meansthat there would have been more catalystavailableto enhancethe
formation of B, which then would have enhancedthe formation of C, and
on and on.
In other words, Kauffrnan realized, if the conditio.nsin your primordial
soup were right, then you wouldn't have to wait for random reactionsat
all. The compounds in the soup could have formed a coherent, self-rein-
forcing web of reactions. Furthermore, each molecule in the web would
have catalyzedthe formation of other molecules in the web-so that all
the molecules in the web would havesteadilygrown more and more abun-
dant relative to molecules that were not part of the web. Thken as a whole,
124 COMPLEXITY
itsownformation.
in short,thewebwouldhavecala.lyzed It wouldhave
been an "a
all this.Hereit wasasain:order.
Kauffmanwasin ad6Thenhe realized
Order for free. ader arising naturally from the laws of physicsand chem-
istry.Order emergingspontaneously from molecularchaosand manifesting
itself as a systemthat grows. The idea was indescribablybeautiful.
But was it life? Well no, Kauffinan had to admit, not if you meant life
as we know it today. An autocatalytic set would have had no DNA, no
geneticcode, no cell membrane. In fact, it would have had no real in-
dependentexistenceexceptas a haze of moleculesfloating around in some
ancient pond. If an extraterrestrialDarwin had happenedby at the time,
he (or it) would have been hard put to notice anything unusual' Any given
molecule participating in the autocatalytic set would have looked pretty
much like any other molecule. The essencewas not to be found in any
individual piece of the set, but in the overall dynamics of the set: its
collective behavior.
And yet in somedeepersense,thought Kauffman, maybean autocatalytic
set would have been alive. Certainly it would have exhibited some re-
markablylifelike properties.It could grow, for example:there wasno reason
in principle why an autocatalyticset shouldn't be open-ended,producing
more and more moleculesastime went on-and moleculesthat weremore
and more complex. Moreover, the set would have possesseda kind of
metabolism: The web molecules would take in a steadysupply of "food"
molecules, in the form of amino acidsand other simple compoundsfoating
all around them in the soup, and catalytically glue them togetherto form
the more complex compounds of the set ibelf.
An autocatalyticsetwould even exhibit a primitive kind of reproduction:
if a set from one little pond happenedto slosh over into a neighboring
pond-in a food, say-then the displaced set could immediately start
growing in its new environment. of course, if another, different set were
alreadyin place, then the two would engagein a competition for resources.
And that, Kauffma.t realized, would immediately open the door for natural
selection to winnow and refine the sets. It was easy enough to imagine
such a processselectingthose setsthat were more robust to environmental
changei or that contained more efficient catalysb and more elaboratere-
actions or that contained more complex and sophisticatedmolecules. Ul-
timately, in fact, you could imagine the winnowing processgiving rise to
DNA and all the rest. The real key was to get an entity that could survive
and reproduce; after that, evolution could do ib work in comparatively
short order.
Secrets of the Old One 125

okax this was admittedly piling up a lot of ifs on top of ifs. But to
Kauffman, this autocatalyticset story was far and away the most plausible
explanation for the origin of life that he had ever heard. If it were true, it
meantthe origin of life didn't haveto wait for someridiculouslyimprobable
eventto producea setof enormouslycomplicatedmolecules;it meant that
life couli indeed have bootstrappedits way into existencefrom very simple
molecules.And it meant that life had not been iust a random accident,
but was part of nature's incessantcompulsion for self-organization'
Kauffman wasenthralled. He immediately plunged into calculationsand
computer simulations and random network models, doing exactly what he
had done at Berkeley:he wanted to understandthe natural laws of auto-
catalytic sets.Okay,he thought, you don't know what compoundsand what
reactionswere really involved way back when. But at least you can think
about probabilities.Was the formation of an autocatalyticset a wildly
unlikely event?Or was it almost inevitable?[,ook at the numbers' Suppose
you have a few different kinds of "food" molecules-amino acids, et cet-
era-and supposethat in the primordial soup, they start linking up into
polymer chains. How many differentkinds of polymerscan you make this
way? How many reactionsare there among the polymers, s'othat you can
make a big web of reactions?And what is the likelihood that if you had
this big web of reactionsit would close on itself and form an autocatalytic
set?
"As I thought it through," he says,"it became perfectly obvious to me
that the number of reactionswent up fasterthan the number of polymers'
So that if there wasany fixed probability that a polymer catalyzeda reaction,
there'd be some complexity at which this thing would have to become
mutually autocatalytic."In other words, it was iust like his genetic net-
works: if the primordial soup passeda certain threshold of complexi$,
then it would undergo that funny phasetransition. The autocatalytic set
would indeed be almost inevitable. In a rich enough primordial soup, it
would hcve to form-and life would "crystallize" out of the soup sPonta-
neously.
The whole storywas iust too beautiful, Kauffman felt.lthad to be true.
"I believe in this scenario as strongly now as I did when I first came up
with it," he says."I believe very strongly that this is how life began."

Arthur was ready to believe it too. He thought that this was great stuff,
and not only becauseit wasa marvelousideaabout the origin of life. The
analogiesbetween autocatalysisand economics were iust too delicious to
126 COMPLEXITY

passup. He and Kauffman kicked it around for days,walking through the

hills or hunched over the lunch tablesat Babe's.
Most obviously, they agreed,an autocatalyticset was a web of transfor-
mationsamong moleculesin preciselythe sameway that an economyis a
web of transformationsamong goodsand services.ln a very real sense,in
fact, an autocatalyticsetwasan economy-a submicroscopic economythat
extracted raw materials (the primordial "food" molecules) and converted
them into useful products (more molecules in the set).
Moreover,an autocatalyticsetcan bootstrapits own evolutionin precisely
the same way that an economy can, by growing more and more complex
over time. This was a point that fascinatedKauffman. If innovations result
from new combinationsof old technologies,then the number of possible
innovations would go up very rapidly as more and more technologiesbe-
came available. In fact, he argued, once you get beyonda certain threshold
of complexityyou can expecta kind of phasetransitionanalogousto the
ones he had found in his autocatalyticsets.Below that level of complexity
you would find countries depbndentupon iust a few maior industries, and
their economies would tend to be fragile and stagnant. In that case, it
wouldn't mafter how much investmentgot poured into the country. "If all
you do is produce bananas,nothing will happen except that you produce
more bananas."But if a country ever managedto diversifyand increase
its complexity abovethe critical point, then you would expectit to undergo
an explosive increase in growth and innovation-what some economists
have called an "economic takeoff."
The existenceof that phasetransition would also help explain why trade
is so important to prosperity,Kauftnan told Arthur. Supposeyou have two
different countries,each one of which is subcriticalby itself. Their econ-
omies are going nowhere. But now supposethey start trading, so that their
economiesbecome interlinked into one large economywith a higher com-
plexity. "l expectthat tradebetweensuch systemswill allow the ioint system
to become supercritical and explode outward."
Finally, an autocatalytic set can undergo exactly the same kinds of ev-
olutionary booms and crashesthat an economydoes. Iniecting one new
kind of molecule into the soup could often hansform the set ufterl5 in
much the sameway that the economywas hansformedwhen the horse was
replacedby the automobile. This was the part of autocatalysisthat really
captivatedArthur. It had the same qualities that had so fascinatedhim
when he first read about molecular biology: upheaval and change and
enormousconsequencesflowing from trivial-seemingevents-and yet with
deep law hidden beneath.
 Secretsof the Old One 127
So Kauffmanand Arthur had a grandtime spinningout ideasand finding
connections.It was like one of the all-time greatfreshmanbull sessions.
Kauffman was especiallyexcited. He felt that they were onto something
really new. Obviously,a network analysiswouldn't help anybodypredict
precisely what new technologies are going to emerge next week. But it
might help economistsget statisticaland structural measuresof the process.
When you introduce a new product, for example, how big an avalancheI
does it typically cause?How many other goodsand servicesdoes it bring I
with it, and how many old onesgo out? And how do you recognizewhen I
a good has becomecentral to an economy,as opposedto being iust another I
hula-hoop? I
Furthermore, Kauftnan felt that it might ultimately be possibleto apply
theseideasfar beyond the economy. "l think thesekinds of models are the
place for contingencyand law at the same time," he says."The point is
that the phasetransitions may be lawful, but the specific details are not.
So maybe we have the startsof models of historical, unfolding processes
for such things asthe Industrial Revolution, for example,or the Renaissance
as a cultural transformation,and why it is that an isolatedsociety,or ethos,
can't stay isolatedwhen you start plugging some new ideas into it." you
can ask the samething about the Cambrian explosion:the period some
570 million yearsago when a world full of algaeand pond scu* suddenly
burst forth with complex, multicellular creaturesin immense profusion.
"Why all of a suddendo you get all this diversity?"Kauffman asks."Maybe
you had to get to a critical diversity to then explode. Maybe it's because
you've gone from algal mats to something that's a little more trophic and
complex, so that there's an explosion of processesacting on processesto
make new processes.It's the same thing as in an economy."
Of course, even Kauftnan had to admit that none of this was any more
than a hope-yet. On the other hand, he told Arthur, it all might tre very
possible.He'd been laying the groundworkfor it since 1982, when he
returned to the subiectof autocatalysisafter a hiatus of more than a decade.

The hiatus beganone day in 1971, Kauftnan remembers,when Chicaeo

chemist Stuart Rice dropped by the theoretical biology group for a uisit.
Rice had a maior reputation in theoretical chemistry, and Kauffman very
much wantedto impresshim. "He camein and askedme what I wasdoing.
So I told him. And he said, 'Why in the world are you doing that?'I don't
know why he said that. I presumehe thought there was no point to it. But
'My
I thought, goodness,Stuart certainly knows what het talking about.
 128 COMPLEXITY
I shouldn'tbe doingthis.' So I wroteit all up and publishedin the Cy-
bemeticsSacietyloumal in 1971,and then put it away.And I forgotall
aboutit."
Kauffman'sreactionwasn'tentirelya matterof insecurity.The fact was
that his autocatalysis modelswereat an impasse.No matterhow many
calculations andcomputersimulations he carriedout on theoriginof life,
theywerestill justcalculations andcomputersimulations. To makea really
compellingcasehe would havehad to takethe experiments of Miller and
Urey one step farther, by demonstrating that their primordial soupcould
actuallygive riseto an autocatalytic set in the laboratory. But Kauftnan
had no ideahow to do that. Evenif he had had the patience and know-
how to do laboratorychemistry,he wouldhave had to look at millions of
possible compounds in all conceivable combinations under a wide range
of temperatures andpressures. He couldhavespenta lifetimeon theprob-
lem and gottennowhere.
No one elseseemedto haveany good ideaseither. Kauffmanwasn't
? alone in thinking about autocatalyticmodels.Severalyearsearlier,the
I Berkeley NobelistMelvin Calvinhadexploredseveral differentautocatalytic
i scenarios for the origin of life in his 1969book,ChemicalEvolution.ln
Germany,meanwhile,Otto Roessler waspursuingautocatalytic ideasquite
independently, aswasManfredEigenin Gdftingen.Eigenwasevenable
to demonstrate a form of autocatalytic cyclein the laboratoryusingRNA
molecules.But no one hadyet beenableto demonshate autocatalytic sets
emergingftom the simplemoleculesof a Miller-urey primordialsoup.
Pendinga new idea,thereseemednowhereto go.
Howiver, even if Kauffman'sreactionto Rice'scommentwasn'tall
because of insecurity,a lot of it undoubtedlywas.He wasfeelinga strong
needto provehimselfin his newfoundprofession. Theory,he wasdiscov-
ering,was in very low repute among biologisb.
"The peopledoingmathin biologywerethe lowestof the low," he says'
It wasexactfuthe oppositeof the situationin physicsor economics,where
the theoristsarekings.Especiallyin molecularanddevelopmental biology,
the experimental to6ls were still so new, and there was such a vast amount
of datato be collected about the details of living systems,that the honor
/ andtheslorvwentto the laboratory. "There is a remarkable certainty among
Lnol."ui", bioloeiststhat all the answenwill be found by understanding
I specific.ol""ul-.r," saysKauffman."There is a greatreluctanceto study
of an athactorstrikesthem
i t o* a systemworks.For example,the concept
\ asgobbledygook"'
ihe atmosphere wasnot quitesohostileto theoryin neuroscience and
 Secretsof the Old One 129
evolutionarybiology.But Kauffman'snetworkideaswereconsidered a little
weird eventhere. He wastalkingaboutorderand the statisticalbehavior
of largenetworlswithout being able to sayanythingspecificabout this
moleculeor that molecule.It washardfor mostresearchers to understand
what he meant."Peoplecertainlyresponded to the geneticnetworkstuff
earlyon," he says."Waddingtonliked it. All sortsof folksliked it. That's
why I got my first iob. And I felt very pleasedand proud of myself.But
thenit sortof quieteddown,andit wasa low tricklefromthe early1970s.
The world didn't particularlycare."
Instead,Kauffmanthrewhis considerable energiesinto learninghow to
do experimental biology.The impulsewasmuch the sameasthe one that
had led him from philosophyto medicalschool:He distrustedhis own
glibnessand theorizing."The emotionalrootswerein part that I needed
to be grounded,"he says."But it's alsoin partthat I reallywant to know
how the worldworks."
In particular,Kauffmanfocusedhis attentionon the tiny fruit fy, Dro-
sophilamelanogaster, which had been intensivelystudiedby geneticists
duringthe earlypartof the twentiethcentury,andwhich hadnow become
a favoriteexperimentalsubjectfor biologistsdoing researchinto the de-
velopmentprocess.Among its manyotherfine qualities,Drosophilahad
the ability to bring forth weird monstersknownas homeoticmutants,in
which a newly hatchedfy turns out to havelegswhereib antennaeare
supposed to be, or genitaliawhereits headis supposed to be, and so on.
ThesemutationsofferedKauffmanampleplayfor model-makingand for
thinking abouthow the developingembryoorganizes itself.
ln 1973,his work on Drosophilacarriedhim to the Nationallnstitures
of Health,iust outsideof Washington, D.C., wherehe had managedto
swinga two-yearappointmentthat allowedhim to fulfill his military re-
quirementwithoutgoingoffto vietnam. (while he wasat chicagohe had
arrangeda four-yeardraft defermentunderwhatwasknownasthe "Berry
Plan," which allowedphysicians to postpone activeservicewhile doing
medicalresearch.) And in 1975,his work on the fruit fly carriedhim
onwardto a tenuredappointmentat the universig of pennsylvania."I
chosePenn," he iokes,i'b.""ur. they havegreatIndian restaurants near
there."More seriously, he felt that he couldn'tbring his famirybackto
live in the Hyde Parkneighborhoodaroundthe universityof chicago,
eventhoughhe had beenofferedtenurethere."The crimeiate and rac-ial
tensionthereweresouglr" he says,"andyoufelt sohelpless to do anything
goodaboutit."
Kauffmancertainlydoesn'tregretthe time he spenton Drosophila.His
 130 COMPLEXITY
discourses on fruit fly development areaspassionate asanythinghe hasto
sayaboutnetworkmodels.And yet, he alsoremembers a vivid moment
in 1982."l wasup in the mountainsof the Sierra,and I realizedI had
not thoughta newthoughtaboutDrosophila for a coupleof years.I'd been
pushingto do the experiments on nuclear transplantation, and cloning,
andotherstuff.But I hadn'tactuallyhad a new thought. And I felt utterly
stuck.l'
Somehow,he knew at that momentthat it wastime to return to his
originalideasaboutnehvorksand autocatalysis. If nothingelse,damn it,
he felt that he had paid his dues: "l've earnedmy ordnright in my own
gutsto think what I want to think. Havinggonedrroughmedicalschool,
havingbecomea doctor,havingdelivered sixtybabies,and doingspinal
tapson infants,and takingcareof cardiacarrests, andall the thingsyou
do as a youngdoctor,havingrun.a lab and learnedhow to do 2-D gels
and l-D gels,and run scintillati6ncounters,havinglearnedhow to do
Drosophila genetics,this,-that,andthe otherthing-even if the biological
communitystill looks'aikance'at theory,I've earnedmy own right to do
what the hell Lwant. I've answered the Oxfordcraving-to not be afraid
to think glibli. I now iust trust myselfas a theoretician,profoundly.It
dqesrr'tmeanthat I'm right. But I trust myself."
In particular,he decidedthat it wastime to go backto theautocatalytic
setidla and do it right. Backin 1971,he says,all he'dreallyhad wasa
verysimplecomputersimulation."I clearlyunderstood that asthe number
of proteins in the solution goes up, the number of reactions goesup even
fasier.So if you get a complicated enough system, you get autocatalysis.
But I didn't havemuch in the wayof analyticwork'"
So he plungedback into his calculations-and,as usual,endedup
inventingthe mathematics that he neededashe wentalong."I spentthe
entirefall of 1983, from October until iust afterChristmas,provingall
sortsof theorems,"he says.Numbers of polymers,numbersof reactions,
probabilities of polymerscatalyzingreactions, phasetransitions in thisgiant
precisely what conditions would autocatalysis
iraph of reactions-under
["ip.". And how could he provethat it would happen?He remembers
deiivi'g a wholeflurry of resultsin November,duringa twenty-four-hour
, flight liome from a meetingin India. "l got back to Philadelphiaand
he says.He scribbledtheoremson ChristmasDay'And by New
"oilrpr.d,"
yeart lg84 he had iL a proofof this funny phasetransitionthat he could
only conjecturebackin 1971.If the chemishywastoo simpleand the
complexityof the interactionswastoo low, then nothingwould happen;
the systemwould be "subcritical."But if the complexi$of the interac-
Secretsof the Old One 131
tionswasrich enough-and Kauffman'smathematics now allowedhim to '
defineprecisely whatthatmeant-then thesystem " ;
wouldbe"supercritical.
Autocatalysis wouldbe inevitable.And the orderreallywouldbe for free. 'I
Wonderful.The next step,obviously,wasto testout thesetheoretical
ideaswith a muchmoresophisticated computersimulation. "I hadtheidea
of subcriticalandsupercritical systems," he says,"andI wasanxiousto see
if the simulationwould behavethat way." But it wasalso importantto
incorporate somethingresembling realchemistryandrealthermodynamics
into the model;if nothingelse,a morerealisticmodelmight giveexperi-
menterssomeguidanceabout how to createan autocatalyticset in the
laboratory.
Kauffmanknew iust the two peopleto help him. He had met one of
them, LosAlamosphysicistDoyneFarmer,duringa conference in Bavaria
in 1982.Thenonlytwenty-nine, Farmerhadprovedto havean imagination
just asfertileand asenergeticasKauftnan's;he wasalsoiust asfascinated
with the conceptof self-organization. They had spentone marvelousday
hiking throughthe Alps, talkingaboutnehvorksand self-organization all
thewhile.Theyhit it offsowell, in fact,thatFarmerarranged for Kauftnan
to startmakingperiodicvisitsto LosAlamosasa consultantand lecturer.
Shortlythereafter,Farmerhad introducedKauffmanto NormanPackard,
a youngcomputerscientist at the University of lllinois.
Farmerand Packardhad beencollaborators sincetheir graduatestudent
daysin the physicsdepartment at the Universityof Califomia,SantaCruz,
in the late 1970s.While theyweretheretheyhad both beenmembersof
the self-styled "DynamicalSystems Collective,"a smallcrewof graduate
studentswho devotedthemselves to whatwasthen the avant-garde subject
of nonlineardynamicsand chaostheory.They had madea number of
seminalconhibutionsto the field-a fact that would eam the Qnamical
SystemsCollectiveits own chapterin famesGleick'sbookCfraos,which
appeared in the fall of 1987at aboutthesametime thatArthur, Kauffman,
and the otherswereconvergingon SantaFefor the economicsmeeting.
And by the time Kauftnan first met them in the early 1980s,Farmer
and Packardwereboth gettingdownrightboredwith chaostheory.
As Farmersays,"So what?The basictheoryof chaoswasalreadyfleshed
out." He wantedthe excitementof beingon the frontier,wherethingswere
not well understood.Packard,for his part, wantedto get his hanis dirty
with somereal complexig.chaoticdynamicswerecomplex,all right think I
of the way a leaf seemsto flutter at randomin a steadybreeze,But the I
complexitywasprettysimpleminded. Thereis one setof forces-from the I
wind, in the caseof the leaf. Thoseforcescan be described by one setof
I
 132 COMPLEXITY
'l
mathematical equations. Andthesystem iustblindlyfollowsthoseequations
I fot.uet.Nothing changes, and nothingadapts. "l wantedto gobeyondthat,
I to richerformsof complexity likebiologyandthemind," saysPackard. He
andFarmerhadbeencastingaroundfor the rightproblemto workon. So
when Kauffmansuggested that they all collaborateon simulatingan au-
tocatalyticsystem,they decidedto giveit a try.
They got down to work in 1985,after Kauffrnanhad returnedfrom a
sabbatical that took him to Parisand |erusalem."It becamean.intense
collaboration,f' saysKauffman.It wasone thing to talk abouta random
networkof reactions; sucha networkcanbedescribed in cleanmathematical
terms.It wasquitesomething elseto modelthosereactions with reasonably
realisticchemistry; things got complicated fast.
What Kauffman,Fatmer,and Packardfinally cameup with wasa sim-
plifiedversionof polymerchemistry. The basicchemicalbuildingblocks-
a.,alogorrsto the amino acids and other simplecompounds that you might
expectto form in the primordial soup by the Miller-Urey process-were
represented in the model by symbols such as a, b, and c. These building
bllcks could then link up into chains to form larger molecules, suchas
accddbacd. Theselargermolecules, in turn, could undergo two kinds of
chemicalreaction.They couldsplitapart:

accddbacd--'>accd + dbacd
Or they coulil do the reverseand ioin at the ends:

bfuad + cccba-->bbcadcccba

Eachof the reactionswould havea numberassociated with it-what a

chemistwould call a rateconstant-andthat numberwould determine
how fastthe reactionwould occurif therewereno catalysts around.
But, of course,the wholepoint of the exercisewasto watchwhat hap-
penedwhen catalysts did appear'So Kauffman,Farmer,and Packardhad
io find a wayto specifywhich moleculecouldcatalyze which reaction.
They tried several-waysof doingthat. one way,which wassuggested_by
K"uff*"n and which workedas well as any other,wassimplyto pick a
seriesof molecules,such as abccd,andarbitrarilyassigneachone to a
reaction,suchasbaba t ccda-+ babaccda.
To run the model,onceall the reactionratesandcatalyticstrengthswere
specified,Kauffrnan,Farmer, and Packard would simplytell the computer
to start enrichingtheir simulatedpond with a steadystreamof "food"
Secrets of the Old One 133

moleculessuch as a, b, andaa. And then they would sit back to seewhat

their simulatedchemistrywould produce.
For quite a while, it didn't produce much of anything. This was frus- r
trating, but not surprising.Reactionrates,catalyticstrengths,rate of food !
supply-any number of parameterscould be off The thing to do was to !
u"iy ttr. parametersand seewhat worked and what didn't. And as they did i
thai, they began to find that occasionally,as they moved the parameters I
into certain favorable ranges, the simulated autocatalytic sets did form. I
Moreover, they seemedto form under iust about the same conditions that I
Kauffman had predicted from his theorems about abstractnetworks.
Kauffman and his collaboratorspublishedtheir resultsin 1986. Farmer
and Packardhad alreadymoved on to other interestsby that point-although
Farmer had taken on a graduatestudent, Richard Bagley, to amplify the
model and speedit up substantially.And Kauffman himself had gone on
to think about other waysin which self-organizationcould have occurred
in evolution. But after that computermodel, he felt more deeplythan ever
that he had truly come face to face with a secretof the Old One.
He remembers taking a solo hike back up into the Sierras near l,ake
Thhoe to one of his favoritespots,HorsetailFalls. It was a lovely summer
day,he recalls.He saton a rock by the falls,thinking abouthis autocatalysis
work and what it meant. "And suddenly,"he says,"l knew that God had
revealedto me a part of how his universeworks." Not a personalGod,
certainly; Kauffman had never been able to believe in such a being. "But
I had a holy senseof a knowing universe,a universeunfolding, a universe
of which we are privilegedto be a part. In fact, it was quite the opposite
of a vainglorious feeling. I felt that God would revealhow the world works
to anyone who cared to listen.
"lt was a lovely moment," he says,"the closestI've ever come to a
religiousexperience. "

Santafu

As the opening day of the economicsworkshopdrew nearer, and as

Arthur began to spend more and more time polishing his talk-he was
scheduledto be the first speakeron the first day-Kauffman went out by
himself for long walks along the dirt roadsnear his house. "I remember
pacing backand forth, trying to put togetherthe central conceptualstructure
of my own talk," he says.By agreement,Arthur would focushis presentation
on increasingreturns,while Kauffman,who had alreadyoutlined what he
134 COMPLEXITY
wanted to say about his various network models, would add a discussion
of their ideason technologyand autocatalysis. The notion of viewing the
economy as an autocatalyticset was just too beautiful to passup, and
Kauftnan couldn't wait to shareit.
Kauffman's Santa Fe home was a good spot for this kind of thinking and
contemplation-in its own way, almost as good as HorsetailFalls. A big
rambling structure with floor-to-ceiling picture windows, it was locatedon
a long dirt road out in the desertnorthwestof town, where it had a spec-
tacular view of the femez Mountains acrossthe Rio Grande Valley. The
settingwassomehowtimeless,almostspiritual.He had boughtthe property
lessthan a year before, for the expresspurposeof spendingmore time at
the institute.
The Santa Fe Institute, of course, was the third of his great intellectual
environments. "As exciting as Oxford and Chicago were," he says,"they
were small potatoescompared with what the institute is like. It's iust an
amazingplace." He had heardabout it from Doyne Farmerin 1985,while
they were both working on the computer model of autocatalysis.But he
had only really experiencedit in August of 1986, when he aftendeda Santa
Fe Institute workshop, "Complex Adaptive Systems," organizedby )ack
Cowan and Stanford evolutionary biologist Marc Feldman. Like Arthur,
he had just fallen in love with the place, and it took him no time at all to
decide that this is where he wanted to be. "The senseof ferment and
intellectual excitement and chaos and seriousnessand joy is here all the
'Thank
time-along with a senseof God, I'm not alone.' "
Liz Kauffman and their two children, Ethan and Merit, had been more
than happy with the idea of spending time in Santa Fe. When Kauffman
brought them out to seethe area,they had immediatelyfallen in love with
it, too. He still remembersthe day they all went out hunting for mushrooms
in the Sangrede Cristo Mountains. Moreover, Liz wasa painter, and there
wasno placein the world that had light like New Mexico. So the Kauffmans
closedon their SantaFe houseOctober 12, 1986,thinking that they might
be spendinga month or so in New Mexico every year.

It was lessthan two weekslater, on October 25, 1986, that Merit Kauff-
man was killed. And in the aftermath of her death, the Santa Fe house
suddenly became a great deal more for the Kauffrnans than a vacation
home. From then on it wasa refuge.Liz and Ethan movedthereessentialiy
full time, while Stuart Kauftnan himself became a kind of exile moving
in between, tethered to Penn only by his students,a salary, and a tenure
Secretsof the Old One 135
slot. His departmentchairman,recognizingthat it wasa matterof emotional
salvation, arrangedfor Kauffman to spendup to half his time in Santa Fe.
"lt was extraordinarilykind," he says."Not many placeswould allow you
to do that."
Kauffman sayshe remembersvery little about the following year. In May
1987 he liarned that he had been awardedone of the MacArthur Foun-
dation'sno-shings"genius" grants.He wasexhilarated-and yet he hardly
felt it. "Thlk about one ofthe worstthingsthat can happenand one ofthe
bestthings that can happen." He basicallyretreatedinto his work. "Being
a scientist,"he says,"is the placeI can go to that feelsnormal." And often
he would walk along thesedirt roadsin the desert,gazingat the mountains
and looking for the secret.
It ta'* 1 t* 1 3 t a s * a r* * f a** r **a a?a o **a *nt*f

Under ordinary circumstances,Brian Arthur didn't get nervouswhen he

gave a talk. But, then, the Santa Fe Institute's economicsconferencewas
no ordinary circumstance.
He'd had a sensethat something big was afoot before he even got there.
"When Ken stoppedme, when I startedto hear that nameslike fohn Reed
and Phil Anderson and Murray Gell-Mann were behind it, when the
presidentof the institute was on the phone-it was clear that this meeting
was being regardedby people at Santa Fe as quite a landmark event." As
organizers, Arrow and Anderson had scheduledthe meeting for ten full
days,which was very long by academicstandards.And GeorgeCowan had
called a pressconferencefor the final day-when fohn Reed himself was
supposedto be there. (lndeed, it wasa testamentthat Andersonwasplanning
to come at all. Sevenmonths earlier,in February1987, everycondensed-
matter physicist in the world had been galvanizedby the announcement
of a new class of grungyJooking ceramic materials that could conduct
electricity without resistanceat the relatively balmy boiling point of liquid
nitrogen, - 321'F. Anderson, like a greatmany other theorists,wasworking
overtime to figure out how these "high-temperature" superconductorsac-
complishedthat feat.)
But for Arthur, the real moment of revelation came when he arrived at
the institute in late August, two weeksbefore the meeting, and finally saw
the roster. He already knew about Arrow and Anderson, of course. And
he'd known about his Stanford colleague Tom Sargent, who was often
mentioned as a Nobel Prize contender becauseof his analvsis of how
"You Cuys ReallyBelieveThat?" 137
"rational" economicdecisionsin the privatesectorare intimatelyinter-
twined with the economicenvironmentcreatedby the government.But
herewerenameslike Harvardemeritusprofessor HollisChenery,a former
directorof researchfor the World Bank(andownerof the racehorse Sec-
retariat);Harvardwhiz-kidlarry Summers,who wascurrentlyservingas
economicsadviserto the Dukakiscampaign;the Universityof Chicago's
fos€Scheinkman,a pioneerin applyingchaostheoryto economics;the
BelgianphysicistDavid Ruelleof the InstitutdesHautesEtudesScienti-
6quesoutsideParis,oneof the foundersof chaostheory.Arthur wasdeeply
impressed. Therewerenearlytwodozennameson thislist, andtheywere
all of this caliber.
He could feel his adrenalinelevel beginningto climb. "l realizedthat
it might well be a crucialmomentfor me, a chanceto presentmy ideas
on increasing returnsto a groupof peopleI verymuchwantedto convince.
I instinctivelyfelt that the physicistswouldbe verymuchat homewith my
ideas.But I didn't reallyknowwhattheywouldhaveto say.Or Arrow, for
thatmatter.And althoughtheeconomisbwerehighJevel,theyweremainly
knownfor conventionaltheory.So I had no ideahow my ideaswould be
received.Therewasno contextat all. I didn't knowwhatthe tone would
be, whetherit would be the kind of bitter attackyou run into sometimes,
or whetherit would be a very friendlyformat."
So, asthe dayapproached opening-Tuesday,Septem-
for the meeting's
ber 8-Arthur spentlessand lesstime walkingand talkingwith Stuart
Kauffman,and moreand moretime polishinghis presentation. He also
remembers doinglotsof tai chi. "Thi chi teachesyou to absorbattacksand
immediately comebackwith a counterhit," he says."I thoughtI might
needthat,Forkeepingyourselfgrounded underfire, there'snothingbetter
than practicingslow-motion martialarts.Because everytime you punch
you can imaginedeliveringsomething to an audience."

The meetingconvenedat 9 e.u. in the convent'schapel-tumed-confer-

ence room. The participantswere seatedarounda long double row of
collapsibletables.And as always,the light from the westernsky came
streamingin throughthe stained-glasswindows.
After a shortintroductorytalkby Anderson,layingout someof the maior
themesthat he washopingto seeaddressed in the workshop,Arthur stood
"Self-Reinforcing
up to beginthe firstformalpresentation: Mechanisms in
Economics."And ashe beganit Arthur somehowhadthe impression that
Arrow waskeyedup, too, asif he wasworriedthat this guy Arthur might
138 COMPLEXIW
givea very peculiarpictureof economicsto the physicists. Arrow himself
doesn'trememberfeelingthatway."I knowBrian'scommunicating ability
is veryhigh," he says.And besides, sofar ashe wasconcerned the meeting
wasonly an experiment."There wasa greatdeal on the line from the
institute'spoint of view," he says,"but there was nothing on the line
intellectually.Therewasno positionto erode.If the experimentfailed, it
failed."
But whetherhe wasconsciously concernedor not, Arrow wascertainly
being his rigorouslyanalyticalself that morning. Only more so. Arthur
startedout by addressinghimselfto the physicists.
When he usedthe words
"self-reinforcing mechanisms," heexplained, hewasbasically talkingabout
nonlinearigin economics ...
"Stop!" saidArrow. "ln preciselywhat sensedo you mean nonlinear?
Aren'tall economicphenomena nonlinear?"
Well, umm, yes.To be mathematically precise,saidArthur, the ordin-
ary assumptionof decreasing returnscorresponds to economicequations
with a "second-order"nonlinearity,which drivesthe economytoward
equilibriumand stability.What he was lookingat were "third-order"
nonlinearities-factorsthat would drive some sectorof the economy
away from equilibrium. This is what an engineerwould call positive
feedback.
That seemedto satisfyArrow. And at variouspoinb aroundthe table,
Arthur could seeAnderson,Pines,and the other physicists beginningto
nod. Increasingreturns,positivefeedback,nonlinearequations-to them
it wasfamiliar stuff.
Then, abouthalfwaythroughthe morning,Phil Andersonstuckup his
handand asked,,"I!qltthe economya lot like a*spinglass?" And Arrow
:
understandably i
As it happens,Arthur hadbeenreadinga lot of condensed-matter physics
overthe pastfew yearsand knewexactlywhata spinglasswas.The name
actuallyreferredto a groupof obscuremagneticmaterialswhosepractical
utility wasnil, but whosetheoreticalpropertieswerefascinating. Anderson
hadpersonally beenstudying themsincetheywerediscovered in the 1960s,
and had coauthoredseveralof the seminalpapersin the 6eld. fust as in
more familiar magneticmaterialssuchas iron, the key componentsof a
spin glassare metalatomswhoseelectronspossess a net whirling motion,
or "spin." And justasin iron, thesespinscauseeachatomto producea
tiny magneticfield, which in tum causesit to exerta magneticforceon the
spinsof its neighbors. Unlikein iron, however, the interatomic forcesin
a qpin glassdo not causeall the spinsto fall into line with one another
"You Cuys ReallyBelieveThat?" 139
and producea large-scalemagneticfield, the kind we seein compassneedles
and refrigeratormagnets.
Instead, the forces in a spin glassare completely random-a statethat
physicistsrefer to as "glassy." (The atomic bonds in a pane of window glass
are random in the same way. Technically, in fact, it's a toss-upwhether J
ordinary glassshould be calleda solid or an exceptionallyviscousliquid') /
Among other things, this atomic-scaledisordermeansthat a spin glassis
a complex mixture of positive and negativefeedbacks,as each atom tries 1
to align its spin in parallel with certain of its neighborsand oppositeto all j
the rest. In general, there is no way to do this consistently;each atom will I
always have to endure a certain amount of frustration at having to align I
with neighbors that it doesn't want to be aligned with. But by the same I
token, there are a vast number of ways to arrange the spins so that the I
frustration is reasonablytolerable for everyone-a situation that a physicist ''lI
would describeas "local equilibrium.'
So yes,agreedArthur. In this sensea spin glasswasquite a good metaphor
for the economy. "lt naturally has a mixture of positive and negativefeed-
bacla, which gives it an extremely high number of natural ground states,
or equilibria." That's exactlythe point he'd been trying to make all along
with his increasing-returns economics.
Arthur could seethe physicistsnodding somemore' Hey, this economics
stuff was all right. "I really resonatedwith Brian," saysAnderson. "We
found ourselvesvery impressed,"
And so it went, for two solid hours:Lock-in. Pathdependence'QWERTY
and possibleinefficiencies.The origin of Silicon Valleys."AIl the time in
my talk, the physicists werenoddingand beamirig,"saysArthur. "But every
'Stop,'asking
ten minutesor so,Arrow would say, me to expandsomething,
or explaining why he didn't agree. He wanted to know exactlywhere each
step of the reasoning was coming from. And when I startedto stateprecise
on
theorems the board,'he, like several of the economists there, wanted to
see the preciseproofs. This slowed me down. But it also sewedup the
arguments."
When Arthur finally sat down at the end he was drained-but he had
the sensethat his careerwasmade. "My ideasgot legitimized that morning,"
he says."Not by me convincing Arrow and the others, but by physicists
convincing the economiststhat what I was doing was bread and butter to
'Oh yes,
them. In effect, they said, this guy knowswhat he'stalking about-
you economistsdon't have to worry.' "
Perhapsit was just projection. But to Arthur, Arrow seemedvisibly more
relaxed.
 140 COMPLEXITY

If Arthur's talk had given the physiciststhe impressionthat they were on

the samewavelengthasthe economists,however,they weresoondisabused.
For the first two or three daysof the meeting, since few of the physicists
had had much exposureto the subiect beyond undergraduateEconomics
l0l, Arrow and Andersonhad askedseveralof the economiststo givesurvey
talkson the standardneoclassicaltheory. "We were fascinatedby this shuc-
ture," saysAnderson,for whom economictheory has long been an intel-
lectual hobby. "We wantedto learn about it."
And indeed,as the axiomsand theoremsand proofsmarchedacrossthe
overheadprojection screen,the physicistscould only be aweshuckat their
counterparts' mathematical prowess-aweshuck and appalled. They had
the sameobiection that Arthur and many other economistshad beenvoicing
from within the 6eld for years. "They were almost too good," saysone
young physicist,who remembersshakinghis head in disbelief."lt seemed
as though they were dazzling themselveswith fancy mathematics, until
they really couldn't see the forest for the trees. So much time was being
spenton trying to absorbthe mathematicsthat I thought they often weren't
looking at what the models were for, and what they did, and whether the
underlyingassumptionswereany good. In a lot of cases,what wasrequired
wasjust somecommon sense.Maybeif they all had lower IQs, they'dhave
been making some better models."
The physicistshad no obiections to the mathematics itself, of course.
Physicsis far and away the most thoroughly mathematizedscience in ex-
istence. But what most of the economistsdidn't know*and were startled
I' to find out-was that physicistsare comparativelycasualabout their math.
"They use a little rigorousthinking, a little intuition, a little back-of-the-
envelopecalculation-so their style is really quite different," saysArrow,
who remembers being pretty surprised himself. And the reason is that
physicalscientistsare obsessiveabout founding their assumptionsand their
theorieson empiricalfact. "l don't know what it's like in fieldslike relativig
theory, where the ratio of theory to observationis much greater," saysArrtrw,
"but the generaltendencyis that you make a calculation,and then find
{
I some experimental data to test it. So the lack of rigor isn't so serious.The
I errorswill be detectedanyway.Well, we don'thayedata of that quality in
economics. We can't generatedata the way the physicistscan. We have to
go pretty far on a small basis. So we have to make sure every step of the
way is correct."
Fair enough. But the physicistswere nonethelessdisconcertedat how
"You Guys ReallyBelieveThat?" 141

seldom the economistsseemedto pay attention to the empirical data that

did exist. Again and again, for example, someonewould ask a question
like "What ibout noneconomic influencessuch as political motives in
OPEC oil pricing, and mass psychologyin the stock market? Have you
consulted sociologists,or psychologists,or anthropologists,or social sci-
entistsin generali' And the economists-when they weren'tcurling their
lips at the thought of these lessersocial sciences,which they considered
horribly mushy-would come back with answerslike "Such noneconomic
forcesieally aien't important"; "They are important, but they are too hard
to treat"; "ihey aren't alwaystoo hard to treatnand in fact, we're doing so
in specific cases";and "We don't need to treat them becausethey're au-
tomatically satisfiedthrough economic effects."
And then there was this businessof "rational expectations."Arthur re-
memberssomeoneaskinghim during his talk that firstday,"lsn't economics
a good deal simpler than PhYsics?" 'agents'-
i,W.ll,"
Arthur replied,"in one senseit is. We call our particles
banks, firms, consumers, governments.And those agentsreact to other
agents, just as particles react to other particles. Only we don't usually
clnsider the spatial dimension in economics much, so that makes eco-
nomics a lot simpler."
However, he added, there is one big difference: "our particles in eco-
nomics are smart, whereas yours in physics are dumb'" In physics, an
elementary particlehasno past,no experience,no goals,no hopesor fears
can talk so freelyabout
aboutthe future. It iust is. That'swhy physicists
,'universallaws":their particlesrespondto forcesblindly, with absolute
obedience. But in economics, saidArthur, "our particleshaveto think
ahead,and try to figure out how otherparticlesmight reactif they were
to undertakecertainactions.Our particleshave to act on the basisof
expectationsand strategies.And regardlessof how you model that, that's;
,,
truly difficult."
' ! f:ff---fr )

whatmakeseconomics
Immediately,he says,he could seeall the physicists in the room sitting
,.Herewasa subjectthatwasn'ttrivial.It waslike theirsubiect,but it
up:
hasthesetwo interestingquirks-strategyand expectations."
Unfortunately, the economists' standard solutionto the problemof ex-
oectations-perfect rationality-drove the nub. Perfectly
physicists rational
agentsdo haE]Eeffiftf5eing perfectly predictable.That is, they know
everythingthat can be known about the choices they will face infinitely
far into ttrefuture,andtheyuseflawless reasoning to foresee all thepossible
implicationsof their actions. So you can safelysay that they will always
takethe most advantageous action in any given situation, based on the
 142 COMPLEXITY
availableinformation.of course,theymaysometimes be caughtshortby
oil shocks,technologicalrevolutions, politicaldecisions aboutinterestrates,
andothernoneconomic surprises. But theyaresosmartandsofastin their
adjustments that they will alwayskeepthe economyin a kind of rolling
equilibrium,with supplypreciselyequalto demand.
The only problem,of course,is that real human beingsare neither
perfectlyrationalnor perfectlypredictable-asthe physicisbpointedout
at greatlength.Furthermore,asseveralof themalsopointedout, thereare
real theoreticalpidalls in assumingperfectpredictions,even if you do
assumethat peopleare perfectlyrational.In nonlinearsystems-andthe
economyis mostcertainlynonlinear-chaostheorytellsyouthattheslight-
estuncertaintyin your knowledge of the initial conditionswill often grow
inexorably.After a while, your predictionsare nonsense.
"They kept pushingus and pushingus," saysArthur. ,,The physicists
wereshockedat the assumptions the economists weremaking-that the
testwasnot a matchagainstreality,but whetherthe assumptions werethe
\ commoncurrencyof the field. I caniustseePhil Anderson,Iaid backwith
'Youguys
\ a smileon his face,saying, reallybelievethat?'"
,,yeah,
\ Th. economisb,backedinto a corner,wouldreply, but thisallows
us to solvetheseproblems.If you don't maketheseassumptions, then you
I
I can't do anything."
\ And the physicisbwould comeright back,"yeah, but wheredoesthat
fou-fou're solvinglhe wrongproblemif that'snot realig."
lCet

Economists asa grouparenotexactlynotedfor theirintellectualmodesty,

and the economisbat santaFe would havebeenlessthan human if thev
hadn'tfelt a touchof resentment aboutall this.Theywereperfectlywillini
to complainaboutthefailingsof theirfieldamongthemselves;Arrow, after
all, haddeliberatelybeenlookingfor well-informedskeptics. Butwho wanb
to hearthe samething from a bunchof outsiders? Everyonewasdoinghis
damnedest to listenand to be polite and to makethe meetingwork. And
yet therewasa distinctundercurrentof "what hasphysicsgot to offerus?
What makesyou guysso damnsmart?"
of course,physicistsaren'texactlynotedfor their intellectualmodesty,
either.For manynonphysicists, in fact, the phrasethat comesto mind is
"insufferablearrogance."It isn't a deliberateaftifudeor evenpersonal.It's
morelike the unconscious superiorityof the Britisharistocracy.Indeed,in
istocracyof science.From the diil
theysignup for Physics -ifr
a"thousandsubtle
"You Guys ReallyBelieveThat?" 143

and not-so-subtleways:they are the heirs of Newton, Maxwell, Einstein,

and Bohr. Physicsis the hardest, purest, toughest science there is' And
physicistshave the hardest,purest, toughestminds around. So if the econ-
omists startedoff the Santa Fe conferencewith a certain touchiness,their
counterpartsmatched it with what Harvard economist Larry Summers
dubbed the "Me Thrzan, You ]ane" attitude: "Give us three weeksto master
this subiectand we'll show you how to do it right."
The potential clash of egoswas a constantconcern of Citicorp's Eugenia
singer, who was sifting in as fohn Reed'srepresentative."I was afraid that
'Tarzan' effect got started, then this whole proiect would be down
if the
the tubesbeforeit evergot offthe ground," sherecalls.And in the beginning
it looked as though it might. "Most of the economistssat on one side of
the table, and most of the physical scientistssat on the other side," she
says."I was horrified by it." She would periodically pull Pines or Cowan
aside and say, "Couldn't we get them to sit a little closer to each other?"
But it never changed.
The potential for total miscommunication was likewise a nightmare for
George Cowan-and not just becausethe institute might lose any hope of
Citicorp funding if the conference failed to come off, The fact was that
this meeting was the most stringent proof of concept yet for the Santa Fe
Institute. In the founding workshopstwo years earlier they had brought
people together for a weekendof talk; but now they were asking two very
iiffere"t and uety proud groups to sit down with each other for ten days
and do something substantive."We were trying to createa communi$ that
didn't exist before," saysCowan. "The odds were that it wouldn't be suc-
cessful, that they'd find nothing to talk about, that there would simply be
polemics."
This was not an idle fear; later Santa Fe workshopshave occasionally
degeneratedinto shouting matchesand sulking. But in September1987,
the godsof interdisciplinaryresearchdecidedto smileonceagain.Anderson
and Arrow had tried hard to recruit peoplewho could listenaswell astalk.
And for all the initial bristling,the participantseventuallybeganto discover
that they had plenty to talk about. Looking back on it, in fact, the two
sidesbeganto find common ground in a remarkablyshort time.
Certainly that was true for Arthur. In his case,the discoverytook about
half a day.
f i*3rt3 |'**3$ $* *cst* trSt*,t *a 6t * a a. aafaa

Accordingto the agenda,the secondpresentation of the economicswork-

shop would begin after lunch on that first day and run the rest of the
afternoon."The Global Economyasan Adaptiveprocess,"it wascalled,
by lgblL_Hslbdgf the Universityof Michigan.
Now that BrianArthur had finishedhis own talk, he had the energyto
feel curiousaboutthis one-and not iust because the title soundedi-nter-
esting.fohn Hollandwasanotherof theinstitute's visitingfellowsthatfall,
and the two of them weresupposed to be sharinga house.But Holland
hadn't arrivedin santa Fe until late the night before,whire Arthur was
downat the conventgoingoverand overhis talk justonelasttime. Arthur
had never laid eyeson the man. All he knew wasthat Hoiland wasa
computerscientistand, accordingto the institute,"a very nice guy."
the instituteappearedto be right aboutthat.As peoplewerefiling
. {ell,
backinto the chapeland settlinginto their chairsaroundthe long row of
foldingtables,Hollandwasup front gettingreadyto start.He prou-.dto b.
a compact,sixtyishmidwesterner with a broad,ruddy facethat seemed
6xedin a perpetualgrin, and a high-pitchedvoicethat madehim sound
like an enthusiastic graduatestudent.Arthur liked him instantly.
But then Hollandbeganhis talk. And within minutes,Arthur wasn't
just sleepilyfollowingalong.He waswide awakeand listeninghard.
Masterof the Came 145

Perpetual Novelty

Holland startedby pointing out that the economyis an examplepar

'
excellence of whatthe SantaFeInstitutehadcometo g*P!.* gd"Plllg\
""ll
systems." In the naturalworldsuchsystems includedbrains,immunesys- I
iffitogies, cells,developing embryos, andantcolonies. In thehuman l
world they includedculturaland socialsystems suchaspoliticalpartiesor I
scientificcommunities.Onceyou learnedhow to recognize them, in fact, I
thesesystems wereeverywhere. But whereveryou found them, saidHol- |
land,theyall seemed to sharecertaincrucialproperties.
First, he said, each of thesesystems is a networkof many"agents"acting
in parailel. In a brain the agents are nerve cells,in an ecoToglifiiagents
arespecies,in a cell the agents areorganelles suchasthe nucleusand the
mitochondria, in an embryo the agents are cells,andsoon. In an economy,
the agentsmight be individualsor households. Or if you werelookingat
businesscycles,the agentsmight be firms. And if you were looking at
international trade,theagents mighteven bewhole nations.But regardless
of howyoude6nethem,eachagent finds itselfin an environment produced
by its interactions with the otheragents in the system. It is constantly acting
and reactingto what the other agents are doing. And because of that,
essentially nothingin its environmentis fixed'
Furthermore,said Holland, the control of a complexadaptivesystem
tendsto be highly dispersed. Ttr*S-hjoqler neuronin the brain, for
example,nor L th.r. *tt[[iiwithinffi&lQembrvo. If there
""y competition
is to beanycoherentbehaviorin thesystem,it hasto arisefrom
and cooperationamongthe agentsthemselves. This is true even in an
economy.Ask any presidenttrying to copewith a stubbornrecession: no
matterwhat Washingtondoesto fiddle with interestratesand tax policy
andthe moneysupply,the overallbehaviorof theeconomyis still the result
of myriadeconomicdecisions madeeverydayby millionsof individual
people.
-
Slcond, said Holland, a complexadaptivesystemhas manv levelsof
organization, with agentsat any one levelservingasthe buildingblocks
f6i ag66;a higherlevel.A groupof proteins,lipids, and nucleicacids
will form a cell, a groupof cellswill form a tissue,a collectionof tissues
will form an organ,an association of organswill form a wholeorganism,
and a groupof organisms will form an ecosystem. In the brain, one group
of neuronswill form the speechcenters,anotherthe motor cortex,and
still anotherthe visualcortex.And in precisely the sameway,a groupof
individualworkerswill compose a department, a groupof departments will
 146 COMPLEXITY
composea division, and so on through companies,economic sectors,
national economies, and finally the world economy.
Furthermore,saidHolland-and this wassomethinghe consideredvery

ranging their
rangrng !!err building blocksas they gain experie1199.
ations of organismswill modify and rearrangetheir
processof evolution. The brain will continually strengthenor weakenmyr-
iad connectionsbetweenib neuronsas an individual learnsfrom his or
her encounterswith the world. A firm will promote individuals who do
w_elland (more rarely) will reshuffie its organizational chart for greater
efficiency. countries will make new trading agreementsor realignlhem-
selvesinto whole new alliances.
At some deep, fundamental level, said Holland, all these Drocesses of
Iearning, evolution, and adaptationare the same. And on. of the funda-
mental mechanismsof adaptationin any given systemis this revisionand
recombinationof the building blocks.
Third, he said, all complexadaptivesystemsanticipatethe future. Ob-
viously,this is no surpriiet" th. ;;oilmi. TE6"a'ntftb;tion
of an ex-
tendedrecession,for example,may lead individualsto deferbuying a new
car or taking an expensivevacation-thereby helping guarantee that the
recessionwill be extended.The anticipationof an oil shortagecan likewise
send shock wavesof buying and selling through the oil ma*ets-whether
or not the shortageever comes to pass.
But in fact, said Holland, this businessof anticipationand prediction
goes far beyond issuesof human foresight, or even consciousness.From
bacteriaon up, every living creaturehas an implicit prediction encodedin
its genes:"In such and such an environment, the organismspecifiedby
this geneticblueprint is likely to do well." Likewise,everycreaiurewith a
brain has myriad implicit predictions encoded in what it has learned: "In
situationABC, action XYZ is likely to pay off."
More generally,said Holland, every complex adaptivesystemis con-
stantly making predictions based on its various internal models of the
world-its implicit or explicit assumptionsabout the way things are out
there. Furthermore,thesemodelsare much rnore than passiveblueprints.
They are active. Like subroutinesin a computer program, they can come
to life in a givensituationand "execute,"producingbehaviorin the system.
In fact, you can think of internalmodelsasthe building blocksof behavior.
And like any other building blocks, they can be tested,re6ned, and rear-
rangedas the systemgainsexperience.
 Game
the Game
Masterof the
Master 147

Finally, said Holland, complexadaptivesystemstypicallyhave,many

niches,."cho.,eofwhich canbe exploitedby an agentadaptedto fill that
ffifrrTlur, the economicworldhasa placefor computerprogrammers,
plumbers,steelmills, and pet stores,iust asthe rain foresthasa placefor
treeslothsandbutterflies.Moreover,theveryactof filling onenicheopens
up more niches-for new parasites, for new predators and prey,for new
symbioticpartners.So new opportunities are alwaysbeing createdby the
,yrt *. And th"t, in turn, means that it's meaningless
essentially to talk
abouta complexadaptivesystem being in eqlriJibriumjthe-syster-n never
"can
eetthers. Itis alwaysunfolding,always in hansition'In fact, if the system
tEiffir t.rch equilibrium,iiisn't iuststable.It's dead.And by the same
token,saidHolland, there'sno point in imaginingthat lhe ?gentsin the
systemcan ever"optimize"their fihress,or their utility, or whatever.The
,p""e of possibilities is too vast;they h-a@he
optimum.The mosttheycaneverdo is to changeand improvethemselvds
,.trtiu" io what the other agentsare doing. In short, complexadaptive
systems are characterized by perpetualnovelty'
Multiple agents,building blocks,internal models,perpehralnovelty-
takingali this together,saidHolland,it's no wonderthat complexadaptive
syste,is*.re ,o hard to analyzewith standardmathematics. Most of the
conventional techniqueslike calculusor linearanalysis areverywell suited
to describeunchangingparticlesmoving in a fixed environment'But to
reallygeta deepunderstanding of theeconomy'or complexadaptive systems
in gen-eral, whatyou needaremathematics andcomputersimulationtech-
niqiresthat empirasizeinternal models,the emergence of new building
blocks,and the rich web of interactions between multiple agents'

By this point in Holland'stalk, Arthur wasscribblingnotesfuriously.

And as Holla"a went on to describethe variouscomputertechniqueshe
had developedoverthe pastthir$ yearsto maketheseideasmoreprecise
and moreuseful,Arthur scribbledevenfaster."It wasincredible,"he says.
"l iust satthereall aftemoonwith my mouth open." It wasn'tsimplythat
Holland'spoint aboutperpetualnovel$ wasexactlywhathe'dbeentrying
to sayfor the pasteight yearswith his increasing-returns economics.Nor
wasit that Holland's point about niches was exactlywhat he and Stuart
Kauffmanhad beenthrashingout for the pasttwo weeks in the contextof
autocatalytic sets.It wasthat Holland's whole wayof looking at thingshad
that
a unity, a clarig, a /rghtnerls made you slapyourforehead andsay,"Of
 148 COMPLEXITY
course!Why didn't I think of that?"Holland'sideasproduceda shockof
recognition,the kind that mademore ideasstartexplodingin your own
brain.
"Sentence by sentence,"
saysArthur,"Hollandwasanswering all kinds
of questionsI'd beenaskingmyselffor years:What is adaptatibn? What is
emergence? And manymorequestions thatI neverrealizedI'd beenasking.',
Arthur had no ideayet how it wouldall fit into economics.In fact, ashe
lookedaroundthe roomhecouldseethata numberof hisfelloweconomists
werelookingpuzzled,too, if not downrightskeptical.(At leastone was
takinga little midafternoon "But I wasconvinced
siesta.) thatHollandwas
ontosomethingmuch, muchmoresophisticated thanwhatwedid." Some-
how, he felt, Holland'sideashad to be incrediblyimportant.

certainly the SantaFe Institutethoughtso. Howevernew Holland's

thinkingmight haveseemed to Arthur and the othervisitorsat the eco-
nomicsmeeting,he had long sincebecomea familiarand profoundry
influentialfigureamongthe SantaFe regulars.
His 6rstcontactwith the institutehadcomein 1985duringa conference
entitled"Evolution,Games,andL,earning," whichhadbeenorganized at
[.osAlamosby Doyne Farmerand NormanPackard.(As it happens,this
wasthe samemeetingin which Farmer,Packard,and Kauffmanfirst re-
portedthe resultsof theirautocatalyticsetsimulation.)
Holland'stalkthere
wason the subjectof emergence, and it seemedto go quite well. But he
remembers beingpeppered with sharp-edged questions
fromthispersonout
in the audience-a white-hairedguy with an intent, slightlycynicalface
peeringout from behinddark-rimmed glasses.
"l wasfairly flip in my
answers," saysHolland."I didn't knowhim-and I'd probablyhavebeen
scaredto deathif I hadt"
Flip answersor not, however,Murray Gell-Mann clearly liked what
Hollandhadto say.Shortlythereafter, Gell-Manncalledhim up andasked
him to serveon what wasthen calledthe santa Fe Institute'sadvisory
board,which wasjustbeingformed.
Hollandagreed."And assoonasI sawthe place,I reallylikedit," he
says."What they were talking about, the way they went at things-my
immediateresponse was,'l surehopetheseguyslike me, becausethis is
for me!'"
The feelingwasmutual. When Gell-Mannspeaks of Holland he uses
wordslike "brilliant"-not a term he throwsaroundcasually. But, then,
it's not oftenthat Gell-Mannhashad his eyesopenedquiteso abruptly.
Masterof the Game 149
In the early days, Gell-Mann, Cowan, and most of the other founders of
the institute had thought about their new science of complexi$ almost
entirely in terms of the physical conceptsthey were alreadyfamiliar with,
such as glqg collective behavior, attd s2gltrpeous-orSanlz.419n.
M oreovdlJher. 6-,,cepili t.aSTE--rned to prdmrse an excePttonat ty rrch
researchagenda,if only as metaphorsfor studying the same ideas-emer-
gence, collective behavior, and spontaneousorganization-in realms such
as economicsand biology. But then Holland came along with his analysis
ofdgplgltg!-not to mention his working computer models' And suddenly I
Gell-Mann and the others realized that they'd left a gaping hole in their I
agenda: -do} How do theY re- I
spondand adaptto their environment? |
Within months they were talking about the institute'sprogram being not
just complex systems,but complex adaptive systems.And Holland's per-
sonal intellectual agenda-to undersiand the intertwining processesof
emergenceand adaptation-essentiallybecamethe agendaof the institute
as a whole. He was accordingly given star billing at one of the institute's
first attemptsat a large-scalemeeting, the Complex AdaptiveSystemswork-
shop organizedin August 1986by fack Cowan and Stanfordbiologist Marc
Feldman. (This wasthe samemeetingthat introducedStuartKauffman to
SantaFe.) David Pineslikewisemade a point of bringing Holland out to
talk to lohn Reedand companyduring the encounterat Rancho Encantado,
which overlappedthe Complex Adaptive Systemsworkshopby a day. And
Anderson saw to it that Holland was in attendanceat the big economics
meeting in September1987.
Holland participated in all this cheerfully-as well he might. He had
labored on his ideasabout adaptationin relative obscurity for a quarter of
a century.And only now, at agefifty-seven,washe beingdiscovered."The
ability to talk one-on-oneand be treatedasan equal with the likesof Murray
Gell-Mann and Phil Anderson-that's great!Incredible!"If therehad been
any way for his wife to leave Ann Arbor-Maurita Holland was head of
the university'ssystemof nine sciencelibraries-he would havebeenspend-
ing a lot more time in New Mexico than he alreadywas'
But, then, Holland wasalmost alwayscheerful. He possessed the guileless
good humor of a genuinely happy rnari who was doing what he genuinely
wanted with his li-fe*and who still seemedamazedat his good fortune. It
was essentiallyimpossiblenot to like fohn Holland.
Arthur, for one, didn't even hy to resist. After Holland's sessionwas
over that first afternoon, he eagerly introduced himself. And over the
ensuing days, the two men quickly becamefast friends. Holland found
 150 COMPLEXITY
Arthur a delight."Fewpeoplecould havepickedup on theseideasabout
adaptation andthenintegrated themsoquicklyandsothoroughlyinto their
own outlook," h€ says."Brian wasvery interested in all partsof it, very
quickto explore."
Arthur, meanwhile,foundthat Hollandwaseasilythemostcomplexand
fascinating intellectthat he hadcomeacrossissrntiFe. Indeed.Holland
wasone of the main reasons tb4-h. spentthe remainingdaysof the
economicsconference in-a-s6l6-ofchronicsleepdeprivation.He and Hol-
-$.g5! spent ryan'"dal; night sittingaroundthe kitchentableat their shared
bggsE-drliiking beer, and discussing the whichness of what.
He remembers one suchconversation in particular.Holland had come
to theconference eagerto learnwhatthecrucialissues werein economics.
("lf you'regoingto do interdisciplinary studiesand entersomeoneelse's
domain,"Hollandsays,"the leastyou shoulddo is taketheir questions
very seriously.They'vespenta long time formulatingthem.") And that
night, as the two of them weresittingaroundthe kitchentable,Holland
put the questionto him straight:"Brian, what is the real problemwith
economics?"
."Chess!"repliedArthur,withoutthinking.
'Cffi
nof.nd didn't understand.
Well, saidArthur,takinga sipof beerandstumbling forwords.He didn't
quiteknowhimselfwhathe meant.Economists werealways talkingabout
systems thatweresimpleandclosed,in thattheywouldquicklyseftledown
into one or two or threesetsof behavior, -and after that, nothing much
happens. Theywerealsotacitlyassuming thateconomicagents areinfinitely
smartandcaninstantlyperceive thebestthingto do in anygivensituation.
But think aboutwhatthat meansin termsof chess.In the mathematical
theoryof gamesthereis a theoremtellingyou that anyfinite, two-person,
zero-sumgame-such aschess-hasan optimalsolution.That is, there
is a way of choosingmovesthat will allow eachplayer,blackand white,
to do betterthan he wouldwith any otherchoiceof moves.
In reality,of course,no one hasthe slightestideawhatthat solutionis
or howto find it. But oneof theseidealeconomicagentsthattheeconomists
talk about could find it instantaneously. Confrontedwith the chessboard
at the start of the game,two such agentswould just formulateall the
possibilities in their mind, and work backwardfrom all the possibleways
you could forcea checkmate. Then theywouldworkbackwardagainand
again,until theyhadconsidered all possible
movesandfoundthe optimal
moveto makeat theopening.Say,pawnto king4. And at thatpoint,there
would be no needto actuallyplay the game.Whicheverplayerheld the
Master of the Game 151

theoretical advaniage-say, white-would immediately claim victory,

knowingthat he would alwayswin. And the otherplayerwould immediately
concededefeat, knowing that he would alwayslose.
"Now, fohn," said Arthur, "does anybodyplay chesslike that?"
Holland just laughed. As it happenshe knew preciselyhow absurd this
was. Back in the 1940s, when computerswere new and researcherswere
making their first attempts to design an "intelligent" program that could
play chess, the father of modern information theory, Claude SJrannonof
b.il L"br, had estimatedthe total number of possiblelrno6in-cT6ffiis
answer, 10120, was a number so vast as to defy all metaphor' There haven't
been thaiTiany microsecondssince the Big Bang. There aren't that many
elementary particles in the observableuniverse. There is no conceivable
computerthat could examineall of thosemoves.And there is certainlyno
human being who could. We human playershave to make do with rules
of thumb-hardlearned heuristic guidesthat tell us what kind of shategies
will work best in a given situation. Even the greatestchessmastersare
alwaysexploring their way in chess,as if they were descendinginto a deep,
deep set of caveswith a tiny lantern. Of course,they do make progress.As
a chessplayer himself, Holland knew that a grand master from the 1920s
wouldn't standa chanceagainsta contemporarygrandmastersuch asGary
Kasparov.But even so, it's as if they had only gotten a few yardsdown into
this immense unknown. That's why Holland would call chessa funda-
mentally "open" system:it is effectively infinite.
Right, said Arthur. "The kinds of patternsthat people can actually per-
ceiveto work on arevery limited comparedwith what's'optimal.'Youhave
to assumeyour agenb are a lot smarterthan the averageeconomist." And
yet, he said, "that's the way we carry out economic problems. Trade with
fapan is at least as complicated as chess.But economistswill start out by
'Assume rational play.' "
saying,
So there you have the economic problem in a nutshell, he told Holland.
How do we make a scienceout of imperfectly smart agentsexploring their
way into an essentiallyinfinite spaceof possibilities?
"A hal" said Holland, the way he alwaysdoeswhen he finally seesthe
light. Chessl Now this was a metaphor he could understand.

The lmmenseSpaceof Possibilities

fohnHollandlovedgames.All kindsof games.

He hadbeena regular
at onemonthlypokergamein Ann Arborfor nearlythirtyyears.Oneof
 152 COMPLEXITY
his earliest memories was of watching the grown-ups play pinochle at his
grandparents' houseand wishing he wasbig enoughto sit at the table. He
learned to play checkersin first grade from his mother, who was also an
expert bridge player. Everyone in the family was a passionatesailor, and
both Holland and his mother frequently competed in regattas.His father
was a first-classgymnast-Holland himself spent severalyears at that in
junior high school-and an avid outdoorsman.Holland'sfamily wasalways
playingsomething.Bridge,golf, croquet,checkers,chess,Go-you name
it.
And yet somehow,very early on, this businessof gamesbegan to be
I
I more than just fun for him. He beganto notice that certaingamesheld a
| rcculiar fascination,a magicthat went well beyondanyquestionof winning
I or losing. Back when he was a freshmanin high school, for example-it
must have been about 1942 or 1943, when his family was living in Van
Wert, Ohio-he and a couple of his buddiesspenta lot of time down in
the basementof Wally Purmort'shousemakingup brandnew games.Their
masterpiece,inspiredby the daily headlines,wasa war gamethat covered
most of the basementfloor. They had tanks and artillery. They had firing
tablesand rangetables.They even had waysofcovering partsofthe board
to simulate smoke screens."It got pretty intricate," saysHolland. "l re-
member using the mimeograph machine at my dad's office to run off lots
of pieces."(The elder Holland had prosperedthrough the Depressionby
founding a stringof soybeanprocessing plantsthroughoutthe Ohio soybean
belt.)
"We wouldn't have said it this way," he says."But we did it becauseall
three of us were interestedin chess.Chesswas a game with just a small
number of rules. And yet the incredible thing to us was that you never
playedthe same game twice. The possibilitieswere just infinite. So we
,would try to invent new gamesthat had that quality."
I In one way or another,he laughs,he'sbeenmakingup gameseversince.
"l iust love thesethings where the situationunfoldsand I say,'Gee whiz!
|
I Did that really come from theseassumptions!?' Becauseif I do it right, if
the underlyingrulesof evolutionof the themesare in control and not me,
then I'll be surprised.And if I'm not surprised,then I'm not very happy,
becauseI know I've built everythingin from the start."
Nowadays,of course,this sort of thing is called"emergence."But long
before Holland had ever heard the word, his fascination with it had led
him into a lifelong love affair with scienceand mathematics.He couldn't
get enough of either one. All during school, he says,"l remembergoing
to the library and getting hold of any book I could on scienceand tech-
Master of the Came 153

nology. By the time I was a sophomorein high school, I was determined

to become a physicist." What captivatedhim wasn't that scienceallowdf;
you to reduceeverythingin the universeto a few simple laws. It was iust \
the opposite:that scienceshowedyou how a few simplelawscould produce \
the enormouslyrich behaviorof the world. "lt reallydelightsme," he says. I
"science and math are the ultimate in reduction in one sense.But if you
turn them on their heads,and look at the syntheticaspects,the possibilities
for surpriseare iust unending. It's a way of making the universecompre-
hensible at one end and forever incomprehensibleat the other end."

At MIT, where Holland arrived as a freshmanin the fall of 1946, it

didn't takehim long to discoverthat samequality of surprisein computers.
"l really don't know where this came from, either," he says,"but even early
'thought processes.'Andthe way you could
on I was alwaysfascinatedby
put just a few bits of data into a computeland then have it do all these
things like integration, and so on-it iust seemedto me that you were
getting an awful lot out for an awful litde in."
At first, unfortunately, there was very little that Holland could learn
about computersother than the secondhandbits and pieceshe picked up
in his electricalengineeringclasses.Electronic computerswere still very
new then, and mostly classified.There were certainly no computer science
coursesto take, even at MIT. One day, however,as he wasbrowsingin the
MIT library-he did a lot of that-Holland cameacrossa setof looseJeaf
lecture notes bound in a simple thesiscover. And as he thumbed through
it, he discovered that the notes contained a detailed account of a 1946
conferenceat the Moore Schoolof ElectricalEngineeringat the Universi$
of Pennsylvania,wherea wartime effortto calculateartillery tableshad led
to the development of the first digital computer in the United States,the
ENIAC. "Those notesare famous," saysHolland. "This wasthe first series
of really detailedlectureson digital computing. They went all the way from
what we would now call computer architecturesright up to software." Along
the way, the lectures dealt with such brand-new conceptsas information
and information processing,and defined a whole new mathematical art:
programming. Holland immediately bought his own copy of the lectures-
in fact, he still has it-and read it cover to cover. Severaltimes.
Then in the fall of 1949, as Holland was startinghis senior year at MIT
and casting about for a topic for hit !3gbglgrdisertation, he found out
about the Whirlwind Proiect MIT's effort to build a "real-time" computer
fast enough to keep track of air traffic. Funded by the Navy at the then-
154 COMPLEXITY
astoundingrate of $l million per year,Whirlwind employedsomeseventy
engineersand technicians,and was by far the largestcomputerproject of
its day. It was also to be among the most innovative. Whirlwind would be
the first computer to use magnetic core memory and interactive display
screens.It would give birth to computer networksand multiprocessing
(running more than one programat once).And as the first real-timecom-
puter, it would pave the way for the use of computersin air-traffic control,
industrialprocesscontrol, ticket reservationsystems,and banking.
But when Holland 6rst heard of it, Whirlwind was still very much an
experiment."l knew Whirlwind was there," he says."It wasn't finished
ye! it was still being built. But you could use it." Somehow,he just had
to get in on this. He startedpoundingon doors.In the electricalengineering
departmenthe found a Czech astronomernamedZednek Kopal, who had
been his instructor in numerical analysis."l convincedhim to chair my
dissertationcommittee. I convincedthe physicsdepartmentto let somebody
from electricalengineeringchair it. And then I convincedthe Whirlwind
peopleto give me accessto their manuals-which were classified!
"That was probably my happiestyear at MIT" he says.Kopal suggested
that, as a topic for his dissertation, he write a program that would allow
Whirlwind to solveLaplace'sequation, which describesa varietyof physical
phenomena ranging from the distribution of electric 6elds around any
electrically chargedobiect, to the vibration of a tightly stretcheddrumhead.
Holland went right to it.
It wasn't exactlythe easiestsenior thesiseverundertakenat MIT. In those
daysno one had ever heard of programming languagessuch as Pascal,or
C, or FORTRAN. Indeed, the very concept of a programming language
wasn'tinventeduntil the mid-1950s.Holland had to write his programin
somethingcalledmachine language,in which commandsto the computer
were encodedas numbers-and not even ordinary decimal numbers, at
that. These were numbers written in hexadecimalnotation, base 16. The
project took him longer than he'd thought; he eventually ended up asking
MIT for twice the time usually allotted for a senior thesis.
And yet he lovedit. "I liked the logicalnatureof the process, " he recalls.
"Programming had some of the same favor as math: you take this step,
and then that lets you take fhe next step, and so on." But more than that,
writing his program for Whirlwind showed him that a computer needn't
be just a fantasticallyfastadding machine. In his arcanecolumns of hexa-
decimal numbers, he could envision a vibrating drumhead, a convoluted
elechic field, or anything else he wanted. In a world of circulating bits he
Master of the Came 155

could createan imaginary universe.All he had to do was to encodethe

proper laws, and everything else would unfold.

Holland never actuallyran his programon the Whirlwind, since it had

been intendedasa papei exercisefrom the beginning.But his seniorthesisI
paid offhandsomely in anotherway:it made him one of the very few Wonle
in the country who knew anything at all about programming' And as I
result, he was snatchedup by IBM as soon as he graduatedin 1950'
The timing couldn't have been better. At its plant in Poughkeepsie,New
York, the giant office equipment manufacturerwas designingits first com-
mercial computer: the Defense Calculatoq eventually to be renamed the
IBM 701, At the time th;;acTine6preiented a maior and rather dubious
;fibl. for the company; many of the oldJine executivesconsideredcom-
putersto be a wasteof money better spenton developingbetterpunch-card
machines.In fact, the product planning departmentspentthe entire year\
of 1950 insistingthat the marketwould neveramount to more than about I
eighteen computers nationwide. IBM was going ahead with the Defense /
Calculator largely becauseit was the pet proiect of a Young Turk known
as Tom funior, son and heir apparentof the company's aging president,
Thomas'8. Watson, Sr.
But the twenty-one-year-oldHolland knew little of that. What he did
know was that he had been transportedto the land of Oz. "Here I was, a
really young guy at a pime place. I was one of the few people to know
what was even going on with the 701." The team leadersat IBM put
Holland on its seven-manlogical planning group, which was in charge of
designingthe instructionsetand generalorganizationof the new machine.
This turned out to be yet another stroke of good luck, becauseit was an
ideal spot to exercisehis programming skills. "After we got through the
initial stagesand had a proto$pe machine, it had to be testedin various
ways,"he says."So the engineerswould work on it all day,tearingit apart,
and then putting it back together again as best they could in the evening.
Then a few of us would go in about ll p.rr,r.and run our stuffall night,
iust to seeif it would work."
Indeed it did work, sort of. By today'sstandards,of course, the 70I was
right out of the Stone Age. It had an enormous control panel full of dials
and switches, with no sign of a video monitor. It performed input and
output via an lBM-standardpunch-cardmachine' It boasteda full 4 kilo-
bytes of memory. (Personalcomputers sold today often have a thousand
 156 COMPLEXITY
timesasmuch.)Andit cquldmultiplytwo numbersin only J0 microsec-
onds.(Almostanymodernhandcalculator cando better.)"It alsohadlots
of quirks," saysHolland. "The meantime betweenfailureswasabqULJQ_
minutesat best,sowe ran everytH"ing@its
'eETy*grcneraifng
spotsof light on the faceof a specialcathode-ray tube.
iSo Holland and his fellow programmers had to tailortheir algorithms to
lavoid writing data too often at the same location in memory; otherwise,
I chargewouldbuild up on the faceof the tubeanddistortnearbybits."lt
wasamazingwe couldgetthe machineto workat all," he laughs.But the
factwas,he didn't care."So far aswe wereconcerned, it seemed like a
giant. We thoughtit wasgreatto havetime on a fastmachineto hy out
our stuff."
There was no shortageof stuff to try out. Those headS early days of
computers were a ferment of new ideas about information, cybemetics,
lyfsmlfl-soncepts that hadn't even existedten yearsearlier. Who knew
where the limits were?Almost anything you tried was liable to break new
ground. And more than that, for the more philosophicallyminded pioneers
like Holland, these big, clumsy banls of wire and vacuum tubes were
opening up whole new waysto think about thinking. Computers might not
be the "Giant Brains"of the more lurid SundaySupplements.In the details
of their structureand operation,in fact, they weren'tanythinglike brains
at all. But it was very tempting to speculatethat computers and brains
might be alike in a deeperand much more important sense:they might
both be information processingdevices.Becauseif that were the case,then
thought itself could be understoodas a form of information processing.
At the time, of course, nobody knew to call this sort of thing "artificial
intelligence"or "cognitivescience."But evenso, the very act of program-
ming computers-itself a totally new kind of endeavor-was forcing people
to think much more carefully than everbeforeabout what it meant to solve
a problem. A computer was the ultimate Martian: you had to tell lt every-
thing: What are the data? How are they transformed?What are the steps
to get from here to there?Thosequestions,in turn, led very quickly to
issuesthat had b_edeviled-philosopherslgr centuriet What is knowledge?
How is it acquired sense impressions?
How is it represented in the
mind?How is it modifiedthroughexperience? How is it usedin reasoning?
How aredecisions transformed into action?
The answers werefar from clearat the time. (ln fact, they arefar from
clearnow.) But the questions werebeingaskedwith unprecedented clarig
andprecision.And IBM'sdevelopment teamin Poughkeepsie, havingsud-
denly becomeone of the premiercollectionsof computertalent in the
Master of the Came 157

country, was in the forefront. Holland fondly remembershow a "regular

irregular"groupwould meet in the eveningseverytwo weeksor so to thrash
out the issuesover a game of poker or Go. One attendeewas a summer
intern named |ohn McCarthy, a young Caltech graduatewho Iater became
one of the founding gurus of artificial intelligence.(In fact, McCarthy was
the one who invented the phrasein 1956to advertisea summer conference
on the subjectat Dartmouth College.)
Another was Arthur Samuel, a soft-spoken,for$ish electricalengineer
who had been brought in by IBM from the Universi$ of Illinois to help
the companyfigure out how to make reliablevacuum tubes,and who was
one of Holland's most frequentcompanionsduring his all-night program-
ming marathons.(He alsohad a daughterin nearbyVassar,whom Holland
dated severaltimes.) Samuel frankly didn't give a damn about vacuum
tubesanymore. For the pastfive yearshe had been trying to write a program
that could play checkers-and not only play the game, but learn to play it
betterand betterwith experience.In retrospect,Samuel'scheckerplayeris
consideredone of the milestonesof artificial intelligenceTeslfrch;-bythe
time he finally finished revisingand refining it in 1967, it was playing at
a world championshiplevel. But even in the 701 days, it was doing re-
markably well. Holland remembersbeing very impressedwith it, partic-
ularly with ib ability to adapt its tacticsto what the other playerwas doing.
In effect, the program was making a simple model of "opponent" and using
that model to make predictions about the bestline of play. And somehow,
without being able to articulate it very well at the time, Holland felt that
this aspect of the checker player captured something essentialand right
about learning and adaptation.

He filed that thought away in his mind as something to mull over. For
the moment he wasmore than busyenoughwith his own proiect:an attempt
to simulate the inner workings of the brain itself. It startedin the spring of
1952,he remembers,ashe waslisteningto J.C.R. Licklider,a psychologist
from MIT who had come for a visit to the Poughkeepsielab and who had
agreedto give the group there a lecture on what wasthen one of the hottest
topics in the field: the new theoriesof learning and memory being advanced
by neurophysiologistDonald O. Hebbpf McGill Universi$ in Montreal.
'The
probiem was this, Licklidffi,i-explained: Througtr a microscope,
most of the brain appearsto'be a study in chaos, with each nerve cell
sendingout thousandsof random filamentsthat connect it willy-nilly to
cells.And yetthisdenselvinterconnected
network
W$ifp!k1.1:T
158 COMPLEXITY
is obviouslynot random.A healthybrain producesperception,thought,
and actionquite coherently.Moreover,the brain is obviouslynot static.It
refinesand adaptsits behaviorthroughexperience. It learns.The question
is, How?
Jhree yearsearlier,in 1949,Hebbhad publishedhis answerin a book
entitledTheOrganizationof Behavior.His fundamentalideawasto assume
that the brain is constantlymakingsubtlechangesin the "synapses," the
pointsof connectionwherenerveimpulsesmakethe leapfrom onecell to
the next. This assumptionwasa bold moveon Hebb'spart, sinceat the
time he had no evidencefor it whatsoever. But havingmadeit, he argued
thatthesesynapticchanges werein factthebasisof all learningandmemory.
A sensoryimpulsecomingin from the eyes,for example,would leaveits
traceon the neuralnetworkby stren$heningall the synapses thatlay along
its path. Much the samething would happenwith impulsescomingin
from the earsor from mentalactivityelsewhere in the brain itself.And as
a result,saidHebb, a networkthat startedout at randomwould rapidly
organizeitself. Experiencewould accumulatethrougha kind of positive
feedbachthe strong,frequentlyusedsynapses wouldgrowstronger,while
theweak,seldom-used synapses wouldatrophy.The favoredsynapses would
eventuallybecomesostrongthat the memorieswouldbe lockedin. These
memories,in turn, would tend to be widely distributedover the brain,
with eachone corresponding to a complexpatternof synapses involving
thousandsor millions of neurons.(Hebbwasone of the first to describe
suchdistributedmemoriesas"connectionist. ")
But therewasmore. In his lecture,Lickliderwenton to explainHebb's
secondassumption: that the selectiveshengthening of the synapses would
causethe brain to organizeitselfinto lsslt4remhlig!"-sgfssg of several
thousandneuronsin which circulatingnerveimpulseswould reinforce
themselves andcontinueto circulate.Hebbconsidered thesecell assemblies
to be the brain'sbasicbuildingblocksof information.Eachone would
correspond to a tone, a fash of light, or a fragmentof an idea. And yet
theseassemblies wouldnot bephysicallydistinct.Indeed,theywouldover-
lap, with any givenneuronbelongingto severalof them. And becauseof
that, activatingone assemblywould inevibbly lead to the activationof
others,so that thesefundamentalbuildingblocls would quicklyorganize
themselvesinto largerconceptsand more complexbehaviors.The cell
assemblies, in short,would be the fundamentalquantaof thought.
Sittingtherein the audience,Holland wastransfixedby all this. This
wasn'tiust the arid stimulus/response view of psychology beingpushedat
the time by behavioristssuchasHarvard'sB. F. Skinner.Hebbwastalking
Masterof the Game 159
about what was going on inside the mind. His connectionisttheory had
the richness,the perpetualsurprisethat Holland respondedso stronglyto.
It felt right. And Holland couldn't wait to do somethingwith it. Hebb's
theory was a window onto the essenceof thought, and he wanted to watch.
He wanted to seecell assembliesorganizethemselvesout of random chaos
and grow. He wanted to seethem interact. He wanted to seethem incor-
porateexperienceand evolve.He wantedto seethe emergenceof the mind
itself. And he wantedto seeall of it happenspontaneously,without external
guidance.
No sooner had Licklider finished his lecture on Hebb than Holland
turned to his leaderon the 701 team, Nathaniel Rochester,and said, "Well,
we've got this proto$pe machine. Let's program a neural network simu-
lator."
And that'sexactlywhat they did. "He programmedone," saysHolland,
"and I programmed another, rather different in form. We called them the
'Conceptors.'
No hubris there!"
In fact, the IBM Conceptors stand as an impressiveaccomplishment
even forty years later, when neural network simulations have long since
become a standard tool in artificial intelligence research.The basic idea
would still look familiar enough. In their programs,Holland and Rochester
modeled their artificial neurons xs "neds5"-in effect, tiny computersthat
can remember certain things about their internal state.They modeled their
artificial synapsesasabstractconnectionsbetweenvariousnodes,with each
connectionhaving a certain "weight" correspondingto the strengthof the
synapse.And they modeled Hebb's learning rule by adjustingthe strengths
as the network gained experience.However, Holland, Rochester,and their
collaboratorsalso incorporatedfar more detailsabout basicneurophysiology
than most neural network simulations do today, including such factors as
how fast each simulatedneuron fired and how "tired" it got if it was fired
too often.
Not surprisingly, they had a tough time getting it all to work. Not only
were their programsamong the 6rst neural network simulations ever, they
markedone of the first times a computer had beenusedto simulateanything
(asopposedto calculatingnumbersor sortingdata).Holland givesIBM a
lot of credit for its corporate patience. He and his colleaguesspent un-
counted hours of computer time on their networks, and even took a trip
up to Montreal to consult with Hebb himself-at company expense.
But in the end, by golly, the simulations worked. "There was a lot of
emergence,"saysHolland, still soundingexcitedabout it. "You could start
with a uniform substrateof neuronsand seethe cell assembliesform." And
 150 COMPLEXITY
whenHolland,Rochester, andtheircolleaguesfinallypublished
theirre-
yearsafterthebulkof theresearch
sultsin 1956,several wasdone,it was
hisfirstpublished
paper,

BuildingBlocks

Lookingbackon it, saysHolland, Hebb'stheoryand his own network

simulationof it probablydid moreto shapehis thinkingoverthe nextthirty
yearsthandid anyothersinglething.But at thetime,themostimmediate
resultwasto goadhim into leavingIBM.
The problemwasthat computersimulationhad somedefinitelimits,
especiallyon the 701. Cell assemblies in a real nervoussystemhaveas
manyas 10,000neuronsdistributed overa largepartof the brain,with as
manyas 10,000synapses per neuron.But the largestsimulatednetwork
Hollandandhiscoworkers everranon the701hadonly 1000neuronsand
16connections perneuron-and theyonly managed to getthatfar by using
everyprogrammingtrick theycouldthink of to speedthingsup. "The more
ll did that," saysHolland,"the moreI realizedthat the distancebetween
f whatwe could reallytestout and whatI wantedto seewasiust too large."
i Th. alternative wasto try to analyzethe networks mathematically. "But
Ithat proved to be reallytough," he says. Everything he attempted ran up
'against
a brickwall. A full-fledgedHebbiannetworkwasiusttoofar beyond
anythinghe could tacklewith the math he had learnedat MIT-and he
hadtakena lot moremathcourses thanmostphysicsmajors."lt justseemed
to me that the key to knowingmore aboutnetworkswasto know more
mathematics," he says.So in the fall of 1952,with IBM'sblessing and a
nice little goingawaypresent-a conhactto continueconsultingfor Big
Blue at the rateof 100hoursler month-he enteredthe Ph.D. program
in mathematics at the Universigof Michiganin Ann Arbor.
Again,his luck held. Of course,Michiganwouldn'thavebeena bad
choice in any case.Not only did it haveone of the bestmathematics
departments in thecounhythen,but-a primeconsideration for Holland-
it had a footballteam."A footballweekendin the Big Ten, with 100,000
peoplecominginto town-I still enioyit!"
The realgoodfortune,however,wasthat at MichiganHollandencoun-
teredArthur Bu&, a philosopherwho wasno ordinaryphilosopher.A
tp"ciaffiiffigmatist phiiosophyof CharlesPeirce,Burkshadgotten
his Ph.D. in l94l at a time whentherewasno hopewhatsoever of finding
Masterof the Came 161
a teachingjob in his field. So the following year,he had takena ten-week
courseat the Moore School of the Universig of Pennsylvaniato turn himself
into an engineer for the war effort. It proved a happy choice. Shortly
thereafter,in 1943, he had been hired to work on the Moore School'stop-
secretENIAC-the first electronic computer. And there he had met the
legendaryHungarian mathematicianfohn von Neumann, who wascoming
in frequently from the Institute of Advanced Study in Princeton to work
asa consultanton the proiect.Undervon Neumann, Burla had alsoworked
on the design of ENIAC's successor,the EDVAC, the first computer to
store its instructionselectronicallyin the form of a program. Indeed, a
1945 paper by von Neumann, Burks, and the mathematicianHerman I
Coldstine-"Preliminary Discussionof the LogicalDesignof an Eie"iio"ic I'
Ccimliiitifrg Instrument"- is now regardedasone of the foundation stonesJ
of modem computer science.In it, the three men had defined the concept
of a program in a preciselogical form, and showedhow a general-purpose
computer could executesuch a program by a continuous cycle offetching
eachinstructionfrom the computer'smemory unit, executingthat instruc-
tion in a central processingunit, and then storing the resultsback in
memory. This "von - Neumann architecture"is still the basisfor almostall
computers today.
When Holland met him at the Universityof Michigan in the mid-1950s,
Burkswasa slim, rathercourtly man who lookedvery much like the minister
he had once thought of becoming. (To this day, Burks never appearson
the notoriouslycasualMichigan campuswithout a coatand tie.)But Burks
also proved to be a warm friend and a superb mentor. He quickly brought
Holland into his logic of Computers group, a coterie of theoristsstudying
computer languages,proving theoremsabout switching networla, and in
general trying to understandthesenew machinesat the rnost rigorous and
fundamentallevel.
Burks also invited Holland into a new Ph.D. programthat he washelping
to organize and that would be dedicatedto exploring the implications of
computersand information processingin asbroada realm aspossible.Soon
known as Communication Sciences,this programwould eventually evolve
into a full-fedged computer department in 1967, when it became known
as Computer and Communication Sciences.But at the time, Burksfelt he
wassimply carryingon the legacyof von Neumann,who had died of cancer
in 1954. "Von Neumann thought of using computersin two ways," he
says.One wes as a general-purposecomputational device, the purposefor
which they had been invented. "The other was as the basisfor a general
162 OoMPLEXITY
theory of automata, natural and arti6cial." Burks also felt that such a
programwould meetthe needsof thosestudents,Holland prominentamong
them, whose minds refusedto flow in the normal channels.
Holland liked what he heard. "The idea wasto developsome very tough
coursesin areaslike biology, linguistics,and psychology,as well as a lot
of the standard stuff such as information theory," he says' "The courses
would be taught by professorsfrom each subiect, so that the studentscould
get the linkagesbetween thesethings and computer models. And the stu-
dents who came out of these courseswould come out with a very deep
understanding of the fundamentals of the field-the problems, the ques-
tions, why the issuesweredifficult, and what computerscould do to help.
They wouldn'tjusthavea surfaceunderstanding."
Holland liked the conceptevenmore because he wasgettingtotally
disenchanted with mathematics. Likemost mathematics groupsin thepost-
WorldWar II era,thedepartment at Michiganwasdominated by theideals
of the.French Bourbakrjghool, whichcalledfor research of almostinhuman
purity anif-abstraction]ffi6'rdingto the Bourbakistandard,it waseven
considered gaucheto illustratethe concepts behindyouraxiomsand theo-
remswith somethingsodownto earthasa drawing."The ideawasto show
thatmathematics couldbedivorcedfrom anyinterpretation," saysHolland.
And yet this wasn'twhat he had comefor at all; he wantedto usemath-
ematicsto understand the world.
SowhenBurkssuggested thatHollarldtransferinto the Communication
Sciences program,he didn't hesitatg- He abandoned hryngg.lycompleted
dissertationin mathematics and startedill-iEl'Iffi6frt lfi-att-iouid ao
a-TiR?rutionfini?J"ill *"r much closerto whatI wantedto do," he
says-namely,neuralnetworls.(lronically,the dissertation topiche finally
decidedon, "Cyclesin l.ogicalNets,"wasdevotedto an analysisof what
happensin a networkof on-offswitches; in it he provedmanyof the same
theoremsthat a youngmedicalstudentnamedStuartKauffmanindepen-
dentlystruggled to provein Berkeley four yearslater')AndwhenHolland
finallyearnedhis Ph.D. in 1959,it wasthe firstthatthe Communication
Sciences programhad everawarded.

None of this deflectedHolland'sattentionfrom the broaderissuesthat

hadbroughthim to Michiganin the 6rstplace.Quite the opposite.Burks'
CommunicationSciences programwasthekindof environment wheresuch
questionscould thrive. What is emergence? And what is thinking?How
doesit work?What are its laws?What doesit reallymeanfor a systemto
Master of the Game 163
adapt?Holland iotteddown reamsof ideason thesequestions,then sys-
tematicallyfiled them awayin manila folderslabeledGlasperlenspiel 1,
Clasperlenspiel 2, et cetera.
Glas-what? "Das Glasperlenspiel," he laughs.It wasHermanHesse's
lastnovel, publishEfii-TyBi:hiGTft author was i n .* il6li$iii ffi I'l
FTollefriftfscovered it one day in a stackof booksthat a roommatehad
broughthome from the library. In Germanthe title literally meansTfre
Class-Bead Came, but in Englishtranslations the book is usuallycalled
Master of the Came. or its l,atin equivalent,MagisterLudi. Laid in a
societyof the far future, the novel describes a gamethat wasoriginally
playedby musicians; the ideawasto setup a themeon a kind of abacus
with glassbeads,and then try to weaveall kindsof counterpointand
variationon the themeby movingthe beadsbackand forth. Over time,
however,the gameevolvedfrom its simpleoriginsinto an inshumentof
profoundsophistication, controlledby a cadreof powerfulpriest-intellec-
tuals."The greatthingwasthatyoucouldtakeanycombinationof themes,"
saysHolland-"somethingfrom astology,something from Chinesehistory,
something from math-and then hy to developthem like a musictheme."
Of course,he says,Hessewasa little vagueaboutexacdyhow this was
done.But Hollanddidn'tcare;morethananythinghe'deverseenor heard
of, the Glass-Bead Gamecapturedwhat had fascinated him aboutchess,
aboutscience,aboutcomputers,aboutthe brain.In a metaphorical sense,
the gamewaswhat he'd beenafterall his life: "l'd like to be ableto take
themesfrom all overandseewhatemerges whenI put them together,"he
says.

A particularly fruiful source of ideas for rhe Glasprtenspiel files *ar y'o lt
another book that Holland came upon one day as he was browsing in the .1,[ L"fu
stacksof the math department library: R. A. Fisher'slandmark 1929 tome t o n b
on genetics,The Cenetical Thnry of Natural Selection. ft{ f
At first, Holland was fascinated."l'd alwaysenjoyedreadingabout ge-- ""t t
neticsand evolution, evenin high school," he says.He lovedthe idea that
genesfrom the parentsare reshuffledin each new generation,and that you
calculate how often specific traits such as blue eyesor dark hair will show
'Wow,
up in their offspring. "l thought, this is really neat!' But reading
Fisher'sbook was the first time I realizedthat you could do anything other
than hivial algebrain this area." Indeed, Fisher used much more sophis-
ticated ideas from differential and integral calculus, as well as probability
theory. His book had provided biologistswith the first really careful math-
1& COMPLEXITY
ematicalanalysisof how the distributionof genesin a populationwill change
as a result of natural selection.And as such it had laid the foundationfor
the modern, "neo-Darwinian" theory of evolutionarychange.A quarter
of a century later, it was still pretty much the stateof the art in the theory
of evolutionarydynamics.
So Holland devouredthe book. "The fact that you could takecalculus
and differentialequationsand all the other thingsI had learnedin my math
classesto start a revolution in genetics-that was a real eye-opener.Once
I saw that, I knew I could never let it go. I knew I had to do something
with it. So I kept messingaround with the ideas in the back of my mind,
scribblingnotes."
And yet, asmuch asHolland admiredFisher'smath, therewassomething
aboutthe wayFisherusedthe math that beganto botherhim. As he thought
about it more and more, in fact, it beganto bother him a lot.
For one thing, Fisher'swhole analysisof natural selectionfocusedon
the evolution of iust one geneat a time, as if each gene'scontribution to
the organism'ssurvival was totally independentof all the other genes. In
effect, Fisher assumedthat the action of geneswas completely linear. "I
knew that had to be wrong," saysHolland. A single gene for green eyes
isn't worth very much unlessit's backedup by the dozensor hundredsof
genesthat specify the structure of the eye itself. Each gene had to work as
part of a team, realizedHolland. And any theory that didn't takethat fact
into account was missinga crucial part of the story.Come to think on it,
that was alsowhat Hebb had beensayingin the mental realm. Hebb'scell
assemblieswere a bit like genes,in that they were supposedto be the
fundamental units of thought. But in isolationthe cell assemblieswere
almostnothing. A tone, a flashof light, a commandfor a muscletwitch-
the only way they could mean anything wasto link up into larger concepts
and more complex behaviors.
For another thing, it bothered Holland that Fisher kept talking about
evolution achievinga stabtq_9q*dfdtgg-that statein which a given species
has attained its optimum size, its optimum sharpnessof tooth, its optimum
fitnessto surviveand reproduce.Fisher'sargumentwasessentially the same
one that economistsuse to define economicequilibrium: once a species'
fitnessis at a maximum, he said, any mutation will lower the fitness.So
natural selection can provide no further pressurefor change. "An awful
'Well,
lot of Fisher is that way," saysHolland: "He says, the systemwill go
to the Hardy-Weinberg equilibrium becauseof the following process.. . .'
But that did not sound like evolution to me."
 Masterof the Came 165
He went back and rereadDarwin and Hebb. No, Fisher'sconcept of
equilibrium didn't sound like n at all. Fisherseemedto be talkingl
about the attainmentof some pristine,eternalperfection."But with Dar- |
win, you seethingsgettingbroaderand broaderwith time, more diverse,"I
saysHolland. "Fisher'smath didn't touch on that." And with Hebb, who
was talking about learning insteadof evolution, you saw the samething:
minds gettingricher,more subtle,moresurprisingasthey gainedexperience
with the world.
To Holland, evolution and learning seemedmuch more like-well, a
game. In both cases,he thought, you have an agent playing againstits
environment, trying to win enough of what it neededto keep going, In
.gyglulionlhaLaayoff is literally survival,
-In
and a chancefor the agentto pass
its geneson to the next gbneration. learning, the payoffis a rewardof
some kind, such as food, a pleasantsensation,or emotional fulfillment.
But either way, the payoff(or lack of it) givesagentsthe feedbackthey need
to improve their performance:if they'regoing to be "adaptive"at all, they
somehow have to keep the strategiesthat pay off well, and let the others
die out.
Holland couldn't help thinking of Art Samuel'schecker-playing program,
which took advantageof exactly this kind of feedback the program was
constantly updating its tactics as it gained experienceand learned more
about the other player. But now Holland was beginning to realize iust how
prescientSamuel'sfocus on gameshad really been. This game analogy
seemed'to be true of any adaptle system. In economfEffil!-a]dffTfG
@is-iii vo'Es,A'ndon and on. At some level, all
these adaptive systemsare fundamentally the same. And that meant, in
turn, that all of them are fundamentallylike checkersor chess:the space
of possibilitiesis vast beyond imagining. An agent can learn to play the
game better-that's what adaptation is, after all. But it has iust about as
much chanceof finding the optimum, stableequilibrium point of the game
as you or I have of solving chess.
No wonder "equilibrium" didn't sound like evolution to him; it didn't
even sound like a war game that a trio of fourteen-year-oldboys could
cobbletogetherin Wally Purmort'sbasement.Equilibrium impliesan end-
point. But to Holland, the essenceof evolution lay in the journey, the
endlesslyunfolding surprise:"It wasbecomingmore and more clear to me
that the things I wanted to understand,that I wascurious about, that would
pleaseme if I found out about them-equilibrium wasn'tan important
part of any of them."
166 COMPLEXITY

Holland had to keepall this on a backburnerwhile he completedhis

Ph.D.dissertation. But oncehehadgraduated in 1959-Burkshadalready
invitedhim to stayon with the Logicof Computersgroupasa postdoc-
he sethimselfthe goalof turninghis visioninto a completeand rigorous
theoryof adaptation."The beliefwasthat if I lookedat geneticadaptation
asthelongest-term adaptation,andthenervoussystem astheshortest term "
he says,'ithen the generaltheoreticalframeworkwould be the same."To
getthe initial ideasstraightin hisown mind he evenwrotea manifesto on
the subject,a forg-eight-page technicalreportthat he circulatedin July
196l under the title "A LogicalTheory of AdaptiveSystemsInformally
Described."
He alsobeganto noticea lot of raisedeyebrows amonghis colleagues
in the logic of Computersgroup.It wasn'ta senseof hostility,exactly.It
wasjust that a few peoplethoughtthat this generaltheoryof adaptation
business sounded weird.Couldn'tHollandbe spending histime on some-
thing a little more-fruidul?
"The questionwas,is it crackpot?"recallsHolland,who cheerfullyad-
mits that he wouldhavebeenskeptical,too, in his colleagues' place."The
stuff I wasdoing didn't fit very well in the nice, familiar categories. It
wasn'thardware,exactly.It wasn'tsoftware,exactly.And at the time it
certainlydidn't fit into artificialintelligence.
So you couldn'tuseanyof
the standard criteriaandcomeup with a judgment."
One personwho didn't needa lot of convincingwasBurls. "I supported
fohn," he says."Therewasa cliqueof logicianswho didn't think thatwhat
fohn wasdoingwaswhat'logic of Computers'should be about.They were
much moretraditional.But I told them that this is whatwe neededto do,
that it wasasimportantfor gettinggrantsastheirstuff."Burkswon the day:
asfounderandguru of the program,his voicecarriedconsiderable weight.
By and by, the skepticsdrifted out of the program.And in 1964,with
Burks'enthusiastic endonement.Hollandwasawardedtenure."An awful
lot of thoseyearsI oweto Art Burksactingasa shield,"he says.
Indeed,Burls'backinggaveHollandthesecurityhe neededto beardown
on the theoryofadaptationashardashe could. By 1962he hadput aside
all his otherresearch projectsandwasdevotinghimselfto it essentially full
time. In particular,he wasdeterminedto crackthis problemof selection
basedon morethanonegene-and not justbecause Fisher'sindependent-
geneassumption hadbuggedhim morethananythingelseaboutthatbook.
Master of the Came 167

Moving to multiple genes was also the key to moving away from this
obsessionwith equilibrium.
In fairnessto Fisher,saysHolland, equilibrium actuallydoesmake a lot
of sensewhen you'retalkingaboutindependentgenes.Forexample,suppose
you had a specieswith 1000 genes,which would make it roughly as com-
plicated as seaweed.And suppose,for simplicity'ssake, that each gene
comesin just two varieties-greencolor versusbrown color,wrinkledleaves
versussmooth leaves,and so forth. How many trials doesit take for natural
selection to 6nd the set ofgenes that givesthe seaweedits highest 6tness?
If you assumethat all the genesare indee(Spendent, saysHolland,
then for each gene you just need two trials to fiiifiilHrc*n-Varietyis better,
Then you have to perform those two trials on each of 1000 genes.So you
need 2QQ0!5!g!gjn all. And that's not very many, he says.In fact, it's such
a comparativelysmall number that you can expectthis seaweedto attain
its maximum fitnessfairly quickly, at which point the specieswill indeed
be at an evolutionaryequilibrium.
But now, saysHolland, look what happenswith that 1000-geneseaweed
when you assumethat the genesare no-t_r:r-dSpsndg$,.To be sure of finding
the highestlevel of fitnessin this case,natural selectionwould now have
to examine every conceivable combination of genes, becauseeach com-
bination potentially has a different fitness.And when you work out the
total number of combinations,it isn't 2 multiplied by 1000.It's 2 multiplied
by itself 1000 times. That's_2rm0,or about 10r00-a number so vastthat it
makeseven the number of moves in chessseem infinitesimal. "Evolution
can't evenbeginto try out that many things," saysHolland. "And no matter
how good we get with computers,we can't do it." Indeed, if every ele-
mentary particle in the observableuniversewere a supercomputerthat had
been number-crunching away since the Big Bang, they still wouldn't be
close.And remember,that'siust for seaweed. Humansand other mammals!
have roughly I00 times as many genes-and most of thosegenescome in
I
many more than two varieties. I
So once again, saysHolland, you have a systemexploring its way into
an immense spaceof possibilities,with no realistichope of ever finding
the single"best" place to be. All evolution can do is look for improvements,
not perfection. But that, of course, was precisely the question he had
resolvedto answerback in 1962:How? Understandingevolution with mul-
tiple genesobviouslywasn't just a trivial matterof replacingFisher'sone-
variableequationswith many-variableequations.What Holland wantedto
know was how evolution could explore this immense spaceof possibilities
168 COMPLEXITY
andfind usefulcombinations of genes-withouthavingto searchoverevery
squareinch of territory.
As it happens, a similarexplosionof possibilitieswasalready wellknown
to mainstreamartificial i;Fell@-ie researchers. At Carnegie Tech (now
Carnegie Mellon Universig)in Pittsburgh,for example, Allen Newelland
HerbertSimon had been conducting a landmark shrdy of human problem-
solvingsincethe mid-1950s. By askingexperimental subiectsto verbalize
theirthoughbastheyshuggled througha wide variety of puzzles andgames,
includingchess,Newelland Simonhad concluded that problem-solving
alwaysinvolvesa step-by-step mentalsearchthrougha vast"problemspace"
of possibilities,with eachstepguidedby qbsu11$i{ule of thumb: "If this
is the situation,then that stepis worthtaking."By buildingtheir theory
into a programknown as GeneralProblemSolver,and by putting that
programto work on thosesamepuzzlesand games,Newell and Simon
had shownthat the problem-space approachcouldreproduce human-s$le
reasoningremarkablywell. Indeed,their conceptof heuristicsearchwas
alreadywell on its wayto becomingthe dominantconventional wisdomin
artificial intelligence.And GeneralProblemSolverstood-as it still
stands-asoneof the mostinfluentialprograms in theyoung6eld'shistory.
But Hollandwasdubious.It wasn'tthat he thoughtNewelland Simon
werewrong about problemspacesor heuristics.Shortlyafter he got his
Ph.D., in fact, he had madeit a point to bringboth of them to Michigan
as part of a maior seminaron artificial intelligence.He and Newell had
beenfriendsandintellectualsparringpartnerseversince,No, it wassimply
thattheNewell-Simon approach didn'thelphim withbiological evolution.
The ap
trthgtlhere-arc.!g3r!q!!c rules,no guidance
6ffsort; succeeding generations explorethe ipaceof possibilities by
mutations and random reshuffiing of genesamong the sexes-in short, by
trial and error. Furthermore, those succeedinggenerationsdon't conduct
their search in a step-by-stepmanner. They explore it in parallel: each
member of the population has a slightly different set of genesand explores
a slightly different region of the space.And yet, despitethesedifferences,
evolution producesiust as much creativity and surpriseas mental activity
does, even if it takesa little longer. To Holland, this meant that the real
unifying principles in adaptation had to be found at a deeper level. But
where?
Initially, all he had wasthis intuitive idea that certain setsof genesworked
well together,forming coherent, self-reinforcingwholes.An examplemight
be the cluster of senesthat tells a cell how to extract energy from glucose
*ol."ffio', th! cluster that conhols cell division, or the cluster that
Master of the Game 169
governshow a cell combineswith other cellsto form a certainkind of
tissue.Hollandcouldalsoseeanalogsin Hebb'stheoryof the brain,where 1
a setof resonating cell assembliesmight form a coherentconceptsuchas
I
"car,"or a coordinated motionsuchasliftingyourarm.
But the more Holland thoughtaboutthis ideaof coherent,self-rein-
forcingclusters,the moresubdeit beganto seem.Foronething,you could
fi nd rg4lggggl-.*egPl"s almost anywhereyou looked.*$u-b" rqgti,u$-in a
computer piogram._-De.pp,*me,nts,ina bureaucracy. Gambils in the larger
strategyof a chess ga;e. Furthermore, you couldT'nd"6ii-mples at eyery
is
level of organization.If a cluster coherentenough anil stabli enough,
thin"ii'ian usuallyserveasa buildingllock for somelargercluster.Cells
maketissues, tissuesmakeotgffidr-nlilile organisms,-organisms make
ecosystems-onandon. Indeed,thoughtHolland,that'swhatthisbusiness
of "emergence" wasall about buildingblocksat onelevelcombininginto
newbuildingbloclaata higherlevel.It seemed to beoneof thefundamental
organizingprinciplesof the world. It certainlyseemedto appearin every
complex,adaptivesystemthat you lookedat.
But why?This hierarchical,building-blockstructureof thingsis ascom-
monplaceasair. It'ssowidespread thatwe neverthinkmuchaboutit. But
when you do think about it, it criesout for an explanation:Why is the
world structuredthis way?
Well, thereareactuallyanynumberof reasons. Computerprogrammers
aretaughtto breakthingsup into subroutines because small,simpleprob-
lemsare easierto solvethan big, messyproblems;it's simplythe ancient
principleof divide and conquer.l,argecreatures suchaswhalesand red-
woodsaremadeof trillionsof tiny cellsbecause the cellscame6rst;when
largeplantsand animals6rstappeared on Earthsome570million years
ago,it wasobviouslyeasierfor naturalselectionto bringtogetherthe single-
celledcreatures that alreadyexistedthan to build big new blobsof proto-
plasmfrom scratch.GeneralMotorsis organized into severalzillion di-
visionsand subdivisionsbecausethe CEO doesn'twant to have half a
million employees reportingto him directly;therearen'tenoughhoursin
theday.In fact,asHerbertSimonhadpointedout in the 1940sand 1950s
in his studiesof business organizations,a well-designed (emphasize well-
designed) hierarchyis an excellentwayof gettingsomeworkdonewithout
anyone personbeingoverwhelmed by meetingsand memos.
As Holland thoughtaboutit, however,he becameconvincedthat the
most importantreasonlay deeperstill, in the fact that a hierarchical, I
building-blockstructureutterlyhansforms a system's
abilityto learn,evolve,'I
and adapt.Think of our cognitivebuildingblocla,which includesuch
170 COMPLEXITV
conceptsas red, car, and road. Once a set of building blockslike this
hasbeentweakedand refinedand thoroughlydebugged throughexperi-
ence,saysHolland, then it can generallybe adaptedand recombinedto
build a greatmany new concepts-say,"A red Saabby the side of the
road."Certainlythat'sa much moreefficientwayto createsomethingnew
than startingall overfrom scratch.And that fact,in turn, suggests a whole
newmechanism for adaptationin general.Instead of movingthroughthat
immensespaceof possibilities stepby step,soto speak,an adaptivesystem
can reshuffleits building blocksand takegiantleaps.
Holland'sfavoriteillustrationof this is the waypoliceartt$Lusedto work
in the daysbeforecomputers,when they needed'tomakea drawingof a
suspectto matcha witness'sdescription.The ideawasto divide the face
up into, say,l0 buildingblocks:hairline,forefuad,eyes,. nose andsoon
downto the chin. Then the artistwouldhavestripsof paperwith a variety
of optionsfor each:say,l0 differentnoses,l0 differenthairlines,and so
forth. That would makea total of 100piecesof paper,saysHolland. And
armedwith that,theartistcouldtalkto thewitness,assemble theappropriate
pieces,and producea sketchof the suspectvery quickly.Of course,the
artistcouldn'treproduce everyconceivable facethat way.But he or she
couldalmostalwaysgetprettyclose:by shufflingthose100piecesof paper,
the artistcouldmakea totalof l0 billion differentfaces,enoughto sample
the spaceof possibilities quite widely."So if I havea process that can
discoverbuilding blocks,"says Holland, "the combinatoricsstartworking
for me insteadof againstme. I can describe a great
many complicated things
with relativelyfewbuildingblocks,"
And that, he realized,wasthe key to the multiple-genepuzzle."The
cut and try of evolutionisn't iustto build a goodanimal,but to find good
buildingblocksthatcanbe put together to makemanygoodanimals."His
challengenowwasto showprecisely andrigorouslyhowthatcouldhappen'
And the first step,he decided,wasto makea computermodel,a "genetic
algorithm"that would both illustratethe process and help him clarifythe
issues in his own mind.

At one time or another,iustabouteveryone in the Michigancomputer

sciencecommunityhad had the experience of seeing|ohn Holland come
runningup with a fisful of fan-foldcomputerprintout.
"Lookat that!"he wouldsay,eagerly pointingto somethingin themidst
of a pagefull of hexadecimal gibberish.
"Oh. CCBl095E.That's-wonderful,|ohn"'
Master of the Game 171

"No! No[ Do you know what that means . l?"

Actually, there were quite a few people in the early 1960swho didn't
know and who couldn't' quite figure it out. His skepticalcolleagueshad
been right about one thing, at least: the genetic algorithm that Holland
finally came up with was weird. Except in the most literal sense,.in fact,
it wasn't really a computer program at all. In its inner workj,ngsit was more
like a simulated ecosystem-a kind of digital Sereugetiin which whole
populations of programswould compete and .haVesex and reproduce for
generatigr*aftergeneration, alwaysevolvjpg'their way toward the solution
of.xhiitever problem the program45zdight set for them.
This wasn't the way programswere usually written, to put it mildly. So
to explain to his colleagueswhy it made sense,Holland usually found it
best to couch what he was doing in very practical terms. Normally, he
would tell them, we think of a computer programas a shing of instructions
written in a special programming languagesuch as FORTRAN or LISP.
Indeed, the whole art of programming is to make sure that you've written
precisely the right instructions in preciselythe right order. And that's ob-
viously the most effectiveway to do it-if you alreadyknow preciselywhat
you want the computer to do. But supposeyou don't know, said Holland.
Suppose,for example, that you're trying to find the maximum value of
some complicated mathematical function. The function could represent
profit, or factory output, or vote counts, or almost anything else;the world
is full of things that need to be maximized. Indeed, programmershave
devisedany number of sophisticatedcomputer algorithmsfor doing so. And
yet, not even the best of those algorithms is guaranteedto give you the
correct maximum value in every situation. At some level, they alwayshave
to rely on old-fashionedtrial and error-guessing.
But if that's the case, Holland told his colleagues,if you're going to be
relying on trial and error an)'way,maybe it's worth seeingwhat you can do
with nature'smethod of trial snd e11q1-namely, natural selection. Instead
of trying to write your programs to perform a task you don't quite know
how to do, evolvethem.
The genetic algorithm was a way of doingthat. To seehow it works, said
Holland, forget about the FORTRAN code and go down into the guts of
the computer, where the program is representedas a shing of binary ones
a n d z e r o s l: l 0 l 0 0 l l l l 0 0 0 l l 0 0 l 0 0 0 l 0 l 0 0 l l l 0 l l , e t c e t e r aI.n t h a t
form the program looks a heck of a lot like a chromosome, he said, with
eachbinary digit being a single"gene." And once you startthinking of the
binary code in biological terms, then you can use that same biological
analogy to make it evolve.
172 COMPLEXITY
First, said Holland, you have the computer generatea population of
maybe 100 of these digital chromosomes,with lots of random variation
from one to the next. Each chromosomecorrespondsto an individual zebra
in a herd of zebras,so to speak.(Forsimplicig's sake,and becauseHolland
wastrying to get at the absoluteessenceofevolution, the geneticalgorithm
leavesaside such details as hoofs and stomachsand brains. It models the
individual as a single piece of naked DNA. As a matter of practicality,
moreover, Holland had to make his binary chromosomesno more than a
few dozen binary digits long, so that they were actually not full-scale
programs but fragments of programs. In his earliest work, in fact, the
chromosome representedonly a single variable. But none of that changed
the basicprinciple of the algorithm.)
Second, said Holland, you test each individual chromosomeon the
problem at hand by running it as a computer program, and then giving it
a score that measureshow well it does. In biological terms, this scorewill
determinethe individual's"fitness"-its probabilig of reproductivesuccess.
The higher the fitness,the higher the individual's chancesof being selected
by the genetic algorithm to passon its genesto the next generation.
Third, said Holland, you takethoseindividualsyou'veselectedas being
fit enough to reproduce,and createa new generationofindividuals through
sexual reproduction. You allow the rest to die off. In practice, of course,
the genetic algorithm leavesaside gender differences,courtship rituals,
foreplay, the union of sperm and egg, and all the other intricacies of real
sexual reproduction, and instead createsthe new generationthrough the
bare-bonesexchangeof genetic material. Schematically,the algorithm
choosesa pair of individuals with chromosomesABCDEFG and abcdefg,
breakseach stringat a random point in the middle, and then interchanges
the piecesto form the chromosomesfor a pair of offspring:ABCDefg and
abcdEFC. (Holland got the idea from real chromosomes,where this sort
of interchange,or "crossover,"happensfairly frequently.)
Finally, said Holland, the offspringproducedby this sexualexchangeof
genesgo on to compete with each other and with their parents in a new
generationalcycle.And this is the crucialstep,both in the geneticalgorithm
and in Darwinian natural selection.Without sexualexchange,the offspring
would have been identical with their parentsand the population would be
well on its way to stagnation.The poor performerswould gradually die off,
but the good performerswould never show any improvement. With sexual
exchange,however,the offspringare similar to their parents,but different-
and sometimesbetter. And when that happens,said Holland, those im-
provementsstand a good chance of spreadingthrough the population and
Master of the Game 173

improving the breedmarkedly. Natural selectionprovidesa kind of upward

ratchet.
In real organisms,of course,quite a bit of variationis also providedby
mutations, $pographical errors in the geneticcode. And in fact, said Hol-
land, the genetic algorithm does allow for an occasionalmutation by de-
liberatelyturning a I into a 0, or vice versa.But for him, the heart of the
genetic algorithm was sexual exchange..Not only does the exchange of
genesthrough sex provide for variation in the population, but it turns out
to be a very good mechanism for searchingout clustersof genesthat work
well togetherand produceabove-average fitness-in short,building blocks.
For example,saidHolland, suPPose that you'veput the geneticalgorithm
to work on one of those optimization problems, where it's looking for a
way to find the maximum value of somecomplicatedfunction. And suppose
that the digital chromosomesin the algorithm's internal population turn
out to get very high scoreswhen they havecertainpatternsofbinary genes
such a-, ll####ll#n###10 or ##1001###lll0l##' (Holland
used # to stand for "doesn't matter"; the digit in that position could be a
I or a 0.) Such patternswill function as building blocls, he said' Maybe
they happento denoterangesofvariableswherethe function doesindeed
havehighervaluesthan average.But whateverthe reason,the chromosomes
that contain such building blocks will tend to plosper and spreadthrough
the population, displacing chromosomesthat don't have them'
Furth.r*ore, he said, sincesexualreproductionallowsthe digital chro-
mosomesto shuffle their geneticmaterial every genelation, the population
will constantlybe coming up with new building blocksand new combi-
nations of the existing building blocks. So the genetic algorithm will very
quickly produce individuals that are doubly and triply blessedwith good
buildingblocks. And if thosebuilding blocksact togetherto confer extra
advantages,Holland was able to show, then the individuals that have them
will spreadthrough the population even fasterthan before. The upshot is
that the genetic algorithm will convergeto the solution of the problem at
hand quite rapidly-without ever having to know beforehand what the
solution is.
Holland remembersbeing thrilled when he first realized this, back in
the early 1960s.It never seemedto get his audiencesvery excited;at the
time, mostof his contemporaries in the still-young6eld of computerscience
felt that they had more than enough to do in laying the groundwork for
conventional programming. In purely practical terms, the idea of evolving
a program seemeda little off the wall' But Holland didn't care. This was
exactlywhat he had been looking for sincehe setout to generalizeFisher's
174 COMPLEXITY
independent-genesassumption. Reproductionand crossoverprovided the
mechanism for building blocls of genesto emergeand evolve together-
and, not incidentally, provideda mechanismfor a population of individuals
to explore the spaceof possibilitieswith impressiveefficiency. By the mid-
1960s, in fact, Holland had proved what he called the schema theorem,
the fundamentaltheoremof geneticalgorithms:in theffi666fffif-
duction,crossover,andmutation,almostanycompactclusterof genesthat
providesabove-average fitnesswill grow in the populationexponentially.
("Schema"washis term for anyspecificpattemof genes.)
"lt waswhenI finally got the schematheoremin a form that I likedthat
I startedwriting my book," he says.

Emergence of Mind

"Jfis fesft"-a compilationof the schematheorem,the geneticalgo-

rithm, and his thinkingon adaptation in general-wassomethingthat
Holland thoughthe might be ableto 6nish in a yearor two. In fact, it
took him a decade.Somehow,asthe writing and the research continued
in parallel,he wasalwaysfindinga new ideato exploreor a newaspectof
the theoryto analyze.He setseveralof his graduatestudentsto work on
computerexperiments-demonstrations that the geneticalgorithmreally
wasa usefulandefficientwayto solveoptimizationproblems.Hollandfelt
thathe waslayingout the theoryandpracticeof adaptation, andhe wanted
it done right-with detail,precision, andrigor.
He certainlyhad that. Publishedin 1975, Adaptationin Natural dnd
Artificial Sysfemswasdensewith equationsand analysis.It summarized
twodecades of Holland'sthinkingaboutthedeepinterrelationships of learn-
ing, evolution,and creativity.It laid out the geneticalgorithmin exquisite
detail.
And in the wider world of computerscienceoutsideMichigan, it was
. In a communityof peoplewho like their
€gjtth rygo!
algorithmsto be elegant,concise,and provablycorrect,this geneticalgo-
rithm stuff wasstill just too weird.The artificialintelligencecommunity
wasa little more receptive,enoughto keepthe booksellingat the rateof
100to 200 copiesper year.But evenso, whentherewereany comments
on the bookat all, they weremostoften alongthe linesof, "fohn'sa real
brightguy,but . ."
Of course,it hasto be saidthat Hollanddidn't do much to makehis
case.Holland publishedhis papers,althoughrelativelyfew of them. He
Masterof the Game 175

gaveseminarswhen people invited him to. But that was iust about it. He
lidn't ,n"k. dramatic ciaims for genetic algorithms at the maior confer-
H. didn't apply genetic algorithms to flashy applications such as
"r,".r. diagnosis,itr.n"a that might make venturecapitalistssit up and
*.Ji*f
"laboratory"for
take notice.-He didn't lobby for big grantsto establisha
geneticalgorithms.He didn't publish a popularbook warning that massive
needed to meet the
F"a.r"t f.ittaing of genetic allorithms was urgently
fapanesethreat.
Inshort,hesimplydidn'tplaythegameofacademicself-promotion.
point,
That seemsto be the one game he doesn't like to play. More to the
he really doesn'tseemto care if he wins it or not. Metaphorically speaking,
"It's like
he still prefersto putter awaywith a few buddies in the basement.
on a sandlot team
playing baseball,'ihe says."fust becauseyou'replaying
not in the maiors-it's the fun that matters.And the kind of science
"na
I do has alwaysbeen a lot of fun for me'"
,,1 listen,"
think it would havebotheredme if nobodyhad been willing to
,,But I've always been very lucky in having bright, interested
he adds.
graduatestudentsto bounce ideasoff of'"
" attitude:
lnd".d, that was the flip side of his buddies-in-the-basement
lot of energy into working with his immediate group at
Holland put a
one time he typically supervised six or seven graduate
Michigan. At any
managed
students-far aboveaverage'Startingin the mid-1960s,in fact' he
with Ph.D.'s at an average of more than one per year'
to graduatethem
iso-. of them havebeenreallybrilliant-and greatfun for that reason,"
he says.Holland deliberatelytook a rather hands-offapproachto guidance,
pub-
having seen too many professorsbuild up a huge publication list by
.,joint" ,.r."r"h papersthat were in fact written entirely by their
lishid
they
grrd,rlt.'rtudents. "So tirey all followed their nosesand did things
we'd all meet around a table about once a
iho,rght were interesting.Then
where he stood on his dissertation, and we'd
,..k] on. of them *ould t ll
all critique it. That was usually a lot of fun for everybodyinvolved." .
In the mid-1970s,Holland also startedmeeting with a group of like-
minded faculty membersfor a free-wheelingmonthly seminaron--well,
just about having to do with evolutionor adaptation.In additron
"nyihing to
io Burks, the group included Robert Axelroi, a political scientisttrying
understa.,d*[y ,.,a whe.r pei]lEilill6ilperate instead of stabbing each
other in the back;Michael bohen, anotherpolitical scientist,specializing
in the social dynamicsof human organizations;and william Hamilton,
an evolutionarybiologistworking with Axelrod to understandsymbiosis,
socialbehavior,and other forms of biologicalcooperation'
 176 COMPLEXITY
"Mike Cohen wasthe catalyst,"recallsHolland. It wasjust after Ad-
aptationcameout. Cohen,who had beensittingin on one of Holland's
courses,cameup afterclassonedayto introducehimself,and said,"you
reallyought to be talkingto Bob Axelrod."Hollanddid, and through
Axelrodsoonmet Hamilton.The BACHgroup-Burks,Axelrod,Cohen,
Hamilton,andHolland-coalescedalmostimmediately.(Theyalmosthad
to work in a "K"; very earlyin the group'sexistence they hied to recruit
StuartKauffman,but lost out to the Universigof Pennsylvania.) ,,What
tiedustogetherwasthatweall hada verystrongmathematical background,"
saysHolland. "we alsofelt very stronglythat the issueswerewider than
any one problem.We beganmeetingon a regularbasis:someonewould
seea paper,and we'd all come in and discussit. There was a lot of
exploratorythinking."
Indeedtherewas-particularly on Holland'spart. The bookwasdone
now, but his conversations with the BACH grouponly underscored what
it had left undone.The geneticalgorithmand the schematheoremhad
capturedsomethingessentialand right aboutevolution;he wasstill con-
vinced of that. But evenso, he couldn't help but feel that the genetic
, algorithm'sbare-bones versionof evolutionwasiust toobare.Something
I had to be missing in a theory in which "organisms" areiustnakedpieces
| 9f DNA that have been designed by a programmer. What could a th.ory
I like thattell you about complex organismsevolving in a complex environ-
I ment? Nothing. The genetic algor!$m was all very nice. But by ibelf, it
I simplywasn'tan adaptive agent. :
1' Nor, @netic algorithma modelof adaptation in
I the human mind, Because it was so explicitlybiological in its design, it
- couldn't tell you anything about how complex concepts grow, evolve, and
/ I
. ! recombinein the mind. And for Holland,that factwasbecomingmore
'
and morefrustrating.Nearlytwenty-fiveyearsafterhe'd 6rst heardabout
Donald Hebb'sideas,he wasstill convincedthat adaptationin the mind
and adaptationin naturewerejusttwo differentaspects of the samething.
Moreover,he wasstill convincedthat if they reallywerethe samething,
they oughtto be describable by the sametheory.
So in the latterhalf of the 1970s,Hollandsetout to Fnd that theory.

Backto basics.An adaptiveagentis constantlyplayinga gamewith its

environment.what exactlydoesthat mean?Distilledto the essence,what
actuallyhasto happenfor game-playingagentsto surviveand prosper?
Two things,Hollanddecided: andfeedb-ac\.It wasan insight
I195!js@r
Master of the Game 177

that he could trace all the way back to his IBM daysand his conversations
with Art Samuel about the checkerplayer.
Prediction is just what it soundslike: thinking ahead. He can remember
Samuel making the point again and again. "The very essenceof playing a
good game of checkersor chessis assigningvalue to the less-than-obvious
stage-iettingmoves," saysHolland-the moves that will put you in an
advantageousposition later on. Prediction is what helps you seize an op-
portunity or avoid getting suckeredinto a hap. An agent that can think
ahead has an obvious advantageover one that can't.
But the concept of prediction also turns out to be at least as subtle as
the conceptof building blocks,saysHolland, Ordinarily, for example,we
think of piediction as being somethingthat humans do consciously,based
on ,o*. explicit model of the world. And there are certainly plenty of
those explicit models around. A supercomputer's simulation of climate
changeis one example.A start-upcompany's business plan is another,as
is an economic projection madeby the Federal ReserveBoard. Even Stone-
henge is a model, its circular arrangementof stones provided the Druid
prieits with a rough but effectivecomputer for predicting the arrival of the
equinoxes.Very often, moreover,the modelsare literally insideour head,
as when a shopper tries to imagine how a new couch might look in the
living room, or when a timid employeetries to imagine the consequences
of t iing off his boss.We use these"mental models" so often, in fact, that
manyprychologists areconvincedtheyarethe basisof all consciousthought.
Buito Holland, the conceptof prediction and modelsactually ran far
deeper than conscious thought-or for that mattet, far deeper than the
existenceof a brain. "All complex,adaptivesystems-economies,minds,
organisms-build models that allow them to anticipatethe world," he
declares.Yes, even bacteria.As it turns out, saysHolland, many bacteria
have special enzyme systemsthat cause them to swim toward stronger
concentrationsof glucose.Implicitly, thoseenzymesmodel a crucial aspect
of the bacterium'sworld, that chemicalsdiffuse outward from their source'
growing less and less concentratedwith distance.And the enzymessi-
multaneouslyencode an implicit prediction:If you swim toward higher
concentrations,then you're likely to find something nutritious. "It's not a
consciousmodel or anythingof that sort," saysHolland. "But it givesthat
organisman advantageover one that doesn'tfollow the gradient."
A similar story can be told about the viceroy bufterfly, he says.The
viceroy is a striking, orange-and-blackinsect that is apparentlyquite suc-
culent to birds-if only they would eat it. But they rarelydo, becausethe
viceroy has evolveda wing patternthat closelyresemblesthat of the vile-
 178 COMPLEXITY
tastingmonarchbutterfly,which everyyoungbird quicklylearnsto avoid.
So in effect,saidHolland,the DNA of the viceroyencodes a modelof the
worldstatingthatbirdsexist,thatthe monarchexists,andthatthe monarch
tasteshorrible.And everyday,the viceroyfluttersfrom flowerto flower,
implicitly bettingits life on the assumption that ib modelis correct.
The samestory can be told yet againabout a very differentkind of
organism, saysHolland:the corporation. Imaginethata manufacturer re-
ceivesa routineorderfor, sa5 10,000widgets.Sinceit's a routineorder,
the employeesprobablydon't give any profoundthoughtto the matter.
Instead,they just set up the productionrun by invokinga "standardop-
eratingprocedsys"-asetof rulesof the form, "lf the situationis ABC,
then takeactionXYZ." And just aswith a bacteriumor the viceroy,says
Holland, thoserulesencodea modelof the company'sworld and a pre-
diction: "lf the situationis ABC, then actionXYZ is a worthwhilething
to do and will lead to god results."The employees involvedin carrying
out the proceduremay or may not know what that model is. After all,
standardoperatingprocedures are often taughtby rote, without a lot of
whysand wherefores. And if the companyhasbeenaroundfor a while,
theremay not be anyoneleft who evenremembers why thingsaredonea
certainway.Nonetheless, asthe standardoperatingprocedurecollectively
unfolds,the companyasa wholewill behave asif it understood thatmodel
perfectly.
I I" the cognitiverealm,saysHolland,anythingwe call a "skill" or "ex-
I pertise"is an implicitmodel-or moreprecisely, a huge,interlocking set
I of standardoperatingprocedures that havebeeninscribedon the nervous
I' system and refinedby yearsofexperience. Showa textbook exercise to an
experienced physicsteacherand he won't wasteanytime scribblingevery
formulain sight,the waya novicewill; his mentalprocedures will almost
always showhim a pathto thesolutioninstantly:'lAha! That'sa conservation
of energyproblem."Lob a tennisball across the net to Chris Evertand
shewon'tspendanytime debatinghowto respond: afteryearsof experience
and practiceand coaching,her mentalprocedures will allow her to slam
the ball backdown your throatinstinctively.
Holland's favoriteexample of implicitexpertiseistheskillof themedieval
architects-who createdthe greatCothic cathedrals. They had no way to
cf,ffi fTfanyiffin g"fse that a modernarchitect
mightdo. Modernphysicsandstructuralanalysis didn't existin the twelfth
century.Instead, theybuilt thosehigh,vaultedceilingsandmassive flying
buttressesusingstandardoperatingprocedures passed downfrom masterto
Master of the Game 179

apprentice-rules of thumb that gave them a senseof which structures

would stand up and which would collapse. Their model of physics was
completely implicit and intuitive. And yet, thesemedieval craftsmenwere
able to createstructuresthat are still standing neadya thousandyearslater.
The examplescould go on and on, saysHolland. DNA itselfis an implicit
model: "Under the conditions we expect to find," say the genes, "the
creaturewe specify has a chance of doing well," Human culture is an
implicit model, a rich complex of myths and symbolsthat implicitly define
a people'sbeliefs about their world and their rules for correct behavior. For
that matter, Samuel'schecker player contained an implicit model, which
it createdby changing the numerical value it assignedto various options
as it gained experiencewith the opponent'splaying s$le.
Indeed, models and predictionsare ever''where,saysHolland' But then,
where do the modelscome from? How can cny system,natural or artificial,
learn enough about its universeto forecastfuture evenb?It doesn't do any
goodto talk about "consciousness,"he says.Most modelsarequite obviously
not conscious:witnessthe nutrient-seekingbacterium, which doesn't even
havea brain. And in any caseit simply begsthe question.Where doesthe
consciousness come from? Who programsthe programmer?
Ultimately, saysHolland, the answerhas to be "no one." Becauseif
there is a programmer lurking in the bacforound-"the ghost in the ma-
chine"-then you haven't really explained anything. You've only pushed
the mystery off someplaceelse. But fortunately, he says,there is an alter-
native: feedbackfrom tlfe-ginns*tt. This was Darwin's great insight,
that an irtpd; ib internal models without any paranormal
"Adil;r
guidance whaboever. It simply has to try the models out, se€ how well
their predictionswork in the real world, and-if it survivesthe experience-
adjust the models to do better the next time. In biology, of course, the
agentsare individual organisms,the feedbackis provided by natural selec-
tion, and the steadyimprovement of the models is called evolution. But
in cognition, the processis essentiallythe same:the agentsare individual
minds, the feedbackcomes from teachersand direct experience,and the
improvement is called learning. Indeed, that's exactly how it had worked
in Samuel's checker player. Either way, saysHolland, an adaptive agent
has to be able to take advantageof what its world is trying to tell it.

The next question,of course,washow?Hollanddiscussed the basic

in the BACH group.But in the end,
conceptat lengthwith his colleagues
1BO COMPLEXITY
therewasonly one wayto pin the ideasdown:he would haveto build a
computer-simulated adaptiveagent,justashe haddonefifteenyearsearlier
with geneticalgorithms.
UnfortunatelShe found that mainsheamartificialintelligencewasno
morehelpful in 1977than it had beenin 1962.The field had admittedly
madesomeimpressive progressin thattime. Out at Stanford,for example,
the artificialintelligencegroupwascreatinga seriesofstartlinglyeffective
programsknownasexpertsystems, which modeledthe expertise of, say,a
doctor,by applyinghundredsof rules:"lf the patienthasbacterialmen-
ingitis, and has been seriouslyburned, then the organismcausingthe
infectionmay be Pseudomonas aeruginosa."Eventhen, the venturecap
italistswerestartingto sit up and takenotice.
But Holland wasn't interestedin applications. What he wanted was a
fundamental theory of adaptiveagents.And so far as he could see,the past
two decadesof progressin artificial intelligence had been achievedat the
price of leaving out almost everything important, startingwith leaming and
feedbackfrom the environment. To him, feedback was the fundamental
issue.And yet, with a few exceptionssuch as Samuel, people in the field
seemedto believe that learning was something that could be set aside for
later, after they had gotten their programs working well with things like
languageunderstanding,or problem-solving,or someotherform of abshact
reasoning. The expert systemsdesignerseven seemed to take a certain
macho pride in that fact. They talked about something called "knowledge
engineering," in which they would createthe hundredsof rules neededfor
a new expert systemby sitting down with the relevant expertsfor months:
"What would you do in fftis sihration?What would you do in thct sihration?"
In fairness,even the knowledgeengineershad to admit that things would
go a lot more smoothly if the programs could only learn their expertise
from teachingand experience,as peopledo-and if someonecould only
figure out how to implement learning without making the softwarefar more
complex and cumbersome than it already was. But to Holland, that was
preciselythe point. Rigging the softwarewith some ad hoc "learning mod-
ule" wasn't going to solve anything. Learning was as fundamental to cog-
nition as evolution wasto biology.And that meant that learninghad to be
built into the cognitive architecturefrom the beginning, not slappedon at
the end. Holland's ideal was still the Hebbian neural network, where the
neural impulses from every thought strengthenand reinforce the connec-
tions that make thinking possiblein the first place. Thinking and learning
were iust two aspectsof the samething in the brain, Holland wasconvinced.
And he wanted to capture that fundamental insight in his adaptiveagent;
Masterof the Came 181

Forall of that, however,Hollandwasn'taboutto gobackto doingneural

networksimulations, Evena quarter-century aftertheIBM 701,computers
werestill not powerfulenough to do a full-fledged Hebbiansimulationon
the scalehe wanted.True, neuralnetworks had enioyed a brief flurry of
fame in the 1960sunder the rubric of "p€rceptrons"-neural networks
specializedto recognizefeaturesin a uisGi-friiiL--B-ut perceptrons were
highly, highly simplifiedversionsof whatHebbhad actually been talking whrl\
about,and couldn'tproduceanythingresembling a resonating cell assem- {'fl^'t
bly. (They alsoweren'tvery goodat recognizing visualfeatures,which is
whytheyhadfallenout of favor.)Nor wasHollandmuchmoreimpressed
with the newergenerationof neural networks,which were iust coming
backinto fashionin the late 1970sandwhich havegottena lot of attention

]H
in the yearssincethen. Thesenetworksaresomewhatmoresophisticated
than perceptrons, saysHolland.But theystill couldn'tsupportcell assem-
blies.Indeed,mostversionshaveno resonance at all; the signalscascade
throughthe networkin one directiononly, front to back."Theseconnec-
tionist networksare very goodat stimulus-response behaviorand pattern
recognition,"he says."But by and largetheyignorethe needfor internal
feedback, which is whatHebbarguedyou neededfor cell assemblies. And
with fewexceptions, theydon'tdo muchwith internalmodels. "
The upshotwasthat Hollanddecidedto designhis simulatedadaptive
agentasa hybrid, takingthe bestof both worlds.For computationaleffi-
ciencyhe would go aheadand usethe kind of ifthen rulesmadefamous
by expertsystems. But he wouldusethem in the6lirit of neuralnetworks.
[la35rher *,,'*^ s

Actually,saysHolland,therewasa lot to like aboutif-thenrulesin any

case.In the late 1960s,long beforeanyonehad evenheardof an expert
system,rule-based systemshadbeenintroducedbyCarnegie-Mellon's Allen
NewellandHerbertSimonasa general-purpose computermodelof human
cognition.Newell and Simon saweachrule ascorresponding to a single
packetof knowledgeor a singlecomponentof skill: "lf Theeg is a bird,
thenTwee$haswings,"for example,or "If there'sa choicebetweentaking
your opponent's pawnand his queen,then takethe queen."Moreover,
theypointedout thatwhena program's knowledge is expressedin thisway,
it automaticallyacquiressomeof the wonderfulflexibilityof cognition.
The condition-action structureof the rules-"lf thisis the case,then do
\ft41"-lngxns that theydon't executein a fixedsequence like somesub-
routinewrittenin FORTRANor PASCAL.A givenrulecomesto life only
when its conditionsare met, so that its responseis appropriateto the
 182 COMPLEXITY
i situation.Indeed,oncea rule is activated it will verylikelyhiggera whole
f sequence of rules:"lf A then B," "lf B then C,. -If C then e" and so
lon-in effect,a wholenewprogramcreatedon the fly and tailoredto the
\poblem at hand.And that,not theblind,rigidbehaviorof a wind-uptoy,
{isexactlywhatyou wantfrom intelligence.
Furthermore,saysHolland,rule-based systems turn out to makea lot of
sense in termsof the neuralarchitecture of thebrain.A rule,for example,
justthe computerequivalentof oneof Hebb'sresonating
- is cell assemblies.
("ln Hebb'sview," he says,"a cell assembly makesa siniplestatement: If
I such and such an eventoccurs,then I will fire for a while at a high rate.',
I The interactionsof the rules,with the activationof one rule settingoff a
I whole cascadeof others,are likewisea naturalresultof the dense-inter-
I connectedness of thebrain,"Eachof Hebb'scellassemblies involves about
onethousandto ten thousandneurons,"saysHolland."And eachof those
neuronshasaboutone thousandto ten thousandsynapses connectingit to
otherneurons.Soeachcell assembly contacbc lof of othercell assembilies. "
In effect,he says,activatingone cell assembly will posta message on a
kind of internalbulletin board,whereit can be seenby mostor all of the
otherassemblies in thebrain:"Cell Assembly 295834108 nowactive!"And
when that message appears,thoseassemblies that are properlyconnected
to the firstonewill alsofire andposttheir own messages, causingthe cycle
to repeatagainandagain.
The internal architectureof a Newell-Simontype rule-basedsystem
actuallyfollowsthis bulletin-board-metaphor quiteclosely,saysHolland.
There is an internaldatashucturethat corresponds to the bulletin board
andcontains a seriesofdigitalmessages. And thenthereisa largepopulation
of rules, bits of computercode that number in the hundredsor even
thousands.When the systemis in operation,eachof the rulesconstantly
scansthe bulletin boardfor the presence of a message that matchesits "if "
part.And wheneveroneof themfindssucha message, it immediatelyposts
a new digital message specifiedby its "then" part.
"Think of the systemas a kind of office," saysHolland. "The bulletin
boardcontainsthe memosthat areto be processed that day,and eachrule
corresponds to a deskin that officethat hasresponsibility for memosof a
givenkind. At the beginningof the day,eachdeskcollectsthe memosfor
which it is responsible. And at the end of the dayltaeh-affiio.-tr tne
memosthatresultfrom itsactivities." In the morning,of course,thecycle
illEE-Jn addition,he says,someof thememosmaybepostedby detectors,
*hi"h k.p the systemup to dateaboutwhat'sgo,T!;G-ih;.Gia;-
world. And still othermemosmayactivatge.fccton*subroutines that allow
Master of the Came 183

the systemto affect the outsideworld. Detectorsand effectorsare the

computeranalogof eyesand muscles,saysHolland' So, in principle, a
rule-basedsystemcan easilyget feedbackfrom its environment-one of
his prime requirements.
Hollandaccordingly usedthissamebulletin-board metaphor in the de-
signof his own adaptiveagent. Having done that, however, he went right
backto beingan iconoclast when it came to the details.
In the standard Newell-Simonapproach, for example,boththe rulesand
the memoson the bulletin boardwere supposed to be writtenin termsof
symbolssuch as "Bird" or "Yellow," which were intended to be the analog
of concepts in the human mind. And for most people in artificialintelli-
genceresearch,this useofsyqbols to represent concepts was utterlynon-
controversial. It hadbeenstandarddoctrine in the field for decades-with
Newelland Simonbeingamongits mostarticulate champions. Moreover,
it did seemto capturemuchof whatactuallygoeson in our heads.Symbols
in the computercouldbe linkedinto elaborate datastruchrres to represent
a complexsituation,just as conceptsare linked and mergedto form the
psychologists' mentalmodels.And thesedatastructures,in turn, couldbe
manipulatedby the programto emulatementalactivitiessuchasreasoning
and problem-solving,iust as mental modelsare remoldedand changed
by the mind during thinking. Indeed, if you took the Newell-Simon
viewliterally,asmanyresearchers did, thiskind of symbol-processing was
thinking.
Yet Holland iust couldn't buy it. "Symbol-processing wasa very good
placeto start,"he says."And it wasa realadvance in termsof understanding
conscious thoughtprocesses. " But*symbols by thelgset^1eg ryerefartoo4gi{,
and they left out far too much. How could a data registercontaining the
characters B-I-R-D really captureall the subtle,shiftingnuancesof that
concept?How couldthosecharacters reallymeananythingto the program
if it had no wayto interactwith realbirdsin the outsideworldTAnd even
leavingthat issueaside,wheredo suchsymbolicconceptscomefrom in
the first place?How do they evolveand grow?How are they moldedby
feedbackfrom the environment?
To Holland,it wasall of a piecewith the mainstream's lackof interestf
in learning."Yourun into thesamedifficultiesyoudo by classifying speciesl
withoutunderstanding how theyevolved,"saysHolland."You can learn
a lot that wayaboutcomparative anatomyandsuch.But in the end, it iust
doesn'tgo far enough."He wasstill convincedthat.concepts had to be
understoodin Hebbianterms,asemergentstructures ffiifi!;fro- rome
deeperneuralsubshate that is constantlybeingadjustedand readjusted by
 1U COMPLEXITY

llinput from the environment. Like cloudsemerging from the physicsand

flchemistry of ivater
vapor, concepts are fuzzy, shifting,dynamicthings.
lf They areconstantly recombining andchangingshape."The mostcrucial
ll thing we've to got getat in understanding complexadaptivesystems is how
I levelsemerge,"he says."lf you ignorethe lawsat the next level below,
I you'll neveFbeableto understand thisone."
To capture thatsense of emergence in hisadaptive agent,Hollanddecided
that his rulesand messages would nof be writtenin termsof meaningful
symbols. They wouldbe arbitrarystringsof binaryI's and0's.A message
mightbe a sequence suchas 10010100, muchlike a chromosome in his
geneticalgorithm.And a rule,asparaphrased in English,mightbe some-
thing like, "If there is a message on the bulletin boardwith the pattern
'don't care,'then post the message
l###0#00, where # standsfor
0 l I l 0 l 0 l ."
This represenlation wasso offbeatthat Hollandeventookto callinghis
rulesby a new name,"classifiers," because of the waytheir if-conditions
classified differentmessages accordingto specificpatterns of bits. But he
considered this abstractrepresentationessential,if only because he'd seen
toomanyartificialintelligence researchers foolthemselves aboutwhattheir
symbol-based programs"knew."In his classifiersystems, the meaningof a
message would haveto emergefrom the wayit causedone classifierrule
to triggeranother,or from the factthatsomeof its bitswerewrittendirectly
by sensors lookingat the real world.-Concepts and *qtl|gdg!9 y-or1ld
liiewisehav"leemerseassclLruppadi;ElilsteEdms:;"rtt"ti"h;b;ra
frGGrably organizeand reorganize in much the samewayas
themselves
autocatalytic sets.
Meanwhile,Hollandwasalsotakingexception to thestandard ideasabout
centralizedcontrol in a rule-basedsystem.According to the conventional
wisdom,rule-based systems weresoflexiblethatsomeform of centralized
control was neededto prevent anarchy.With hundredsor thousandsof
ruleswatchinga bulletin board crammedwith messages, therewasalways
thechancethatseveral ruleswould suddenly hop up and startarguingover
who got to postthe nextmessage. The presumption was that they couldn't
all do so,because theirmessages mightbe utterly inconsistent. ("Thke the
queen." "Takethepawn.") Or theirmessages might lead to entirelydifferent
cascades of rules,and thusto entirelydifferentbehaviorof the systemas
a whole. So to preventthe computerequivalent of schizophrenia, most
systems implementedelaborate "cog$gllg9lutiol" strategies to makesure
thatonly one rule couldbe activeat a time.
Masterof the Game 185
Holland, however,sawsuch asprecisely
conflictresolution the
wTg_qg3jl_lggglsthe worldsucha simple L.taI;G;ffiCiHti-
alwavsknowfre
always Hard And if the systemftasbeen
Hardly.
the bestrule in advance?
ITX
told what to do in advance, then it's a fraud to call the thing artificial'
intelligence:the intelligenceisn't in the programbut in the plogrammer.
No, Holland wanted control to be leamed. He wanted to seeit emerging
from the bottom up, iust as it did from the neural substrateof the brain.
Consistencybe damned: if two of his clasifier rules disagreedwith one \
another, then let them fight it out on the basisof their performance,their I
proven conhibution to the taskat hand-nof somePreprogrammedchoice I
made by a softwaredesigner.
"ln contrast to mainstream artificial intelligence, I see competition as-
much more essentialthan consistency,"he says.Consistencyis a chimera,
b."""*i";comFli-cafed *oild thetl is no guaranteethat experiencewill
be consistent.But for agentsplaying a gameagainsttheir environment,
competition is forever."Besides,"saysHolland, "despiteall the work in
economicsand biology, we still haven't extractedwhat's central in com-
petition." There's a richness there that we've only lust begun to fathom.
bonsider the magical fact that competition can produce a very strong in-
centive for cooperation,as cettain playersspontaneouslyforgealliancesand
symbiotic relationshipswith each other for mutual support. It happensat
every level and in every kind of complex, adaptivesystem,from biology
to economics to politics. "Competition and cooperation may seem anti-
thetical," he says,"but at some very deep level, they are two sidesof the
samecoin."
To implement this competition, Holland decidedto make the posting of
messages into a kind olauctiog*His basicidea wasto think of the classifiers
not as computer commands but as hypotheses,coniechrresabout the best
messagesto post in any given situation. By assigningeach hypothesisa
numerical value measuringits plausibility, or strength,he then had a basis
for bidding. In Holland's versionof messaggposting, each cycle startediust
as before, with all the classifiersscanning the bulletin board in search of
a match. And iust asbefore, thosethat found a match would stand up and
get ready to post their own messages.But insteadof posting them imme-
diately, each one would first shout out a bid proportional to its shength.
A classifieras solidly grounded in experienceas "The sun will rise in the
easttomorrowmorning" might bid 1000,while a classifieraswell-grounded
as "Elvis is alive and appearingnightly at the Walla Walla Motel 6" might
bid onlv l. The systemwould then collect all the bids and choosea set of
186 COMPLEXITY
winners by lottery, with the highestprobabilityof winning going to the
highestbidders.The chosenclassifiers would posttheir messages, and the
cycle would repeat.
Complex? Holland couldn't deny it. As things stood, moreover,the
auctionsimply replacedarbitraryconf ict resolutionshategies with arbitrary
plausibility values. But assumingfor the moment that the systemcould
somehowlearn theseplausibilityvaluesfrom experience,then the auction
would eliminate the central arbiter and give Holland exactly what he
wanted. Not every classifiercould win: the bulletin board was big, not
infinite. Nor would the race alwaysgo to the swift even Elvis might get a
chance to post his messageif he got a lucky break. But on the average,
control over the system'sbehavior would automatically be given to the
strongestand most plausible hypotheses,with off-the-wall hypothesesap-
pearing just often enough to give the systema little spontaneity.And if
someof thosehypotheses were inconsistent,well, that shouldn'tbe a crisis
but an opportunity, a chancefor the systemto learn from experiencewhich
onesare moreplausible.

So once again, it all came back to learning:How were the classifiers

supposedto prove their worth and earn their plausibility values?
To Holland, the obvious answerwas to implement a kind of Hebbian
reinforcement. Whenever the agentdoessomethingright and g.tr-ffi;dF
feedbackfrom the environment, it should strengthenthe classifiersrespon-
sible. Whenever it doessomethingwrong, it should likewiseweakenthe
classifiersresponsible.And either way, it should ignore the classifiersthat
were irrelevant.
The trick, of course,wasto figure out which classifierswere which. The
agent couldn't just reward the classifiersthat happen to be active at the
moment of payoff. That would be like giving all the credit for a touchdown
to the player who happened to carry the ball acrossthe goal line-and
none to the quarterbackwho called the play and passedhim the ball, or
to the linemen who blocked the other team and opened up a gap for him
to run through, or to anyone who carried the ball in previous plays. It
would be like giving all the credit for a victory in chessto the final rnove
that trapped your opponent's king, and none to the crucial gambit many
moves before that set up your whole endgame. And yet, what was the
alternative?Ifthe agent had to anticipatethe payoffin order to reward the
correct classifiers,how was it supposedto do so without being prepro-
Master of the Came 187
grammed?How wasit supposed to learn the value of thesestage-setting
moveswithout knowingaboutthem already?
Good questions.Unfortunately,the generalideaof Hebbianreinforce-
ment wastoo broad-brush to provideanyanswers. Hollandwasat a loss-
until one day he happenedto think backon the basiceconomicscourse N.+
he'd takenat MIT from PaulSamuelson, authorof the famouseconomicsA tit*
textbook,and realizedthat he'd almostsolvedthe problemalready.By }r*at.
auctioningoff spaceon the bulletinboard,he had createda kind of mar-
ketplace within the system.By allowingthe classifiersto bid on the basis
oftheir strength,he hadcreateda currency.Sowhy not takethe nextstep?
Why not createa full-fledgedfree-marketeconomy,and allow the rein-
forcementto takeplacethroughthe profit motive?
Why not, indeed?The analogy was obvious when you finally saw it. If
you thought of the messagesposted on the bulletin board as being goods
and servicesup for sale, Holland realized,then you could think of the
classifiersas being firms that producethosegoodsand services.And when
a classifierseesa messagesatisfyingits if-conditions and makesa bid, then
you could think of it as a firm trying to purchasethe suppliesit needsto
make its product. All he had to do to make the analogy perfect was to
arrangefor eachclassifierto-payfor the suppliesit used.When a classifie{
won the right to postits m"r'r6$, he decided,it wouldtransfersomeof its
strengthto its suppliers:namely,the classifiersresponsible
for postingthe
messages that triggeredit. In the process,the classifierwould then be
weakened.But it would have a chanceto recoup its strengthand even make
a profit during the next round of bidding, when its own messagewent on
the market.
And where would the wealth ultimatelycome from? From the 6nal
consumer,of course:the environment,the sourceof all payoffsto the
system.Exceptthat now, Hollandrealized,it would be perfectlyall right
to rewardthe classifiers that happento be activeat the momentof payoff.
Sinceeachclassifierpaysits suppliers,the marketplace will seeto it that
the rewardspropagate throughthe wholecollectionof classifiers and pro-
duceexactlythe kind of automaticrewardandpunishmenthe waslooking
for. "lf you producethe right intermediateproduct,then you'll makea
pro6t,"he says."lf not, thennobodywill buy it andyou'llgo bankrupt."
All the classifiers
that leadto effectiveactionwill be shengthened, and yet
none of the stage-setting classifierswill be neglected.Over time, in fact,
as the systemgainsexperience and getsfeedbackfrom the environment,
the shengthof eachclassifier will cometo matchitshue valueto theagent.
 188 COMPLEXITY
Hollanddubbedthisportionof his adaptive agentthe "bucket-brigade',
algorithmbecauseof the wayit passedrewardfrom each;'EGi6;;-the
previous classifier.It wasdirectlvanalogolrlt9 thestrengthening of synapses
in Hebb'stheoryof the brain-or, for that mafter,to thc kind of reinforce-
lSru$$d to train a simulatedneuralnetworkin a ma;ffi
he hai-FF,itland-I"new ir€;;s; mosfI6d. Econonii "fi;fiimar
c rii nforcement
via the profit motive wasan enormouslypowerfulorganizingforce, in
much the samewaythat Adam Smith'sInvisibleHandwasenormously
powerfulin the real economy.In principle,Holland realized,you could
startthesystem off with a setof totallyrandomclassifiers, sothatthe agent
just thrashedaroundlike the softwareequivalentof a newbornbaby.And
then, as the environment reinforced certainbehaviors and asthe bucket
brigadedid its work,you couldwatchthe classifiers organizethemselves
into coherentsequences that would produceat leasta semblanceof the
desiredbehavior.Learning,in short,wouldbe built into the systemfrom
the beginning.
So,Hollandwasalmosthome-but notquite.Byconstructing thebucket-
brigadealgorithmon top of the basicrule-based system,Hollandhadgiven
his adaptive agentone form of learning.But therewasanotherform still
missing.It wasthe differencebetweenexploitationandexploration.The
bucket-brigade algorithmcould strengthenthe classifiers that the agent
alreadypossessed. It could honethe skillsthatwerealreadythere.It could
consolidate the gainsthat had alreadybeenmade.But it cpuldn'tcreate-
anythingryIv, By itself,it couldonly ieadthe systeminto higTii6fiiffi'ile-
;ffi;b. il n"a no wayto explorethe immense,p"". o1porribl. .,.*
classifiers.
This, Hollanddecided,wasa job for the geneticalgorithm.When you
thoughtaboutit, in fact, the Darwinianmetaphor and the Adam Smith
metaphorfit togetherquitenicely:Firmsevolveovertime, sowhyshouldn't
classifiers?
Holland certainlywasn'tsurprisedby this insight;he'd had the genetic
algorithmin the backof his mind all along.He'd beenthinkingabout
it when he first set up the binary representation of classifiers. A classi-
fier might be paraphrased in Englishas somethinglike, "If there are
two messages with the patternsl###0#00 and0#00####, then post
the message 01110101."In the computer,however,its variousparts
would be concatenated togetherand written simply as a string of bits:
^l###0#090#00####r0lll0l0l." And to the geneticalgorithm,that
lookediust like a digitalchromosome. So the algorithmcouldbe carried
out in exactlythe sameway.Mostof the time, the classifiers wouldmerrily
Master of the Came 189

buy and sell in their digital marketplaceas before.But everyso often, the
systemwould selecta pair of the shongestclassifiersfor reproduction' These
classifierswould reshuffe their digital building blocls by sexualexchange
to producea pair of offspring.The offspringwould replacea pair of weak
classifiers.And then the offspringwould havea chanceto prove their worth
and grow strongerthrough the bucket-brigadealgorithm'
The upshot was that the populationof rules would changeand evolve
over time, constantly exploring new regions of the spaceof possibilities.
And there you would have it: by adding the geneticalgorithm as a third
layer on top of the bucket brigadeand the basicrule-basedsystem,Holland
could make an adaptive agent that not only learned from experiencebut
could be spontaneousand creative.
And all he had to do was to turn it into a working program.

Hollandstartedcodingthe firstclassifier systemaround1977' And oddly

enough,it didn't turn out to be as straighforward a iob ashe hadhoped.
"I reallythoughtthat in a coupleof monthsI'd havesomething up and
running that wasusefulto me," he says. "Actually, it was the betterpart
of a yearbeforeI wasfully satisfied."
On the otherhand, he didnt exactlymakethingseasyfor himself.He l'
codedthat firstclassifiersystemin trueHollands$le:by himself.At home. 1
In hexadecimal code,the samekind that he'dwrittenfor the Whirlwind
i
thirty yearsearlier.On a Commodorehomecomputer.
Holiandt BACH colleagues still roll their eyeswhentheytell this story.I
The wholecampuswascrawlingwith computersr VAXs,mainframes, even;
high-powered graphicsworkstations. Why a Commodore? And hex!Almost,
nobodywrotein hex anymore.If you werereallya hard-core computer
,'.
iock trying to squeeze the last ounce of performance out of a machine,
you might write in somethingcalledassemblylanguage,which at leastt
replaced the numberswith mnemonics like MOV tMZ, andSUB.Other-1
wise,you wentwith a highJevellanguagesuchasPASCAL,C, FORTRAN, I
or LlSP-somethingthata humanbeingcouldhopeto understand. Cohen,
in particular, remembers arguinglongandhardwith Holland:Who'sgoing I
to believethat this thing worla if it's written in alphanumericgibberish?,
And evenif anybody doesbelieveyou,who'sgoingto usea classifier system:
if it only runs on a homecomputer?
Hollandeventuallyhadto concedethe point-although it waswell into I
the early 1980sbeforehe agreedto handoverthe classifiersystemcodeto I
a graduatestudent,Rick Riolo, who transformed it into a general-purpose
190 COMPLEXITY
package that would run on almostanytypeof computer."This is iust not
my instinct,"Hollandadmits."My tendency is to do piecesof something
until I seeit can reallybe implemented.Then I tendto loseinterestand
go backto theory."
Be that asit may,he still maintainsthat the Commodoremadea lot of
senseat the time. The campuscomputershad to be shared,he explains,
and that madethem a pain: "l wantedto fusswith the programonJine,
and nobodywaslikely to give me eight hoursat a stretch."Holland saw
the personalcomputerrevolutionasa godsend."l realizedthat I coulddo
my programmingon my machine,that I could haveit in my own home
andbe beholdento nobody."
Besides, havingcomeof ageprogramming WhirlwindandtheIBM 701,
Hollanddidn't find theselittle desktopmachines primitiveat all. When
he finally got the Commodore,in fact, he considered it quite a stepup.
He hadactuallytakenhis firstplungeinto personalcomputingwith some.
thing calledthe Micromind,which he boughtin 1977whenit lookedlike
a seriousrival to the brand-newApple II. "lt wasa very nice machine,"
he recalls.True, it wasn'tmuchmorethana bunchof circuitboardsin a
blackbox that could be hookedup to a teletypemachinefor input and
output.It had no screen.But it did have8 kilobytesof 8-bit memory.And
it costonly $3000.
And asfor hex-well, theMicrominddidn't haveanyotherprogramming
Ianguages availableat that point, and Hollandwasnot aboutto wait. "I
wasusedto writing in assembler," he says,"and I coulddo hex almostas
easilyas I could assembler, so it wasn'thard."
All told, Holland says,it's reallytoo badthat the Micromindcompany
went bankruptso quickly. He movedto the Commodoreonly when he
beganto feel the constrainbof that tiny 8 kilobytememory.At the time,
it wasthe ideal choice,he says.It usedthe samemicroprocessor chip as
the Micromind, which meant that it could run his hexadecimalcode
virtuallyunchanged.It hadmuch morememory.It hada screen.And best
of all, he says,"The Commodorewould let me playgames."

His colleagues'exasperation
aside,Holland's6rst classifiersystemran
well enoughto convincehim that it reallywouldworkthewayhe intended
it to-and, not incidentallXthatit reallydid holdtheseeds
of a full-fedged
theoryof cognition.In testsof an earlyversionof the system,which he
publishedin 1978in collaborationwith Michigan psychologyprofessor
|udy Reitman,their agentleamedhow to run a simulatedmazeaboutten
Master of the Came 191

timesfasterwhen it usedthe geneticalgorithmthan it did withoutthe I

algorithm.The sametestsalsoprovedthat a classifier systemcouldexhibit I
what psychologists call trander:it could applyruleslearnedin one maze
I
to run othermazeslateron.
Theseearlyresults wereimpressive enoughthatwordof classifier systems
beganto spread, evenwithoutHollandpushing them.In 1980,for example,
StephenSmith at the Universi$of Pittsburghbuilt a classifiersystemthat
couldplaypoker,and pittedit againstan olderpoker-playing programthat
wasalsoableto learn.It wasn'tevena contest; the classifier systemwon
in a walk. In 1982,Slewail{ll1,l_ of the PolaroidCorporationuseda
systemto cooidinatethemotionof a TV cameraanda mechanical
classifier
arm. He showedthat the bucket-brigade and geneticalgorithmscauseda
spontaneous organizationof the classifierrules, so that they segregated
themselves into groupsthat couldfunctionascontrolsubroutines andpro-
ducespeci6c,coordinated actionsasneeded. Alsoin 1982,Holland'sstu-\
dentLashonBookercompleted a Ph.D.thesisin whichheplaceda classifier \
systemin a simulatedenvironmentwhereit had to find "food" and avoid I
"poison."The systemsoonorganizedits rulesinto an internalmodel of I
thatenvironment-in effect,a mentalmap. )
For Holland, however,tte m6Fliilffig demonstration wasthe one
producedin 1983by D. a Ph.D,-boundcivil engineerwho
"ufdSeldkg,
adaptive systems courseseveralyearsbeforeand
had enrolledin Holland's
hadbecomea truebeliever. Persuading Holland to cochairhisdissertation
committee,Goldberg wrote a thesisdemonstrating how geneticalgorithms
and classifiersystems could be used to control a simulated Bj=UEl!!g' At
the time it wasby far the most complex problem that a classifier system
had everbeenpresented with. The obiective in any pipeline system is to
meetdemandat the end of the pipelineaseconomically as possible' But
a pipeline consistsof dozensor hundredsof compressors pumping gas
throughthousands of milesof large-diameter pipe.The customers'demand
for gaschanges on an hourlyand seasonal basis.Compressors and pipes
springleaks,compromising the system's abilityto deliver gas at the appro-
priatepressure. Safetyconstrainbdemandthat pressure and fow rateshave
to be keptwithin properbounds.And everything affectseverything else.
Optimizingevena very simplepipelineis far beyondthe reachof math-
ematicalanalysis.Pipelineoperators learntheir craftthroughlongappren-
ticeships-and then "drive" their systemby instinctand feel, the waythe
restof us drivethe family car.
The pipelineproblemseemed soinhactable,in fact,thatHollandfretted
that Goldbergmight havebittenoff morethanthe classifiersystems could
192 COMPLEXITY
chew. He needn't have worried. Goldberg'ssystemlearned to operatehis
simulated pipeline beautifully: starting from a set of totally random clas-
sifiers,it achievedexpert-levelperformancein about 1000daysof simulated
experience.Moreover,the systemwasincrediblysimplefor what it did. Ib
messages were only 16 binary digits long, ib bulletin board held only 5
messages at a time, and it containedonly 60 classifierrules, total. In fact,
Goldbergran the whole classifiersystem,plus the pipelinesimulation,on
his Apple II computer at home with iust 64 kilobytesof memory. "He's a
guy after my own heart," laughsHolland.
The pipeline simulation not only earnedGoldberga Ph.D. in 1983, it
won him a PresidentialYoungInvestigator Awardin 1985.Holland himself
considers his work a milestone for classifier systems."lt was very con-
vincing," he says."lt really worked on a real problem-or, at least, a
simulation of one." In a deliciousirony, moreover,this most "practical"
of the classifiersystemsdeviseduntil that time also turned out to have the
most to say about basic cognitive theory.
You could seeit most clearly in the way Coldberg'ssystemorganizedits
knowledgeabout leaks,saysHolland. Startingfrom a random set of clas-
sifiers, it would first learn a seriesof broadly applicable rules that worked
quite well for normal pipeline operations. An example that actually ap-
peared in one run was a rule that could be paraphrasedas "Always send a
'No
leak' message."Clearly, this was an overgeneralrule that worked only
if the pipeline was normal. But the systemdiscoveredthat fact soon enough
when Goldberg startedpunching simulatedholes in varioussimulatedcom-
pressors.Its performance immediately declined drastically. However, by
means of the genetic algorithm and the bucket brigade, the systemeven-
tually recoveredfrom ib errors and startedproducing more specific rules
such as "If the input pressureis low, the output pressureis low, and the
rate of changeof pressureis very negative,then sendthe'Leak'message."
Whenever this rule applied, moreover, it would give a much strongerbid
than the first rule and knock it right off the bulletin board. So, in effect,
the first rule governed the dghgllbehavior of the systemunder normal
conditions, while the secondrule and otherslike it would kick in to give
the correctbehaviorunder exceptionalconditions.
Holland was hemendouslyexcitedwhen Goldbergtold him about this.
In psychologythis kind of knowledge organization is known as a default
hierarchy, and it happenedto be a subiectthat wasvery much on Holland's
mind at the time. Since 1980, he had been involvedin an intensecollab-
oration with three colleaguesat Michigan-psychologists Keigf,I{g\93k
and Ric.lag!*[!s!e{ and philosopher Pllf Thagard-to build a general
 Master of the Came 193

:gg@lbg+sf leamine,res$oninc.
an{ tntelgqdiscc'vgv. Asthev
Iaterrecounted
in their l2!9lg&-nAucti,qrr,illTour ofiEem-t-ad
inde-
pendentlycometo believethat sucha theoryhad to be foundedon the
three basicprinciplesthat happenedto be the samethreethat underlay
Holland'sclasifier system:namely,that knowledgecan be expressed in
termsof mentalstructures thatbehaveverymuchlike rules;thattheserules
are in competition,sothat experience causes usefulrulesto growstronger
and unhelpfulrulesto growweaker;and that plausiblenew rulesaregen-
eratedfrom combinations of old rules.Their argument,which theybacked
up with extensive observations andexperiments, wasthattheseprinciples
could accountfor a wide varietyof "Ahal" type insights,rangingfrom
Newton'sexperiencewith the appleto such everydayabilitiesas under-
standingan analogy.
In particular,they arguedthat thesethreeprinciplesoughtto causethe
sP9lltanri!)ulrmergetSC-of -{glaglt hierarchiesas the basicorganizational
structureof all humanknowledge-iiiriaeertneyappearto do. The cluster
of rulesforminga defaulthierarchyis essentially synonymous with what
Hollandcallsan ietemel model.We useweakgeneralruleswith shonger
exceptionsto makepredictionsabout how thingsshouldbe assignedto
categories: "If it's sheamlinedand hasfins and livesin the water,then it's
a fish"-but "lf it also has hair and breathesair and is big, then it's a
whale."We usethe samestrucfureto makepredictionsabouthow things
shouldbe done "It's always'i' before'e' exceptafter 'c' "-but ,,If it's a
wordlike neighbor, weigh,or weird,thenit's'e'before'i."'And we use
thesameshuctureagainto makepredictions aboutcausality: "lf youwhistle
to a dog,then it will cometo you"-but "If thedogis growlingandraisir,rg
its hackles,then it probablywon'tcome."
The theory saysthat thesedefault-hierarchy modelsought to emerge
whetherthe principlesare implementedasa classifiersystemor in some
otherway,saysHolland.(In fact,manyof thecomputersimulations quoted
inlnduction weredonewith PI, a somewhat moreconventional rule-based
programdevisedby Thagardand Holyoak.)Nonetheless, he says,it was
thrilling to seethe hierarchies actuallyemergein Goldberg's pipelinesim-I
ulation.The classifier systemhadstartedwith nothing.Itsinitiaisetof rulesI
had been totally random,the computerequivaleniof primordialchaos.I
And yet, herewasthis marvelousshuctureemergingout of th. chaosto
astonish and surprise them.
"we wereelated,"saysHolland. "lt wasthe first caseof whatsomeone
couldreallycall an emergent model." i
 194 COMPLEXITY
A Place to Come Home To

The kitchen-tableconversation betweenHollandand Arthur wandered

on and on for houts,askitchen-table conversations arewont to do. By the
time theyfinallycalledit a night,theirdiscussion hadgonefrom chessto
economics,from economicsto checkers,andthen on to internalmodels,
geneticalgorithms,and chessagain.Arthur felt he wasfinally beginning
io understandthefull implications of learningandadaptation. And thetwo
of them had rathersleepilybegunto bat aroundan approachthat might
iust crackthis problemof rationalexpectations in economics:insteadof
I assumingthat your economic agents are perfectlyrational,why not iust
I modela bunchof them with Holland-sgle systems
classifier andlet them
[ 1."r.,from experience like recleconomic agents?
Why not indeedeBeforehe turnedin, Hollandmadea noteto dig out
an old setof overheadtransparencies on Samuel'scheckerplayerthat he
happenedto havewith him. Arthur had beenenchantedwith the ideaof
a la-e-playingprogramthat learned; he'd neverheardof sucha thing'
Hollandlhoughfhe might give the meeting participantsan impromptutalk
on the subiectthe next daY'

The talkwasa hit-especiallywhenHollandpointedoutto hisaudience

that Samuel'sprogramwasstill prettymuchthe stateof the art in checkers
playingro*. ihitty yearslater.But, then, Holland'swholeapproachhad
L"." hit at the meeting.Nor weresuchimpromptuinterchanges at all
"
unusualby that point. Participants find it hard to pinpoint exactlywhen
themoodof theeconomics meetingbeganto change.But somewhere about
the third day,after they had clearedawaythe earlybarriersof iargonand
mutualconfusion,the meetingbeganto catchfire'
,,I found it very exciting,"saysstuart Kauffman,who felt primedfor
economicsaftertwo *e.ks of talkingto Arthur. "ln a funnywayit waslike
kindergarten, when you get exposedto all sortsof new thingslike finger
paintirig.Or it wastit. Uiing a pupPy,runningarounds11ffinsat things'
with this wonderfulsenseof discovery,that the wholeworldwasthis won-
drousplaceto explore.Everythingwasnew.And somehowthat'swhatthis
meetingwaslike to me. wonderinghow theseotherguysthink. what are
the criieria?What are the questionsin this new field?That'svery much
my style, personally.But I think it had that flavorfor a fair number of
people.We went on talkingat one anotherlong enoughto hearone an-
othet."
 Master of the Came 195

lronically,consideringthe physicists'earlyskepticism aboutmathematical

abstractions,it wasmathematicsthat providedthe common language."As
I look back on it, I think Ken madethe right decision," saysEugenia Singer,
who had originally been disappointedat Arrow's failure to include soci-
ologistsand psychologistsin the group. "He had the most highly, technically
trained economistshe could get. And as a result, there was a credibility
that was built. The physicalscientistswere amazedat their technical back-
ground. They were familiar with a lot of the technical concepts,even some
of the physicalmodels.So they wereableto startusingcommon termsand
building alanguagethey could talk to each other in. But if they had gotten 1
a lot of social scientistsin there with no technical background,I'm not'
sure the gulf could haie been crossed. /
After most of the formal presentationswere finished, the participantsat
the workshop startedbreaking up into informal working groupsto focus on
particular subjects.One of the most popular topics wasgfugg the domain
of one group that frequently gatheredaround David Ruelle in the small
conferenceroom. "All of us knew about chaosand had readarticles,"says
Arthur. "Some of the economistshad done considerableresearchin that
area. But I remember that there was an awful lot of excitement in seeing
some of the physicists'models."
Anderson and Arthur, meanwhile, were part of a group that met out on
the terraceto discusseconomic"patterns"such astechnologicallock-in or
regional economic differences. "I was almost too tired to do an awful lot
of talking or listening,"saysArthur. "I usedthe workinggroupto quiz phil
Andersonon variousmathematicaltechniques."
Arthur actually found himself feeling very much in tune with Anderson
and the other physicists."l liked their emphasison computerexperiments,"
he says.Among economists,computer models had gotten a very bad odor
back in the 1960sand 1970sbecauseso many ofthe early oneshad been
rigged to give results supporting the programmers'favorite policy recom-
mendations. "So it fascinatedme to seecomputer modeling used properly
in physics. And I think the opennessof the Eeld appealedto me. It was
intellectually open, having a willingness to look at new ideas, and being
nondogmatic about what was acceptable."
Arthur wasalso gratified to find that increasingreturns wasmaking quite
an impact at the meeting. Quite asidefrom his own presentation,a number
of the other economists had been thinking about it independentry.one
day, for example, the participanb listened via telephone link to a iecture
by Harvard emeritus professorflollis chenery, who had fallen too ill to
travel. chenery's lecture wasabou['pitternsof development-why countries
 196 COMPLEXITY
show differences in how they grow, especiallyin the Third World. And
during it, he mentioned increasingreturns. "So after he hung up," says
'Hollis
Arthur, "Arrow jumped up to the blackboard,and said, Chenery
mentionedincreasingreturns.Let me tell you more about it'-and spon-
taneously gave an hour and a half lecture on the history of thought on
increasing returns, along with what it had to say in trade theory, with no
notes whatsoever.I would never have suspectedthat Arrow knew so much
about the subject."
It was just a few dayslater that jos6 Scheinkman, who had alreadydone
seminal work applying increasingreturns to international trade, stayedup
until threein the morning alongwith UCLAT Michele Boldrin to formulate
a theory of economic developmentunder increasingreturns.
Inevitably, saysArthur, there was also a discussionabout whether the
stock market could get into a positive feedbackloop, with stocksbeing bid
higher and higher just becausepeopleseeother investorscoming in. Or,
conversely,could there be an oppositeeffect, a crash, if people saw other
investorsgetting out? "Civen that the market was somewhatoverheatedat
the time," saysArthur, "there was quite a lot of discussionof whether that
wasfeasible,whether it did happenin reality-and whether itmight happen
soon."
The consensuswas "Maybe." But the posibility seemedreal enoughto
David Pines that he called his broker with an order to sell off some of his
stock.The broker talked him out of it-and a month later, on October 19,
1987, the Dow fell 508 points in a day.
i "That led to this rumor that the conference had predicted the stock
l' market collapsea month beforeit happened,"saysArthur. "We didn't. But
the crashcertainlyhad this positivefeedbackmechanismthat we had dis-
, cussedat length."

And so it went: a ten-day marathon with only one Saturdayafternoon

off. Everyone was exhausted-gloriously so. "By the end of the ten daysI
wason a huge scientifichigh," saysArthur. "l couldn't believethere were
peoplewilling to listen."
Indeedthere were. Becauseof a prior commitment to delivera paperin
San Francisco on Friday, September 18, Arthur had had to miss the last
day of the meeting, when the group had scheduleda wraP-uPsessionand
a press conference. (Reed, unable to get away from New York, sent a
congratulatorymessage on video.) But as soonas he walkedin the door of
Master of the Game 197

the convent that next Monday afternoon, Pines came up to him in the
hallwaywith a smile on his face.
"Did the conferenceget over okay?" askedArthur.
"Oh, we're very pleased,".saidPines.Eugenia Singerhad been partic-
ularly enthusiastic, and was preparing a glowing report for Reed. Mean-
while, he added, the scienceboard had met right after the conferencewas
over, and first off, they wanted to invite Arthur to ioin the scienceboard.
Arthur wasastonished.The scienceboardwasthe institute'sinner sanc-
tum, the seatof all real policy-makingpower. "Certainly," he said.
"And there'sbeen a further thought," said Pines."We're very anxious
not to let this opportunity slip. Everybody'sso excitedabout the conference
that we want to expand it into a full-scale researchprogram. We'd been
discussingthat, and we were wondering if you and fohn Holland could
come next year [meaning the next academicyear,twelve months from then]
and get the programup and running."
It took Arthur about two secondsto work that through. The scienceboard
wasaskinghim and Holland to run the program.He stammeredout some-
thing to the effectthat he did havea sabbaticalcoming up, as a matter of
fact, and it soundedlike greatfun. And-yes, he'd be delighted.
"l was enormouslyflattered,"he says,"and I felt very humble indeed.
But running throughout that-and still to this day-was this notion of
'Who,
me?'I mean, this is Phil Anderson,or Ken Arrow, and here I am,
and they're askingme what I think about this or that. So I had the reaction
that-didn't they really mean somebodyelse?Certainly nothing like that
had happenedto me before in my academiclife. "
"You know," he adds, "it's perfectly possiblefor a scientist to feel that
he has what it takes-but that he isn't acceptedin the community. John
Holland went through that for decades.I certainly felt like that-until I
walked into the SantaFe Institute, and all theseincredibly smart people,
people I'd only read about, were giving me the impressionof 'What took
you so long to get here?"
For ten days, he had been talking and listening nonstop. His head was
so full of ideasthat it hurt. He was exhausted.He neededto catch up on
about three weeksof sleep. And he felt as though he were in heaven.
"From then on," he says,"l stoppedworrying about what other econ-
omists thought. The people I cared about sharing my work with were the
people in SantaFe. SantaFe was a placeto come home to."
;s*&* **#**s*136*s 8st*s*sn*i sa+t ttt t*i*a

On Tuesday,September 22, 1987,all too brightandearlyon the morning

after he'd been offered the codirectorship of the SantaFe Institute'snew
economics program, a sleepy Brian Arthur climbedinto the carwith fohn
Hollandand droveup to Los Alamos to visit the @,
a five-dayhappeningthat had started the day before.
Arthur wasa little hazy about what "artificial life" actuallymeant.In
fact, consideringhow exhausted he still felt after the economicsmeeting
of the previousweek,he wasa little hazy about a lot of things' But as
fl Hollandexplained it, artificiallife wasanalogous artificialintelligence.
to
I Th" diff.tence was that, insteadof using computersto model thought
I pro""st.t, you usedcomputersto modelthe basicbiologicalmechanisms
I of evolutionand life itself. It wasa lot like what he'd beentrying to do
with the geneticalgorithmand classifiersystems, saidHolland, but even
morebroad-ranging andambitious.
The wholething wasthe brainchildof a postdocup at the Los Alamos,
Chris l,angton,who had beena studentof Holland'sand Art Burks'at
Michigan.Langtonwassomethingof a latebloomer,saidHolland.At age
thirty-nine,in fact, he wasaboutten yearsolderthan mostpostdocs. And
he still hadn't quite put the finishingtoucheson his Ph.D. dissertation.
But he'd beenan extraordinary sfudent."A veryfertileimagination,"said
Holland. "Very goodat gatheringin experience of all kinds."And he was
puttingtremendous energyinto thisworkshop.Artificial life wasLangton's
baby.He'd inventedthe name.He'd spentmostof the pastdecadetrying
to articulatethe concept.He'd organizedthis workshopto hy to turn
Life at the Edge of Chaos 199
artificial life into a real scientificdiscipline-without evenknowinghow
manypeoplewouldshowup. He'd inspiredenoughconfidence that the
LosAlamosCenterfor NonlinearStudieshadput up $15,000to payfor
the workshop,while the SantaFe Institutehadput up another$5000and
agreedto publishthe proceedings aspartof its newbookserieson com-
plexity.And fromwhatHollandhadseenof theworkshop's kickoffyesterday,
l,angtonwasbringingit off beautifully.It was-well, Arthur would have
to seefor himself.
Indeed,Arthur did. Whenhe andHollandwalkedinto the auditorium
buildingat LosAlamos,he formedtwoimpressions veryquickly.The first
wasthathe'dbadlyunderestimated his housemate. "lt waslike walkingin
with Gandhi," he says."I'd thoughtI wasroomingwith a short,pleasant
computerwhiz. And herepeoplewereheatinghim like the greatguru of
this field: 'lohn Hollandl' Peoplewould rush up to him in the hallway.
What do you think of thisTWhat do you think of that?Did you get the
paperI sentyou?"
Arthur'shousemate tried to takeit all in stride.But therewasno getting
aroundit: muchto his own embarrassment, fohn Hollandwasbecoming
famous,Indeed,therewasn'tmuch he could do to stopit. He'd been
turning out one or two freshPh.D.sper yearfor twenty-fiveyears,so that
by now therewerea lot of believersout therespreading the word,And in
the meantime,the worldhad beencatchingup to him. Neuralnetworks
wereverymuchbackin vogue.And learning,by no coincidence, hadnow
emerged asone of the hottesttopicsin mainstream artificialintelligence.
The firstinternational conference on geneticalgorithms hadbeenheldin
1985,and more werein the offing. "It seemedto be the standardintro-
ductionto everyone's talk,"saysArthur:"fohn Hollandhassuchandsuch
to say.Now here'smy version."
Arthur'ssecondimpression wasthatartificiallife was-shange.He never
did getto meetl,angton,who provedto be a tall, lankyguy with a mane
of brown hair and a rumpledfacethat madehim look strikinglylike a
young, amiableWalter Matthau. Lan$on wasconstantlyon his way to
somewhere else-coping,fixing,worrying,andfranticallytryingto make
it all happen.
So, instead,Arthur spenta goodpartof the daywandering amongthe
computerdemonstrations that had beensetup in the hallwaysaroundthe
auditorium. It wassomeof the damnedest stuff he'd everseen:darting
flocksof animated, electronic birds,strikingly realisticplantsthatgrewand
developed on screenbeforeyoureyes,bizarre,fractal-like creatures, patterns
that undulatedandsparkled. lt wasfascinating. But whatdid it mean?
200 COMPLEXITY
And the talkslThe onesthat Arthur heardwerea disconcerting mix of
wild-eyed speculationandhard-nosed empiricism. It wasasif no oneknew
what the speakersweregoingto saybeforetheygot up to sayit. There were
a lot of peopletherein ponytailsand blue ieans.(Onewomangot up to
give her talk in bare feet.)The word "emergence"seemedto crop up
ftequently.And most of all, therewasthis incredibleenergyand cama-
raderiein the air-a senseof barrierscrumbling,a senseof new ideaslet
loose,a senseof spontaneous, unpredictable, open-ended freedom.In an
sort
odd,intellectual of way, the artificial
life workshop felt like a throwback,
like somethingright out of the Vietnam-era counterculture.
And, of course,in an odd, intellectualsortof way,it was.

Epiphany at MassachusettsCeneral

Chris l,angtoncan rememberthe precisemomentwhen artificial life

wasborn, if not the precisedate. It waslate 1971, early1972-winter,
anyway.And in classichackersgle, Langtonwasall aloneup on the sixth
foor of the Massachusetts GeneralHospitalin Boston,sittingat the big,
desklikeconsoleof the psychology departrnent'sPDP-9computerand de-
buggingcodeat threein the morning.
He likedworkingthat way."We didn't haveto be thereat anyparticular
time," I-angtonexplains."The guy who ran this place,FrankErvin, was
a very creattve,very hip kind of guy' He basicallyhireda wholebunch of
brightkidsto do the coding,and he gavethem a prettyfreehand. So the
straightpeople,who weredoing the real boringstuff, had the machine
during the day.And we got into the habit of comingin at four or five in
the afternoonand stayinguntil threeor four in the morning,when we
couldjustplay."
Indeed,so far as langton wasconcerned,programmingwasthe best
gameeverinvented.lt hadn't exactlybeena deliberatecareerchoice;he
had iust sort of driftedinto Ervin's groupabouttwo yearsbefore,shortly
afterhe arrivedat Massachusetts Generalasa collegedropoutfulfilling his
alternativeservice requirement asa conscientious obiectorto the Vietnam
War. Except for a few summer cou$es backin high school,in fact, his
programmingskillswere entirelyself-taught.But once he really started
rn.rting aroundwith computers,he'd startedhavingso much fun that he
stayedon, evenafterhis requirementwasfinished.
"It was great,"he says."I'm a mechanicat heart. I like to construct
things.I'd ilke to seethis thing actuallywork." And with the kind of stuff
Life at the Edgeof Chaos 201
he wasdoing on the PDP-9,he says,"You had to go knuckle-to-knuckle
with the hardware.Your programshad to takeinto accountwhatthe ma-
chinewasreallydoing,like'loadtheaccumulator from thisspecificaddress
and then put it back.'It waslogic, but it wasalsovery mechanical."
But he alsoliked the weird kind of abshactions he wasgettinginto. A
good examplewashis very first proiectthere, when he had gottenthe
experimentalpsychologists up and runningon the PDP-9' For yearsthey
hadbeenrecordingtheir dataon an ancientandsl-o-w PDP-8S,andthey
weregettingsickof it. But the problemwasthat, in the process, they had
createdall mannerof special-purpose softwarethat nobodywantedto re-
write-and that wouldn't run on the PDP-9.So langton'staskhad been
to write a programthat would hick the old sqftwareinto thinking it was
still running on the old machine.In effect,he wassupposed to re-create
the PDP-8Sasa "virtual machine"insidethe new one.
"l hadn'thadanyformalcourses in computationtheory,"saysLangton.
"So I got my first, visceralexposure to the conceptof a virtual machine
by having to createone. And I iust fell in love with the concept.The
notionthatyou couldtakea realmachineandabstract its lawsof operation
into a programmeantthat the programhad capfuredeverythingthat was
importantaboutthe machine.You'dleft the hardware behind."
Anyway, he says,on that particular night he was debugging code.And
sincehe knew he wouldn't actually be running anything for a while, he
pulled out one of the papertapesthat wasalwayssittingin a box in front
of the computer'sbig cathode-ray tube, and had run it throughthe tape
readerto set the computer going with the Gameof Life'
It wasone of hisfavorites."We'd gotten holdof thecodefromBill Gosper
and his group, who were hacking on the Gameof Life overat MIT," says
langton, "and we were playing around with it, too." The thing wasdown-
rightaddictive.Developed the previous by the Englishmathematician
year
fohn Conway,the Gameof Life wasn'tactuallya gamethat you played;it
wasmorelike a miniatureuniversethatevolvedasyouwatched.Youstarted
out with the computerscreenshowinga snapshot of this universe:a two-
dimensional grid full of blacksquares thatwere "alive" andwhitesquares
thatwere "dead. you
" The initial patterncouldbeanything liked' But once
yousetthegamegoing,thesquares wouldliveor diefromthenon according
to a few simplerules.Eachsquarein eachgeneration would first look
aroundat its immediateneighbors.If too manyof thoseneighborswere
alreadyalive, then in the next generationthe squarewould die of over-
crowding.And if too fewneighbors werealive,thenthe squarewoulddie
of loneliness. But if the numberof neighbors wasiust right, with either
 202 COMPLEXITY
hvoliving squares or threeliving squares, then in the nextgeneration that
centralsquaierould be alive-either by survivingif it werealready alive
or by being"born" if it weren't.
That wasall. The ruleswerenothingbut a kind of cartoonbiology.But
what madethe Gameof Life wonderfulwasthat when you turnedthese
simplerulesinto a program,theyreallydid seemto makethe screencome
alive. Comparedwith what you would seeon a present-day computer
screen,the actionwasratherslowand ierky,asif it werebeingplayedback
on a VCR in slowmotion. ln yourmind'seye,howevet,the screenalmost
boiled with activity,as if you werelookingthrougha microscopeat the
microbesin a drop of pond water.You could startup the gamewith a
randomscattering of live squares, andwatchtheminstantlyorganizethem-
selves into all manner of coherent structures.You could 6nd shuctures
thattumbled and structures that oscillatedlike beasts breathingin and out.
You could find "gliders," little clustersof live cellsthat movedacrossthe
screenat constantvelocity. You could 6nd "gliderguns" that firedoff new
glidersin a steadystream,and otherstructures that calmly atethe gliders.
if you ,"r. luckyyou might evenfind a "Cheshire Cat" that slowlyfaded
away,leavingnothingbehindbut a smileand a paw print. Every run was
different.and no one had everexhausted the possibilities."The first con-
figurationI eversawwasa large,stable, diamond-shaped structure," says
..Butthenyoucouldintroducea gliderfrom outsideandit would
Gngton.
inteirupt the perfectcrystallinebeau$. And the structurewould slowly
decayinto nothing,as if the gliderwasan infectionfrom outside.It was
like the Andromeda strain."
Sothat night, saysl,angton,the camputerwashumming,the computer
screenwasboiling with theselittle critters,and he.wasdebuggingcode.
"One time I glancedup," he says."There'sthe Gameof Life cranking
awayon the screen.Then I glancedbackdown at my computercode-
and at the sametime, the hairson the backof my neckstoodup. I sensed
"ro'
the presence of someone elsein the room."
Langtonlookedaround,surethat one of his fellow Programmers was
sneakingup on him. It wasa crowdedroom, crammedwith the big blue
cabinetsof the PDP-9,alongwith standingracksfor electronicequipment,
an old electroencephalograph machine,oscilloscopes, boxespushedinto
cornerstrailing tubes and wires, and a lot of stuff that was neverused
anymore. It was the classic hacker's paradise. But no-no onewasbehind
him; no one was hiding. He was definitely alone.
Langtonlookedbackat the computerscreen."I realizedthat it must
havebeenthe Gameof Life. Therewassomething aliveon that screen.
Life at the Edge of Chaos 203
And at thatmoment,in a wayI couldn'tput into wordsat the time, I lost
any distinctionbetweenthe hardwareand the process.I realizedthat at
somedeeplevel,there'sreallynotthatmuchdifferencebetween whatcould
happenin the computerand what could happenin my own personal
hardware-that it wasreallythe sameprocess that wasgoingon up on the
screen.
"I rememberlookingout the windowin the middle of the night, with
all this machineryhumming away.lt wasone of thoseclear,frostynights
whenthestarsweresortof sparkling.AcrosstheCharlesRiverin Cambridge
you could seethe ScienceMuseumand all the carsdrivingaround.I
thoughtaboutthe patternsof activity,all the thingsgoingon out there.
The city wassittingthere, just liting. And it seemedto be the samesort
of thingasthe Cameof Life. It wascertainlymuchmorecomplex.But it
wasnot necessarilydifferentin kind."

The Self-Assemblyof the Brain

Lookingbackon it with the perspective of twentyyears,saysLangton,

that night of epiphanychangedhis life. But at the time it waslittle more
than an intuition, a certainfeelinghe had. "It wasone of thosethings
whereyouhavethisfash of insight,andthenit'sgone.Likea thunderstorm,
or a tornado,or a tidalwavethatcomesthroughandchanges thelandscape,
andthen it's past.The actualmentalimageitselfwasno Iongerreallythere,
but it hadsetme up to feelcertainwaysaboutcertainthings.Thingswould
comealongthat justsmelledright,thatwouldremindme of this pattern
of activity.And for the restof my careerI've tried to follow that scent.Of
course,"he adds,"that scenthasoftenled me somewhere and then just
left me, not knowingwhereto go next."
That'sactuallyan understatement. Not only wasthe Chris Langtonof
l97l almostcluelessasto whatthis feelingmeant,he wasa longwayfrom
beinga systematic scholar.His ideaof followingthe scentwasto wander
aroundthe libraryor throughbookstores, pickingup articleshereandthere
that somehowrelatedto virtualmachines, or to emergent, collectivepat-
terns,or to local rulesmakingglobaldynamics.And everyso often he
would takea randomcourseat Harvard,BostonUniversi$,or wherever.
But basicallyhe wascontentto takethingsasthey came.Therewasiust
too much otherstuffgoingon with his life. His realpassionwashis guitar;
he anda friendweretrying(unsuccessfully) to starta profesionalbluegrass
band. He wasstill puttinga lot of energyinto draft resistance and protests
2M COMPLEXITY
againstthe Vietnam War. And so far as he was concerned,the whole
counterculture scenearound the periphery of the universitiesmade Cam-
bridge and Boston a greatplace to be. The fact wasthat Chris langton was
happierthan he had been in a long while.
"High school was a disasterfor me," he says.In 1962, when he was
fourteen, Langton had gone from a very small elementaryschool in his
hometown of Lincoln, Massachusetts, to Lincoln-Sudbury High, a big
regionalschool in nearby Sudbury. "It was like going to iail, every day,"
he says. "This was an industrial-strength high school, where kids were
treated like iuvenile delinquents unless they somehow proved themselves
otherwise and escapedinto the special classes. And I was iust not of the
right mental demeanor to play along with that whole system. I had long
hair. I playedguitar and was into folk music. I was a hippie without there
being any other hippiesaround. So I was very much a loner."
It probably didn't help that his parents,mysterywriter fane Langton and
physicistWilliam Langton, had been "radicals" from the earliestdaysof
the civil rightsmovementand the Vietnam War. "During high school, my
parents wbuld occasionally take me out and we'd go into the city and
participate in sit-ins and teach-insfor equality. We went to a lot of inner-
city schools. We also took busesto Washington, and we'd protestthis and
that, and I got arrestedfor whateverexcusethere was to arrestprotesters."
Finally, Langton graduatedin 1966. "This was the beginning of the
hippie era," he says."So a friend of mine and I hoppedon a bus and went
to California that summer, where things were a lot more advancedalong
that particular axis. We went straightto Haight-Ashbury. Listened to fanis
foplin, the feffersonAirplane. It wasa greatsummer."
In the fall, unfortunately, he had to report back for duty at Rockford
College in lllinois. Personally,he didn't give a damn about college.And
the feeling seemedto be mutual: with his high-school gradeaveragehov-
ering around a C, the Harvardsand MITs of the world had given Langton's
applicationsa decisivethumbs-down. But his parentshad insistedthat he
go to college somewhere.And Rockford, having iust converted itself from
a finishing school for girls ("The Vassarof the Midwest") into a general
liberal arts college, was eagerlyrecruiting.
To Langton, RocKord's brand-new campus out in the cornfields looked
like a minimum-securityfarm prison. "lt might as well have had barbed
wire and razor wire along the top," he says.Becausethe schoolhad done
so much recruiting, however,it had managedto draw about ten EastCoast
hippie gpes that year out of a total studentbody of 500. "We got there and
looked around, and there were these incredibly redneck, extreme right-
Life at the Edge of Chaos 205
wing-well, thisbasicareawasthehomeof theMinutemen,"he says."At
leaston the EastCoast,thingswerestartingto happen.But out in the
cornfieldsof Illinois, it wasstill somewhere backin the McCarthyera.And
a hippiein the middleof lllinois in 1966wasbasically deadmeat.I got
signedup for women'sgym by the registrarafterthey sawme. One time
severalofus guyswalkedinto a doughnutshop,anda coupleofstatecops
walkedin behindus, andoneof themsaid,'I don'tknowwhichoneit is,
but one of you guyshasa pretg ugly girlfriend.'We got thrown out of
everyrestaurant; nobodywould serveus becausewe had long hair. The
administrationimmediatelystartedsuspecting us of drugsand all kindsof
otherstuff."
The only thing to do, obviously,wasto headnorth. Langtonand his
fellow"undesirables" startedhitchhikingup to the Universityof Wisconsin
in Madison,often stayingweeksat a time. "This waswhereI belonged,"
he says."The whole socioculturalupheavalthat wasthe 1960swashap-
peningin Madison,and therewaszip goingon in Rockford.Therewasa
lot of antiwaractivity goingon in Madison.There werelots of hippies
startingto experiment with drugs,soI did. I hadan electricguitar,anda
friendof minehadbeenexposed to Appalachian bluegrass, sowedid some
incrediblejamming.There werelots of thingsgoingon-but nothing
havingto do with what you weresupposed to be in collegefor."
By the startof his sophomore yearat Rockford,not surprisingly, Langton
wason academicprobation.At the end of that fall semester, the school
administrationtold him to leaveand he told them he wasquitting.
"I wantedto stayup in Madison,"he says."But I didn'thavea iob, and
I didn't haveany realwayto supportmyself.So I endedup goingbackto
Boston,where I got a lot more political and a lot more involvedwith
antiwaractivity."With no morestudentdeferment,he filed for consci-
entiousobjectorstatus.And after a long fight, he got his draft boardto
acceptit. "Then I did alternative serviceat Massachusetts GeneralHospital
startingin 1968."

Oncethere,of course,Langtonwasconvinced he'dfoundhis nicheat

last.He wouldhavehappilystayed in hisprogramming "[t
iob indefinitely.
wasa greatjob. I waslearninga lot, I washavinga greattime with the
peoplethere.And therewasno reasonto leavethat."But in 1972he had
no choice:the groupleader,FrankErvin, accepted a positionat the Uni-
versityof Californiain Los Angelesand essentiallypackedup the lab to
takeit with him. Left at looseends,Langtonhookedup with anothergroup
 206 COMPLEXITY
of psychologists, who werestudyingsocialinteractions amongshort-tailed
macaquemonkeysfrom SoutheastAsia. And by Thanksgiving1972he
found himselfout in the iunglesaboutforg milesfrom Sanfuan, Puerto
Rico, at the CaribbeanPrimateResearch Center.
This, asit turnedout, wasnota greatiob. Langtondid like the monkeysr
he spenteightto ten hoursa daymonitoringthemduringthe experiments,
becomingfascinated by their cultureand how theypassed that cultureon
to each new generation.The problem, unfortunately, was that the humans
on the primatecenterstaffwere entirelytoo similar to their subiecb. "One
of the experiments therewas to understand how the social system responded
to stress,"saysLangton."So they would slightlydrug a monkeyat some
positionin the hierarchy,andthen seehow the hierarchyresponded when
that monkeydidn't do what it was supposed to do. The top male, for
example,wassupposed to threatenall the others,matewith all the females,
settlethe arguments, and chasecertainonesaround.So when he wasa
little out of it, the colonyresponded by breakinginto factions.The sub-
leadersmight be veryattentiveto the chiefmonkey,but occasionally attack
him, but then backoff reallyquickly.You couldseethat theyweretrying
to supporthim in his role, but alsohavingto takeon leadershiprespon-
sibility.But he wasstill there,so therewasthis funny tension.
"Well, the headof the research centerat the time wasa complete,total
t alcoholic.He'dstartoff themorningwith abouta gallonof BloodyMarys,
and then he'd be out of it for the restof the day.He couldn'tfunction in
t
I
his role. So the restof the staffweren'treallyempowered to do things,but
'You
had to do things.And therewereall thesefightsabout, shouldhave
!
consultedwith me!'I couldhavetakenthedatasheetsI wasusingto observe
the monkeys,lifted the roof off the research centet,and seenexactlythe
samething. It brokeup into factions,therewasa kind of revolution,and
I waspart of one factionthat ultimatelylost. I wasaskedto leave,and I
wasreadyto leave."
At looseendsagainaftera yearin PuertoRico, Langtonrealizedthat it
wastime to siartthinkinga little moreseriously aboutlife. "I couldn'tiust
keepiumpingaroundand living for eachday, without anykind of long-
termideaof whereI was going," he says. But where was that? He wondered
if that mysterious scentmight be hying to tell him something. He'd been
followingit the wholetime he was in Puerto.Rico, and he was beginning
to think that maybe,iust maybe he'd found the trail: cosmology and astro-
physics.
"l didn't haveany access to computers down there,so I wasn'tdoing
Life at the Edge of Chaos 207

any computer work to speakof. But I really did tons of reading," he says.
The origin of the universe, the structure of the universe, the nature of
time-it all seemedto have the right smell. "So when the situation dete-
riorated, I went back to Boston and startedtaking coursesin mathematics
and astronomyat Boston University."
He had taken a lot of the mathematicsbefore,of course.But Langton
thought it might be a goodmove to startall overfrom scratch."l just wasn't
paying attention before. I wasn't in school becauseI wanted to be. I went
becausethat's what you did. You just got squeezedout of the tube of high
school, onto the toothbrush of college." He could only afford to take a few
coursesat a time on an outstudent basis,while he worked at various odd
jobs. Yet he threw himself into those courseswholeheartedly,and started
doing remarkably well. Finally, one of his teachers,who had become a
good friend, said, "Look, if you really want to do astronomy, go to the
Universityof Arizona." Boston Universitywas 6ne for a lot of things, he
said. But Arizona was one of the astronomicalcapitalsof the world. The
campus in Tucson was right in the middle of the SonoraDesert, where
you could find some of the clearest,driest, darkestskieson the planet. The
mountaintopsin the areasproutedtelescopedomeslike mushrooms.Kitt
PeakNational Observatorywas only forty miles away,and ib headquarters
was right there on campus. Arizona was the place to be.
That made senseto Langton. He applied to the University of Arizona,
which acceptedhim for the fall of 1975.

When he was in the Caribbean, saysLangton, he had learned to scuba

dive. And there, among the coralsand fishes,he had come to love moving
in that third dimension.It wasintoxicating.But oncehe wasbackin Boston,
he'd soon discoveredthat scuba diving in the cold, brown watersof New
England just wasn'tthe same. So as a substitutehe'd tried hang-gliding.
And he'd becomehookedthe 6rstday.Sailingoverthe world, riding upward
from thermal to thermal-this was the ultimate in three dimensions. He
becamea fanatic, buying his own hang-gliderand spendingevery spare
minute aloft.
All of which explainswhy, at the beginning of the summer of 1975,
langton set out for Tucson along with a couple of hang-glidingbuddies
who were moving to San Diego and who had a truck. Their plan was to
spend the next few months making their way acrossthe country at the
slowestspeedpossible,while they went hang-glidingoffany hill that looked
208 COMPLEXITY
halfwayinviting. And that'sexactlywhattheystartedto do, workingtheir
way down the Appalachians until they cameto GrandfatherMountain,
North Carolina.
As the highestpeakin the BIue Ridge,CrandfatherMountain boasted
a spectacular view;in fact, it wasa privatelyownedtouristattraction.And
it turnedout to be just asspectacular a placeto fly: "When the wind was
right you couldstayup for hours!"saysLangton.Indeed,whenthe owner
of the mountainrealizedhowmanyhot dogsandhamburgers andsouvenirs
hewassellingwhilethetouristsstoodaroundto watchtheselunaticsdefying
gravity,he offeredthem $25 a dayif theywouldstayall summer.
"Well, it wasquite unlikelywe'dFnd a betterplace,"saysLangton.So
they agreed.And asa touristattraction,they werea tremendoussuccess.
Moreover,the ownergot sointerested in hang-glidingthat he arrangedfor
a nationalchampionshipto be held at GrandfatherMountain at the end
of the summer.Langton,figuringhe wouldhavea homecourt advantage
if he competed,spentthe restof the summerpracticing.
The accidentwason August5, he says.His friendswith the huck had
alreadyleft. And he wasplanningto headout the nextdayhimseli figuring
that he would go to Tucson,register,then come back to Grandfather
Mountainforthechampionship beforeclassesstarted. Butin themeantime,
he'dwantedto get in a few morepracticerunsfor the spotJandingevent,
whereyou had to hit a bull's-eyeon the ground.
Sotherehe was,saysLangton,comingin for his lastaftemptof the day,
This particularspotlandingwasa hicky maneuverat best,sincethe target
wasin a smallclearingin the hees;the only wayto do it wasto comein
high and then spiraldown almostat stallspeed.But on that daythe wind
wassofunky anduncooperative thatit seemediustaboutimpossible. lang-
ton had had to abort his landingfour timesalready,and he was really
gettingfrustrated.This would be his last chancebeforethe competition
itself.
"l remember 'Damn, I'm too closeandI'm too high. I'll
thinking, iust
try for it anyway. What the hell.'And thenasI settleddownbeneaththe
levelof the trees,at aboutfifty feet, I sankinto deadair. I wastoo slow,
andI stalledoutat iustthewrongaltitude.I remember thinking,'Oh, shit.'
I realizedI wasgoingto crash.It wasgoingto be a badcrash.I remember
'God, I'm goingto breaka legnow.Shit.' " In a desperate attempt
thinking,
to pick up speedand regaincontrol,he put the gliderinto a dive. No go.
Then, as he'd beentaughtto do, he held out his legsto absorbsomeof
the shock."You knowyou'regoingto breakyour legs,but you don't pull
themup," he says."Because if youhit on yourbutt you'llbreakyourback.
Life at the Edgeof Chaos 209
"I don'tremember hitting.My memoryisspoftyafterthat.I do remember
lying there,realizingthat I wasbadlyhurt and knowingthat I shouldlie
still. My friendsran over.Peopleheardaboutit at the top of the mountain
andcamedown.The ownerof themountainwastakingpictures.Somebody
wastherewith a two-wayradioand calledfor an ambulance.A long time
laterI rememberthe medicsshowingup andsaying,'Wheredoesit hurt?'
and me saying,'All over.'I rememberthem mumblingto themselves,
rollingme ontothe stretcher."
The ambulancetook langton down the mountainto the nearestemer-
gencyroom, at Cannon Memorial Hospitalin tiny BannerElk, North
Carolina.Much later,when he waslying semiconscious in the intensive-
careunit, he remembers the nursestellinghim, "Oh, you brokeyour legs.
You'll be here a coupleof weela.Then we'll haveyou out of here and
you'll be runningaroundin no time."
"I wason morphine,"he says,"so I believed them."
In fact, langton wasa mess.His crashhelmethad savedhis skull, and
his legshad cushionedthe impactenoughto savehis backandpelvis,But
he had shatteredthirty-fivebones.The impacthad brokenboth legsand
both arms,almostrippinghis right arm out of its socket.It had fractured
mostof his ribs and had collapsedone lung. And it had drivenhis knees
into his face, smashingone knee, his iaw, and almosteverythingelse.
"Basically,"saysl,angton,"my facewaspaste."His eyeswouldn't track:
his cheekbones and the floorsof his eyesocketswerebroken,and there
wasnothingsolidfor the eyemuscles to pullagainst.And hisbrainwasn't
workingright: the crushingof his face had causedtraumadeepinside.
"Theyseta lot of bonesand reinflated my lung in the emergency room,"
saysLangton."But I didn't comeout of the anesthesia for a day longer
than I shouldhave.They wereworriedabouta coma."
He did wakeup, eventually. But it wasa longtimebeforehewascoherent.
"I had this weird experience of watchingmy mind comeback," he says.
"I could seemyselfas this passiveobserverbacktheresomewhere. And
therewereall thesethingshappeningin my mind that weredisconnected
from my consciousness. It wasvery reminiscent of virtualmachines,or
like watchingthe Game of Life. I could seethesedisconnected patterns
self-organize, cometogetheE andmergewith mein someway,I don'tknow
how to describeit in any obiectivelyverifiableway,and maybeit wasiust
a figmentof all thesefunny drugstheyweregivingme, but it wasasif you
took an ant colonyand tore it up, and then watchedthe antscomeback
together,reorganize,and rebuildthe colony.
"So my mind was rebuildingitself in this absolutelyremarkableway.
210 COMPLEXITY
And yet, still, therewerea numberof pointsalongthe waywhen I could
tell I wasn'twhat I usedto be, mentally.There were thingsmissing-
thoughI couldn'tsaywhatwasmissing.It waslike a computerbootingup:
I couldfeel differcntlevelsof my operatingsystembuilding up, eachone
with morecapabilitythan the last.I'd wakeup one morning,and like an
elechicshockalmost,I'd sort of shakemy headand suddenlyI'd be on
'Boy,I'm back!'ThenI'd realizeI wasn't
somehigherplateau.I'd think,
really quite back. And then at somerandompoint in the future, I'd go
throughanotheroneof those,and-am I backyetor not?I still don'tknow
until this day.A coupleof yearsagoI wentthroughanotherone of those
episodes,a fairly maior one. So who knows?When you'reat one level,
you don'tknowwhat'sat a higherlevel."
Langtonwasoneof the worstaccidentcases they'deverseenin Banner
EIk, wherethe hospitalwasmuch more usedto gunshotwoundsand ski
accidents.Moreover,he was in traction from head to foot, and in no
conditionto be moved.l,angtondid, however,haveone incrediblepiece
of luck. Dr. lawson Thte,the directorof CannonMemorialand the son
of the founder,hadpracticedin a numberof maiormedicalschoolsbefore
comingbackto BannerElk, and wasan orthopedicsurgeonof national
caliber.Over the nextseveralmonthshe reconstructed langton'scrushed
cheekbones andput in sheetsofreinforcing plasticto theeyesockets.
rebuild
He pulledthe sinuscavitiesbackopenand rebuiltthe facialbones.He
reconstructed the shatteredknee from piecesof Lan$on's hip. And he
rebuiltthe dislocated rightshouldersothatthe nervescouldgrowbackinto
theparalyzed arm.By Christmastime 1975,whenlangton wasfinallyflown
to EmersonHospitalin Concord,Massachusetts, nearhis parentsin Lin-
coln, Thtehad performedfourteenoperations on him. "The doctorsthere
wereamazedthat one guy had done all theseseparate operations,"says
Langton.
In Concord,Langtonwasfinallywell enoughto beginthe long, slow
process of learningto usehis bodyagain."l'd beenflaton my backfor six
months,"he says,"a lot of thattime in tractionwith my iaw wiredshut.
I wentfrom 180to ll0 pounds.And I gotno physicaltherapythatentire
time. So a lot of thingshappenin that situation.Youloseall your muscle;
it iust disappears.All of your ligamentsand tendonstightenup. And you
becomevery stiff, becauseif your iointsaren'tconstantlybeingflexedto
keepcleara certainrangeof motion,theyfill up with thisstuffthafssecreted
to replaceworn-outcartilage, until there'sno roomfor yourioint to move
at all.
"So I wasthis skeletal-looking anolexic,"saysLangton."Of course,I'd
Life at the Edge of Chaos 211
had my jawswiredshut,so I'd losta lot of the musculature that controls
the iaws.It took me a long time to regainthe abilityto reopenmy mouth
morethan justan inch or so. Eatingwasdifficulgchewingwasdilficult.
Thlking-J talkedalmostthroughclenchedteeth.And my facehung in
funny ways.My cheekbones werewaybackinsteadof beingpushedout.
So I had this ghoulishexpression. My eye socketswere very different
shapes-theystill are."
The physicaltherapistsat EmersonHospitalgotlangon up andwalking.
And theytriedto gethis rightarmbackto work."A lot of thewayI regained
controlwasby playingguitarlying fat on my back,"he says."l forced
myself.I didn't carewhatelsehappened, but I wasn'tgoingto not play
guitaranymore."
In themeantime,Langtonwasreadingeverything aboutsciencehecould
get his handson. He'd startedin BannerElk, assoonas his eyesockets
werebackin placeandhe no longerhaddoublevision."l hadpeoplemail
me books,"he says."I had bookscomingin by the truckload,and I was
devouringthem. Someof it wason cosmology. I readmathbooksand did
problems.But I alsospenta lot moretime on the historyof ideasand
biologyin general.I readLewisThomas'sThe Livesof a CeIL I reada lot
aboutthephilosophyof scienceandthephilosophy of evolution,"He wasn't
up to a truly concentrated effort,he says.The hospitalin BannerElk had
put him on antidepressants and enoughof the pain-killerDemerolto get
him thoroughlyaddicted.Moreover,his mind wasstill in the midstof this
funny proces of reorganization. "But I wasa sponge.I did a lot of non-
specific,nondirected genericthinkingaboutbiology,physical science,and
ideasof the universe, andabouthowthoseideaschanged with time.Then
therewasthisscentI talkabout.Throughall of this,I wasalways following
it, but withoutanydirection.Cosmology andastronomy smelledgood.But
basicallyI still didn't understand it. I wasstill lookingbecause I didn't
knowwhatwasout there."

Artificial Life

When Langtonfinallymadeit to the Universigof Arizonacampusin

Tucsonin the fall of 1976,he wasableto hobblearoundwith the aid of
a cane,althoughtherewerestill moreoperations to comeon his kneeand
rightshoulder.But he wasalsoa twenty-eight-year-old
sophomore, crippled
and cadaverous-looking.He felt grotesque,
like somethingout of a circus
sideshow.
212 COMPLEXITY

"lt wasbizarre,"he says,"becausethe Universi$of Arizona is a real

Ken and Barbiekind of place,with fratsandsororities andlotsof beautiful
people.AIso,my mentalstatewassuchthat I'd find myselframblinga lot.
I'd get off on tangentsof whateverthe conversation was,and suddenly
.realizethat I didn't have a clue of where this conversation hadstartedfrom.
My attentionspanwas fairlynarrow. So I felt mentally a freak, andphysically
a freak."
On the otherhand,oneof the reallygoodthingsaboutArizonawasthe
universityhospitaland itr first-class Programin physicaltherapyand sports
medicine."I reallybenefited a lot ftom thatprogram,"says[an$on. "They
insistedthat you keep plugging away, that you makeProgress. And I saw
therethat you had to passa threshold; you had to go through a transition
in your mind of acceptingthe way you were and work from there: not
feelingbad aboutit, but feelinggoodaboutProgress' lust So I resolved to
live with the ostracismand weirdness that I felt. And I still would answer
questions in class-eventhoughthingswouldsometimes bea little strange,
becauseI'd getoff the topic. I iust had to keeppluggingaway."
Unfortunately,however,evenashis mind andbodywereslowlycontin-
uingto heal,l,angtonwasdiscovering onethingthatArizonawasnot good
for: astronomy.He had neverthoughtto checkwhetherthe Astronomy
Capital of the World offeredan undergraduate maior in the subiect'It
didn't. The university did havean astronomy Ph.D. program,which was
superb.But to getthere,undergraduates weresupposed to maiorin physics
and then switch.The only problemwasthat, so far asLangtonwascon-
cerned,the Arizonaphysicsdepartmentwasabysmal."It wascompletely
disorganized," he says."None of the peopleteachingthe classes spoke
English.The lab manualswereancient.The equipmentdidn't match'
Nobodyknewwhatwe weresupposed to be learning."
This wasnot the kind of sciencethat Langtonhadsignedup for. Within
a semester he wasout of physicsand out of astronomy. After all that, the
elusive scent had led him straight into a dead end' (langton wasn'tthe
only person to feel this way about the physics department; in 1986,the
universitybrought a in new chairman frOm l-os Alamos to revitalize it:
PeterCarruthers.)
The goodnewswasthathe hadno regrets. Arizonadid havean excellent
department of philosophy, a subiect that appealed to Langtonbecauseof
his fascination with the history of ideas. And it had an equallyfine de-
partmentof anthropology, which appealed to him because of the affection
he'dfelt for the Puerto Rican monkeys. That first semester he tookcourses
Life at the Edge of Chaos 213

in both subjects to fill out his comprehensiveprogramrequirements' And

by the time he left physicshe waspursuing what eventually turned into a
philosophy-anthropology doublemaior.
It was an odd combination,to saythe least' But to langton they fit
togetherperfectly.He'd sensedit the dayhe walkedinto wesleysalmon's
ph"ilosophy of scienceclass."salmon had this very nice perspective," says
Lngto., who quickly askedhim to be his adviserin the philosophy de-
..H"'i b...r a studentof HansReichenbach, a philosopher of
prrti,..,i.
science fromtheviennacircle. Those guys had been doing very technical
stuff-the philosophy of space andtime,andquantummechanics, andthe
curvatureof rp"."-ti*" by gravity.And I very rapidly realized that l_was
much more interested,not in our specific,currentunderstanding the of
universe,but in how our world view had changedthroughtime. what I
*", ,."liy interestedin wasthe historyof idea!.And cosmology..iust hap-
penedto be one of the mostacc6i6iE arenasfor studyingthat'"
In theanthropology department, meanwhile,langton washearingabout
the rich varietyof hu*"n mores,beliefs,andcustoms; aboutthe riseand
fall of civilizations;abouttheoriginsof humankindoverthreemillionyears
of hominid evolution.Indeed,his adviserthere,physicalanthropologist
StephenZeguta,wasboth a superbteacherand a man with a cleargrasp
of evolutionarytheorY.
So on everyside,saysLangton,"I wasiust immersedin this ideaof the
evolutionof information.That quickly becamemy chief interest.It iust
smelledright." Indeed,the scentwasoverpowering' Somehow,he says,he
knew he was getting verY close.

One of Langton'sfavoritecartoonsis a panelof Gary Larson'sThe Fdr

side, which showsa fully equippedmountaineeraboutto descendinto an
immensehole in the ground.As a reporterholdsup a microphone,he
proclaims,"Because it is not therel"
"That'show I felt," laughslangton. The morehe studiedanthropology,
he says,the morehe sensed thatthe subiecthada gapinghole' "lt wasa
fundamental dichotomy. the onehand,herewasthisnice,clearfossil
on
recordof biologicalevolution,togetherwith a nice bodyof Darwinran
theorythatexplained it. Thattheoryinvolvedtheencoding of information,
and the mechanisms by which that informationwas passed down from
generation to generation. On the other hand,herewas this nice, clearfossil
iecord of cultural evolution, as discoveredby the archeologists. And yet
 214 COMPLEXITY

I peoplein cultural anthropologywouldn't think about, or talk about, or

/ euenlistento you talk abouta theoryto explainthat record.They seemed
I to be avoidingit."
Langton'simpression wasthatevolutionary theoriesof culturestill carried
a stigmafrom the time of spcial Daryinlgp in the nineteenthcentury,
when peoplewere defending both war and grosssocialinequityon the
groundsof "the survivalof the But while he couldcertainlysee
fittest."
the problem-after all, he'd beenprotestingwar and socialinequitymost
of his life-he just couldn'tacceptthe gapinghole. If you could createa
realtheoryof culturalevolution,asopposed to somepseudoscientific ius-
tificationfor the statusquo, he reasoned, then you might be able to un-
derstand how culturesreallyworked-andamongotherthings,actuallydo
something aboutwarand socialinequity.
Now herewasa goalworthpursuing.And mostof all, herewassomething
that smelledright.lt wasn'tiust cultural evolution,Langtonrealized.It
wasbiologicalevolution,intellectualevolution,culturalevolution,con-
ceptscombiningand recombiningand leapingfrom mind to mind over
milesandgenerations-allwrappedtogether.Somehow, at theverydeepest
level,theywereall iustdifferentaspects of the samething. More thanthat,
they wereall iust like the Came of Life-or for that matterlike his own
mind, still reassembling itselffrom fragmentsscattered by the fall. There
wasa unity here,a commonstorythat involvedelementscomingtogether,
structures evolving,andcomplicated systems acquiring thecapacity to grow
and be alive.And if he could only learnto look at that uni$ in the right
way,if he could only abstractits lawsof operationinto the right kind of
computerprogram,then he would havecapturedeverythingthat wasim-
portantaboutevolution.
"This waswherethingsfinally startedto cometogetherfor me," says
[,an$on.As a visionit wasstill almostimpossible to articulateto anyone
else. "But nothing elsedrove me. This is what I thoughtaboutall the
time."
ln the springof 1978,Langtonlaid out histhinkingin a twenty-six-page
paperentitled "The-Evolution of Belief." His basicargumentwas that
Liologi"al .nd c.iiiut?G'irr-ii6il--w-ei6-simplytwo aspectsof the same
phenomenon,andthat the "qenes"of culturewerefu[iefs-which in turn
wererecordedin the basic"DNA" of culmffitGie' In retrospectit
wasa pretg naive attempt, he But
says. it was his manifesto-and not
incidentally,his proposalfor an interdisciplinary, self-designed Ph.D' pro-
gram that would allow him to do researchon this stuff. Moreover,the
paperwasenoughto convincehis anthropology adviser,Zegura'"He was
 Life at the Edgeof Chaos 21s
a reallygoodguy, a goodteacher,a believer,"saysLangton."He wasthe
only one who reallygrokkedwhat I wastalkingabout.His attitudewas,
'Go for "
it!' But Zeguraalsowarnedhim thatfor a specialPh.D. program,
langton would haveto haveadvisersin other departments as well; as a
physicalanthropologist he simplywasnot competentto giveLangtonthe
guidancehe would needin physics,biology,and computerscience.
So Langtonspenthis senioryearat Arizonamakingthe rounds."That's
when I startedcallingit 'artificiallife,' by analogywith artificialintelli-
gence,"he says."I wanteda nice,shorthandlefor it thatwouldat least
put peoplein the ballpark,Most peopleknewwhat artificialintelligence
was,moreor less.Well, artificiallife meanttryingto captureevolutionin i
the samekind of waythat artificialintelligencewastrying to captureneu- [
ropsychology. I wasn'tgoingto mimic exactlythe evolutionof the reptiles./
I wasgoingto capturean abshactmodelof evolutionin the computerand
experimenton that. So that phraseat leastopenedthe door."
Unfortunately, thedoorusuallyslammedshutagainassoonashe opened\
his mouth."l talkedto guysin computerscience, andtheydidn't havea \
clue,"hesays."Theywereintocompilers anddatastrucfures andcomputerI
languages. They didn't evendo artificialintelligence,sotherewasnobodyI
therewho could evencomecloseto listeningto me. They noddedtheir J
headsand said,'This hasnothingto do with computers.'" I
Langtongot exactlythe sameresponse from the biologistsand the phys-
icists."I keptgettingthis look you getwhentheythink you'rea crackpot,"l
he says."It wasvery discouraging-especially comingas it did after the I
accident,when I felt unsureof what I wasor who I was."ObiectivelX{
Langtonhadmadeenormous progressby thispoint;he couldconcentrate,
he wasstrong,andhe couldrun fivemilesat a stretch.But to himself,he
stillfeltbizarre,grotesque, andmentallyimpaired."I couldn'tfell.Because
of this neurologicalscrambling,I couldn'tbe sureof anyof my thoughts
anymore.So I couldn'tbe sureof this one. And it wasn'thelpingthat
nobodyunderstood whatI wastrying to say."
And yet he keptpluggingaway."I felt like it wasthe thing to do," he
says. "I waswillingto keeppushingbecause I knewthisstuffhadconnections
to whatI'd thoughtaboutwhenI wassaneandrational,beforetheaccident.
And I keptseeingthingsout therethatrelatedto it. I didn'tknowanything
aboutnonlineardynamicsat the time, but therewereall theseintuitions
for emergentproperties, the interactionof lotsof parts,the kindsof things
thatthe groupcoulddo collectively thatthe individualcouldn't."
Intuitions,unfortunately, weren'tgoingtocut it. By theendof his senior
year,for all his plugging away,l,angtonhadto admithe wasstuck.Zegura
216 COMPLEXITY

wasbehindhim. But zeguracouldn'tdo it all by himself.It wastime to

fall backand regroup.

In the middle of all this, on December22, 1979,l.angtonhad goften

marriedto Elvira segura,a feisty,blunt-talkingmaster'sin library science
whom he had met in one of SteveZegura'santhropologyclasses'"We
startedoff asgoodfriends,and it wentfrom there,"he says.And when he
graduatedwith his doublemaior in May 1980-largely becausehe'd ac-
cumulatedsomanycreditsthatthe universityinsistedon it-he andElvira
Segura-langtonmovedinto a little two-bedroomrentalhouseiust north
"t;iTiT;r. in a stablepositionfor the time being.His wife had a good
positionatthe universitylibrary,andLangtonhimselfwasworkingathourly
iobs,both asa carpenterwith a homeremodelingcompany-he thought
of the exercise asgoodtherapy-and asan assistant at a stained-glass shop.
Indeed,there was a part ofhim that could havehappily gone on doing that
forever. "Good glasstakes on a life of its own," he says. "You have lob of
little pieces,but you're putting them all together to form a nice global
effect."But l,angtonalsoknewthathe hadsomeseriousdecisions to make,
andsoonerratherthan later.With Zegura'sencouragement, he had already
beenacceptedinto the anthropology departmenfs graduate program' But
without an agreement to do an interdisciplinary Ph.D. on artificial life,
that meantwastinga lot of time on courseshe didn't want or need' So,
shouldhe iust bagartificiallife entirely,or what?
Not a chance."By now I'd had the epiphanyand I was a religious
convert,"he says."This wasclearlymy life from now on. I AnewI wanted
to go on anddo a Ph.D. in thisgeneralarea.It's iustthatthe pathto take
wasn'tobvious."
The thing to do, he decided,wasto geta computerand work someof
theseideasout explicitly.That way,he could talk aboutartificiallife and
at leasthavesomethingto showpeople.Sowith a loanfrom the proprietor
of the stained-glass shop,he boughtan Apple II homecomputerand set
it up in the secondbedroom.He alsoboughta little colortelevisionsetto
useasa monitor.
"I gpically workedon it at night because I had to be on the iob during
the day," saysLangton."l'm almostalwaysawakefor two or threehours
everynight.For somereason,my mind is at its mostactive,mostaware,
in iis bestmodefor free,creativethinkingthen. I'll wakeup with an
".d
ideaon the tip of my mentaltongue,and I'll iust get up and pursueit' "
 Life at the Edgeof Chaos 217
His wife wasn'texactlyhappyaboutthis. "Will you comebackto bed
and go to sleep!"he'd hearfrom the otherbedroom."You'll be exhausted
tomorrow."Today,with 20/20hindsight,ElviraSegura-langton looksback
on Chris Langton'snighttimehackingasbeingwell worth the effort. But
at the time shefound it intenselyannoyingto haveher husbandtreatthe
placelike an office.To her the housewashome,a placefor family and an
escapefrom the outsideworld, And yet-she could alsoseethat this was
very clearlywhatChris Langtonneededto do.
Iangton'sfirstaftemptat artificiallife wasexceedinglysimple:'brganisms',
that consistedof little more than a table of genes."Each entry in the
hble wasthe genotypeof the organism,"he says."lt would havethingslike,
How long is this organismsupposed to live?How manyyearsbetweenpro-
ducingoftpring? What color is it? Whereis it in space? And then there'd
be someenvironment,like birdsgoingthroughandpickingthingsoff that
stoodout too much from the background. Sothe creatureswereevolving,
because when they producedoffspringthere'dbe a chanceof mutation."
Once he got this programup and running, langton wasquite happy
with it-at first. The organisms did indeedevolve;you couldwatchthem
do it. And yet he quicklybecamedisillusioned. "It wasall prettydamn
linear,"he says.The organisms weredoingobviousthings.They didn't I
takehim beyondwhat he alreadyunderstood. "There wereno real orga- |
nisms,"he says."WhatI hadwasthistableof genesbeingmanipulated by I
someexternalgod-the program.Reproduction iust happenedmagically. I
what I wantedwasa liftle moreclosure-so thattheprocess of reproduction t/
would arisespontaneously and be part of the genotypeibelf." l|J
Not knowingevenhow to begin,langton decidedit wastime to go to
the Universityof Arizonalibrary,wherehe could do a computerized lit-
eraturesearch.He hied the keywords"self-reproduction."

"Wow!" saysLangton,"This wasright.When I found all that, I said,

'Hey,
I maybe crazy,but thesepeopleareat leastas rrazyasI am!' " He
checkedout the booksby von Neumann,Burks,and Codd, alongwith
everythingelseon the list that he could find in the universitylibrary,and
devouredthem. YeslIt wasall there:evolution,the Gameof Life. self-
assembly,emergentreproduction,everything.
218 COMPLEXITY

Von Neumann,he discovered, hadgotteninterested in the issueof self-

reproduction back in the late 1940s, in the aftermath of his workwith Burks
andGoldstine on the design of a programmable digital computer. At a time
when the very concept of a programmable computer was still fresh and
new, and when mathematicians and logicians were eager to understand
what programmable machinescould and couldn't do, the questionwas
al-osl inevitable: Could a machinebe programmed to makea copyof
itself?
Von Neumanndidn't haveany doubtthat the answerwasyes'at least
in principle.Afterall, plantsandanimalshavebeenreproducing themselves
for severalbillion years, and at the biochemical level they are iust "ma-
chines"followingthe samenatural laws as the stars and planets. But that
factdidn't help him verymuch. Biological self-reproduction is immensely
complicated, involvinggenetics,sex,the union of spermand egg,cell
divisions,and embryodevelopment-tosaynothingof the detailedmo-
lecular chemistryof proteinsand DNA, which was still almosttotally
unknownin the 1940s.Machinesobviouslyhad none of that. So before
von Neumanncouldanswerthequestionaboutmachineself-reproduction,
he had to reducethat processto its essence, its abshactlogicalform. In
effect,he had to operatein the samespiritthat programmers would years
laterwhen theystartedto build virtual machines:he had to find out what
wasimportanfaboutself-reproduction, independentof the detrailed bio-
chemicalmachinery.
To get a feel for the issues,von Neumannstertedout with a thought
experiment. Imaginea machinethatfoatsaroundon thesurface of a pond,
he said, togetherwith lots of machineparts.Furthermore,imaginethat
thismachineis a uniyersalconstructor: givena description of anymachine,
it will paddlearoundthe ponciuntil it locatesthe properparts,and then
constructthat machine.In particular, givena description of itself,it will
construct a copy of itself. i
Now thatsoundslike self-reproduction, saidvonNeumann'But it isn't-
at least,not quite. The newly created copy of the 6rst machinewill have
all the right parts.But it won't have a description of ibelf, which means
that it won't be able to make any further copies of itself.Sovon Neumann
alsopostulated that the original machine should have a desciptioncopiet:
deui"ethatwill take the original description, duplicate it, andthenattach
"the duplicatedescriptionto the offspringmachine.Oncethat happens,he
said,tire offspringwill haveeverythingit needsto continuereproducing
indefinitely.And then that will be self-reproduction'
 Life at the Edge of Chaos 219
As a thought experiment,von Neumann'sanalysisof self-reproduction
was simplicity itself. To restateit in a slightly more formal way, he was
sayingthat the genetic material of any self-reproducingsystem,natural or
artificial, has to play two fundamentally different roles. On the one hand,
it hasto serveasa program,a kind of algorithmthat can be executedduring
the construction of the offspring. On the other hand, it has to serve as
passivedata, a descriptionthat can be duplicatedand given to the offspring.
But as a scientific prediction, that analysisturned out to be breathtaking:
when Watson and Crick finally unraveledthe molecular structureof DNA
a few yearslater, in 1953, they discoveredthat it fulfilled von Neumann's
two requirements precisely.As a genetic program, DNA encodesthe in-
structions for making all the enzymesand structural proteins that the cell
needsto function. And as a repository of genetic data, the DNA double
helix unwinds and makesa copy of inelf every time the cell divides in two.
With admirable economy,evolution hasbuilt the dual nature of the genetic
material into the structure of the DNA molecule itself.
That was still to come, however. In the meantime, von Neumann knew
that a thought experiment alone wasn't enough. His image of a self-
reproducing machine on a pond was still too concrete, too tied to the
material detailsof the process.As a mathematicianhe wantedsomething
that was completely formal and abstract.The solution, a formalism that
eventually becameknown as the cellular dutomaton, was suggestedby his
colleagueStanislasUlam, a Polishmathematicianwho had taken up res-
idence at Los Alamos and who had been thinking about many of these
issueshimself.
What UIam suggestedwas the same framework that fohn Conway was
to use more than twenty yearslater when he invented the Game of Life;
indeed, as Conway was well aware, the Game of Life was just one special
caseof a cellular automaton. Essentially,Ulam's suggestionto von Neu-
mann was to imagine a programmableuniverse."Time" in this universe
would be definedby the ticking of a cosmic clock, and .,space"would oe
defined to be a discrete lattice of cells, with each cell occupied by a very
simple, abstractly defined computer*a finite automaton. At any given
time and in any given cell, the automaton could be in only one of a finite
number of states,which could be thought of as red, white, blue, green,
and yellow, or 7,2,3,4, or living and dead, or whatever. At each tick of
the clock, moreover,the automatonwould make a transition to a new state,
which would be determined by its own cu*ent stateand the current state
of its neighbors. The "physical laws" of this universe would therefore be
220 COMPLEXITY
encoded in its stcte transition table: the rule that tells each automaton
which state to change to for each possibleconfiguration of statesin its
neighborhood.
Von Neumann loved the cellular automaton idea. Here was a system
that was simple and abstractenough to analyzemathematically,yet rich
enough to capture processeshe was trying to understand' And, not inci-
dentally, it was exactly the sort of systemthat you could simulate on a real
computer-at least in principle. Von Neumann'swork on the theory of
cellular automata was left unfnished at the time of his death from cancer
in 1954. But Art Burks, who had been askedto edit his paperson the
subiect, subsequentlyorganized what was there, Flled in the remaining
details, and published the collection asTheory of Self-ReproducingAuto'
mata in 1966. One of the highlightswasvon Neumann'sproof that there
existedat least one cellular automaton pattern that could indeed reproduce
itself. The pattern he'd found wasimmenselycomplicated,requiring a huge
lattice and 29 different statesper cell. It was far beyond the simulation
capacig of any existing computer. But the very fact of its existencesettled
the essentialquestionofprinciple: self-reproduction, onceconsideredto be
an exclusive cha.racteristicof living things, could indeed be achieved by
machines.

In reading all this, saysLangton, "All of a sudden I felt very confident.

I knew I was on the right track." He went back to his Apple II and quickly
wrote a general-purposecellular automaton program that would let him
watch the cellular world as a grid of colored squareson the screen. The
memory limitations of the Apple-it only had 64 kilobytes-meant that
he could allow no more than eight statesper cell. That ruled out von
Neumann's 29-stateself-reproducer. But it didn't rule out the possibility
of finding c self-reproducingsystemwithin those limits' Langton had set
up his program so that he could try out any set ofstates and any transition
ta-blehe wanted. And with eight statesper cell, that left him with only
about l0ro,0mdifferent tablesto explore. He went to it.
Langton alreadyknew that his quest wasn't as hopelessas it seemed' In
his reading he'd discoveredthat Ted codd had found an eight-stateself-
reproducingpattern more than a decadeearlier,when he wasa grad student
at Michigan working under some guy named fohn Holland. And while
Codd's prtt ..t was still too complex for the Apple II, l,angton thought that
by playing with the variouscomponentsof it, he might be able to implement
somethingsimpler that would fit within his own constraints.
Life at the Edge of Chaos 221

"All of Codd'scomponentswerelike datapaths,"saysLan$on' That is,

four of the eight automatonstatesin Codd'ssystemactedas bits of data,
while the other four playedvariousauxiliary roles.In particular,there was
one statethat functioned as a conductor and another statethat functioned
as a kind of insulator, so that togetherthey defined channelsthrough which
the data would flow from cell to cell as if the path were a copper wire' So
l,angton began by implementing Codd's "periodic emitter" structure: es-
sentially a loop in which one bit of data circulated around and around like
the secondhand of a clock, along with a kind of arm that grew out from
the side of the loop and periodicallyfired off a copy of the circulatingbit.
And then langton startedmodifying the emitter, putting a cap on the arm
so the signalswouldn't escape,addinga secondcirculatingsignalto make
the cap, tweakingthe rule table, ad infinitum' He knew he could do it, if
only he could make that arm grow out, curl back on itself, and make a
loop identical with the first.
It was slow going. Langton worked into the wee hours every night, while
his wife Elvira did her bestto be patient."She caredthat I was interested
in it and that I thought somethingwashappening,"saysLangton. "But she
was more concernedwith, What are we going to do?Where is all this work
going to take us? How is it conhibuting to the progressof the domestic
iituation? Where are we going to be in two years?That was very hard to
explain. So you've-d-onethis.Now you do wfrat with it? And I didn't know'
I iust knewjt<ff, important. "
I-angton could only keep plugging away. "l kept getting this piece and
that piece," he says."I would start with a rule, then I'd modify it, and
modify it again, and then I'd paint myselfinto a corner' I filled up fifteen
floppy diskswith preservedrule tables,so that I could back up and take off
in a different direction. So I had to keep very careful recordsof what rule
produced what behavior, and what changed, and what I'd backed up to,
and what disk I'd storedit on."
All told, saysLangton, it was about two months from the time he first
read von Neumann until he finally got what he wanted. One night, he
says, the pieces iust finally came together. He sat staring at loops that
extendedtheir arms, curled thosearmsaround to form new, identical loops,
and went on to form still more loops ad infinitum. It looked like the growth
of a coral reef. He had createdthe simplest sg!fueproducillce'llular au-
"l hadiF s ili66di6le-volcano of emoti6ffiE"
-says.
@
"This is possible.ltdoes work. This is true. Evolutionmadesense
now. This wasn'tan externalProgramthat iust manipulateda table.This
hadclosureon itself,sothattheorganismwcstheProgram.It wascomplete.
 222 COMPLEXITY

And now all thesethingsthat I'd beenthinkingof that might be the case
if I coulddo this-well, theywereall possible,too. It waslike a landslide
The dominoesfell, and iust keptfallingand fallingand
of possibilities.
falling."

The Edgeof Chaos

"l'm in part a mechanic,"saysl.angton."l haveto get my handson

something,put it together,seeit work.And onceI'd actuallyput something
together,anydoubtsI had weregone.I could seewhereartificiallife had
to go from there."In his own mind it wasall crystalclear:now that he
had self-reproduction in the cellularautomatonworld, he would haveto
addon the requirementthat thesepatternsqgrforrn;9rne-tasL-before thgy
can relg){qce, like collectenoughenergyor--noughof the right comPo-
*ffi-ts.
He would haveto build wholepopulationsof suchpatterns,sothat
theycompetedwith eachotherfor theseresources. He wouldhaveto give
them the ability to move around and sense each other. He would haveto
allow for the possibilig of mutations and errorsin reproduction. "These
wereall problems to be solved,"he says."But okay-in thisvon Neumann
world, I knewI could embedevolution'"
Armedwith his newself-reproducing cellularautomaton,Langtonwent
backto campusand startedmaking the roundsagain,hying to drum up
supportfor his interdisciplinary Ph.D. Program'"f[[5-," he would tell
p"opt., pointingto the unfoldingstructures on the screen,"fftis is what I
wantto do."
No go. If anything,the response waschillier than before."There was
,o *,r.h to explainat thispoint," he says,"The peoplein the anthroPology
departmentdidn't know aboutcomputers,period,let alonecellular au-
tomata.'How is this any differentfrom a videogame?'Thepeoplein the
computersciencedepartment didn't knowaboutcellularautomata,either.
'What'sself-reproduction got
And they weren'tinterested in biologyat all.
to do with computerscience?'So when you tried to paint the whole pic-
ture-hey, you soundedlike a complete,babblingidiot'
"Well, I knew I wasn'tcrazy,"he says."I felt very saneby then' I felt
sanerthan werybodyelse.In fact, I worriedaboutthat-I'm surethat's
. what crackpotsfeel." But saneor not, he clearlywasn'tgettinganywhere
at Arizona. It wastime to startlookingelsewhere.
 Life at the Edgeof Chaos 223

Langton wrote to his former philosophy adviser,Wesley Salmon, who

had moved to the Universig of Pithburgh: "What do I do?" Salmon wrote
back with a suggestionfrom his wife: "Study with Art Burks."
Burks?"l'd just assumedhe wasdead,"saysLangton."Almost everybody
else from that era was." But Burks turned out to be very much alive at the
University of Michigan. More than that, once Langton startedcorrespond-
ing with him, Burks was very encouraging.He even arrangedfor l,angton
to get financial supportas a teachingassistantand researchassistant.]ust
apply, he wrote.
l,angton wasted no time. By that point he d learned that Michigan's
Computer and Communication Sciencesprogram was famous for exactly
the kind of perspectivethat he was after. "To them," saysLangton, ,'in-
formation processingwas writ large acrossall of nature, However infor-
mation is procesed is worthwhile understanding.So I applied under that
philosophy."
By and by he got a letter back from ProfessorGideon Frieder,the chair-
man of the department. "Sorry," it said, "you don't have the proper back-
ground." Application denied.
Langton was enraged.He fired back a seven-pageletter, the gist of which
was, What the helltt "Here is your whole philosophy, the purpose you
claim to exist,live, and breathefor. This is exactlywhat I've beenpursuing.
And you're telling me no?"
A few weeksafterward Frieder wrote again, sayingin effect, ..Welcome
aboard." As he told l,angton later, "I iust liked the idea of having somebody
around who would saythat to the chairman."
Actually, as Langtonwasto learn, there had beenmuch more to it than
that. Neither Burksnor Holland had everseenhis originalapplication.For
a variety ofbureaucratic and financial reasons,the broad-baseddepartment
they had spent thirty yearsbuilding was on the verge of merging with the
departmentof electricalengineering,where peoplehad a much more hard-
nosed,practicalidea of what constitutedworthwhile research.And in an-
ticipation, Frieder and others were already trying to deemphasizesuch
things as "adaptivecomputation." Burks and Holland were Fghting a rear-
guard action.
Fortunately or unfortunately, however, Langton didn't know this at the
time. He was iust glad to have goftenaccepted."I iust couldn't passthis
up," he says,"especiallywhen I knew I was right." Elvira Segura-l,angton
 224 COMPLEXIW

waswilling to giveit a try. True, shewouldhaveto giveup her iob at the

And theywouldn'tbe nearherfamilyin Arizonaanymore.But
university.
consideringthat shewasnowpregnantwith their firstchild, it wouldn'tbe
sucha bad ideato be coveredby chris'sstudenthealthplan.And besides,
/as much astheyboth lovedthe southwest,it might be fun to seea cloud
headednorth'
l.u.ry onceand a while. So in the fall of 1982,they

Intellectually,at least,langton had a grandtime at Michigan' As a

teachingassistant for Burks'historyof computingclass,he absorbed Burks'
fevewitness account of thoseearly days-and helped Burks put together an
l.*t iUit on ENIAC using some of the originalhardware. He met fohn
Holland. And for his classin integratedcircuitry,Langtondesignedand
built a chip for the ulha-fastexecutionof a part of Holland'sclassifier
system.
But mostly,Langtonstudiedlike mad. Formallanguagetheory,com-
putationalcomplexitytheory,datastructures,compilerconstruction-he
wassystematically learningmaterialhe'd only absorbed in bits and pieces
it.
before.And he loved Burks, Holland, and company were nothingif not
demanding;once while langton was at Michigan they funked almost
everybody takingthe oral "qualifier"exams to getinto the Ph,D. plogram.
(Theflunkeesdid getanotherchance.) "Theycouldaskyoustuffthatwasn't
evenin the courses,andyou'dhaveto saysomethingintelligentaboutit,"
saysLangton."Well, I reallyenioyedthatkind of learningProcess. There's
. tig diff.t.nce betweenhaving passed the courses and knawing the ma-
terial."
on the academicpoliticsfront, however,thingswerenot so grand.By
late 1984,when Langtonhadfinishedhis coursework,gottenhis master's
degree,passed his qualifiers,and wasreadyto begina Ph-D. dissertation,
it had becomepainfully obviousthat he wouldn't be allowedto do the
^ thesishe wantedon the evolutionof artificial life in a von Neumann
Nr
'i1"aCe I u.,iurrr.. Burks' and Holland'srear-guardaction was failing. The old
\ Computerand CommunicationSciences departmenthad beenabsorbed
lA llnto in. engineering
i I /{ I't schoolin 1984.And in the dominantengineering
V "naturalsystems" curric-
cultureof ii newhJme, the Burks-Holland-s$le
ulum waseffectivelybeingphasedout. (The situationwas and is one of
the few thingsthat i{olland geb visiblyangryabout;he had initially been
one of the stiongestvoicesin favorof the merger,havingbeen convinced
that the naturaliystemsapproachwouldbe protected,and he now felt as
 life at the Edge of Chaos 225
thoughhe had beensuckered.Indeed,the situationgaveHollandconsid-
erableextraincentiveto startgettinginvolvedwith the santaFe Institute
aboutthis time.) But asthe betterpart of valor,Burhsand Holland both
urgedLangtonto do a thesisthat wasa little lessbiologicaland a little
morecomputer-sciencelike. And langtonhadto admittheyhada point-
if only as a mafterof practicality."r had enoughsawy by then to know
that this von Neumannuniversewasgoingto be an extiemelydifficult
systemto setup and getgoing,"he says.So he startedlookingaroundfor
somethingthat might be doablein the spaceof a yearor twJ, insteadof
decades.
well, he thought,insteadof trying to build a completevon Neumann
universe,why not startby trying to understanda little more about the
"physics"of that universe? why not try to understand why certaincellular
automatonrule tablesallowedyouto build interesting structures,andothers
didn't?That wouldat leastbea stepin therightdireciion.It wouldprobably
containenoughhard-coreeomputerscienceto satisfytheengineering types.
And with anyluck, it mightproducesomeinteresting connection, to ,."t
physics.Indeed,this cellularautomata-physics connection had becomea
certifiablyhot topic of late. As physicswhiz-kid stephenwolfram had
pointedout in 1984,when he wasstill at caltech, automatanot
only havea rich mathematical structurebut havedeep".lrul*
similaritiesto non-
lineardynamicalsystems.
- whlt Langtonfoundparticularlyfascinating waswolfram'scontention
that all cellular automatarulesfall into one of fou, universality,lorr"r.
wolfram'sclassI containedwhatyou mrghtcall
ffi
whatpatternof living anddeadcellsyou startedthem ouiwith, everything
would just die within one or two time steps.The grid on the computer
screenwouldgo monochrome.In the language ofdynamicalsystems, such
rulesseemedto havea single"point athactor."Thrt is, the system,"r-ra
to be mathematically like a rnarblerollingtowardthebottomtf a big cereal
bowl:no matterwherethe marblestartedout on the sidesof the 6owl, it
would alwaysroll down very quickly to a point in the center*the dead
state.
wolfram'sclass II ruleswerea little morelively, but not much. with
theserules,an initial patternthat scattered living and deadcellsoverthe
screenat randomwould quickly coalesceinto a set of staticblobs,
with
perhaps a fewotherblobsthatwouldsitthereperiodically oscillating.'Such
automatastill gavethe generalimpression of frozenstagnation and death.
In the languageof dynamicalsystems, theseruresseei-,edto havefalen
226 COMPLEXITY

into a set of periodic attractors-that is, a seriesof hollows in the bottom

or u"-py cerealbowl where the marble could roll around and around
"
the sidesindefinitelY.
wolfram's class III rules went to the opposite extreme:they were too
lively. They producedso much activi$ that the screenseemedto be boiling'
Nothing *as stableand nothing was predictable:structureswould break up
these
almost L roon asthey formed. ln the languageof dynamical systems,
rules correspondedio "strange" attractors-more commonly known as
around inside that cereal bowl so
$p-oq. Thei were like a marble rolling
iast and so hard that it could never settle down'
Finally, there were wolfram's class IV rules. Included here were those
but
rare, impossible-to-pigeonholerules that didn't produce frozen blobs,
that didn't producetotal chaos,either.What theydid producewere coherent
structuresihat propagated,grew, split apart, and recombined in a wonder-
they
fully complex *"y. Th.y essentiallyneversettleddown. In that sense
*er" all u.ry *u"h like the most famousmember of class IV the Game
of Life. And in the languageof dynamical systems,they were ' ' '
Well, that was just the pioblem. There was nothing in the conventional
rule.
theory of dynamical systemsthat looked anything like a Class IV
Wotfiam had coniectured that these rules represented a kind of behavior
had the
that was unique to cellular automata.But the fact wasthat no one
they represented,'N9r,-!or that matter' did anyone
slightest idea of what
produced Class IV behavior and another
ha"vethe slightestideawhy onerule
to was
didn't; the only way to find out which classa given rule belonged
to try it out and seewhat haPPened.
For Langton, the situation not only was intriguing, but revived the-old
..Becauseil's not there" feeling he'd once had about anthropology. Here
Neumann
were the very rules that seemedessentialto his vision of a von
so much of what was important about the
universe, that seemedto capfure
And yet they seemed
spontaneousemergenceof iife and self-reproduction.
unknown realm of dynamics. so he decided to
to lie in ,o*.
"o-pl.tely classes related to one
tackle the problem head on: How were wolfram's
to?
another, ,.,d *h"t determined the classa given rule belonged
One id.a struck him almost immediately. At about that same time, as
it happens,he had been doing quite a bit of reading on dynamical systems
systems, he knew' the equation of
And in *"ny ,."[onlinear
"nd "i"or. functions as a kind of tuning
motion containsa @
teattris"*If thesystem
knob,controllingiuttEtuifiaotl9_.4f9-yrtem were
theparameter would be the rateof
drippingwate;f;;;i5l;;;ple,
"*.t , noi. or if the system of
werea population rabbits, the parameter
 Life at the Edge of Chaos 227
would involvea ratio betweenthe rabbits'birth rateand their deathrate
due to overcrowding.In general,a small value of the parameterwould
I
usuallycorrespondto stablebehavior:equal-sized drops,a constantpop- |
ulation,andsoforth.And that,in turn, seemed highlyreminir"..,toitli" t
staticbehaviorseenin wolframt classI andclasslt. But astheparameterI
got progressively largerand larger,the behaviorof the systemwould get j
progressively morecomplicated-different-sized drops,a fluctuatingpop-,i
ulation,andsoon-until ultimatelyit became compietely chaotic.aitr,rt
point it would resembleWolfram'sClassIIL
l,angtonwasn'ttoo clearwhereclassIV wouldfit into this picture.But
the analogywastoo goodto ignore.If he could only find ,om. w"y
of
assigning a similarkind of parameter to thecellurarautomaton ,ules,tien
thewolframclasses mightstartto makesomesense. of course,he couldn,t
just assignthe numbersarbitrarily;this parameter, whateverit turnedout
to be, would haveto be derivedfrom the rule itself. He might measure
eachrule'sdegreeof reactivity-for example,howoftenit causJs thecentral
cell to changeits state.But therewereany numberof thingsto try.
So Langtonstartedprogramming his computerfo, .uery!"rameterthat
seemedhalfwayreasonable. (one of the firstthingshe haddonewhen he
cameto Michigan wasto implementa moresophisticated versionof his
Apple II cellular automatonprogramon a high-powered, blazinglyfast
Apollo workstation.) And he got absorutery no*h.r.-uniil, on" i"y, h"
hied one of the simplestparameters he could think of. The Greekietter
lg*?91 f\),.as he ca.lledit, wasiust the probabilitythat any givencell
wourdbe "arlve"rn the nextgeneration. so if a rule hada lamMavalue
of precisely 0.0, for example,then nothingwouldbe ariveafterthe irst
timestepandtherulewouldclearlybein crassL If therulehada lambda
of 0.50, then the grid wouldbe boilingwith activig,with half ,r,.
..rt
a-liveon the average and half dead.presumably, sucha rule would be in
the chaotic class III. The questionwas whetherlamMa would reveal
anythinginteresting in between.(Beyond0.50theroleof ,,alive"and,dead,,
wouldbe reversed, and_things wouldpresumably getsimpler ,rntit
you werebackat classI whenyou reached1.0;iiwould be lik"e "g"in
looki.,g
at the samebehaviorin a photographic negative.)
To testthe parameter, Langtonwrotea liftle programtelringhis Apolro
to automaticallygenerateruleswith a specificvalueof lambd"a,
then
run a cellularautomatonon the screento showhim whatthat ,ule "nd
would
do' "well, the firsttime I ran it," he says,"I setlambdaequ"r
to o.io,
thinking-I wassettingit to a totallyrandomstate-and I suddenly
started
gettingall theseclassIV rules,oneafteranother!I thought,.God,
thisis
225 COMPLEXITY

too goodto be true!'So I discovered, sureenough,that therewasa.bug

in thi prog.am that was actually setting lambda to a differentvalue-which
to be the critical value for that class of automata'"
'iusthappened
Fixingthe bug in his program, langton startedexploring variouslambda
valuessistematilally. Ai uety low values around 0.0 he found nothingbut
the dead,frozenCiass I rules. As he increased the values a little bit, he
startedfindingperiodic Class II rules. As he increased the value a little
more,he that the Class II rules took longer and longer-to settle
"otL.a jumped all the way to 0.50, he found himself in the
down.Then if he
total chaosof ClassIII, iust as he expected. But right there in between
ClassesII and III, clusteredtightly aroundthis magic*ritt9el-vah9-9f.
lambda(abpur_0J71), he foundwholethicketsof complexclassIV rules.
;;Afr-:[. Gameof Life wasamongthem. He wasflabbergasted. Some-
how,'thissimpleminded lambda parameter had put the wolfram-classes
into exactlyttret ind of sequence he'dwanted-andhadfounda placefor
the ClassIV rulesto boot, right at the transitionpoint:
I & Il + "IV" -+ III
Moreover,that sequencesuggested transitionin
an equallyprovocative
dynamicalsystems:
Order---r"Complexity"-> Chaos
where "complexity" referredto the kind of eternally surprisingdynamical
behavior shown by the Class IV automata'
"lt immediatelybroughtto mind somekind olpbrt. tttnt!$," he says'
like temperah'rre'
Supposeyou thought of the parameterlamMa as being
would
Then theClass I and lI rules that you found at low valuesof lambda
rigidly locked
.orr.rpond to a solid like ice, wherethe watermoleculesare
i.,to n lattice. The Class III rules that you found at high values of
"ryrt"l correspondto a vapor }ike steam, where the molecules are
lambda would
flfing and slamminginto each other in total chaos'And the Class
"ro,r.d
you found in betwien would correspondto-what? Liquids?
iV,ifo
"l knew very little about phasetransitions at the time"' sayslangton'
,,but I dug into what w", k rorr about the molecular structureof liquids."
discovered'
At first it'iooked very promising:the moleculesin a liquid' he
over and around eachother, bonding and clustering
are constantlytumbling
apart again billions of times per second-a lot like the shuc-
and breakinj
oilif.. "It seemed to me quite plausible that something
tures in the bame
level iust in a
like the Game of Life might be going on at the molecular
glassof water," he saYs.
 Life at the Edge of Chaos 22g
langton lovedthat idea. And yet, as he thoughtaboutit somemore,
he beganto realizethat it wasn'tquite right. classIV rurestypicalrypro-
duced"extendedtransients,"suchasthe Gameof Life'sgliders:strucfures
that could surviveand propagate for an arbitrarilylonj time. ordinary
liquidsdidn't seemto haveanythinglike thatat a molecularlevel.So far
as anyonecould tell, they werealmostas completelychaoticas gaseous
matter.Indeed,Langtonhad learnedthat by increasingthe temperature
a-ndpressure enough,you couldgo from steamto waterwithoutevergoing
througha phasetransitionat all; in general,gases andliquidsa.e luJ trvo
aspects of a singlefluid phaseof matter.so the distinctionwasn'ta fun-
damentalone, and the resemblance of liquidsto the Gameof Life was
only superhcial.
Langtonwentbackto hisphysicstextsandkeptreading.,,Finally,I came
acrossthebasicdistinctionbetween,first-adt{ andsecond-order phasetran-
sitions,"he says.lU*:orfuJrg-ryi!1o-qgpre ttrekind we'reall familiarwith:
sharpand precise.Raisethe temperatureof an ice cube past iz"F, for
example,and the changefrom ice to waterhappens all at once.Basically,
what'sgoingon is thatthe moleculesareforcedto makean either-orchoice
betweenorderand chaos.At temperatures belowthe transition,they are
vibratingslowlyenoughthat they can makethe decisionfor crystalline
order(ice).At temperatures abovethe transition,however,the molecules
arevibratingsohardthatthe molecularbondsarebreakingfasterthanthey
can reform,so theyareforcedto opt for chaos(water).
S:"""4-qtltt qh' aremuchlesscommonin nature,L,ang-
ton learned.(At least,theyarJit the temperatures andpressures humans
areusedto.) But theyaremuchlessabrupt,largelybecause the molecules
in sucha systemdon't haveto makethat either-orchoice.They combine
chaosand order.Abovethe transitiontemperature, for e*ample,mostof
the moleculesaretumblingoveroneanotherin a completelyciraotic,fluid
phase.Yettumblingamongthemaremyriadsof submi".oscopic islandsof
orderly,latticework solid,with molecules constantly dissorving andrecrys-
tallizingaroundthe edges.Theseislandsareneitherverybig.,-o,uerylo.,g-
Iasting,evenon a molecularscale.so the system is stili mJstlychaos.But
asthetemperature is lowered,the largestislandsstartto getverybig indeed,
andtheybeginto livefor a correspondingly longtime.Thebalan"Jetween
chaosand order has begunto shift. of .ourr., if the temperaturewere
takenall the waypastthe transition,the roleswould reverse, the material
would go from beinga seaof fuid doftedwith islandsof solid,to beinga
continentof soliddottedwith lakesof fuid. But rightat the transition,I-,.
balanceis perfect:the orderedstructuresfill a volumeprecisely.q,r"l to
 230 COMPLEXITY

that of the chaotic fluid. Order and chaos intertwine in a complex, ever-
changing dance of submicroscopicarms and fractal filaments. The largest
ordereditructures propagatetheir fingers acrossthe material for arbitrarily
long distancesandlast for an arbitrarily long time. And nothing ever really
settlesdown.
l,angton was electrified when he found this: "There was the critical con-
nectioil There wasthe analog to Wolfram's ClassIV!" It was all there. The
propagating,gliderlike "extendedtransients,"the dynamicsthat took forever
L r.ttf. dowt, th. inhicate dance of struchrresthat grew and split and
recombined with eternally surprising copplexity-it practically defined a
! second-orderphasetransition.
i So now Langton had yet a third analogy:

CellularAutomatdClasses:
I&Il+"1!"+III
DynamicalSysfems:
Order-+ "Complexity"'+ Chaos
Motter'.
Solid+ "Phasetransition"+ Fluid
The questionwas,wasit anythingmorethan an analogy?Langtonwent
right io work, adaptingall mannerof statisticaltestsfrom the physicists'
w"orl<larrdapplyinjthemto thevon Neumannworld.And whenhe plotted
his resultsai a functionof lamMa, the graphslookedlike somethingright
out of a textbook.To a physicisttheywouldhavescreamed "second-order
I phase transition." Langton had no idea why his lamMa parameter_wgrked
lro *.11. or whv it seemed soclosely analogous to (Indeed'
temperature' no
it yet.) But therewasno denyingthat it did' The
J on. ,.rlly understands
phasetransition wasn'tiust an analogy'lt wasreal'

At one time or another,Langton usedeveryname for this phasetransition

that he could think ofi the "hansition to chaos,"the "boundaryofchaos,"
feeling
the "onset ofchaos." But the one that really capturedthe visceral
it gavehim was "the edge of chaos'"
zlt
reminded me of the feelings I experiencedwhen I learned to scuba
fairly
dive in Puerto Rico," he explains."For most of our dives we were
the water was crystal clear and you could see the
close to shore, where
feet down. However, one day our inshuctor took
bottom perfectly about 60
gaveway
u, to the edgeof the continental shelf, where the 60-foot bottom
 Life at the Edge of Chaos 23'l
to an 8O-degreeslope that disappearedinto the depths-l believe at that
point the transition was to about 2000 feet. It made you realize that all the
diving you had been doing, which had certainlyseemedadventurousand
daring,wasreallyjust playingaroundon the beach.The continentalshelves
are like puddlescomparedto 'The Ocean.'
"Well, life emerged in the oceans," he adds, "so there you are at the
edge,alive and appreciatingthat enormousfluid nursery.And that's why
'the
edgeof chaos'carriesfor me a very similar feeling:becauseI believe
life also originated at the edgeof chaos. So here we are at the edge, alive
and appreciating the fact that physics itself should yield up such a nur-
sery. ."
That's a poetic way to say it, certainly. But for Langton, that belief was
far more than iust poetics.The more he thought about it, in fact, the more
he became convinced that there was a deep connection between pliase
transitions and computation-and betweencomputation and life itself. \\\
The connectiongoesright backto-of course-the Game of Life. After
the game was discovered in 1970, saysLangton, one of the first things
peoplenoticed wasthat propagatingstructureslike gliderscould carry signals
acrossthe von Neumann universefrom one point to another. Indeed, you
could think of a flock of glidersgoing singlefile as being like a streamof
bits: "glider present" : l; "glider absent" = 0. Then, as people played
around with the game still more, they discoveredvarious shuctures that
could store such data, or that could emit new signals that encoded new
information. Very quickly, in fact, it became clear that Game of Life
structurescould be used to build a complete computer, with data storage,
information-processingcapability,and all the rest.This Game of Life com-
puter would have nothing to do with the machine the game was actually
running on, of course;whateverf/rat machine was-a pDp-g, an Apple
II, an Apollo workstation-it would just serveas the engine that made the
cellular automaton go. No, the Game of Life computer would existentirely
within the von Neumann universe,in exactlythe sameway that Langton,s
self-reproducing patterndid. It would be a crudeand inefficientcomputer,
to be sure. But in principle, it would be right up there with Seymour -ray's
finest. It would be a univercal computer, with the power to compute any-
thing computable.
Now, that's a pretty astonishingresult, saysLangton-especially when
you considerthat only comparativelyfew cellular automatonrulesallow it
to happen' You couldn't make a universalcomputer in a cellular automatonf
governedby class I or II rules, becausethe structuresthey produce are too{
static;you could storedata in such a universe,but you would have no way
 232 COMPLEXITY
to propagate the informationfrom placeto place.Nor could you makea
in a chaoticClassIII automaton; the signalswouldgetlost in
"o*puto
the noise,and the storagestructures would quickly getbafteredto pieces.
Indeed,saysLangton,the only rules that allow you to build a universal
computerarethosethat arein class IV like the Game of Life. Theseare
the only rulesthatprovideenoughstabili$to store information ondenough
fuidity to sendsignalsoverarbitrarydistances-the two things that seem
essential for computation. And, of course,thesearealsothe rules that sit
right in this phasetransitionat the edgeof chaos.
So phasetransitions,-eolnplelity, qL{-euUUtelig were all -wrapped
together,I-angtonrealized.-Or,at least,they werein the von Neumann
u.riu..r.. But LangtonwaSconvincedthatthe connections heldtrue in the
realworldaswell-in everything fromsocialsystems to economies to living
(-cells.Because onceyou gotto computation,youweregettingawfullyclose
I to the essence of life itself. "Life is basedto an incredibledegreeon ib
I abiliW to process information,"he says."lt storesinformation'It maps
lr.nrory iniormation.And it makessomecomplextransformations on that
informationto produceaction. [The EnglishbiologistRichard]Dawkins
hasthis reallynice example.If you takea rock and tossit into the air, it
tracesout a nice parabola.Itfs at the mercyof the lawsof physics.It can
only makea simpleresponse to theforcesthatareactingon it from outside.
Bui now if you take a bird and throwit into the air, its behavioris nothing
like that. It fliesoff into the trees somewhere. The sameforcesarecertainly
actingon this bird. But there's an awful lot of internalinformationpro-
c.rriig goingon that's responsible for its behavior. And that'strue evenif
you go down to simplecells: they aren't iust doing what inanimatematter
do.r. Th.y aren'tiustresponding to simple forces. So one of the interesting
questionswe can askabout living things is, Under what conditionsdo
systemswhosedynamics are dominated by information processing arise
from thingsthat iust respond to physical forces? when and where does the
processinjof information and the storage of information become impor-
tant?"
In an attemptto answerthat question,saysLangton,"I took my phase
transitionglasses and lookedat the phenomenology of computation.And
therewereenormousnumbersof analogies." When you takea classin the
I th.orv of computation,for example,one of the 6nt things youlearnabout
[ii, th"'dirtincti-onbetween programs that "halt"-that is, take in a stringof
I'd"t" produce an answer in a finite amount of time-and thosethat
"nd But that, l,angton says, is iust like the
\lurt t..p churning awayforever.
I
 Life at the Edgeof Chaos 233
distinction betweenthe behavior of matter at temperaturesbelow and abovei-.
a phasetransition. There is a sensein which the material is constantlyi
hying to "compute" how to arrangeitself at a molecular level: if it's cold,
then it reachesan answervery fast-and crystallizescompletely. But if it's
hot it can't reach an answerat all, and remainsfluid.
In much the same way, he says,the distinction is analogousto the one
between Class I and II cellular automata that eventually halt by freezing
into a stable configuration, and chaotic Class III cellular automata that
boil along nonstop. For example, supposeyou had a program that just
printed out one m€ssage on the screen-"HELLO WORLDI"-and quit.
Such a program would correspondto one of thoseClassI cellular automata
down around lamMa : 0.0, which go to quiescencealmostimmediately.
Conversely,supposeyou had a program with a seriousbug in it, so that it
printed out a steadystreamof gibberishwithout ever repeatingitself. Such
a program would correspond to Class III cellular automata out around
lambda = 0.50, where the chaosis maximal.
Next, saysLangton, supposeyou moved awayfrom the extremes,toward
the phase transition. In the material world, you would find longer and
longer transients:that is, as the temperaturegets closer to the phasetran-
sition, the molecules require more and more time to reach their decision,
Likewise,as lambda increasedfrom zero in the von Neumann universe,
you would start to find cellular automata that would churn around a bit
before they reachedquiescence,with the amount of churning depending
on just what their initial state was. These would correspondto what are
known as polynomial-time algorithmsin computer science-the kind that
have to do a significant amount of work before they finish, but that tend
to be relatively fast and efficient at it. (Polynomial-time algorithms often
crop up when the problem involves choressuch as sorting a list.) As you
went further, however,and as lambda beganto get very close to the phase
transition,you would begin to find cellular automatathat churned around
for a very long time indeed. These would correspondto nonpolynomial-
time algoiithms-the kind that might not halt for the lifetime of the uni-
v€rse,or longer. Such algorithms are effectively useless.(An extreme ex-
ample would be a program that tried to play chessby looking aheadat every
possiblemove.)
And right at the phasetransition?In the materialworld, a given molecule
might wind up in the ordered phaseor the fluid phase;there would be no
way to tell in advance,becauseorder and chaosare so intimately intertwined
at the molecular level. In the von Neumann universe,likewise,the class
 234 COMPLEXITY

IV rulesmighteventuallyproducea frozenconfiguration, or theymight

not. But eitherway, Langton says, the phasetransition at the edgeofchaos
would correspondto what computer call
scientists "undecidable" algo-
rithms.Thesearethe algorithmsthat might halt very quickly with certain
inputs-equivalent to startingoff the Gameof Life with a known stable
structure.But theymightrun on foreverwith otherinputs.And the point
is, you can'talwaystell aheadof time whichit will be-even in principle'
In fact, saysLangton,there'sevena theoremto that effect:the "undecid-
ability theorem"provedby the British logicianAlan Turing backin the
I930s.Paraphrased, the theoremessentially saysthat no matterhow smart
you think you are,therewill alwaysbe algorithmsthatdo thingsyou can't
predictin advance. The only wayto find out whattheywill do is to run
them.
And, of course,thoseare exactlythe kind of algorithmsyou want for
modelinglife and intelligence.So it's no wonderthe Gameof Life and
otherClassIV cellularautomataseemso lifelike.They existin the only
dynamicalregimewherecomplexity,computation,and life ilself arePos-
sible:the edgeof chaos.
I Langtonnow had four verydetailedanalogies-
CellularAutomataC/csses:
I & II --+ "[V" + III
DynamicalSystems:
Order--+ "Complexity"-> Chaos
Matter
Solid + "PhaseTransition"-+ Fluid
ComPutation;
Halting -+ "Undecidable"-+ Nonhalting
-along with a fifth and far more hypothetical one:
l

Life:
\ Too static -+ "Life/lntelligence" -+ Too noisy
But what did they all add up to? Just this, l,angton decided:"solid" and
..fluid"
are not iust hro fundamental phasesof matter, as in water versus
ice. They are two fundamentalclasses of dynamicalbehaviorin general-
including dynamical behavior in such utterly nonmaterialrealms as the
spaceof cellular automaton rules the spaceof abstractalgorithms. Fur-
or
tirermore, he realized, the existenceof these two fundamental classesof
Life at the Edgeof Chaos 235
dynamicalbehaviorimplies the existenceof a third fundamentalclass:
.,phase transition"behaviorat theedgeof chaos,whereyouwouldencounter
complexcomputationand quite possiblylife itself.
So did this meanthat you might onedaybe ableto writedown general
physicallawsfor the phasetransition,lawsthatwouldsomehowencompass
both the freezingandthawingof waterand the originof life?Maybe'And
maybelife begansomefour billion yearsagoas somekind of real phase
hansitionin the primordialsoup.langton had no idea.But he did have
this irresistiblevision of life aseternallytrying to keepib balanceon the
edgeof chaos,alwaysin dangerof falling off into too much orderon the
on"eside,and too much chaoson the other.Maybethat'swhat evolutionf
is, he thought:iusta process of life'slearninghow to seizecontrolof moreI
and moreof its own param€ters, so that it hasa betterand betterchanceI
to staybalancedon the edge. I
Who knew?Sortingit all out couldbethework of a lifetime. Meanwhile,
by 1986l,angtonhad finally gottenthe engineeringschoolto agreethat,
asa thesistopic, his ideason computation,dynamicalsystems, and phase
transitionsin cellular automatawere quite acceptable. But he still had
plentyof workto do in gettinghis basicframeworkfleshed out enough to
satisfyhis thesiscommittee.

Go, Co, Co, Yes, Yes!

Two yearsearlier,in fune 1984,langton had goneto a conferenceon

cellularautomataat MIT and had happened to sit downat lunch one day
nextto a tall, rangyguy with a ponytail.
"What areyou workingon?" askedDoyneFarmer.
"l don't reallyknow how to describeit," admittedLangton."I've been
callingit artificiallife."
"Artificial life!" exclaimedFarmer."Wow, we gottatalk!"
Sotheyhadtalked.A lot. After the conference, moreover,theyhadkept
on talkingvia elechonicmail. And Farmerhad madeit a point to bring
Langtonout to LosAlamoson severaloccasions to givetalksandseminars.
(lndeed,it wasat the "Evolution,Games,and learning" conference in
May 1985that Langtongavehis first public discussion of his lambdapa-
rameterandthephasetransitionwork.Farmer,Wolfram,NormanPackard,
andcompanywereprofoundlyimpressed.) This wasthe sameperiodwhen
Rrmer wasbusywith Packardand StuartKauffmanon the autocatalytic
236 COMPLEXITY

setsimulationfor the origin of life-not to mentionhelpingto get the

SantaFe Instituteup and running-and he wasgettingdeeplyinvolved
with issuesof complexighimself.ChrisLangton,he figured,wasprecisely
thekind of guyhewantedto havearound.Furthermore, asa formerantiwar
activisthimself,he wasableto convincelangton that doingscienceat a
nuclearweapons laboratory wasn'tquiteasspookyasit mightseem:since
the kind of researchthat Farmerand his groupdid wascompletelynon-
classifiedandnonmilitary,youcouldthinkof it asa wayof divertingsome
of that "dirty" moneyto a goodpurpose.
The upshotwasthat in August1986,langton, his wife, and their two
infantsonsheadedsouthfor New Mexicoand a postdoctoral appointment
at Los Alamos'Centerfor NonlinearStudies. The move was a big relief
to Elvira Segura-Langton; afterfour yearsof Michigan snow and rain, she
it
couldn'twaitto getbackto thesun.And sounded wonderful to Langton
himself;the Centerfor NonlinearStudies was exactly
the kind of place he
wantedto be. True, he still had a few more computer runs to do before
he wasquite finishedwith his dissertation. But that wasn'tunusualfor
peopleiust startingtheir first postdoc.He shouldbe ableto wrapit all up
andactuallygethis Ph.D. in iusta few months.

It didn't quite work out that way.To 6nishhis computerrunsat Los

Alamos,l,angtonneededthe useof a workstation.That wasno problem
in principle.By the time he got therethe Centerfor NonlinearStudies
had alreadyreceiveda wholeshipmentof worlstationsfrom Sun Micro-
systems, alongwith all the cablesand hardwareneededto link them into
a local areanetwork.But it wasa nightmarein practice.The machines
werescattered aroundin variousbuildingsand hailers,andthe centerwas
who had not a clue how to makethe systemwork' "Well,
full of physicists
I wasa computerscientist,"saysLangton,"so they iust assumedI knew
how. So I becamethe defaultsystemmanagerfor the machinesin my
area. "
JohnHolland,who wascochairmanof Langton'sdissertation committee
alongwith Burks,and who arrivedat Los Alamosshortlyafter Langton
did to begina yearas a visitingscholar,wasappalled."Chrisis iust too
gooda guy," he says."Any time anyonehad a problemwith the network,
or with usinga Sun,they'dcometo him. And Chris,beingChris,would
spendwhatevertime wasnecessary out. The
trying to get it straightened
firstseveralmonthsI wasthere,he spentmoretime on that than anything
Life at the Edge of Chaos 237

else.So here he is pulling wiresthroughthe wall. But all that time his
dissertation is standingstill.
"Art Burl.sand I were continuallyat Chris on this-as wasDoyne,"
'Look,you'vegotto getyour union
Hollandadds."The themewasalways,
card,or you'regoingto regretit later.'"
langton understood that message completely;he wantedto gethis dis-
sertatiJnfinishediust asbadlyastheywantedhim to. But evenwhen the
networkwasup andrunning,of course,he still hadto takeall thecomputerI
codeshe'dwrittenfor the Apolloworkstation at Michiganandrewritethem !
to run on his Sun at Los Alamos.Nas$ iob. And then therewere the i
preparations for his artificiallife workshopin September 1987.(Partof the
whenhe'dcometo LosAlamoswasthathe couldorganizesuch
"gr..ment
a workshop.) And-"Well, thingsiustran awayfrom me," he admits'"I
didn't get anything doneon cellularautomatathat firstyear'"
what tangton did getdonewastheworkshop.Indeed,he threwhimself
into it with everything he had. "l wasdesperate to getbackinto artificial
life," he says. "At Michigan I had done tons of computerizedliterature
'Self-
searches, and they were frustrating. If you used the keyword
Reproduction,'you'd geta floodof stuff.But if you tried'Computers and
Seif-Reproduction,' you got nothing. And yetI kept stumbling across articles
in weird, out-of-the-way places."
He couldsense it. Somewhere out therewerethe authorsof thoseweird,
out-of-the-way articles: people iust like he hadbeen,lonelysoulstrying to
follow this bizarre scent all by themselves without quite knowingwhat it
was,or who else might be doing it too. l,angton wantedto find thesepeople
and bring them together, so that they could begin to forgea real scientific
discipline.The question was how
In the end, saysLangton,there was only one way to do it: "l iustl
announcedthat therewouldbea conference on artificiallife, andwe'dseeI
f
who showedup." Artificial life wasstill a goodlabel,he decided."I'd been
usingit sincethe Universi$of Arizona,and people immediately grokked
whai t meant." On the other hand, he considered it crucial that people
understandthat term very, very clearly,or else he might startpulling in
everyflakein the countrywho wantedto show a whacked-out videogame'
"I spenta long time, aboutone man-month, wording the invitation," he
says."We didn't want the conference to be too far out, or too science
fictiony.But we didn't wantto iustlimit it to DNA databases, either.So
I passed the invitationaroundhereat LosAlamos. I refined it. I wentover
and overand overit again."
238 COMPLEXITY
Then, once he had the invitation worded the way he wanted it, there
twas the question of how to broadcastit. Via nationwide electronic mail,
,maybe? In the UNIX operating systemthere was a utility program called
SENDMAIL, which had a well-known bug that could be exploitedto make
an electronic messagegeneratemultiple copiesof itself as it traveled."l
thought about using that bug to send out a self-reproducingmessagethat
lwould spreadthrough the networkto announcethe conference-and then
lcancel itself," saysLangton. "But I thought better of it. That wasn't the
hssociationI wanted."
In retrospectthat wasprobably just aswell. Two yearslater, in November
1989, a Cornell Universig graduatestudent named Robert Morris tried to
exploit that same bug to write a computer virus-and ended up nearly
crashingthe whole nationwide researchnetworkwhen a programmingerror
allowed the virus to propagateout of control. But even in 1987, says
Langton, computer viruses were one of the few subiectsthat he actually
wanted to discourageat the meeting. In one sense,they were a natural.
Computer virusescould grow, reproduce,respondto their environment,
and in general do almost everything carbon-basedlife-forms did; it was
(and is) a fascinatingphilosophicalquestionwhetheror not they are truly
"alive." But computer viruses were also dangerous."I didn't really want
to stimulatepeople to go out and play with them," saysLangton. "And
frankly, at that time, it was pretty iffy whether, if we said anything about
'No you
virusesin the workshop,the lab might actuallystep in and say,
can't do this.' We didn't want to attracthackersto come to Los Alamos
and try get into the securecomputers."
In any case,saysLangton, he finally iust mailed the invitationto all the
people he knew about who might be interested,and askedthem to spread
the word. "l had no idea how many peoplewould come," he says'"Five
to 500-I was clueless."

In fact, the turnout wasabout 150,includinga handful of slightlybaffled-

looking reportersfrom places'lilil]% e New York Times and Nature mag-
azine. "In the end, we attractediust the right setofpeople," saysLangton'
"We had some on the lunatic fringe and some hard-nosedscoffers,but a
lot of solid peoplein the middle." There were the usualsuspects from Los
Alamos and the SantaFe Institute,of course-people like Holland, Kauff-
man, Packard, and Farmer. But the British biologist Richard &yhlt'
author of The SelfishCene, camefrom Oxford to talk about his "Biomorph"
program for simulated evolution. Aristid Lindenmeyer came from Holland
Life at the Edge of Chaos 239

to talk about his computer simulationsof embryonic developmentand plant

growth. A. K..Dgdryr who had alreadypromotedthe conferencein his
"Computer Recreations"column in ScientificAmeican, came and orga-
nized the computerdemonstrations; he alsoran the "Artificial 4-H" contest
for the best computer creature. Craham aaimq-Slq-lth came from Glas-
gow to talk about his theory of the origin of life on the surfaceof micro-
scopic clay crystals.Hans Moravec came from CarnegieMellon to talk
about robots, and his convictionlii'at they would one day be the heirs of
humanity.
The list went on and on. Langton had no idea what most of the speakers
would say until they got up to say it. "The meeting was a very emotional
experiencefor me," he says."You'll neverre-createthat feeling.Everybody
had been doing artificial life on his own, on the side, often at home. And
'There
everybodyhad had this feeling of, must be somethinghere.' But
they didn't know who to turn to. It wasa whole collection of people who'd
had the sameuncertainties,the samedoubts, who'd wonderedif they were
crazy.And at the meetingwe almostembracedeachother. There wasthis
'l
real camaraderie,this senseof may be crazy-but so are all these other
people.' "
There were no breakthroughsin any of the presentations,he says.But
you could see the potential in almost all of them. The talla ranged from
the collectivebehaviorof a simulatedant colony,to the evolutionof digital
ecosystemsmade out of assembly-language computer code, to the power
of sticky protein moleculesto assemblethemselvesinto a virus. "It was
fascinatingto seehow far people had gotten on their own," saysl,angton.
And more than that, it was fascinatingto see how the same theme kept
cropping up again and again:in virtually everycase,the essenceof fluid,
natural, "lifelike" behavior seemedto lie in such principles as bottom-up
rules, no centralcontroller,and emergentphenomena.Already,you could
sensethe outlines of a new sciencetaking shape."That's why we told people
not to turn their papersin until after the conference," he says."Because
it was only after having listenedto all theseother ideasthat peoplecould
seemore clearly what they had been thinking.
"lt was hard to say exactly what was happening at the workshop," he
says."But 90 percentof it wasto givepeoplethe confidenceto keeppushing.
By the time we all went away,it was as if we had each risen aboveall the
things that were blocking us. Before,everythinghad been 'Stop,"Wait,'
'No,'
and like my not being able to do a thesison artificial life at the
Universityof Michigan. But now, everythingwas"Go, Go, Go, Yes,Yes"!
"l was so hyped up, it was like an alteredstateof consciousness," he
 COMPLEXITY
says."l havethisimageof a seaof graymatter,with ideasswimmingaround,
ideasrecombining, ideasleapingfrom mind to mind."
For that spaceof five days,he says,"it waslike beingincrediblyalive."

'
Sometime after the meetingwasover,Lan$on receivedan elechonic
mail message from Eiiti Wada,who had cometo the meetingfrom the
; Universi$of Tokyo. "The workshopwasso intensive,"wroteWada,"that
I I had no time to confessto you that I wasin Hiroshimawhen the first
i atomicbomb wasdroppedthere."
\ H. wishedto thank Langtonagainfor that mostexcitingweekin Los
\ Alamos,discussing the technology
of life.
 tia;$ta Gsst*c
Is*o*89S$**S6**S**$**S6tf

About five o'clockthat sameTuesday, September 22.,1987,)ohn Holland

and Brian Arthur left the artificial life workshop at Los Alamosto drive
backdown the mesatowardSantaFe. And asidefrom stop
an occasional
to savorthe late afternoonvista to the east, where the Sangrede Cristo
Mountainsrosenearly7000feetabovethe valleyof the Rio Grande,they
spentthe wholehourjong drive talkingabout"boids":a simulationpre-
sinted at the workshopby Craig Reynoldsof the SymbolicsCorporation
in los Angeles.
Arthur wasfascinated by the thing. Reynoldshad billed the programas
an attemptto capturethe essence of flockingbehaviorin birds,or herding
behaviorin sheep,or schoolingbehaviorin fish.And asfar asArthur could
tell, he had succeeded beautifully'Reynolds'basicidea was to place a
largecollectionof autonomous, birdlikeagents-"boids"-into an on-
screenenvironmentfull of wallsand obstacles. Eachboid followedthree
simplerulesof behavior:

l. It tried to maintaina minimum distancefrom otherobiectsin the

environment, includingotherboids.
2. It tried to matchvelocitieswith boidsin ib neighborhood.
3. It tried to move towardthe perceivedcenterof massof boids in its
neighborhood.

What wasshikingabouttheseruleswasthat noneof them said,"Form

a flock." Quite the opposite:the ruleswereentirelylocal, referringonly
 242 COMPLEXITY
to what an individual boid could seeand do in its own vicinity. If a flock
was going to form at all, it would have to do so from the bottom up,
as an emergent phenomenon. And yet flocks did form, every time,
Reynoldscould start his simulation with boids scatteredaround the com-
puter screen completely at random, and they would spontaneouslycol-
lect themselves into a flock that could fly around obstaclesin a very
fluid and natural manner. Sometimes the fock would even break into
subfocks that fowed around both sidesof an obstacle,reioining on the
other side as if the boids had planned it all along. [n one of the runs, in
fact, a boid accidentallyhit a pole, fluttered around for a moment as
though stunned and lost-then darted forward to reioin the flock as it
moved on.
Reynolds had insisted that this last bit of businesswas proof that the
behaviorof the boids was huly emergent.There was nothing in the rules
of behavioror in any of the other computercode that told that particular
boid to act that way. So Arthur and Holland had startedto chew the question
over almost as boon as they got into the car: How much of the boids'
behavior was built in, and how much of it was truly emergent behavior
that was unexpected?
Holland remained to be convinced; he had seenall too many examples
of "emergent" behavior that was tacitly built into the ptogmm from the
start. "I wassayingto Brian that you have to be cautious.Maybe everything
that'shappeningin there, including the one bumping into the pole, is so
obviousfrom the rules that you aren't learninganythingnew. At least,I'd
want to have the ability to put other sorts of obiecb in, change the envi-
ronment, and seeif it still behavedin a reasonable way."
Arthur couldn't very well argue with that. "But for myself," he says,"l
couldn't see how you could define'truly'emergentbehavior."In some
,sense, everything that happens in the universe, ineluding life itself, is
I aheadybuilt into the rules that govern the behavior of quarks. So what is
you seeit? "That
[ "*.rg.n"e, anyway?And how do you recognizeit when
\ goesto the heart of the problem in artificial life," he says.
Since neither Holland nor anyoneelse had an answerto that, he and
Arthur never did reach a firm conclusion.But in rehospect,saysArthur,
that discussiondid plant a seedin his sleepdeprivedmind. In earlyOctober
1987, exhaustedbut happy, he endedhis stint as a visiting scholarat the
SantaFe Instituteand returnedto Stanford.And there,oncehe had caught
up on sleep, he continued to mull over what he had learned in Santa Fe.
"l had been enormouslyimpressedby Holland'sgeneticalgorithms,and
classifiersystems,and the boids, and so on. I thought a good deal about
PeasantsUnder Class 243
them, and the possibilities theyopenedup. My instinctwasthat this was
an answer.The problemwas,What wasthe questionin economics?
"Now, my earlierinterests hadbeenhow economies changeanddevelop
in the Third World," he says."So aboutNovember1987I phonedfohn
and saidthat I'd had a vision of how theseideasmight be appliedto an
economy.I had this notion that you could havewithin your officein the
universitya little peasanteconomydevelopingundera bubbleof glass.Of
course,it would reallybe in a computer.But it wouldhaveto be all these
little agents,preprogrammed io getsmartand interactwith eachother.
"Then in this dreamlikeidea,you'dgo in one morningand say,'Hey,
lookat theseguys!Tho or threeweeksagoall theyweredoingwasbartering,
andnow they'vegot joint stockcompanies.' Then the nextdayyou'dcome
in andsay,'Oh-they've discovered centralbanking.'Then a fewdayslater
you'dhaveall yourcolleagues aroundandyou'repeeringin:'Wow!
clustered
They'vegot laborunions!What'll theythink of next?'Orhalf of themhave
goneCommunist.
"At the time I still couldn'tarticulate
it verywell," saysArthur.But he
knewthatsuchan economy-under-glass wouldbeprofoundlydifferentfrom
conventional economicsimulations,in whichthe computeriustintegrated
a bunchof differential equations.His economicagents wouldn'tbe math-
ematicalvariables,but cgents-entitiescaughtup in a webof interaction
andhappenstance. Theywouldmakemistakes andlearn.Theywouldhave
histories.They wouldno morebegoverned by mathematical formulasthan
humanbeingsare.As a practicalmatter,of course,theywouldhaveto be
far simplerthan realhumanbeings.But if Reynolds couldproducestar-
tlingly realisticflockingbehaviorwith iust threesimplerules,then it was
at leastconceivable that a computerfull of well-designed adaptiveagents
might producestartlinglyrealisticeconomicbehavior.
"I thoughtvaguelywe could cookup theseagentsvia fohn'sclassifier
system,"saysArthur."I couldn'tseehowto do it. fohn hadno immediate
suggestion of how to do it, either.But he wasalsoenthused."The two
menagreedthatwhentheSantaFeInstitute'seconomics programgotunder
waythe next year,this would haveto be a top priority.

The FledglingDirector

In the meantime,Arthur had plentyto keephim busywith the orga-

nizationof that program.Indeed,he wasiustbeginningto graspthe full
implicationsof whathe had committedhimselfto.
24 COMPLEXITY
Holland,it soondeveloped, wouldnot be ableto sharethe directorship.
He had alreadyusedup his sabbatical time the yearbefore,whenhe had
spentthe 1986-87academicyearasa visitingscholarat LosAlamos.Back
at Michigan he wasstill embroiledin the academicpoliticsof his depart-
menfs mergerinto the engineeringschool.And his wife, Maurita, was
tied to her iob as headof the sciencelibrary system.At most, Holland
wouldbe ableto cometo SantaFe for a month or so at a time.
Sothe iob fell entirelyto Arthur, who hadneverrun a research program
in his life, muchlesscreated one.
What doesfohn Reedwantusto do here?he askedEugeniaSinger,who
wasto be hisliaisonwith theCiticorpchairman."He saysdo anythingyou
want," sherepliedafter checkingwith Reed,"so long asit's not conven-
tional."
Whatdo youwantusto do here?heaskedKenArrowandPhil Anderson.
They said they wantedhim to createa completelyrigorousnew way of
doingeconomics basedon the complexadaptive systemspointof view-
whativerthat was.
What doesthe institutewantusto do here?he askedGeorgeCowanand
the otherpowersat SantaFe. "The scienceboardis hopingthat you'll set
radicallynew directionsin economics,"thsy told him. And by the way,
yourbudgetfor the firstyearwill be $560,000-pqrtlyfrom Citicorp,partly
from the MacArthur Foundation,and partlyfrom the NationalScience
Foundation andtheDepartment of Energy,whichwe'resurewill begiving
us a pair of hefty grants.Well, prettysure,anyway.Of course,this will
alsobe the 6rstand the biggestmaiorresearch programat the institute,so
we'll all be watchingcloselyto seehow well you do,
"I left shakingmy head,"saysArthur. "Half a million dollarsis about
mid-sizedon an academic scale.But this wasan enormous challenge. It
'Here'san ice ax and rope-go climb Mount Everest.'
waslike beingtold,
I wasamazed.I wasawestruck. I found it overwhelming"'
In practice,of course,Arthur was far from alone' Both Arrow and
Andersonweremorethan willing to givemoral support,advice,and en-
couragement. "Th"y werevery much the bedrock,the gurusof the pro-
gtam,"saysArthur. Indeed,he considered it theirProgram.But theyalso
madeit very clearthat Arthur wasto be the chiefexecutiveofficer."They
kepthandsoff," saysArthur. "lt wasup to me to directthe thing and get
it going."
. He madetwo key decisionsearlyon, he says'The first had to do with
I topics.He wasdistinctlyunenthused by the ideaof applyingchaostheory
I and nonlinear dynamics to economics, whichseemed to be a big partof
Peasants
UnderGlass 245
what Arrow had in mind. There were plenty of other groupsdoing that
kind of thing already-and with very few worthwhile results, so far as he
could tell. Nor was Arthur interestedin having the programbuild some
huge economic simulation of the whole world. "This may have been in
fohn Reed'smind," he says,"and it seemsto be the first thing engineers
'You're
or physicistswant to do. But it's as if I said to you, an astrophysicist,
why don't you build a model of the universe?'" Such a model would be
iust about as hard to understandas the real universe,he says,which is why
astrophysicistsdon't do it that way. Instead, they have one set of models
for quasars,another set for spiral galaxies,another set for star formation,
and so on. They go in with a computational scalpel to dissect specific
phenomena.
And that's exactly what Arthur wanted to do in the Santa Fe program.
He certainly didn't intend to back away from his economy-under-glass
vision. But he also wanted people to learn how to walk before they hied
to run. In particular, he says,he wanted to seethe program take some of
the classicalproblems in economics, the hoary old chestnutsof the field,
and seehow they changedwhen you looked at them in terms of adaptation,
evolution, learning, multiple equilibria, emergence,and complexity-all
the Santa Fe themes. Why, for example, are there speculativebubbles and
crashesin the stock market? Or why is there money? (That is, how does
one particulargood such as gold or wampum becomewidely acceptedas
a medium of exchange?)
That emphasison the old chestnutsgot the program into hot water later,
saysArthur, when a number of people on the institute's science board
accusedthem of being insufficiently innovative. "But we thought it was
just good science,good politics, and good procedureto approachthe stan-
dard problems," he says."These are problems that economistsrecognize.
Above all, if we could prove that changing the theoretical assumptionsto
be more realistic made major differencesto the insights you got, maybe
gettinga feelingof more realismin thoseinsights,then we could show the
field that we had really contributed something."
Formuchthesamereason, he says,he resistedwhenMurrayGell-Mann1
urged him to come out with a manifestofor the economicsprogram,i
something to nailto theChurchDoor."severaltimeshe pushedtheidea,"j
saysArthur. "He wantedsomethingthat said:'The dayhasdawnedfor ai
differentform of economics,'et cetera,But I thoughtaboutit anddecided!
no. It would be far betterto tackleit problemby problem-these oldi
chestnuts in economics-andwe'll be convincing."
The secondkeydecisionhad to do with the kind of researchers Arthur
246 COMPLEXITY
recruited for the program. He neededpeople who were open-minded and
sympatheticto the Santa Fe themes, of course,The ten-dayeconomics
workshop had proven to him iust how fruidul and exciting such a group
could be. "l realizedearly on that neither I nor Arrow, nor Anderson,nor
anyoneelsewasgoing to lay down the frameworkfor the SantaFe approach
from on top," he says."lt would haveto emergefrom what we did, from
the way we tackled problems, with everybodyhaving his own ideas."
paperpub-
After his own fiascotrying to get that first increasing-returns
lished, however,Arthur alsoknew that it wascritically important to establish
the program'scredibilityamongmainstreameconomists.So he wantedthe
participantsto include economictheoristswith impeccable,diamond-hard
reputations-the likes of Arrow himself or Stanford'sTom Sargent.Not
only would having them there help ensurethat this still+o-be-definedSanta
Fe approach met every existing standardof rigor, but if tftey went away
talking SantaFe-ism,then peoplewere going to listen.
Unfortunately, assemblingsuch a team turned out to be easiersaid than
done. After consultingwith Arrow, Anderson,Pines,and Holland to draw
a list of candidates,Arthur wasable to get pretty much everyonehe wanted
on the noneconomicsside.Phil Andersonagreedto comefor a shortwhile,
as did his former student, Richard Palmer of Duke University. Holland
would come, of course. So would David Lane, a sharp-mindedand ar-
gumentative probability theorist from the Universi$ of Minnesota. Arthur
evengot commitmentsfrom his Sovietcoauthors,probabili$ theoristsYuri
Ermoliev and Yuri Kaniovski.And then therewereStuartKauffrnan, Doyne
Farmer,and all the rest of the crew from Los Alamos and SantaFe. But
when Arthur startedcalling the economists,he quickly discoveredthat his
concernsabout credibilitywere not misplaced.Almost everyonehad heard
rumors that somethinghad happenedin SantaFe. Arrow was talking it up
everywherehe went. But who or what wasthe SantaFe Institute?"So when
'Well, yes,
I called," saysArthur, "the economistswere inclined to say, it's
a bit late, and I've made plans.'l approachedseveralwho said that they'd
wait and see how things worked out. Basically,it was very, very difficult
to get economistswho weren't at the meetingto be interested."
The good news was that the economistswho had been at the economics
meeting were a superbgroup-they'd been selectedby Arrow, after all.
Moreover, the responsefrom outside that group was not entirely bleak.
Arrow agreedto come for severalmonths, as did Sargent.|ohn Rust and
William Brock agreedto come down from the Universityof Wisconsin'
Ramon Marimon would likewisecome down from Minnesota.fohn Miller
would come from the Universig of Michigan, wherehe had iust completed
PeasantsUnder Class 247

a Ph.D. dissertation that madeheavyuseof Holland'sclassifiersystems.

And in whatArthur regarded coup,FrankHahn
asa particularlysatisfying
agreedto come over from Cambridge,wherehe reignedas the leading
economictheoristin England.
So all in all, saysArthur, abouttwentypeopleparticipated to a greater
or lesserdegreein the economicsprogramthat 6rst year,with no more
thansevenor eightin residence at anyonetime.That wasaboutthe size
of an economics department at a smallcollege.And togethertheywere
supposed to reinrtnt the field.

The Santafu APProach.

The economicsprogramwasdue to startat the institutein September

1988,with a second week-long economics workshop to kickit off. SoArthur
tookup full-timeresidence in
there )une, to givehimself thewholesummer
to getieady.He needed every minute of it. And he found thatthingsonly
goi*or. intensewhenthe participants actually started arriving in the fall.
"Peoplewouldcomein to me daily,"he says, "like one guy who didn't
know how to changelight bulbs.Could I do that? And the place was so
small that occasionallyI had to solveproblemslike, What office should
you put someonein who smokes? Or shouldone personsharean of6ce
with someone elsewho had hairylegsandworeshortsall the time?This
personreallyobiectedto that. lt alsofell to me to organizethe workshops.
And partof it wastravelingto recruitpeople,to talkto them,to getadvice,
and to spreadthe word."
When you'rethe boss,Arthur wasdiscovering, you can'talwaysgo out \
and play with the other kids. You haveto spendentirelytoo much of your J
time being the grown-up.Even with the yeomanhelp of the institute'sI
permanentstaff,Arthur foundthat about80 percentof his time wasbeing I
takenup by the nonscientificsideof his iob, and it wasn'ta lot of fun. At I
one point, he says,he camehometo their rentedhousein SantaFe and
startedcomplainingto his wife, Susan,abouthow little time he had for
research. "She finallysaid,'Oh, knockit off. You'veneverbeenhappier
in your life,' And shewasright."
Shewasright. Because for all the administrative nitty-gritty,saysArthur,
that other20 percentmadeup for everything.By thefall of 1988the Santa
FeInstitutewaspulsingwith energy-and not iustbecause of theeconomics
program.Latethe previousautumn,the long-promised federalgrantshad
indeedcomethroughfrom the NationalScienceFoundationand the De-
248 COMPLEXIW
partmentof Energy.Cowanhadn't beenableto talk thoseagenciesinto
nearlyasmuch moneyashe'd wanted-therewasstill no funding to hire
a permanentfaculty,for example-but he hadgottenthemto commit$l'7
million overthe nextthreeyears,startingin fanuary1988'Sothe institute
now had financialsecurityuntil the beginningof 1991.And there was
finally enoughcashfor peopleto getseriousaboutdoingwhatthe institute
had beencreatedto do.
The science board,underGell-MannandPines,hadaccordingly given
thego-ahead for fifteennewworlshops.Someworlahops promised to'come
atthe issueof complexityfroma hard-core physics perspective, with a prime
examplebeing Physicsof Information,Enhopy,and Complexi$, to be
organizedby a youngPolishphysicistfrom Los AlamosnamedWoiciech
Zurek. Zurek's idea was to startfrom the conceptsof informationand
computationalcomplexity,as definedin computerscience,and explore
their deepconnectionswith quantummechanics,thermodynamics, the
quantumradianceof blackholes,and the (hypothetical) quantumorigins
of the universe.
Other workshopspromisedto approachcomplexi$ from the biological
side,with two prime examples beinga pair of worlshopson the immune
systemto be organizedby l-os AlamosbiologistAlan Perelson'Indeed,
Perelsonhad alreadyrun a maior SantaFe workshopon immunologyin
June1987,and wasnow leadinga smallongoingresearch programthere.
The ideawasthat the body'simmunesystem,which consisbof billionsof
highly responsive cellsthat floodthe bloodstream with antibodiesto neu-
tralizean invadingvirus or bacteriumassoonas it appears,is a complex
adaptivesystemin preciselythe samesensethatecosystems andbrainsare.
So the ideasand techniquesfoating aroundthe instituteought to help
illuminateimmune-related problems suchasAIDS, andautoimmune dis-
eases such as multiple sclerosisor arthritis.And convenely, because so
much wasknown about the molecular details of the immune system, a
researchprogramdevoted to that system ought to help keep some of the
morehigh-flyingideasaroundSantaFe pinneddownto reality.
Meanwhile,the scienceboardalsostronglyendorsed the ideaof bringing
in visitorsandpostdocs not associated with any particular programor work-
shop.lt wasa continuationof the approach the institutehad followedfrom
the beginning:iust getsomevery good, very smart people in here,and see
what happens.The ioke amongthe science board members wasthat the
SantaFe Institutewasan emergent phenomenon all by itself. It wasa ioke
theyactuallytook quite seriously.
PeasantsUnder Glass Z4g
All this wasfine by Cowan.He wasalwayseagerto find more people
with that indefinablefire in their soul. It wasn'ta matterof seekingout
talentper se,he says.You could talk aboutthe institutewith an an{ul lot
of excellentpeoplewho just didn't understand whatyou weredriving at.
Insteadyouhadto lookfor a kind of resonance: "Eithersomeone getsglassy-
eyedor the communicationbegins,"he says."And if it does,then you're
exercising a form of powerthat'sextremelycompelling:intellectualpower.
If you can geta personwho understands the conceptsomewhere down in
the bowelsof the brain, wherethat sameidea'sbeensittingforever,then
you havea graspon that person.Youdon't do it by physicalcoercion,but
by a kind of intellectualappealthatamounbto a coercion.Yougrabthem
by the brainsinsteadof by the balls."
It wasno easierto 6nd suchpeoplethan it hadeverbeen.But theywere
out there.And in ever-increasing numbers,theywerebeginningto come
throughSantaFe-to the point wherethe tiny conventwasoften filled to
overflowing.Indeed,with workshops and seminarsconstantlyunder way
in the chapel,with as manyasthreeor four peoplecrowdedinto offices
thatwouldhavebeencrampedfor one,with officematesendlessly drawing
on the blackboards and arguing,with bull sessions formingand reforming
in the hallwaysor out on the patio under the trees,it wasoften almost
impossible to think. And yetthe energyandthecamaraderie wereelechic.
As StuartKauffmansays,"I waslearninga wholenew wayof lookingat
the world abouttwicea day."
Theyall were."On a $pical daythere,"saysArthur, "mostpeoplewould
disappearinto their officesin the morningand you'd heara lot of click-
clackingof computerterminalsand keyboards. But then someonewould
peeraroundyour door. Haveyou done this?Haveyou thoughtof that?
Haveyou got half an hour to talk to somevisitor?Then we'dgo to lunch,
andusuallyto theCanyonCafe,whichwecalled
'Faculty
the Club.' We became so well known there that the waitresses
hardly even broughtus menusanymore.We'd alwayssay,'Give me a
numberfive,'sotheydidn't evenhaveto ask."
The talk wasendlessand, for the mostpart, fascinating.Indeed,says
Arthur, what he remembersmore vividly than anythingelsewere the
lqplomptu seminar/bullsessions thatwereforeverspringingup in the late
moihingi6i-affimoonil'\Ifi: did that three,four, fiue ti*er a week,"he
says."Someonemight go downthe corridorand say,'Hey, let'stalk about
X.' So abouthalf a dozenof us wouldmeet,eitherin the chapelor more
oftenin the smallmeetingroomby the kitchen.It wasbadlylit, but it was
250 COMPLEXITY
next to the coffeeand the Coke machine.It had Navahodecorationsdom-
inating the room. And there was a photograph of Einstein in an Indian
headdress, beaming down on us.
"So we would iust sit around the table," he says,"maybe with Stuart
leaning on the mantelpiece.Someonemight put a problem on the board,
and we'd startkicking around an awful lot of questions.Thesewere actually
very good arguments.They were neveracrimonious.But they were quite
hard-edged, becausethe issuesthat kept coming up were fundamental.
They weren't the sort of technical problemsthat come up in academic
economics, like how do you solve this or that fixed-point theorem, or in
physics,like why doesthis materialgo superconducting at minus 253 de-
grees, or whatever. They were questionsof where the science should go
next. How do you deal with boundedrationali$?How is economicssup-
posedto proceed when the ptoblems start to get truly complicated, as in
chess?How would you think about an economythat is alwaysevolving,
that never settlesdown to an equilibrium? If you do computer experimen-
tation in economics,how would that work?
"I think that's where SantaFe came into its own," he says."Becausethe
answerswe were startingto give and the techniqueswe were borrowing
were, in my opinion, what began to define a Santa Fe approach to eco-
nomics."

There was one seriesof discussionsthat Arthur remembersin particular,

becauseit did so much to crystallizehis own thinking. Arrow and Cam-
bridge'sFrank Hahn were both there, he says,so it must havebeensometime
'We
during their visit in Octoberor November1988. would meet-myself
and Holland, Arrow and Hahn, maybe Stuart and one or two others-and
we would thrash out what economists could do about hgd.d o!io!3|!.bC'
That is, what would really'hlpfl6ili6-economic theory if they quit assuming
that people could instantaneously compute their way to the solution of any
economic problem-even if that problem were iust as hard as chess?
They had almostdaily meetingson the questionin the small conference
room. Arthur remembersHahn pointing out once that the reasonecon-
omistsuse perfectrationality is that it's a lenchqrk. If peopleare perfectly
rational, then theoristscan sayexactlyhow they ilill react. But what would
perfect inationalig be like? Hahn wondered'
"Brian!" he said. "You're Itish' You might know."
Seriously, Hahn continued while Arthur tried to laugh, there is only
 Peasants Under Class ZS1
one way to be perfectly rational, while there are an infinity of ways to be
partially rational. So which way is correctfor human beings?"Where," he
asked,"do you set the dial of rationality?"
Where do you set the dial of rationality? "That was Hahn's metaphor,"
saysArthur, "and it really stuck.in my mind. I thought about it for a long
time afterward. I chewed a lot of pencils. 'fhere was a lot of discussion."
And slowly, like watching a photographic image emerging in a developer
ha5 he and the others began to see an ansrwer:the way to set the dial of
rationali$ was to leave it alone. Let the agenb set it by themselves.
"You'd pull a fohn Holland on it," saysArthur. "You'd iust model all
these agentsas classifiersystems,or as neural nets, or as some other form
of adaptive learning system,and then allow the dial to vary as the agents
learn from experience.So all the agentscould start off as perfectly stupid.
That is, they would just make random, blundering decisions.But they
would get smarterand smarteras they reactedto one another." Maybe they
would get very smart indeed, and maybe they wouldn't; it all depended
upon what they experienced.But either way, Arthur realized, these adap-
tive, artificially intelligent agentswere exactly what you wanted for a,real
theory of dynamics in the economy. If you put them down in a stable,
predictableeconomic situation, you might very well find them making
exactly the kind of highly rational decisionsthat neoclassicaltheory pre-
dicts-not becausethey had perfectinformation and infinitely fastreasoning
ability, but becausethe stability would give them time to learn the ropes.
However,if you put thosesameagentsin the midst of simulatedeconomic
change and upheaval, they would still be able to function. Not very wetl,
perhaps:they would stumble and fail and make any number of false starts,
iust as humans do. Nonetheles, under the influence of their built-in
Iearning algorithms they would slowly grope their way toward some rea-
sonablenew courseof action. And by the sametoken, if you put the agents
into a competitive situation analogousto chess,where they had to choose
moves against each other, you could watch them make their choices. If
you put the agentsinto a simulatedeconomyundergoinga simulatedboom,
you would watch them explore their immense spaceof possibilities. No
matter where you put them, in fact, the agentswould try to do something.
So unlike the neoclassicaltheory, which has almost nothing at all to say
about dynamics and changein the economy,a model full of adaptiveagents
would come with dynamicsalreadybuilt in.
This was obviouslythe same intuition as in his economyunder glass,
Arthur realized.Indeed,it wasessentiallythe samevision he'd had almost
252 COMPLEXITY
a decade earlier, after reading The Eighth Day of Creation, Except that
now he could see that vision with crystal clarity. Here was this elusive
"SantaFe approach":Insteadofemphasizingdecreasing returns,staticequi-
librium, and perfectrationality,as in the neoclassical view, the SantaFe
team would emphasize increasing returns, bounded rationali$, and the
dynamics of evolution and learning. Instead of basing their theory on
assumptionsthat were mathematically convenient, they would try to make
models that were psychologicallyrealistic. Insteadof viewing the economy
as some kind of Newtonian machine, they would see it as something
organic,adaptive,surprising,and alive. Insteadoftalking about the world
asif it were a'staticthing buried deepin the frozen regime, asChris Langton
might have put it, they would learn how to think about the world as a
dynamic, ever-changingsystempoised at the edge of chaos.
"Of course,that wasnot a totally new point of view in economics,"says
Arthur. The greateconomistJosephSchumpetermay not haveknown words
like "the edgeofchaos," but he had pushedfor an evolutionaryapproach
to economicsas far back as the 1930s.RichardNelsonand SidneyWinter
of Yale Universityhad been trying to foment an evolutionarymovement
in economicssincethe mid-1970s,with somesuccess. And other research-
ers had even made some early attemptsto model the effectsof learning in
economics."But in theseearlierlearning models," saysArthur, "you as-
sumed that the agentshad alreadyformed a more-or-lesscorrect model of
the situationthey werein, and that learningwasiust a matterof sharpening
up the model a bit by adiustinga few knobs.What we wantedwassomething
much more realistic.We wantedthese'internalmodels'to emuge-tofotm
inside the agents'minds, so to speak-as they learned. And we had a slew
of methods we could use to analyze that process.There were Holland's
classifiersystemsand geneticalgorithms.RichardPalmerwasiust finishing
a book on neural nehvorks.David lane and I knew how to mathematically
analyze systemsihat learned on the basis of probability. Ermoliev and
Kaniovski were experts on stochastic learning. And we had the whole
literature in psychology.Theseapproachesgaveus a really fine-grainedway
to model adaptation, to make it algorithmically precise.
"In fact," Arthur adds, "the key intellectual influence that whole first
year was machine learning in generaland fohn Holland in particular-
not condensed-matterphysics, not increasing returns, not computer sci-
ence, but learning and adaptation.And as we startedto kick it around with
Arrow and Hahn and the others, it was clear that what was exciting to all
of us was an instinct that economicscould be done in this very different
way."
 Peasants
UnderGlass 253

If the Santa Fe economistsfound the prospectexciting, however, they

alsofound it vaguelydisturbing.And the reason,saysArthur, wassomething
that he didn't put his finger on until much later. "Economics, as it is
usually practiced, operatesin a putely deductive mode," he says."Every
economic situation is 6rst translatedinto a mathematical exercise.which
the economicagentsaresupposed to solveby rigorous,analyticalreasoning.
But then herewereHolland, the neuralnet people,and the other machine-
learning theorists. And they were all talking about agentsthat operate in
an rtSlWIF_ mode, in which they hy to reasonfrom fragmentarydata to
a useful internal model." Induction is what allowsus to infer the existence
of a cat from the glimpseof a tail vanishingaround a corner. Induction is
what allows us to walk through the zoo and classifysome exotic feathered
creatureas a bird, even though we've never seena scarlet-crested cockatoo
before. Induction is what allows us to survive in a messy,unpredictable,
and often incomprehensibleworld.
"lt's as if you'veparachutedinto somenegotiatingsessionin
fapan," says
Arthur. "You've never been in Japan before, you don't know how the
fapanesethink, or act, or work. You can't quite understandwhat is going
on. So most of the things you do are completely out of cultural context.
And yet over time, you notice that someof the things you do are successful.
So slowly, you and your companysomehowlearn to adaptand behave.,'
(Of course,it's anotherstorywhetherthe Japanese actuallybuy your prod-
ucts.) Think of the situationas a competitionlike chess,he says.players
have fragmentary information about their opponents'intentions and abil-
ities. And to fill in the gaps,theydo indeed uselogical, deductivereasoning.
But they can only use it to look a few movesahead, at most. Much more
often the playersoperate in a world of induction. They try to fill in the
gaps on the fly by forming hypotheses,by making analogies,by drawing
from past experience,by using heuristic rules of thumb. Whatever works,
works-even if they don't understandwhy. And for that very reason, in-
duction cannot dependupon precise,deductivelogic.
At the time, Arthur admits, even he found this troubling. ,.Until I went
to SantaFe," he says,"I thoughtthat an economicproblem had to be well
defined before you could even talk about it. And if it wasn't well defined.
what the hell could you do with it? You certainly couldn't apply logic.
"But then |ohn Holland taught us that this isn't so. When we talked to
fohn and read his paperswe beganto realizethat he wastalking about cases
where the problem context isn't well defined, and the environmentisn't
 254 COMPLEXITY
'fohn, how can you eventecm in
stationaryovertime. We saidto him,
thatenvironment?' "
Holland'sanswerwasessentially that you learn in that environment
becauseyou haveto: "Evolutiondoesn'tcarewhetherproblemsare well
definedor not." Adaptiveagentsareiustresponding to a reward,he pointed
out. Theydon'thaveto makeassumptions aboutwheretherewardiscoming
from. In fact, that wasthe wholepolnt of his classifier systems. Algorithm-
ically speaking,thesesystems weredefinedwith all the rigoryou couldask
for. And yettheycouldoperatein an environmentthatwasnot well defined
at all. Sincethe classifierruleswereonly hypotheses aboutthe world, not
,,facts,"theycouldbemutuallycorrhadictory. Moreover,because thesystem
wasalways testing those hypotheses to 6nd out which ones wete useful and
led to rewards, it could continue to learn even in the face of crummy,
incompleteinformation-andevenwhile the environment waschanging
in unexpected ways.
"But its behaviorisn't optimal!"the economists complained,having
convincedthemselves that a rationalagent is onewho optimizes his "utility
function."
"Optimal relativeto what?"Hollandreplied.Thlkaboutyour ill-defined
criterion,in anyrealenvironment,the spaceof posibilitiesis sohugethat
thereis no wayan agentcanfind the optimurn-or evenrecognizeit. And
that'sbeforeyou takeinto accountthe factthat the environmentmight be
.changingin unforeseen ways.
"This wholeinductionbusiness fascinated me," saysArthur' "Hereyou
couldthink aboutdoingeconomics wheretheproblemfacingtheeconomic
agentwasnot evenwell defined,wheretheenvironmentisnot well defined,
wherethe environmentmightbe changing,wherethe changes weretotally
rt unknown.And, of co[rse, you iust had to think for about tenth of a a
Peopleroutinelymakede-
J" ,e"o.rdto realize,that'swhat life is all about.
cisionsin contextsthat arenot well de6ned,withoutevenrealizingit. You
muddlethrough,you adaptideas,you copy'you try whatworkedin the
past,you try out things.And, in fact, economisbhad talkedaboutthis
i.ina of behaviorbefore,But we werefindingwaysto makeit analytically
precise,to build it into the heartof the theory."
Arthur remembers one keydebateduringthat sameperiodthat wentto
the heartof the difficulty."It wasa longdiscussion in October,November
1988,"he says."Arrow,Hahn, Holland,myself,maybehalf a dozenof
us. We had iust begunto realizethat if you do economics this way-if
there was this Santa Fe approach-then there might be no equilibrium in
Peasants
UnderGlass 2Ss
the economy at all. The economy would be like the biosphere:always
evrilving, alwayschanging, alwaysexploring new territory.
"Now, what worried us wasthat it didn't seempossibleto do economics
in that case," saysArthur. "Becauseeconomics had come to mean the
investigation of equilibria. We'd gotten used to looking at problems as
though they were butterflies,nailing them down to cardboardby holding
them in equilibrium while we examinedthem, insteadof letting them go
pastyou and fly around. So FrankieHahn said,'[f thingsarenot repeating,
if things are not in equilibrium, what can we, as economists,say?How
could you predict anything? How could you have a science?'"
Holland took the questionvery seriously;he'd thought a lot about it.
Look at meteorology, he told them. The weather never settlesdown. It
never repeatsitself exactly. It's essentiallyunpredictablemore than a week
or so in advance.And yet we can comprehend and explain almost every-
thing that we see up there. We can identify important featuressuch as
weatherfronts, jet streams,and high-pressuresystems.We can understand
their dynamics. We can understandhow they interact to produce weather
on a local and regionalscale.In short, we h4vea real scienceof weather-{
without full prediction. And we can do it becauseprediction isn't the I
essenceof science. The essenceis comprehensionand explanation. And I
that's preciselywhat Santa Fe could hope to do with economicsand other
social sciences,he said: they could look for the analog ofweather fronts-
dynamical socialphenomenathey could understandand explain.
"Well, Holland's answer was to me a revelation," saysArthur. "It left
me almost gasping.I had been thinking for almost ten yearsthat much of
the economy would never be in equilibrium. But I couldnt see how to
'do'economics
without equilibrium. John'scommentcut throughthe knot
for me. After that it seemed-straighdorward."
Indeed,saysArthur, it wasonly during thoseconversations in the fall of
1988 that he really began to appreciatehow profoundly this Santa Fe
approach would change economics. "A lot of people, including myself,
had naively assumedthat what we'd getfrom the physicistsand the machine-
learning people like Holland would be new algorithms, new problem-
solving techniques, new technical frameworks.But what we got was quite
different-what we got was very often a new attitude, a new approach, a
whole new world view."
256 COMPLEXITY
The DarwinianPrincipleof Relativity

Holland, meanwhile, was having the time of his life at Santa Fe' He
loved nothing better than to sit down with a bunch of very sharp people
and kick ideasaround. More impor$nt, however,theseconversationswere
leading him to make a maior coursechangein his own research-that is,
the conversationsplus the fact that he hadn't been able to 6gure out how
to say no to Murray Gell-Mann.
"Murray's a greatarm twister," laughs Holland. In the late summer of
1988, he says,Gell-Mann had telephonedhim in Michigan: "fohn," he
said, "you're doing all this stuff on genetic algorithms. Well, we need an
example we can use againstthe creationists."
The fight against"creation science"wasindeedone of Gell-Mann's many
passions.He had gotten involved severalyearsbefore, when the Louisiana
SupremeCourt was hearing argumentsfor and againsta statelaw requiring
. that creation sciencebe taught on an equal footing with Darwin's theory.
lc.ll-Mrnn had persuadedalmost all the U.S. recipienb of what he calls
fl.,the Swedishprizes" in scienceto sign an amicuscuriae brief urging repeal.
And the court did vote 7 to 2 to throw out the law. But in the wake of that
decision, ashe readthe newspapercorrespondence,Gell-Mann had realized
that the problem went well beyond the activities of a few religious fanatics:
'Of
"People wrote in saying, course I'm not a fundamentalist, and I don't
believe all that nonsense about creation science.However,the name brand
of evolution they teach in our schools seemsto have something wrong with
it. Surely it couldn't be by blind chance that it all happened,'et cetera, et
cetera.So these weren't creationists. But they couldn't be convinced, some-
how, that iust chance and selection have produced what we see'"
So what he had in mind, he told Holland, was a seriesof computer
programs, or even computer games, that would slrou people how it could
happen. They would reveal how chance and selection plessule, operating
over a vast number of generations, could produce a huge amount of evo-
Iutionary change. You would fust set up the initial conditions-essentially
a planet-and then let things rip. In fact, said Gell-Mann, he wasthinking
of organizinga workshopat the institute to talk about such games.Wouldn't
Holland like to conhibute something?
Well no, actually, Holland wouldn't. He was certainly sympatheticto
what Gell-Mann was trying to do. But the fact was that he had a very full
plate of researchproiects already*not the least of them being a classifier
systemhe was writing for Arthur to apply to economic models. From his
point of view, Gell-Mann's evolution simulator would be a dishaction.
 Peasants
Under Glass ZS7
Besides,he'd alreadydonethegeneticalgorithm,and he couldn'tseethat
d_oing it all overagainin anotherform wouldteachhim anythingnew.so
Hollandsaidno asfirmly ashe could.
Okaythen,saidGell-Mann,whydon'tyouthinkaboutit. And not long
after,he calledagain:fohn, this is reallyimportant.wourdn't Holland
changehis mind?
Hollandtried to sayno oncemore-although he courdalreadyseethat
he wasgoingto havea toughtime makingthatanswerstick.so in the end,
aftera longconversation, he abandoned all furtherresistance. "All right,"
he told Gell-Mann,"I'll try."
Actually,Hollandadmib, he wasn'tputtingup much of a fight by then
anyway'In betweenthe phonecalls,ashe'd thoughtabouthow to make
Gell-Manntakeno for an answer,he had startedto think moreand more
aboutwhat he would do if he had to sayyes.And he'd realizedthat there
might actuallybe a rich opportunityhere.Evolution,of course,wasa lot
morethan iust randommutationand naturalselection.It wasalsoemer-
genceand self-organization. And that, despitethe besteffortsof Stuart
Kauffman,Chris langton, and a greatmanyotherpeople,wassomething
that no one understoodvery well. Maybethis wasa chanceto do better.
"l startedlookingat it," saysHolland,"and realizedthat I coulddo a model
that would satisfyMurray-or at leasta pieceof it might-and still do
somethinginterestingfrom the research viewpoint."
The model wasactuallya revivalof somethinghe'd done backin the
early1970s,Hollandexplains.At the time he wasstiil workinghardon
geneticalgorithmsand his bookAilaptation.But he wasinvitedto sive a
talk at a conference in the Netherlands. And for the fun of it, he dJcided
to tacklesomethingcompletelydifferent the origin of life.
He calledthe talkand the paperbasedon it "spontaneous Emergence,"
he says.And in rehospect,it took an approach thrt *", quite siriilar in
spirit-to the autocatalyticmodelsbeingindependently pursuedby stuart
Kauffman,IVlanfredEigen,andotto R<issler at aboutihl sametime. "My
paperwasn'ta computermodel,assuch," saysHolland. "It wasa formal
modelin which you could do mathematics. I wastrying to showthat you
could designautocatalyticsystemsin which you could-geta simpleself-
replicatingentity, and that this would occur many ordirs of magnitude
fasterthan the usualcomputations predicted."
Thoseusualcalculations-stilllovinglyquotedby creationists-were6rst
put forwardby quitelegitimatescientists backin the 1950s.The argument
was_that self-replicating
life-formscould not possiblyhaveoriginatJdfro*
randomchemicalreactions in theprebioticsoup,because theti'merequired
 255 COMPLEXITY

wouldhavebeenvastlygreaterthan the ageof the universe.It wouldhave

beenlike waitingfot ihor" fabledmonkeysin the basement of the British
Museum to producethe completeworks of Shakespeare by banging away
on typewriteis:they will getthere,but it will take them a long time'
Hoiland, however,*"*'t morediscouraged by this argumentthan
"ny reactions wereall very well,
Kauffmanand the othershad been.Random
he thought.But whataboutchemicalcatalysis, which is decidedlynon-
randomiSoin hismathematical modelHolland postulated a soupof "mol-
ecules,,-arbitrarysymbolslinked into stringsof various lengths-which
areacteduponby free-floating catalytic"enzymes": oPerators thatdid things
\ to thestrines."Thesewereveiyprimitiveoperators likeCOPY,whichcould
1 attachto ariyshingwhatsoever andmakea copyof thatstring,"saysHolland.
,.1*r, actuallyableto provea theorem.If you hada systemwith someof
i
I th.r. operatorsfoating around,and if you allowedrecombination among
that
I a.bitrarystringsof differentlengths-in effect,buildingblocks-then
entity much more rapidly than
! systemwouldproduce a self-replicating
i tryingto do thingspurelyat random."
That spontaneous emergence paperwaswhat Hollandcalls"a singular
point"; it was like nothing he had donebeforeor since.And yet,the issues
of .-.rg.r,"" and self-organization werestill very much in his mind. |ust
the year-before, in fact, he had,spent a lot of time battingthem around
wittr Doyne Farmer, Chris LangLon, Stuart Kauftnan, and othersduring
his stayat Los Alamos. "so with Murray's arm-twisting, I thoughtmaybe
the time wasripe to do more along those lines," he says. "And maybenow
I'd build a realcomputermodel of these things'"
Having spentall ihoseinterveningyearson classifiersystems,h-esays'
the wayt:o*ake a computermodelseemed obvious.Sincethe free-floating
operatorsin the original paper had had the effectof rules-"IF you en-
countersuchandsucha string, THEN do the followingto it"-the thing
to do wasto writethem thatwayin the program, andmakethe wholething
look as much like a classifiersysternas possible. And yet, as soonashe
startedthinking in thoseterms,Hollandalso realized that he wasgoingto
haveto fr"e uito the maiorphilosophical in
faw classifier systems. In the
spontaneous emergence papei he says,the spontaneity had been real' and
ti. e*.rg"nc. h"i b"err"ompletelyintrinsic.But in classifier systems, for
all their llarning abilityandfor all theirpowerto discover emergent clusters
of rules, there wasstiil a deusex machina; the systemsstill dependedon
'A-classif-ersysterneet_s_l_PrySff
the shaiowy hand of the programmer.
onlv becauseI assignwinning or losing,'aiys Holland'
{ffiffi'.,et6fitif;t md aTways buggedhim. Lcavingasidequestions
 Peasants
UnderClass 2Sg
of religion, he says,the real world seemsto get along just fine without a
cosmic referee. Ecosystems,economies, societies-they all operate ac-
cordingto a kind of Darwinian principleof relativity:everyoneis constantly
adapting to everyoneelse. And becauseof that, there is no way to look at
any one agentand say,"It's fitnessis 1.j75." Whatever..fitness,'means_
and biologistshave been arguingabout that sincethe time of Darwin-it
cannot be a single, fixed nurnber. That's like askingif a gymnast is a better
or worse athlete than a sumo wrestler;the question is meaninglessbecause
there'sno common scaleto measurethem. Any given organism'sability
to survive and reproduce dependson what niche it is filling, what other
organismsare around, what resourcesit can gather,evenwhat its pasthistory
has been.
"That shift in viewpoint is very important," saysHolland. Indeed, ev-
olutionarybiologistsconsiderit soimportantthatthey'vemadeup a special
word for iL organismsin an ecosystem don't just evolve,thev coevolve.
Organismsdon't changeby climbinguphiil-t ffil p""k of ,o*.
abstractfitnesslandscape,the waybiologistsof R. A. Fishei'sgeneration
had it. (The fitness-maximizingorganismsof classicalpopulatio-ngenetics
actually look a lot like the utility-maximizingagentsof neoclassicaleco-
nomics.) Real organismsconstantlycircle and chaseone another in an
infinitely complex danceof coevolution.
on the faceof it, coevolutionsoundslike a recipefor chaos,saysHolland.
At the institute, stuart Kauftnan liked to compare it to climbing around
in a fitnesslandscapemade of rubber, so that the whole thing deforl, euery
time you take a step. And yet somehow, saysHolland, this dance of co_
evolution producesresultsthat aren't chaotic at all. In the natural world it
has produced flowers that evolved to be fertilized by bees, and bees that
evolved to live off the nectar of flowers. It has produced cheetahsthat
evolved to chasedown gazelles,and gazellesthat evolved to escapefrom
cheetahs.It has produced myriad creaturesthat are exquisitelyadaptedto
each other and to the environment they live in. In ihe human world,
moreover, the dance of coevolution has produced equally exquisite webs
of economic and political dependencies-alliances,tiuarries,customer-
supplier relationships,and on and on. It is the dynamic that underlay
Arthur's vision of an economyunder glass,in which artificial economic
agentswould adapt to each other as you watched. It is the dynamic that
underlay Arthur and Kauftnan's analysis of autocatalytic technology
change. It is the dynamic that underlies the affairs of naiions in a worid
that has no central authority.
Indeed, saysHolland, coevolution is a powerful force for emergenceand
 260 COMPLEXITY

self-organizationin any complex adaptivesystem'And thafs why he knew

that, ii he was ever going to understandthesephenomena at the deepest
I.u.i, h" was going tl tr"u" to start by eliminating this businessof,outside
,.*"rd. Ut fotiu"at ln however,he alsoknew that the assumptionof outside
reward was intimately bound up in the classifiersystem'smarketplacemet-
aphor. Holland had set up a systemin which each classifierrule wasa very
tiny, u.ry simple agent that participatedwith the other rules in an internal
."o.ro-y where the currency was "strength," and where the only source
of wealth wasthe payofffrom the final consumer-that is, the programmer.
And there was no way to get around that fact without changingthe classifier
systemframework comPletelY.
So that'swhat Holland did. what he needed,he decided,wasa different
g"{, and more elemental metaphor for interaction: .cs@!^ And what he came
up with was Echo, a highly simplified biological community in which
digital organismsroam the digital environment in searchof the resources
J,ou they.,eed to stayalive and reproduce the digital analogsofthe water,grass,
la* nuts, berries, et cetera. When the creaturesmeet, of course, they also try
goth- to make resourcesout of each other. ("Echo" is short for ecosystem.)
"I

e\"(.- compare it to a game that my daughter Mania has, called Mail order
Monsters," saysHolland. "You have a bunch of possibilitiesof offcnseand
fr3"x defense,and how you put thesetogetherdetermineshow well you do against
other monsters."
More specifically,explainsHolland, Echo represents the environment
as a large flat plain dotted here and there with "fountains," which gush
forth vaiious kinds of resources represented by the symbols c, b, c, and d.
lndividual organisms randomly move about this environment in sheep
mode, placidly grazingon whatever resources they come across and placing
them in an internal reservoir. whenever two organisms encounter one
another, however,they instantly shift from sheep mode to wolf mode, and
attack.
In the battlethat follows,saysHolland,the outcomeis determinedby
eachorganism's pair of "chromosomes," which are iusta setof resource
Asvmbols strunetoeetherinto two sequences such as aabc and bfud' "lf
"then
I uou'r. on. of tf,eoiganisms,"he explains, youmatchyourfirststring,
I which is calledthJ'offense,'withthe otherorganism's secondstring,the
.d.f.nr..'And if theymatch,you geta high score.So it's much like the
\'
irrr-un" system:if youroffensecomplements the otherguy'sdefense,-then
yo,r'u"*ri. breach.He doesthesamethingreciprocally back.His offense
" hereis awfullysimple.
is a matchagainstyour defense.Sothe interaction
Do your offensiveand defensive overwhelm
capabilities his?"
 Peasants Under Class 261
If the answeris yes,he says,then you get a meal:all the resourcesymbols
in youi opponent'sreservoirand in both his chromosomesgo into your
reservoir. Furthermore, saysHolland, if eating your erstwhile opponent
means that you now have enough resourcesymbols in your reservoir to
make a copy of your own chromosomes,then you can reproduceby creating
a whole new organism-perhaps with a mutation or two. If not-well,
back to grazing.
Echo wasn'texactlywhat Gell-Mann had had in mind, to put it mildly.
It had nothing obvious for a user to play with, and nothing at ail in the
way of fancy graphics. Holland simply couldn't be bothered with any of
that. To run the thing he would type in a shing of cryptic numbers and
symbols, and then watch as a cascadeof even more cryptic output came
pouring down the screenin columns of alphanumericgibberish.(By this
point he had graduatedto a Macintosh II computer.) And yet Echo was
Holland's kind of game. In it he had finally eliminated this businessof
explicit, outside reward. "It's the closure of the loop," he says.,,you really
are going clear back to the point of, If I don't gather enough resourcesto
make a copy of myself, I don't survive." He had capturedwhat he regarded
as the essenceof biological competition. And now he could use Echo as
an intellectual playground, a place to explore and understand what co-
evolution could really do. "I had a list of severalphenomenathat occur in
ecological systems,"he says."And I wanted to show that even with this
very simple structure, each of thesephenomenawould show up in one
way or another."
At the head of that list was what the English biologist Richard Dawkins
called the evolutionary arms race. This is where a plant, say,evolvesever
tougher surfacesand ever more noxious chemical repellents to fend off
hungry insects,evenas the insectsare evolvingeverstrongerjawsand ever
more sophisticatedchemical resistancemechanismsto pressthe attack.Also
knownas th-e.Re9 Qu:en hypgthe-lis,in honor of the lrwis carro[ character
who told Alice that she had to run as fastas shecould to stayin the same
place, the evolutionaryarms race seemsto be a maior impetusfor ever-
increasingcomplexity and specializationin the natural world-iust as the
real arms race was an impetus for ever more complex and specialized
weaponryduring the Cold War.
Holland certainly wasn't able to do much with the evolutionaryarms
race in the fall of 1988;at that point Echo was barerymore than a design
on_paper.But within a year or so, it wasall working beautifully. '.If I startld
off with very simple organismsthat used only one letter for their offense
chromosomeand one for their defensechromosome,"saysHoiland, ',then
 262 COMPLEXITY
LrJtwrLEAt t r
i
thatusedmultipleletters'[Theorga-
^-,,trF l{whatI beganto seewereorganisms
t{-J
Y\'^n3 ,-lilUt nisms
- ' nisms coJd
coftd lengthen,,l'.ii,.:l'"i::"T'_
lengthentheir chromosomes throughmutations.lThey co-
:ry**.T*in':r-3:l^:i;
evolved.One wouldadda little moreoffensivecapability;
^.,^t.,-l f)-- .,,^,,11 o,ll o little -.t" ^ffensive canebilifu: the other would
the other would

<Cf f f ,dd defensivecapability.So they got progressively more complex.And

LY:*^sometimes theywouldsplit,so got
I essentially a
so_l.esentialll.coltnew species'
!S i*.ro-etimes :.Y.:P:"itt'
..lt wasat that point,';saysHolland,"with the fact .., ,
-:"lnb'', that evenwith such
\f " I r -- l --^^.:^d^- rL^+ f L-^^- +^ k-
simple apparatusI could get arms racesand speciation,that I began to be
II
much more interested."
t In particular,he says,he wantedto understand a deepparadoxin evo-
I
t lution: the fact that the samerelentless competitionthat givesriseto evo-
I lutionary arms racescan also give rise to symbiosisand other forms of
cooperation. Indeed,it wasno accidentthatcogpeEtio!in itsvariousguises
actuallyunderlayquitea fewitemson Holland'slist, It wasa fundamental
problem in evolutionarybiology-not to mention economics,political
science,andall of humanaffairs.In a competitive world,whydo organisms
cooperate at all?Why do theyleavethemselves oPento "allies"who could
easilyturn on them?
The essence of the problemis neatlycapturedby a scenarioknown as
in the branchof
lbtrisongp*Dileqrm.a, which wasoriginallydeveloped
-math-e]tiii&
cailedgametheory.Two prisoners arebeingheld in separate
rooms,goesthe story,and the policeareinterrogating both of them about
a crimetheycommittediointly. Each prisoner has a choice he can inform
on his partner ("defect")or elseremain silent("cooperate"-with his part-
ner, not the police). Now, the prisoners know that if both of them remain
silent,then both of them will be released; the police can't pin a thing on
them without a confession. The police, however, are perfectly well aware
of this. So they offerthe prisonersa little incentive:if one of them defecb
and informson his partner,then that prisonerwill be grantedimmuni$
and go free-and will geta rewardto boot.The partner,meanwhile,will
b" se,-ntenced,to the maximum-and to addinsultto iniury, will be assessed
a fine tO,ffvdltfL ntrt prisoner'sreward.Of course,if bothof the prisoners
rat on eachother,thenboth of them servethe maximumand neithergeb
a reward.
So what do the prisonersdo-cooperate or defect?On the face of it,
theyoughtto cooperate with eachotherandkeeptheirmouthsshut,because
th"i *"y th.y bothgetthebestresultfreedom.But thentheygetto thinking.
PrisonerA, beingno fool, quicklyrealizesthat there'sno wayhe canhust
his partnernot to turn state'sevidenceand walk off with a fat reward,
leauinghim to payfor the privilegeof sittingin a iail cell. The temptation
is iustto great.He alsorealizesthat his partner,beingno fool either,is
 Peasants
Under Glass 262
thinking exactlythe samething about him. So prisonerA concludesthat
the only sane responseis to defect and tell the police everything, because
if his partner is crazyenoughto keephis mouth shut, then prisonerA will
be the one walking out with the cash. And if his partnerdoesthe logical
thing and talks-well, since prisonerA has to servetime anyway,at least
he won't be paying a fine on top of it. So the upshot is that both prisoners
are led by ruthlesslogic to the leastdesirableoutcome:jail.
In the real world, of course,the dilemma of trust and cooperationis
rarely so stark. Negotiations, personalties, enforceablecontracts, and any
number of other factorsaffect the players'decisions.Nonetheless,the pris-
oners'Dilemma doescapturea depressing amount of truth about mistrust
and the need to guard againstbetrayal.Considerthe Cold War, when the
hvo superpowerslocked themselvesin to a forty-year arms race that was
ultimately to the benefit of neither, or the seemingly endlessArab-lsraeli
deadlock, or the eternal temptation for nations to erect protectionist hade
barriers.or in the natural world, considerthat an overly trustingcreature
might very well get eaten. So once again: Why should any organism ever
dare to cooperatewith another?
A big part of the answer came in the late 1970swith a computer tour-
nament organizedat Michigan by Holland'sBACH colleagueRobertAx-
elrod, a political scientistwith a long-standinginterestin the cooperation
question. Axelrod's idea for the tournament was straighforward: anyone
who liked could enter a computer program that would take the role of one
of the prisoners.The programswould then be paired up in variouscom-
binations, and they would play the Prisoners'Dilemmagame with each
other by choosingwhether to cooperateor defect. But there was a wrinkle:
insteadof playing the game iust once, each pair of programswould play it
over and over again for 200 moves. This would be what game theorists
called an iferctdPllg-orylglQileggra, arguablya much more realisticway
of ,.pr.r.ffiti*rh1p, weusuallygetintowith
each other. Moreover, this repetition would allow the programs to base
their cooperate/defectdecisions on what the other program had done in
previous moves.If the programsmeet only once, then defectionis obviously
the only rational choice. But when they meet many times, then each
individual program will developa history and a reputation. And it was far
from obvious how the opposingprogram should deal with that. Indeed,
that was one of the main things Axelrod wanted to learn from the tour-
nament: what strategieswould produce the highest payoff over the long
run? Should a programalwaysturn the other cheekand cooperateregardless
of what the other playerdoes?should it alwaysbe a rat and defect?should
 2U COMPLEXITY

it respond to theotherplayer's movesin somemorecomplexmanner? And

if so, what?
In fact, the fourteenprogramssubmittedin the 6rst round of the tour-
namentembodieda varietyof complexstrategies. But much to the aston-
ishmentof Axelrod and everyoneelse, the crown went to the simplest
strategyof all: TIT FOR TAT. Submitted by psychologist Anatol Rapoport
of the Universityof Toronto,TIT FOR TAT would start out by cooperating
on the Srstmove,and from thereon out would do exactly whatthe other
programhaddoneon the movebefore.That is, the TIT FOR TAT strategy
incorporatedthe essence of the carrotand the stick. It was "nice" in the
sensethat it would neverdefectfirst. It was"forgiving" thein sensethat it
would rewardgoodbehaviorby cooperating the nexttime. And yet it was
"tough" in the sensethat it would punishuncooperative behavior de-
by
fecting the next time. Moreover, it was "clear" in the sensethat its strategy
*", ,o simple that the opposingprogramscould easilyfigure out what they
were dealingwith.
Of course,with only a handfulof programs enteredin the tournament,
therewasalwaysthe possibilitythat TIT FOR TAT'ssuccess wasa fluke.
But maybenot. Of thefourteenprograms submitted,eightwere"nice" and
would neverdefect6rst. And everyone of them easilyoutperformedthe
I six not-nicerules.So to settlethe questionAxelrodheld a secondround
\of the tournament,specificallyinviting peopleto hy to knockTIT FOR
/tA.I' offitr throne.Sixty-twoentrantstried-and TIT FORTAT wonagain.
The conclusionwas almostinescapable. Nice guys-or more precisely,
nice,forgiving, tough, and clearguys-can indeedfinishfirst'
,/ Holla.d and the other members of the BACH group wele naturally
i enchantedby all this. "l'd alwaysbeen tremendously botheredby the Pris-
\ onerr'Dilemma,"saysHolland."lt wasone of thosethingsI iustdidn't
Iike.So to seethe resolutionwasa delight.Justinvigorating.Greatstuff."
It waslost on no one that TIT FOR TATs success had profoundim-
plicationsboth for biologicalevolution and for human affairs.In hi{984
R-
Usgk--&ggjuttqt*pigqpqclLo.!'AxelrodpointedoutthatTlTffi
rar int.i""fd-.rn leadto cooperation in a wide varie$of socialsettings-
including someof themostunpromising
-*as situationsimaginable. Hisfavorite
exa*ple the "live-andlet-live" system that spontaneously de.reloped
durini world war I, when units in the frontline trenches would refrain
from shootingto kill, solongasthe otherside refrainedaswell. The troops
on one sideof no-man's-land had no chance communicate their
to with
counterpartson the other side, and they certainlyweren'tfriends.But
whatmadethe systemworkwasthat the sameunitswereboggeddownon
 Under Class
Peasants 265

both sidesfor monthsat a time, giving them a chanceto adaptto each

other.
In a chapterof the bookcoauthored with biologistandBACH colleague
William Hamilton (and adapted from a prize-winningI98l paperin the
journal science), Axelrodalsopointedout thatTIT FOR THT interactions
leadto cooperationin the naturalworld evenwithoutthe benefitof intel-
ligence.Examplesincludelichens,in which a fungusextracts nutrientsI
from the underlyingrock while providinga home for algae that in turn I
providethe funguswith photosynthesis; the ant-acacia hee, which housesI
andfeedsa typeofant that in turn protectsthe tree;andthe fig tree,whoseI
fowers serveasfood for fig waspsthat in turn pollinatethe flowersand
I
scatterthe seeds.
More generally,Axelrodsaid,the processof coevolutionshouldallow
TIT FOR TAT-stylecooperation to thriveevenin a world full of treach-
eroussleazoids. supposea fewTIT FOR TAT individualsarisein sucha
worldby mutation,he argued.Then solongasthoseindividualsmeetone
anotheroften enoughto havea stakein future encounters, theywill start
to form little pocketsof cooperation.And once that happens,they will
performfar betterthan the knife-in-the-back typesaroundthem. Their
numberswill thereforeincrease.Rapidly.Indeed,saidAxelrod,TIT FOR
Tm-style cooperation will eventuallytakeover.And onceestablished, the
cooperative individualswill be thereto stay;if less-cooperative $pes try to
invadeandexploittheirniceness, TIT FORTAT'spolicyof toughness will
punishthem so severely thattheycannotspread. "Thus,"wroteAxelrod,')
"the gearwheelsof socialevolutionhavea ratchet." I
Shortlyafter the book waspublished,Axelrodproduceda computer',
simulationof this scenarioin collaboration with Holland'sthen-graduate \
studentStephanieForrest.The question was whethera populationof in- \
dividualscoevolvingvia the geneticalgorithmcould discoverTIT FOR i
TAT. And the answerwasyes:in the computerruns,eitherTIT FOR Tlff I
or a strategyvery much like it would apPearand spreadthroughthe pop- |
ulationveryquickly."When it did that,"saysHolland,"we all threwour /
handsup and said,Hooray!"
This TIT FORTAT mechanism for the originof cooperation wasexactly
the sortof thing Holland meantwhen he saidthat people at the institute
oughtto be lookingfor the analogof "weather fronts" in the social sciences.
And the wholeissueof cooperation wasthereat the backof his mind while
he wasdeveloping Echo,he says.It wascertainlynot something thatcould
arisein the first versionof the program,because he had built in the as-
sumptionthatindividualorganisms will always fight.But in a more recent
 266 COMPLEXITY
version,he has tried to broadenthe organisms' repertoireto includethe
possibilityof cooperation.Indeed,he hasbeentrying to makeEcho into
a kind of "unified"modelof coevolution.
"At the institutewe'venow hadthreeongoingmodelsoutsideof Echo,"
he explains."We had a stockmarketmodel,the immunesystemmodel,
and a model that lStanfordeconomist]Tom Sargentmadethat involved
trading. I realizedthat theseall had very similar features.They all had
'hade,'in that thereweregoodsbeingexchanged in onewayor another.
'resource
I They all had transformation,'such as might be producedby
'mateselection,'
enzymes or productionprocesses. And theyall had which
actedas a sourceof technologicalinnovation.So I beganon the unified
model. I can rememberStephanieForrestand fohn Miller and I sitting
down and trying to 6gure out, What'sthe minimum apparatus that we
couldput into Echoto imitateall thesethings?It turnedout you coulddo
it by addingthingsto the offensiveand defensive chromosomes without
much changein the basicmodel. I addedthe possibilityof trading by
providingadditionalidentifiersdefinedby the chromosomes; thesewould
be analogousto trademarks, or molecularmarkerson the surfaceof the
cell. And the minuteI did thatI hadto add,for the firsttime, something
'lf
that lookedlike a rule in Echo: the otherguy showssuchand suchan
t identifyingtag,then I'm goingto attemptto tradeinsteadof fight'' That
I 'aberrations' aslying and
allowedthe evolutionof cooperation,and such
mimicry.With that, I sketched out howto do a versionof Sargent's model.
Then I startedsketchingout how I could makeEcholooklike an immune
systemmodelby stretchingit in anotherdirection,andsoon. The current
versionof Echo stemsfrom that."
This unified versionof Echo hasbeenquite successful, he says.With
it he'sbeenableto demonstrate both the evolution of cooperation and the
evolutionof predator-prey relationshipssimultaneously, in the same eco-
system.And that successhas inspired him to start work on still more
sophisticated variationsof Echo:"There'sa laterversionthat I'm program-
ming rightnowthatallowsthisthingto evolvemg!tJeg[-e(organisms," he
says."So now, insteadof talkingabouttradingind so on' my hope is that
I cantalkabouttheemergence andorganizations.
of individuals Thereare
nice thingsto be learnedwheneachagentis trying to maximize its repro-
ductiveratebut is constrained by the necessityof the continuation of the
overallorganization. Cancers area goodexample of failurein this dimen-
sion-I'll not talk aboutthe U.S. automotiveindustry!"
Practical of suchmodelsarestillin thefuture,saysHolland.
applications
But he'sconvincedthat a fewgoodcomputersimulationsalongtheselines
Peasants
Under Glass 267
might do more for the world than almostanythingelseon the SantaFe
agenda."lf we do thisright," he says,"thenpeoplewho arenot scientists-
peoplein Washington,for example-will be ableto createmodelsthat can
givethem somefeelingfor the implicationsof variouspolicyoptions,with-
out havingto knowall the detailsof how that modelactuallyworks."In
effect,he says,suchmodelswouldbe like flight simulatorsfor policy,and
would allow politiciansto practicecrashJandingthe economywithout
taking250 million peoplealongfor the ride. The modelswouldn't even
haveto be terriblycomplicated,solong astheygavepeoplea realisticfeel
for the way situationsdevelopand for how the most importantvariables
interact.
Hollandadmitsthat listenershavebeenprettyunderwhelmed when he
talksaboutthis flight simulatorideain Washington;mostpracticingpoli-
ticiansaretoo busydodgingthepunchescomingat themrightnowto think
aboutstrategyfor their nextfight. On the otherhand,he is clearlynot the
only one thinking in simulatorterms. In 1989the Maxis Companyof
Orinda,California,broughtouta simulationgamecalledSimCig, in which
the playertakesthe role of a mayorand triesto nursdGiiTher city to
prosperity in thefaceof crime,pollution,trafficcongestion,andtaxrevolts.
The gamequicklywentto thetopof thebest-seller charts.Realcity planners
sworeby it; assimpleasthe simulationwas,theysaid,andasmanydetails
as it left out, SimCityfelt right. Holland, of course,boughta copy im-
mediately-and lovedit. "SimCity is one of the bestexamples I know of
thisflight simulatoridea," he says.The SantaFeInstituteis talkingseriously
with Maxisaboutadaptinga SimCity-sgleinterfaceto usewith someof
its own simulations.And Hollandis now workingwith Maxisto develop
a user-friendlyversionof Echo that anyonecan usefor computerexperi-
ments.

Wet Labsfor the Mind

All through those early daysof the SantaFe economicsprogram, mean-

while, Brian Arthur was taking a keen interestin computerexperiments,
too. "Mostly in the program we were doing mathematical analysis and
proving theorems,iust as in standardeconomics,"he says."But because
we were studyingincreasingreturns,learning,and this ill-definedworld of
adaptationand induction, the problems often got too complicated for the
mathematics to handle. So then we had to resort to the computer to see
268 COMPLEXITY

how thingswould work out. The computerwaslike a wet lab wherewe

couldseeour ideasplayout in action."
Arthur'sproblem,however,wasthat the thoughtof computermodeling
gavea lot of economists the willies,evenin SantaFe."l guesswe'regoing
to haveto do simulationin economics,"Arrow glumly told him one day
overlunch. "But I think I'm too old to change."
"Thank god, my boy, I'm retiring," saidthe sixtyishHahn on another
occasion. "If the eraof theorems is passing, I don'twantto be there."
Arthur had to admit that therewasgod reasonfor the economists to
be leery; in many ways,he felt the sameway himself. "The history of
simulationin the fieldwasabsolutely dismal,"he says."Earlyin niy own
career,my colleagueCeoff McNicoll and I spenta lot of time lookingat
in
simulationmodels economics, and we cameto twoconclusions-which
were widely shared. The first was that, by and large,only peoplewho
couldn't think analytically resorted to computer simulations.The whole
for
cultureof ourdisciplinecalls deductive, logicalanalysis, andsimulation
runs counterto that. The secondconclusion was that you could prove
anythingyou wantedby tweakingthe assumptions deep in your model.
Often peoplewould startfrom basicallypolitical positions-say, we need
lower taxes-and then twiddle the assumptions to showthat lower taxes
would be wonderful.Geoffand I madea gameof goinginto modelsand
finding the one assumption you couldtweakthat wouldchangethe entire
outcome.Otherpeopledid thisaswell. Socomputersimulation gota very
badnamein socialscience andespecially in economics. It waskind of the
resortof the scoundrel."
Evenafterall theseyears,in fact, Arthur findsthat he'sstill allergicto
the word "simulation";he muchprefersto call whathe andhis colleagues
did in the economicsprogram"computerexperiments"-aphrasethat
capturesthe kind of rigorand precisionhe sawbeingpracticedby Holland
andthe SantaFephysicists. At thetime, hesays,theirapproach to computer
modelingwasa revelation. "I thoughtit waswonderful,"he says."In the
handsof someonewho was being extremelycareful,where all the as-
sumptions werecarefullylaidout,wheretheentirealgorithmwasexplicitly
given,wherethe simulationwasrepeatable andrigorous,like a lab exper-
iment-then I sawthat computerexperiments wouldbe perfectlyfine. In
fact,the physicistsweretellingusthattherewerethreewaysnowto proceed
in science:mathematicaltheory, laboratoryexperiment,and computer
modeling.You haveto go backand forth. Youdiscoversomethingwith a
computermodelthat seemsout of whack,and then you go and do theory
to try to understandit. And then with the theory,you go backto the
Peasants
Under Glass 269
computer or to the laboratory for more experiments.To many of us, it
seemedas though we could do the same thing in economicswith great
profit. We beganto rcalizethat we'd been restrictingourselvesin economics
unnaturally,by exploringonly problemsthat might yield to mathematical
analysis. But now that we were gefting into this inductive world, where
things startedto getvery complicated,we could extendourselvesto problems
that might only be studiedby computer experiment.I sawthis asa necessary
development-and a liberation."
The hope, of course,was that the SantaFe economicsprogram could
come up with computer modelscompelling enough to convincethe rest
of the professiqn-s1, at least, that would not turn them off any more than
they alreadywere. And, indeed, by the fall of 1988Arthur and his team
already had severalsuch computer experimentsunder way.
Arthur's own effort, begun in collaboration with Holland, was a direct
descendantof his original economy-under-glass vision. "By the time I got
to Santa Fe in fune 1988," he says,"I'd realizedthat we needed to start
with a more modestproblem than building a whole artificial economy.
And that led to the artificial stock market."
Of all the hoary old chestnutsin economics,he explains,stockmarket
behavior is one of the hoariest. And the reasonis that neoclassicaltheory
finds Wall Street utterly incomprehensible.Since all economic agentsare
perfectly rational, goesthe argument, then all investorsmust be perfectly
rational. Moreover, since theseperfectlyrational investorsalso have exactly
the same information about the expectedearnings of all stocksinfinitely
far into the future, they will alwaysagreeabout what every stockis worth-
namely, the "net presentvalue" of its future earningswhen they are dis-
counted by the interestrate. So this perfectlyrational market will never get
caught up in speculativebubblesand crashes;at most it will go up or down
a little bit as new information becomesavailableabout the various stocks'
future earnings. But either way, the logical conclusion is that the floor of
the New York Stock Exchange must be a very quiet place.
In reality, of course, the floor of the New York Stock Exchange is a
barelycontrolledriot. The placeis wrackedby bubblesand crashesall the
time, not to mention fear, uncertainty,euphoria,and mob psychologyin
every conceivablecombination. Indeed, saysArthur, a martian who sub-
scribed to the interplanetary edition of the Wall Street loumal might very
well end up thinking that the stock market was a living thing. "The stories
refer to the market almost as if it has psychologicalmoods," he says."The
market is jittery, the market is depressed, the market is confident." The
placeis a form of artificial life all by itself. So in 1988,saysArthur, it only
270 COMPLEXITV

seemedappropriateto try to model the stock market Santa Fe style: "Our

idea was that we would go in with a scalpel, excisethe perfectly rational
agentsfrom the standardneoclassicalmodels, and in their place slide in
artificially intelligent agentsthat could learn and adapt the way humans
do. So the model would have one stock, which the agentscould buy and
sell. And as they leamed rules for hading, you could watch what kinds of
market behavior emerged."
The question, obviously, was what that emergent behavior would be.
Would the agenb iust calmly settle down and start trading stock at the
standardneoclassicalprice? Or would they go into a more realistic pattern
of constant upheaval?Arthur and Holland had no doubt it would be the
latter. But in fact there was quite a bit of skepticismon that score, even
around the institute.
Arthur particularly remembersone meeting in March 1989, when Hol-
land was back down from Ann Arbor for a time, and severalother people
had come in for an economicsworlshop in the convent'ssmall conference
room. When the subject of the stockmarket model came up, Tom Sargent
and Minnesota'sRamon Marimon both arguedvery stronglythat the prices
bid by the adaptive agentswould very quickly settle down to the stock's
"fundamental value"-that is, the one predicted by neoclassicaltheory.
The market might show a few random fluctuations up or down, they said.
But the agents couldn't really do anything else; the fundamental value
would draw them in like an immensegravitational6eld.
"Well, fohn and I looked at each other and lust shook our heads," says
Arthur. "We said no-it was our shong instinct that the stock market we
were building had so much potential to self-organizeits own behavior, to
become complex, that rich new behavior would have to emerge."
It got to be quite a debate, Arthur recalls. He knew, of course, that
Sargenthad been an enthusiastfor the Holland approachto learning since
that first economicsworkshopin September1987. Indeed, Sargenthad
startedto study the impact of learning on economic behavior well before
that. And Marimon, meanwhile, was as enthusiasticfor computer exper-
imentationas Arthur himself was. But to Arthur, it didn't seemthat Mar-
imon and Sargentsaw learning as anything really new in economics.They
seemedto see it as a way of strengtheningthe standardideas-as a way of
understandinghow economic agentswill grope their way toward neoclass-
ical behavior even when they aren't perfectly rational.
Well, fair enough. Arthur had to admit that the two men had good
reasonto feel that way. Quite asidefrom theory-where Sargent'swork on
"rational expectations"was well known*they had quite a bit of experi-
 UnderGlass
Peasants 271

mentalevidenceon their side.In a numberof laboratorysimulations,with

,trrd"nt,playingthe roleof tradersin simplestockmarkets,researchers had
subiects converged on the fundamental price
shownthatihe experimental
veryquickly.Morlover,MarimonandSargent werewellalongon a Santa
Fe-stylecomputermodel of their own: another old chestnutknown as
Wicksells hiansle.The scenario here is that three $pesof agenbproduce
fr4ffif. mr.e typesof goods, one of which eventually emergesasa
mediumof exchange:-*on.y. And when Marimon and sargent replaced
the rationalagentsof the originalmodelwith classifiersystems,they found
that the ,yrt .n converged on the neoclassical solution everytime. (That
is, the medium of exchangewasthe good with the lowest storage cost-
metaldisks,say,insteadof freshmilk.)
Nonetheless, Arthur and Hollandstuckto their guns."The question
was,,,says Arthur,"doesrealistic adaptive behaviorleadyouto the rational
expectat'ions outcome?To *y mind, the answerwasyes-but only if the
probl.* is simple enough,or if the conditionsare repeatedagainand
again. Basically,rationaiexpectationis sayingthat peoplearen't stupid.
Thenit'sliketic-tac-toe: aftera fewtimesI Iearnto anticipate my opponent,
and we both play games.
perfect But if it's an on-off situationthat'snever
goingto happenagain,or if the situationis verycomplicated,sothat your
Ig.r; l,ru. io do an awful lot of computing,then you'reaskingfor a hell
oI a lot. Becauseyou're askingthem to have knowledgeof their own
expectations, of thedynamics of themarket,of otherpeople's expectations,
of other people's expectations about other people's expectations,et cetera.
And pretg soon, economics is loading almost impossible conditions-onto
these-hapiess agents."Underthosecircumstances, Arthur and Holland
argued,th. ,g.ntr areso far from equilibrium that the "gravitationalpull"
of-the rationalexpectations outcome becomes very weak. Dynamicsand
surpriseare everything.
'itre
debatewasboth affableand intense,Arthur recalls,and it wenton
like that for sometime. In the end, of course,neithersideconceded.But
Arthur definitelyfelt a challenge: If he and Hollandbelievedthat their
stockmarketmodelcould show realistic emergentbehavior,then it wasuP
to them to proveit.

Unfortunately,the programmingof the stockmarketmodel had pro-

ceededonly in fitsandstartsby thatpoint.Mostlyfis. ArthurandHolland
had roughedout an initial draftof the simulationoverlunch one dayin
fune 1988,whenboth of themwerein SantaFeto lectureat the institute's
 272 COMPLEXITY
first complexsystems summerschool.During the summer,backin Ann
Arbor, Holland had codedup a full-fledgedclassifiersystemand genetic
algorithmin the only computerlanguagethatArthur knew:BASIC. (This
waswhatfinallyled Hollandto giveup writinghisprograms in hexadecimal
notation;he hadto teachhimselfthe language,but he'swriftenin BASIC
eversince.)And duringthe fall, onceHollandwasbackin SantaFe for
the first few monthsof the economicsprogram,theyhad tried to develop
the stockmarketmodelfurther.But with Hollandgettingdeflectedoff into
Echo, and with Arthur boggeddown in adminishativeduties, nothing
happenedvery fast.
Worse,Arthur wasbeginningto realizethat classifiersystems,for all
their conceptualbrilliance,could often be a bearto work with. "In the
beginning,"he says,"the atmosphere at SantaFewasthatclassifier systems
coulddo anything.They couldcrackthestockmarketproblem.Theycould
makeyour coffeein the morning.So I usedto pull |ohn'sleg:'He5 John,
is it true that classifiersystems can producacold fusion?'
"But then in early 1989David lane and RichardPalmerorganizeda
studygroup on fohn Holland'sideas,wherewe would meet maybefour
timesper weekbeforelunch. fohn hadleft by then, but for abouta month
we workedour waythroughhis book Induction,took it apart.And aswe
got into classifiersystems technically,I discovered that you had to havea
very carefuldesignto makesurethe architecture workedin practice.You
had to be very carefulabouthow you hitchedone rule to another.Also,
you could have'deep'cl1l6.t syjtems-that is, onewhereruleshiggered
rutestnggereqrutesrn {onfiMfis-or you could have'wide'ones:stim-
ulus-response-type systems wheretherewouldbe 150diFmlfiays to act
underslightlydifferentconditions,but wherethe rulesweren'tlookingat
what eachother did. My experience
gru. rvr, v^P!.
- - _ . systems
wasthat wide ._, learnedvery
I /w-eJ,-anddg€p-UstSms *
didn't. "
QZ '
Arthur hada lot oftilki iSoutthisproblemwith Holland'sformerstudent
StephanieForrest,who wasnow at the Universityof New Mexico in Al-
I buquerque and a frequent visitor to the institute. The problem, she told
( him, was in Holland'sbucket-brigade algorithmfor assigningcredit to the
rules.Ifa bucketbrigadehasto passcreditbackthroughseveralgenerations

A of rules, it will usuallyhavevery little credit left to go around once it gets

back to the ancestors.So it's no wonder that the shallow systemslearn
better, Indeed, coming up with refinementsand alternativesto the bucket-
brigadealgorithm had become one of the most intensiveareasof classifier
systemresearch.
"For thoser€asons of classifier
I becameskeptical systems,"
saysArthur.
PeasantsUnder Glass 273

"With familiarity,the drawbacks had becomeclear' And yet the more I

lookedat them, the more I admiredthe thinking behindthem. I really
lovedthe idea that you could havemanyconficting hypotheses in your
mind, and that thesehypotheses could compete,so that you didn't have
to preprogram theexperiise.I began to conceiveof Holland'ssystems slightly
ditret from fohn. I thought of them asbeing like an ordinary computer
program "tlywith manymodulesandbranchingpoints,but wherethe program
it .if nrr to learnwhich moduleto higgerat any giventime, ratherthan
higgeringthem in a fixedsequence. And onceI beganto conceive-of t:-
as]r.[-'ad"pting computer program, I becamemuch morecomfortable.
That'swhat I thoughtfohn had achieved."
In any case,he iays, they finally did get a versionof the stockmarket
model up and running. sargenthimselfsuggested a numberof waysto
simplifyihe originaldesig.t,which helpeda lot. And in the latespringof
l98b Duke UniversigphysicistRichardPalmerioined proiect,bringing
the
with him his inestimable programming skills.
Palmer,meanwhile,was intriguedby the model for much the same
reasons thatHollandandArthurwere."It relatedto self-organization, which
is an areathat fascinates me a lot," he says. "How the brain is organized,
the natureof self-awareness, how life spontaneously arose-a fewbig ques-
tionsI keepin the backof mYmind'"
Besides,he says,he wasgettinga little restless with the proiectthat had
alreadydevouredmostof his time at SantaFe:the DoubleOral Auction
Tournament,ajointeffortwithCarnegie-Mellon',@
The tournament,which wasultimatelyheld in early
coiimfrTJofifi-Rust.
1990,hadbeenconceived duringthe firsteconomics workshopin Septem-
ber 1987.It wasvery similarin spiritto Axelrod'stournamenta decade
before.But insteadof playingthe iteratedPrisoners' Dilemmagame,the
programswould embodyvariousstrategies for brokersin a commodities
-"rk t suchasthe stockexchange.Is it bestto announceyour bid at the
start?Shouldyou keepquiet and wait for a betterprice?What?Sincethe
buyersandthe sellersin sucha marketarebothbiddingsimultaneously-
thusthe name"doubleauction"-the answerwasnot at all obvious'
Well, the tournamentpromisedto be a lot of fun, saysPalmer,and the
programmingthat he and his colleagues weredoingto getreadyfor it was
a challenge.But the agentsin the modelwereessentiallystatic.
".rLinly thetournament didn'thavethemagicof ArthurandHolland's
Forhim, iust
model,whereyou could hopeto seeagentsgrowingmoreand morecom-
plex,anddeveloping a realeconomiclife of theirown.
So in the earlyspring,Palmerpitchedin. By May 1989he andArthur
274 COMqLEXITv
hada preliminaryversionof thestockmarketup andrunning.As planned,
they startedtheir agentsoff from total stupidig-random rules-and let
them learn how to bid. As expected,they sawthe agentslearningener-
getically.
And on everyrun, they watchedthe damn thingsdoing exactlywhat
Tom Sargenthad saidthey would do. "We had a singlestockwith a $3
dividendin the model,"saysArthur. "We had a l0 percentdiscountrate.
So the fundamentalvaluewas$30.And the pricesindeedsettleddownto
fluctuations around$30.It provedthe standard theory!"
Arthur waschagrinedand disgusted. It seemedthat the only thing left
to do wasto call Sargent backat Stanfordandcongratulate him. "But then
Richardand I walkedin one morningand ran it on my Macintosh.We
keptlookingat it, discussing how to improveit. And we noticedthat every
timethepricehit 34,theagents wouldbuy.Wecouldgraphthat.It seemed
anomalousbehavior.We thoughtit wasa bugin the model.But thenafter
thinking hard about'it for an hour or so, we realizedthat therewas no
mistake!The agentshaddiscovered a primitiveform of technicalanalysis.
That is, theyhadcometo the beliefthat if the pricewentup enough,then
it wouldcontinueto go up. Sobuy. But, of course,thatbeliefbecamea
self-fulfillingprophecy:if enoughagentstriedto buy at price34,thatwould
causethepriceto go up."
Furthermore,he says,exactlythe inversehappenedasthe price fell to
25: the agentshied to sell, therebycreatinga self-fulfillingprophecyfor
fallingprices.Bubblesandcrashes! Arthurwasexuberant. AndevenPalmer,
normallya mostcautiousman,foundtheenthusiasm infectious.The result
would be confirmedagainand againin later,morecompleteversionsof
the model,saysArthur. But that morningin May 1989theyknewthey
had it.
"lmmediately,"he says,"we realizedthat we had the 6rstglimmerof
an emergentpropergin the system.We hadthe first glimmerof life."
s*t*a****s4s6*Se68**C***€s*6e*t

At the very end of November1988,the secretaries at the LosAlamosCenter

for Nonlinear Studies handed chris I-,angtona sealed, officialJooking en-
velope, inside of which he found a memorandum signed by laboratory
director Siegfried Hecker:

It has recently come to our attention that you have entered the third
yearof your postdoctoralfellowship,at the sametime that you havenot
yetcompletedyourPh.D. dissertation. Accordingto DOE Rule40-l 130,
we are not allowedto employ postdoctoral fellowsbeyondtheir third year
unlessthey haveobtainedthe Ph. D. degree. In your case,dueto a clerical
error, we neglectedto warn you in advanceof the possibleviolation of
this rule. In view of this fact, we have obtainedan extensionfrom the
DOE office, so that you will not be liable to returnPaymentfor the FY89
portion of your fellowship.However,until you haveobtainedyour Ph'D'
degree,we will not be ableto extendyour appointmentbeyondlZll/88'

In short, "You're fired." Panic-shicken,langton ran to Center associate

director Gary Doolen, who gravelyassuredhim that yes, there was such a
rule. And yes, Hecker really could do this-
l,angton still shuddersat the memory. Those bastardsleft him in freak-
out mode for a full two hours before they actually sprang the surprise
party. "The DOE rule number should have given it away," saysDoyne
h"r-.r. who had written the memo and organized the entire charade.
"Chris was turning forty, and his birthday was I l/10."
276 COMPLEXIW
Happily,oncel,angtonhadrecovered, the partyturnedout to bea pretty
goodone. After all, Ph.D. candidates don't turn forty everyday.Farmer
hadevengottenl.angton'scolleagues at the Centerand in the lab'stheory
divisionto chip in andbuy him a newelechicguitar."But I wasseriously
trying to prod him to finish up," saysFarmer,"becauseI wasgenuinely
worriedthat the shit wasgoingto hit the fan overhis not havinga degree.
And I suspected that therereallymight be somekind of rule againstit."

The A-Life Papers

langton gotthe message loudandclear.He'dgottenit longago.Nobody

wantedto seethe end of his dissertation morethan he did. And in the
yearsincethe artificiallife workshop,he had actuallymadequite a bit of
progresson it. He had convertedhis old cellular automatacode from
Michiganto run on Los Alamos'Sun workstations. He had exploredthe
edge-of-chaos phase transitionwith innumerable computer experiments.
He hadevengonedeepinto the physics literature,learninghow to analyze
the phasetransitionwith hard-core statistical
mechanics,
But asfor actuallywritingthe thing-well, the yearhadiustgoftenaway
fromhim. The factwasthatmostof thetimesincetheartificiallife workshop
hadbeendevouredby the workshop's aftermath.GeorgeCowanandDavid
Pineshad invitedhim to collectwritten versionsof the talksand publish
them in the nameof the SantaFe Institute,asoneof a seriesof booksthe
institutewaspublishingon thesciences of complexig.But PinesandCowan
had alsoinsistedthat thosepapersbe rigorouslyreviewedby outsidesci-
entists,in exactlythe samewaythat they would havebeenin any other
researchpublication.The institutecouldn't affordto be associated with
fakiness,they told him. This had to be science, not videogames.
That wasfine by Langton,who'd alwaysfelt that wayhimself.But the
upshotwasthat he'd spentmonthsplayingeditor-which meantreading
forty-fivepapers an average offour timesapiece,sendingeachofthem out
to severalreviewers,sendingthe reviewers' commenbbackto the authors
with demands andgenerally
for rewrites, caiolingeverybody to getit done
beforethe sun grewcold. Then he'dspentmoremonthswritinga preface
a"afntrffitrli5"pl.t to the book."It iust took enormousamountsof
time,"he sighs.
On the otherhand,thewholeprocess hadbeenenormously educational.
"It waslike studyingfor your qualifiers,"he says."'What'sgood?What's
BS?It reallvmademe a masterof thatmaterial." And nowthatthe volume
Waitingfor Carnot 277

was finally done-with all the rigor that Cowan and Pinescould askfor-
l,angton ielt that he had createdsomething much more than a seriesof
p.p.rr. His dissertationmight be off in limbo somewhere,but the workshop
uoi,r*" promisedto lay the foundation for artificial life asa seriousscience.
Moreover, by taking all the ideasand insights that people had brought to
the workshop, and distilling them into a preface and a forty-seven-page
introductory article, l,angton had written his clearestand most articulate
manifesto yet for what artificial life was all about.

Artificial life, he wrote, is essentiallyiust the inverse of conventional

biology. Insteadof being an effort to understandlifeby analysis-dissecting
Iiving communities into species,organisms,organs' tissues,cells, organ-
elles, membranes,and finally molecules-artificial life is an effort to un-
derstandlife by s/ntfresisipufting simple piecestogetherto generatelifelike
behavior in ml-n-made iystems. lb credo is that life is not a property of
matter per se, but the organizationof that matter. Its operatingprinciple is
that the laws of life must be laws of dynamical form, independent of the
details of a particular carbon-basedchemistry that happenedto arise here
on Earth four billion yearsago.Its promiseis that by exploringolher possible
biologies in a new medium-computers and perhapsrobots-artificial life
researcherscan achieve what space scientistshave achieved by sending
probesto other planets:a new understandingof our own world through a
cosmic perspectiveon what happenedon other worlds. "Only when we are \
able to view life-as-we-know-itin the contextof.life-as-it-could-bewill we I
really understandthe nature of the beast," Langton declared' I
The idea of viewing life in terms of its abstractorganization is perhaps
the single most compelling vision to come out of the workshop, he said.
And it's no accident that this vision is so closelyassociatedwith computers:
they share many of the same intellectual roots. Human beings have been
searchingfor the secretof automata-machines that can generatetheir own
behavior-at leastsincethe time of the Pharaohs,when Egyptiancraftsmen
createdclocks basedon the steadydrip of water through a small orifice. In
the first century A.D., Hero of Alexandria producedhis treatisePneumatics,
in which he described(among other things) how pressurizedair could
generatesimple movements in various gadgetsshapedlike animals and
humans. In Europe, during the greatage of clockworksmore than a thou-
sandyearslater, medievaland Renaissance craftsmendevisedincreasingly
elaboratefigures known as "iacks," which would emergefrom the interior
of the clock to strike the hours; some of their public clockseventually grew
278 COMPLEXITY
to include large numbers of figures that acted out entire plap. And during
the Indushial Revolution, the technology of clockwork automata gave rise
to the still more sophisticatedtechnologyof processcontrcl, in which factory
machines were guided by intricate sets of rotating cams and interlinked
mechanical arms. Moreover, by incorporatingsuch refinementsasmovable
cams, or rotating drums with movable pegs, nineteenth-centurydesigners
soon developed controllers that could be adiusted to generatemany se-
quences of action from the same machine. Along with the development
of calculating machinesin the early twentieth century, noted Langton, "the
introduction of such programmable controllers was one of the primary
developmentson the road to general-purposecomputers."
Meanwhile, he said, the foundations for a generaltheory of computing
were being laid by logicians who were hying to formalize the notion of a
prrcedure, a sequenceof logical steps.That effort culminated in the early
decadesof the twentieth century with the work of Alonzo Church, Kurt
G<idel, Alan Turing, and others, who pointed out that the essenceof a
mechanical process-the "thing" responsiblefor its behavior-is not a thing
at all. It is an abshact conhol structure, a program that can be expressed
as a set of rules without regard to the material the machine is made of.
Indeed, said l,angton, this abstractionis what allows you to take a piece of
software from one computer and run it on another computer: the "ma-
chineness"of the machine is in the software,not the hardware.And once
you've acceptedthat, he said, echoing his own epiphany at Massachusefts
General Hospital nearly eighteen yearsbefore, then it's a very small step
to say that the "aliveness"of an organism is also in the software-in the
organization of the molecules, not the moleculesthemselves.
Now admiftedlx said Langton, that step doesn't alwayslook so small,
especiallywhen you consider how fuid, spontaneous,and organic life is,
and how controlled computers and other machines are. At first glance it
seemsludicrous even to talk about living systemsin those terms.
But the answer lies with a secondgreat insight, which could be heard
at the workshop again and again: living systemsare machines, all right,
but machineswith a very different kind of organizationfrom the oneswe're
used to. Instead of being designedfrom the top down, the way a human
engineerwould do it, living systemsalwaysseemto emergefrom the bottom
up, from a population of much simpler systems.A cell consistsof proteins,
DNA, and other biomolecules.A brain consistsof neurons.An embryo
consistsof interacting cells. An ant colony consistsof anb. And for that
mafter, an economy consistsof firms and individuals.
Of course, this was exactly the point that fohn Holland and the rest of
Waiting for Carnot 279
theSantaFecrowdlikedto makeaboutcomplexadaptive systems in general.
The differencewasthat Hollandsawthis populationstructuremainly asa
collectionof building blocksthat could be reshuffledfor very efficient
evolution,whereas Langtonsawit mainlyasanopportunity for rich, lifelike
dynamics."The mostsurprising lessonwe havelearnedfrom simulating
complexphysicalsystems on computersis that complexbehaviorneednotl
haw complexroofs,"he wrote,completewith italics."lndeed,tremendously I
interestingand beguilinglycomplexbehaviorcanemergefrom collections
of extremelysimplecomponents. "
Langtonwasspeaking from the hearthere,sincethatstatement soclearly
reflectsthe experience of discoveringhis self-reproducing cellularautom-
aton. But the statementappliedequallywell to one of the most vivid
demonstrations at the artificial life workshop:Craig Reynolds'flock of
"boids."Insteadof writingglobal,top-downspecifications for howthe flock
shouldbehave,or telling his creatures to follow the leadof one BossBoid,
Reynolds had usedonly the threesimplerulesof local,boid-to-boid in-
teraction.And it wasprecisely thatlocalitythatallowedhis flockto adapt
to changingconditionssoorganically. The rulesalways tendedto pull the
boids together,in somewhatthe sameway that Adam Smith'sInvisitrle
Hand tendsto pull supplyinto balancewith demand.But just as in the
economy,the tendencyto converge wasonly a tendency,the resultofeach
boid reactingto what the otherboidsweredoingin its immediateneigrr-
borhood.So when a flock encountered an obstaclesuchasa pillar, it had
no trouble splittingapartand flowing to either sideas eachboid did its
own thing.
Tiy doing that with a singleset of top-levelrules,said langton. The
systemwould be impossiblycumbersome and complicated,with the rules
telling eachboid preciselywhat to do in everyconceivable situation.ln
fact,he hadseensimulationslike that;theyusuallyendedup lookingierky
andunnatural,morelikean animated cartoonthanlikeanimated life.And
besides, he said,sinceit's effectivelyimpossible to covereveryconceivable
situation,top-downsystems areforeverrunningintocombinations of events
they don't know how to handle.They tend to be touchyand fragile,and
theyall too oftengrindto a halt in a ditherof indecision.
The samekind of bottom-up,populationthinking wasresponsible for
the graphicalplantspresented by AristidLindenmayer of the Universityof
Utrechtand Prezemyslaw Prusinkiewcz of the Universityof Reginain Sas-
katchewan.Theseplantsweren'tjustdrawnon the computerscreen.They
wercgrown.They siartedfrom a singlestem,and then useda handful of
simplerulesto tell eachbranchhow to makeleaves,flowers,and more
280 COMPLEXITY
branches.Once again,the rulessaidnothingaboutthe overallshapeof
the final plant. They weremeantto modelhow a multitudeof cellsdif-
ferentiateand interactwith one anotherduring the courseof the plant's
development.Nonetheless, theyproducedshrubs,or trees,or flowersthat
lookedstartlinglyrealistic.If the ruleswerechosencarefullyenough,in
fact, they could producea computerplant that lookedvery much like a
known species.(And if thoseruleswerethen changedevenslightly,they
might producea radicallydifferentplant, thus illustratinghow easyit is
for evolutionto makelargeleapsin outwardappearances by makingonly
tiny changesin the courseof development.)
The themewasheardoverandoveragainat theworkshop,saidLangton:
the way to achievelifelike behavioris to simulatepopulationsof simple
units insteadof one big complexunit. Uselocal conhol insteadof global
conhol. Let the behavioremergefrom the bottom up, insteadof being
specifiedfrom the top down. And while you'reat it, focuson ongoing
behaviorinsteadof the 6nal result.As Hollandlovedto point out, Iiving
systems neverreallysettledown.
Indeed,saidl,angton,by takingthis bottom-upideato ib logicalcon-

v
clusion, you could seeit as a new and thoroughlyscientificversionof
vitalism:the ancientideathat life involvessomekind of energy,or force,
or spirit that transcends merematter.The fact is that life doestranscend
merematter,he said-not becauseliving systems are animatedby some
vital essence operating outsidethe laws of physics and chemistry,but be-
causee population of simplethings following simple rulesof interaction
can behavein eternallysurprisingways.Life may indeed be a kind of
biochemical machine,he said.But to animatesucha machine"is not to
bringlife to a machine;rather,it is to organize a populationof machines
'alive.'"
in sucha waythat their interactingdynamicsare
Finallx saidLangton,therewasa third greatideato be distilledfrom
the workshoppresentations: the possibilitythat life isn't iust lifte a com-
putation,in the senseof beinga propertyof the organizationratherthan
the molecules.Life literallyis a computation.
To seewhy,saidLangton,startwith conventional, carbon-based biology.
As biologistshavebeenpointingout for morethan a century,one of the
moststrikingcharacteristics of anylivingorganismisthedistinctionbetween
its genotype-thegeneticblueprintencodedin its DNA-and if pheno'
type-the structurethat is createdfrom thoseinstructions.In practice,of
course,the actualoperationof a living cell is incrediblycomplicated,with
eachgeneservingasa blueprintfor a singletypeof proteinmolecule,and
with myriadproteinsinteractingin the body of the cell in myriadways.
Waitingfor Carnot 281

But in effect, said langton, you can think of the geno$peas a collec-
tion of little computerPrograms executingin parallel,one programPer
gene.When activated,eachof theseprogramsentersinto the logicalfray
Ly competingand cooperatingwith all the other activeprograms'And
collectively,theseinteractingprogramscarry out an overallcomputation
that is the phenogpe:the structurethat unfoldsduring an organism's
development.
Next, movefrom carbon-based biologyto the more generalbiologyof
artificial life. The sameflotionsapply,saidl,angton.And to capturethat
fact, he coinedthe term generalized genotype,or GTYPE, to referto any
collectionof lowlevel rules.He likewisecoinedthe termgeneralized phe-
notype,or PTYPE, to referto the shuctureand/orbehaviorthat results
whenthoserulesareactivatedin somespecificenvironment.[n a conven-
tional computerprogram,for example,the GTYPE is obviouslyiust the
computercodeibelf, andthe FIYPE is whatthe programdoesin resPonse
to input from the user.In Langton'sown self-reproducing cellularauto-
maton,the GTYPE is the setof rulesspecifringhow eachcell interactedwith
its neighbors,and the P|YPE is the overallpattern.In Reynolds'boids
program,the GTYPE is the setof threerulesthat guidesthe fight of each
boid, and the PTYPEis the focking behaviorof the boidsasa grouP.
More generally,said[an$on, this conceptof a,GTYPE is essentially
identicalwith John Holland'sconceptof an "inte?nTl*ffi6ifel"; the only
differenceis that he placedevenmore emphasis than Holland did on its
role asa computerprogram.And by no coincidence,the GTYPE concept
appliesperfectlywell to Holland'sclassifiersystems, wherethe GTYPE of
a givensystemis justits setof classifier rules.It likewiseappliesto his Echo
model,wherea creature's GTYPE consists of its offenseanddefensechro-
mosomes.It appliesto BrianArthur'seconomy-under-glass models,where
an artificial agent'sGTYPE is its hardlearnedset of rulesfor economic
behavior.And it applies,in principle, to any complexadaptivesystem
whatsoever-anythingthat has agenb interactingaccordingto a set of
rules.As their GTYPE unfoldsinto a PTYPE,they are all performinga
computation.
Now, what'sbeautifulaboutall this, saidLangton,is that onceyou've
madethe link behveenlife and computation, you canbringan immense
amountof theoryto bear.For example,Why is life quite literallyfull of
surprises? Because,in general,it is impossibleto startfrom a givensetof
GTYPE rules and predictwhat their ffYPE behaviorwill be-even in
principle.This is the undecidabilig theorem,oneof thedeepest resultsof
computerscience:unlessa computerprogramis ufterlytrivial, the fastest
 . tf lf,^,1 ? ?o$n
U",lit,JrU,lb 1.cfrm).owt l,o( lt"u.
282 * v 9^l), Aq (nflr$a/ o A-'cno*, COMPLEnTTY
way to find out what it will do is to run it and see.There is no general-
purposeprocedure that can scan the code and the input and give you the
answer any fasterthan that. That's why the old saw about computers only
doing what their programmerstell them to do is both perfectly true and
virtually irrelevant;any pieceof codethat'scomplexenoughto be interesting
will alwayssurpriseits programmers.That's why any decentsoftwarepackage
has to be endlesslytestedand debuggedbefore it is released-and that's
why the users alwaysdiscover very quickly that the debugging was never
quite perfect.And, most important for artificial life purposes,that'swhy a
living systemcan be a biochemical machine that is completely under the
control of a program, a GTYPE, and yet still havea surprising,spontaneous
behavior in the PTYPE.
Conversely,said Langton, there are other deeptheoremsfrom computer
sciencestatingthat you can't go the other way,either. Civen the specification
for a certain desiredbehavioq a PTYPE, there is no generalprocedure for
finding a set of GTYPE rules that will produceit. In practice,of course,
those theorems don't stop human programmersfrom using well-testedal-
gorithms to solve preciselyspecifiedproblems in clearly defined environ-
ments, But in the poorly defined,constantlychangingenvironmentsfaced
by living systems,said Langton, there seemsto be only one way to proceed:
hial and error, also known as Darwinian natural selection.The process
may seemterribly cruel and wasteful, he pointed out. In effect, nature does
its programming by building a lot of different machines with a lot of
randomly differing CTYPES, and then smashingthe onesthat don't work
very well. But, in fact, that messy,wasteful processmay be the best that
nature can do. And by the same token, fohn Holland's genetic algorithm
approachmay be the only realisticway of programming computersto cope
with messy, ill-defined problems. "It is quite likely that this is the only
efficient, general procedurethat could find GTYPEs with specific ffYPE
traits," Langton wrote.

In writing his introductory chapter,l,angton very carefully avoidedmak-

ing any claim that the entitiesbeing studiedby the artificial lifers were
"really" alive. Obviously, they weren't. Boids, plants, self-reproducingcel-
lular automata-none of them were anything more than a simulation, a
highly simplified model of life having no existenceoutsideof a computer.
Nonetheless,since the whole point of artificial life researchwas to grapple
with the most fundamental principles of life, there was no avoiding the
question:Could human beingsultimately createartificial life for real?
Waitingfor Carnot 283

langton foundthatquestionto bea toughone,not leastbecause neither

he nor anyoneelsehad a clearideaof what "real" artificiallife would be
like. Somekind of geneticallyengineered superorganism, perhaps? A self-
reproducing robot?An overeducated computervirus? What is life, precisely?
How do you knowfor surewhen you'vegot it and when you haven't?
Not surprisingly,therehadbeena gooddealof discussion on this point
duringthe workshop, not only during the sessions, but in the hallways and
in loud, lively debates over dinner. Computer viruses wele a particularly
hot topic:manyof theparticipants felt thatviruseshadcomeuncomfortably
closeto crossing the line already. The peskythingsmet almosteverycri-
terionfor life thatanyone could think of. Computervirusescouldreproduce
andpropagate by copyingthemselves into anothercomputeror to a floppy
disk. They could storea representation of themselves in computercode,
analogous to DNA. They couldcommandeer the metabolism of their host
(a computer)to carryout their own functions,much real as viruses com-
mandeerthe molecularmetabolismof infectedcells' They could respond
to stimuli in their environment(the computeragain).And-courtesy of
certainhackerswith a warpedsenseof humor-they could evenmutate
and evolve.Tiue, computerviruseslived their lives entirelywithin the
cyberspace of computersand computernetworks.They didn't haveany
independent existence out in the materialworld.But thatdidn't necessarily
rule them out as living things.If life wasreallyiust a questionof organi-
zation,asLangtonclaimed,thena properlyorganized entitywouldliterally
be alive,no mafterwhat it wasmadeof.
Whateverthestatusof computerviruses,however,Langtonhadno doubt
that"real" artificiallife wouldonedaycomeinto being-and soonerrather
than later.What with biotechnology, robotics,and advanced softwarede-
velopment,moreover,it wasgoingto happenfor commercialand/ormil-
itary reasons whetherhe andhis colleagues studiedthe subiector not. But
that iust madethe research all the moreimportant,he argued:if we really
are headedinto the BraveNew World of artificial life, then at leastwe
oughtto be doing it with our eyesopen.
"By the middle of this century,"he wrote,"mankindhad acquiredth. -V
'
powerto extinguishlife on Earth. By the middle of the nextcentury,he
will be ableto createit. Of the two, it is hardto saywhich placesthe larger
burdenof responsibiligon our shoulders.Not only the specifickinds of
living thingsthatwill exist,but theverycourseof evolutionitselfwill come
moreand moreunderour control."
Given that prospect,he said,he felt that everyoneinvolvedin the field
shouldgo right out and rcadFrankenstein: it's clearin the book(although
2U COMPLEXITY
not in the movie) that the doctor disavowedany responsibilityfor his cre-
ation. That could not be allowed to happen here. The future effect of the
changeswe makenow areunpredictable,he pointedout, evenin principle.
Yet we are responsiblefor the consequences,nonetheless.And that, in
turn, meant that the implicationsof artificial life had to be debatedin the
open, with public input.
Furthermore, he said, supposethat you could ueate life. Then suddenly
you would be involved in something a lot bigger than some technical
definition of living versusnonliving. Very quickly, in fact, you would find
yourself engagedin a kind of empirical theology. Having createda living
creature, for example, would you then have the right to demand that it
worship you and make sacrificesto you? Would you have the right to act
as its god? Would you have the right to destoy it if it didn't behave the
way you wanted it to?
Good questions, said l.angton. "Whether they have correct answersor
not, they must be addressed,honestly and openly. Artificial life is more
than just a scientificor technicalchallenge;it is a challengeto our most
fundamental social, moral, philosophical, and religious beliefs' Like the
Copernican model of the solar system,it will force us to reexamineour
place in the universeand our role in nature."

The New Second Law

rhetoricsometimes
If Langton's to soara bit higherthanmost
seems
scientificprose-well, that wasn'tat all unusual in his corner of Los Ala-
mos. Doyne Farmer, for one, was famous for his cruiSesin the conceptual
stratosphere.A prime example was "Artificial Life: The Coming Evolu-
tion," a nontechnical paper that he coauthoredin 1989 with his wife,
environmental lawyer Alletta Belin, and then delivered at a Caltech sym-
posiumcelebratingMurray Gell-Mann'ssixtiethbirthday:"With the advent
of artificial life," they wrote, "we may be the first creaturesto createour
own sllccessors. . If we fail in our task 4s creators,they may indeed be
cold and malevolent. Howevet, if we succeed,they may be glorious, en-
lightened creaturesthat far surpassus in their intelligence and wisdom. It
is quite possiblethat, when the consciousbeings of the future look back
o., ihi, era, we will be most noteworthy not in and of ourselvesbut rather
for what we gave rise to. Artificial life is potentially the most beautiful
creationof humanitY."
Rhetoric aside,however,Farmerwasperfectlyseriousabout artificial life
Waitingfor Carnot 285
as a new kind of science.(Most of the "Coming Evolution" paperwas in
fact a reasonablysoberassessment of what the field might hope to accom-
plish.) By no coincidence,he was equally seriousabout supportingChris
Langton. It was Farmer, after all, who brougtrt Langton to Los Alamos in
the first place. And despitehis exasperation over Langton'smuch-delayed
dissertation, he found no reason to regret having done so. "Chris was
definitely worth it," he says."Peoplelike him, who have a real dream, a
vision of what they want to do, are rare. Chris hadn't learned to be very'
efficient. But I think he had a good vision, one that was really needed'
And I think he wasdoing a really good iob carrying it out. He wasn't afraid
to tacklethe details."
Indeed, Farmer was wholeheartedlyserving as a mentor to Langton-
even though Langton happenedto be 6ve yearsolder than he was. Down
the hill, where Farmer was one of the very few young scientistsincluded
in the inner circle of the SantaFe Institute, he had persuadedCowan to
contribute$5000towardLangon's artificial life workshopin 1987.Farmer
had made sure that Langton was invited to speakat institute meetings. He
had served as an advocateon the in$itute's scienceboard for bringing in
visiting scientiststo work on artificial life. He had likewiseencouraged
Langton to set up an ongoing seriesof artificial life seminarsup at Los
Alamos, with occasionalsessionsdown in Santa Fe. And perhapsmost
important, when Rrmer had agreedin 1987 to head the new Complex
Systems groupwithin the LosAlamostheorydivision,he had madeartificial
life one of the group'sthree maior researchefforts,along with machine
learningand dynamical systemstheory.
Farmer wasn't exactly a natural-born administrator type. At age thirty-
five, he wasa tall, angular New Mexican who still wore a graduatestudent's
ponytailand tee-shirtsthat saidthingslike, "QuestionAuthority!" He found
bureaucraticbusywork to be a pain, and he found the writing of proposals
beggingmoney from "some boneheadback in Washington"to be even
more of a pain. Yet Farmer had an undeniable gift for generating both
funding and intellectual excitement. In the field of mathematical predic-
tion, where he had originally made his reputationand wherehe still spent
most of his researchtime, he wasat the forefront of finding waysto project
the future behaviorof systemsthat seemedhopelesslyrandom and chaotic-
including certain systems,such as the stock market, where people had
incentiye to project the future. Moreover, Farmer had no compunction
about channeling most of the group's "general-purpose"money toward
langton and the tiny cadreof artificial life researchers,while making his
own nonlinear forecastingwork and other effortspay for themselves."Fore-
286 COMPLEXITY
castingproducespracticalresults,so that I could promisethe funding
agenciesa payoffwithin a year,"he says,"whereaspracticalresultsfrom
artificiallife are fartherin the future. With the currentfunding climate,
this makesAlife almostunfundable. This wasdrivenhometo me when
one of the agencies that fundedmy predictionwork calledto askabouta
proposaltheyhad receivedto studyartificiallife. Fromtheir attitudeit was
quiteclearthat they viewedartificiallife on a par with flyingsaucers or
astrology. They were upset to seemy name appearingin the list of refer-
ences. "
This wasnot Farmer'sideaof the idealsituation,by a long shot.He
genuinelylovedthe forecasting work. But betweenthat and the adminis-
trativeBS, he had very little time left overto work on artificiallife. And
artificiallife somehow strucka chordin him thatnothingelsedid. Artificial
life, he says,waswhereyou could getright down into the deepquestions
of emergence andself-organization,questions thathadhauntedhim all his
life.

"I wasalreadythinkingaboutself-organization in naturewhen I wasin

high school,"saysFarmer,"althoughinitially it wason a vaguelevel,from
readingsciencefiction stories."He particularlyremembers one storyby
IsaacAsimov,"The Final Question,"in which humansof the far future
consulta cosmicsupercomputer abouthow to repealthe secondlaw of
thermodynamics: the inexorable tendencyfor everythingin the universeto
cool off, decay,and run downasatomstry to randomizethemselves. How
canwe reverse the increasein entropy,theyask,referringto the physicisb'
namefor molecular-scale disorder.Eventually,long afterthe human race
hasvanishedand all the starshavegrowncold, the computerlearnshow
to accomplishthis feat-whereuponit declares,"Let therebe light!" and
givesriseto a fresh,new,low-entropyuniverse.
Farmerwasfourteenwhenhe readAsimov'sstory,andit seemed to him
eventhen to point the waytowarda profoundquestion'If enhopyis always
increasing,he askedhimself,and if atomic-scale randomness anddisorder
are inexorable,then why is the universestill ableto bring forth starsand
planebandcloudsandhees?Why is matterconstantly becomingmoreand
moreorganized on a large at
scale, the same time that it is becomingmore
and more disorganized on a small scale?Why hasn't everythingin the
universelong since dissolvedinto a formles miasma? "Frankly," says
Farmer,"my interestin those questions wasone of my driving concernsin
becominga physicist. Bill Wootters fphysicistWilliam Wootters, now at
Waitingfor Carnot 287

Williams Collegein Massachusetts] and I spenta lot of time at Stanford

sittingaroundon the lawnafterphysicsclass,talkingaboutthesequestions.
Ideasjust seemedto springinto our heads.It wasonly yearslater that I
discoveredthat other peoplehad thoughtaboutthis, too, and that there
wasa literatureout there-Norbert Wienerandcybernetics, Ilya Prigogine
andself-organization, HermannHakenandsynergetics." In fact,he says,
you couldevenfind the sameissueslatentin the workof HerbertSpencer,
the Englishphilosopher who helpedpopularize Darwin'stheorybackin
the 1860s by coining such phrases as"survival ofthe fiftest,"andwho saw
Darwinian evolution as iust a special case of a broader forcedrivingthe
spontaneous origin of structure in the universe.
So thesewerequestionsthat sprangforth in manyheadsindependently,
saysFarmer.But at the time he felt frustrated: "l couldn'tseea forum for
pursuingthem. Biologists weren't doing it-they weremiredin the nitty-
grittyof which protein interacted with which, missing thegeneralprinciples.
Yetasfar as I couldtell, physicists didn't seem to be doinganythinglike
I
this either.That'sone of the reasons iumped into chaos."
The story of that iump rateda whole chapterin famesGleick'sbest-
sellingbookChaos:how Farmerand his lifelongfriend Norman Packard
becamefascinated with roulettein the late 1970swhile theyweregraduate
studentsin physicsat the Universityof California'scampusin SantaCruz;
how the effortof calculating the movingball'straiectory on the fly, so to
speak,gavethem an exquisitefeel for the waya tiny change of the initial
conditionsin a physicalsystemcan producea dramaticchangein the
outcome;how they and hro other graduatestudents-RobertShawand
)amesCrutchfield-cameto realizethatthissensitivity to initialconditions
could be describedby the emergingscienceof "chaos,"more generally
knownas"dynamicalsystems theory";and how the four of themwereso
determinedto pursueresearch in this field that theybecameknownasthe
DynamicalSystems Collective.
"After a while, though,I got prettyboredwith chaos,"saysFarmer."l
felt 'So what?'The basictheoryhad alreadybeen fleshedout. So there
wasn'tthat excitementof beingon the frontier,wherethingsaren'tunder-
stood."Besides, he says,chaostheoryby itselfdidn'tgofarenough'It told
you a lot about how certainsimplerules of behaviorcould give rise to
astonishingly complicated dynamics. But despite all the beautifulpictures
of fractalsand such,chaostheoryactuallyhad very little to sayaboutthe
fundamentalprinciplesof living systems or of evolution.It didn't explain
how systemsstartingout in a stateof randomnothingness could then
organizethemselves into complexwholes.Mostimportant,it didn't answer
288 COMPLEXITY
his old,questionaboutthe inexorablegrowthof orderand shucturein the
universe.
Somehow,Farmerwasconvinced,therewasa wholenew level of un-
derstandingyet to be reached.Thus his work with Stua* Kauftnan and
Norm Packardon autocatalytic setsand the origin of life, and his enthu-
siasticsupportof l.angton'sartificiallife. Like somanyof the otherpeople
aroundLosAlamosand SantaFe,Farmercouldfeelit-an understanding,
an answer,a principle,a law hoveringalmostwithin reach.
"I'm of the schoolof thoughtthat life andorganization areinexorable,"
he says,"just asinexorableasthe increasein entropy. They iustseemmote
fluky because theyproceedin fit andstarb,andtheybuild on themselves.
Life is a refectionof a much moregeneralphenomenon that I'd like to
believeis describedby somecounterpartof the secondlaw of thermody-
nsrnigs-5srnelaw that woulddescribethe tendencyof matterto organize
itself, and that would predictthe generalpropertiesof organizationwe'd
expectto seein the universe."
Farmerhasno clearideaof whatthisnewsecondlaw wouldlooklike.
"If we knewthat," he says,"we'dhavea big clue how to getthere.At this
somethingthat intuition suggests
point it's purely speculative, when you
standbackandstrokeyourbeardandcontemplate." In fact,he hasno idea
whetherit wouldbe one law, or several.What he doesknow,however,is
that peoplehave recentlybeen finding so many hints about things like
emergence, adaptation,andthe edgeof chaosthattheycanbeginto sketch
at leasta broadoutline of whatthis hypotheticalnewsecondlaw might be
like.

Emergence

First,saysFatmer,thisputativelawwouldhaveto givea rigorousaccount

of emergence: What doesit really meanto saythat the whole is greater
than the sum of its parts?"It's not magic,"he says'"But to us humans,
with our crudelittle humanbrains,itfeelslikemagic."Flyingboids(and
realbirds)adaptto theactionsof their neighbors, therebybecominga fock.
Organismscooperateand eompete in a dance of coevolution,thetebybe-
coming an exquisitelytuned ecosystem. Atoms searchfor a minimum
energystateby formingchemical bonds with each other,therebybecoming
the emergentstructuresknownas molecules. Human beings hy to satisfr
theirmaterialneeds bybuying,selling, and tradingwith each thereby
other,
creatingan emergentstructure known as a market.Humans likewiseinteract
Waitingfor Carnot 289
with eachotherto satisfylessquantifiablegoals,thereby-formingTamilies,

modation and self-consiste -groufC-ofagentsmanageto transcendthem-

selvesand more. The trick is to figure out how, without
into sterilephilosophizing
or NewAgemysticism.
And that, saysFarmer,is the beau$ of computersimulation in general
and artificial life in particular: by experimentingwith a simple model that
you can run on your desktop,you can try out ideasand seehow well they
really work, You can try to pin down vaguenotionswith more and more
precision. And you can hy to extract the essenceof how emergencereally
works in nature. These days, moreover, there is a wide variety of models
to choosefrom. One that Farmerhas given particularattentionto is con-
ne,ctionism:the idea of representinga population of interacting agentili:
ffi;rEi "nodes" linked by "connectionr." And in that, he has plenty of
company. Connectionist models have been popping up everywherein the
past decadeor so. Exhibit A has to be the neural network movement, in
which researchersuse webn of artificial neurons to model such things as
perceptionand memory retrieval-and, not incidentally,to mount a radical
attackon the symbol-processing methodsof mainstreamartificial intelli-
gence. But closebehind are many of the modelsthat havefound a home
at the SantaFe Institute,including |ohn Holland'sclassifiersystems,Stuart
Kauffman's genetic networks, the autocatalyticset model for the origin of
life, and the immune systemmodel that he and Packarddid in the mid-
1980swith [,os Alamos'Alan Perelson.Admittedly, saysFarmer,some of
thesemodelsdon't look very connectionist,and a lot of peopleare surprised
the first time they hear the things being describedthat way. But that's only
becausethe models were createdby different people at different times to
solvedifferentproblems-and then describedin differentlanguage."When
you peel everything back," he says,"they end up looking the same. You
can literally map one model into another."
In a neural network, of course, the node-and-connection structure is
obvious. The nodescorrespondto neurons,and the connectionscorrespond
to synapseslinking the neurons. If a programmer has a neural network
model of vision, for example,he or she can simulatethe patternof light
and dark falling on the retina by activatingcertain input nodes,and then
letting the activation spreadthrough the connections into the rest of the
network. The effect is a bit like sending shiploadsof goodsinto a few port
citiesalong the seacoast, and then lettinga zillion truckscart the stuffalong
the highwaysamong the inland cities. But if the connectionshave been
properly arranged,the network will soon seftleinto a self-consistentpattern
290 COMPLEXITY
ofactivation that corresponds to a classification ofthe scene:"That's a catl"
Moreover, it will do so even if the input data are noisy and incomplete-
or, for that mattet, even if some of its nodes are burned out.
In fohn Holland's classifiersystemthe node-and-connectionstructure is
considerablyless obvious, saysFarmer, but it's there. The set of nodes is
just the set of all possibleinternalmessages, such as l00l00lll0lllll0.
And the connections are iust the classifierrules, each of which looks for
a certain messageon the system'sinternal bulletin board, and then responds
to it by posting another. By activatingcertain input nodes-that is, by
posting the correspondinginput messages on the bulletin board-the pro-
ir"*-., can cause the classifiersto activate more messages,and then still
more. The result will be a cascade of messages analogousto the spreading
activation in a neural network. And, iust as the neural net eventuallysettles
down into a self-consistent state, the classifier system will eventually setde
down into a stable set of active messages and classifiersthat solves the
problem at hand-or, in Holland's picture, that represents an emergent
mental model.
The network structure is also there in the model he did with Kauffman
and Packard on autocatalysisand the origin of life, saysFarmeI. In this
case the set of nodes is the set of all possible polymer species,such as
abbcaad.And the connectionsarethe simulatedchemicalreactionsamong
those polymers: polymer A catalyzesthe formation of polymer B, and so
on. By activating certain input nodes-that is, by seedingthe systemwith
a steadyflow of small "food" polymersfrom the simulated environment-
the three of them could set off a cascadeof reactions, which eventually
settled down into a pattern of active polymers and catalytic reactionsthat
could sustain itself: an "autocatalytic set'' that presumably correspondsto
some sort of protoorganismemerging from the primordial touPi -
The analysis is much the same in Kauffman's network models of the
genome in number of other models, saysFarmer' Underlying them
"nd "r,y
Itt i, tttir same node-and-connectionframework. lndeed, when he first
recognizedthe parallelsseveralyearsago, he wasso delightedthat he wrote
it alfup for puLlication in a paper entitled "A RosettaStone for Connec-
tionism." If nothing else, he says,the very existenceof a common frame-
work is reassuring,in the sensethat most of the blind men at least seem
to have their hands on the sameelephant. But more than that, a common
frameworkshould help the peopleworking on thesemodelsto communicate
a lot more easilythan they usually do, without the babel of different iargons'
.,The thing I considered important in that paPerw:rs that I hammered out
Waitingfor Carnot 291
the actual translationmachinery for going from one model to another. I
could take a model of the immune systemand say, 'If that were a neural
net, here'swhat it would look like.' "
But perhapsthe most important reasonfor having a common fiamework,
saysFarmer, is that it helps you distill out the essenceof the models, so
that you can focus on what they actually have to sayabout emergence.And
in this case,the leson is clear:the power really doeslie in the connections.
That's what getsso many people so excited about connectionism. you can
startwith very, very simple nodes-linear "polymers,""messages" that are
just binary numbers, "neurons" that are essentiallyiust on-off switches-
and still generatesurprising and sophisticatedoutcomes just from the way
they interact.
Thke learning and evolution, for example. Since the nodes are very
simple, the behavior of the networkasa whole is determinedalmost entirely
by the connections. Or in Chris Langton's language,the connectionsen-
code the GTYPE of the network. So to modify the system'spTypE be-
havior, you simply have to changethoseconnections. In fact, saysFarmer,
you can change them in two different ways. The first way is to leave the
connectionsin place but modify their "shength." This correspondsto whaVtrl.#
-
Holland calls exploitafion learning: improving what you already have. Irf\" i -
Holland's own classifier system this is done through the bucket-brigade
ftF
algorithm, which rewardsthe classifierrules that lead to a good resurt. In
neural networks it's done through a variety of learning algorithms, which
present the network with a seriesof known inputs and then tweak the
connection strengthsup or down until it givesthe right responses.
The second, more radical way of adiusting the connectionsis to change
the network'swhole wiring diagram. Rip out some of the old connections
and put in new ones. This correspondsto what Holland callsexploration
learning: taking the risk of screwing up big in retum for the chance of
winning big. In Holland's classifier systems,for example, this is exactly
what happenswhen the genetic algorithm mixes rules togetherthrough ib
inimitable version of sexual recombination; the new rules that resuli will
often link messagesthat have never been linked before. This is also what
happensin the autocatalyticset model when occasionalnew polymers are
allowed to form spontaneously-as they do in the real world. The resulting
chemical connections can give the autocatalyticset an opening to explore
a whole new realm of polymer space.This is nof what usuailylappens in
neural networla, since the connectionswere originally supposedto model
synapsesthat can't be moved. But recently,saysFarmer,a number of neural
 292 COMPLEXITY
network aficionadoshave been experimentingwith networksthat do rewire
themselvesas they learn, on the grounds that any fixed wiring diagram is
arbitraryand ought to be open to change.
So, in short, saysFarmer,the connectionist idea showshow the capacity
for learning and evolution can emergeeven if the nodes,the individual
agents,are brainlessand dead. More generally,by putting the power in the
connections and not the nodes, it points the way toward a very precise
theory of what Langton and the artificial lifers mean when they say that
the essenceof life is in the organization and not the molecules. And it
likewisepointsthe way towarda deeperunderstandingof how life and mind
could have gotten startedin a universethat began with neither'

The Edgeof Chaos

However,saysFarmer,asbeautiful asthat prospectmay be, connectionist

models are a long way from telling you everythingyou'd like to know about
the new secondlaw. To begin with, they don't tell you much about how
emergenceworks in economies, societieS,or ecosystems,where the nodes
are "smart" and constantly adapting to each other. To understandsystems
like that, you have to understandthe coevolutionary dance ofcooperation
and competition. And that means studying them with coevolutionary
modelss,rchas Holland's Echo, which havestartedto get popular only in
the past few years.
More important, saysFarmer, neither connectionist models nor coevo-
lutionary models tell you what makes life and mind possible in the first
pl"ce. Iih"t is it about the universethat allows thesethings to happen?It
isn't enough to say"emergence";the cosmosis full of emergentstructures
like galaxiesand clouds and snowflakesthat are still iust physical obiects;
th.y1"u" no independent life whatsoever.Something more is required.
And this hypothetical new secondlaw will have to tell us what that some-
thing is.
Ciearly, this is a job for models that try to get at the basic physicsand
chemistry of the world, such as the cellular automata that Chris l,angton
is so foni of. And by no coincidence, saysFarmer, l,angton's discovery of
be
this weird, edge-of-chaosphasehansition in cellular automataseemsto
a big part of ihe answer. l.angton kept a_.discreetsilen_ceon this subiect
i
I auriin giuenifEiia-ltifiiilfissertationatthe
theartificiallife confere'nce,
"g,,rt
I ti-.. fromthestart,saysFarmer, a lot of peoplearoundLosAlamos
I
I
Waitingfor Carnot 293
and SantaFe have found the edge-of-chaosidea awfully compelling. Lang-
ton is basically sayingthat the mysterious"something" that makeslife and
mind possibleis a certain kind of balancebetweenthe forcesof order and
the forces of disorder. More precisely,he's sayingthat you should look at
systemsin terms of how they behave instead of how they're made. And
when you do, he says,then what you find are the two extremesof order
and chaos.It'sa lot like the differencebetweensolids,wherethe atomsare
locked into place, and fluids, where the atoms tumble over one another at
random. But right in betweenthe two extremes,he says,at a kind of abstract
phasetransitioncalled"the edgeof chaos,"you also 6nd complexityra class
of behaviorsin which the componentsof the systemneverquite lock into
place, yet neverquite dissolveinto turbulence, either. Theseare the systems
that areboth stableenoughto storeinformation,and yet evanescent enough
to transmit it. These are the systems that can be organized to perform
complex computations, to react to the world, to be spontaneous,adaptive,
and alive.
Strictly speaking,of course,Langton demonstratedthe connectionbe-
tween complexity and phasehansitions only in cellular automata. No one
really knows if it holds true in other models-or in the real world, for that
matter. On the other hand, saysFarmer, there are some strong hints that
it might. With 20/20 hindsight, for example, you can see that phase-\\
transitionlikebehavior has been cropping up in connectionistmodelsfor ll
years.Back in the 1960sit wasone of the 6rst thingsthat StuartKauffman | |
discoveredabout his genetic'networks.If the connectionswere too sparse,
the networks would basically iust freeze up and sit there. And if the
connections were too dense, the networks would churn around in total
chaos. Only right in between, when there were preciselytwo inputs per
node, would the networks produce the stable state cycles Kauftnan was
looking for.
Then, in the mid-1980s,saysFarmer,it was much the samestory with
the autocatalyticset model. The model had a number of parameterssuch
as the catalytic shength of the reactions, and the rate at which "food"
molecules are supplied. He, Packard,and Kauffman had to set all these
parameters by hand, essentiallyby trial and error.And one of the firstthings
they discoveredwas that nothing much happenedin the model until they
got those parametersinto a certain range-whereupon the autocatalytic
setswould take off and develop very quickly. Again, saysFarmer, the be-
havior is shongly reminiscentof a phasehansition-although it's still far
from clear how it relatesto phase transitions in the other models. "One
 294 CAMPLEXITY
senses theanalogies, but it'smoredifficultto makethemprecise,"he says.
"That'sanotherareawheresomebody needsto do somecarefulcrosscom-
parisons,analogous to the RosethStonepaper."
Meanwhile,saysFarmer,it's evenlessclearwhetherthe edge-of-chaos
ideaappliesto coevolutionary systems. When you getto somethinglike an
ecosystem or an economy,he says,it's not obvioushowconcepts like order,
chaos,and complexi$ can evenbe definedvery precisely,much lessa
phasetransitionbetweenthem. Nonetheless, he says,there'ssomething
abouttheedge-of-chaos principlethatstill feelsright.ThketheformerSoviet
Union, he says:"It's now prettyclearthat the totalitarian,centralizedap-
proachto the organizationof societydoesn'tworkverywell." In the long
run, the systemthat Stalinbuilt wasjust too stagnant,too lockedin, too
rigidlycontrolledto survive.Or lookattheBigThreeautomakers in Detroit
in the 1970s.They had grownso big and so rigidly lockedin to certain
waysof doingthingsthattheycouldbarelyrecognizethegrowingchallenge
from fapan,much lessrespondto it.
On the otherhand,saysFarmer,anarchydoesn'tworkverywell, either-
as certainpartsof the former SovietUnion seemeddeterminedto prove
in the aftermathof thebreakup.Nor doesan unfettered laissez-fairesystem:
witnessthe Dickensianhorrors the of Indushial Revolution in England or,
morerecently,thesavings andloandebaclein the United States'Common
sense,not to mention recentpolitical experience,suggesbthat healthy
economiesand healthysocietiesalike haveto keeporder and chaosin
balance-and not just a wishy-washy, average, middle-of-theroadkind of
balance,either.Like a living cell, theyhaveto regulatethemselves with a
denseweb of feedbacks and regulation,at the sametime that they leave
plenty of room for creativity,change,and response to new conditions'
"Evolutionthrivesin systems with a bottom'uporganization,which gives
rise to flexibility," saysFarmer."But at the sametime, evolutionhas to
I channelthe bottom-upapproachin a way that doesn'tdestroythe orga-
I
I nization.Therehasto bea hierarchy of control-with informationfowing
I from the bottom up as well as from the top down." The dynamicsof
complexig at the edgeof chaos,he says,seemsto be idealfor this kind of
behavior.

The Crowth of ComPlexitY

In any case,saysFarmer,"at a vague,heuristiclevelwe think we know

somethingabout the domain wherethis interestingorganizational phe-
Waiting for Carnot 295

nomenonappears."However,this can't be the wholestory,either.Even

if you assume,for the sakeof argument,that this specialedge-of-chaos
domain really exists,the hypotheticalnew secondlaw will still have to
explainhow-qr,nergent get there,how they&eepthemselves
systems there,
' '-"i_'"tii!'|'r'''r*-.+% %.d)

and what thev do there.

At that u"rnffitr., heuristiclevel, saysFarmer,it's easyto persuade

that direction. And if it is on the edgeof chaos,then you'd expectlearning

and evolution to pull it back if it ever startsto drift away. In other words, i
you'd expect learning and evolution to make the edgeof chaosstable, the
natural place for complex, adaptivesystemsto be.
The third question, what systemsdo once they get to the edgeof chaos,
is a bit more subtle. ln the spaceof all possibledynamical behaviors, the
edgeof chaosis like an infinitesimally thin membrane, a region of special,
complex behaviors separatingchaos from order. But then, the surface of
the ocean is only one molecule thick, too; it's iust a boundaryseparating
water from air. And the edgeof chaosregion, like the surfaceof the ocean,
is still vast beyond all imagining. It contains a near-infinity of waysfor an
agentto be both complex and adaptive.Indeed,when John Holland talks
about "perpetual novel$," and adaptiveagentsexploring their way into an
immense spaceof possibilities,he may not sayit this way-but he's talking
aboutadaptiveagentsmoving aroundon this immenseedge-of-chaos mem-
brane.
So, what might the new second law have to say about that? Partly, of
course,it could talk about building blocks,internal models,coevolution,
and all the other adaptationmechanismsthat Holland and others have
studied. But Farmer, for one, suspectsthat at heart it will not be about
mechanism so much as direction: the deceptivelysimple fact that evolution
is constantlycoming up with things that are more complicated,more so-
phisticated, more structured than the ones that came before. "A cloud is
more structured than the initial miasma after the Big Bang," saysFarmer,
"and the prebiotic soup was more structuredthan a cloud." We, in turn,
are more structuredthan the prebiotic soup. And, for that matter, a modern
economy is more structured than those of the Mesopotamian cig-states,
 296 COMPLEXIW
iust asmoderntechnologyis more sophisticated than that of Rome. It seems
that learning and evolution don't just pull agents to the edge of chaos;
f\ slowly, haltingly, but inexorably,learning and evolution move agentsclong
I the edge of chaos in the direction of greaterand greatercomplexity. Why?
"Ifs a thorny question,"saysFarmer."It's very hard to articulatea notion
'progress'in
of biology." Whatdoes it mean for one creatureto be more
advancedthan another?Cockroaches,for example, have been around for
severalhundred million yearslonger than human beings,and they are very,
very good at being cockroaches. Are we more advancedthan they are, or
just different?Were our mammalianancestors of 65 million yearsago more
advanced than Tyrannosdurusrex, or iust luckier in surviving the impact
of a marauding comet?With no obiectivedefinition of fitness,saysFarmer,
"survival of the fittest" becomesa tautology: survival of the survivors.
"But I don't believe in ninnism,'iiiE-er*atntlf iHf ;;ihing-is better
th"n that evolution led inevitably toward
"nffi't
us; that's silly. But if you stand back and hke in the broad sweep of the
entire evolutionaryprocess,I do think you can talk meaningfully about
progress.You see an overall trend toward increasingsophistication, com-
plexity, and functionality; the difference betweena model-T and a Ferrari
is nothing comparedwith the differencebetweenthe earliestorganismsand
the latestorganisms.As elusiveas it is, this overall trend toward increasing
'quality'
of evolutionarydesignis one^ofthe mostfascinatingand profound
to what life is all about." l,r$r^*-.'Y -t,rJ
e*frCg. 'ol W
"iu., ", evolution worls in the iutoca-'
One of his favorite examplesis the way
talytic set model that he did with Packard and Kauffman. One of the
wonderful things about autocatalysisis that you can follow emergencefrom
the ground up, he says.The concentration of a few chemicals gets spon-
taneouslypumped up by ordersof magnitude over their equilibrium con-
centration becausethey can collectively catalyzeeach other's formation.
And that means that the set as a whole is now like a new, emergent
individual stickingup from the equilibrium background-exactlywhat you
want for explaining the origin of life. "lf we knew how to do this in real
chemical experiments, we'd have something poised between living and
nonliving," he says."These autocatalyticindividualsdon't have a genetic
code. And yet on a crude level, they can maintain and propagatethem-
selves-not nearly as well as seeds,for example, but much better than a
pile of rocks."
In the original computer model, of course, there was no evolution of
the sets becausethere was no interaction with any kind of outside envi-
ronment. The model assumedthat everythingwashappeningin one well-
 Waitingfor Carnot 297

stirredpot of chemicals,sooncethe setsemergedtheywerestable.In the

realworld of four billion yearsago,however,the environmentwould have
subiectedthesefuzzily definedautocatalytic individualsto all mannerof
buffetingand fuctuations.So to seewhat would happenin that kind of
situation,Farmerand graduate studentRick Bagleysubiected the model
autocatalytic setsto fuctuationsin their "food" supply:the sheamof small
moleculesthat servedas raw materialfor the seb. "What wasreallycool
wasthat somesetswerelike Pandabears,which canonly digestbamboo,"
saysFarmer."lf you changedtheir food supplythey iust collapsed.But
otherswerelike omnivores; theyhadlotsof metabolicpathways thatallowed
them to substituteone food moleculefor another'So when you played
aroundwith the foodsupplytheywerevirtuallyunchanged," Suchrobost
sets, presumably, would have been the kind that survived on the early
Earth.
More recently,saysFarmer,he, Bagley,andLosAlamospostdocWalter
-- Fontanamadeanothermodificationin the autocatalytic \gk
modelto allowfor
*7 o""asionaJ6[6ilGF6ilE$6ni, whichareknownto happenin realchem- I aft.
ical systemY. Thesespontanous reactionscausedmanyof the autocatalyticIry"a
setsto fall apart.But theones crashed
that pavedthewayfor an evolutionaryI
leap."They triggeredavalanches of novelty," he says'"Certainvariationst
would get amplified,and then would stabilizeagainuntil the nextcrash.
We saw a succession of autocatalyticmetabolisms,each replacingthe
other,"
Maybethat'sa clue,saysFarmer."lt will be interesting to seeif we can
'progress'that would involveemergentshuctures
articulatea notion of
havingcertainfeedbackloops[fot stability]that weren'tpresentin what
went before.The key is that therewould be a sequence of evolutionary
eventsstructuringthe matterin the universein the Spencerian sense,in
which eachemergence setsthe stageandmakesit easierfor the emergence
of the nextlevel."

"Actually,"saysFarmer,"I'm frustratedin talkingaboutall this. There's

a reallanguageproblem.Peoplearethrashingaroundtryingto definethings
like'complexity' and'tendency for emergent I canonlyevoke
computation.'
vagueimagesin your brain with wordsthat aren'tpreciselydefinedin
mathematical terms.It's like the adventof thermodynamics-butwe're
wherethey werein about 1820.They knewtherewassomethingcalled
'heat,' but they weretalkingaboutit in termsthat would later soundri-
diculous."In fact, he says,they weren'tevensurewhatheatwas,much
298 COMPLEXITY
less how it worked. Most reputable scientistsof the day were convinced
that a red-hot poker, say,wasdenselyladen with a weightless,invisible fluid
known ascaloic, which would flow out of the pokerinto cooler, lesscaloric-
rich bodies at the slightestopportunity. Only a minorig thought that heat
might representsome kind of microscopic motion in the poker's atoms.
(The minority was right.) Moreover, no one at the time seemsto have
imaginedthat messy,complicatedthings like steamengines,chemical re-
actions, and electric batteriescould all be governedby simple, generallaws.
It was only in 1824 that a young French engineernamed Sadi Carnot
published the first statementof what would later be known as the second
law of thermodynamics:the fact that heat will not spontaneouslyfow from
cold obiects to hot ones. (Carnot, who was writing a popular book about
steam engines meant for his fellow engineers,quite correctly pointed out
that this simple, everydayfact placedseverelimits on how efficient a steam
engine could be-not to mention internal combustionengines,powerplant
turbines,or any other enginethat runs on heat.The statisticalexplanation
for the second law, that atoms are constantly trying to randomize them-
selves,came some seventyyearslater.)
Likewise, saysFarmer, it was only in the l8'10sthat the English brewer
and amateur scientistfamesfoule laid the experimentalfoundationsfor the
ttrsf law of thgrmodynamics,also known as the conservationof energy:the
:-:_---.--.--
fact that energy can-Fin*ge from one form to aniTterE-enrat-ftechan-
ical, chemical, elechical-but can never be created-or*destpyed.And it
wasonly in the 1850sthat the hvo lawswerestatedin explicit,mathematical
form.
"We're creeping toward that point in self-organization," says Farmer.
"But organization turns out to be a lot harder to understandthan disor-
ganization. We're still missing the key idea-at least in a clear and quan-
titative form. We need something equivalent to the hydrogen atom,
something we can pull apart to get a nice, clear descriptionof what makes
it tick. But we can't do that yet. We only understandlittle pieces of the
ptzzle, each in its own isolatedcontext. For example, we now have a good
understandingof chaosand fractals,which show us how simple systems
with simple parts can generatevery complex behaviors. We know quite a
bit about gene regulation in the fruit fly, Drosophila.In a few very specific
contextswe have some hints as to how self-organizationis achievedin the
'toy
brain. And in artificial life we arecreatinga new repertoireof universes''
Their behavior is a pale reflection of what actually goes on in natural
systems.But we can simulate them completely, we can alter them at
will, and we can understandexactly what makesthem do what they do.
Waitingfor Carnot 299
The hope is that we will eventually be able to stand back and assemble
all these fragments into a comprehensivetheory of evolution and self-
organization.
"Jhis not a field for people who like sharply d,gfuedplqblems," Farmer
th"ttt'iner7'-'l hid in
"aaffie-GtG;t
stone.It's still happening.I don't seeanybodywith a clear path to an
answer.But there are lots of little hints flying around.Lob of little toy
and vagueideas.So it's conceivable
systems to me that in twentyor thirty
yearswe will havea realtheory."

The Arc of a Howitzer Shell

StuartKauftnan, for his part, is devoutlyhopingthat it will takefar less

time than that.
"I've heardDoynesaythatthisislikethermodynamics beforeSadiCarnot
he ;'I he'sright. What we'rereallylookingfor in
camealong," says. think
the science of complexig is the general law of patternformation in non-
equilibrium systems throughout the universe. And we need to inventthe
properconceptsto do it. But with all thesehinb, like the edgeof chaos,
I feel asthoughwe'reon the vergeof a breakthrough, asthoughwe'reiust
a few yearsbeforeCarnot."
Indeed,Kauffmanclearlyhopesthat the new Carnotwill be named-
well, Kauftnan. Like Farmel,he envisionsa newsecondlaw that explains
how emergententitieswill do the mostinterestingthingswhen they'reat
the edgeof chaos,and how adaptation will inexorablybuild theseentities
up into higherand higherlevelsof complexity.But unlike Farmer,Kauff-
man hasn'tbeentied down and frustratedby the bureaucraticnecessities
of runninga research group.He'sbeenthrowinghimselfheadlonginto
the problemalmostsincethe dayhe arrivedat the SantaFe Institute.He
talkslike someonewho needsto find the answer-as thoughthirty years
of tryingto understand themeaningof orderandself-organization hasmade
the closeness of it like a physicalache.
"For me, this ideathat there'san evolutionto the edgeof chaosis just
the next stepin an enormousstruggleto understand the marriageof self-
organization andselection," he says."It'ssoannoying because I canalmost
tasteit, almostseeit. I'm not beinga carefulscientist.Nothing'sfinished.
I'veonly hada firstglanceat a bunchof things.I feelmorelike a howitzer
shellpiercingthroughwall afterwall, leavinga messbehind.I feelthat
I'm rushingthroughtopic after topic, trying to seewherethe end of the
 3OO COMPLEXITY
arcof the howitzershellis, withoutknowinghowto cleanup anythingon
the wayback."
That arc beganbackin the 1960s,saysKauffman,whenhe first started
playingaroundwith autocatalytic setsand his networkmodelsof the gen-
ome. In thosedayshe reallywantedto believethat life wasshapedalmost
entirelyby self-organization, thatnaturalselection wasiusta sideshow. And
nothingshowedthatbetterthanembryonicdevelopment, whereinteracting
genesorganizedthemselves into differentconfigurations corresponding to
differentcell types,andwhereinteracting cellsorganized themselves into
varioustissues andstructures withina developing embryo."l neverdoubted
that natural selectionworked,"he says."It iust seemedto me that the
deepestthingshad to do with self-organization.
"But then one day in the early 1980s,"he says,"l wasvisitingfohn
MaynardSmith," an old friendandan eminentpopulationbiologistat the
Universityof Sussex. This wasat a time whenKauffmanhad iuststarted
thinkingseriously aboutself-organization again,aftera ten-year hiatuswhen
he workedon embryonicdevelopment in Drosophilafruit fies. "Johnand
his wife Sheilaand I wentout for a walkon the Downs,"he says,"and
Johnpointedout thatwe weren'tfar from Darwin'shome.Then he opined
that, by and large,thosewho would takenaturalselectionseriouslywere
Englishcountrygentlemen-likeDarwin.And thenhe lookedat me and
he gaveme his little smile,and he said,'Thosewho thoughtthat natural
selectiondidn't havemuchto do with biologicalevolutionhavebeenurban
|ewsl'Well,it crackedme up. I iustsatdownlaughingin the hedgerow.
But then he said.'You
didn't wantto. I it all to be spontaneous."
But Kauffmanhad to admit that MaynardSmith was right. Self-
organization couldn'tdo it all alone.After all, mutantgenescan self-
organizethemselves justaseasilyasnormalonescan.And whenthe result
is, say,a fruit fly monstrosigwith legswhereits antennae shouldbe, or
no head,thenyoustill neednaturalselection to sortout what'sviablefrom
what'shopeless.
"So I satdown in 1982and I outlinedmy book."(TheOriginsof Order,
a muchrevisedsummaryof Kauffman's thinkingoverthepastthirtyyears,
wasfinally publishedin 1992.)"The bookwasto be aboutself-organization
and selection:How do you put the two together? And the wayI conceived
I of it at the beginning was that there is a a!1ggg!gbelglhq Selection
I may want one thing, but there arelimits to whatthe self-organizing behavior
I of the system will allow. So they tug on one another until they cometo
someequilibriumwhereselection can'tbudgethings.That imagestayed
Waiting for Carnot 301

with me through the 6rst two-thirds of the book"-or more precisely,until

the mid-1980s,when Kauffman arrived in Santa Fe and startedhearing
about the edge of chaos
Ultimately, saysKauftnan, the edge-of-chaosconcept transformed his
point of view on the self-organization-versus-selection questionyet again.
At the time, however,he had decidedlymixed feelingsabout it. Not only
had he been seeing phase-transitionlikebehavior in his genetic networks
since the 1960s,but in 1985he had come very clpseto working out the
edge-of-chaos idea himself A"d 1n(X hrr,1e(l' rc.l,l' {'" r^ltAY.
i'Th"t,"
he sayswith the air of /man still kicking himself, "is one'of ,', .^-. .)
those papersthat I never wrote, and that I've alwaysregretted." The ide^w-'i;J'
u"ul"
had come to him in the summer of 1985, he says,when he was spending
a sabbaticalyear at the Ecole Normale Sup'drieurein Paris.Along with
physicist G6rard Weisbuch and graduate student Frangoise Fogelman-
Soule, who was doing her thesison Kauffman'sgeneticnetworks,he had
gone off from Paris to spend a few months at the HadassahHospital in
ferusalem. And one morning there, Kauftnan got to thinking about what
he called "frozen components" in his nehrorks. He'd first noticed them
back in 1971, he says.In his light bulb analogy,it was as if connected
clustersof nodes here and there in the network would either all light up
or all go dark, and then stay that way while the light bulbs elsewherern
the networkcontinuedto ficker on and off. The frozencomponentsdidn't
appearat all in the denselyconnectednetworks, which were a solid mass
of chaotic flickering. And yet they seemedto dominate the very sparsely
connectednetworks,which is why thosesystemstendedto freezeup entirely.
But what happensin the middle, he wondered?That's whereyou find the
more-or-lesssparselyconnected networks that seem to correspond most
closely to real genetic systems.And that's where the networksare neither
completelyfrozen nor completelychaotic. . ' .
"l remember bursting in on Frangoiseand Gdrard that moming," says
'Look, you guys,
Kauffman. "I said, iust where the frozen componentsare
melting and are tenuously connected to one another, and the isolated
unfrozen islandsare iust beginning to ioin tendrils,you ought to be able
to get the most complex computations!'We talked about it extensivelythat
morning, and we all agreed that was interesting.I noted it down as some-
thing to get to. But-we got off onto other things. Besides,it was still in
'Ah, going to care
that period of my own life when I thought, nobody's
about thesethings.'So I never focused on it again."
The upshot was that Kauffman listened to all this talk about the edgeof
chaos with an odd mix of d6ii vu, regret, and excitement. Here was an
302 OoMPLEXITY
idea he couldn't help feeling a bit proprietary about. And yet he had to
admit that l,angton was the one who made the connection betweenphase
transitions,computation,and life into somethingfar more than a morning's
passingfancy. l,angton had done the hard work of making the idea rigorous
and precise. Moreover, l,angton had recognizedwhat Kauftnan had not:
that the edge of chaos was much more than just a simple boundary sepa-
rating purely ordered systemsfrom purely chaotic systems.Indeed, it was
langton who finally got Kauffman to understandthe point after several
long conversations:the edge of chaoswas a special region unto iself, the
place where you could find systemswith lifelike, complexbehaviors.
So l,angton had clearly done an elegant and important piece of work,
saysKauftnan. Nonetheless,what with his own involvementin economics,
autocatalysis,and all sortsof other projectsat the Santa Fe Institute-not
to mention the time he wasputting in on his book about the tensionbetween
self-organizationand selection-it was severalyean before the full impli-
cations of the edgeof chaos finally hit home, It didn't really happen until
the summer of 1988, in fact, when Norman Packardcame through the
institute on a visit from Illinois and gavea seminarabout his own work
regarding the edge of chaos.
Packard,who had independently hit upon the phasehansition idea at
about the same time l,angton did, had also been giving a good deal of
thought to adaptation. So he couldn't help but wonder: Are the systems
that can adapt the bestalso the ones that can compute the best-the ones
at this funny boundary?It was an appealingthought. So Packardhad done
a simple simulation. Startingwith a lot of cellular automatarules, he had
demanded that each of them perform a certain calculation. He had then
applied a Holland-sgle genetic algorithm to evolve the rules according to
how well they did. And he had discoveredthat the final rules, the ones
that could do the calculation fairly well, did indeed end up clumped at the
boundary. In 1988 Packardhad published his results in a paper entitled
"Adaptation to the Edge of Chaos"-which, as it happened,was the 6rst
time that anyone had actually used the phrase "edge of chaos" in print.
(When Langton was being formal he still said "onsef of chaos.")
Listening to this, Kauffman was thunderstruck. "lt was one of those
moments when I said, 'Of course!' It was almost a shockof recognition. It
had crossedmy mind that you could get complex computation at the phase
transition. But the thought that I hadn't had, which was silly, was that
selection would get you there. The thought just didn't crossmy mind."
Now that it fiad crossedhis mind, however, his old problem of self-
organizationversusnatural selectiontook on a wonderful new clarity. Living
Waitingfor Carnot 303

systemsarc not deeply entrenched in the ordered regime, which was es-
slntially what he'd been sayingfor the pasttwenty-fiveyearswith his claim
that self-organizationis the most powerful force in biology. Living systems
are actually very close to this edge-of-chaosphasetransition, where things
are much looserand more fuid. And natural selectionis nof the antagonist
of self-organization.It's more like a law of motion-a force that is constantly
pushing
- emergent, self-organizingsystemstoward the edgeof chaos'
,.Letis
talk about networksas a model of the geneticregulatory system,"
saysKauffman with the enthusiasmof a convert. "My claim is that sparsely
connected networks in the ordered regime, but not too far from the edge,
do a pretty good iob of fitting lots of featuresabout real embryonic devel-
opment, and real cell types, and real cell differentiation. And insofar as
that's hue, then it's a good guessthat a billion yearsof evolution has in
fact tuned real cell gpes to be nearthe edgeof chaos.So that'svery powerful
evidence that there must be something good about the edge of chaos'
"So let's say the phase transition is the place to be for complex com-
'Mutation
putation," he says."Then the secondassertionis soniethinglike,
and selectionwill get you there."'Packard, of course,had alreadydem-
onstratedthis assertionin his simplecellularautomatamodel. But that was
just one model. And anyway,Kauffman wantedto seeit happen in his own
genetic networks, if only to bolster his argument about evolution bringing
ieal cell typesto the edgeofchaos. So shortly after he heard Packard'stalk,
he worked up a simulation in collaboration with a young Programmer'
Sonkefohnsen, who had iust graduatedfrom the Universityof Pennsylvania.
Following Packard'ssame basic strategy,Kauffman and fohnsen presented
pairs of simulated networkswith a challenge:the "mismatch" game' The
id.r *"r to wire up each networkso that six of their simulatedlight bulbs
were visible to its opponent, and then set them to fashing the light bulbs
at each other in various patterns;the "fittest" network was the one that
could flash a seriesof patterns that were as different as possibleftom its
opponent'spatterns.The mismatchgamecould be adiustedto makeit more
complicated or less complicated for the networks, saysKauffman. The
question was whether selection Pressure,coupled with the genetic algo-
rithm, would be enough to lead the nehvorksto the phasetransition zone
where they were iust on the vergeof going chaotic. And the answerwas
yes in every case,he says.In fact, the answercontinued to be yeswhether
he and fohnsen startedthe networksfrom the orderedregime or the chaotic
regime. Evolution alwaysseemedto lead to the edgeof chaos.
So does that prove the coniecture?Hardly, saysKauffman. A handful
of simulations don't prove anything by themselves."If it turned out to be
304 COMPLEXITY
true for a widevarietyof complicatedgamesthatthe edgeof chaoswasthe
bestplaceto be, andthat mutationandselectiongotyouthere,thenmaybe
the whole loose,wonderfulconiecturenuld be answered,"he says.But
that, he admits,is one of thosepilesof rubblehe hasn'thadtime to clean
up. He'sfelt too manyotherwonderfulconjectures beckoning.

The Danish-bornphysicistPerBakwassomethingof a wild cardin the

edge-of-chaos game.He and his colleagues at long Island'sBrookhaven
Nationall,aboratoryhad firstpublishedtheir ideason "self-organized crit-
icality"in 1987,and Phil Anderson, for one,hadbeenravingabouttheir
workeversince,WhenBakfinallycamethroughLosAlamosandtheSanta
Fe Instituteto talk aboutit in the fall of 1988,he provedto be a rotund
youngman in his mid-thirties,with cherubicfaceanda Teutonicmanner
that rangedfrom brusqueto confrontational."I know what I'm talking
about," he repliedwhen Langtonaskeda questionat one seminar."Do
you know whatyou'retalkingabout?"He wasalsoundeniablybrilliant. His
formulationof thephasetransition ideawasatleastassimpleandaselegant
asl,angton's,and yet so utterlydifferentthat it wassometimes hard to see
how they relatedat all.
Bakexplainsthat he and his coworkers ChaoThngand Kurt Wiesenfeld
discovered self-organized
criticalityin I 986astheywerestudying anesoteric
condensed-matter phenomenonknown ascharge-density waves.But they
quickly recognizedit assomethingmuch moregeneraland far-reaching.
For the bestand mostvivid metaphor, he says,imaginea pile of sandon
a tabletop,with a steadydrizzleof new sandgrainsrainingdown from
above.(This experimenthas actuallybeen done, by the way, both in
computersimulationsand with real sand.)The pile growshigherand
higheruntil it can'tgrowanymore:old sandis cascading downthe sides
and off the edgeof the tableasfastasthe new sanddribblesdown. Con-
versely,saysBak,you could reachexactlythe samestateby startingwith a
hugepile of sand:the sideswould fustcollapse until all the excess sand
had fallen off.
Eitherway,he says,the resultingsandpile is self-organized, in the sense
that it reachesthe steadystateall by itselfwithout anyone explicitlyshaping
it. And it's in a stateof citicality, in the sensethat sandgrainson the
surfaceare iust barelystable.In fact, the critical sandpile is very much
like a criticalmassof plutonium,in whichthechainreactionis iustbarely
on the vergeof running awayinto a nuclearexplosion-but doesn't.The
microscopicsurfaces and edgesof the grainsareinterlockedin everycon-
Waitingfor Carnot 305

ceivablecombination,and are iust readyto giveway. So when a falling

grain hits there'sno telling what might happen.Maybenothing' Maybe
iust a tiny shift in a few grains.Or maybe,if one tiny collisionleadsto
anotherin justthe rightchainreaction,a catastrophic landslide will take
off one whole faceof the sandpile' In fact, saysBak, all thesethingsdo
happenat one time or another.Big avalanches arerare,and smallones
ari lrequent.But the steadilydrizzlingsandtriggers cascades of all sizes-
a fact that manifestsitselfmathematically asthe avalanches' "power-law"
behavior:the average frequency of a given size of avalanche is inversely
proportionalto some power of its size.
Now the point of all this, saysBak,is that powerlawbehavioris very
commonin nature.It's beenseenin the activityof the sun, in the light
from galaxies,in the flow of currentthrougha resistor,and in the flow of
waterthrougha river.Largepulsesarerare'smallonesarecommon,but
all sizesoccurwith thispowerJaw relationship in frequency. The behavior
is so common,in fact, that explaining its ubiquity has become one of the
naggingmysteries of physics:WhY?
fhe-sand pile metaphorsuggesban answer,he says.fust as a steady
trickle of sanddrivesa sandpile to organizeitself into a critical state,a
steadyinput of energyor wateror electronsdrivesa greatmanysystems in
natureto organizethemselves the same way. They become a mass of in-
tricatelyinterlocking subsystems iustbarelyon theedgeof criticality-with
breakdowns of all sizesrippingthroughand rearranging thingsiustoften
enoughto keepthem poisedon the edge.
A prime exampleis the distributionof earthquakes, saysBak' Anyone
who livesin Californiaknowsthat little earthquakes that rattlethe dishes
are far more commonthan the big earthquakes that makeinternational
headlines. In 1956,geologists BenoGutenberg andCharlesRichter(of the
famousRichterscale)pointedout that thesetremorsin factfollow a Power
law:in anygivenarea,the numberof earthquakes eachyearthatrelease a
certainamountof energyis inversely proportional to a certainpowerof the
energy.(Empirically,the poweris about l/2') This soundedlike self-
organized criticalityto Bak.Sohe andChaoThngdid a computersimulation
of a fault zonelike the SanAndreas,wherethe two sidesof the fault are
beingpulled in oppositedirectionsby the steady,inexorablemotionof the
earth'scrust.The standardearthquake modelsaysthat the rockson either
sideare lockedtogetherby enormouspressure and friction;they resistthe
motionuntil suddenly theyslipcatastrophically. In BakandTang'sversion,
however, the rockson eithersidebendanddeformuntil theyareiustready
to slip pasteachother-whereuponthe fault undergoes a steadycascade
 306 COMPLEXITY
of little slipsand biggerslipsthat are just sufficientto keepthe tensionat
thatcriticalpoint. so a powerlawfor earthquakes is exactlywhatyouwould
expect,theyargued;it's justa statement thattheearthhaslongsincetortured
all its fault zonesinto a stateof self-organized criticality.And indeed,their
simulatedearthquakes follow a powerlaw verysimilarto the onefoundby
Gutenbergand Richter.
Soon after that paperwaspublished,saysBak, peoplestartedfinding
evidencefor self-organizing criticalityin all sortsof areas.Fluctuationsin
stockprices,for example,or the vagaries of city haffic. (Stop-and-go haffic
iams correspond to criticalavalanches.) Thereis still no generaltheory
specifying which systems will go to a criticalstateand which won't, ne
admits.But clearlya lot of systems do.
Unfortunately,he adds,self-organizing criticalig only tells you about
the overallstatistics ofavalanches; it tellsyou nothingaboutanyparticular
avalanche.This is yet anothercasewhereunderstanding is not the same
thing as prediction.The scientistswho try to predict earthquakes may
ultimatelysucceed,but not because of self-orga.nized criticality.They'rein
thesamepositionasan imaginary groupof tinyscientists livingon a critical
sandpile. Thesemicroscopic researchers can certainlyperforrna lot of
detailedmeasurements on thesandgrainsin theirimmediateneighborhood,
and-with a tremendous effort-predict whenthoseparticularsandgrains
aregoingto collapse.But knowingthe globalpowerJawbehaviorof the
avalanches doesn'thelp them a bit, becausethe globalbehaviordoesn't
dependon the local details.In fact, it doesn'tevenmakeanydifferenceif
the sandpile scientisbtry to preventthe collapsethey'vepredicted.They
can certainlydo soby puttingup bracesand supportstructures and such.
But they just end up shiftingthe avalanchesomewhere else.The global
powerlaw staysthe same.

"Absolutelywonderful sfuff," declaresKauffman."When Per came

throughthe institute,I fell in lovewith hisself-organized
criticality." I-ang-
ton, Farmer,and all the restof the SantaFecrewfelt muchthe sameway,
despite the prickliness
of the messenger. Here,clearly,wasanothercrucial
pieceofthe edge-of-chaos puzzle.The trickwasto figureout exactlywhere
it fit in.
Self-organized criticalilywasobviouslyon the edgeof something. And
in manyways,that something wasvery much like the phasetransitions
that Langtonwastryingso hardto writeaboutin his dissertation. In the
kind of "second-order" phasetransitionsthat he thoughtwereimportant
Waitingfor Carnot 307
for the edge of chaos, for example, a real substanceshows microscopic
density fluctuationson all size scales;right at the transition, in fact, the
fluctuations follow a power law. In the more abstractsecond-orderphase
transitions that Langton had discoveredin the von Neumann universe,
moreover, Class IV cellular automata such as the Game of Life also show
structuresand fuctuations and "extendedtransients"on all size scales.
In fact, you could even make the analogymathematicallyprecise.Lang-
ton's ordered regime, where systemsalwaysconvergeto a stablestate, was
like a subcriticalpieceof plutonium wherechain reactionsalwaysdie out,
or like a tiny sand pile where the avalanchesnever really get going. His
chaotic regime, where systemsalwaysdivergeinto unpredictablethrashing,
was like a supercriticalpiece of plutonium where the chain reactionruns
awayor like a huge sandpile that collapsesbecauseit can't supportitself.
The edge of chaos, like the stateof self-organizedcriticality, lies right at
the boundary.
However,there were also some puzzling differences.The whole point
of Langton's edge of chaos was that systemsat the edge had the potential
to do complex computationsand show lifelike behaviors.Bak'scritical state
didn't seemto have anything to do with life or computation.(Can earth-
quakescompute?)Furthermore, there wasnothing in l,angton'sformulation
that said systemshad Io be at the edge of chaos;as Packardhad pointed
ogt, they can get there only through someform of natural selection.Bak's
systemsmoved to the critical state spontaneously,driven by the input of
sand, or energy, or whatever. It was (and is) an unsolved problem to un-
derstandpreciselyhow these two phasetransition conceptsfit together.
Kauffman, however, wasn't terribly worried about that. It was clear
enough that the conceptsdid fit together;whatever the details, there was
somethingabout self-organized criticality that felt righf. Betterstill, Bak's
way of looking at thingsclarifiedsomethingthat had beenbugginghim for
a while now. It was one thing to talk about individual agentsbeing on the
edgeof chaos.That's preciselythe dynamical region that allows them to
think and be alive. But what about a collectionof agentstakenas a whole?
The economy, for example: people talk as if it had moods and responses
and passingfevers.Is it at the edgeof chaos?Are ecosystems? Is the immune
system?Is the global community of nations?
Intuitively you'd like to believethat they all are, saysKauffman, if only
to make senseof emergence.Moleculescollectivelymakea living cell, and
the cell is presumablyat the edgeof chaosbecauseit is alive. Cells col-
lectively make an organism,and organismscollectivelymake ecosystems,
et cetera. So arguing by analogy, it seemsreasonableto think that each
 308 COMPLEXITY
new level is "alive" in the samesense-by virtue of being at or very near
the edge of chaos.
But that was iust the problem: Reasonableor not, how could you even
testsuch a notion? Langton had been able to recognizea phasetransition
by watching for cellular automatathat showedmanifestlycomplex behavior
on a computer screen. Yet it was not at all obvious how to do that for
economies or ecosystemsout in the real world. How are you supposedto
tell what'ssimple and what'scomplexwhen you're looking at the behavior
of Wall Street?Preciselywhat doesit mean to saythat global politics or the
Brazilian rain forest is on the edge of chaos?

rif Bak's self-organizedcriticality suggestedan answer,Kauffman realized.

You can tell that a systemis at the critical stateand/or the edge of chaos
if it shows wavesof change and upheaval on all scalesand if the size of
the changesfollows a power law. Of course,that was iust a mathematically
more preciseway of sayingwhat l,angton had been sayingall along: that a
systemcan exhibit complex, lifelike behavior only if it has just the right
balance of stability and fluidity. But a power law was somethingyou could
hope to measure.
To see how it might work, saysKauftnan, imagine a stable ecosystem
or a mature industrial sectorwhere all the agentshave gotten themselves
well adaptedto each other. There is little or no evolutionary pressureto
change. And yet the agentscan't staythere forever,he says,becauseeven-
tually one of the agentsis going to suffer a mutation large enough to knock
him out of equilibrium. Maybethe agingfounderof a firm finally dies and
a new generation takesover with new ideas. Or maybe a random genetic
crossovergives a speciesthe ability to run much faster than before, "So
that agent startschanging," saysKauftnan, "and then he induces changes
in one of his neighbors, and you get an avalancheof changesuntil every-
thing stopschangingagain." But then someoneelsemutates.Indeed, you
can expect the population to be showeredwith a steadyrain of random
mutations, much as Bak's sand pile is showeredwith a steadydrizzle of
sand grains-which means that you can expect any closely interacting
population of agentsto getthemselvesinto a stateof self-organizedcriticality,
with avalanchesof change that follow a power law.
In the fossil record, saysKauftnan, this processwould show up as long
periods of stasisfollowed by rapid bursts of evolutionary change-exactly
the kind of "punctuated equilibrium" that many paleontologists,notably
Stephenf. Gould and Niles Eldridge, claim that they do seein the record.
Thking that idea to ib logical conclusion, moreover,you can argue that
theseavalancheslie behind the greatextinctioneventsin the earth'spast,
Waitingfor Carnot 309
where whole groupsof speciesvanish from the fossilrecordand are replaced
by totally new ones. A falling asteroidor comet may very well have killed
off the dinosaurs some 65 million yearsago; all the evidence points that
way, But most or all of the other greatextinctionsmay have been purely
internal affairs-just bigger-than-usualavalanchesin a global ecosystemat
the edge of chaos. "There's not enough fossil data on extinction eventsto I
persuadeyourself," saysKauffman. "But you can plot what there is to see{
if you find a power law, and you sort of roughly do." Indeed, he made[
such a plot not long after hearing Bak's talk, The graph wasn't a perfect
power law by any means. It was bent over, so that there weren't enough
big avalanchescompared with little avalanches.On the other hand, the
graph didn't have to resemblea power law at all. So the resultsmay not
have been compelling, he says,but given the uncertainties in the data,
they were certainly suggestive.
This tentative successled Kauffman to wonder if powerJaw cascadesof \
changewould be a generalfeatureof "living" systemson the edgeof chaos- /
the stockmarket, interdependentwebsof technology,rain forests,et cetera. (
And while the evidence isn't all in yet, by a long shot, he feels that that
predictionis still quite plausible.In the meantime,however,thinking about
ecosystemsat the edgeof chaoshad deflectedhis attention to another issue:
@w do thgy4et$r_ere?
P;;Eftfb original answer,and his own, had been that systemsget to the
edgeof chaos through adaptation. And Kauffman still believed that this
was basically the right answer. The problem, however, was that when he
and Packardhad actually done their models, they had both demandedthat
their systemsadapt to some arbitrary definition of fitness that they had
imposed from the outside. And yet in real ecosystems,fitnessisn't decreed
from the outside at all. It arisesfrom the dance of coevolution, as each
individual constantly tries to adapt to all the others. This is exactly the
issuethat had led fohn Holland to start his work with the Echo model:
imposing a definition of fitness from the outside is cheating. So the real
question wasn't whether adaptation per se could get you to the edge of
chaos,Kauffman realized.The real questionwaswhether coevolutioncottld
get you there.
To find out-or, at least,to clarify the issuesin his own mind-Kauffman
decidedto do yet another computer simulation, again in collaboration with
SonkeJohnsen.As ecosystemmodelsgo, he admits, the simulationwas a
pretty good connectionistnetwork. (At the heart of the programwasa variant
of his "NK landscape"model, which he had been developingover the past
few yeii-t6-@7-56tter understanding of natural selection, and what it
 310 COMPLEXITY
really meansfor a species'fitnessto dependupon manydifferentgenes.
The namerefersto the fact thateachspecies hasN genes,with the Etness
of each one dependingon K other genes.)the model was even more
abstractthat Holland'sEcho, which wasalreadypretg rarefied.But con-
ceptually,saysKauftnan, it wasprettystraighdorward. you startby imag-
ining an ecosystem wherespecies arefreeto mutateand evolveby natural
selection,but wherethey can interactwith eachotheronly in certain
specifiedways.so thefrogalwayshiesto catchthe ffy with itsstickytongue,
the fox alwayshuntsthe rabbit,and so on. Altematively,you could tf,ink
of the model as an economywhereeach firm is free to organizeitself
howeverit likes internally,but whereib relationships to oth'erfirms are
fixedby a networkof conhactsand regulations.
Either way, saysKauftnan, there is still plenty of room within those
constraints for coevolution.If thefrogevolves a longertongue,for example,
thefly hasto learnhowto makea fastergetaway. If thefly evolvesa chemlcal
to makeitselftasteghastly,the froghasto learnhow to toleratethat taste.
Sohow can you visualizethis?one way,he says,is to look at eachspecies
in turn, startingwith, say,the frog.At anygiveninstant,the frogwiil find
that somestrategies work betterthan others.So at any given instant,the
se-to-fall strategies availableto the frogformsa kind of imaginarylandscape
of "fitness,"with the mostusefulshategies beingat the peaksandthe least
usefulbeingsomewhere downin thevalleys.Asthefrogevorves, moreover,
it movesaroundon this landscape. Everytime it undergoes a mutationit
takesa stepfrom itscurrentstrategy to a newshategy. And naturalselection,
of course,ensuresthat the average motion is alwaysuphiil toward greater
fitness:mutationsthat movethe organismdownhill tendto die out]
It's the samestorywith the fly, the fox, the rabbit,et cetera,sayslGuff-
man. Everyoneis moving aroundon his own landscape.However,the
wholepoint of coevolutionis that theselandscapes are not independent.
They arecoupled.what's a goodstrategy for the frogdepends on whatthe
fly is doing,and viceversa."so aseachagentadapts,it changes the fitness
landscape of all the otheragents,"saysKauftnan. ,.you haveto imagine
this pictureof a frog climbingup hills towarda peakin its strategysplce,
anda fly climbingup hills towarda peakin its space,but the randscapes
aredeformingastheygo." It's asif everyonewerewalkingon rubber.
Now, saysKauftnan:What kind of dynamicsdo you get in sucha sys_
tem?what kind of globalbehaviors do yousee,andhowdo thosebehaviors
relateto oneanother? This iswherethesimulation camein, he says.when
he and|ohnsengottheir NK ecosystem modelup andrunning,theyfound
Waitingfor Carnot 311
exactly the same three regimes that langton had: an ordered regime, a
chaotic regime, and an edge-of-chaoslikephasetransition.
That wasgratifying,saysKauftnan. "lt didn't haveto be, but it was,
In retrospect, however,it'seasyenoughto seewhy."Picturea big ecosystem,
coupled.Well, thereareonly two thingsthat could
with all the landscapes
happen.Eitherall the species keepwalkinguphill andthe landscape keeps
deformingunderthem, so that theykeepmushingaround and never stop
moving. Or a group of neighboringspecies does stop, because they've
reachedwhat fohn Maynard-Smith would call an-evolutionarilystablestrat-
egy." That is, each speciesin the group has gotten so
others that there's no immediate incentive to change.
"Now both processes canoccurin the sameecosystem at the sametime,
depending on theprecise structure
oflandscapes andhowtheyarecouPled,"
saysKauffman."So look at the setof playerswho'vequit movingbecause
they'reat a local optimum. Paintthoseguysred. And paint the others
green."He and Johnsen actuallydid that asa wayof displaying the sim-
ulation on a computerscreen,he says.When the systemis deepin the
chaoticregime,then almostnobodyis standingstill' Sothe displayshows
a seaof green,with iust a few islandsof redtrvinklinginto existence here
and therewherea handful of species manageto 6nd a temporaryequilib-
rium. When the systemis deepin the orderedregime,conversely, then
almosteveryoneis lockedinto an equilibrium'Sothe displayshowsa 6eld
of red, with bits of greensnakingaroundhereand therewhereindividual
speciescan't quite manageto settledown.
When the systemis at the phasetransition,of course,orderand chaos
arein balance.And Ettinglyenough,the displayseemsto pulsewith life.
Red islandsand greenislandsintertwinewith each other, shootingout
tendrils like randomfractals.Partsof the ecosystem are foreverhitting
equilibrium and turning red, while otherpartsare forever twinklingand
turninggreenasthey find new waysto evolve.Waves of change washacross
the screenon all size scales-including the occasional huge wavethat
spontaneously washesacrossthe screenand transformsthe ecosystem be-
yond recognition.
It lookslike punctuatedequilibriumin action,saysKauffman.But as
much fun as it wasto seethe threedynamicalregimesdisplayed in this
the
way-and asgratifyingasit wasto seethat coevolutionary model really
did havean edge-of-chaos phasehansition-that wasonly half the story.
It still didn't explainhow ecosystems couldget to this boundary region.
On the other hand, he says,evenwith all this businessabout rubbery,
 312 COMPLEXITY
deformablefitnesslandscapes, the only thing he'stalkedaboutso far has
beenthe processof mutationin individualgenes.what aboutchangesto
rhestructureof eachspecies'genome-theinternalorganization chait that
tells how one geneinteractswith another?That structureis presumably
just as much a productof evolutionasthe genesthemselves are, he says.
"So you can imaginean evolutionarymetadynamic, a processthat would
tune the internalorganizationof eachagentso that they all resideat the
edgeofchaos."
To test that idea, Kauftnan and fohnsenallowedthe agentsin their
simulationto changetheir internalorganization.This wastantamountto
whatJohnHollandcalls"exploratory learning." It wasalsoverymuchlike
the radicalrewiringthat Farmertalkedaboutin his Rosettastonepaperon
connectionist models.But the upshotwasthatwhenspecies weregiventhe
ability to evolvetheir internalorganization,the ecosystem asa wholedid
indeedmovetowardthe edgeof chaos.
Once again,saysKauftnan, it's easyenoughin rehospectto seewhy.
"lf we'redeepin the orderedregime,"he says,"then everybody is at a peak
in fitnessand we'reall mutuallyconsistent-buttheseare lousyp"rkr."
Everybodyis trappedin the foothills,so to speak,with no way to break
looseand headfor the crestof the range.In termsof humanorganizations,
it's as if the iobsare so subdividedthat no one hasany latitude;all they
can do is learn how to performthe one iob they'vebeenhired for, and
nothing else.whateverthe metaphor,however,it's clearthat if eachin-
dividual in the variousorganizations is alloweda little more freedomto
march to a differentdrummer,then everyonewill benefit.The deeply
frozensystemwill becomea little morefluid, saysKauffman,theaggregate
fitnesswill go up, and the agentswill collectivelymovea bit closer-to1he
edgeof chaos.
Conversely, saysKauffman,"lf we'redeepin the chaoticregirne,then
everytime I changeI screwyou up, and vice versa.We neverget to the
peaks,because you keepkickingme offand I keepkickingyou off, and it's
like sisyphustrying to roll the rock uphill. Therefore,my overallfitness
tendsto be prettylow, andsodoesyours."In organizational terms,it,sas
if the lines of commandin eachfirm are so screwedup that nobodyhas
the slightestideawhatthey'resupposed to do-and half the time they are
workingat cross-purposes anyway.Either way,it obviouslypaysfor indi-
vidual agentsto tightenup their couplingsa bit, so that they can beginto
adaptto what other agentsare doing. The chaoticsystemwill becomea
litde morestable,saysKauffman,the aggregate fitnesswill go up, andonce
again,theecosystem asa wholewill movea bit closerto theedgeof chaos.
Waitingfor Carnot 313
Somewherein between the orderedand chaotic regimes, of course, the
aggregatefitnesshasto reacha maximum. "From the numerical simulations
that we've done," saysKauffman, "it turns out that the maximum fitness
is occurringright at the phasetransition.So the crux is, asif by an invisible
hand, all the playerschangetheir landscape,each to its own advantage,
ff
and the whole systemcoevolvesto the edgeofchaos."
So there it is, saysKauffman: the evidenceconsistsof a sort of power
law in the fossil record suggestingthat the global biosphereis near the edge
of chaos;a couple of computer modelsshowing that systemscan adapttheir
way to the edgeof chaosthrough natural selection;and now one computer
model showing that ecosystemsmay be able to get to the edge of chaos
through coevolution. "So far," he says,"that's the only evidencethat Ilil
know that the edgeof chaos is actually where complex systemsgo in order f ll
to solve a complex task. It's pretty sketchy.So, while I'm absolutelyin love I | |
with this hypothesis-I think it's absolutelyplausibleand credible and
intriguing-l don't know if it's generallytrue.
. "But if this hypothesisis generally true," he says,"Then it's really im-
portant. It would applyto economicsystemsand everythingelse."It would
help us make senseof our world in a way we never were able to before. It
would be a linchpin of this hypothetical new secondlaw And, not inci-
dentally, it would go a long way toward Stuart Kauftnan's thirty-year quest
for a lawful marriageof self-organization and selection.

Finally, saysKauffman, there has to be at leastone further aspectof the

new secondlaw: "There has to be somethingin it about the fundamental
fact that organismshave gotten more complex since life began. We need
to know why organismshave gotten more complex. What's the advantage?"
The only honestanswer,of course,is that nobody knows-yet. "But that
question is behind this whole other strand in my thinking," he says."lt
startswith the origin-of-life-autocatalytic-polymer-setmodel, and goeson
through a theory of complexity and organization that may follow from
that." That theory is still nebulousand tentativein the extreme,he admits.
He can't claim that he's satisfiedwith it. "But it's where my own deepest
hopes for Carnot's whispersremain."
From his own point of view, ironically, the autocatalytic set idea sat
around in limbo for quite a while. By the time that he, Farmer,and packard
had publishedthe origin-of-lifesimulationin 1986,saysKauffman,Farmer
had gone off to do some more work on prediction theory, packard was
helping Stephen Wolfram set up a complex systemsinstitute at the Uni-
 314 COMPLEXITY
versityof lllinois, and he justdidn't feelhe coulddevelopthe modelany
furtherby himself.Quite asidefrom wantingto pursueabouta dozenhot
ideasper dayat the SantaFe Institute,he lackedeitherthe patienceor the
programming skillsto sit in frontof a computerscreendayafterdayhunting
downbugsin a complicated pieceof software.(lndeed,workon the origin-
of-lifemodelresumed againonly in 1987,whenFarmerfounda graduate
student,RichardBagley,whowasinterested in takingit up ashisdissertation
topic. Bagleygreadyimprovedthe simulationby includinga morerealistic.
accountof thermodynamics andseveralotherrefinements, not to mention
speeding up the computercodeabout1000-fold. He received his Ph.D. in
1991.)
The upshotwasthat he did relativelylittle on autocatalysis for aboutfour
years,saysKauffman-not until May 1990, in fact, when he heard a
seminargiven by Walter Fontana,a youngGerman-Italianpostdocwho
had recently.ioinedFarmer'scomplexsystems groupup at Los Alamos.
a-
Fontana started with one of thosecosmic observations that soundso
/deceptivelysimple.When we look at the universeon sizescalesranging
/ from quarksto galaxies,he pointedout, we find the complexphenomena
I associated with life only at the scaleof molecules.Why?
Well, saidFontana,one answeris just to say"chemishy":life is clearly
a chemicalphenomenon, andonly molecules canspontaneously undergo
complexchemicalreactionswith one another.But again,Why?What is
it that allowsmoleculesto do whatquarksand quasars can't?
Two things, he said. The first sourceof chemistry'spower is simple
variety:unlike quarks,which can only combineto makeprotonsand neu-
-Tffi
groupsof three,atomscan be ananged and rearranged to form a
hugenumberof structures. The spaceof molecularpossibilities iseffectively
limitless.The secondsourceof poweris reactivity:structureA can manip
ulatestructureB to form somethingnew-_m;lfre C.
Of course,this definition left out a lot thingslike rateconstantsand
n temperature dependence, which are crucial to understanding real chem-
'\ istry.But
thatwasintentional,saidFontana.Hiscontentionwasthat"chem-
L$lr' it " conceptthatactuallyappliesto a widevarietyof complexffi-
f includingeconomies,
'interact
technologies, andevenminds.(Goodsandservices
with goodsand servicesto producenew goodsand services,ideas
reactwith ideasto producenew ideas,et cetera.)And for that reason,he
said,a computermodelthatdistilledchemishydownto ib purestessence-
varietyand reactivi$-ought to give you a whole new way to studythe
growthof complexityin the world.
To accomplishthat, Fontanawent back to the esence of computer
Waitingfor Carnot 315
programmingitselfto definewhathe calledalgorithmicchemistrSor "AI-
chemy." As fohn von Neumannhadpointedout longago,he said,a piece
of computercodeleadsa doublelife. On the one hand it's a program,a
seriesof commandstellingthe computerwhatto do. But on the otherhand
it's just data,a stringof symbolssittingsomewhere insidethe computer's
memory.So let'susethat fact to definea chemicalreactionbetweentwo
programs,said Fontana:programA simply readsin programB as input
data,and then "executes"to producea stringof outputdata-which the
computernow interpretsas a new program,programC. (Sincethis ob-
viouslywouldn't workvery well with a computerlanguagesuchasFOR-
TRAN or PASCAL, Fontanaactuallywrote his reactiveprogramsin a
variant of the computerlanguageLISP, in which almostany string of
symbolscan represent a valid program.)
Next, saidFontana,takea few zillion of thesesymbol-string programs\
put them in a simulatedpot wheretheycan interactat random,andwatchI
whathappens.In fact, the resultsarenot unlikethoseof the autocatalytic I
model of Kauftnan, Farmer,and Packard,he said,but with weird andJ
wonderfulvariations.Thereareself-sustaining autocatalytic
sets,of course.
But there are alsosetsthat can growwithout bound. Thereare setsthat
can repairthemselves if someof their component"chemicals"aredeleted,
and therearesebthat adaptandchangethemselves if newcomponenbare
injected.There areevenpairsof setsthat haveno membersin cornmon,
but that mutually catalyzeeachother'sexistence.In short, he said, the
Alchemyprogramsuggests of,gureprocess$Fhissymbol-
that.popqlaEqls
stringprograms-areenoughfor the spontaneous emergence of somevery
livelystructures indeed.
'Well,
I wasreallyvery,veryexcitedaboutwhatWalterhaddone,"says
Kauftnan. "I had beenthinkingaboutmy autocatalytic polymerstuff for
a longtime asa modelof economicandtechnological webs,but I couldn't
seemy waypastpolymers.But assoonasI heardWalter,that wasit. He
had figuredit out."
Kauftnan immediatelydecidedto follow Fontana'slead and get back
into the autocatalysis gamein a big way-but with his own twist. Fontana
had identifiedabstractchemistryas a whole new way of thinking about
emergenceand complexig, he realized.But were his resultsa general
propertyof abstractchemishy?Or iust of the way he implementedhis
Alchemyprogram?
It wasthe samequestionKauffmanhadaskedaboutgenomicregulatory
systems in 1963,whenhe 6rstdevisedhis networkmodels."fustasI wanted
to find the genericpropertiesof geneticnetworks,"he says,"I wantedto
 316 coMPLEXtTY
look at genericpropertiesof abstractchemistries.As you tune the complexity
of the chemistry,and other things, such as how much diversityyou have
in the initial set of molecules, what are the generic consequencesfor the
unfolding behavior?"So insteadof following Fontana'sAlchemy approach
directly-it was Fontana's,in any case-Kauffman abstractedthe idea even
f further. He still used symbol stringsto representthe "molecules" of the
I system.But he didn't even insist that they be programs.They could lust
- | be stringsof symbols:ll0l00lll, 10, llllll, et cetera.The "chemistry"
| | of his model was then just a set of rules specifyinghow certain symbol
; I' stringscould transformceitain other symbol strings.And since stringsof
\ symbolsare like words in a language,he called that set of rules-what
else?-a "grammar." (In fact, such grammars of symbol-string transfor-
mationshavebeenextensivelystudiedin the contextof computer languages,
which is where Kauffman got the idea.) The upshot was that he could
sample the kind of behaviorsthat result from various chemistriesby gen-
erating a set of grammar rules at random, and then seeingwhat kinds of
autocatalytic structuresresult.
"Here's the intuition," he says."Start with a pot of symbol strings,and
let them act on one another according to the grammar rules. It might be
the casethat the new shings are alwayslonger than the old strings, so that
you can never make a string you've made before." Call that a "iet": in the
spaceof all possiblestrings, it's a structure that shootsfarther and farther
outwardwithout everlookingback."Or, when you makea cloud of strings,"
he says, "you might start making a string you've made before, but by a
different route. Call that a'mushroom.'Those are my autocatalyticsets;
they're a model of how you can bootstrapyourself into existence. Then
you might get a set of stringsthat collectively make themselvesand nothing
else, just hovering there in string space.Let me call that an'egg.'lt's a
self-reproducingthing, but no single entig in there reproducesitself. Or
you might makewhai I'll call a 'filigreedfog'-you makeall kindsof strings
all over the place, but there are certain stringsyou cannot make, like
ll0ll0ll0. So here'ssomenew kindsof objectsto playwith."
And what doesall this haveto do with the mysterious,inexorablegrowth
of complexig? Maybe a lot, saysKauffman. "The growth of complexi$
reallydoeshavesomethingto do with far-from-equilibriumsystems building
themselvesup, cascadingto higher and higher levels of organization.
Atoms, molecules, autocatalytic sets, et cetera. But the key thing is that
once those higher level entities emerge,they can also interact among them-
selves."A molecule can connectto a moleculeto make a new molecule.
And much the samething happenswith theseemergentobiectsin the string
Waitingfor Carnot 317
world, he says:the same chemistry that createsthem allows them to have
a rich setof interactions,simplyby exchangingstrings."For example,here's
an egg,you throw in a string fiom the outside,and it might turn into a
iet, or change to another egg, or become a filigreed fog. And the samefor
any of theseother obiects."
In any case,saysKauffman, once you havethe interactions,it ought to
be true in general that autocatalysisoccurs whenever the conditions are
right-whether you're talking about moleculesor economies."Once you've
accumulateda sufficient diversity of obiectsat the higher level, you go
through a kind of autocatalytic phase transition-and get an enormous
proliferation ofthings at that level." Theseproliferatingentitiesthen proceed
to interact and produce autocatalyticsetsat a still higher level. "So you get
a hierarchical cascadefrom lower-orderthings to higher-orderthings-
each going through something like this autocatalyticphasetransition."
If that's really true, saysKauffman, then you can begin to see why the
growth of complexityseemsso inexorable:it's just a refection of the same
law of autocatalysisthat (perhaps)was responsiblefor the origin of life. And
surely that has to be a part of the putative new second law That said,
however, he's also convinced that it's not the whole story-for exactly the
samereasonsthat he finally came to realizethat self-organization isn't all
there is to biology. When you think about it, in fact, this upward cascade
oflevels upon levelsis iust anotherkind ofself-organization.So how is the\
cascadeshapedby selection and adaptation? I

ffil;,ftl* j:';",?:
n"3:il.:l"T:,H,.,,:':i'L::r::ffi";;t'i::
me onedayrecently:if you startwith somefoundersetof strings,they may
give rise to autocatalytic setsof shings, they may give rise to autocatalytic
ti_
'" li.]h -
qrt
sets
" - + . squirting
;-+.
"-,,i,+l-- ieb, theymaygiveriseto mushrooms,
+L-,, --., ^i.,- ,:-- eggs,or whatever.But ll .
l^ -,.^L- ^-..,1-^+^.,^- n,.r

decd qtrinor 'dead'

'dead'slrinq obv'laqf
they may
thew nlsn make
mev also meke dead strings. A
A meenc one
string means nne that's
ihqt's inert.
inerl It
Il --"--]
f^{
,t
can't be a catalyst, and :itr doesn't
^^-tr L^ ^ ^^r^1.,-r ^-l -1^^--t^ -^^^!
react -with
--:rL ----^L:--
anything."
Now clearlS he says,if the systemends up *"ki;;'n lot of dead ,tringr,t{/t'
then it's not goingto expandveryfar-rather like an economythat diverts
mostof its outputinto knickknacks that no onewantsto buy andthat can't
be madeinto anythingelse."But if the 'live,'productivestringscansome-
how organizethemselves so they don't makeso manydeadstrings,then
therearemorelive strings." Sothe netproductiviggoesup, andthisgroup\
of live stringshas a sglgg[g advantage over groupsthat don't organizeI
themselves in thisway.And, in fiii, whenyoulookatthecomputermodelsI
you find that the flow of stringsinto deadstringsdoesdecrease as the
simulationproceeds.
318 COMPLEXITY
"I also think that there are refinementsof that idea," he says."suppose
you have two jets coming out of a founder set. The two jets can compete
with one another for strings.But if one jet can learn to help a secondjet
avoid making dead strings,and the secondiet can learn to help the first
jet avoid making dead strings,then you have Lr-rutualists." That pair of
cooperativeiets might then becomethe founffitiirn ffi;"n;w ..*ulti|.f,
structurethat would emergeas a new and more complex individual at a
still higher level. "l havea hunch that higher-orderthingsemergebecause
they can suck more flow of stuff into themselves,faster," saysKauftnan,
"whetherwe'retalkingabout E. coli, prebioticevolution,or firms. So what
I'd like to seeall this lead to is a theory of coupled processes that build
themselvesup into things that comDetefor and yg_lb. f"y-and that get
themselves
to theedge
if
"h"fi;ffi;1il.1i.."
Now admittedly, saysKauffman, none of this is anything more than
intuition so far. "But it feelsright to me. Somehow,the next stepin the
new second law is to understand the natural unfolding of this upward,
billowing cascade.If I can iust show that those entities that happen fastest
and suck the most flow through them are what you see,with some char-
acteristicdistribution. that will be it."

At Home in the Universe

Science is about a great many things, saysDoyne Farmer. It's about the
systematicaccumulation of facts and data. It's about the construction of
logically consistenttheoriesto account for those facts. It's about the dis-
covery of new materials,new pharmaceuticals, and new technologies.
But at heart, he says,scienceis about the telling of stories-storiesthat
explain what the world is like, and how the world came to be as it is. And
like older explanations,such as creation myths, epic legends,and fairy
tales,the storiesthat sciencetells help us understandsomethingabout who
we are as human beings, and how we relate to the universe. There is the
story of how the universeexplodedinto existencesome l5 billion yearsago
at the instant of the Big Bang;the story of how quarks,elechons,neuhinos,
and all the rest came fying out of the Big Bang as an indescribablyhot
plasma;the storyof how thoseparticlesgraduallycondensedinto the matter
we seearound us today in the galaxies,the stars,and the planets;the story
of how the sun is a star like other stars, and how Earth is a planet like
other planets;the story of how life aroseon this Earth and evolved over 4
Waitingfor Carnot 319
billion yearsof geological time;the storyof how the humanspecies first
aroseon theAfricansavannah some3 millionyears agoandslowlyacquired
tools,culture,and language.
And now thereis thestoryof complexity. "I almostviewit asa religious
issue,"saysFarmer."For me asa physicist,asa scientist,my deep-down
motivationhasalways beento understand the universe aroundme. Forme
as a pantheist,natureis Cod. So by understanding natureI get a little
closerto God. Up until I wasin my third yearof gradschool,in fact, I I
neverevendreamed I wouldbeableto geta iobasa scientist. I iustviewedI
it aswhatI wasdoinginsteadof ioininga monastery. I
"Sowhenweaskquestions likehowlife emerges, andwhylivingsystems
are the waythey are-these arethe kind of questionsthat are reallyfun-
damentalto understanding whatwe areandwhatmakesus differentfrom
inanimatematter.The morewe knowaboutthesethings,the closerwe're
goingto getto fundamental questions like, 'What is the purposeof life?'
Now, in sciencewe can neverevenattemptto makea frontalassaulton
questions like that, But by addressing a differentquestion-like,Why is
therean inexorablegrowthin complexity?-wemaybeableto learnsome-
thing fundamentalaboutlife that suggests ib purpose,in the samewaythat
Einsteinshedlight on whatspaceand time areby tryingto understand
gravity.The analogyI think of is avertedvisionin astronomy: if you want I
to seea veryfaint star,you shouldlook a little to the sidebecause your eye I
is more sensitiveto faint light that way-and assoonasyou look right - at 'J
the star,it disappears."
Likewise,saysFarmer,understanding the inexorable growthof com-
plexity isn't goingto give us a full scientifictheoryof morality.But if a
newsecondlawhelpsusunderstand whoandwhatweare,andtheprocesses
that led to us havingbrainsanda socialstructure,then it mighttell us a\
lot moreaboutTloralitvthan we know now. I
"Religionstry to'imposerulesof moralityby writing them on stone
tablets,"he says."We do havea real problemnow, becausewhen we
abandonconventionalreligion,we don't know what rulesto follow any-
more.But whenyou peelit all back,religionand ethicalrulesprovidea \
way of structuringhuman behaviorin a way that allowsa functioning I
society.My feeling is that all of morality operatesat that level. It's an-.|
evolutionaryprocessin which societiesconstandyperformexperiments,
and whetheror not thoseexperimenbsucceeddetermines which cultural
ideasand moralpreceptspropagate into the future." If so, he says,then a
theorythat rigorouslyexplainshow coevolutionary systems are driven to
 320 COMqLEXTTy
the edgeof chaosmight tell us a lot aboutculturaldynamics,and how
societiesreachthat elusive,ever-changing
balancebetweenfreedomand
control.

"l drawa lot of fairly speculative conclusions aboutthe implicationsof

all this," saysChrisl,angton."lt comesfrom viewingtheworldverymuch
throughthesephasetransitionglasses: you can applythe idea to a lot of
thingsand it kind of fits."
witnessthe collapseof communismin the formerSovietUnion and its
EasternEuropeansatellites,he says:the whole situationseemsall too
reminiscent of the power-law distributionof stabiligand upheaval at the
edgeof chaos."When you think of it," he says,"the Cold War wasone
of theselong periodswherenot muchchanged.And althoughwe can find
faultwith theU.s. andsovietgovernments for hotdinga gun to theworrd's
head-the only thing that kept it from blowingup wasMutual Assured
Destruction-there wasa lot of stability.But now that periodof stability
is
, ending.We'veseenupheaval in the Balkans andall overthe place.I'm
lmore scared about what's coming in the immediate future.Because in the
[models, once you get out of one of these metastableperiods, you get into
loneof thesechaoticperiods wherea lot of change happens. The possibilities
for war are much higher-includingthe kind that couldleadto a world
war. It's much moresensitive now to initial conditions.
"So what's the right course of action?" he asks."l don't know, except
,
that this is Iike punctuatedequilibrium in evolutionaryhistory. It doesn't
happenwithout a greatdeal of extinction. And it's not necessarilya step
for the better. There are models where the speciesthat dominate in the
stableperiod after the upheaval may be lessfit than the speciesthat dom-
inated beforehand. So these periods of evolutionary change can be pretty
nastytimes. This is the kind of era when the united statescould disappear
as a world power. Who knows what's going to come out the other end?
"The thing to do is to try to determine whether we can apply this sort
of thing to history-and if so, whether we also seethis kind of punctuated
equilibrium. Things like the fall of Rome. Becausein that case,we really
are part of the evolutionary process.And if we really study that process,
we may be able to incorporatethis thinking into our political, social, and
economic theories, where we realize that we have to be very careful and
put some global agreementsand heaties in place to get us through. But
then the question is, do we want to gain conhol of our own evolution or
not?If so, doesthat stopevolution?It's goodto haveevolutionprogress.
If
Waitingfor Carnot 321
single-celledthings had found a way to stop evolution to maintain them-
selvesasdominant life-forms,then we wouldn't be here. So you don't want
to stop it. On the other hand, maybe you want to understandhow it can
keep going without the massacresand the extinctions.
"So maybe the lessonto be learned is that evolution hasn't stopped,"
saysLangton. "lt's still going on, exhibiting many of the samephenomena
it did in biological history-except that now it's taking place on the social-
cultural plane. And we may be seeinga lot of the samekindsof extinctions
and upheaval."

"I have partial answersto what it all means," saysStuart Kauftnan, who
speaksas a man who's had reasonto be reflective of late. Shortly before
Thanksgiving1991,he and his wife Liz werepassengers in a car crashthat
left them both severelyiniured and could easilyhave been fatal; they spent
months recuperating.
"For example, supposethat these models about the origin of life are
correct. Then life doesn'thang in the balance.It doesn'tdependon whether
some warm little pond iust happensto produce template-replicatingmol-
eculeslike DNA or RNA. Life is the natural exDressionof complex matter.
It's a verydeeppropertyoTifrEfrltry--"r-a-*&Alts-K
andbeingfar from
equilibrium. And that meansthat we'reat home in the universe.We're to
be expected. How welcoming that is! How far that is from the image of
organisms as tinkered-togetherconhaptions, where everything is bits of
widgetrypiled on top of bits of ad hocery,and it's all blind chance.In that
world there are no deep principles in biology, other than random variation
and natural selection;we're not at home in the universe in the same way.
"Next," saysKauftnan, "supposethat you come back many yearslater,
after the autocatalytic sets have been coevolving with one another and
squirting stringsat one another. The things that would still be around would
be thosethings that had come to evolvecompetitiveinteractions,food webs,
muhralism, symbiosis.The things that you would seewould be those that
made the world they now mutually live in. And that reminds us that we
make the world we live in with one another. We're participantsin the story
as it unfolds. We aren't victims and we aren't oubiders. We are part of the
universe, you and me, and the goldfish. We make our world with one
another.
"And now supposeit's really hue that coevolving, complex systemsget
themselvesto the edgeof chaos," he says."Well, that'svery Gaialike. It
saysthat there's an attractor,a statethat we collectively maintiii-iTiSiBlVes
 322 COMPLEXITY

in, an ever-changing statewherespecies arealwaysgoingextinctand new

onesare cominginto existence. Or if we imaginethat this reallycarries
overinto economicsystems, then it's a statewheretechnologies comeinto
I existence and replaceothers,et cetera.But if this is hue, it meansthat the
i edgeof chaosis, on average, the bestthat we can do. The ever-openand
, ever-changing world that we must makefor ourselves is in somesenseas
'
goodasit possiblycan be.
"Well, that'sa storyaboutourselves," saysKauffman."Matterhasman-
agedto evolveasbestit can. And we'reat home in the universe.It'slot -
.p.anglossiaq bqc,ause there'sa lot of pain. You can go extinct,or broke.
But herewei66n-IhilCctfe oTTliaosbecause that'swhere,on average, we
all do the best."

Roasted

In late 1989it finally happened,just asDoyneFarmerhad alwaysbeen

afraidit would.Chrisl,angtonappliedfor an internalgrantfrom LosAlamos
headquarters. And in thecourseof processing thepaperwork, thelaboratory
higher-upsdiscovered that Langtonhadbeena postdoctherefor threefull
yearsand sfill didn't havethe "doc." 'lt hit the fan," saysFarmer."l
rememberit because I wasin Italy at the time on vacation.They somehow
trackedme down in this smalltown on the Ligurian coast,and I had to
makea seriesof phone callswhereI wasplunkingdown thesethousand-
lira coins into a phone that lookedlike it had beenmadeby Alexander
GrahamBell himself.And when I got back,I had to meetin front of the
postdoccommitteeto defendChris-and to defendmyselfashissupervisor.
'How couldthis havehappened?' et cetera.All I could
I reallggotroasted.
do waspoint to the fact that Chris wasthe founderof a wholenew field
calledartificial life. Of course,all that accomplished wasto raisetheir
suspicions evenmore.In the end,because he still hadn'tfinished,we.even
had to askfor a three-monthextensionof his postdocappointment."
Rrmer andDavidCampbell,directorof theCenterfor NonlinearStudies
whereLangtonworked,continuedto besupportive. But therewasno doubt
in their mindsor in Langton'sthat the pressure wason. On top of that, a
secondartificial life conferencehad alreadybeenscheduledfor February
1990.And while langton had someorganizational help this time around
from Farmerand severalothers,the worlcshop wasstill his baby.He had
to getthis damneddissertation out of the way.So he workedlike a fiend.
And in November1989he flew to Ann Arbor, readyto defendthe thing
Waiting for Carnot 923
in frontof hisdissertation
committee, whichwascochaired byfohnHolland
and Art Burls. If they found the work acceptable, theywould awardhim
his Ph.D. on the spotand the agonywouldbe over.
The opinion of the committee,unfortunately, wasunanimous:.,Not \
yet." The basicedge-of-chaos ideais wonderful,theysaid,andyou'vedone \
lots of computerexperiments to backit up. But you'vealsomadesome \
prettysweepingstatements in hereaboutthe Wolframclasses, the emer- I
genceof computation,andsuch,andthe link to the datais prettyslippery. I
So the thing to do is to tone down the statements, make them more j
supportable, and getthem lined up betterwith your data. I
But that meansrewritingthe wholething! saida despairinglangton.
Then you'd bettergetstarted,saidHolland,Burks,and the others.
"This wasa very depressing time," saysl,angton."Here I wasthinking
I wasreadyto defend.But I couldn't.And thenA-LifeII wascomingup
that February. So I had to put it asideagain."
t!tatttaat*aaa;aaa*a;t*ttttlt+ aataa{

ShortlybeforeChristmasof 1989,asBrianArthur drovewestfrom Santa

Fe with a car packedfull of booksand clothesfor his return home to
Stanford,he found himselfstaringshaightinto a spectacular New Mexico
sunsetthat bathedthe-desertin a vastred glow. "I thought, 'This is too
bloodyromanticto be true!' " he laughs.
But appropriate."I had beenat the institutejustabouteighteenmonths
at that time," he says,"and I felt that I neededto go home-to write, and
think, andgetthingsclearin my mind. I wasjustloadeddownwith ideas.
I'd felt that I waslearningat SantaFemorein a month than I wouldhave
in a yearat Stanford.The experience had almostbeentoo rich. And yet
it wasa wrenchto leave.I felt very, very,very sad,in a goodway,and
verynostalgic.The wholescene-the desert,thelight, thesunset-brought
home to me that thoseeighteenrironthsmight well havebeenthe high
point of my scientificlife, and they wereover.That time would not be
easilyrecaptured.I knewotherpeoplewouldcomeand follow up. I knew
I could probablygo back-even go backand run the economicsprogram
againin somefuture years.But I suspected that the institutemight never
be the same,I felt luckyto havebeenin on a goldentime."

The Tao of Complexity

Threeyearslater,sittingin his cornerofficeoverlooking

the tree-shaded
of StanfordUniversig,the DeanandVirginiaMorrisonProfessor
walkways
Work in Progress 325
of PopulationStudiesand Economicsadmitsthat he sfill hasn'tgottenthe
SantaFe experiencecompletelyclear in his mind. "l'm beginningto ap-
preciateit more as time passes,"saysArthur. "But I think the story of what's
been accomplishedin SantaFe is still very much unfolding."
Fundamentally, he says,he's come to realize that the Santa Fe Institute
was and is a catalystfor changesthat would have taken place in any case-
but much more slowly. Certainly that was the case for the economics
program, which continued after his departureunder the ioint directorship
of Minnesota'sDavid L,aneand Yale's)ohn Geanakoplos. "By about 1985,"
saysArthur, "it seemsto me that all sortsof economistswere gefting anby,
starting to look around and sniff the air. They sensedthat the conventional
neoclaisicalframeworkthat had dominatedover the past generationhad
reacheda high water mark. It had allowed them to explorevery thoroughly
the domain of problems that are treatableby static equilibrium analysis.
But it had virtually ignoredthe problemsof process,evolution,and pattern
formation-problems where things were not at equilibrium, where'there's
a lot of happenstance,where history mattersa greatdeal, where adaptation
and evolution might go on forever. Of course,the field had kind of gotten
stymied by that time, becausetheories were not held to be theories in
economicsunlessthey could be fully mathematized,and people only knew
how to do that under conditions of equilibrium' And yet some of the very
best economists were sensingthat there had to be other things going on
and other directionsthat the subiectcould go in.
"What Santa Fe did was to act as a gigantic catalystfor all that. It was
a place where very good people-people of the caliber of Frank Hahn and
Ken Arrow-could come and interact with people like fohn Holland and
Phil Anderson, and over a period of severalvisits there realize, Yes! We
can deal with inductive learning F-thgr-than deductive logic, we can cut
the Gordian kn-ot d;f=Elllifiiti-ni and deil ri,iiliipen:Cnded evolution,
becausemany of theseproblems have been dealt with by other disciplines.
Santa Fe provided the iargon, the metaphors, and the expertisethat you
neededin order to get the techniquesstartedin economics. But more than
that, SantaFe legitimizedthis differentvision of economics'Becausewhen
word got around that people like Arrow and Hahn and Sargentand others
were writing papersof this sort, then it became perfecdy reasonableand
perfectly kosher for others to do so."
Arthur seesevidence for that development every time he goes to an
economicsmeeting thesedays."The peoplewho were interestedin process
and changein the economywere there all along," he says.Indeed, many
of the essentialideas were championed by the great Austrian economist
326 COMPLEXITY

|osephSchumpeterasfar backasthe 1920sand 1930s."But my senseis

that in the pastfour or fiveyears,thepeoplewhothink thiswayhavegotten
much more confident.They aren'tapologeticany moreaboutjust being
ableto givewordy,qualitativedescriptions of economicchange.Nowthey're
armed. They have technique.They form a growingmovementthat is
becomingpartof the neoclassical mainstream everywhere."
That movementhas cerhinly madehis own life easier,notesArthur.
His ideason increasingretums,oncevirtuallyunpublishable, now havea
following. He finds himselfgettinginvitationsto givethis or that distin-
guishedlecturein far-offplaces.In 1989he wasinvitedto write a feature
articleon increasins-reh,rrns economics f.orScientificAmeican. "That was
one of the ligsii-tlHt:ifl-s;'h" r.yile"a thaGifiEl6SGlish;[ in Februrry
l?90-helpei-hi* b""o-e a co-winnerof the IniernationalSfrffiFffi
Society's1990 SchumpeterPrize for the bestresearchon evolutionary
economics.
For Arthur, however,the most gratifyingassessment of the SantaFe
approachcamein September1989,asKenArrowwassummarizinga big,
weekJongworkshopthat had reviewedthe program'sprogress to date.At
the time, ironically,Arthur barelyheardwhat Arrow wassaying.That
noontime,he says,ashe'dheadedout the front doorof the conventon his
wayto lunch, he'dmanaged to trip andsprainhisankleterribly.He'dspent
that wholeafternoonin the convent'schapel-furned-conference room lis-
teningto the closingsession of the workshopthrougha hazeof pain, with
his foot carefullywrappedby Dr. Kauffmanand proppedup with a bagof
ice on the chair in front of him. In fact, the full impactof Arrow'swords
only hit him a few dayslater,after he'd defiedall adviceof doctors,col-
leagues,and wife and hobbledoffto a long-planned conference in lrkutsk,
on the shoresof LakeBaikalin Siberia.
"It wasone of theseflashesof extremeclarity you get at threein the
morning," he says."The Aeroflotjet wasjust coming into lrkutsk, and
therewasthis guy riding a bicycledown the runway,wavinga light stick
to showus whereto taxi. And whenI thoughtaboutwhatArrow had said
in his closingsummary,it finally shuckhome. He said,'I think we can
safelysaywe haveanothertypeof economics here.Onetypeis the standard
stuffthat we'reall familiarwith'-he wastoo modestto call it the Arrow-
Debreusystem,but hebasicallymeanttheneoclassical, generalequilibrium
theory-'and then this other gpe, the SantaFe-styleevolutionaryeco-
nomics.'He madeit clearthat,to his mind, whatthe programhad dem-
onshatedin a yearwasthat this wasanothervalid wayto do economics,
equal in statusto the traditionaltheory.It wasn'tthat the standardfor-
Work in Progress 327
mulation was wrong, he said, but that we were exploring into a new way
of looking at parts of the economy that are not amenable to conventional
methods. So this new approachwas complementary to the standardones.
He also said that we didn't know where this new sort of economics was
taking us. [t was the beginnings of a researchprogram. But he found it
very interesting and exciting.
';That pleasedme enormously,"saysArthur. "But Arrow said a second
thing also. He comparedthe SantaFe program of researchwith the Cowles
Foundationprogramthat he had been associated with in the early 1950s.
And he said that the Santa Fe approach seemedto be much more accepted
at this stage, given that it's now at most two yearsold, than the Cowles
Foundation group had been at the samepoint. Well, I was amazedto hear
that, and tremendously flattered. Becausethe Cowles Foundation people
were the Young Turks of their day-Arrow, Koopmans, Debreu, Klein,
Hurwicz, et ceteta. Four of them got Nobel Prizes,with maybea few more
to come. They were the people who mathematizedeconomics.They were
the people who had setthe agendafor the following generations.They were
the peoplewho had actually revolutionizedthe field."

From the Santa Fe Institute's point of view, of course' this effort to

catalyzea sea change in economics is only a part of its effort to catalyze
the complexity revolution in scienceas a whole. That quest may yet prove
quixotic, saysArthur. But nonetheless,he'sconvincedthat GeorgeCowan,
Murray Cell-Mann, and the others have gotten hold of exactly the right
set of issues.
"Nonscientiststend to think that scienceworksby deduction," he says.\
"But actually scienceworks mainly by metaphor. And what's happening isJ
that the kinds of metaphorpeoplehave in mind are changing'" To put it
in perspective,he says,think of what happenedto our view of the world
with the adventof Sir IsaacNewton. "Beforethe seventeenth century," he
"it
says, was a world of trees, human
disease, psyche, and human behavior'
It wasmessy and organic. The heavenswere alsocomplex. The traiectories
of the planets seemedarbitrary. Trying to figure out what was going on in
the world wasa matterof art. But then along comesNewton in the 1660s'
He devisesa few laws, he devisesthe differential calculus-and suddenly
the planets ale seento be moving in simple, predictableorbi*l
"This had an incredibly profound effect on people'spsyche,right up to
the present," saysArthur. "The heavens-the habitat of God-had been
explained, and you didn't need angelsto push things around anymore. You
 COMPLEXIW
didn't needGod to hold thingsin place.So in the absence of God, the
agebecamemoresecular.And yet, in the faceof snakes and earthquakes,
stormsandplagues,therewasstill a profoundneedto knowthat something
hadit all undercontrol.Soin the Enlightenment,which lastedfrom about
1680all throughthe 1700s,the era shiftedto a belief in the primacyof
nature:if you just left thingsalone,naturewouldseeto it that everything
workedout for the commongood."
The metaphorof the age,salg;fdlur, becamethe clockworkmotionof
the planets:a simple,regular,predictableNewtonianmachinethat would
run of itself. And the model for the next two and a half centuriesof
-g!l!g&uig sciencebecameNewtonianphysics."Reductionistscience
tendsto say,'Hey, the world out there is complicatedand a mess-but
looktTwo or threelawsreduceit all to an incrediblysimplesystem!'
"So all that remainedwasfor AdamSmith, at the heightof the Scottish
EnlightenmentaroundEdinburgh,to understand the machinebehindthe
economy,"saysArthur. "In 1776,inTheWealthof Nations,he madethe
casethat if you left peoplealoneto pursuetheir individualinteresb,the
'lnvisible Hand' of supply
and demandwould seeto it that everything
workedout for the commongod." Obviously,thiswasnot thewholestory:
Smith himselfpointedto suchnaggingproblemsasworkeralienationand
exploitation.But there was so much about his Newtonianview of the
economythat wassimpleand powerful and ight that it has dominated
Westerneconomicthoughteversince."Smith'sideawassobrilliantthat
/ it just dazzledus," saysArthur. "Once, longago,the economistKenneth
'What would you like to do in economics?'
/ Bouldingaskedme, Being
I young and brash, I said veryimmodestly,'I wantto bring economicsinto
'Don't you think you
I the twentiethcentury.'He lookedat me and said,
I shouldbring it into the eighteenthcenturyfirst?'"
In fact, saysArthur, he feelsthat economicsin the twentiethcentury
haslaggedabouta generation behinda certainlossofinnocencein all the
sciences. As the centurybegan,for example,philosophers suchasRussell,
Whitehead,Frege,and Wittgensteinset out to demonshatethat all of
mathematicscould be foundedon simplelogic. They werepartly right.
Much of it canbe.But notall:in the 1930s, themathematician Kurt G0del
showedthat evensomeverysimplemathematical systems-arithmetic,for
example-are inherentlyincomplete.Theyalwayscontainstatements that
cannotbe provedhue or falsewithin the system,evenin principle. At
about the sametime (and by usingessentially the sameargument),the
logicianAlan Turingshowedthatevenverysimplecomputerprograms can
be undecidable: you can'ttell in advancewhetherthe computerwill reach
Work in Progress 329

an answeror not. In the 1960sand 1970s,physicists got much the same

message from chaos theory: even very simple equations canproduceresults
that aresurprisingand essentially unpredictable. Indeed, saysArthur, that
message hasbeenrepeatedin field after field. "People realized that logic
and philosophyare messy,that languageis messy, that chemical kinetics
is missy,that physicsis messy,and finally that the economy is naturally
messy.And it's not that this is a messcreatedby the dirt that'son the
microscope glass.It's that this messis inherentin the systems themselves.
You can't captureany of them and confinethem to a neat box of logic"'
The result, saysArthur, hasbeenthe revolution in complexity' "ln a
senseit's the oppositeof reductionism.The complexity revolution began
'Hey, I can startwith this amazinglysimple
the 6rst time someonesaid,
system,and look-it givesriseto theseimmenselycomplicatedand un-
predictable ronr.qu.n".r.' " Instead of relyingon theNewionianmetaphor
of clockworkpredictability,complexityseemsto be basedon metaphors
moreclosely to thegrowthof a plantfrom a tiny seed,or the unfolding
"kin
of a computerprogramfrorn a few lines of code, or perhapoeven the
organic,self-organized flockingof simpleminded birds.That'scertainlythe
kina of metaphorthat Chris Langtonhasin mind with artificiallife: his
wholepoint is that complex,lifelikebehavioris the resultof simplerules
unfoldingfrom the bottomup. And it's likewisethe kind of metaphorthat
infuencedArthur in the santaFeeconomics program:"If I hada purpose, i
or a vision, it was to show that the messiness and the livelinessin thel
economycangrowout of an incrediblysimple,eveneleganttheory.That's\
why we createdthesesimplemodelsof the stockmarketwherethe market
appearsmoody, showscrashes,takesoff in unexpecteddirections,and
acquiressomethingthat you could describeasa personality."
While he wasactuallyat the institute,ironically,Arthur hadalmostno
time at all for Chris Langton'sartificiallife, or the edgeof chaos,or the
hypothetical newsecondlaw.The economics Program wastakingup ll0
peicentof his workdayas it was. But what he did hear he foundfascinating.
it reemedto him thatartificial life and the restcaptured something essential
about the spirit of the institute. "Martin Heidegger once said that the
fundamental philosophical question is being," notes Arthur. "What arewe
doinghereas conscious entities? why isn't the universe iust a turbulent
messof particlestumbling aroundeachother?Why are therestructure,
form, and pattern?Why is consciousness possibleat all?" Veryfew people
attheinstituteweregrapplingwith that problem quiteasdirectlyasLangton,
Kauffman,and Farmerwere.But one in way or another,saysArthur, he
sensedthat everyonewasworkingon a pieceof it.
 330 COMPLEXITY
Furthermore,he felt that the ideasresonatedshonglywith what he and
his coconspirators
weretryingto accomplishin economics, When you look
atthe subiectthroughChrislangton'sphasehansitionglasses, for example,
all of neoclassical
economics is suddenlytransformed into a simpleassertion
thatthe economyis deepin the orderedregime,wherethe marketis always
in equilibriumand thingschangeslowlyif at all. The SantaFe approach
is likewisetransformedinto a simpleassertion that the economyis at the
edgeofchaos,whereagentsareconstantly adaptingto eachotherandthings
are alwaysin flux. Arthur alwaysknew which assertionfte thought was
more realistic.

Like otherSantaFefolk, Arthur is hesitantwhenit comesto speculating

aboutthe largermeaningof all this. The resultsarestill so-embryonic.
And it's entirelytoo easyto comeoff soundingNew Age and flaky.But
like everyoneelse,he can't help thinkingaboutthe largermeaning.
You can look at the complexityrevolutionin almosttheologicalterms,
I he says."The Newtonianclockworkmetaphoris akin to standardProtes-
vte-glituJa-d6littfr;;t.bil;ln-ihe universe.*liTiof ifiit we'relyonGod
rfororder.That'sa liftle too Catholic.It's that God hasarrangedthe world
bo that the order is naturallythere if we behaveourselves.If we act as
lindividualsin our own right, if we pursueour own righteous self-interest
i andworkhard,anddon'tbotherotherpeople,thenthe naturalequilibrium
: of the world will assertitself.Then we getthe bestof all possibleworlds-
i the one we deserve.That's probablynot quite theological,but jt's the
i impressionI haveof one brandof Christianity.

, matterof patternsthat change,that partlyrepeat,but neverquite repeat,

1,thatarealwaysnew and different.
'
"What is our relationto a world like that?Well, we are madeof the
sameelementalcompositions. So we area partof thisthingthat is never
changingand alwayschanging.If you think that you'rea steamboat and
cango up the river,you'rekiddingyourself.Actually,you'reiustthecaptain
of a paperboatdriftingdownthe river.If you try to resist,you'renot going
to getanywhere. On theotherhand,if youquietlyobserve theflow,realizing
Work in Progress 331

that you'repart of it, realizingthat the flow is ever-changing and always

leadingto new complexities, then everyso oftenyou can stickan oar into
the river and punt yourselffrom one eddyto another.
"So what'sthe connectionwith economicandpoliticalpolicy?Well, in
a policycontext,it meansthatyou observe,andobnerve,.and observe,and
occasionallystick your oar in and improve something for the better.It
meansthat you try to seereality for what it is, and realize that the game
you are in keepschanging, so that it's up to you to figure out the current
iulesof the gameasit'sbeingplayed.It meansthatyouobserve thefapanese
like hawla, you stop being naive, you stop appealing for them to playfair,
you stop adheringto standard theories that are built on outmoded as-
'Well, if only we could
sumptionsaboutthe rulesof play, you stop saying,
,.""[, thi, equilibriumwe'd be in fat city.' You iust observe.And where
you can makean effectivemove,you makea move."
Notice that this is nof a recipefor passivi$,or fatalism,saysArthur. ;
,.This is a powerful approachih"t -"k., use of the natural nonlinear
\
',
dynamicsof the system.You applyavailableforceto the maximum effect.
You don't wasteii. This is exactlythe differencebetweenWestmoreland's:
approachin south vietnam versusthe North Vietnamese approach.west- .
morelandwouldgoinwithheavyforcesandartilleryandbarbedwirean :
burn the villages.lnd the North Vietnamese wouldiustrecedelike a tide.
Then threedlys hter they'dbe back,and no one knewwherethey came
from. It's alsothe principlethat lies behindall of Oriental martial arts.
You don't try to stopyour opponent,you let him comeat you-and then
give him a tap in iust the right directionashe rushesby. The idea is to;
obr.ru", to act courageouslS andto pickyourtimingextremely well.''
Arthur is reluctantto getinto the implicationsof all thisfor policyissues.
But he doesrememberone smallworkshopthat Murray Gell-Mann per-
suadedhim to cochairin thefall of 1989,shortlybeforehe left the institute.
The purposeof the workshoPwasto look at whatcomplexitymight have
to sayaboutthe interplayof economics,environmental values,andpublic
policyin a regionsuchasAmazonia,wherethe rain forestis beingcleared
lor roadsand farmsat an alarmingrate.The answerArthur gaveduring
his own talk wasthat you can approachpolicy-makingfor the rain forest\
(or for any othersubiect)on threedifferentlevels. I
The first level,he says,is the conventional goj!-beq-e5!-eplroagh: What
are the costsof eachspecificcoiilfiFiltonl!fi"tite the benefits,and
how do you achievethe optimumbalancebetweenthe two?"There is a
placefor that kind of science,"saysArthur. "lt doesforceyou to think
throughthe implicationsof the alternatives. And certainlyat that meeting
332 COMPLEXITY
we had a numberof peoplearguingthe costsand benefitsof rain forests.
The troubleis that this approachgenerallyassumes that the problemsare
well defined,thatthe optionsarewell defined,andthatthe politicalwhere-
withal is there,so that the analyst'sjob is simplyto put numberson the
costsandbenefitsof eachalternative.It'sasthoughtheworldwerea railroad
switchyard:We're goingdown this one track,and we haveswitcheswe
cantum to guidethehain ontoothertracks."Unfortunately for thestandard
theory,however,the real world is almostneverthat well defined-partic-
ularly when it comesto environmentalissues.All too often, the apparent
objectivityof cost-benefitanalysesis theresultof slappingarbitrarynumbers
on subjectivejudgments,andthenassigning the valueof zeroto the things
that nobodyknowshow to evaluate."I ridiculesomeof thesecost-benefit
analyses in my classes,"he says."The 'bencit' of havingspottedowls is
definedin termsof how manypeoplevisit the forest,how manywill see
a spottedowl, and what'sit worth to them to seea spottedowl, et cetera.
It'sall thegreatestrubbish.Thistypeof environmental cost-benefit analysis
makesit seemasthoughwe'rein frontof theshopwindowof naturelooking
in, and saying,'Yes,we wantthis, or this,or this'-but we'renot inside,
we'renot partof it. So thesestudieshaveneverappealed to me. By askingll
only what is goodfor human beings,they are beingpresumptuous and |l
arrogant." / I
The second levelof policy-making isa full institutional-politicalanalysis,
saysArthur: figuring out who'sdoing what,?i?t'Wtfl"lTrG6-yilr-it"if iJ
do that for, say,the Brazilianrain forest,you 6nd that thereare various
players: landowners, settlers,squatters,politicians,rural police,roadbuild-
ers,indigenous peoples. Theyaren'tout to getthe environment, but they
are all playingthis elaborate,interactiveMonopolygame,in which the
environmentis beingdeeplyaffected.Moreover,the politicalsystemisn't
someexogenous thing that standsoutsidethe game.The politicalsystem
is actuallyan outcomeof the game-the alliancesandcoalitionsthat form
asa resultof it."
In short,saysArthur, you lookat the systern asa system, the waya Thoist
in his paperboat would observethe complex,ever-changing river. Of
course,a historianor a politicalscientist wouldlook at the situationthis
wayinstinctively.And somebeautifulstudiesin economicshaverecently
startedto takethis approach.But at the time of the workshopin 1989,he
says,the ideastill seemedto be a revelationto manyeconomists. "ln my
talk I put in a strongpleafor this kind of analysis,"he says."lf you really
want to get deeplyinto an environmentalissue,I told them, you haveto
askthesequestionsof who haswhat at stake,what alliancesare likely to
Work in Progress 333

form, and basicallyunderstand the sihration.Then you might find certain

pointsat which interventionmaybe possible.
"So all of that is leadingup to the.thfuillerl of analysis,"saysArthur. t
"At this levelwe might look at whattwo differentworld viewshaveto say$--.
']K
aboutenvironmentaiissues.One of theseis the standard equilibriumuiew-
4'|
point that we'veinheritedfrom the Enlightenment-the ideathat there's
a dualitybetweenman and nature,and lhat there'sa naturalequilibrium2al
betweenthem that'soptimalfor man. And if you believethis view, then
you can talk about'the optimizationof policydecisionsconcerningenvi-
ronmentalresources,' which wasa phraseI got from one of the earlier
speakers at the workshop.
"The otherviewpointis complexity,in whichthereisbasicallyno duality
betweenman and nature,"saysArthur. "We are part of natureourselves.
We'rein the middleof it. There'sno divisionbetween doersanddone-to
because we areall part of this interlocking network. If we, ashumans,hy
to takeaction in our favor without knowing how the overallsystemwill
adapt-like choppingdown the rain forest-we set in motion a train of
eventsthat will likely come back and form a different pattern for us to
adjustto, like globalclimatechange.
"So once you drop the duality," he says,"then the questionschange.
You can'tthen talk aboutoptimization, because it becomes meaningless.
'us
It would be like parentstrying to optimizetheir behavior termsofin
versusthe kids,' which is a strangepoint of view if you seeyourselfas a
family. You haveto talk aboutaccommodation and coadaptation-what
would be goodfor the family asa whole.
"Basically,what I'm sayingis not at all new to Easternphilosophy'Ifs
neverseenthe worldasanythingelsebut a complexsystem.But it's a world
view that, decadeby dicade,is becomingmoreimportantin the West-
both in scienceand in the cultureat large.Very,veryslowly,there'sbeen
a gradualshift from an exploitativeview of nature-man versusnature-
to an approachthatstresses the mutualaccommodation of manandnature.
What has happenedis that we'rebeginningto loseour innocence,or
naivet6,abouthow the world works.As we beginto understand complex
systems, we beginto understandthat we'repart of an ever-changing, in-
terlocking,nonlinear,kaleidoscopic world.
"So the questionis how you maneuverin a world like that. And the
answeris that you want to keepasmanyoptionsopenaspossible.You go
for viability,something that'sworkable,ratherthanwhat's'optimal.'Alot
'Aren't you then acceptingsecondbest?'No, you're
of peoplesayto that,
not, becauseoptimizationisn't well definedanymore.What you'retrying
 334 COM4LEX1W
to do is maximizerobustness, or survivability,in the faceof an ill-defined
fuhrre.And that, in turn, putsa premiumon becomingawareof nonlinear
relationships and causalpathways as bestwe can. You observethe world
very,very carefully,and you don't expectcircumstances to last."
so what is the roler-f the santaFe Institutein all thiql_certainlynot to
l.-"91qi.4s"ffi.rt$!*E k, says
Arthur, th.r;
to be a few peoplewho expecti[ to. No, he says, "Ithoujh "fi;t;r;.rn
the institute,s
role is to
ohelg_us look at this ever-changingriverand understand whatwe'reseeing.
"lf you havea truly complexsystem,"he says,..thenthe exactpatterns
f
are not repeatable.And yet there are themesthat are recognizable.In
/
I hrstory'tor example,you can talk about 'revolutions,'eventhough one
revolutionmight be quitedifferentfrom another.Sowe assignmetaphors.
It turns out that an awful lot of policy-makinghasto do with finding the
appropriatemetaphor.Conversely,bad policy-rrtaking almostalwaysin_
volvesfinding inappropriate metaphors. Forexample,it maynot be appro_
priateto think abouta drug'war,'withgunsandassaults.
"So from this point of view, the purposeof havinga santaFe Institute
is that it, and placeslike it, arewherethe metaphors and a vocabularyare
being createdin complexsystems.So if somebodycomesarongwith a
beautifulshrdyon the computer,thenyou cansay'Here's n.* *-.t phor.
[et's call this one theedgeof chaos,'orwhatever.Sowhat"the SFI wiil do,
if it sh',diesenoughcomplexsystems,is to showus the kindsof patterns
we might observe,andthe kindsof metaphorthatmight be fo,
systems that are movingand in process "ppropiirt
and complicated,rather than the
metaphorof clockwork.
"So I would arguethat a wiseuseof the SFI is_,krlefit*d-q_science,', he
says."To makeit into a policy shopwould be a greatmistake.It would
cheapenthe wholeaffair.And in the end it would be counterproductive,
because whatwe'remissingat the momentis anyprecise understanding of
how complexsystems operate.This is the next maior taskin sciencefor
the next 50 to 100years."

"l think there'sa personality that goeswith this kind of thing," Arthur
says."It's peoplewho like process and pattern,asopposedto peoplewho
are comfortablewith stasisand order.I know that everytime in my life
thatI've run acrosssimplerulesgivingriseto emergent,complex*.rrin.rr,
I've just said,'Ah, isn't that lovelyl'And I think that sometimes, when
otherpeoplerun across it, theyrecoil."
In about1980,he says,at a time whenhe wasstill strugglingto articulate
Work in Progress 33s
his own vision of a dynamic,evolvingeconomy'he happenedto reada
bookby the geneticistBisbad-kwonJin. And he wasstruckby a passage
in which Lewontinsaidthat scientists cgglp.iLhua"rypgs. Scientists of the
first type seethe world as being basicallyin equilibrium. And if untidy
forcessometimes pusha systemslightlyout of equilibrium,then they feel
the whole trick is to push it backagain.Lewontincalledthesescientists
"Platonists,"after the renownedAthenianphilosopherwho declaredthat
ffiefiessy, imperfectobiectswe seearoundus are merelythe reflections
of perfect"archetypes. "
ofthe secondtype,however,seethe worldasa process
Scientists of flow
and change,with the samematerialconstantly going around and-around
in endlesscombinations.Lewontincalledthesescientists "H%SlSegq,f
after the Ionian philosopherwho passionately and poeticallyarguedthat
the world is in a constantstateof fux. Heraclitus,who lived nearly a
centurybeforePlato,is famousfor observingthat "Upon thosewho step
into the sameriversflow otherandyetotherwaters,"a statement thatPlato
himselfparaphrased as"You can neverstepinto the sameriver twice'"
"When I readwhat l,ewontinsaid,"saysArthur, "it wasa momentof
revelation.That'swhen it finally becameclearto me whatwasgoingon.
'YeslWe'refinallybeginningto recoverfromNewton'' "
I thoughtto myseli

The Hair Shirt

Meanwhile,at aboutthe sametime that Brian Arthur wasdriving off

into the sunset,the Heraclitian-in-chief backin SantaFewasgettingready
to call it quits. For all the undeniable success of the economicsprogram,
and for all the intellectual ferment over the edge of chaos,artificiallife,
andtherest,GeorgeCowan was acutely aware that the institute'spermanent
endowmentfund still stood at zero. And after six years, he wastired of
constantlybegging people for operating cash. He was tired of frettingover
the economics program,Iestit become the 800-pound gorilla thattookover
theinstitute.And speaking of 800-pound gorillas,he was tiredof theendless
contest of willswith MurrayGell-Mannto define whatthe Santa FeInstitute
wasall about-including, not incidentally, what the complexity revolution
couldtell us aboutbuildinga more sustainable future for the human race.
Cowanwas just-tired. Now thathe'd gotten the Santa Fe Institute up and
running, he wantedto spendthe time he had left in life working on the
science of the institute,this strangenew scienceof complexity. So at the
firstopportunity-the annualmeetingof the institute's board of hustees in
336 COMPLEXIW
March 1990-Cowan submittedhis formallefterof resignation. One more
year,he told the boardmembers.He'd givethem one moreyearto pick a
successor, while he did his bestto getthe institute'sfundingstabilized.But
that wasit.
"l iust felt it wastime for a freshfaceat the helm," he says.'"Theboard
meetingwastheweekaftermy seventieth birthday.I suppose thatI promised
myselfas a much youngerpersonthat I wouldn'tair* t wasessentialto
anythingwhen I wasseventy.I've seentoo manyold fartsiust gettingin
the way.Therewerelots of otherpeoplewith their own ideas.It wastime
for them to havea chance."
Cowan'sannouncement didn't exactlycomeas a surpriseto anyone
who'd spentany time aroundthe convent.He'd beenlookingso beaten
down of latethat his colleagues had startedto worryabouthis health.His
temperwas erratic;he could be all smilesone day, then thunderyand
doleful the next. He frequentlytold peoplethat he'd announcedhis res-
ignationfrom the presidencythe day he took the job in 1984,that he'd
only takenit to keepthe chair warm for a youngerperson.He'd already
threatened morethan onceto resign,andhadbeentalkedout of it. At the
previous boardmeetingin March1989,in fact,'hehadhintedbroadlythat
the time had comeand he had appointeda searchcommitteeto find his
replacement-acommitteethatnowhadto shiftinto highgearandactually
do something.
But thatwasjusttheproblem,for thesearchcommitteeandfor everyone
else.cowan wasthe onewho hadconceived the institutein the 6rstpiace.
He wasthe onewho hadenvisioned a scienceof complexitybeforeanyone
hadevenknownwhatto call it. He wasthe onewho haddonemorethan
anyoneelseto makethe SantaFe Institutehappen,to makeit the most
intellectuallyexcitingplacethat any of them had everbeenin. As chris
langton said,when you sawGeorgesittingtherein the mothersuperior's
office, you somehowknewthat everythingwasokay.And it wasnt at all
clearwhetheranyoneelsecould carrythat off.
So if not GeorgeCowan,thenwho?

Cowanhimselfhad not the slightestidea.And for the moment,at least,

he didn't havemuch time to worryaboutit; for the nexttwelvemonths,
the pressurewasonly goingto get morerelentless. "BeforeI could, in all
goodconscience, stepdown,"he says,"I wantedthe fundingfor the next
threeyearsto be reasonablywell in hand, so that my successorwouldn't
Work in Progress 337

be immediatelyimpoverished." That meantthat his mosturgentpriority

had to be the completionof a pair of massiveproposals to the National
ScienceFoundation(NSF) and the Departmentof Energy(DOE)' The
originalthree-year grantsfromthoseagencies-atotalof some$2 million-
hadbeenawardedin 1987,andwerecomingup for renewal;if theyweren't
renewed,there wouldn't be much of an instituteleft for anyoneto be
presidentof.
To Cowan, however,therewasa lot more at stakein thoseproposals
than moneyper se. If it had iust beena matterof money,in fact, his life
would havebeena gooddealeasier.The institutecouldhavesimplydone
whatmanyuniversityscienceandengineering departments do, which is to
insistthatindividualresearchers hustleup theirown grants from thefunding
agencies.It wouldn't have been hard; the place was full of smart and
experienced academics who'dbeenhustling grantsall their adult lives.They
knewhow the gamewas played. But Cowan was convinced that such an
approachwould haveendedup destroyingthe very thing that had made
the instituteso special.
"To me," saysCowan,"the overridingissuewasthat we wereinventing
a newkindof scientific community-onethatwasmoreor lessecumenical,
coveringall of the aspects of the hardsciences, the mathematical arts,and
the socialsciences.We startedwith the very bestpeoplewe could 6nd,
togetherwith someblackmagicthat I can only defineasmattersof taste,
in that we went out of our wayto bring togetherthe kindsof peoplewho
would inevitablyproducea kind of intellectualdonnybrook.I think that
the communitywe'vebuilt is uniquein it's breadthandquality.I havenot
seena similarrosterin anyotherscientificinstitutionin history-and I've
lookedfor them, in an effortto emulatethe successes.
"But if we'dgottenstrictlybits and piecesof funding,"he continues,\
"we wouldhavefragmented right away."Quite asidefrom the factthat the I
funding agenciesgenerallyreshicttheir individual granb to cover one
specificpieceof researchin one specific,recognizeddiscipline-exactly
the oppositeof the SantaFe approach-individualgrantstend to create
individualbaronies."You see,when somebody writesin for a grant,he 1
spends a lot of timeworkingon it," saysCowan."Thenhegetshis $50,000I
or $100,000,and he becomes,in effect,an entrepreneur who ownsthat I
money.If youtry in anywayto infringeon hisautonomy,you'vecommitted
a mortalsin." So evenwith the bestwill in the world,he says,evenwith
everybodytrying hard to stayloose,collegial,and interdisciplinary,the
individual investigatorsinexorablyend up spendingmoreand moretime
 338 COMPLEXITY
doingtheir own thing on their own proiects,and lessandlesstime com-
municating with eachother."There'sno centralcoordination. you become
academic all overagain."
In practice,of course,the SantaFe communi$ went out hustlingfor
specializedgrantsanyway.Financesbeingwhat they were, the institute
didn't havetheluxuryof foregoing thementirely.Indeed,citicorp'sfunding
for theeconomics programwasiustoneverylargeexample of a specialized
grant-and BrianArthurhadspenta goodbit of histimeasdireciorwriting
proposals to thisfoundationor thatfoundation,askingfor evenmoremoney.
So, to counteractthe cenhifugalforces,cowan had verybadlywantedto
get what he calledan "umbrellagrant":moneythat could coveranyone
who seemed to havea goodideain complexity, whetheror not that idea
fit into a predefined slot.A chris Langton,for example, or a |ohn Holland,
or a StuartKauffrnan."lf you want a coherentprogremon complexity,"
sayscowan, "then you'vegot to inventa communityin which thrt
herenceemerges from the bottomup-without trying to tell peoplewhat "o-
to do. The umbrellagrantwasan essential partof that."
That'swhy he had goneto the NSF anj the DOE in the firstplace.
until suchtime asan angelappeared andgavetheinstitutean endowment,
thoseagencies werethe only placeshe couldhopeto getumbrellamoney
withoutforcingthe programinto a disciplinary mold. And that'swhy he
felt it wassocrucialto getthe renewals: if the umbrellafolded,the creative
fermentthat Arthur, Kauffman,Holland, and the othershad found so
incrediblyexcitingwould curdleand sourin veryshortorder.
so cowan spentendlesshoursworkingon the newproposalthat spring,
alongwith his executive vice president, Mike Simmons,and the vario,.r,
scienceboardmembers.They all knewthat this had to be one hell of a
persuasive document.It had beentoughenoughto talk the two agencres
into funding the instituteon the first go-aroundin 1987,*he., all the
SantaFe team had to do wasprovethat they had somevery goodpeople
and a very goodidea. It wasgoingto be a far toughersell on this second
go-around,when they wereaskingthe NSF and the DOE to
lointly up
the anteby a factorof l0-from $2 million overthreeyearsto some$20
million over 6ve years.Moreover,they wereproposingthis increaseat a
time whenfederalsciencebudgets wereinexorablygeftingtighter;research-
ersin the conventional disciplines werescreaming for cashmorefiercely
thanever,andmidlevelmanagers in boththeNSFandtheDoE hadbeen
heardto wonderaloudwhy moneywasgoingto thisspeculative, interdis-
ciplinary stuff while perfectlysolid proiectsout in the universitieswent
begging.
Work in Progress 339
SoCowan,Simmons,andcompanyobviously couldn'tmaketheircase
on promisesanymore.They had to showthat they had actuallyaccom-
plishedsomethingin the pastthreeyears,and that theyhad the ability to
do somethingworth $20 million overthe next five years.This wasa bit
tricky, of course,sincethey cpuldn't honestlyclaim to havesolvedthe
whole mysteryof complexity.At bestthey'dmadea sbrt. But what they
could and did claim wasthat in threeyearsof full-time operation,they
had createda viableinstitutiondevotedto attackingthe problemof com-
plexity.As promisedin the original 1987proposal,they wrote,the Santa
FeInstitute"hasdeveloped a comprehensive program,an innovativesystem
ofgovernance,a cadreofremarkably qualiEed investigators,anda support
basethat hasonly begunto matchthe largeoverallneed."
And, in fact,makingdueallowancefor theproposalese in thatstatement,
Cowanand Simmonscouldmakea prettystrongargument.In threeyears,
theypointedout, the institutehadsponsored 36interdisciplinary workshops
attendedby morethan 700people.It hadalsohostedmoreextendedvisits
by another100researchers, who in turn hadpublishedsome60 paperson
complexityin established scientificiournals.It had launchedan annual
ComplexSystems SummerSchool,with monthJongsessions designedto
teachthe mathematicaland computationaltechniquesof complexi$ re-
searchto some150scientists at a time. It had startedpublishinga series
of volumesknown as the Sanfa Fe lnstitute Studiesin the Sciences of
Complexity.And at the time the proposalwasbeingwritten, it wasnego-
tiating with severalacademicpublisherswho wantedto launch a new
researchjournal on complexity.
Then therewasthe research inelf. "lt is particularlynoteworthy," Cowan
and Simmonswrote, "that the commihnentto SFI programsof gifted
associates, rangingfrom brilliant graduate studentsto Nobellaureates, sen-
ior corporateexecutives, and prominentpublic officials,hasgrownand is
no longer an untestedaspectof the SFI approach.The formationand
supportof interactivegfoupsand networksrepresentative of the highest
levelsof ability in the manydisciplinesof interestrank amongthe most
significantconhibutionsmadeto dateby the institute."
Onceagain,theycouldbackup theproposalese with a longlist of specific
accomplishmenb. In fact, the bulk of the proposalwasdevotedto doing
iust that, with extendeddiscussion of programsrangingfrom artificiallife
to economics."The most matureof the SantaFe Institute'sprograms,"
Cowan and Simmonssaid of economics,"this initiative is viewedas a
paradigm,in termsof bothsubstance andorganization, for thedevelopment
of otherinstituteendeavors."
 340 COMPLEXITY

of course,like any reasonably happyfamily puttingon their bestface

for company,the santaFeteamhada fewintimatedeLilsth"t theydidn,t
spellout in the proposal-suchasthe factthai the economics programhad
doneasmuch asanythingelseto driveGeorgeCowannub.
Partof it wasjust the sameold problemof moneyrin his lesscharitable
moments,cowan sometimesfelt as though the economistswantedthe
instituteto raiseall the cashwhile theywentoffand hadall the fun. And
evenwhen he wasn'tbeingquite that dyspeptic,he wasacutelyawarethat
the-economics programhad beenfar moresuccessful intellecfuallythan it
hadeverbeenfinancially.citicorpwashappyenough,andhadcontinued
to renewib $125,000-per-year fundingfor the program.But that hadn't
coveredthe full costby anymeans.And Arthur'sefforbto getmoremoney
out of the big foundations-RussellSage,sloan, Mellon-had all fallen
flat. The brutalfactwasthattherewasviry litde research fundingout there
evenfor mainstreameconomics,much lessfor this speculative- SantaFe
stuff.
"lt turns oyj tfrat economicsis very poorlysupportedin this counhy,"
1
sayscowan- "lndividual economists arepaidverywell, but theydon't get
{
rid for doing basicresearch.They get paid from corporate,our".r, fo,
/
programma-tic things.At the sametime, the field getsremarkably
f *1"g.
I little in the way of research moneyfrom the NationalScience Foundation
and othergovernmentagencies because it's a sociar science, and the gov-
i .planning,'
.T.T.l, is.nota big-pahonof the socialsciences. It smacks of
I
I which is a badword." As a result,he says,manyof the economisbseemed
to look uponthe santaFe Instituteasjustanothersourceofsupport,and
broughtvery little additionalsupportto it. So the institutet nra to
supplementthe citicorp funding with chunksof ib own federal "igrants-
moneythat Cowanhad hopedto usefor otherprojecb
And then on top of all that, saysCowan,Ken Arrow hadbeentrying to
recruita top-notcheconomist to replaceBrianArthur asresident director
afterhe left at the end of 1989."well, we werefundingourselves from
yearto year,and couldhardlylookpastthecurrentyear'sbudget,"he says.
"But when you're trying to attractpeoplewho can go ,rry*h.r. and do
anything,you haveto startmakingpromises aboutwhatresources aregoing
to be availabledown the road. And althoughthe chancynatureo1 the
institutewasveryobviouswhenthe economicsprogramstarted,it became
lessobviousaftera yearor two. It startedto look moresubstantial than it
was.Sothepeoplewetalkedto beganto heatusasthoughwewerestanford
Work in Progress Y1

or Yale. And since there was no endowmentwe either had to disillusion

them or else act as though they were absolutely right and create some
resources.It was a different kind of pressure.The nature of the game
changed."
Once again, however, Cowan's real concern wasn't the money per se'
but the fragile Santa Fe community. The very successof the program
threatened to turn the place into a full-time economics institute, which
was not the idea at all. "Creatingan institutionwithout departments,and
then just pursuingone discipline,is a conhadictionin terms," saysCowan'
"You might as well set up a department in the first place. We had to start
somewhere,but we also had to make sure from the beginning that eco-
nomics didn't becomethe one interestof the institute'"
Perhapsnot surprisingly, this had led to more than one dustup with
Arthur over the funding and pace of the economics program. "On the
scienceboard," saysCowan, "Brian took a partisaneconomist'sposition,
which was that the program was a great success-and that as long as this
programwasproceedingassuccessfullyasit was,we shouldn't divert support
from it for anything else. Don't stop betting on a winning horse. Now,
Brian is a fierce defenderof his views.That's great,But the entire philosophy
of the institute was that complex systemsconsist of many aspects.These
include-especially if you're going to talk about complex systemsthat in-
volve people-neural behavior, human behavior, societal behavior, and
many other things that economicsdoes not specificallydeal with. So I
pushed hard to support at least one other Programthat would be equal in
size to the economics program. We needed to broaden our academic
agenda,and spreadour bets.And the scienceboardas a whole was quite
supportive of that-though with a lot of discussionabout it' "
The particular program Cowan had in mind was'ldAptiye computation":.
an effort to develop a set of mathematical and computational tools that'
could be appliedto all the sciencesof complexig-including economics.
"lf there's a common conceptual framework," he says,"there ought to k
a common analytic framework." In part, he adds, starting such a program
would just be a matter of recognizingwhat wasalreadythere and giving it
broadersupport.fohn Holland'sideasaboutgeneticalgorithmsand classifier
systemshad long since permeatedthe institute, and would presumablyform
the backboneof adaptivecomputation.But therewerealsoall thesesimilar
ideasgrowing out of Stuart Kauffman'sBoolean networksand autocatalytic
sets, Chris Langton's artificial life, and the various economy-under-glass
models that Brian Arthur and the economistswere building. A lively cross-
fertilization was well under wav-witness Dovne Farmer's"Rosetta Stone
 y2
COMPLEXITY
for connectionism" paper, in which he pointed out that neural networla,
the immune systemmodel, autocatalytic sets,and classifiersystemswere
essentially just variations on the same underlying theme, Indeed, Mike
Simmons had inventedthe phrase"adaptivecomputation" one day in 19g9,
as he and cowan had been sitting in cowan's office kicking around names
that would be broad enough to cover alr these ideas-bu1 that wouldn't
carry the intellectual baggageof a phraselike "artificial intelligence."
so_at one level, sayscowan, an adaptivecomputation program would
simply give this ferment some formal recognition a.td coordinition, not to
mention some exha money for graduatestudents,visitors,and workshops.
In the long run, however,he wasalso hoping that the program would give
economists, sociologists,political scientisb, and even historians some of
the same precision and rigor that Newton brought to physics when he
invented calculus. "what we're still waiting for-it may taki ten or fifteen
years-is a really rich, vigorous, generalset of algorithmic approachesfor
quantifying the way complex adaptiveagentsinteractwith one another,"
he says."The usual way debatesare conducted now in the social sciences
is that each persontakesa two-dimensionalslice through the problem, and
then arguesthat theirs is the most important slice. 'My srice is more im-
portant than your slice, becauseI can demonstratethat fiscal policy is much
more important than monetary policy,' and so forth. But you can't dem-
onstrate that, becausein the end it's all words, whereasa computer sim-
ulation providesa catalogof explicitly identifiedparametersandvariables,
that people at leasttalk about the samethings. And a computer lets you
1o
handle many more variables.so if a simulatiorrhas both fiscalpolicy and
monetary policy in it, then you can start to say why one turns out io be
more important than the other. The resultsmay be right or they may be
wrong. But it's a much more structureddebate.Even when the models are
wrong' they have an enormousadvantageof structuringthe discussions."
whether or not the .simulationsever got that good, however,starting an
adaptivecomputation program would certainly have at leastone h"ppy-ride
effect: it would give cowan and company an excuseto hire
;ohn itolland
away from Michigan to be their first full-time faculg member. Not only
was he the natural and unanimous choice to be the program's resident
director, but he was a nonstop font of energy and ideas. piople
iust liked
having him around.
cowan and simmons accordinglygave adaptivecomputation its own
special lO-pagesection in the NSF-DOE proposal-much of it written by
an enthusiasticfohn Holland himself-and shippedthe whole r 50-page
packageoff to washington on fuly 13, 1990. From there on out. about all
Work in Progress v3
they could do waswait with their fingerswell crossed,and hopethat the
reviewerswould be kind.

Therewasa certainirony in the institute'scourtingof Holland.Backin

the earlydaysof the institute,Cowanandtheotherfoundershadhadevery
intentionof hiring a permanentfacultyand makingthe placeinto a full-
fedgedresearch institutionalongthe linesof RockefellerUniversityin New
York. But fiscalrealityhad intervened.And by 1990,Cowan,Simmons,
and quite a numberof the other SantaFe regularshad begunto suspect
that this particularrestrictionhad at leastone largevirtue the institute
might actuallybe much befteroff withauta permanentfaculty.
"The virtue wasthat we weremoreflexiblethan we wouldhavebeen,"
saysCowan.After all, he'd realized,onceyou hire a bunchof peoplefull
time, yourresearch programisprettywellcastin concreteuntil thosepeople
leaveor die. Sowhy not just keepthe institutegoingin its catalystroleTIt
had certainlyworkedbeautifullyso far. Keepgoingwith a rotatingcastof
visitingacademics who wouldstayfor a while, mix it up in the intellectual
donnybrook,andthen go backto their homeuniversities to continuetheir
collaborations longdistance-and,not incidentally,to spread therevolution
amongtheir stay-at-home colleagues.
Thatsaid,howeveleveryone wasmorethanwilling to makean exception
in Holland'scase.And bestof all, a sourceof fundsto supporthim had
alreadypresentedibelf, in the flamboyantform of one RobertMaxwell:
formerCzechresistance fighter,self-madebillionairepres baronin Lon-
don, and-it turnedout-a mangivento quirkyenthusiasms aboutthings
like complexity.
In retrospect,of course,RobertMaxwellisalsofamousfor his mysterious
deathby drowningin late 1991,and the spectacular collapseof his debr
riddenmediaempireimmediatelythereafter.But at the time, he looked
like a fairy godmother.The institute'scontactwith Maxwellhad begun
morethan a yearearlier,whenMurrayGell-Mannhad happened to meet
Maxwell'sdaughter,Christine.ChristineMaxwell,in turn, had arranged
for Gell-Mann to havelunch with her fatherin May 1989.And when
Gell-Mann had reportedbackto Cowanthat the elderMaxwellseemed
intriguedby what the institutewasdoing, the SantaFe teamhad imme-
diatelygoneinto fund-raisingmode.Nobodyhad the slightestideawhat
Maxwellwasworth, but it had to be-zillions. .
Many faxesand phonecallslater,in February1990,therearrivedone
particularfax from London makingtwo key points.First, saidMaxwell,
 34 COMPLEXITY
he wantedto beginhis association with the institutewith a conhibutionof
$100,000to be usedfor the studyof adaptivecomplexsystems.second,
he liked the institute'sideaof foundinga new scientificjoumal on com-
plexity, and would be interestedin publishingthat journal through his
subsidiary,the academic publishinghousepergamon press.
wantedto fogin his association!?cowan andsimmonsmulledoverthat
little gemfor a while. Finally,cowan decidedto takea gambleand up the
ante:"l want to askhim for more." In his replyhe encloseda draft of the
wo* of the institute'sjournalcommittee,outliningwhattheyhad in mind
for the journal, and addeda proposal, that the publith.t establish',The
RobertMaxwell Professorship" at the institute,funded at the level of
$300,000 peryear.ThatsumwouldcovernotjustthesalaryfortheMaxwell
professor alone,cowan explained,but wouldalsopayfor postdocs, graduate
-
sfudents,travelmoney,a secretary, and assorted otherexpenses.
The responsefrom London took sometime in coming. Maxwell, as
cowan and Simmonshad long sincelearned,delegated almostnothing.
All they coulddo waskeepthe fax lineshot with reminders, alongwith
lettcrs,telephonecalls,and contactsthroughGell-Mann,christine-Max-
well, and her brothers.The answer-'Accept in principle"-finally came
just in time for the boardmegtingin March 1990to formally
oft'e,the
Maxwellprofessorship to |ohn Hollandfor five years.

up in Michigan, Holland proceeded to parlaythat offer for all it was

worth. By that point, still bitteroverthe mergerof his old computer and
communicationsSciencedepartment into the engineering school,and
hatingthe kind of short-term,applications-oriented mindset-that prevailed
there, Holland had alreadyleveraged himself halfwayout. A f.* y.r*
earlier,ucLA hadstartedto hint thatit mightofferhim an endowedchair.
SoHolland,showinga previouslyunsuspected talentfor academicgames-
manship,had immediatelygoneto the university's provost."To stayhere,"
he said,"I needat leasta half-timeappoinhnentin psychology"-alarge,
nationallyrankeddepartmentwherehe had extensivecontactsfrom his
daysof workingon the booklnduction. The provost,Edie Gordenberg,
beingboth sympathetic andeagerto keephim at Michigan,hadmadethe
necessary arrangements
Now, with the offerfrom SantaFe in his pocket,Hollandwentto Gol-
denbergagain. "This Maxwell professorship is almostideal in termsof
doing research,"he told her, "and I'm very inclinedto hke it-un/ass I
can spendmoretime on research hereat Michigan."Onceagain,Gol-
Work in Progress 345
denbergwas readyto listen. She found money, made arrangements,and
helped him work out a quid pro quo: Holland would get a full-time ap-
pointment in psychology,plus a reduced teaching load to give him more
time for research.And in return, he would setup a permanentlink between
the Santa Fe Institute and the university-an arrangementwherein pro-
fessors,postdocs, and graduate students from Michigan would regularly
spend time in Santa Fe, and the two institutions would regularly sponsor
joint conferences.It would be a kind of SantaFe outpostin the snowsof
Ann Arbor.
The deal was consummatedby the summer of 1990. To inaugurate the
outpost, Holland organized a two-week seminar in the fall of 1990, with
a special kickoff symposium starring Brian Arthur, Stanford'sMarc Feld-
man, and Murray Gell-Mann, Holland had a great time, and from all
accounts, so did everyone else. "[University president]ames] Duderstadt
came to the kickoff symposiumand stayedthe whole time!" saysHolland.
"He eventook notes.It wasa lot offun, and everybodywaspleased."From
then until no% moreover, with the exceptionof foraysto Santa Fe and to
various conferences,Holland has spent the majority of his time happily
ensconcedwith his Macintosh II computer in the study of his home, a
striking hilltop chateau overlooking the rolling woodlands west of Ann
Arbor. Lately, in fact, he's even begun to talk seriouslyabout retiring from
the university altogether,so that he can have even more time for research.
"lt's the finite horizon effect," he says."I'm getting old enough [he's sixty-
three], and I have so many ideas in my folder that I want to work on
more. ."

Back in Santa Fe, Cowan was sorry to hear Holland say no to the
professorship.But he had to admit he was impressedby the way Holland
had finessedhis way out of a bad situationat home. And he wasevenmore
impressedby the fact that Holland had played"bet your iob" to securethe
ongoing link with Michigan-something the institute was overioyed to
have, and probably never would have had otherwise.
In the meantime, howevel Cowan had to deal with Maxwell. He and
Simmons spent the early summer of 1990 keeping the fax lines hot with
very polite reminders to [,ondon: pleasedon't forget to send the money.
Maxwell's personalcheck for $150,000-the first installmentof the first
year's$300,O00-finally arrivedin August.And it wasonly then that they
told him that Holland wouldn't be able to accept. "Do you think it would
help if I went to Michigan and talked to him?" Maxwell responded.
 346 COtvlpLEXtTy
Well, no. But SantaFe wasableto offera compromise: Hollandand
Gell-Mann would sharethe professorship for the fall 1990semester that
wasjust shrting, during which time Holland would lay the groundwork
for this new adaptivecomputationprogram.In l99l the positionwould
rotatebehveenStuartKauftnan and David Pines.And in the meantime,
the institutewould usethe fexibility to bring in someFrst-class younger
people,suchasSethLloyd, jamesCrutchfield,and Alfred Hubler.
That, said the fax machine,wasquite acceptable to Maxwell. It also
provedto be quite acceptableto everyonethat the newcomplexityjournal
be publishedthroughMaxwell'sPergamon Press.The detailswereworked
out by cowan and Maxwellduringa long transatlantic telephoneconver-
sation-shortly beforeMaxwellsuddenlydecidedto sellpergamonto help
financehis otheracquisitions.And in lateRbruary 1991,aftera seriesof
increasingly urgenthansatlanticreminders,Maxwellevenremembered to
sendalonganother$l 50,000to payfor thesecondhalfofthe professorship's
academic year.

All throughthe summerandfall of 1990,wheneverthe subjectof Cow-

an'ssuccessorcameup, MurrayGell-Manncouldbe heardto sighin tones
of resignation,"l guessI'll haveto do it."
Gell-Manncertainlydidn't wcnf to.be president of the institute,one
wasgiven to understand.He loathedbureaucraticbusywork.He'd been
turning down iobs like this all his life-the chairmanship of caltech's
Division of Physics,Mathematics,and Astronomy,for example.But the
SantaFe Instituteand the sciences of complexigwereso incrediblyim-
portantthat-well, who elsehadsucha clearvisionof whatneededto be
done?Who elsecould articulatethe sciences of complexityso well?Who
elsehad the prestigeand the networkof contactsto givethe institutethe
cloutit needed?
Who elseindeed?The institute'ssearchcommifteeimmediatelywent
into paralysis.No one wasfooled:Murray Gell-Mannwantedvery badly
to be presidentof the SantaFe Institute.The questionwaswhetherthey
daredlet him do it. Somepeoplefelt that theyoughtto seriouslyconsider
the possibility.After all, they said,whatwe havehereis a seminalfigure
in the historyof science-with a Nobel Prizeto boot. If he reallywanb
the iob, why not givehim a shot?
Others,who knewhim befter,wereappalledat the thoughtof Murray
Cell-Mannactuallyhying to run anything.No onedoubtedhis intellectual
vision, his energy,or his fund-raisingpower.He wasa nonstopsourceof
Work in Progress 347
ideasaboutwhat scientificquestionswould be interestingto tackle.And
he seemedto know everybody;he had an incredibleknackfor getting
togethergroupsof peoplewho were absolutelytops in their fields. The
SantaFe Institutewouldn't be what it waswithout him. But-president?
They had visionsof his deskaccumulating geologicalshataof unsigned
papersand unreturnedphonecallswhile Cell-Mannwasoff savingthe
rain forest.Worse,theyhadvisionsof the SantaFeInstitutebecoming,de
facto,"The Gell-MannInstitute."
"Murray hasmoreof a purelyintellectualapproachto life than anyone
I've everknown," saysone physicistwho hasknownGell-Mannfor many
years."All of his conversationsand everythingelseabouthis life is driven
by his intellectualconcerns. He caresdeeplyaboutthe intellectual agenda
of the SantaFe Institute.He seesthe directionthat he wantsit to go. He's
thoughtaboutit verydeeply,andwanbto besurewemovein thatdirection.
"Now, that'sbothgoodandbad.I think it's goodfor the instituteto have
a strongintellectlike Murraydrivingit in productive directions.But the
flip sideis that when Murray is around,it's hardfor anyoneelseto get a
word in edgewise. Once he analyzesa problem,he feelsthat it hasbeen
fully analyzed.If someonedisagrees, he tendsto think that they mustnot
haveheardhim or mustnot haveunderstood. And if he doesn'twritethem
off entirely,he tendsto repeathis argumentfor greaterclarity.Soby sheer
intellectualpowerandforceofpersonality,he tendsto displaceeveryother
point of view.The dangerthateveryonesawwasthatthe SantaFeInstitute
would just becomea vehiclefor Gell-Mann'spersonalenthusiasms."
That wascertainlythe dangerthat Cowansaw.In fairnes, he'd heard
Gell-Mannsayall the rightwordsaboutthe needfor diversityandmultiple
pointsof view at the institute.But he wasalsoconvincedthat Gell-Mann
as presidentwould wreck the institute'stumultuous,multifacetedcom-
munity without evenmeaningto, asall the truly originalthinkersleft to
preserve theirsanity."MurraywouldbetheHerrProfessor whowasrunning
things," saysCowan. "He alwaysfeelsthat his point of view is the only
possiblepoint of view. He'salwaysstraightening peopleout."
Cowanhad reasonto know In one wayor another,he'd beenfighting
this battlewith Gell-Mannsincethe institutewasfounded.He did his best
to keepa lid on it, of course.Cowanwasacutelyawareof how much he
and the instituteneededGell-Mann;he'd felt compelledto deferto the
man sooftenthat manypeoplewonderedif he weresimplyintimidatedby
NobelPrizewinners.But thereweredayswhenCowanjustcouldn'tstand
it anymore.
Thke their long-runningdebateover the propersubjectmafterof the
 348 COMPLEXITY
institute."I think of thesubjectasthestudyof simplicitycnd complexig,"
saysGell-Mann."The simplelaw of the universeand it's probabilistic
characterseemto me to underliethe wholesubiect-that, and the nature
of informationand quantummechanics. well, we havehad information
andthe universediscussed twiceat SantaFe.And in the earlydayswe had
a wonderfulworkshopon superstrings, with an overviewof mathematics,
cosmology,and particlephysics.But there was all this pressureagainst
studyingsimplicity,andwe'veneverdonesuperstrings again.The president
of the institute,GeorgeCowan,hatedthesethingsbitterly.I don't know
why."
Actually,Cowandidn't hatethesethings.Superstring theory-a hypo-
thetical"Theory of Everything"that attemptsto describeall elementary
particlesasinfinitesimally tiny,furiouslyvibratingthreads
of pureenergy-
waswonderfulstuff. It's just that therewereplentyof otherplaceswhere
peoplecouldworkon strings,quarksandcosmology, andhe didn't think
that the institutehad any time or moneyto wasteon duplicatingthem.
(Nor wascowan the only onewho thoughtthat:a majorityof the science
boardlookedat that superstring worlshopand said, .,Neveragain.',)But
for Cowan,the reallyannoyingthingaboutGell-Mann's..simplicig', was
that it soundedto him like reductionismin disguise.He found it ielli.,g
that Gell-Mann still took suchobviousrelishin cleverput-downsof any-
thing he wasn'tpersonallyinterestedin, suchas chemistryor solid-state
physics.(He wouldcall the latter"squalid-state physics"to phil Anderson's
face,apparently justto irritateAnderson.) MaybeGell-Mannwassimply
trying to be funny, sayscowan. But the not-very-veiled message wasthat
the studyof collectivebehaviorwassomehowpragmatic,messy,and not
"intellectual.'
To outsiders, thetestiness overGell-Mann's notionof simplicitysounded
a bit like one of thosearcanemedievaldebates overthe finer pointsof
theology.But Cowanand Gell-Manngot quiteangryaboutit, with the
subiectleadingto any numberof arguments and abruptlyterminatedtele-
phone calls. The one occasionthat Cowan particularlyrememberswas
about 1987,when five or six of the seniorSantaFe regularsweresitting
aroundthe tablein a smallprivatemeetingand discussing how the Santa
Fe Institute should describeitself. "Every time we would saywe were
interested in the sciences of complexity,"saysCowan,"Murraywouldadd,
'and
the fundamentalprinciplesof which it wascomposed'-meaning
quarks.The implicationwasthat socialorganizations weremadeup of a
Iot of quarks,and you could follow the quarksthrough to the various
aggregations.
Work in Progress 349
"Now, this is what I would call the religionof theoreticalphysics,"says
Cowan:"this beliefin symmetryand totalreductionism. I sawno reason
to adoptthat statement,and I saidwe weren'tgoingto do that." Cowan's
argument,supportedby most of the othersaround the table, was that
emergent,complexsystemsrepresented somethingnew-that the funda-
mentalconceptsneededto understand their macroscopic behaviorgo well
beyondthe fundamentallawsof force.
"Murraysaidfatly thathewouldn'tgoalongwithit," says Cowan."Well,
this wasthe 6rst time I realizedthat Murray,merelyby asserting that this
washow he wantedit to be, expectedotherpeopleto do thingshis way.
And I felt that thiswassomonumentallyegocentric that I lostmy temper."
lndeed,Cowanwassofuriousthathethrewwhatwas,for him, a tantrum:
he pickedup his papers,said"l quit," and walkedout of the room-with
Ed Knappand PeteCarruthersin hot pursuit,shouting,"George,come
back!"
He did, eventually.And afterthatincident,Cell-Mannrarelymentioned
"simplicity"again.
However,Cowan'sannoyance oversimplicitywasnothingcompared with
whathe felt aboutthe institute's"GlobalSustainability" program.To begrn
with, it had startedout as f,is program,a smalleffort reflectinghis most
deeplyheld concernsabouthuma.nity's future on this planet.Moreover,
he hadn't evencalledit "sustainability" then. His originalconceptwas
"Global Stability"or "Global Security"-the latterbeingthe title of the
6rstsmallworlshophe organizedin December1988."The subiectstarted
out as somethinglike nationalsecurig,but it rapidlybecamemuch
broader,"saysCowan:"How do we survivethe nexthundredyearswithout
'ClassA' That is, somethingthat can't be setshaightin a
a catastrophe?
generation." In edge-of-chaosterms,avoiding suchcataclysms wouldmean
findingsomewayto dampout theverylargest,mostdestructive avalanches
of change."Originally,numberoneon my list of ClassA catastrophes was
nuclearwar," saysCowan, "with a ClassB catastrophe beingsomething
like World War II. But by the time of the first meeting,rapprochement
betweenRussiaandthe UnitedStates wassuchthatthenuclearwarproblem
wasdown aroundnumberfiveon the list. And whatemergedveryquickly
insteadwasthe populationexplosion,the [Paul]Ehrlich-gpecatastrophe.
Then camepossibleenvironmentalcatashophe, such as the greenhouse
warming,whichI myselfdidn'tthinkof asa ClassA catastrophe, butwhich
othershaveseizedupon."
This effort perkedalongfor a while in a low-keyway,largelybecause
Cowankeptorganizingsmallmeetingson his own wheneverhe could find
 3s0 COMpLEXtTy
the time. But then Gell-Mannbeganto get more interested. The ideaof
takinga global,integrated lookat humanity'slong-termviabilitywassome-
thing that resonated shonglywith him. After all, cel-Mann's firstintro-
duction to sciencehad beenhis naturewalksin central hrk at agefive.
His mostdeeplyheld concernwasthe preservation of the globareiviron-
mentin general,andthebiological diversigof therainforestin particular.
so he wadedin, inexorablypushingcowan'sglobalstabirityprogramin
the direction he wanted.And by 1990,he had effectivelyredefinedthe
agendaand madeit his own.
It wasa far more activistagendathan cowan's.Gell-Mannwasn'tin-
terestedsimply in avoidingcatashophe. He wantedto achievea stateof
global "sustainability"-whateverthat notoriouslyslipperyword might
mean.
Speakingat a SantaFe workshopin May 1990-by now he wasa co-
chairmanwith cowan-Gell-Mann pointedout that "sustainability" has
in truth becomea trendyclichc of late, the sourceof endlessplatitudes.
Fbr mostpeopleit seemsto meansomethinglike business asusual-but,
you know,sustainable. And yetbusiness asusualis preciselythe problem,
he said.At the world Resources Instifute,a washington-tased environ-
mentalthink tank that Gell-Mannhad helpedsetup in his capacityas a
directorof the MacArthur Foundation,foundingdiiectorGns spetirand
othenhavearguedthatglobalsustainability ispossibleonryif humansociety
undergoes at leastsix fundamentalhansitionswithin a very few decades:

l. A demographic hansitionto a roughlystableworld population.

7. A technological transitionto a minimal environmentalimpactper per-
son.
3. An economichansitionto a world in which seriousattemptsare made
!
to chargethe realcosbof goodsandservices-includingenvironmenhli
cosb-so that there are incentivesfor the world economyto live off |
nafure's"income" ratherthan depletingib "capital." I
4. A srcial transitionto a broadersharingof that income, along with I
increasedopportunitiesfor nondeshuctiveemploymentfor the poor
familiesof the world.
5. An institutionalhansitionto a setof supranational alliancesthat facil-
itate a globalattackon globalproblemsand allow variousaspectsof
policy to be integratedwith one another.
6. An informationaltransitionto a world in which scientificresearch,
education,and global monitoringallow large numbersof peopleto
understand the natureof the challenges theyfrce.
Work in Progress 351
The trick,of course,is to getfrom hereto therewithoutoneof Cowan's
ClassA globalcatastrophes. And if we'reto haveanyhopeof doingthat,
saidGell-Mann, the studyof complexadaptivesystems is clearlycritical.
Understanding thesesixfundamental transitionsmeansunderstanding eco-
nomic,social,andpoliticalforcesthataredeeplyintertwined andmutually
dependentupon one another.You can't iust look at eachpiece of the
problemindividually,ashasbeendonein the past,and hopeto describe
the behaviorof the systemasa whole.The only wayto do it is to look at
theworldasa shonglyinterconnected system-evenifthe modelsarecrude.
But more than that, saidGell-Mann,the trick in geftingfrom here to
thereis to makesurethat"there"is a worldworthlivingin. A sustainable
human socieg could easilybe someOrwelliandystopiacharacterized by
rigid controland narrow,confinedlivesfor almosteveryonein it. What it
shouldbeisa societythatisadaptable, robust,andresilientto lesser disasters,
that can learnfrom mistakes, that isn'tstatic,but that allowsfor growthin
the quality of human life insteadof just the quantityof it.
Achievingthis will clearlybe an uphill battle,he said.In the West,
intellectualsandmanagers aliketendto behighlyrationalistic, emphasizing
the meansby which undesirableeffectsoccur and lookingfor technical
fixesthat will blockthoseeffects.Thus, we havecontraceptives, emission
controls, arms control agreements, and so forth. And those things are
But
certainlyimportant. the real solution will requiremuch he
more, said.
It will requirethe renunciationor sublimationor transformation of our
traditionalappq!!E:s: to outbreed,outconsume,and conquerour rivals,
especiallyour rivalsin other tribes.Theseimpulsesmay oncehavebeen
adaptive.Indeed.thev maygvenbe had=f-u,lld_g{o3_ug.bry{n9. But we no
-faen
longerlravethe fuy_1ygf tolgglinAlbg"r. Lhal rr. 'k*l>
And yet thereinliesone of the crucialproblems, saidGell-Mann.On
the one hand,humanityis gravelythreatened by superstition and myth,
the stubbornrefusalto recognizethe urgentplanetaryproblems,and gen-
eralizedtribalismin all its forms.To achievethosesix fundamentaltran-
sitionswill requiresomekind of broad-scale agreement on principlesand
a more rationalway of thinking about the future of the planet, not to
mentiona morerationalwayof governingourselves on a globalscale.
But on theotherhand,he said,"Howdo youreconcile theidentification
andlabelingof errorwith thetolerance-notonlytolerance, but celebration
and preservation-of cultural diversig?"This isn't a mafterof political
correctness, but ofhard-edged practicality.Culturescan'tbe eradicated by
fiat; witnessthe violent reactionto the Shah'seffortsto westernizeIran.
The world will haveto be governedpluralisticallyor not at all. Moreover,
 352 COMPLEXITY
qulturaldiversitywill be justasimportantin a sustainable worldasgenetic
diversityis in biology.we needcross-cultural ferment,saidGell-Mann.
"of particularimportance maybe discoveries abouthow [ourown culture
can] restrainthe appetitefor materialgoodsand substifutemore spiritual
appetites." In the long run, he said,solvingthis dilemmamay require
much morethan sensitivity.It mayrequireprofoundnewdevelopmenb in
the behavioralsciences.After all, the cure of individual neurosesis not
easy;the sameis true of socialneuroses.
Of course,saidGell-Mann,lookingat suchmultifaceted, denselyin-
terconnected systemsis exactlywhat the santaFe Institutewasset up to
do. But he arguedthat the institutewasfar too smallto undertakea study
of global sustainabilityby itself. It neededpartnerssuch as the world
Resources Institute,the BrookingsInstitution,and the MacArthur Foun-
dation (by no coincidence,the cosponsors of that particularworkshop).
with them to lookat the policyaspecb,he said,and the santaFeInstitute
to focuson the basicresearch,they could beginto attaekthe problemof
sustainabilityasa whole,

By the time of the May 1990workshop,of course,whatwasnow called

the "Global Sustainability" programhad longsinceslippedfrom Cowan,s
control.And aboutthe only thing he coulddo aboutit waswatchin quiet
fury. After all, Gell-Mannwascochairmanof the institute'sscienceboard,
which gavehim far moresayin the directionof anygivenprogramthan
Cowan had. Gell-Mann could and did definethe progremthe way he
wantedit, while Cowan,aspresident,had the responsibiligto go out and
raisethe money.
And as if that weren'tinfuriatingenough,therewasthe actualcontent
of Gell-Mann'sagenda.Cowandidn't think it waswrong,exactly.Cowan
wasthe first to agreethat the world is far from sustainablenow, and that
somefundamentalchangesare sorelyneeded.No, what Cowan found
enragingwas that Gell-Mann and his buddiesfrom Brookingsand
MacArthurand the world Resources Institutewereso-sure. Despiteall
of Gell-Mann'sprotestations to the conhary,whenyou actuallylistenedto
them you couldn't help feelingthat they knewthe problems,they knew
the solutions,and all they reallywantedto do wasgeton with preserving
the rain forest,
, Cowanwashardly alonein that feeling.Then and nou manypeople
f at the instituteweredeeplysuspicious thattheGlobalsustainabilityproiect
\ wouldturn into somekindofglobalenvironmental activism."If youalready
Work in Progress 353
know what to do, then it's not a researchprogram,"saysone SantaFe
regular."lb a policy implementation progrem,and that'snot our role.,' )
YA the factwasthatCowan,for one, justdidn't havethe energyto fight
with Gell-Mannanymore.let him havethe damnGlobalSustainability
program.cowan would get backto his own visionof globalstabilityafter
he steppeddown aspresident,if then. "l havea feelingthat Murrayand I
don'thavereallydeepjntellectualdifferences," saysCowan...We'reboth
more similar than we ought to be. Maybethat'sthe problem.His social
skillsaresuchthat I find myselfeasilyoffendedby Munay. I'm not alone
in that. But I haveno reasonto put up with it, soI probablylosepatience
moreeasily.If I werea moreperfectpenon,therewouldn'tbeanyproblems.
I've iust reachedthe agewhen I don't botherwith peopleI haveto make
allowances for."

Towardthe end of 1990,at a time when Gell-Mannwasstill the only

seriouscandidatefor the Sanh Fe Institutepresidency, Cowanhappened
to be chattingwith Ed Knapp,who wasnow backat LosAlamosheading
up the laboratory'smesonphysicsfacility.Knapp,a tall, easygoingphysicist
with a distinguished crop of wavysilverhair, mentionedthat los Alamos
wasofferinga veryattractiveearlyretirementpackage, at leastpartlyto ease
the pinch of post-ColdWar defensecutbacks.In fact,saidthe fifty-eight-
year-oldKnapp,he wasthinkingof takingadvantage of it.
Neither man remembenexactlywho saidwhat at that point. But very
quickly,the obviousquestionwasin the air: Would Knappbe interested
in becomingpresidentof the SantaFe Institute?
lt madea lot of senseto cowan. Knapphadbeenpresentat the creation,
backwhen the institutewasstill just an ideabeingkickedaroundamong
the laboratory'sseniorfellows.He'd alwaysbeenwilling to help out when
he could-even agreeingto serveaschairmanof the boardof husteesfor
two years.He'd beenheadof the NationalscienceFqundationback in
washington, then head of the universitiesResearchAssociation.the
seveng-two-member universityconsortiumthat runs the Fermi National
Accelerator Laboratory outside of Chicago and the DOE's new
superconducting supercolliderproiect.He clearlycaredaboutthe institute
and what it stoodfor. And yet, unlike certainother candidates,Knapp
had no personalaxesto grind regardingwhat the institute should or
shouldn'tdo.
"George,"protestedKnapp,"rememberthat I'm not a theoreticalsci-
entist,I'm an administrator."
354 COMPLEXITY
"That'sgreat,"saidCowan.
The discussion wentfrom there.Knappagreedthat if offeredthe job by
the institute'sboardof trustees,he would takeit. And onceCowanhad
passedthat wordalong,the senseof reliefamongthe boardmemberswas
tangible.The questionhadalwaysbeenwhetherGell-Mannwouldor could
remakehimselfinto an adminishator,andwhetherhe wouldbe willing to
takeenoughtime awayfrom his manyotherinteresbto do a decentiob in
SantaFe. By late 1990,the generalconsensus wasthat he wouldn't. And
now that an acceptable altemativecandidatewasavailable,it very quickly
becameclearto everyone,includingGell-Mann,thatif he forcedthe issue
to a voteof the board,he would lose.
Meanwhile,Gell-Mann himselfhad begunto geta senseof what he'd
beenaskingfor. David Pines,amongothers,hadspentquite a lot of time
trying to explainwhat it meantto be an administrator-thebudgets,the
meetings,theendless hasslingwith personnel."Murray,"Pineskeptsaying,
"this isn't the iob you want at the SantaFe Institute;you want to be a
professor."
So in the end it wasall very gentlemanly.A specialboardof hustees
meetingwascalledfor December1990.Gell-Mannhimselfput Knapp's
namein nomination.And Knappwasthe unanimouschoice.
"l wasa little disappointed,"saysGell-Mann. "I would haveliked the
iob. This wasthe first time in my life I'd everexpressed
interestin sucha
iob. But I wasquitepleased with the choiceof Ed Knapp.I washappythat
the personwe chosewasa goodone and easyto work with."

As he'd promiseda yearearlier,CeorgeCowansteppeddownfrom the

presidencyof the SantaFe Instituteat the boardof trusteesmeetingin
March 1991.And ashe'dhoped,he wasableto do soin goodconscience.
The NSF and the DOE had renewedtheir granb for threeyearsinstead
of five years,and for an unchanged$2 million insteadof $20million. But
the granb had indeedbeenrenewed.Meanwhile,the MacArthur Foun-
dation had decidedto boostib annual conhibutionfrom $350,000to
$500,000.Increases had beenpromisedby severalindividualdonors,in-
cluding Gordon Getty and William Keck,fr. And RobertMaxwell was
committedto funding his profesonhipat the rateof $300,000per year-
althoughhe wasstill dolingout the moneya semester at a time. SoCowan
wasindeedleavingthe institutein soundfinancialshapefor the nearterm;
his successor,Ed Knapp,wouldhavetheluxuryof punuingan endowment
 Work in Progress 355
without having to constantly scroungefor dayto-day operating expenses.
(Life has not been quite that rosy in practice: the evaporationof the pro-
fessorshipafter Maxwell's death in late l99l left a rather large hole in
Knapp's 1992 budget, and forced the institute to cut back on the number
of visitors and postdocs.Fortunately, the shortfall was temporary and re-
parable.)
Once the mantle had safelypassed,however,Cowan was out of there.
He was now seventy-oneyears old, and after seven yearsof anxiety and
administrativehassle,he badly neededa rest-which in his casemeant
reimmersinghimself in the double betadecayexperimentthat he and several
los Alamos colleagueshad been planning for the better part of a decade,
and which was now nearingcompletion.For months he rarelyshowedhis
face around the institute. (The double betadecayexperimenthad been one
of a long list of Cowan's researchproiectscited by the DOE the previous
October when the department named him as the corecipient of its presti-
gious Fermi Award, which honors outstandingscientificachievementin
the development, use, or control of atomic energy.Previousrecipientshad
included such figures as fohn von Neumann and f. Robert Oppenheimer.
Double beta decayis an exotic and exceedinglyrare form of radioactivity
that providesa sensitiveexperimenhl testof the shndard elementaryparticle t
theories. Much to Cowan's delight, he and his colleagueswere able to I
detectthe decayand show that it wascompletelyconsistentwith the standard
theories.)
j
For Cowan, however,the break apparentlyhad its healing effect. By the
fall of 1991, he had once again becomea regularat the institute, where
he shared a small office with Chris Langton. And more than one person
remarked how healthy and enthusiastiche was looking.
"I don't know quite how to explain how I felt about stepping down,,'
Cowan saysnow. "One way is to repeatthe story of the guy who sat in the
presenceof this continuously loud sound, and when it stoppedsaid, ,What
was that!?' Or, if you constantly wear a hair shirt, you feel a little funny
when you take it off. If you've got a puritan streak, you even feel a little
guilty when you take it off. But now I've put on a modified version of the
hair shirt, and I feel a lot better."
In particular, he says,now that he has so much more time to think about
the new sciencesof complexig, he 6nds himself more enthralledthat ever.
"Talk about the coercivepower of an intellectual idea! I feel as though I've
been coerced more than anyone else. These things have grabbedme and
kept me in a state of perpetual excitement. I feel as though I've taken a
 356 COMPLEXITY
new leaseon life, at the craniumpart of my body.And that, to me, is a
maior accomplishment.It makeseverythingI've everdone here worth-
while."
The issuethat grabshim the hardest,he says,is adapbtion-or, more
precisely,adaptationunderconditionsof constantchangeand unpredict-
ability. Certainlyhe considersit one of the centralisues in the elusive
\ questfor globalsustainabilig.And, not incidentally,he 6ndsit to be an
issuethat's consistendyslightedin all this talk about "hansitions"to a
\!|sustainable
world. "Somehow,"he says,"the agendahasbeenput into the
$orm of talkingabouta setof transitions from stateA, the present,to a state
fB that's sustainable. The problem is that thereis no suchstate.You have
r to assumethat the hansitionsare goingto continueforeverand everand
ever.You haveto talk aboutsystems that remaincontinuouslydynamic,
and that are embeddedin environmenbthat themselves arecontinuously
dynamic." Stability,asJohn Holland says,is death;somehow,the world
hasto adaptitselfto a conditionof perpetualnovelty,at the edgeof chaos.
"I still haven'tfound the right wordsfor that," saysCowan."fust recently
I wastoyingwith the title of HavelockEllis'sbook,T[e Danceof Lifu. But
that isn't quite right. It isn't a dance.There'snot evena giventempo.So
if anythingwe'regeftingbackto Heraclitus:'EverythingMoves.'A term
like 'sustainable' doesn'treallycapturethat."
Of course,addsCowan, it may be that concepbsuch as the edgeof
chaosand self-organized criticalityaretellingus that ClassA catashophes
are inevitableno matterwhatwe do. "PerBakhasshownthat it's a fairly
fundamentalphenomenonto haveupheavals andavalanches on all scales,
includingthe largest,"he says."And I'm preparedto believethat." But he
also finds reasonfor optimism in this mysterious,seeminglyinexorable
increasein complexityovertime. "kryrtq*r thtP_eqpa\loola at don't
have
-lIyou memory oq cultu,*rg,"he says. "And for me it's an articleof faith that
can aA[memory and accurateinformationfrom generationto gen-
eration-in somebetterway than we have in the past-then you can
accumulatewisdom.I doubtvery much whetherthe world is goingto be
transformedinto a wonderfulparadisefree of traumaand tragedy.But I
think it's a necessary part of a human visionto believewe can shapethe
future. Even if we can'tshapeit totally,I think that we can exercisesome
kind of damagecontrol. Perhaps we can getthe probabilityof catastrophe
to decrease in eachgeneration.For example,ten yearsago,the probability
of nuclearwar wasmaybea few percent.Now it's waydown. Now we're
more concernedwith environmentaland populationcatastrophes. So I
suspectthat if we can iterateday to day,and constantlydo somecourse
Work in Progress 357
corrections,then we will helpprovidea somewhat befterfuturefor society
than if we say,'Well, it's all the handof God.' "
On anotherfront, Cowanis characteristically cautiouswhenit comesto
evaluatingwhat he'saccomplished as a foundingfatherof the SantaFe
Institute."I feelverygoodabouthavingaftempted it," he says."The iury's
still out abouthow successful it will be. Onesignthat it wasn'ttime wasted
is that manypeoplefeelthat we havelegitimizedphysicalscientists'getting I
into whattheythink of as'soft'science-whether we call it economics, or I
socialsciences, or somethingelse.In effect,thesepeoplearegivingup one
of thethingsto whichthey'veheldveryshonglyin theirprofessional careers,
which is to dealonly with phenomena that can be handledanalytically
and rigorously,and they'vegoneinto fieldsthat they'vealwayscriticized
as being 'f.uzzy.'That opensthem to criticismfrom someof their more
conservative colleagues for havingbecomefuzzythemselves. But thenotion
that thereis a disciplinecomingalongcalledthe scienceof complexityhas
madeit morerespectable to do this-to becomeconcerned with questions
that are centralto the nation'swelfareand, for that matter,the welfareof
the world. And I think that this represents a hend that can only payoffto
both the countryand to the academiccommunity.Because if it works,
somethingvery importanthashappened.It represents, to me, a reintegra-
tion of a scientificenterprisethat hasbecomealmosttotally fragmented
overthe pastfewcenturies-a recombiningof the analysisand rigorof the
physicalsciences with the visionof the socialscientists andthe humanists.,,
So far, he adds,this efforthasbeena remarkable success at SantaFe,
particularlyin the economics program.But who knowshow long it can
last.One day,despitethe besteffortsofeveryoneinvolved,eventhe Santa
R Institutecould growsettled,conventionaland old, Institutionsdo. "It \
maybe like a floatingcrapgame,"saysCowan."You mayhaveto shut it I
down in one place and start it up in another.I think it's a necessary$
enterprise. And whetherit's sustained or not here,I think it hasto go on." /

A Moment in the Sunfight

Shordyafter lunchtimeon a Fridayafternoonin lateMay 1991,asthe

clearNew Mexico sunshinefloodedthe tiny courtyardof the Cristo Rey
Convent,ChristopherG. l.angton,Ph.D., sat at one of the blindingly
white patiotablesand did his bestto answerthe questions
of a particularly
persistentreporter.
Dr. Langtonwaslooking markedlymore relaxedand confidentthese
3sB COM4LEXTTy
days.Havingfinally andsuccessfully defendedhis dissertationon the edge
of chaossomesix monthsearlier,in November1990,he had removedan
enormousblackcloudfrom his life-and, not incidentally, hadearneda
scientist's essentialunion card.The SantaFe Institutehad immediately
madehim a memberof its "externalfaculty"-the list of researchers whose
association with the instituteisconsideredquasi-permanent, andwhohave
a strongvoicein its scientificdirection.Indeed,with LosAlamos'budgets
growingincreasingly constrained andsurvival-orientedin theaftermathof
the Cold War, the SantaFe Institutehad becomethe maiorsupportfor
artificiallife. Langtoncouldfeelat homeat the institutein a wayhe never
had before.
He clearlywasn'tthe only onewho felt at home.In the earlyafternoon
sunshinethe courtyardwascrowdedwith visitorsand residentsalike. At
one table, StuartKauffmanwasholding forth with Walter Fontanaand
severalothersabouthis latestideason autocatalysis and the evolutionof
complexity.At another,economics codirectorDavidLanewastalkingwith
his graduatestudent,Francesca Chiaromonte,aboutthe economicspro-
gram'snewestefforl a computerstudythat washying to explorethe dy-
namicsof multiple adaptivefirms engagedin technologicalinnovation.
And at still anothertable, D.oyneFarrnelwastalkingwith severalother
YoungTurls abouthis
reachedthe end of his limited patidEffiiih-ffi6Ti-gEFb-ddge6 and the
bureaucraticpettifoggery up at los Alamos,Farmerhad recentlydecided
that the only sanewayto pursuehis realresearch interestswasto go offfor
a fewyearsand usehis forecasting algorithmsto makesomuchmoneythat
he would neverhaveto write a grantproposalagain.He felt so shongly
aboutit, in fact, that he'd evenhimmedoff his ponytail,the befterto deal
with the business types.
Of course,therewasa certainwisful senseof beingat the end of an era
that Fridayafternoon.For morethan four years,the CristoReyConvent
had been small, primitive, overcrowded, and somehowperfect.But the
instifutewascontinuingto grow,andthefactwasthatthe staffjustcouldn't
keepputtingmoredeslsin the hallway.And in anycase,the leasewasup
andthe CatholicChurch neededits conventback.Sowithin a month, the
SantaFe Institutewasscheduledto moveto largerrentedquartersin the
land of the Lawyers-a new ofhcecomplexout on the Old PecosTrail.
So far asanyonecould tell, it wasa perfectlyfine space,but-well, there
wouldn't be manymorelunchtimeson the sun-drenched patio.
As Langtoncontinuedhisattempbto educatethereporteronthenuances
of artificial life and the edgeof chaos,severalof the institute'syounger
Work in Progress 359
postdocsbegan to pull up chairs around the table, not realizing that this
was supposedto be an interview. The architect of artificial life was some-
thing of a celebrity in their circle, and alwaysworth listening to. The
interviewquickly turned into a generalbull session.How do you recognize
emergencewhen you see it? What makesa collection of entities an indi-
vidual? Everyonehad opinions, and no one seemedto be particularlyshy
about offeringthem.
Melanie Mitchell, a computer sciencepostdocfrom Michigan, where
she is the newestmember of the BACH group, asked,"Are there degrees1
of being an individual?" Lan$on had no idea."l can't think of evolution I
acting on individuals any more," he said. "lt's alwaysacting on an eco-
|
a population,with one part producingsomethinganotherlart
Ljil, I
That sparkedoff otherquestions: Is evolutiona matterof survivalof the
fittestor survivalof the moststable? Or is it justsurvivalof the survivors?
And whatexactlyis adaptation, anyway? The SantaFeline isthatadaptation
requireschangingan internalmodel,i Ia fohn Holland.But is that the
only wayto look at it?
And speaking of emergence, someone asked,is theremorethanone&ind
of emergence? And if so, how many differentkindsare there?langton
started to answer,groundto a halt,andendedup justlaughing."l'm going
to haveto punt on that one," he said."l justdon't havea goodanswer.
All thesetermslike emergence, life, adaptation, complexity-thesearethe
thingswe'restill tryingto 6gureout."
 Bib,liography

For those who want to learn a bit more about complexity, the book and
articles listedbelow will be a start.Few ofthem are intendedfor casualreaders:
the field r'sst// so new that most of the written accountsare in the form of
conferenceproceedingsand iournal articles.Noneftleless,many of the refer-
enceshereshould be reasonablyaccessibleto nonexperb.And virtually all of
them contain further referencesto the technical litenfure.

The SantaFe Institute

and the Sciencesof Complexity in General

Davies, Paul C. W., ed. The New Physics.New York:CambridgeUniversig

Press(1989). Conteins a number of survey articles on condensed-matter
physics,collectivephenomena,nonlineardynamics,and self-org4nization,
all written by leadingscientisbin the field.
fen, Erica, ed. 1989 l-ecturesin ComplexSystems.SantaFe Institute Studies
in the Sciencesof Complexity,lectures vol. 2. RedwoodCity, CA: Addison-
Wesley(1990). This volume, along with the two edited by Daniel Stein
listed below, is basedon lecturesgiven at the SanteFe Institute'sannual
summer school on complexity.They provide a broad overviewof mathe-
matical and computationaltechniques.
Nicolis, Grdgoire, and Ilya Prigogine. Exploing Complexity. New York:
W H. Freeman(1989).
Perelson, Alan S., ed. Theoreticdl lmmunology, Part One and Theoretical
Immunology,Part Two. SanteFe Institute Studiesin the Sciencesof Com-
Bibliography 361
plexig, Proceedingsvols. 2 and J. RedwoodCity, CA: Addison-Wesley
(te88).
Perelson,Alan S., and Stuart A. Kauftnan, eds.Molecular Eyolution on
Rugged Landscdpes:Proteins, RNA, and the Immune System. Santa Fe
InstituteStudiesin the Sciencesof Complexity,Proceedings
vol. 9. Redwood
City, CA: Addison-Wesley(1990).
Pines, David, d. Emerging Syntheses in Science.SantaFe Institute Studies
in the Sciencesof Complexig, Proceedingsvol. l. RedwoodCity, CA:
Addison-Wesley(1986). This is the proceedingsvolume of the institute's
founding workshopsin the fall of 1984.
Prigogine, llya. From Being to Becoming.San Francisco:W. H. Freeman
(re8o).
SantaFe Institute, Bulletin d the SantaFe Inttitute (1987-present).Published
hvo or three times per year,the institute'sbulletin containsextendedinter-
views with some of the major figures there, as well as summariesof its
variousworkshopsand meetings.
Stein, Daniel L., ed,.kctures in the SciencesdComplexity. SantaFe Institute
Studiesin the Sciencesof Complexity,Lecturesvol. L RedwoodCity, CA:
Addison-Wesley( 1989).
Stein, Daniel L., and Lynn Nadel, eds. 1990Itctures in ComplexSysfems.
SantaFe Institute Studiesin the Sciencesof Complexity,Lecturesvol. 3.
RedwoodCity, CA: Addison-Wesley(1991).
Zurek, Woiciech H., ed. Complexity,Entropy,dnd thePhysicsof Infuimation.
SantaFe Institute Studiesin the Sciencesof Complexity,Proceedingsvol.
8. RedwoodCity, CA: Addison-Wesley(1990).

Economicsand the SantaFe EconomicsProgram

Anderson,PhilipW, Kennethf. Arrow, and David Pines,eds.TheEconomy

as an EvolvingComplexSysfem.SantaFe InstituteStudiesin the Sciences
of Complexity,vol. 5. RedwoodCity, CA: Addison-Wesley (1988).This is
the proceedingsvolume of the institute'sfirst big economicsmeeting in
September1987.
Arthur, W. Brian. "PositiveFeedbacks in the Economy." ScientificAmerican
(February1990):92-99.
Arthur, W Brian, et al. EmergentStructures:A Newsletterof the Economic
Research Program(March 1989and August 1990).SantaFe:The SantaFe
Institute. Detailed accounb of the variousproiecb undertakenduring the
program'sfirst l8 months.
 362 Bibliography
Evolution and Order

|udson, Horace Freeland.The Eighth Day of Creation. New yorkr Simon &
Schuster(1979).
. -Kauffman, StuartA. "Antichaosand Adaptation."ScientificAmericdn(August
l99l): 78-8,1.
y Kauftnan, Stuart A. oigins of order: self-organizationdnd selectionin
v Evolution. Oxford: Oxford UniversityPress(1992).

Neural Networks,Cenetic Algorithms,

ClassifierSystems,and Coevolution

4/ )fAnderson, JamesA., and EdwardRosenfeld,eds.Neurocom puting: Founda-

tionsof Ruearcft.Cambridge,MA: MIT Press(1988).Excerptsfrom many
of the basicbooksand papersin the field of neural networks.Among them
are some of the original works of McCulloch and pitb, Hebb, and von
Neumann. Also included is the 6rst neural network paperby Rochester,
, Holland, and their colleagues.
b"l y' Axelrod, Robert.The Evolutionof cooperation New York BasicBooks(
I 984).
Forrest,Stephanie, ed. EmergentComputation:Self-Organizing,Coilective,
and cmperative Phenomenain Natural and Artificial computing Nefworfrs.
Cambridge,MA: MIT Press(1991).The proceedings of a confeiencespon-
sored by Los Alamos' Center for Nonlinear Studies.Contains p"p"ri by
Chris L,angtonon computationat the edgeof chaos,Doyne Farmer'onthe
"RosettaStonefor Connectionism,"and much more.
Goldberg,David E. Cenetic Algoithmsin Search,Optimization,and Machine
. Izaming. Reading,MA: Addison-Wedey(1989).
./ Holland, fohn H. Adaptation inNatural and Artificiat Systems.Ann Arbor:
Universityof MichiganPress(1975).
Holland, fohn H., Keith f . Holyoak,RichardE. Nisbeft,and paul R. Thagard.
Induction: Prrcessesof lnference,L*aming, and Discotery.Cambridge,MA:
MIT Press(1986).

Cellular Automata,Artificial Life, the Edgeof Chaos,

and Self-OrganizedCriticality

.-Bak, Per, and Kan Chen. "Self-OrganizedCriticality." Scientific Ameican

( ) a n u a r yl 9 9 l ) : 4 6 - 5 3 .
Burks, Arthur W., ed. Essayson CeIIuIar Automata. Champaign-Urbana:
Universityof Illinois Press(1970).
 Bibliography g0g
Dewdney,A. K., "ComputerRecreations." ScientificAmeican (May 1985).
A basicdiscussion of cellularautomataand computation.
---
Farmer,Doyne, Alan lapedes, Norman Packard,and Burton Wendroff, eds.
Evolution, Cames,and l*aming. Amsterdam:North-Holland (1986). [Re-
printed from Physica D 22D (1986)Nos- l-3,1 The proceedings of a con-
ferencesponsoredby I-os Alamos'Center for Nonlinear Studies.Includes
one of the first public discussionsof Farmer,Kauftnan, and Packard'sau-
tocatalyticorigin-ofJife model, as well as one of the first presentationsof
Chris Langton'sdiscoveryof a phasetransitionin cellular automata.
'r/ Langton, ChristopherG., ed. Artificial Life. SantaFe InstituteStudiesin the
Sciencesof Complexity, Proceedingsvol. 6. RedwoodCity, CA: Addison-
Wesley(1989).Proceedingsof the first artificial life workshopin September
1987. IncludesChris l,angton'sintroduction and overviewof the artificial
. life concept,as well as extensivebibliographiesof the field.
, fangton, ChristopherG., CharlesThylor,f. Doyne Farmer,and SteenRass-
1' mussen,eds. Artifrcial Life II. SantaFe InstituteStudiesin the Sciencesof
Complexig, Proceedingsvol. 10. RedwoodCig CA: Addison-Wesley
(1992).Proceedings ofthe secondartificiallife workshopin 1990.lncludes
papersby Chris Langton on the edgeofchaos, StuartKauftnan and Sonke
fohnsenon coevolutionto the edgeof chaos,WalterFontanaon "algorithmic
chemistry,"and Rick Bagley,Doyne Farmer,and Walter Fontanaon further
. developmentof the autocatalyticorigin-ofJife model.
{ von Neumann, fohn. Theoryd Self-Reproducing Automata.Completedand
editedby Arthur W. Burks.Champaign-Urbana: Universityof lllinois Press
(r966).
Wolfram, Stephen."Computer Softwarein Scienceand Mathematics."Sci-
entificAmerican(September1984).Includesa basicdiscussionof cellular
automata.
- Wolfram, Stephen, ed. Theoryand Applicationsof Cellular Automata. Sin-
gapore:World Scientific(1986).Reprintsof many of the cellularautomata
paperspublishedby Wolfram and his colleaguesin the early 1980s.
 Acknowledgments

To the dozensof people who have given so generouslyof their time and
patienceto help makethis book what it is: a heardeltthank'you. All booksare
a team effort, and this one is more so than most, Many of you have received
only a passingmention in the text, or none at all. Believeme, that doesnot
diminish either the importanceof your contribution or the gratitudeI feel.
To Brian Arthur, GeorgeCowan, fohn Holland, Stuart Kauftnan, Chris
Langton,Doyne Farmer,Murray Gell-Mann, KennethArrow, Phil Anderson,
and David Pines: a special thank-you for enduring endlessinterviews and
telephonecalls, for sharingyour own struggles,for educatingme in the ways
of complexity and, not least, for readingvariousdrafts of the manuscriptin
whole or in part. For me, at least,the processhas been a ioy. I hope it has
been for you as well.
To Ed Knapp, Mike Simmons,and the staffof the SanteFe Institute:thank
you for hospitalityand assistance aboveand beyondthe call ofduty. The Santa
Fe Institute is truly a place to come home to.
To my agent, PeterMabon: thank you for your guidance,advice, and re-
assurance,which alwayshad a calming infuence on a frequently nervous
author.
To my editor, Gary Luke, and the production staff at Simon & Schuster:
thank you for your enthusiasticsupport-and your yeomanefforb on behalf
of a book that wasvery late.
And to Amy: thankyou. For everything.
lndex
Academyof Sciences,Budapest,48 on emergentptoperties,82-83
Adams, RobertMcCormack, 86, 88, 9l meetingwith Reed,92-93
Adaptation, 146. SeealsoComplex on reductionism,8l
adaptivesystems Anthropology,86
ariving at edgeofchaos through, 309 Arrow, Kenneth, lZ, 74, 52, 97, l)6,
coevolutionand, ]09 tt7-t8,140, 142, l+t, r9r,744,
evolution and, 179 215, 246, 250, tZ', t26, 140
Holland'sanalysisof, 144-19, 166 Arthur and. 52-53
in humanmind, 176 on evolutionaryeconomics,J26-27
Adaptiveagents,145, 176 on increasingteturns economics,
in evolution,179 t95-96, tZ7
computersimulationof, l8l-91 personalitycharacteristics,52
Hebb'ssearchfor theory of, 180 on SantaFe Institute versusCowles
Hofland on, 176-77 Foundationresearchprograms,327
in inductivemode, 253, 254 Arthur, SusanPeterson,15, 29, ll7
simulated,l8l-91 Arthur, William Brian, l5-54, 98, 240
Adaptivecomputation, 341-'13 academicbackground,li-16, 19-28
origin of term, 342 at American EconomicsAssociation
Adaptivesystems.SeeComplex adaptive meeting,50
systems Arrow and, 52-51
Agents,in adaptivesystems.SeeAdap- in Bangladesh,26-27
tive agenb codirectonhipof SantaFe Institute of-
Agnew, Harold, 62 fered to, 199
AIDS, 60, 248 "Competing Technologies,Increasing
Alchemymodel, ll5-16 Retums,and lockln by Historical
Anderson, Herb, 84 Events."49-50
Anderson,loyce,92 on complexityas scientific revolution,
Anderson,Philip, 12, 79-8\, 95-96, 727-70
tt6, t37, ltg, t40, l4'' 149, 195, asdirectorof SanhFelnstitute,244-
244, )04,318 245.247-18,tZ4-t5
economicsworkshoporganizedby, 97 24, 25-26
dissertation,
 370 lndex
Arthur, William Brian (Conf.) Kauffmanon, lll, 315-18
StewartDreyfus and, 24 node-and-connect.ionshucture of. 290
in Germany,21, 25 selectionand adaptationin, ll7-18
Hofland and, 144, 147-48. 194.242- simulating,ll2-33
243 Automatatheory, I09
ideasfor SantaFe Institute, 245-46 Axelrod,Rohrt, 176, 2$,277
on increasingreturnseconomics,29, Evolution of Cooperation, The, 264
14_46
at InternationalInstitute for Applied
SystemsAnalysis,28-47 Bagley,Richard,ltl, 297, 314
Kauffmanand, 100-101, lt7-20. Bak, Per, 304-6
113-i4, tJ7 Balance,betweenordet and chaos.294
personalitycharacteristics, Bangladcshpopulationand develop.
49
point of view, 38-46 ment,26-27
on SantaFe Institute approach,326- Behavior
))5 collective,emergent,95
at SantaFe Institute economicswork- patternsof, 66
shops,99-101, ll7-20, lll. l16- Belin, Alletta, 284
t43, t97-99 Bethe,Hans,57,68
speculationsof, 127-)5 Big Bangtheory,62, 64, 76, 81, 295, ll8
StanfordUniversityprofessorship, Biology,16,29-30
46-
,2 developmental,106
as visiting professorat SantaFe Insti- molecular,62, 76
tute,5l-54 physicists and,62-67
A*ificial intelligence,71, 157, 180, Bloch, Eric, 90
185. Seea/so Neural netwotk simu- Bohr, Niels. 68
lations Boldrin, Michele, 196
A*ificiaf life, 199-240, 277-81, 299 Booker,Lashon, l9l
Farmeron, 284-99 Bottom-up organization,and evolution,
l,angtonand, 215-40 294
langton'suseof term, 198,2\7 Branscomb,Lnuis, 87
metaphorof, 329 8rock, William, 246
possibifityof creation of, 282'84 Brookingslnstitute, 352
"rcal," 283-84 Bucket-brigade algorithm,188-89, 19l,
rulesof, 281-82 272, 291
scienceof, langton as founder of, Bulletin of the Atomic Scientists, Tfte,
)22, )18-59 foundingof, 58
value oi 289 Burks,A*hur, 160-61,162, 165,22i,
Artificial Life Worlshop, los Alamos, 224
t99-240
Asimov,Isaac,"The Final Question," Cairns-Smith, Graham,219
286 Calvin,Melvin,128
AspenCenter for Physics,71, 83 Campbell,David,122
Atomic EnergyCommission,58 Campbell, fack,79
Augusztinovics, Maria, 48 Carruthers, Peter,69-70, 84, 87, 749
Autocatalyticsets/autocatalytic
models, supportof SantaFeftrstitute,69-70
t24-28, t70, t12_)7, t48,235_ Cell, asself-organizing
system,34
236, Z9l Cellassemblies, 182
Alchemyprogramand, 3l 5 Cellularautomata, 87, 217,219
analogyto economics,125-27 Class I rules,225,228,2t1,2t]
emergencein, 289, 296-97 ClassIf rules,225-26,ZZ8,73l.z]^t
 lndex 371
ClassIII rules,226,227,228,82, C o d d ,T e d ,Z l 7 , 2 Z 0 , 2 Z l
Lt5 Coevolution. J09
IV rules,226,227,228-30,
Class edge-of-chaos principle and, 294
B1 as force for emergence,259-60
phasetransition in,
edge-of-chaos Coevolutionarymodels, Z9Z, )09-lt.
292-9) See alsoEcho model
evolution and,227 edge-of-chaos phasehansition, 3l I
Came of Life as specialcaseof, 219 Cognition
l,angton and,72l-22 computermodel of, I8l-82
nonlinear systemsand, 225 learningand, 180
rules followed by, 225-)0 Cognitive science,71, 76
self-reproductionand, 721-22, 279, as interdisciplinary,7l
281 Cognitive theory, 192-203
Cellular Automata (Codd, editor), 217 clasifier systemand, 190, I9l
Cellular differentiation, 106 Cohen, Michael, 175,176, 189
Change, power-lawcascades of' ]08-9 Colgate,Stirling, 70
Chaos. 65-66, 3ll. Seea/soChaosthe- Combat, as metaphorfor interaction,
ory 260
classifiersystemand, 193 Competition, in adaptivesystems,185
evolution and,103 and cooperation,l8r, 262-6,
hansition to, 230 (seealso Edge of in evolution, 262
chaos) Compfex adaptivesystems,294-99
Ciaos (Gleick), l3l, 287 buitding block structureand, 170
Chaostheory, l3l-12, ]29. Seealso chemishyand, 314-15
Edge of chaos economyas example of, 145 (seealso
as emergingscience,287 Economics;Economy)
Farmeron, 287-88 emergentbehavior in (seeEmergence)
Los Alamos as world centerfor, 70 generalizedgenotypeversusPhenotyP€
and self-organizingcomplex systems, of,281-82
t2 Holland's work and, 166-94
useof, 95 levels of organization, 145-46
Chemical Evolution (Calvin), 128 nichesof, I47
Chemishymodel, 314-15 prediction and, 146, 176-77
Chenery,Hollis, l17, 195-96 realignmentof, 146
Chessanalogy,I50-51 rulesof. 280-81
Chiaromonte, Francesca,J58 Complex systems.SeealsoComplex
Church, Alanzo,278 adaptivesystems;Complexity; Edge
Citicorp, 91, 92, 94-95, 96, 181-94, of chaos
242-43, 244, 252, 258, ll8, 140 characteristics of, I l-12
Classifiersystems,188-89, 272, 289, computersand, 67
290 defined,ll
Arthur on, 272-73 spontaneousself-organizing,I I
bucket-brigadealgorithm, 188-90, Complexity, 294. Seea/soComplexity
19l,291 research;Complexity theory
\ cognitivetheory and, 192-93 in economics,18
\ deusex machina, 258-59 at edgeofchaos, 293
systemand, 193 growthof, 6r, 165,294-99,316-18,
".emergent
geneticalgorithm,188-90, l9l, I99, l19
t03
242-4t,Z5Z,272, phasetransitionsand, 732, 283
node-and-connection shuctureof, 290 SantaFe Institute approachesto, 248
as self-adapting,273 scientists'fascinationwith. 6l-64
 372 Index
Complexity (Cont.) first lectureson, l5l
Wolfram on, 86-87 newkindsof,71,76
Complexity research rule-basedsystems,Newell-Simon.
computersin, 87 I8l-8t
exampleof questionsdealt with b* 9- softwarepackages,debugging,282
ll _
Connectionism,289-92
troublein definine,9 defined,289
Complexig theory models,289
metaphors and, ll4 Connectionisttheoryof memory, 158-
asscientificrevolution,J27-10 l t 9 , 1 6 0 ,l 8 l
asworld view, Jll Connections,289-90
Computation(s), of complexadaptivesys- adjusting/changing, 291
tems, 106, 212, 280-Bt, 31t-4i powerof, in learningand evolution,
asessenceof life, 212, 280 29t.292
Computergraphics,95 sparse,303
Computersimulation,63-64, gj, 177, Consistency, in adaptivesystems,185
251. Seealso Echo model; Model- Conway,fohn, 201, 219
ing Cooperation
of adaptiveagenb, t8l-93 and competition, 262-65
of autocatalysis,B2-13 (seealso Au- in evolution, 262-64
-tocatalyticsets/autocatalytic
models) in real world, 263
of complex adaptivesystems,needfor, Cost-benefitanalyses,33I-32
t47 Cowan, GeorgeA., l), 57-69, l4t.
connectionist,289 241, 276-77, 95_17
of cooperation,262-65 academicbackground,5t-56, 57
ofearthquakes,305-6 dissatisfaction with academicapproach
of economics,93,95, 2$, ZSl, ZiZ* to science,60-61
255,268-74 Gell-Mann and,77-74, ]3i, 717-14
ofecosystems, 309-13 ideasfor interdisciplinaryinstitute for
emergentbehavior in, 242, 270 study of complexity,68-69
ofevolution, 256-58 at L,osAlamos, ,7-r9, 66-69
of flocking behaviorof birds, Z4l-4). as ManhattanProjectscientist,56-58
279, 129 personalig characteristics,54*5 5
ofgeneticnetworks,lll-14, l16 at Santa,FeInstifute, after presidency,
of immune system,266 t > > - >/
lessonlearnedabout complex adaptive as SantaFe Institute chairman of
systemsfrom, 279 boardand president,77-71,77-79,
local controlin, 280 84-8r.735-14
of neuralnetworks,I ll, lt9-60, White HouseScienceCouncil and.
l8l,289-90 59-61
origin-ofJife,124-25, 313, )14 Cowan,|ack,85, 87,134
of plant development,279-80 Kauffmanand, I 16
of policy options, 266-67 Cowles Foundation,j27
rule-based systems,l8l-82 Crick, Francis,29, 67.,Zl9
of stockmarket, 266, 27)-74, 729 Critical states,self-organized,304-7
value of, 289 Crutchfield, lames, 287, 346
Computer virr-rses, 238, 279, 78\
propertiesof, 283 Danceof Life, The (Ellis), 156
Computers Darwin, Charles,75, 102, 179,287,295
in complexityresearch,63, 87 David,Paul,50
conflict resolutionstrategies,184-85 Dawkins,Richard,?32,?38, 26l-62
development of, at IBM: 155-t7 Debreu, Gerard, 24
Index 373
Default-hierarchymodels, 192, 19) historians,on increasingreturnseco-
Detectors,182-83 nomics, 50
Dewdney,A. K., 279 lack of attention to empirical fact, l4l
DNA, l2l, 122 openness to new view of economics,
as molecular-scalecomputer, 3l )25-26
structure of, 29-30, 62, 219 theoretical,on increasingreturnseco-
Von Neumann'sanalysisof self-repro- nomics,49-50
ductionand, 219 Economy
Doofen,Cary,274 as complex adaptivesystem,145
Double Oral Auction Tournament, 271 computer-simulationoi with adaptive
Dreyfus, Stuart, 24 agents,243, Z5l, 257-55
Dynamical behavior,Wolfram's classes Third World, pafternsof develop-
ot- /1>-t> ment, 195-96
Dynamical systems,287 world, computer simulationsof, 9J
nonlinear, and cellular automata, Ecosystems
225-30 (seealsoCellular automata; computersimulationoi j09-ll
Nonlineardynamics) at edgeofchaos, 108-ll
Edge of chaos,210-31, ?92-94,
t56
Earthquakes as balancepint, 12
computermodels,305*6 cellularautomataat, 714
diskibution of, 305 complexityand, 295
power law for, 305-6 defined.l2
self-organized criticalityand, l0l-4 ecosystems at, 108-ll
EaslWest PopulationInstitute (Hono- Farmeron, 293
lulu), 25, 29 as l,anglon'sphrase,230
Echo model, 7ffi-62, 266, 292, 709, langton on, 302
310 origin oflife and, 215
Econometricmodels,9l as region rather than boundary, 302
Economics self-organizedcriticality and, 107
analogyto autocatalysis models, 125- theory of, difficulty of testingin real
r27 world, 107-8
analogyto chess,150*51 EDVAC, 16I
computersimulationsof, 9), 95, 24), Effectors,182-81
2rr, 252-55,268-74 Ehrlich. Paul, 25
evolutionaryapproachto, 252 Eighth Day of Creation, Tie (fudson),
"increasingreturns" conceptin (see 29, 74
Increasingreturnseconomics) Einstein,Albert, l0l
instabilitvin. I7 Ellis, Havelock,156
neoclassicalview of, 17, 22-24, 77, Emergence,152, ZA0, 2t8
74-)5, )7-38, 43, 140-41, 378, and adaptation,149, I8l-84
ll0 Andersonon, 82-83
new (seeIncreasingreturnseconomics) bottom-up, 278-79
asnonlinear,65 coevolutionand, 259-60
SantaFe Institute approachto, 250, in computersimulations,242,270
252, 7.54-5r, 110 (seea/soIncreas- Farmeron, 288-92
ing returnseconomics) hierarchicalbuilding-block structure
SantaR Institute programin, 95-98 and, 169-70
asself-organizing system,34 meaningof, 242
spinglassmetaphor,138-39 Emergentmodels,270
Economisk default-hierarchiesmodelsas. I9l
attitudesof. l'12 E N T A C .l 5 l . 1 6 l
 374 Index
Enhopy, 286, 288 positive,ll8, ll9, 196
Equifibrium, 165, 167 rule-basedsystemsand, 183
complexadaptivesystemsand, 147 stockmarket and, 196
in economics,255 Feldman,Marc, 87, I l4
Ermoliev,Yu.ri,46,246 Rrmi, Enrico, 56, 68
Ervin, Frank,200, 205 Feynman,Richard,68
Evolution. SeealsoCoevolution Fisher, R. A., 163-61, 166-67, ZS9
cellular automataand, 222 Fishlow, 50
competitionand cooperationin, 262- Flocking-of-birdscomputer simulation,
265 2 4 t - 4 7 . 2 7 9 .1 2 9
computer
simulation
of,256-58 Fogelman-Soule,Frangoise,301
developmentof theory of, 299 Fontana,Walter,297, ll4-15, lt8
edgeof chaosand, 103 Forrest;Stephanie,272
Fisher versusHebb on, 164-65 Free-market ideal, 47 -48
hierarchyof control in, 294 increasingrefums economicsand, 48
as longestterm adaptation,Hebb's Freeman,Christopher,43
study of, 166-74 Friedez,Gideon. 221
movementof, towardgreatercomplex-
ity,296 Game of Life, Z0l-1, 209, Zl8, ZZ9,
paradoxin, 262 z7t,zJ4
self-organizedcriticality and, 108-9 artificiallifeand,202-l
Evolution of Cnpration, The,261 ascellularautomaton, 219
"Evolution, Games,and lrarning" con- Gamestheory,150,262
ference,l,os Alamos. 1,18 Gamow,Ceorge,62
"Evolutionaryarmsrace," 261-62 Geanakoplos, lohn,725
Expectations Gefl-Mann,Murray,12,53-54,73-77,
in economics,l4l 80,8t, 83,84, 136,t49,24r,258.
rational.270-71 26t,284,37t,346
Expertsystems,180, l8l Cowanand,7,-74, )35,347-51
Exploitationlearning, 291 asheadof SantaFeInstituteboardof
Explorationleaming, 291, Jlz trustees, 89-90
"Extendedtransients,"229,?30, B1 asheadof SantaFe lnstituteScience
Board,90
Rrmer, Doyne, l)l-12, l?'1, 148,23j. on Holland,148-,19
246,258, 276, 284-99. 3t3. tt4, interests
of, 71-75,256-58
329,)r8 personality
characteristics,
74
"Artificial Life The Coming Evolu- Generalizedgenotype, 28I -82
tion," 284, 285 Ceneralizedphenotype, 281-82
on chaostheorx 287-88 Genes,regulatory, 106,107,108
on emergence,288-92 Geneticalgorithms,188-90,l9l, 242-
Langtonand,235, )22 243,2r2, 272.)03
personalitycharacteristics,285 international conferenceson. 199
"RosettaStonefor Connectionism," Geneticcircuih,106
290,294,312,34t_42 Geneticcode,29-10
on science,Jl8-19 strucfureof, 62
on self-organization,286-88 Ceneticnetworkmodels.
Feedback,29 289
Holland on, 180 neuralnetworlaand, lll
intemal, l8l orderly statesof, I I l-l l
naturalselectionand, 179 phase-transitionlike behaviorin, 301
negative,139 regulatory,I l0-ll, 120-21
patternsand, 16 Geneticorder,Kauffmanon, 108-ll
Index 375
Genetics Arthur and, 144, 117-48, 194,242-
Fisheron, 16l-64 243
molecular,106-7 Burksand, 16l-62
Cenomes,106*8, 290 on emergence and adaptation,148,
regulatorysystems,I I 5- l6 149-50
Getg, Gordon, 354 on equilibrium, 167
Cleick, fames,Lll, 287 geneticalgorithm,I88-90, l9l, 247-
Global behavior,106 z4l, 252, Z7Z, 101
Global sustainabilityprogram, 349-52 and Hebb'stheory, 158-60
Ccidel,Kurt, 778,328 at IBM, 155-60
Goldberg,David, l9l-92, 193 Langton and,224
Goldenberg,Edie, 144-45 MIT seniorthesis,l5l-55
Goldstine,Herman, 16l on personalcomputen, 189-90
Coodwin, Brian, I I6 programmingknowledgeof, 154-55
Gould, Stephenf ., 308 SantaFe Institute Maxwell professor-
Cutenberg,Beno, 105 ship offeredto,741,144
Holyoak,Keith, 192-91
Hahn, Frank,247,250, Zrr,175 Hubler, Alfred, 146
Haken,Hermann, 287
Hamilton, William, 175, 176
Hebb. Donald O., 157-60, 169,176, tBM. 156-57,159
182, t83-84, 188 computerdevelopment,I 55-56
theoryof learningand memory, 164- Hollandat, 155-60
165 IIASA. SeeInternationa'lInstitute for
Hecker,Siegfred,275 AppliedSystems Analysis
Heidegger,Martin, 329 Immune systemmodel, 289
"Heraclitians," 335 Increasingrefumseconomics,17-18,
Heraclitus,J8 79, )4-18, 42
Hewlett-Packard, 69 acceptanceof, 195-96, 325-26
Hicks,fohn R.,47 Arthur on, \8-46 (seealso Arthur,
High technologyindustries William Brian)
governmentpolicyand, 41-42 free-marketideal and, 48
standardsand,,42-47 Kauffmanon, Il8
Historicalaccidents,\6-17, 19, 4l "legitimization'rof, by physicists,139-
Holistic approach,versusreductionism, 140
60-61 Marshall on,41-:45
Holland, fohn H., 144-94, 197, ZZ0, mathematicalanalysisof, 14-46
246. 2t2, 257-5+, 278-79, 289, propertiesof,16-J7
290,295,109,l4l, '12, )46 technologicalchangeand, I 19-20
academicbackground,I 5l-55, l6l- Induction.25t,2t4
162. t66. t75 Induction (Holland, Holyoak, Nisbett,
Adaptationin Natural and Artificial and Thagard),'191,272, )44
Systems,174 Information processing,I 53
on adaptivesystems,144-47, 166-94 Insight, 193
at Artificial Life Workshop,los Ala- Interdisciplinary research
mos, 198, 199-200,2]8-40 cognitivescienceas, 7l
bucket-brigade algorithm,188-90, universitiesand, 67
t9t, z7z,791 Internalcombustionengine,40-41
on classifiersystems(seeClassifiersys- InternationalInstitute for Applied Sys-
tems) temsAnalysis, 28, )2, 39-40
Gell-Mann and, 256-57 Arthur at, 28-47
at SantaFe Institute, 2'14 InternationalSchumpeterSociety,326
 376 Index
Jacob,Franqois,
3l intuition of, 202-7
Japan, economic of, ,13
development at Los Alamos Center for Nonlinear
fohnsen, Sonke,
103,109-12 Studies,2)6-40,722
foule,fames,298 marriage,216
fudson,HoraceFreeland,
29-30,31, speculationsof, 320-21, t29, ?30,
34 3t8*59
Leaming,291
Kaniovski, Yuri, 46, 246 cognitionand, 180
Kauftnan,Stuart,162,l9+,235,Z4f, computermodpling of, 185-89
249,257,2r8, 2r9, 289,299_104. in economics,modeling effecb of,
329,111,316,358 t>t
Arthurand, 100,ll7-20, 133-34. Hebb'sneurophysiologicaltheory of,
t)7 155-56
on autocatalysis, l2l-ll, ll5-18 lack of interestin, l8l-84
automobile accident, 321 l,evins, Dick, I 16
background, 103- 13 Lewontin,Richard,116, ll5
Licklider,I. C. R., I57, lt8, 159
JackCowanand,116
deathofdaughter, l17, ll4-15 Life
ecosystem modelof, 309-13 artificial (seeArtificial life)
Farmer and,288 behaviorof. 280
fruit fly development work,129-J0 as computation, 272, 280
Goodwinand,I 16 computersimulation of (seeArtificial
HorsehilFallshome,134 life)
marriage,I 14 edge-of-chaos phasetransitionand,
McCullough and,lll-16 102-1
on origin oflife, l2l-29 as emergentproperty,82, 278
Oigins of Order,The, )00 organizationas essenceof, 288, 292
personality characteristicsof, l0 l-Z orfgrnol, tt>, /.>/->6
on self-organized criticality,307-8 Lindenmeyer,Aristid, 238-39,
speculations of, 321-22 279
Kaysen, Carl, 92 Linear systems,versusnonlinear sys-
Keck,William,Jr.,154 tems, 6,f-65. See alsoReduciion-
Keynes, ism
fohnMaynard, 23
Lloyd, Seth, 346
Keyworth,Ceorge(lay),II, 60,69
Knapp,Ed, 88, 349,353-54 Lock-in, ofeconomicadvantage, 36, 37,
^ FeInstitutepresident, 354* 39-40,50,139
i;i" examples
of,40-42
Knief,Byron,92 fapaneseeconomy and, 13-44
Koopmans, Tialling,24 Uccello clock and. 40
Kopal,Zednek,154 los Alamos, ,4, 60, 61, 62
Artificial Life Worlshop, 199-240
Carruthersat. 69-70
lane,David,272,325,)r8 Center for Nonlinear Studies,216,
langton,Christopher C., 199*240, aat
Lta
257,258,274-84,291,3u, tZ} chaostheory research,70
andartificiallife,science
of,715-10, Complex Systemsgroup, 285-86
322,358-19 GeorgeCowan at, 57*59, 66-69
background,200-38 "Evolution,Games,and Learning"
Farmerand,2J5 conference,235
hang-glidingaccidentandrecovery, missionof, 68'
208-lI nonlinear dynamicsresearchat, 66-
Hollandand.224 70
Index 377
MacArthur Institute.]r0. 3r2 neuralnetworkmodels,I I l, I 59-60,
McCarthy,fohn, 157 18r,289-90
McCullough,Warren,I l3-16 nodeand connectionstructureof, 289
death, I 16 (seealn Connectionism)
Kauffmanand, 113-16 random.of reactions.lJ2
McKenzie,Lionel, 2'1 propertiesol 120-21
statistical
McKinseyand Company,20-21,27, Ncural networksimulations.lll. 159-
24,25 160. l8l. 289-90
McMahon bill (1946).t8 McCullough-Pittsmodel, I l3
McNicoll, Ceoffrcy.76-27 Neurosciencc,l6
Manhattan Prciect, 67-68 Newell,Allen, 168, l8l
MANIAC, 70 Nisbett.Richard.192-91
Marimon, Ramon, 746, 270,271 Nonlincardynamics,64, 65-66,76,
Marshall, Alfred, 44 l2l
Mathematics.fermentin. 7l Carruthersand, 69-70
Maxwell, Robert,341-46, 355 cellular automata and, 225
MaynardSmith,tohn, l16, 120,100, Easternphilosophyand, 331
il l formation of patternsand, 65
Memory, Hebb'sneurophysiological the- [,os Alamos researchon, 66-70
ory of, 155-56 self-organizingsystems as, 65
Menezes,Victor, 92 Novelty, perpetual, 147
Metaphors,and complexsystems,329, Nuclearpower,4l-43, 60
]J4 wastedisposal,60
Metropolis, Nick, 71, 72,76,78-79 weaponresearch,55-59
supportof SantaFe Institute, 70-71
Miller, Stanley,I2l, 128, lt2 Operationsresearch
Miller, fohn, 246-47, 266, 273 defined, 20
Mind, studyof. SeeCognitivescience Williams as studentof, 20
Minsky,Marvin, ll5 Oppenheimer,I. Robert, 58, 68
Mitchell. Melanie. 159 Order,29), 291, )11
Modeling/models,76. Seea/soCom- Christianityand, ll0
puter simulations genetic,108-13
of complexadaptivesystems,needfor, Kauffmanon, 102
t47 from molecularchaos,124
econometric,9l Prigogineon,12-33
emergent,252,25) Organization.Seealn Self-organization/
internal.146.147 self-orga ni zing systems
Kauftnan and, 105 as essenceof life, 288, 292
mental. 177. 278 levelsof, in complexadaptivesystems,
predictionand,177 t45-46
Monod, Jacques, 3l propertiesof, 288
Moravec, Hans, 219 Origin of life, 121-25, 119
Morris, Robert,2J8 autocatalytic setand, 124-25
Mountain Bell, 78 computersimulatirinof, 121-25, )13,
)14
Nagle,Darragh,72
Naturallaw, beliefin, 8l Packard.David. 69
Nelson,Richard,252 Packard,Norman, l3l, 148,235,287,
Networks 288, 302-j, 3t3
"frozencomponents"in, l0l Palmer,Richard,246, 252, 272,271
geneticnetworkmodels,ll0-14, 289, Papert,Seymour,I 15
301,303 Pathdependence, 50, 139
 378 Index
kttems understandingversus,J06
ofactivegenes,31, 107 Prigogine,llya, 32, 287
Arthur's awarenessof, 27-28, 1l-32 Pinciples of Economiu (Marshall), 44
of behavior,66 Prisoners'Dilemma, 262, 26'-65
computersimulation of, 61-64 iterated,263
ofconnections, 289-90 TIT FOR TAT strategy,264-6,
of economicdevelopment,195-96 Processcontrol, 278
increasingcomplexityof, 66 Programminglanguages,154
nonlineardynamicsand, 65 Prusinkiewcz,Prezemyslaq279
reasonsfor, 36-J7 Psychologists,71, 195
Perelson, Alan, 248, 289 Purmort,Wally, 165
PergamonPress,1,14,346
Periodicattractors,226 Qu"rkt, 53-51,74, 311, 348
Perpetualnovelty, 1,17 QWERTY keyboard,as exampleof lock-
Phasetransitions,228-10, 213-3r, i n , 1 5 ,4 0 , 4 2 , 5 0 , 1 3 9
292-94
to chaos,210-31 Rationalexpectations,Sargents'swork
complexity and, 232, 29j on,270-71
edgeof chaos, 292-9), 302-) Rationality, l4l-42
origin of life and, 2)5 bounded,250-51
self-organizedcriticality and, 107 perfect,250-51
Physicists,62,287 Reaganadministration, 46, 47
attitudesof,112-4] ' "Star Wars"
StrategicDefenseInitia-
economishand, 119-40, l4l-43 tive, 60
and empirical fact, 140 Red Queen hypothesis,26J
Physics,16-17 Reductionism.Seeallo Linear systems
Arthur's studyof, 32 Andenon on, 8l
Pines,David, 7l-72, 75, 79, 80, 81, versuscomplexity, 329
92,96-97,196-97,
246,n6-71, as deadend approach,60, 6l
t46 versusholism, 60-61
academicbackground,7l Reed,/ohn, 91,93, l)6, l1), 119, 196,
at Los Alamos. 7l t97, 241, 215
Pipelinesimulation,19l-93 meetingwith SantaFe Institute scien-
Pitts, Wdlter, I 13 tists, 92-96
"Platonists,"335 personalitycharacteristics,92
Policy-making, Arthur on, JJ0-34 Reichenbach, Hans, 2ll
Political policy, Taoistapproachto,771- Reitman,Judy,190-91
3t4 Religion, )]0. SeeolsoTaoism
Polynomial-time
algorithms,
233 Farmeron, ll9
Council(NewYork),26, 27,
Population Researchstyle, Arthur on, 29
t6 Reynolds,Craig, 741-4\, 779
Populationgrowth, Third World, 75-27 Rice, Stuart, 127,128
PowerJawbehavior,105 Richardson,Ginger, 100
global, 106 Richter, Charles, ]05
possibilityof testing, 308-9 Rio Grande Institute, 79, 89. Seealrc
Predictability,)9, 142, 729 SantaFe Institute
Prediction Riolo. Rick. 189-90
complexadaptivesystemsand, 146, Risking. SeeExploration learning
176-77 RNA, I2I
implicit, l,16 Robotics.281
modelsand, 176-79 Rochester,Nathaniel, 159, 160
in relationto science.255 Rockefeller,fohn D.; IIl, 26
lndex 379
Roessler,
Otto, 128 seniorfellows at, 69-77
Rota,Gian-Carlo,70-7l, 72,75 staffing of, 99
Rothenberg, 50 Universityof Michigan, Ann Arbor,
Ruelle,David,lJ7, 195 link with, 145
RussellSageFoundation,91, 92 Sargeant, Tom, 116-37, 246,266, Z7l,
Rust,fohn, 273 271
Scheinkman,Josd,I17, 196
Salmon,Wesley, 213 Scholes,Christopher,35
Samuel, Arthur,157,165,180,194 Schriidinger,Erwin, 62
Samuelson, Paul,24 Schumpeter,foseph,46, 119,2r2, t26
Sandpile metapho4304-5 Schwartz,Douglas, 85-86
SantaFeInstitute,12, ll Science,318-19
accomplishments of, 339 and prediction,255
approach of, 247-55, 326-)5 Scientists
approach to complexity issue,248 typesof, 315
approach to economics, 250-55 world views,ll4-15
Arthurasvisitingfellowat, 57-54 Segura-Langton, Elvih, 216, 217, 221,
Arthur'sintroductionto, 57-54 221-74, 86
ArtificialLife Worhhop,279 Selection,versusself-organization,100-
ArtificalLife II Workshop, 722,32) tOl, lll
atmosphere of, 100,249-50,329 Self-assembly, 218
attainment of name,89 Self-fulfilling prophecy,274
building,99-100, 158 Self-organization/self-organizing systems,
ascatalyst, 325 317
ComplexSystems SummerSchool, adaptive,I I
7)9 autocatalyticsetsand, 133
conhibutions to complexity science, defined, 102
324-27 developmentof theory of, 299
Cowanand,5)-16, 66-69,78-79 economicsas, 34
(seeaIsoCowan,GeorgeA.) Farmeron, 286-88
DoubleOral AuctionTournament, in geneticself-regulatorynetworks,
273 l2l
earlypfansfor,68-73 Kauffman on, 300
economics program,244-15,tt9, knowledgeof,298-99
)41 living systemsand, I0Z
economics workshops, 95-98,l13- as nonlinear,65
t1i, t49-rt, t94-97, 247-r5 Prigogine on,17-)4,65
focusof studyat, 5\, 54-55 rise and fall of civilizationsand, 86
foundingworkshops, 79, 84-89 versusselection,100-101, ll3
fundingof, 77-78,89,90, 91, 96, spontaneous,I I
99, 244,247-48, 7t5-r9, t54-55 Woolfram on, 86-87
ClobalSustainability program,349- Self-organizedcriticality, )04-5, )56
t>L of earthquakefault zones,306
Hollandand,225,345(see alsoHol- Langton'sformulationsin relation to,
land,fohnH.) t06-8
impromptuseminar/bull sessions
at, Self-reinforcingmechanisms,Arthur on,
249-50 t77-39
incorporationof, 79 Self-reproducingsystems,217, 2\7. See
mission,72-73,79 a/soComputer viruses
asnewkind of scientificcommunitv. cellular automata and, 221-27
337.t)8 computersand, 2)7
of, 87
organization SelfishCene, The (Dawkins),238
 2an A .,
Index
#nlt )&ol P'Uo's.L99
Shanriofi.Claude. l5l T heory of Self-ReproducingAutomata,
Shaw, Robert, 287 Tfte(Von Neumann),217
SimCity, 267 Thom, Rene, l16
Simmons,Michael, 89, 1'8-79, )42, Time-delayedcontrol theory, 24, 25
)44 Tobin, fames,97
Simon, Herbert,168, l8l Trivelpiece,Alvin, 90
Simulation l,aboratories,267 Turing, Alan, 234, 278, 328-29
Singer,Eugenia,92,%,95, l1), 195,
211 Uccelloclock,40
Skills, as implicit models, 178-79 Ulam, Stanislas. 219
Smith, Adam, 279,328 Uncertainties,and chaos,66
Smith, Stephen,l9l Universalconstructor,218
Society,Farmer on, )19-20 Unpredictability,36- 37
Sociologists, 195 Urey,Harold,l2l, 128, ll2
Spencer,Herbert, 287
Spiegel,Arthur, 78 Vaupel, fames,.10
Spin glassmetaphor,for economy, 138- VHS, lock-in to, 15-16, 40
l19 Volker, Paul, 94, 95
Standards,economic rewardsof, 42 Von Neumann,fohn, 68, 70, 16l,217
StanfordUniversity,Arthur as professor analysiso[ self-reproduction,2 17- 19
at, 46-52 work on cellularautomata,218-20
Stockmarket, 269-70
computer simulation of, 266, 271-74, Wada, Eiitr,240
)29 Waddington,Conrad, I l6
positivefeedbackmechanismand, 196 Watson,fames,29, 62, Zl9
self-fulfilling prophecyin, 274 Weisbuch,Cdrard, l0l
Strangeattractors,226.SeealsoChaos Weisskopf,Victor, 8l
theory What Is Life? (Schriidinger\,62
Strategies,in economics,l4l Whirlwind Proiect,at MIT, l5l-55
Strauss,Lewis, 58 White HouseScienceCouncil. 59-61
Strucfure.SeealsoOrder; Patterns Wicksell'striangle,271
hierarchicalbuilding-block,169-70 Wiener, Norbert, 287
increasein, 295 Wiesenfeld.Kurt. l0.l
Summers,Lany, 117, 11) Wigner, Eugene,68
Superstringtheory, 348 Wilson. Robert. 87
SymbolicsCorporation, 240 Wilson, Stewart,l9l
Symbols,l8l, 184 Winter, Sidney,252
Syntheses in science,75-76,87-88 Wolfram, Stephen,86-87, 225, )17-14
Wolpert,lrwis, l16
Wootters,William, 286-87
Tang,Chao, ]04, )05-6 World Resources Institute,75, )50, )rZ
Thoism World views, l3J
policy-makingand, 130-14 new,219,255
world view of, ll0. t33-74 of scientists,lJ'l-J5
Tate, l,awson, 210 Wriston. Walter, 9l
Taucher,George,2l
Technological change,ll8-20 Zegura, Steve,2l'1-16
Thagard, hul, 192-91 Zurek, Woiciech, 248