T H E ARCHITECTURE O F COMPLEXITY

HERBERT A. SIMON*
Professor of Administration, Carnegie Institute of Technology
(Read April 26, 1962)

A N U M B E R of proposals have been advanced in and to analyze adaptiveness in terms of the theory
recent years for the development of "general sys- of selective i n f ~ r m a t i o n . ~The ideas of feedback
tems theory" which, abstracting from properties and information provide a frame of reference for
peculiar to physical, biological, or social systems, viewing a wide range of situations, just as do the
would be applicable to all of them.l IVe might ideas of evolution, of relativism, of axiomatic
well feel that, while the goal is laudable, systems method, and of operationalism.
of such diverse kinds could hardly be expected to In this paper I should like to report on some
have any nontrivial properties in common. Meta- things we have been learning about particular
phor and analogy can be helpful, or they can be kinds of complex systems encountered in the be-
misleading. All depends on whether the simi- havioral sciences. The develooments I shall dis-
larities the metaphor captures are significant or cuss arose in the context of specific phenomena,
superficial. but the theoretical formulations themselves make
I t may not be entirely vain, however, to search little reference to details of structure. Instead
for common properties anlong diverse kinds of they refer primarily to the complexity of the sys-
complex systems. The ideas that go by the name tems under view without specifying the exact
of cybernetics constitute, if not a theory, at least content of that complexity. Because of their
a point of view that has been proving fruitful over abstractness, the theories may have relevance-
a wide range of application^.^ It has been useful application would be too strong a term-to other
to look at the behavior of adaptive systems in kinds of complex systems that are observed in
terins of the concepts of feedback and homeostasis, the social, biological, and physical sciences.
In recounting these developments, I shall avoid
* T h e ideas in this paper have been the topic of many technical detail, which can generally be found
conversations with my colleague, Allen Xewell. George elsewhere. I shall describe each theory in the
\V. Corner suggested important improvements in biologi-
cal content as well as editorial form. I am also particular context in which it arose. Then, I shall
indebted, for valuable comments on the manuscript, to cite some examples of complex systems, from
Richard H. Meier, John R. Platt, and Warren Weaver. areas of science other than the initial application,
Some of the conjectures about the nearly decomposable to which the theoretical framework appears rele-
structure of the nucleus-atom-molecule hierarchy were
checked against the available quantitative data by Andrew
vant. In doing so, I shall make reference to areas
Schoene and William Wise. My \vork in this area has of knowledge where I an1 not expert-perhaps
been supported by a Ford Foundation grant for research not even literate. I feel quite comfortable in doing
in organizations and a Carnegie Corporation grant for so before the members of this society, representing
research on cognitive processes. T o all of the above, my as it does the whole span of the scientific and
warm thanks, and the usual absolution.
See especially the yearbooks of the Society for Gen- scholarly endeavor. Collectively you 11-ill have
eral Systems Research. Prominent among the exponents little difficulty, I am sure, in distinguishing in-
of general systems theory are L. von Bertalanffy, K. stances based on idle fancy or sheer ignorance
Boulding, R. W. Gerard, and J. G. Miller. For a more from instances that cast some light oil the ways
skeptical view-perhaps too skeptical in the light of the
present discussion-see H. A. Simon and A. Newell, in which complexity exhibits itself wherever it
Models : their uses and limitations, in L. D. White, ed., is found in nature. I shall leave to you the final
T h e state of the social sciences, 66-83, Chicago, Univ. of judgment of relevance in your respective fields.
Chicago Press, 1956. I shall not undertake a formal definition of
2 N. Wiener, Cybernetics, New E'ork, John Wiley &
Sons, 1948. For an imaginative forerunner, see A. J. C. Shannon and W . Weaver, T h e rnathematicnl theory
Lotka, Elelnetits of nzathelnatical O i o l o g ~ ~New
, York, o f commttnicatioti, Urbana, Univ. of Illinois Press, 1949 ;
Dover Publications, 1951, first published in 1924 as Ele- W. R. Ashby, Dcsigtz for a brain, Kew York, John Wiley
wzents of Physical biology. & Sons, 1952.
468 HERBERT A. SIMON [PROC.
AMER.
PHIL.SOC.

"complex systems." Roughly, by a complex HIERARCHIC SYSTEMS

system I mean one made up of a large number By a hierarchic system, or hierarchy, I mean
of parts that interact in a nonsimple way. In such a system that is composed of interrelated sub-
systems, the whole is more than the sum of the systems, each of the latter being, in turn, hier-
parts, not in an ultimate, metaphysical sense, but archic in structure until we reach some lowest
in the important pragmatic sense that, given the level of elementary subsystem. In most systems
properties of the parts and the laws of their inter- in nature, it is somewhat arbitrary as to where
action, it is not a trivial matter to infer the prop- we leave off the partitioning, and what subsystems
erties of the whole. In the face of complexity, we take as elementary. Physics makes much use
an in-principle reductionist may be at the same of the concept of "elementary particle" although
time a pragmatic h o l i ~ t . ~ particles have a disconcerting tendency not to
The four sections that follow discuss four as- remain elementary very long. Only a couple
pects of complexity. The first offers some com- of generations ago, the atoms themselves were
inents on the frequency with which complexity elementary particles; today, to the nuclear physi-
takes the form of hierarchy-the complex system cist they are complex systems. For certain pur-
being composed of subsystems that, in turn, have poses of astronomy, whole stars, or even galaxies,
their olvn subsystems, and so on. The second can be regarded as elementary subsystems. I n one
section theorizes about the relation between the kind of biological research, a cell may be treated
structure of a complex system and the time re- as an elementary subsystem; in another, a protein
molecule; in still another, an amino acid residue.
quired for it to emerge through evolutionary proc- Just why a scientist has a right to treat as ele-
esses : specifically, it argues that hierarchic systems mentary a subsystem that is in fact exceedingly
will evolve far more quickly than non-hierarchic complex is one of the questions we shall take up.
systems of comparable size. The third section F o r the moment, we shall accept the fact that
explores the dynamic properties of hierarchically- scientists do this all the time, and that if they are
organized systems, and shows how they can be careful scientists they usually get away with it.
decon~posed into subsystems in order to analyze Etymologically, the word "hierarchy" has had
their behavior. The fourth section examines the a narrower meaning than I am giving it here.
relation between coillplex systems and their de- The term has generally been used to refer to a
complex system in which each of the subsystems
scriptions.
is subordinated by an authority relation to the
Thus, the central theme that runs through my system it belongs to. More exactly, in a hier-
reinarks is that complexity frequently takes the archic formal organization, each system consists
form of hierarchy, and that hierarchic systems of a "boss" and a set of subordinate subsystems.
have some common properties that are independent Each of the subsystems has a "boss" who is the
of their specific content. Hierarchy, I shall argue, immediate subordinate of the boss of the system.
is one of the central structural schemes that the \Ve shall want to consider systems in which the
architect of complexity uses. relations among subsystems are more complex
than in the formal organizational hierarchy just
4 W . TVeaver, in: Science and complexity, .4merican described. IVe shall want to include systems in
S c i e ~ l t i s t 36: 536, 1948, has distinguished two kinds of which there is no relation of subordination among
complexity, disorganized and organized. W e shall be subsystems. ( I n fact, even in human organiza-
primarily concerned with organized complexity.
5 See also John R. Platt, Properties of large molecules
tions, the formal hierarchy exists only on paper;
that go beyond the properties of their chemical sub-groups, the real flesh-and-blood organization has many
Jolrr.. Tllrorrt. Biol. 1. 342-358, 1961. Since the reduc- inter-part relations other than the lines of formal
tionism-holism issue is a major cazrse de gzlerre between authority.) For lack of a better term, I shall use
scientists and humanists, perhaps we might even hope hierarchy in the broader sense introduced in the
that peace could be negotiated between the two cultures
along the lines of the compromise just suggested. As I previous paragraphs, to refer to all complex sys-
go along, I shall have a little to say about complexity in tems analyzable into successive sets of subsystems,
the arts as well as in the natural sciences. I must empha- and speak of "formal hierarchy" when I want to
size the pragmatism of my holism to distinguish it sharply refer to the more specialized concept6
from the position taken by W. M. Elsasser in T h e physi-
cal fo~ct~datiofz of hiology, Xew York, Pergamon Press, 6 The mathematical term "partitioning" will not do for
1958. what I call here a hierarchy; for the set of subsystems,
VOL.106, KO.6, 19621 T H E ARCHITECTURE O F COMPLEXITY 469

SOCIAL SYSTEXS tend to reserve the word hierarchy for a system

I have already given an example of one kind of that is divided into a small or mdderate numbev
hierarchy that is frequently encountered in the of subsystems, each of which may be further sub-
social sciences : a formal organization. Business divided. Hence, we do not ordinarily think of or
firins, governments, universities all have a clearly refer to a diamond or a gas as a hierarchic struc-
visible parts-within-parts structure. But formal ture. Similarly, a linear polymer is simply a
organizations are not the only, or even the most chain, which may be very long, of identical sub-
cominon, kind of social hierarchy. Almost all parts, the monomers. A t the molecular level it
societies have elementary units called families, is a very flat hierarchy.
which may be grouped into villages or tribes, and I n discussing formal organizations, the number
these into larger groupings, and so on. If we of subordinates who report directly to a single boss
make a chart of social interactions, of who talks is called his span of co?ztvol. I will speak analo-
to whom, the clusters of dense interaction in the gously of the spa% of a system, by which I shall
chart \\-ill identify a rather well-defined hierarchic mean the number of subsystems into which it is
structure. The groupings in this structure may partitioned. Thus, a hierarchic system is flat at
be defined operationally by some measure of fre- a given level if it has a wide span at that level.
quency of interaction in this sociometric matrix. A diamond has a wide span at the crystal level,
but not at the next level down, the molecular level.
BIOLOGICAL Ah-D PHYSICAL SYSTEhIS I n most of our theory construction in the fol-
The hierarchical structure of biological systems lowing sections we shall focus our attention on
is a failliliar fact. Taking the cell as the building hierarchies of moderate span, but from time to
block, we find cells organized into tissues, tissues time I shall comment on the extent to which the
into organs, organs into systems. Moving down- theories might or might not be expected to apply to
ward froill the cell, well-defined subsystems-for very flat hierarchies.
example, nucleus, cell membrane, microsomes, There is one important difference between the
n~itochonclria,and so on-have been identified in physical and biological hierarchies, on the one
animal cells. hand, and social hierarchies, on the other. Most
The hierarchic structure of many physical sys- physical and biological hierarchies are described
tems is equally clear-cut. I have already men- in spatial terms. Jj7e detect the organelles in a
tioned the two main series. A t the microscoDic cell in the way we detect the raisins in a cake-
level we have elementary particles, atoms, mole- they are "visibly" differentiated substructures lo-
cules. macromolecules. A t the macroscopic level calized spatially in the larger structure. O n the
we have satellite systems, planetary systems, gal- other hand, we propose to identify social hier-
axies. Matter is distributed throughout space in archies not by observing who lives close to whom
a strikingly non-uniform fashion. The most nearly but by observing who interacts with whom. These
random distributions we find, gases, are not ran- two points of view can be reconciled by defining
dom distributions of elementary particles but hierarchy in terms of intensity of interaction, but
random distributions of complex systems, i.e. observing that in most biological and physical sys-
molecules. tems relatively intense interaction implies rela-
a\
considerable range of structural types is sub- tive spatial propinquity. One of the interesting
sunled under the term hierarchy as I have defined characteristics of nerve cells and telephone wires is
it. By this definition, a diamond is hierarchic, that they permit very specific strong interactions
for it is a crystal structure of carbon atoms that at great distances. T o the extent that interactions
can l ~ efurther decon~posedinto protons, neutrons, are channeled through specialized communications
antl electrons. However, it is a very "flat" hier- and transportation systems, spatial propinquity
archy, in which the number of first-order sub- becomes less determinative of structure.
syst=ms belonging to the crystal can be indefinitely
large. volume of n~oleculargas is a flat hier- SYMBOLIC SYSTEMS
archy in the same sense. In ordinary usage, we
One very important class of systems has been
and the successive subsets in each of these defines the onlitted from my examples thus far : systems of
partitioning, independently of any systems of relations
among the subsets. By hierarchy I mean the partitioning human synlbolic production. A book is a hier-
in conjunction with the relations that hold among its archy in the sense in which I am using that term.
parts. It is generally divided into chapters, the chapters
470 HERBERT A. SIMON [PKOC. AMEK. P H I L . SOC.

into sections, the sections into paragraphs, the semble l / p parts (the expected number asselllhlecl
paragraphs into sentences, the sentences into before interruption). O n the other hand, Hora
clauses and phrases, the clauses and phrases into has to complete one hundred eleven sub-assemblies
words. Jj7e may take the words as our elementary of ten parts each. The probability that he will not
units, or further subdivide them, as the linguist be interrupted while completing i n y one ot' these
often does, into smaller units. If the book is nar- is (1-p)lO, and each interruption will cost only
rative in character, it may divide into "episodes" ahout the time required to assemble five parts.'
instead of sections, but divisions there will be. Now if p is about .Ol-that is, there is one
The hierarchic structure of music, based on such chance in a hundred that either watchmaker will
units as movements, parts, themes, phrases, is well he interrupted while adding any one part to an as-
known. The hierarchic structure of products of sembly-then a straightforward calculation shows
the pictorial arts is more difficult to characterize, that it will take Tempus, on the average, about
but I shall have something to say about it later. four thousand times as long to assemble a watch
T H E EVOLGTIOX O F COMPLEX SYSTEMS
as Hora.
ll'e arrive at the estimate as follows :
Let me introduce the topic of evolution with a 1. Hora must make 111 times as many complete
parable. There once were two watchmakers, assemblies per watch as Tempus; but.
named Hora and Tempus, who manufactured very
2. Tempus will lose on the average 20 times as
fine watches. Both of them were highly regarded,
much work for each interrupted assembly as
and the phones in their workshops rang frequently
Hora [ l o 0 parts, on the average, as against 5 1 ;
-new customers were constantly calling them.
and,
However, Hora prospered, while Tempus became
Tempus will complete an assembly only 44
poorer and poorer and finally lost his shop. JVhat
times per million attempts (.99'(""' = 44
was the reason?
The watches the men made consisted of about
x while Hora will complete nine out of
ten (.991° = 9 x lo-'). Hence Tempus will
1,000 parts each. Tempus had so constructed his
have to make 20,000 as many attempts per
that if he had one partly assembled and had to put
completed assembly as Hora. ( 9 x lo-' ) '(44
it down-to answer the phone say-it immediately
fell to pieces and had to be reassembled from the
x = 2x lo4. ~ ~ u l t i p l y i n tghese three
elements. The better the customers liked his ratios, we get :
watches, the more they phoned him, the more diffi- 1,'lll x 100/5 x .9910j.991000
cult it became for him to find enough uninterrupted = 1/11 1 x 20 X 20,000 -- 4,000.
time to finish a watch.
7 The speculations on speed of evolution Lvere first sug-
The watches that Hora made were no less com-
gested by H . Jacobson's application of information theory
plex than those of Tempus. But he had designed to estimating the time required for biological evolution.
them so that he could put together subassemblies See his paper, Information, reproduction, and the origin
of about ten elements each. Ten of these subas- of life, in Americatz Scietztist 43: 119-127, January, 1955.
semblies, again, could be put together into a larger From thermodynamic considerations it is possible to esti-
mate the amount of increase in entropy that occurs when
subassembly; and a system of ten of the latter sub- a complex system decomposes into its elements. t See,
assemblies constituted the whole watch. Hence, for example, R. B. Setlow and E. C. Pollard, Jlolcc~tlnr
when Hora had to put down a partly assembled biophysics, 63-65, Reading, l l a s s . , Addison-LVesley Puh-
watch in order to answer the phone, he lost only a lishing Co., 1962, and references cited there.) But entropy
small uart of his work, and he assembled his is the logarithm of a probability, hence information, the
negative of entropy, can be interpreted as the logarithm
watches in only a fraction of the man-hours it of the reciprocal of the probability-the "improbability,"
took Tempus. so to speak. T h e essential idea in Jacobson's model is
I t is rather easy to make a quantitative analysis that the expected time required for the system to reach
of the relative difficulty of the tasks of Tempus a particular state is inversely proportional to the proha-
bility of the state-hence increases exponentially with the
and H o r a : Suppose the probability that an inter- amount of information (negentropy) of the state.
ruption will occur while a part is being added to Following this line of argument, but not introducing
an incomplete assembly is p. Then the probabil- the notion of levels and stable subassemblies, Jacohson
ity that Tempus can complete a watch he has arrived at estimates of the time required for evolution so
large as to make the event rather improbable. Our analy-
started without interruption is (l-pjlOoO-a very sis, carried through in the same way, but with attention
small number unless p is .001 or less. Each in- to the stable intermediate forms, produces very much
terruption will cost, on the average, the time to as- smaller estimates.
VOL. 106, SO. 6 , 19621 T H E ARCHITECTURE O F COMPLEXITY 47 1

BIOLOGICAL EVOLUTION scheme by the stability of the complex forms, once

What lessons can we draw from our parable these come into existence. But this is nothing
for biological evolution? Let us interpret a par- more than survival of the fittest-i.e., of the stable.
tially completed subassembly of k elementary parts Second, not all large systems appear hierarchi-
as the coexistence of k parts in a small volume- cal. F o r example, most polymers-e.g., nylon-
ignoring their relative orientations. T h e model are simply linear chains of large numbers of
assumes that parts are entering the volume at a identical components, the monomers. However,
constant rate,-but that there is a constant prob- for present purposes we can simply regard such a
ability, p, that the part will be dispersed before structure a s a hierarchv with a sDan of one-the
another is added, unless the assembly reaches a limiting case. F o r a chain of any length repre-
stable state. These assumptions are not particu- sents a state of relative e q u i l i b r i ~ m . ~
larly realistic. They undoubtedly underestimate Third, the evolution of complex systems from
the decrease in probability of achieving the assem- simple elements implies nothing, one- way or the
bly with increase in the size of the assembly. other, about the change in entropy of the entire
Hence the assumptions understate-probably by a system. If the process absorbs free energy, the
large factor-the relative advantage of a hier- complex system will have a smaller entropy than
archic structure. the elements; if it releases free energy, the oppo-
Although we cannot, therefore, take the nu- site will be true. T h e former alternative is the
merical estimate seriously the lesson for biological one that holds for most biological systems, and
evolution is quite clear and direct. The time re- the net inflow of free energy has to be supplied
quired for the evolution of a complex form from from the sun or some other source if the second
simple elements depends critically b n the numbers law of thermodynamics is not to be violated. F o r
and distribution of potential intermediate stable the evolutionary process we are describing, the
forms. In particular, if there exists a hierarchy equilibria of the intermediate states need have only
of potential stable "subassenlblies," with about the local and not global stability, and they may be
same span, s, at each level of the hierarchy, then stable only in the steady state-that is, as long as
the time required for a subassemblv can be ex- there is an external source of free energy that may
pected to be about the same at each level-that is be drawn u ~ o n . @
proportional to l / ( l - p ) s . T h e time required for Because organisms are not energetically closed
the assembly of a system of n elements will be systems, there is no way to deduce the direction,
proportional to logs n, that is, to the number of much less the rate, of evolution from classical
levels in the system. One would say-with more thermodynamic considerations. All estimates in-
illustrative than literal intent-tllat the time re- dicate that the amount of entropy, measured in
quired for the evolution of multi-celled organisms physical units, involved in the formation of a one-
from single-celled organisms might be of the same celled biological organism is trivially small-about
order of magnitude as the time required for the - 10-l1 cal/degree.1° T h e "improbability" of evo-
evolution of single-celled organisms from macro- lution has nothing to do with this quantity of
molecules. The same argument could be applied entropy, which is produced by every bacterial cell
to the evolution of proteins from amino acids, of every generation. The irrelevance of quantity of
molecules from atoms, of atoms from elementary 8 There is a well-developed theory of polymer size,
particles. based on models of random assembly. See for example
A whole host of objections to this oversimplified P . J. Flory, Principles of polymer chemistuy, ch. 8,
Ithaca, Cornell Univ. Press, 1953. Since all subassem-
scheme will occur, I am sure, to every working blies in the polymerization theory are stable, limitation
biologist, chemist, and physicist. Before turning of molecular growth depends on "poisoning" of terminal
to matters I know more about, I shall mention groups by impurities or formation of cycles rather than
three of these problems, leaving the rest to the upon disruption of partially-formed chains.
@ T h i spoint has been made many times before, but it
attention of the specialists. cannot be emphasized too strongly. F o r further discus-
First, in spite of the overtones of the watch- sion, see Setlow and Pollard, op. cit., 49-64; E. Schro-
maker parable, the theory assumes no teleological dinger, W h a t i s life? Cambridge Univ. Press, 1945; and
n~echanisnl. The complex forms can arise from H. Linschitz, The information content of a bacterial cell,
in H. Questler, ed., Infoutnation theory i n biology, 251-
the simple ones by purely random processes. ( I 262, Urbana, Univ. of Illinois Press, 1953.
shall propose another model in a moment that 1 0 See Linschitz, op. czt. This quantity, 10-11 cal/de-
shows this clearly.) Direction is provided to the gree, corresponds to obout 1013 bits of information.
472 HERBERT A. SIMON [PROC.
AMER. PHIL. SOC.

information, in this sense, to speed of evolution A little reflection reveals that cues signaling
can also be seen from the fact that exactly as much progress play the same role in the problem-solving
information is required to "copy" a cell through process that stable intermediate forms play in the
the reproductive process as to produce the first cell biological evolutionary process. I n fact, we can
through evolution. take over the watchmaker parable and apply it
The effect of the existence of stable intermediate also to problem solving. In problem solving, a
forms exercises a powerful effect on the evolution partial result that represents recognizable progress
of complex forms that may be likened to the dra- toward the goal plays the role of a stable sub-
matic effect of catalysts upon reaction rates and assembly.
steady state distribution of reaction products in Suppose that the task is to open a safe whose
open systems.ll I n neither case does the entropy lock has ten dials, each with one hundred possible
change provide us with a guide to system behavior. settings, numbered from 0 to 99. How long will
it take to open the safe by a blind trial-and-error
PROBLEM SOLVIXG AS NATURAL SELECTION
search for the correct setting? Since there are
Let us turn now to some phenomena that have 10010 possible settings, we may expect to esamine
no obvious connectioil with biological evolution: about one-half of these, on the average, before
human problem-solving processes. Consider, for finding the correct one-that is, fifty billion billion
example, the task of discovering the proof for a settings. Suppose, however, that the safe is de-
difficult theorem. T h e process can be-and often fective, so that a click can be heard when any one
has been-described as -a search through a maze. dial is turned to the correct setting. Now each
Starting with the axioms and previously proved dial can be adjusted independently, and does not
theorems, various transformations allowed by the need to be touched again while the others are being
rules of the mathematical systems are attempted, set. The total number of settings that has to he
to obtain new expressions. These are modified in tried is only 10 X 50, or five hundred. The task
turn until, with persistence and good fortune, a of opening the safe has been altered, by the cues
sequence or path of transforn~ationsis discovered the clicks provide, from a practically impossible
that leads to the goal. one to a trivial one.13
The process usually involves a great deal of trial A considerable amount has been learned in the
and error. Various paths are tried; some are past five years about the nature of the mazes that
abandoned, others are pushed further. Before a represent common human problem-solving tasks-
solution is found, a great many paths of the maze proving theorems, solving puzzles, playing chess,
may be explored. The more difficult and novel making investments, balancing assembly lines, to
the problem, the greater is likely to be the amount mention a few. All that we have learned about
of trial and error required to find a solution. At these mazes points to the same conclusioi~: that
the same time, the trial and error is not com- human problem solving, from the most blundering
pletely random or blind ; it is, in fact, rather highly to the most insightful, involves nothing more than
selective. The new expressions that are obtained varying mixtures of trial and error and selectivity.
by transforming given ones are examined to see The selectivity derives from various rules of
whether they- represent
- progress toward the goal. amplifier, 215-233 in C. E. Shannon and J. LlcCarthy,
Indications of progress spur further search in the Automata studies, Princeton, Princeton Univ. Press,
same direction; lack of progress signals the 1956.
abandonment of a line of search. Problem solving 13The clicking safe example was supplied by D. P .
requires selective trial and error.'? Simon. Ashby, op. cit., 230, has called the selectivity
involved in situations of this kind "selection by compo-
11 See H. Kacser, Some physico-chemical aspects of nents." The even greater reduction in time produced by
biological organization, Appendix, pp. 191-249 in C. H. hierarchization in the clicking safe example, as compared
Waddington, T h e strategy of the getzes, London, George with the watchmaker's metaphor, is due to the fact that
Allen & Unwin, 1957. a random search for the correct combination is involved
See A. Newell, J. C. Shaw, and H. A. Simon, in the former case, while in the latter the parts come to-
Empirical explorations of the logic theory machine, Pro- gether in the right order. It is not clear which of these
ceedings o f the 1957 W e s t e r n Joint Cotnputer Colzferetzce, metaphors provides the better model for biological evo-
February, 1957, New York: Institute of Radio Engi- lution, but we may be sure that the watchmaker's meta-
neers; Chess-playing programs and the problem of com- phor gives an exceedingly conservative estimate ot' the
plexity, IBM Journal o f Research and Developmetzt 2 : savings due to hierarchization. The safe may give ar.
320-335, October, 1958; and for a similar view of prob- excessively high estimate because it assumes all possible
lem solving, W. R. Ashby, Design for an intelligence arrangements of the elements to be equally probable.
VOI,. 106, NO. 6 , 19621 T H E A R C H I T E C T U R E O F COMP1,EXITY 473

thumb, or heuristics, that suggest which paths Macedonian empire and gave it to his son, to be
should be tried first and which leads are promising. later combined with the Persian subassembly and
W e do not need to postulate processes more others into Alexander's greater system. O n Alex-
sophisticated than those involved in organic evo- ander's death, his empire did not crumble to dust,
lution to explain how enormous problem mazes but fragmented into some of the major subsystems
are cut down to quite reasonable size.14 that had composed it.
The watchmaker argument implies that if one
T H E SOURCES O F SELECTIVITY would be Alexander, one should be born into a
When we examine the sources from which the world where large stable political systems already
problem-solving system, or the evolving system, exist. Where this condition was not fulfilled, as
as the case may be, derives its selectivity, we dis- on the Scythian and Indian frontiers, Alexander
cover that selectivity can always be equated with found empire building a slippery business. So
some kind of feedback of information from the too, T . E. Lawrence's organizing of the Arabian
environment. revolt against the Turks was limited by the charac-
Let us consider the case of problem solving first. ter of his largest stable building blocks, the sepa-
There are two basic kinds of selectivity. One we rate, suspicious desert tribes.
have already noted: various paths are tried out, The profession of history places a greater value
the consequences of following them are noted, and upon the validated particular fact than upon ten-
this information is used to guide further search. dentious generalization. I shall not elaborate upon
In the same way, in organic evolution, various my fancy, therefore, but will leave it to historians
complexes come into being, at least evanescently, to decide whether anything can be learned for the
and those that are stable provide new building interpretation of history from an abstract theory
blocks for further construction. It is this infor- of hierarchic complex systems.
mation about stable configurations, and not free
energy or negentropy from the sun, that guides CONCLUSION :THE EVOLUTIONARY E X P L A K A T I O X
OF HIERARCHY
the process of evolution and provides the selectivity
that is essential to account for its rapidity. W e have shown thus far that complex systems
The second source of selectivity in problem will evolve from simple systems much more rapidly
solving is previous experience. W e see this par- if there are stable intermediate forms than if there
ticularly clearly when the problem to be solved is are not. The resulting complex forms in the for-
similar to one that has been solved before. Then, mer case will be hierarchic. W e have only to turn
by sin~plytrying again the paths that led to the the argument around to explain the observed pre-
earlier solution, or their analogues, trial-and-error dominance of hierarchies among the complex sys-
search is greatly reduced or altogether eliminated. tems nature presents to us. Among possible
What corresponds to this latter kind of informa- complex forms, hierarchies are the ones that have
tion in organic evolution? The closest analogue the time to evolve. The hypothesis that complex-
is reproduction. Once we reach the level of self- ity will be hierarchic makes no distinction among
reproducing systems, a complex system, when it very flat hierarchies, like crystals, and tissues, and
has once been achieved, can be multiplied indefi- polymers, and the intermediate forms. Indeed, in
nitely. Reproduction in fact allows the inheritance the complex systems we encounter in nature, ex-
of acquired characteristics, but at the level of amples of both forms are prominent. A more
genetic material, of course; i.e., only characteristics complete theory than the one we have developed
acquired by the genes can be inherited. W e shall here would presumably have something to say
return to the topic of reproduction in the final about the determinants of width of span in these
section of this paper. systems.
ON E M P I R E S A N D E M P I R E - B U I L D I N G NEARLY DECOMPOSABLE SYSTEMS
W e have not exhausted the categories of com- In hierarchic systems, we can distinguish be-
plex systems to which the watchmaker argument tween the interactions among subsystems, on the
can reasonably be applied. Philip assembled his one hand, and the interactions within subsystems
l4A. Newell and H. A. Simon, Computer simulation -i.e., among the parts of those s u b s y s t e m s ~ n
of human thinking, Science 134: 2011-2017, December 22, the other. The interactions at the different levels
1961. may be, and often will be, of different orders of
 HERBERT A. SIMON

( a ) in a nearly decomposable system, the short-

run behavior of each of the component subsystems
is approximately independent of the short-run be-
havior of the other components; ( b ) in the long
run, the behavior of any one of the components
depends in only an aggregate way on the behavior
of the other components.
Let me provide a very concrete simple example
of a nearly decomposable system.15 Consider a
building whose outside walls provide perfect
thermal insulation from the environment. W e
shall take these walls as the boundary of our sys-
tem. The building is divided into a large number
of rooms, the walls between them being good, but
FIG. 1. A hypothetical nearly-decomposable system. In not perfect, insulators. The walls between rooms
terms of the heat-exchange example of the text, A l , are the boundaries of our major subsystems. Each
X2, and A3 may be interpreted as cubicles in one room is divided by partitions into a number of
room, B1 and B2 as cubicles in a second room, and C1,
C2, and C3 as cubicles in a third. The matrix en- cubicles, but the partitions are poor insulators.
tries then are the heat diffusion coefficients between A thermometer hangs in each cubicle. Suppose
cubicles. that at the time of our first observation of the sys-
tem there is a wide variation in temperature from
cubicle to cubicle and from room t o room-the
various cubicles within the building are in a state
of thermal disequilibrium. When we take new
temperature readings several hours later, what
magnitude. In a formal organization there will shall we find? There will be very little variation
generally be more interaction, on the average, be- in temperature among the cubicles within each
tween two employees who are members of the single room, but there may still be large tempera-
same department than between two employees ture variations awzottg rooms. When we take
from different departments. I n organic sub- readings again several days later, we find an al-
stances, intermolecular forces will generally be iliost uniform temperature throughout the build-
weaker than molecular forces, and molecular forces ing; the temperature differences among rooms
than nuclear forces. have virtually disappeared.
In a rare gas, the intermolecular forces will be W e can describe the process of equilibration
negligible compared to those binding the molecules forillally by setting up the usual equations of heat
-we can treat the individual particles, for many flow. The equations can be represented by the
purposes, as if they were independent of each matrix of their coefficients, rij, where rij is the rate
other. \Ye can describe such a system as decoi~z- at which heat flows from the ith cubicle to the jth
posable into the subsystems comprised of the indi- cubicle per degree difference in their temperatures.
vidual particles. As the gas becomes denser, mo- If cubicles i and j do not have a common wall,
lecular interactions become more significant. But rig will be zero. If cubicles i and i have a common
over some range, we can treat the decomposable wall, and are in the same room, ri, will be large.
case as a limit, and as a first approximation. W e If cubicles i and j are separated by the wall of a
can use a theory of perfect gases. for example, to l5 This discussion of near-decomposability is based
describe approxin~ately the behavior of actual upon H. -4. Simon and .A. .Ando, .Aggregation of variables
gases if they are not too dense. As a second ap- in dynamic systems, Econovzetrica 29: 111-138, April,
1961. The example is drawn from the same source,
proximation, we may move to a theory of nearly 117-118. The theory has been further developed and
decomposable systems, in which the interactions applied to a variety of economic and political phenomena
among the subsystems are weak, but not negligible. by Ando and F. M. Fisher. See F . M. Fisher, On the
At least some kinds of hierarchic systems can be cost of approximate specification in simultaneous equation
Economctrica 29: 139-170, .April, 1961, and
approximated
-- successfully as nearly decomposable estimation,
F. hi. Fisher and A. Ando, T W O theorems on Ceteris
systems. The main theoretical findings from the Paribus in the analysis of dynamic systems, American
approach can be summed up in two propositions: Political Sciettce Revie.rc1 61 : 103-113, March, 1962.
VOL.l o b , S O . 6, 19621 T H E ARCHITECTURE OF COMPLEXITY 475

room, v i j will be nonzero but small. Hence, by economic activity. The large linkage coefficients
grouping all the cubicles together that are in the are associated, in general, with the main flows of
same room, we can arrange the matrix of coeffi- raw materials and semi-finished products within
cients so that all its l a r"~ kelements lie inside a and between industries. An input-output matrix
string of square submatrices along the main di- of the economy, giving the magnitudes of these
agonal. All the elements outside these diagonal flows, reveals the nearly decomposable structure
squares will be either zero or small (see figure 1). of the system-with one qualification. There is a
W e may take some small number, E , as the upper consumption subsystem of the economy that is
bound of the extradiagonal elements. W e shall linked strongly to variables in most of the other
call a matrix having these properties a nearly subsystems. Hence, we have to modify our no-
decot~zposablewzatrix. tions of decomposability slightly to accommodate
Now it has been proved that a dynamic system the special role of the consumption subsystem in
that can be described by a nearly decomposable our analysis of the dynamic behavior of the
matrix has the properties, stated above, of nearly economy.
decomposable system. In our simple example of In the dynamics of social systems, where mem-
heat flow this means that in the short run each bers of a system communicate with and influence
room will reach an equilibrium temperature (an other members, near decomposability is generally
average of the initial temperatures of its offices) very prominent. This is most obvious in formal
nearly independently of the others; and that each organizations, where the formal authority rela-
room will remain approximately in a state of equi- tion connects each member of the organization
librium over the longer period during which an with one immediate superior and with a small
over-all temperature equilibrium is being estab- number of subordinates. Of course many com-
lished throughout the building. After the intra- munications in organizations follow other channels
room short-run ecluilibria have been reached, a than the lines of formal authoritv. But most of
single thermometer in each room will be adequate these channels lead from any particular individual
to describe the dynamic behavior of the entire to a very limited number of his superiors, sub-
system-separate thermometers in each cubicle will ordinates, and associates. Hence, departmental
be superfluous. boundaries play very much the same role as the
walls in our heat example.
X E A R DECOMPOSABILITY O F SOCIAL SYSTEMS

As a glance at figure 1 shows, near decomposa- P H Y S I C O - C H E M I C A L SYSTEMS

bility is a rather strong property for a matrix to In the complex systems familiar in biological
possess, and the matrices that have this property chemistry, a similar structure is clearly visible.
will describe very special dynamic systems-van- Take the atomic nuclei in such a system as the
ishingly few systems out of all those that are elementary parts of the system, and construct a
thinkable. How few they will be depends, of matrix of bond strengths between elements. There
course, on how good an approximation we insist will be matrix elements of quite different orders
up011. If we demand that epsilon be very small, of magnitude. The largest will generally corre-
correspondingly few dynamic systems will fit the spond to the covalent bonds, the next to the ionic
definition. But we have already seen that in the bonds, the third group to hydrogen bonds, still
natural world nearly decomposable systems are smaller linkages to van der Waals forces.16 If we
far from rare. O n the contrary, systems in which select an epsilon just a little smaller than the mag-
each variable is linked with almost equal strength nitude of a covalent bond, the system will de-
with almost all other parts of the system are far compose into subsystems-the constituent mole-
rarer and less typical. cules. The smaller linkages will correspond to the
In economic dynamics, the main variables are intermolecular bonds.
the prices and quantities of commodities. It is It is well known that high-energy, high-fre-
empirically true that the price of any given com- l 6 For a survey of the several classes of molecular and
modity and the rate at which it is exchanged de- inter-molecular forces, and their dissociation energies
pend to a significant extent only on the prices and see Setlow and Pollard, op. cit., chapter 6. The energies
quantities of a few other commodities, together of typical covalent bonds are of the order of 80-100 k
with a few other aggregate magnitudes, like the cal/mole, of the hydrogen bonds, 10 k cal/mole. Ionic
bonds generally lie between these two levels, the bonds
average price level or some over-all measure of due to van der Waals forces are lower in energy.
476 H E R B E R T A. SIMON [PROC.
AMER. PHIL.
SOC.

quency vibrations are associated with the smaller as the number associated with the individual atoms,
physical subsystems, low-frequency vibrations the bonding process can go on indefinitely-the
with the larger systems into which the subsystems atoms can form crystals or polymers of indefinite
.
are assembled. For e x a m ~ l.e ,the radiation fre- extent. If the number of bonds of which the
quencies associated with molecular vibrations are composite is capable is less than the number as-
much lower than those associated with the vibra- sociated with each of the parts, then the process
tions of the planetary electrons of the atoms; the of agglomeration must come to a halt.
latter, in turn, are lower than those associated with W e need only mention some elementary es-
nuclear processes.17 Molecular systems are nearly amples. Ordinary gases show no tendency to ag-
deconiposable systems, the short-run dynamics glomerate because the multiple bonding of atoms
relating to the internal structures of the subsys- "uses up" their capacity to interact. While each
tems ; the long-run dynamics to the interactions of oxygen atom has a valence of two, the O1 niole-
these subsystems. cules have a zero valence. Contrariwise, indefinite
A number of the important approximations em- chains of single-bonded carbon atoms can be built
ployed in physics depend for their validity on the up because a chain of any number of such atoms,
near-decomposability of the systems studied. The each with two side groups, has a valence of e s -
theory of the thermodynamics of irreversible proc- actly two.
esses, for example, requires the assumption of Now what happens if we have a systetii of ele-
macroscopic disequilibriun~ but microscopic equi- ments that possess both strong and weak inter-
librium.ls exactli the situation described in our action capacities, and whose strong bonds are ex-
heat-exchange example. Similarly computations haustible through combination? Subsystems will
in quantum mechanics are often handled by treat- form, until all the capacity for strong interaction
ing weak interactions as producing perturbations is utilized in their construction. Then these sub-
on a system of strong interactions. systems will be linked by the weaker second-order
bonds into larger systems. For example. a water
SOAIE OBSER\'ATIOh-S O N H I E R A R C H I C SPAN
molecule has essentially a valence of zero-all the
T o understand why the span of hierarchies is potential covalent bonds are fully occupietl by the
sometimes very broad-as in crystals-sometimes interaction of hydrogen and oxygen molecules.
narrow. we need to examine more detail of the in- But the geometry of the molecule creates an elec-
-
teractions. In general, the critical consideration
is the extent to which interaction between two (or
tric dipole that permits weak interaction between
the water and salts dissolved in it-whence such
a few) subsystems excludes interaction of these phenomena as its electrolytic conductivity.'"
subsystems with the others. Let us examine first Similarly, it has been observed that, although
some physical examples. electrical forces are much stronger than gravita-
Consider a gas of identical molecules, each of tional forces, the latter are far more important
which can form covalent bonds, in certain ways, than the former for systems on an astronomical
with others. Let us suppose that we can associate scale. The explanation, of course, is that the elec-
with each atom a specific number of bonds that trical forces, being bipolar, are all "used up" in the
it is capable of maintaining simultaneously. (This linkages of the smaller subsystems, and that sig-
number is obviously related to the number we usu- nificant net balances of positive or negative charges
ally call its valence.) S o w suppose that two atoms are not generally found in regions of macroscopic
join, and that we can also associate with the com- size.
bination a specific number of external bonds it is In social as in physical systems there are gen-
capable of maintaining. If this number is the same erally limits on the simultaneous interaction of
17Typical wave numbers for vibrations associated with large numbers of subsystems. In the social case,
various systems (the wave number is the reciprocal of these limits are related to the fact that a human
wave length hence proportional to frequency) : being is more nearly a serial than a parallel in-
steel wire under tensi~n-lO-~~to lo-' cm-I formation-processing system. H e can carry on
molecular rotations-lo0 to l o 2 cm-l
molecular vibrations-10' to lo3 cm-' only one conversation at a time, and although this
planetary electrons-lo4 to 10' cm-I does not limit the size of the audience to which a
nuclear rotations-10' to 1010 cm-I mass communication can be addressed, it does
nuclear surface vibrations-1011 to 1012 cm-l.
1s S. R. de Groot, Thermodynanzics of irreversible proc- 1 9 See, for example, L. Pauling, General ch~inistry,
esses, 11-12, Xew York, Interscience Publishers, 1951. ch. 15.
VOL.106, NO. 6 , 19621 T H E ARCHITECTURE OF COMPLEXITY

limit the number of people simultaneously involved in the outline and is likely to be lost. The loss of
in most other forms of social interaction, Apart such information and the preservation mainly of
from requirements of direct interaction, most roles information about hierarchic order is a salient
impose tasks and responsibilities that are time con- characteristic that distinguishes the drawings of a
suming. One cannot, for example, enact the role child or someone untrained in representation from
of "friend" with large numbers of other people. the drawing of a trained artist. ( I am speaking of
It is probably true that in social as in physical an artist who is striving for representation. )
systems, the higher frequency dynamics are associ-
ated with the subsystems, the lower frequency dy- NEAR DECOMPOSABILITY A N D COXPREHENSIBILITY
namics with the larger systems. It is generally From our discussion of the dynamic properties
believed, for example, that the relevant planning of nearly decomposable systems, we have seen that
horizon of executives is longer the higher their comparatively little information is lost by repre-
location in the organizational hierarchy. It is senting them as hierarchies. Subparts belonging
probably also true that both the average duration to different parts only interact in an aggregative
of an interaction between executives and the aver- fashion-the detail of their interaction can be ig-
age interval between interactions is greater at nored. In studying the interaction of two large
higher than at lower levels. molecules, generally we do not need to consider
in detail the interactions of nuclei of the atoms
SUMMARY : NEAR DECOXPOSABILITY
belonging to the one molecule with the nuclei of
W e have seen that hierarchies have the property the atoms belonging to the other. I n studying the
of near-decomposability. Intra-component linlz- interaction of two nations, we do not need to study
ages are generally stronger than intercomponent in detail the interactions of each citizen of the
linkages. This fact has the effect of separating first with each citizen of the second.
the high-frequency dynamics of a hierarchy-in- The fact, then, that many complex systems
volving the internal structure of the components- have a nearly decomposable, hierarchic structure
from the low frequency dynamics-involving inter- is a major facilitating factor enabling us to under-
action among components. W e shall turn next stand, to describe, and even to "see" such systems
to some important consequences of this separation and their parts. O r perhaps the proposition
for the description and comprehension of complex should be put the other way round. If there are
systems. important systems in the world that are complex
T H E DESCRIPTIOS O F COMPLEXITY without being hierarchic, they may to a consider-
able extent escape our observation and our under-
If you ask a person to draw a complex object-
standing. Analysis of their behavior would in-
e.g., a human face-he will almost always proceed
volve such detailed knowledge and calculation of
in a hierarchic fashion.*O First he will outline
the interactions of their elementary parts that it
the face. Then he will add or insert features:
would be beyond our capacities of memory or
eyes, nose, mouth, ears, hair. If asked to elabo-
computati~n.~~
rate, he will begin to develop details for each of
the features-pupils, eyelids, lashes for the eyes, 2 1 I believe the fallacy in the central thesis of W. hf.

and so on-until he reaches the limits of his ana- Elsasser's T h e physical foundation of biology, mentioned
earlier, lies in his ignoring the simplification in description
tomical knowledge. His information about the of complex systems that derives from their hierarchic
object is arranged hierarchicly in memory, like structure. Thus (p. 155) : "If we now apply similar
a topical outline. arguments to the coupling of enzymatic reactions with
When information is put in outline form, it is the substratum of protein molecules, we see that over a
sufficient period of time, the information corresponding to
easy to include information about the relations the structural details of these molecules will be commu-
among the major parts and information about the nicated to the dynamics of the cell, to higher levels of
internal relations of parts in each of the subout- organization as it were, and may influence such dynamics.
lines. Detailed information about the relations of While this reasoning is only qualitative, it lends credence
subparts belonging to different parts has no place to the assumption that in the living organism, unlike the
inorganic crystal, the effects of microscopic structure can-
George ,4. Miller has collected protocols from sub- not be simply averaged o u t ; as time goes on this influ-
jects who were given the task of drawing faces, and finds ence will pervade the behavior of the cell 'at all levels.' "
that they behave in the manner described here (private But from our discussion of near-decomposability it
communication). See also E. H. Gombrich, Art and would appear that those aspects of microstructure that
illzrsion, 291-296, New York, Pantheon Books, 1960. control the slow developmental aspects of organismic
 478 H E R B E R T A. S I M O K [PROC. AMER.
PHIL.SOC.

I shall not try to settle which is chicken and AB

which is egg: whether we are able to understand the pattern CD for example, occurs four times
the world because it is hierarchic, or whether it in the total pattern, it is economical to represent
appears hierarchic because those aspects of it it b y the single symbol, a.
which are not elude our understanding and ob- If a complex structure is completely unre-
servation. I have already given some reasons for dundant-if no aspect of its structure can be in-
supposing that the former is at least half the ferred from any other-then it is its own simplest
truth-that evolving complexity would tend to be description. W e can exhibit it, but we cannot
hierarchic-but it may not be the whole truth. describe it by a simpler structure. The hierarchic
structures we have been discussing have a high
S I M P L E DESCRIPTIONS O F COMPLEX SYSTEMS
degree of redundancy, hence can often be described
One might suppose that the description of a in economical terms. The redundancy takes a
complex system would itself be a complex struc- number of forms, of which I shall mention three:
ture of symbols-and indeed, it may be just that. 1. Hierarchic systems are usually composed of
But there is no conservation law that requires only a few different kinds of subsystems, in vari-
that the description be as cumbersome as the ob- ous combinations and arrangements. A familiar
ject described. A trivial example will show how example is the proteins, their multitudinous vari-
a system can be described economically. Suppose ety arising from arrangements of only twenty
the systenl is a two-dimensional array like this : different amino acids. Similarly, the ninety-odd
elements provide all the kinds of building blocks
A B M.1-RS H I needed for an infinite variety of molecules. Hence,
C D O P T C ' J K we can construct our description from a restricted
M S A B H I R S alphabet of elementary terms corresponding to the
O P C D J K T C ' basic set of elementary subsystems from which
R S H I A B M S the complex system is generated.
T C J K C D O P
H I R S M.lTAB
2. Hierarchic systems are, as we have seen,
J K T I W O P C D often nearly decomposable. Hence only aggre-
aative properties of their parts enter into the de-
Let us call the array IAB a , the array
CD 1
op 1 m,
/ *WN
scriptiin o f the interactions of those parts. A
generalization of the notion of near-decomposabil-
ity might be called the "empty world hypothesis"
the arrajr r, and the array h. Let us
-most things are only weakly connected with
am most other things; for a tolerable description of
w ,and the array Then
8 8
reality only a tiny fraction of all possible interac-
the entire array is simply II wxx ' w I . While the tions needs to be taken into account. By adopting
a descriptive language that allows the absence of
original structure consisted of 64 symbols, it re- something to go unmentioned, a nearly empty
quires only 35 t o write down its description: world can be described quite concisely. Mother
Hubbard did not have to check off the list of pos-
sible contents to say that her cupboard was bare.
3. By appropriate "recoding," the redundancy
that is present but unobvious in the structure of
a complex system can often be made patent. The
most cornillon recoding of descriptions of dy-
namic systems consists in replacing a description
Liveachieve the abbreviation b y making use of of the time path with a description of a differential
the redundancy in the original structure. Since law that generates that path. The simplicity, that
is, resides in a constant relation between the
dynamics can be separated out from the aspects that con- state of the system at any given time and the state
trol the more rapid cellular metabolic processes. For of the system a short time later. Thus, the struc-
this reason we should not despair of unravelling the web
of causes. See also J. R. Platt's review of Elsasser's
ture of the sequence, 1 3 5 7 9 11 . . ., is most
book in Pcvspcctives in biology utzd wzedicitze 2 : 243-245, simply expressed by observing that each member
1959. is obtained by adding 2 to the previous one. But
 VOL. 106, NO. 6, 19621 T H E ARCHITECTURE O F COMPLEXITY 479

this is the sequence that Galileo found to describe lation between the state and process descriptions
the velocity at the end of successive time intervals of the same complex reality. Plato, in the Meno,
of a ball rolling down an inclined plane. argued that all learning is remembering. H e
It is a familiar proposition that the task of sci- could not otherwise explain how we can discover
ence is to make use of the world's redundancy to or recognize the answer to a problem unless we
describe that world simply. I shall not pursue the already know the answer.23 Our dual relation to
general methodological point here, but shall instead the world is the source and solution of the para-
take a closer look at two main types of description dox. W e pose a problem by giving the state
that seem to be available to us in seeking an description of the solution. The task is to dis-
understanding of complex systen~s. I shall call cover a sequence of processes that will produce
these state description and process description, the goal state from an initial state. Translation
respectively. from the process description to the state descrip-
tion enables us to recognize when we have suc-
STATE DESCRIPTIONS A N D PROCESS DESCRIPTIONS ceeded. The solution is genuinely new to us-
and we do not need Plato's theory of remembering
"A circle is the locus of all points equidistant to explain how we recognize it.
from a given point." "To construct a circle, rotate There is now a growing body of evidence that
a compass with one arm fixed until the other arm the activity called human problem solving is basi-
has returned to its starting point." It is implicit cally a form of means-end analysis that aims at
in Euclid that if you carry out the process speci- discovering a process description of the path that
fied in the second sentence, you will produce an leads to a desired goal. The general paradigm is :
object that satisfies the definition of the first. given a blueprint, to find the corresponding
The first sentence is a state description of a circle, recipe. Much of the activity of science is an
the second a process description. application of that paradigm : given the descrip-
These two modes of apprehending structure are tion of some natural phenomena, to find the differ-
the warp and weft of our experience. Pictures, ential equations for processes that will produce
blueprints, most diagrams, chemical structural the phenomena.
formulae are state descriptions. Recipes, differ-
ential equations, equations for chemical reactions T H E DESCRIPTION O F COMPLEXITY I N
are process descriptions. The former characterize SELF-REPRODUCIxG SYSTEMS
the world as sensed; they provide the criteria for
identifying objects, often by modeling the objects The problem of finding relatively simple descrip-
themselves. The latter characterize the world as tions for complex systems is of interest not only
acted upon; they provide the means for producing for an understanding of human knowledge of the
or generating objects having the desired charac- world but also for an explanation of how a com-
teristics. plex system can reproduce itself. In my discus-
The distinction between the world as sensed and sion of the evolution of complex systems, I touched
the world as acted upon defines the basic condition only briefly on the role of self-reproduction.
for the survival of adaptive organisms. The or- Atoms of high atomic weight and complex in-
ganism must develop correlations between goals organic molecules are witnesses to the fact that the
in the sensed world and actions in the world of evolution of complexity does not imply self-
process. \{'hen they are made conscious and ver- reproduction. If evolution of complesity from
balized, these correlations correspond to what we simplicity is sufficiently probable, it will occur
usually call means-end analysis. Given a desired repeatedly; the statistical equilibrium of the system
state of affairs and an existing state of affairs, the will find a large fraction of the elementary par-
task of an adaptive organism is to find the differ- ticles participating in complex systems.
ence between these two states, and then to find If, however, the existence of a particular com-
the correlating process that will erase the differ- plex form increased the probability of the creation
en~e.?~ of another form just like it, the equilibrium be-
Thus, problem solving requires continual trans- tween complexes and components could be greatly
altered in favor of the former. If we have a de-
" See H. A. Simon and A. Newell, Simulation of scription of an object that is sufficiently clear and
human thinking, i n M. Greenberger (ed.), Management
and the computer of the future, 95-114, esp. pp 110 ff., 2 3 T h e zuorks of Plato, B. Jowett, trans., 3 : 26-35, New
New York, Wiley, 1962. York, Dial Press.
480 HERBERT A. SIMOX [PROC.
AMER.
PHIL. SOC.

complete, we can reproduce the object from the of symbolic structures. Let me spin out some of
description. Whatever the exact mechanism of the consequences of the latter comparison.
reproduction, the description provides us with the If genetic material is a program-viewed in its
necessarv information. relation to the organism-it is a program with
S o w we have seen that the descriptions of com- special and peculiar properties. First, it is a self-
plex systems can take many forms. I n particular, reproducing program ; we have already considered
we can have state descriptions or we can have its possible copying mechanism. Second, it is a
process descriptions ; blueprints or recipes. Re- program that has developed by Darwinian evolu-
productive processes could be built around either tion. On the basis of our watchmaker's argument,
of these sources of information. Perhaps the we may assert that many of its ancestors were also
simplest possibility is for the complex system to viable programs-programs for the subassemblies.
serve as a description of itself-a template on Are there any other conjectures we can make
which a copy can be formed. One of the most about the structure of this program? There is a
plausible current theories, for example, of the well-known generalization in biology that is ver-
reproduction of deoxyribonucleic acid ( D N A ) bally so neat that we would be reluctant to give
proposes that a D N A molecule, in the form of a it up even if the facts did not support it: ontogeny
double helix of matching parts (each essentially recapitulates phylogeny. The individual organism,
a "negative" of the other), unwinds to allow each in its development, goes through stages that re-
half of the helix to serve as a template on which semble some of its ancestral forms. The fact that
a new matching half can form. the human embryo develops gill bars and then
On the other hand, our current knowledge of modifies them for other purposes is a familiar
how D S X controls the metabolism of the organ- particular belonging to the generalization. Biolo-
ism suggests that reproduction by template is only gists today like to emphasize the qualifications of
one of the processes involved. According to the the principle-that ontogeny recapitulates only the
prevailing theory, D S X serves as a template both grossest aspects of phylogeny, and these only
for itself and for the related substance ribonucleic crudely. These qualifications should not make us
acid (RS-A). RNA, in turn, serves as a template lose sight of the fact that the generalization does
for protein. But proteins-according to current hold in rough approximation-it does summarize
knowledge-guide the organism's metabolism not a very significant set of facts about the organism's
by the template method but by serving as catalysts development. How can we interpret these facts?
to govern reaction rates in the cell. IVhile R N A One way to solve a complex problem is to
is a blueprint for protein, protein is a recipe for reduce it to a problem previously solved-to show
metabolisn~.'~ what steps lead from the earlier solution to a solu-
tion of the new problem. If, around the turn of
ONTOGENY RECAPITULATES PHYLOGENY the century, we wanted to instruct a workman
The D S A in the chromosomes of an organism to make an automobile, perhaps the simplest way
contains some. and ~ e r h a most,~ s of the informa- would have been to tell him how to modify a
tion that is needed to determine its development wagon by removing the singletree and adding a
and activity. IVe have seen that, if current theo- motor and transmission. Similarly, a genetic
ries are even approximately correct, the informa- program could be altered in the course of evolu-
tion is recorded-not as a state descri~tionof the tion by adding new processes that would modify
organism
" but as a series of "instructions" for the a simpler form into a more complex one-to con-
construction and maintenance of the organism struct a gastrula, take a blastula and alter it!
from nutrient materials. I have already used the The genetic description of a single cell may,
metaphor of a recipe ; I could equally well compare therefore, take a quite different form from the
it with a computer program, which is also a se- genetic description that assembles cells into a
quence of instructions, governing the construction multi-celled organism. Multiplication by cell divi-
14 C. B. Anfinsen, T h e ?~zolcctllar basis of e v o l u t i o n ,
sion would require, as a minimum, a state descrip-
chs. 3 and 10, New York, Wiley, 1959, will qualify this tion (the DNA, say), and a simple "interpretive
sketchy, oversimplified account. For an imaginative dis- processn-to use the term from computer language
cussion of some mechanisms of process description that -that copies this description as a part of the
could govern molecular structure, see H. H. Pattee, On
the origin of macromolecular sequences, B i o p h y s i c a l Jolrr- larger copying process of cell division. But such
~ t a l1 : 683-710, 1961. a mechanism clearly would not suffice for the
 VOL. 106, SO.6, 19621 T H E ARCHITECTURE O F COMPLEXITY 481

differentiation of cells in development. It appears realm of biology. It can be applied as readily,

illore natural to conceptualize that mechanism as for example, to the transmission of knowledge in
based on a process description, and a somewhat the educational process. In most subjects, par-
more conlplex interpretive process that produces ticularly in the rapidly advancing sciences, the
the adult organism in a sequence of stages, each progress from elementary to advanced courses is
new stage in development representing the effect of to a considerable extent a progress through the
an operator upon the previous one. conceptual history of the science itself. Fortu-
It is harder to conceptualize the interrelation of nately, the recapitulation is seldom literal-any
these two descriptions. Interrelated they must more than it is in the biological case. \Ve do not
be, for enough has been learned of gene-enzyme teach the phlogiston theory in chemistry in order
mechanisins to show that these play a major role later to correct it. ( I am not sure I could not
in development as in cell metabolism. The single cite examples in other subjects where we do ex-
clue we obtain from our earlier discussion is that actly that.) But curriculum revisions that rid us
the description may itself be hierarchical, or nearly of the accumulations of the past are infrequent
decomposable, in structure, the lower levels gov- and painful. S o r are they always desirable-par-
erning the fast, "high-frequency" dynamics of the tial recapitulation may, in many instances, provide
individual cell, the higher level interactions gov- the most expeditious route to advanced knowledge.
erning the slow, "low-£requencyn dynamics of the
developing multi-cellular organism. SUMMARY : T H E DESCRIPTION O F COMPLEXITY
There are only bits of evidence, apart from the How complex or simple a structure is depends
facts of recapitulation, that the genetic program is critically upon the way in which we describe it.
organized in this way, but such evidence as exists Most of the complex structures found in the world
is compatible with this notion.25 T o the extent are enormously redundant, and we can use this
that \ve can differentiate the genetic information redundancy to simplify their description. But to
that governs cell metabolism from the genetic in- use it, to achieve the sinlplification, we must find
fornlation that governs the developn~entof differ- the right representation.
entiated cells in the multi-cellular organization, The notion of substituting a process description
we simplify enormously-as we have already seen for a state description of nature has played a cen-
-our task of theoretical description. But I have tral role in the development of modern science.
perhaps pressed this speculation far enough. Dynamic laws, expressed in the form of systems
The generalization that in evolving systems of differential or difference equations, have in a
whose descriptions are stored in a process lan- large number of cases provided the clue for the
guage, vie might expect ontogeny partially to re- simple description of the complex. In the pre-
capitulate phylogeny has applications outside the ceding paragraphs I have tried to show that this
characteristic of scientific inquiry is not accidental
2 5 There is considerable evidence that successive genes
or superficial. The correlation between state de-
along a chromosome often determine enzymes controlling
successive stages of protein syntheses. F o r a review of
scription and process description is basic to the
some o i this evidence, see P . E. Hartman, Transduction: functioning of any adaptive organism, to its ca-
a comparative review, in W. D. McElroy and B. Glass pacity for acting purposefully upon its environ-
(eds.), T i ~ cchemical basis o f heredzty, Baltimore, Johns ment. Our present-day understanding of genetic
Hopkins Press, 1957, a t pp. 442-454. Evidence for dif- mechanisms suggests that even in describing itself
ferential activity of genes in different tissues and a t differ-
ent stages of development is discussed by J. G. Gall, the multi-cellular organism finds a process descrip-
Chromosomal Differentiation, ill W. D. McElroy and tion-a genetically encoded program-to be the
B. Glass (eds.), T h e chenzical basis of developnzetzt, parsimonious and useful representation.
Baltimore, Johns Hopkins Press, 1958, a t pp. 103-135.
Finally, a model very like that proposed here has been CONCLUSION
independently, and far more fully, outlined by J. R. Platt,
A 'book model' of genetic information transfer in cells
Our speculations have carried us over a rather
and tissues, itz Kasha and Pullman (eds.), Hori,-otzs ipt alarming array of topics, but that is the price we
bioi.hc~r~istry,N ew York, Academic Press, forthcoming. must pay if we wish to seek properties common
Of course, this kind of mechanism is not the only one in to many sorts of complex systems. My thesis has
which development could be controlled by a process de- been that one path to the construction of a non-
scription. Induction, in the form envisaged in Spemann's
organizer theory, is based on process description, in which trivial theory of complex systems is by way of a
metabolites in already formed tissue control the next theory of hierarchy. Empirically, a large propor-
stages of development. tion of the complex systems we observe in nature
482 H E R B E R T A. SIMON [PROC.
AMER. PHIL. SOC.

exhibit hierarchic structure. On theoretical In both science and engineering, the study of
grounds we could expect complex systems to be "systems" is an increasingly popular activity. Its
hierarchies in a world in which complexity had to popularity is more a response to a pressing need
evolve from simplicity. In their dynamics, hier- for synthesizing and analyzing complexity than it
archies have a property, near-decomposability, that is to any large development of a body of knowl-
greatly simplifies their behavior. Near-decom- edge and technique for dealing with complexity.
posability also simplifies the description of a com- If this popularity is to be more than a fad, neces-
plex system, and makes it easier to understand sity will have to mother invention and provide
how the information needed for the development substance to go with the name. The explorations
or reproduction of the system can be stored in reviewed here represent one particular direction
reasonable compass. of search for such substance.