THE ARCHITECTURE OF COMPLEXITY

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

A NUMBER of proposals have been advanced in sis, and to analyze adaptiveness in terms of the
recent years for the development of “general sys- theory of selective information.3 The ideas of
tems theory” which, abstracting from properties feedback and information provide a frame of ref-
peculiar to physical, biological, or social systems, erence for viewing a wide range of situations,
would be applicable to all of them. *1 We might just as do the ideas of evolution, or relativism, of
well feel that, while the goal is laudable, systems axiomatic method, and of operationalism.
of such diverse kinds could hardly be expected to In this essay 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 similari- havioral sciences. The developments I shall dis-
ties the metaphor captures are significant or su- cuss arose in the context of specific phenomena,
perficial. but the theoretical formulations themselves make
It may not be entirely vain, however, to search little reference to details of structure. Instead they
for common properties among diverse kinds of refer primarily to the complexity of the systems
complex systems. The ideas that go by the name under view without specifying the exact content
of cybernetics constitute, if not a theory, at least a of that complexity. Because of their abstractness,
point of view that has been proving fruitful over the theories may have relevance—application
a wide range of applications.2 It has been useful would be too strong a term— to other kinds of
to look at the behavior of adaptive systems in complex systems that are observed in the social,
terms of the concepts of feedback and homeosta- biological, and physical sciences.
In recounting these developments, I shall avoid
* The ideas in this paper have been the topic of many technical detail, which can generally be found
conversations with my colleague, Allen Newell. George elsewhere. I shall describe each theory in the par-
W. Corner suggested important improvements in biologi- ticular context in which it arose. Then, I shall cite
cal content as well as editorial form. I am also indebted, some examples of complex systems, from areas
for valuable comments on the manuscript, to Richard H. of science other than the initial application, to
Meier, John R. Platt, and Warren Weaver. Some of the
conjectures about the nearly decomposable structure of which the theoretical framework appears rele-
the nucleus-atom-molecule hierarchy were checked vant. In doing so, I shall make reference to areas
against the available quantitative data by Andrew of knowledge where I am not expert—perhaps
Schoene and William Wise. My work in this area has not even literate. I feel quite comfortable in doing
been supported by a Ford Foundation grant for research
in organizations and a Carnegie Corporation grant for so before the members of this society, represent-
research on cognitive processes. To all of the above, my ing as it does the whole span of the scientific and
warm thanks, and the usual absolution.
1
scholarly endeavor. Collectively you will have
See especially the yearbooks of the Society for Gen- little difficulty, I am sure, in distinguishing in-
eral Systems Research. Prominent among the exponents stances based on idle fancy or sheer ignorance
of general systems theory are L. von Bertalanffy, K.
Boulding, R. W. Gerard, and J. G. Miller. For a more from instances that cast some light on the ways in
skeptical view—perhaps too skeptical in the light of the which complexity exhibits itself wherever it is
present discussion—see H. A. Simon and A. Newell, found in nature. I shall leave to you the final
Models: their uses and limitations, in L. D. White, ed., judgment of relevance in your respective fields.
The state of the social sciences, 66-83, Chicago, Univ. of
Chicago Press, 1956. I shall not undertake a formal definition of
2
N. Wiener, Cybernetics, New York, Wiley, 1948. For
3
an imaginative forerunner, see A. J. Lotka, Elements of C. Shannon and W. Weaver, The mathematical the-
mathematical biology, New York, Dover Publications, ory of communication, Urbana, Univ. of Illinois Press,
1951, first published in 1924 as Elements of physical 1949; W. R. Ashby, Design for a brain, New York,
biology. Wiley, 1952.

PROCEEDINGS OF THE AMERICAN PHILOSOPHICAL SOCIETY, VOL. 106, NO. 6, DECEMBER, 1962. Reprinted with permission.
VOL. 106, NO. 6, 1962] THE ARCHITECTURE OF COMPLEXITY 468

“complex systems.”4 Roughly, by a complex sys- HIERARCHIC SYSTEMS

tem I mean one made up of a large number of By a hierarchic system, or hierarchy, I mean a
parts that interact in a nonsimple way. In such system that is composed of interrelated subsys-
systems, the whole is more than the sum of the tems, each of the latter being, in turn, hierarchic
parts, not in an ultimate, metaphysical sense, but in structure until we reach some lowest level of
in the important pragmatic sense that, given the elementary subsystem. In most systems in nature,
properties of the parts and the laws of their inter- it is somewhat arbitrary as to where we leave off
action, it is not a trivial matter to infer the proper- the partitioning and what subsystems we take as
ties of the whole. In the face of complexity, an elementary. Physics makes much use of the con-
in-principle reductionist may be at the same time cept of “elementary particle,” although particles
a pragmatic holist.5 have a disconcerting tendency not to remain ele-
The four sections that follow discuss four as- mentary very long. Only a couple of generations
pects of complexity. The first offers some com- ago, the atoms themselves were elementary parti-
cles; today, to the nuclear physicist they are
ments on the frequency with which complexity
complex systems. For certain purposes of astron-
takes the form of hierarchy —the complex sys- omy, whole stars, or even galaxies, can be re-
tem being composed of subsystems that, in turn, garded as elementary subsystems. In one kind of
have their own subsystems, and so on. The sec- biological research, a cell may be treated as an
ond section theorizes about the relation between elementary subsystem; in another, a protein
the structure of a complex system and the time molecule; in still another, an amino acid residue.
required for it to emerge through evolutionary Just why a scientist has a right to treat as ele-
processes; specifically, it argues that hierarchic mentary a subsystem that is in fact exceedingly
systems will evolve far more quickly than nonhi- complex is one of the questions we shall take up.
erarchic systems of comparable size. The third For the moment, we shall accept the fact that sci-
section explores the dynamic properties of hier- entists do this all the time and that, if they are
archically organized systems and shows how they careful scientists, they usually get away with it.
can be decomposed into subsystems in order to Etymologically, the word “hierarchy” has had a
analyze their behavior. The fourth section exam- narrower meaning than I am giving it here. The
term has generally been used to refer to a com-
ines the relation between complex systems and plex system in which each of the subsystems is
their descriptions. subordinated by an authority relation to the sys-
Thus, the central theme that runs through my tem it belongs to. More exactly, in a hierarchic
remarks is that complexity frequently takes the formal organization, each system consists of a
form of hierarchy, and that hierarchic systems “boss” and a set of subordinate subsystems. Each
have some common properties that are independ- of the subsystems has a “boss” who is the imme-
ent of their specific content. Hierarchy, I shall diate subordinate of the boss of the system. We
argue, is one of the central structural schemes shall want to consider systems in which the rela-
that the architect of complexity uses. tions among subsystems are more complex than
in the formal organizational hierarchy just de-
scribed. We shall want to include systems in
4
W. Weaver, in: Science and complexity, American which there is no relation of subordination among
Scientist 36: 536 1948, has distinguished two kinds of subsystems. (In fact, even in human organiza-
complexity, disorganized and organized. We shall be
concerned primarily with organized complexity. tions, the formal hierarchy exists only on paper;
5
See also John R. Platt, Properties of large molecules the real flesh-and-blood organization has many
that go beyond the properties of their chemical sub- interpart relations other than the lines of formal
groups, Journ. Theoret. Biol. 1: 342-358 (1961). Since authority.) For lack of a better term, I shall use
the reductionism-holism issue is a major cause de guerre “hierarchy” in the broader sense introduced in the
between scientists and humanists, perhaps we might even
hope that peace could be negotiated between the two previous paragraphs, to refer to all complex sys-
cultures along the lines of the compromise just sug- tems analyzable into successive sets of subsys-
gested. As 1 go along, I shall have a little to say about tems, and speak of “formal hierarchy” when I
complexity in the arts as well as in the natural sciences. 1 want to refer to the more specialized concept.6
must emphasize the pragmatism of my holism to distin-
guish it sharply from the position taken by W. M. El-
6
sasser in The physical foundation of biology, New York, The mathematical term “partitioning” will not do for
Pergamon Press, 1958. what I call here a hierarchy; for the set of subsystems,
469 HERBERT A. SIMON [PROC. AMER. PHIL. SOC

SOCIAL SYSTEMS chy in the same sense. In ordinary usage, we tend

I have already given an example of one kind of to reserve the word “hierarchy” for a system that
hierarchy that is frequently encountered in the is divided into a small or moderate number of
social sciences: a formal organization. Business subsystems, each of which may be further subdi-
firms, governments, universities all have a clearly vided. Hence, we do not ordinarily think of or
visible parts-within-parts structure. But formal refer to a diamond or a gas as a hierarchic struc-
organizations are not the only, or even the most ture. Similarly, a linear polymer is simply a
common, kind of social hierarchy. Almost all chain, which may be very long, of identical sub-
societies have elementary units called families, parts, the monomers. At the molecular level it is
which may be grouped into villages or tribes, and a very flat hierarchy.
these into larger groupings, and so on. If we In discussing formal organizations, the number
make a chart of social interactions, of who talks of subordinates who report directly to a single boss
to whom, the clusters of dense interaction in the is called his span of control. I will speak analo-
chart will identify a rather well-defined hierarchic gously of the span of a system, by which I shall
structure. The groupings in this structure may be mean the number of subsystems into which it is
defined operationally by some measure of fre- partitioned. Thus, a hierarchic system is flat at a
quency of interaction in this sociometric matrix. given level if it has a wide span at that level. A
diamond has a wide span at the crystal level, but
BIOLOGICAL AND PHYSICAL SYSTEMS not at the next level down, the molecular level.
In most of our theory construction in the fol-
The hierarchical structure of biological systems lowing sections we shall focus our attention on
is a familiar 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
ward from the cell, well-defined subsystems—for to very flat hierarchies.
example, nucleus, cell membrane, micro-somes, There is one important difference between the
mitochondria, 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. We detect the organelles in a cell
tioned the two main series. At the microscopic in the way we detect the raisins in a cake—they
level we have elementary particles, atoms, mole- are “visibly” differentiated substructures local-
cules, macromolecules. At the macroscopic level ized spatially in the larger structure. On the other
we have satellite systems, planetary systems, gal- hand, we propose to identify social hierarchies
axies. Matter is distributed throughout space in a not by observing who lives close to whom but by
strikingly nonuniform fashion. The most nearly observing who interacts with whom. These two
random distributions we find, gases, are not ran- points of view can be reconciled by defining hi-
dom distributions of elementary particles but erarchy in terms of intensity of interaction, but
random distributions of complex systems, i.e. observing that in most biological and physical
molecules. systems relatively intense interaction implies rela-
A considerable range of structural types is sub- tive spatial propinquity. One of the interesting
sumed under the term hierarchy as I have defined characteristics of nerve cells and telephone wires
it. By this definition, a diamond is hierarchic, for is that they permit very specific strong interac-
it is a crystal structure of carbon atoms that can tions at great distances. To the extent that interac-
be further decomposed into protons, neutrons, tions are channeled through specialized commu-
and electrons. However, it is a very “flat” hierar- nications and transportation systems, spatial pro-
chy, in which the number of first-order subsys- pinquity becomes less determinative of structure.
tems belonging to the crystal can be indefinitely
large. A volume of molecular gas is a flat hierar- SYMBOLIC SYSTEMS
One very important class of systems has been
and the successive subsets in each of these defines the omitted from my examples thus far: systems of
partitioning, independently of any systems of relations human symbolic production. A book is a hierar-
among the subsets. By “hierarchy” 1 mean the partition- chy in the sense in which I am using that term. It
ing in conjunction with the relations that hold among its is generally divided into chapters, the chapters
parts.
VOL. 106, NO. 6, 1962] THE ARCHITECTURE OF COMPLEXITY 470

into sections, the sections into paragraphs, the tion will cost, on the average, the time to assem-
paragraphs into sentences, the sentences into ble 1/p parts (the expected number assembled
clauses and phrases, the clauses and phrases into before interruption). On the other hand, Hora has
words. We may take the words as our elementary to complete 111 subassemblies of ten parts each.
units, or further subdivide them, as the linguist The probability that he will not be interrupted
often does, into smaller units. ]f the book is nar- while completing any one of these is (1 — p)10,
rative in character, it may divide into “episodes” and each interruption will cost only about the
instead of sections, but divisions there will be. time required to assemble five parts.7
The hierarchic structure of music, based on such Now if p is about 0.01—that is, there is one
units as movements, parts, themes, phrases, is well chance in a hundred that either watchmaker will
known. The hierarchic structure of products of be interrupted while adding any one part to an
the pictorial arts is more difficult to characterize, assembly—then a straightforward calculation
but I shall have something to say about it later. shows that it will take Tempus, on the average,
about four thousand times as long to assemble a
THE EVOLUTION OF COMPLEX SYSTEMS watch as Hora.
We 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 2. Tempus will lose on the average 20 times as
very fine watches. Both of them were highly re- much work for each interrupted assembly as
garded, and the phones in their workshops rang Hora [100 parts, on the average, as against
frequently—new customers were constantly call- 5]; and
ing them. However, Hora prospered, while Tem- 3. Tempus will complete an assembly only 44
pus became poorer and poorer and finally lost his times per million attempts (0.991000 = 44 ×
shop. What was the reason? 10-6), while Hora will complete nine out of
The watches the men made consisted of about ten (0.9910 = 9 × 10-1). 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 × 10-1)/(44 ×
it down—to answer the phone, say—it immedi- 10-6) = 2 × 104. Multiplying these three ra-
ately fell to pieces and had to be reassembled tios, we get
from the elements. The better the customers liked 1/111 × 100/× 0.9910/0.991000
his watches, the more they phoned him and the = 1/111 × 20 × 20,000 ~ 4,000.
more difficult it became for him to find enough
uninterrupted time to finish a watch. 7
The speculations on speed of evolution were first sug-
The watches that Hora made were no less gested by H. Jacobson’s application of information theory to
complex than those of Tempus. But he had de- estimating the time required for biological evolution. See
signed them so that he could put together subas- his paper Information, reproduction, and the origin of
semblies of about ten elements each. Ten of these life, in American Scientist, 43: 119-127, January 1955.
subassemblies, again, could be put together into a From thermodynamic considerations it is possible to
estimate the amount of increase in entropy that occurs
larger subassembly; and a system of ten of the when a complex system decomposes into its elements.
latter subassemblies constituted the whole watch. (See for example, R. B. Setlow and E. C. Pollard, Molecular
Hence, when Hora had to put down a partly as- biophysics, 63-65, Reading, Mass., Addison-Wesley Pub-
sembled watch in order to answer the phone, he lishing Co., 1962, and references cited there.) But en-
tropy is the logarithm of a probability, hence information,
lost only a small part of his work, and he assem- the negative of entropy, can be interpreted as the loga-
bled his watches in only a fraction of the man- rithm of the reciprocal of the probability—the “improb-
hours it took Tempus. ability,” so to speak. The essential idea in Jacobson’s
It is rather easy to make a quantitative analysis model is that the expected time required for the system to
of the relative difficulty of the tasks of Tempus reach a particular state is inversely proportional to the
probability of the state—hence it increases exponentially
and Hora: Suppose the probability that an inter- with the 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 probability the notion of levels and stable subassemblies, Jacobson
that Tempus can complete a watch he has started arrived at estimates of the time required for evolution so
large as to make the event rather improbable. Our analy-
without interruption is (1 – p)1000—a very small sis, carried through in the same way, but with attention to
number unless p is 0.001 or less. Each interrup- the stable intermediate forms, produces very much
smaller estimates.
471 HERBERT A. SIMON [PROC. AMER. PHIL. SOC

BIOLOGICAL EVOLUTION plex forms, once these come into existence. But
What lessons can we draw from our parable for this is nothing more than survival of the fittest—
biological evolution? Let us interpret a partially i.e., of the stable.
completed sub-assembly of k elementary parts as Second, not all large systems appear hierarchi-
the coexistence of k parts in a small volume— cal. For example, most polymers—such as ny-
ignoring their relative orientations. lon—are simply linear chains of large numbers of
The model assumes that parts are entering the identical components, the monomers. However,
volume at a constant rate, but that there is a con- for present purposes we can simply regard such a
stant probability, p, that the part will be dispersed structure as a hierarchy with a span of one—the
before another is added, unless the assembly limiting case. For a chain of any length repre-
reaches a stable state. These assumptions are not sents a state of relative equilibrium.8
particularly realistic. They undoubtedly underes- Third, the evolution of complex systems from
timate the decrease in probability of achieving simple elements implies nothing, one way or the
the assembly with increase in the size of the as- other, about the change in entropy of the entire
sembly. Hence the assumptions understate— system. If the process absorbs free energy, the
probably by a large factor—the relative advan- complex system will have a smaller entropy than
tage of a hierarchic structure. the elements; if it releases free energy, the oppo-
Although we cannot, therefore, take the nu- site will be true. The former alternative is the one
merical estimate seriously, the lesson for biologi- that holds for most biological systems, and the
cal evolution is quite clear and direct. The time net inflow of free energy has to be supplied from
required for the evolution of a complex form the sun or some other source if the second law of
from simple elements depends critically on the thermodynamics is not to be violated. For the
numbers and distribution of potential intermedi- evolutionary process we are describing, the equi-
ate stable forms. In particular, if there exists a libria of the intermediate states need have only
hierarchy of potential stable “subassemblies,” local and not global stability, and they may be
with about the same span, s, at each level of the stable only in the steady state—that is, as long as
hierarchy, then the time required for a subassem- there is an external source of free energy that
bly can be expected to be about the same at each may be drawn upon.9
level—that is, proportional to 1/(1 — p)s. The Because organisms are not energetically closed
time required for the assembly of a system of n systems, there is no way to deduce the direction,
elements will be proportional to logs n, that is, to much less the rate, of evolution from classical
the number of levels in the system. One would thermodynamic considerations. All estimates in-
say—with more illustrative than literal intent— dicate that the amount of entropy, measured in
that the time required for the evolution of multi- physical units, involved in the formation of a
celled organisms from single-celled organisms one-celled biological organism is trivially
might be of the same order of magnitude as the small—about –10-11 cal/degree.10 The “improb-
time required for the evolution of single-celled ability” of evolution has nothing to do with this
organisms from macromolecules. The same ar- quantity of entropy, which is produced by every
gument could be applied to the evolution of pro- bacterial cell every generation. The irrelevance of
teins from amino acids, of molecules from atoms,
of atoms from elementary particles. 8
There is a well-developed theory of polymer size,
A whole host of objections to this oversimpli- based on models of random assembly. See, for example,
fied scheme will occur, I am sure, to every work- P. J. Flory, Principles of polymer chemistry, ch. 8, Ithaca,
Cornell Univ. Press, 1953. Since all subassemblies in the
ing biologist, chemist, and physicist. Before turn- polymerization theory are stable, limitation of molecular
ing to matters I know more about, I shall mention growth depends on “poisoning” of terminal groups by
three of these problems, leaving the rest to the impurities or formation of cycles rather than upon disrup-
attention of the specialists. tion
9
of partially formed chains.
First, in spite of the overtones of the watch- This point has been made many times before, but it
cannot be emphasized too strongly. For further discus-
maker parable, the theory assumes no teleologi- sion, see Setlow and Pollard, op cit., 49-64; E. Schrod-
cal mechanism. The complex forms can arise inger, What Is life? Cambridge Univ. Press, 1945; and H.
from the simple ones by purely random proc- Linschitz, The information content of a bacterial cell, in
esses. (I shall propose another model in a mo- H. Questler ed., Information theory in biology, 251-262,
ment that shows this clearly.) Direction is pro- Urbana, Univ. of Illinois Press, 1953.
10
See Linschitz, op. cit. This quantity, 10-11 cal/degree,
vided to the scheme by the stability of the com- corresponds to about 1013 bits of information.
VOL. 106, NO. 6, 1962] THE ARCHITECTURE OF COMPLEXITY 472

quantity of information, in this sense, to speed of A little reflection reveals that cues signaling
evolution can also be seen from the fact that ex- progress play the same role in the problem-
actly as much information is required to “copy” a solving process that stable intermediate forms
cell through the reproductive process as to pro- play in the biological evolutionary process. In
duce the first cell through evolution. fact, we can take over the watchmaker parable
The effect of the existence of stable intermedi- and apply it also to problem solving. In problem
ate forms exercises a powerful effect on the evo- solving, a partial result that represents recogniz-
lution of complex forms that may be likened to able progress toward the goal plays the role of a
the dramatic effect of catalysts upon reaction stable subassembly.
rates and steady-state distribution of reaction Suppose that the task is to open a safe whose
products in open systems.11 In neither case does lock has 10 dials, each with 100 possible settings,
the entropy change provide us with a guide to numbered from 0 to 99. How long will it take to
system behavior. open the safe by a blind trial-and-error search for
the correct setting? Since there are 10010 possible
PROBLEM SOLVING AS NATURAL SELECTION settings, we may expect to examine about one
Let us turn now to some phenomena that have half of these, on the average, before finding the
no obvious connection with biological evolution: correct one—that is, 50 billion billion settings.
human problem-solving processes. Consider, for Suppose, however, that the safe is defective, so
example, the task of discovering the proof for a that a click can be heard when any one dial is
difficult theorem. The process can be—and often turned to the correct setting. Now each dial can
has been—described as a search through a maze. be adjusted independently and does not need to
Starting with the axioms and previously proved be touched again while the others are being set.
theorems, various transformations allowed by the The total number of settings that have to be tried
rules of the mathematical systems are attempted, is only 10 x 50, or 500. The task of opening the
to obtain new expressions. These are modified in safe has been altered, by the cues the clicks pro-
turn until, with persistence and good fortune, a vide, from a practically impossible one to a triv-
sequence or path of transformations is discovered ial one.13
that leads to the goal. A considerable amount has been learned in the
The process ordinarily involves much trial and past five years about the nature of the mazes that
error. Various paths are tried; some are aban- represent common human problem-solving tasks—
doned, others are pushed further. Before a solu- proving theorems, solving puzzles, playing chess,
tion is found, many paths of the maze may be making investments, balancing assembly lines, to
explored. The more difficult and novel the prob- mention a few. All that we have learned about
lem, the greater is likely to be the amount of trial these mazes points to the same conclusion: that
and error required to find a solution. At the same human problem solving, from the most blundering
time, the trial and error is not completely random to the most insightful, involves nothing more than
or blind; it is, in fact, rather highly selective. The varying mixtures of trial and error and selectivity.
new expressions that are obtained by transform- The selectivity derives from various rules of
ing given ones are examined to see whether they
represent progress toward the goal. Indications of
lem solving, W. R. Ashby, Design for an intelligence
progress spur further search in the same direc- amplifier, 215-233 in C. E. Shannon and J. McCarthy,
tion; lack of progress signals the abandonment of Automata studies, Princeton, Princeton Univ. Press, 195).
13
a line of search. Problem solving requires selec- The clicking safe example was supplied by D. P.
tive trial and error.12 Simon. Ashby, op. cit., 230, has called the selectivity
involved in situations of this kind “selection by compo-
11
nents.” The even greater reduction in time produced by
See H. Kacser, Some physico-chemical aspects of hierarchization in the clicking safe example, as compared
biological organization, Appendix, pp. 191-249, in C. H. with the watchmaker’s metaphor, is due to the fact that a
Waddington, The strategy of the genes, London, George random search for the correct combination is involved in
Allen
12
& Unwin, 1957. the former case, while in the latter the parts come to-
See A. Newell, J. C. Shaw, and H. A. Simon, Em- gether in the right order. It is not clear which of these
pirical explorations of the logic theory machine, Pro- metaphors provides the better model for biological evolu-
ceedings of the 1957 Western Joint Computer Confer- tion, but we may be sure that the watchmaker’s metaphor
ence, February 1957, New York: Institute of Radio Engi- gives an exceedingly conservative estimate of the savings
neers; Chess-playing programs and the problem of com- due to hierarchization. The safe may give an excessively
plexity, IBM Journal of Research and Development 2: high estimate because it assumes all possible arrange-
320-335, October 1958; and for a similar view of prob- ments of the elements to be equally probable.
473 HERBERT A. SIMON [PROC. AMER. PHIL. SOC

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 promis- later combined with the Persian subassembly and
ing. We do not need to postulate processes more others into Alexander’s greater system. On 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 subsys-
are cut down to quite reasonable size.14 tems that had composed it.
The watchmaker argument implies that if one
THE SOURCES OF SELECTIVITY would be Alexander, one should be born into a
When we examine the sources from which the world where large stable political systems al-
problem-solving system, or the evolving system, ready exist. Where this condition was not ful-
as the case may be, derives its selectivity, we dis- filled, as on the Scythian and Indian frontiers,
cover that selectivity can always be equated with Alexander found empire building a slippery busi-
some kind of feedback of information from the ness. So too, T. E. Lawrence’s organizing of the
environment. Arabian revolt against the Turks was limited by
Let us consider the case of problem solving first. the character of his largest stable building blocks,
There are two basic kinds of selectivity. One we the separate, 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
In the same way, in organic evolution, various upon my fancy, therefore, but shall leave it to
complexes come into being, at least evanescently, historians to decide whether anything can be
and those that are stable provide new building learned for the interpretation of history from an
blocks for further construction. It is this informa- abstract theory of hierarchic complex systems.
tion about stable configurations, and not free en-
CONCLUSION: THE EVOLUTIONARY EXPLANATION
ergy or negentropy from the sun, that guides the
process of evolution and provides the selectivity OF HIERARCHY
that is essential to account for its rapidity. We have shown thus far that complex systems
The second source of selectivity in problem will evolve from simple systems much more rap-
solving is previous experience. We see this par- idly if there are stable intermediate forms than if
ticularly clearly when the problem to be solved is there are not. The resulting complex forms in the
similar to one that has been solved before. Then, former case will be hierarchic. We have only to
by simply trying again the paths that led to the turn the argument around to explain the observed
earlier solution, or their analogues, trial-and-error predominance of hierarchies among the complex
search is greatly reduced or altogether eliminated. systems 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 is the time to evolve. The hypothesis that complex-
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 inheri- the complex systems we encounter in nature, ex-
tance of acquired characteristics, but at the level amples of both forms are prominent. A more
of genetic material, of course; that is, only char- complete theory than the one we have developed
acteristics acquired by the genes can be inherited. here would presumably have something to say
We shall return to the topic of reproduction in the about the determinants of width of span in these
final section of this paper. systems.
ON EMPIRES AND EMPIRE BUILDING NEARLY DECOMPOSABLE SYSTEMS
We 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—
i.e., among the parts of those subsystems—on the
14
A. Newell and H. A. Simon, Computer simulation of other. The interactions at the different levels may
human thinking, Science 134: 2011-2017, December 22, be, and often will be, of different orders of mag-
1961.
VOL. 106, NO. 6, 1962] THE ARCHITECTURE OF COMPLEXITY 474

A1 A2 A3 B1 B2 C1 C2 C3 (a) in a nearly decomposable system, the short-

run behavior of each of the component subsys-
A1 — 100 — 2 — — — — tems is approximately independent of the short-
A2 100 — 100 1 1 — — — run behavior of the other components; (b) in the
long run, the behavior of any one of the compo-
A2 — 100 — — 2 — — — nents depends in only an aggregate way on the
B1 2 1 — — 100 2 1 — behavior of the other components.
B2 — 1 2 100 — — 1 2 Let me provide a very concrete simple example
of a nearly decomposable system.15 Consider a
C1 — — — 2 — — 100 — building whose outside walls provide perfect
C2 — — — 1 — 100 — 100 thermal insulation from the environment. We
shall take these walls as the boundary of our sys-
C3 — — — — 2 — 100 — tem. The building is divided into a large number
FIG. 1. A Hypothetical Nearly Decomposable Sys-
of rooms, the walls between them being good,
but not perfect, insulators. The walls between
tem. In terms of the heat-exchange example of
the text, Al, A2, and A3 may be interpreted as
rooms are the boundaries of our major subsys-
tems. Each room is divided by partitions into a
cubicles in one room, Bl and B2 as cubicles in a
second room, and Cl, C2, and C3 as cubicles in a
number of cubicles, but the partitions are poor
third. The matrix entries then are the heat diffu-
insulators. A thermometer hangs in each cubicle.
sion coefficients between cubicles.
Suppose that at the time of our first observation
of the system there is a wide variation in tem-
perature from cubicle to cubicle and from room
Al Bl Cl to room—the various cubicles within the building
A2 C2 are in a state of thermal disequilibrium. When we
A3 B2 C3 take new temperature readings several hours
later, what shall we find? There will be very little
nitude. In a formal organization there will gener- variation in temperature among the cubicles
ally be more interaction, on the average, between within each single room, but there may still be
two employees who are members of the same large temperature variations among rooms. When
department than between two employees from we take readings again several days later, we find
different departments. In organic substances, in- an almost uniform temperature throughout the
termolecular forces will generally be weaker than building; the temperature differences among
molecular forces, and molecular forces weaker rooms have virtually disappeared.
than nuclear forces. We can describe the process of equilibration
In a rare gas, the intermolecular forces will be formally by setting up the usual equations of heat
negligible compared to those binding the molecules— flow. The equations can be represented by the
we can treat the individual particles, for many matrix of their coefficients, rij, where rij is the
purposes, as if they were independent of each rate at which heat flows from the ith cubicle to
other. We can describe such a system as decom- the jth cubicle per degree difference in their tem-
posable into the subsystems comprised of the indi- peratures. If cubicles i and j do not have a com-
vidual particles. As the gas becomes denser, mo- mon wall, rij will be zero. If cubicles i and j have
lecular interactions become more significant. But a common wall and are in the same room, rij will
over some range, we can treat the decomposable be large. If cubicles i and j are separated by the
case as a limit and as a first approximation. We
can use a theory of perfect gases, for example, to 15
describe approximately the behavior of actual This discussion of near decomposability is based
gases if they are not too dense. As a second ap- upon H. A. Simon and A. Ando, Aggregation of vari-
ables in dynamic systems, Econometrics 29: 111-138,
proximation, we may move to a theory of nearly April 1961. The example is drawn from the same source,
decomposable systems, in which the interactions 117-118. The theory has been further developed and ap-
among the subsystems are weak, but not negligible. plied to a variety of economic and political phenomena
At least some kinds of hierarchic systems can be by Ando and F. M. Fisher. See F. M. Fisher, On the cost
of approximate specification in simultaneous equation
approximated successfully as nearly decomposable estimation, Econometrica 29: 139-170, April 1961, and
systems. The main theoretical findings from the F. M. Fisher and A. Ando, Two theorems on Ceteris
approach can be summed up in two propositions: Paribus in the analysis of dynamic systems, American
Political Science Review 61: 103-113, March 1962.
475 HERBERT A. SIMON [PROC. AMER. PHIL. SOC

wall of a room, rij will be nonzero but small. are associated, in general, with the main flows of
Hence, by grouping together all the cubicles that raw materials and semifinished products within
are in the same room, we can arrange the matrix and between industries. An input-output matrix
of coefficients so that all its large elements lie of the economy, giving the magnitudes of these
inside a string of square submatrices along the flows, reveals the nearly decomposable structure
main diagonal. All the elements outside these of the system—with one qualification. There is a
diagonal squares will be either zero or small (see consumption subsystem of the economy that is
Figure 7). We may take some small number, , linked strongly to variables in most of the other
as the upper bound of the extra-diagonal ele- subsystems. Hence, we have to modify our no-
ments. We shall call a matrix having these prop- tions of decomposability slightly to accommo-
erties a nearly decomposable matrix. date the special role of the consumption subsys-
Now it has been proved that a dynamic system tem in our analysis of the dynamic behavior of
that can be described by a nearly decomposable the economy.
matrix has the properties, stated earlier, of a In the dynamics of social systems, where mem-
nearly decomposable system. In our simple ex- bers of a system communicate with and influence
ample of heat flow this means that in the short other members, near decomposability is generally
run each room will reach an equilibrium tempera- very prominent. This is most obvious in formal
ture (an average of the initial temperatures of its organizations, where the formal authority relation
offices) nearly independently of the others; and connects each member of the organization with
that each room will remain approximately in a one immediate superior and with a small number
state of equilibrium over the longer period during of subordinates. Of course, many communica-
which an over-all temperature equilibrium is be- tions in organizations follow other channels than
ing established throughout the building. After the the lines of formal authority. But most of these
intraroom short-run equilibria have been reached, channels lead from any particular individual to a
a single thermometer in each room will be ade- very limited number of his superiors, subordi-
quate to describe the dynamic behavior of the nates, and associates. Hence, departmental
entire system—separate thermometers in each boundaries play very much the same role as the
cubicle will be superfluous. walls in our heat example.
NEAR DECOMPOSABILITY OF SOCIAL SYSTEMS PHYSICO-CHEMICAL SYSTEMS
As a glance at Figure 1 shows, near decom- In the complex systems familar in biological
posability is a rather strong property for a matrix chemistry, a similar structure is clearly visible.
to possess, and the matrices that have this prop- Take the atomic nuclei in such a system as the
erty will describe very special dynamic sys- elementary parts of the system, and construct a
tems—vanishingly few systems out of all those matrix of bond strengths between elements.
that are thinkable. How few they will be depends, There will be matrix elements of quite different
of course, on how good an approximation we orders of magnitude. The largest will generally
insist upon. If we demand that epsilon be very correspond to the covalent bonds, the next to the
small, correspondingly few dynamic systems will ionic bonds, the third group to hydrogen bonds,
fit the definition. But we have already seen that still smaller linkages to van der Waals forces.16 If
in the natural world nearly decomposable sys- we select an epsilon just a little smaller than the
tems are far from rare. On the contrary, systems magnitude of a covalent bond, the system will
in which each variable is linked with almost decompose into subsystems—the constituent
equal strength with almost all other parts of the molecules. The smaller linkages will correspond
system are far rarer and less typical. to the intermolecular bonds.
In economic dynamics, the main variables are It is well known that high-energy, high-
the prices and quantities of commodities. It is frequency vibrations are associated with the
empirically true that the price of any given com-
modity and the rate at which it is exchanged de- 16
For a survey of the several classes of molecular and
pend to a significant extent only on the prices and intermolecular forces, and their dissociation energies, see
quantities of a few other commodities, together Setlow and Pollard, op. cit., Chapter 6. The energies of
with a few other aggregate magnitudes, like the typical covalent bonds are of the order of 80-100 k
average price level or some over-all measure of cal/mole, of the hydrogen bonds, 10 k cal/mole. Ionic
economic activity. The large linkage coefficients bonds generally lie between these two levels; the bonds
due to van der Waals forces are lower in energy.
VOL. 106, NO. 6, 1962] THE ARCHITECTURE OF COMPLEXITY 476

smaller physical subsystems, low-frequency vi- same as the number associated with the individ-
brations with the larger systems into which the ual atoms, the bonding process can go on indefi-
subsystems are assembled. For example, the ra- nitely—the atoms can form crystals or polymers
diation frequencies associated with molecular of indefinite extent. If the number of bonds of
vibrations are much lower than those associated which the composite is capable is less than the
with the vibrations of the planetary electrons of number associated with each of the parts, then
the atoms; the latter, in turn, are lower than those the process of agglomeration must come to a halt.
associated with nuclear processes.17 Molecular We need only mention some elementary exam-
systems are nearly decomposable systems, the ples. Ordinary gases show no tendency to ag-
short-run dynamics relating to the internal struc- glomerate, because the multiple bonding of at-
tures of the subsystems, the long-run dynamics to oms “uses up” their capacity to interact. While
the interactions of these subsystems. each oxygen atom has a valence of two, the O2
A number of the important approximations em- molecules have a zero valence. Contrariwise, in-
ployed in physics depend for their validity on the definite chains of single-bonded carbon atoms
near decom-posability of the systems studied. can be built up, because a chain of any number of
The theory of the thermodynamics of irreversible such atoms, each with two side groups, has a va-
processes, for example, requires the assumption lence of exactly two.
of macroscopic disequilibrium but microscopic Now what happens if we have a system of ele-
equilibrium,18 exactly the situation described in ments that possess both strong and weak interac-
our heat-exchange example. Similarly, computa- tion capacities and whose strong bonds are ex-
tions in quantum mechanics are often handled by haustible through combination? Subsystems will
treating weak interactions as producing perturba- form, until all the capacity for strong interaction
tions on a system of strong interactions. is utilized in their construction. Then these sub-
systems will be linked by the weaker second-order
SOME OBSERVATIONS ON HIERARCHIC SPAN bonds into larger systems. For example, a water
To understand why the span of hierarchies is molecule has essentially a valence of zero—all the
sometimes very broad—as in crystals—and potential covalent bonds are fully occupied by
sometimes narrow, we need to examine more the interaction of hydrogen and oxygen molecules.
detail of the interactions. In general, the critical But the geometry of the molecule creates an elec-
consideration is the extent to which interaction tric dipole that permits weak interaction between
between two (or a few) subsystems excludes in- the water and salts dissolved in it— whence such
teraction of these subsystems with the others. Let phenomena as its electrolytic conductivity.19
us examine first some physical examples. Similarly, it has been observed that, although
Consider a gas of identical molecules, each of electrical forces are much stronger than gravita-
which can form covalent bonds, in certain ways, tional forces, the latter are far more important
with others. Let us suppose that we can associate than the former for systems on an astronomical
with each atom a specific number of bonds that it scale. The explanation, of course, is that the elec-
is capable of maintaining simultaneously. (This trical forces, being bipolar, are all “used up” in
number is obviously related to the number we the linkages of the smaller subsystems, and that
usually call its valence.) Now suppose that two significant net balances of positive or negative
atoms join and that we can also associate with the charges are not generally found in regions of
combination a specific number of external bonds macroscopic size.
it is capable of maintaining. If this number is the In social as in physical systems there are gener-
ally limits on the simultaneous interaction of
17 large numbers of subsystems. In the social case,
Typical wave numbers for vibrations associated with these limits are related to the fact that a human
various systems (the wave number is the reciprocal of
wave length hence proportional-10to frequency): being is more nearly a serial than a parallel in-
steel wire under tension—10 to 10-9 cm-1 formation-processing system. He can carry on
molecular rotations—100 2to 102 3cm-1-1 only one conversation at a time, and although this
molecular vibrations—10 to 10 cm does not limit the size of the audience to which a
4
planetary electrons—10 to 105 cm-1
nuclear rotations—109 to 1010 cm-1 mass communication can be addressed, it does
nuclear
18
surface vibrations—1011 to 1012 cm-1.
S. R. de Groot, Thermodynamics of irreversible
19
processes, 11-12, New York, Interscience Publishers, See, for example, L. Pauling, General chemistry, ch.
1951. 15.
477 HERBERT A. SIMON [PROC. AMER. PHIL. SOC

limit the number of people simultaneously involved subparts belonging to different parts has no place
in most other forms of social interaction. Apart in the outline and is likely to be lost. The loss of
from requirements of direct interaction, most roles such information and the preservation mainly of
impose tasks and responsibilites that are time con- information about hierarchic order is a salient
suming. One cannot, for example, enact the role characteristic that distinguishes the drawings of a
of “friend” with large numbers of other people. child or someone untrained in representation
It is probably true that in social as in physical from the drawing of a trained artist. (I am speak-
systems, the higher-frequency dynamics are as- ing of an artist who is striving for representation.)
sociated with the subsystems, the lower-
frequency dynamics with the larger systems. It is NEAR DECOMPOSABILITY AND COMPREHENSIBILITY
generally believed, for example, that the relevant From our discussion of the dynamic properties
planning horizon of executives is longer, the of nearly decomposable systems, we have seen
higher their location in the organizational hierar- that comparatively little information is lost by
chy. It is probably also true that both the average representing them as hierarchies. Subparts be-
duration of an interaction between executives and longing to different parts only interact in an ag-
the average interval between interactions are gregative fashion—the detail of their interaction
greater at higher than at lower levels. can be ignored. In studying the interaction of two
large molecules, generally we do not need to
SUMMARY: NEAR DECOMPOSABILITY consider in detail the interactions of nuclei of the
We have seen that hierarchies have the prop- atoms belonging to the one molecule with the
erty of near decomposability. Intracomponent nuclei of the atoms belonging to the other. In
linkages are generally stronger than intercompo- studying the interaction of two nations, we do not
nent linkages. This fact has the effect of separat- need to study in detail the interactions of each
ing the high-frequency dynamics of a hierar- citizen of the first with each citizen of the second.
chy—involving the internal structure of the com- The fact, then, that many complex systems
ponents—from the low-frequency dynamics— have a nearly decomposable, hierarchic structure
involving interaction among components. We is a major facilitating factor enabling to us under-
shall turn next to some important consequences stand, to describe, and even to “see” such sys-
of this separation for the description and compre- tems and their parts. Or perhaps the proposition
hension of complex systems. should be put the other way round. If there are
important systems in the world that are complex
THE DESCRIPTION OF COMPLEXITY without being hierarchic, they may to a consider-
able extent escape our observation and our un-
If you ask a person to draw a complex object— derstanding. Analysis of their behavior would
such as a human face—he will almost always involve such detailed knowledge and calculation
proceed in a hierarchic fashion.20 First he will of the interactions of their elementary parts that it
outline the face. Then he will add or insert fea- would be beyond our capacities of memory or
tures: eyes, nose, mouth, ears, hair. If asked to computation.21
elaborate, he will begin to develop details for
each of the features—pupils, eyelids, lashes for
21
the eyes, and so on—until he reaches thelimitsof I believe the fallacy in the central thesis of W. M.
his anatomical knowledge. His information about Elsasser’s The physical foundation of biology, mentioned
earlier, lies in his ignoring the simplification in description
the object is arranged hierarchically in memory, of complex systems that derives from their hierarchic
like a topical outline. structure. Thus (p. 155): “If we now apply similar argu-
When information is put in outline form, it is ments to the coupling of enzymatic reactions with the
easy to include information about the relations substratum of protein molecules, we see that over a suffi-
among the major parts and information about the cient period of time, the information corresponding to the
structural details of these molecules will be communi-
internal relations of parts in each of the suboutli- cated to the dynamics of the cell, to higher levels of or-
nes. Detailed information about the relations of ganization as it were, and may influence such dynamics.
While this reasoning is only qualitative, it lends credence
20
to the assumption that in the living organism, unlike the
George A. Miller has collected protocols from sub- inorganic crystal, the effects of microscopic structure can-
jects who were given the task of drawing faces and finds not be simply averaged out; as time goes on this influ-
that they behave in the manner described here (private ence will pervade the behavior of the cell ‘at all levels.’”
communication). See also E. H. Gombrich, Art and illu- But from our discussion of near decomposability, it
sion, 291-296, New York, Pantheon Books, 1960. would appear that those aspects of microstructure that
VOL. 106, NO. 6, 1962] THE ARCHITECTURE OF COMPLEXITY 478

I shall not try to settle which is chicken and AB

which is egg: whether we are able to understand pattern CD , for example, occurs four times in the
the world because it is hierarchic or whether it total pattern, it is economical to represent it by
appears hierarchic because those aspects of it the single symbol, a.
which are not elude our understanding and ob- If a complex structure is completely unredun-
servation. I have already given some reasons for dant—if no aspect of its structure can be inferred
supposing that the former is at least half the from any other—then it is its own simplest de-
truth—that evolving complexity would tend to be scription. We can exhibit it, but we cannot de-
hierarchic—but it may not be the whole truth. scribe it by a simpler structure. The hierarchic
SIMPLE DESCRIPTIONS OF COMPLEX SYSTEMS structures we have been discussing have a high
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 that only a few different kinds of subsystems, in vari-
the description be as cumbersome as the object ous combinations and arrangements. A familiar
described. A trivial example will show how a example is the proteins, their multitudinous vari-
system can be described economically. Suppose ety arising from arrangements of only twenty
the system is a two-dimensional array like this: different amino acids. Similarly, the ninety-odd
A B M N R S H I elements provide all the kinds of building blocks
C D O P T U J K needed for an infinite variety of molecules. Hence,
M N A B H I R S we can construct our description from a restricted
O P C D J K T U alphabet of elementary terms corresponding to
R S H I A B M N the basic set of elementary subsystems from
T U J K C D O P which the complex system is generated.
H I R S M N A B 2. Hierarchic systems are, as we have seen, of-
J K T U O P C D ten nearly decomposable. Hence only aggregative
Let us call the array 
AB  MN properties of their parts enter into the description
CD a, the array  OP  of the interactions of those parts. A generaliza-
m, the array 
RS  HI tion of the notion of near decomposability might
TU r, and the array JK h. Let us be called the “empty world hypothesis”—most
things are only weakly connected with most other
call the array 
am rh
ma w, and the array hr x. Then things; for a tolerable description of reality only a
tiny fraction of all possible interactions needs to
the entire array is simply 
wx be taken into account. By adopting a descriptive
xw. While the origi- language that allows the absence of something to
nal structure consisted of 64 symbols, it requires go unmentioned, a nearly empty world can be
only 35 to write down its description: described quite concisely. Mother Hubbard did
wx not have to check off the list of possible contents
s = xw to say that her cupboard was bare.
am rh 3. By appropriate “receding,” the redundancy
w = ma x = hr that is present but unobvious in the structure of a
AB MN RS HI complex system can often be made patent. The
a =CD m = OP r = TU h =JK commonest receding of descriptions of dynamic
systems consists in replacing a description of the
We achieve the abbreviation by making use of time path with a description of a differential law
the redundancy in the original structure. Since the that generates that path. The simplicity, that is,
resides in a constant relation between the state of
the system at any given time and the state of the
control the slow developmental aspects of organismic system a short time later. Thus, the structure of
dynamics can be separated out from the aspects that con- the sequence 1 3 5 7 9 11 ... is most simply ex-
trol the more rapid cellular metabolic processes. For this pressed by observing that each member is ob-
reason we should not despair of unraveling the web of
causes. See also J. R. Platt’s review of Elsasser’s book in tained by adding 2 to the previous one. But this is
Perspectives in biology and medicine 2: 243-245, 1959. the sequence that Galileo found to describe the
479 HERBERT A. SIMON [PROC. AMER. PHIL. SOC

velocity at the end of successive time intervals of argued that all learning is remembering. He could
a ball rolling down an inclined plane. not otherwise explain how we can discover or
It is a familar proposition that the task of sci- recognize the answer to a problem unless we al-
ence is to make use of the world’s redundancy to ready know the answer.23 Our dual relation to the
describe that world simply. 1 shall not pursue the world is the source and solution of the paradox.
general methodological point here, but I shall We pose a problem by giving the state descrip-
instead take a closer look at two main types of tion of the solution. The task is to discover a se-
description that seem to be available to us in quence of processes that will produce the goal
seeking an understanding of complex systems. I state from an initial state. Translation from the
shall call these state description and process de- process description to the state description en-
scription, respectively. ables us to recognize when we have succeeded.
The solution is genuinely new to us—and we do
STATE DESCRIPTIONS AND PROCESS DESCRIPTIONS not need Plato’s theory of remembering to ex-
“A circle is the locus of all points equidistant plain 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 ba-
has returned to its starting point.” It is implicit in sically a form of means-end analysis that aims at
Euclid that if you carry out the process specified discovering a process description of the path that
in the second sentence, you will produce an ob- leads to a desired goal. The general paradigm is:
ject that satisfies the definition of the first. The given a blueprint, to find the corresponding rec-
first sentence is a state description of a circle, the ipe. Much of the activity of science is an applica-
second a process description. tion of that paradigm: given the description of
These two modes of apprehending structures some natural phenomena, to find the differential
are the warp and weft of our experience. Pictures, equations for processes that will produce the
blueprints, most diagrams, and chemical struc- phenomena.
tural formulas are state descriptions. Recipes,
differential equations, and equations for chemical THE DESCRIPTION OF COMPLEXITY IN
reactions are process descriptions. The former SELF-REPRODUCING SYSTEMS
characterize the world as sensed; they provide the The problem of finding relatively simple descrip-
criteria for identifying objects, often by modeling tions for complex systems is of interest not only
the objects themselves. The latter characterize the for an understanding of human knowledge of the
world as acted upon; they provide the means for world but also for an explanation of how a com-
producing or generating objects having the de- plex system can reproduce itself. In my discus-
sired characteristics. sion of the evolution of complex systems, I touched
The distinction between the world as sensed only briefly on the role of self-reproduction.
and the world as acted upon defines the basic Atoms of high atomic weight and complex in-
condition for the survival of adaptive organisms. organic molecules are witnesses to the fact that
The organism must develop correlations between the evolution of complexity does not imply self-
goals in the sensed world and actions in the reproduction. If evolution of complexity from
world of process. When they are made conscious simplicity is sufficiently probable, it will occur
and verbalized, these correlations correspond to repeatedly; the statistical equilibrium of the sys-
what we usually call means-end analysis. Given a tem will find a large fraction of the elementary
desired state of affairs and an existing state of particles participating in complex systems.
affairs, the task of an adaptive organism is to find If, however, the existence of a particular com-
the difference between these two states and then plex form increased the probability of the crea-
to find the correlating process that will erase the tion of another form just like it, the equilibrium
difference.22 between complexes and components could be
Thus, problem solving requires continual trans- greatly altered in favor of the former. If we have
lation between the state and process descriptions a description of an object that is sufficiently clear
of the same complex reality. Plato, in the Meno, and complete, we can reproduce the object from
the description. Whatever the exact mechanism
22
See H. A. Simon and A. Newell, Simulation of hu-
man thinking, in M. Greenberger, ed., Management and
23
the computer of the future, 95-114, esp. pp. 110 ff., New The works of Plato, B. Jowett, trans., 3: 26-35 (New
York, Wiley, 1962. York: Dial Press.
VOL. 106, NO. 6, 1962] THE ARCHITECTURE OF COMPLEXITY 480

of reproduction, the description provides us with If genetic material is a program—viewed in its

the necessary information. relation to the organism—it is a program with
Now we have seen that the descriptions of special and peculiar properties. First, it is a self-
complex systems can take many forms. In par- reproducing program; we have already considered
ticular, we can have state descriptions, or we can its possible copying mechanism. Second, it is a
have process descriptions— blueprints or recipes. program that has developed by Darwinian evolu-
Reproductive processes could be built around tion. On the basis of our watchmaker’s argument,
either of these sources of information. Perhaps we may assert that many of its ancestors were also
the simplest possibility is for the complex system viable programs—programs for the subassemblies.
to 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 deoxyribo-nucleic acid (DNA) bally so neat that we would be reluctant to give it
proposes that a DNA molecule, in the form of a up even if the facts did not support it: ontogeny
double helix of matching parts (each essentially a recapitulates phylogeny. The individual organ-
“negative” of the other), unwinds to allow each ism, in its development, goes through stages that
half of the helix to serve as a template on which a resemble some of its ancestral forms. The fact
new matching half can form. that the human embryo develops gill bars and
On the other hand, our current knowledge of then modifies them for other purposes is a famil-
how DNA controls the metabolism of the organ- iar particular belonging to the generalization. Bi-
ism suggests that reproduction by template is ologists today like to emphasize the qualifica-
only one of the processes involved. According to tions of the principle—that ontogeny recapitu-
the prevailing theory, DNA serves as a template lates only the grossest aspects of phylogeny, and
both for itself and for the related substance ribo- these only crudely. These qualifications should
nucleic acid (RNA). RNA, in turn, serves as a not make us lose sight of the fact that the gener-
template for protein. But proteins—according to alization does hold in rough approximation—it
current knowledge—guide the organism’s me- does summarize a very significant set of facts
tabolism not by the template method but by serv- about the organism’s development. How can we
ing as catalysts to govern reaction rates in the interpret these facts?
cell. While RNA is a blueprint for protein, pro- One way to solve a complex problem is to re-
tein is a recipe for metabolism.24 duce it to a problem previously solved—to show
what steps lead from the earlier solution to a so-
ONTOGENY RECAPITULATES PHYLOGENY lution of the new problem. If, around the turn of
The DNA in the chromosomes of an organism the century, we wanted to instruct a workman to
contains some, and perhaps most, of the informa- make an automobile, perhaps the simplest way
tion that is needed to determine its development would have been to tell him how to modify a
and activity. We have seen that, if current theo- wagon by removing the singletree and adding a
ries are even approximately correct, the informa- motor and transmission. Similarly, a genetic pro-
tion is recorded not as a state description of the gram could be altered in the course of evolution
organism but as a series of “instructions” for the by adding new processes that would modify a
construction and maintenance of the organism simpler form into a more complex one—to con-
from nutrient materials. I have already used the struct a gastrula, take a blastula and alter it!
metaphor of a recipe; I could equally well com- The genetic description of a single cell may,
pare it with a computer program, which is also a therefore, take a quite different form from the
sequence of instructions, governing the construc- genetic description that assembles cells into a
tion of symbolic structures. Let me spin out some multicelled organism. Multiplication by cell divi-
of the consequences of the latter comparison. sion would require, as a minimum, a state de-
scription (the DNA, say), and a simple “interpre-
24
tive process”—to use the term from computer
C. B. Anfinsen, The molecular basis of evolution, language —that copies this description as a part
chs. 3 and 10, New York, Wiley, 1959, will qualify this
sketchy, oversimplified account. For an imaginative dis- of the larger copying process of cell division. But
cussion of some mechanisms of process description that such a mechanism clearly would not suffice for
could govern molecular structure, see H. H. Pattee, On the differentiation of cells in development. It ap-
the origin of macromolecular sequences, Biophysical pears more natural to conceptualize that mecha-
Journal 1: 683-710, 1961.
481 HERBERT A. SIMON [PROC. AMER. PHIL. SOC

nism as based on a process description, and a the educational process. In most subjects, particu-
somewhat more complex interpretive process that larly in the rapidly advancing sciences, the pro-
produces the adult organism in a sequence of gress from elementary to advanced courses is to a
stages, each new stage in development representing considerable extent a progress through the con-
the effect of an operator upon the previous one. ceptual history of the science itself. Fortunately,
It is harder to conceptualize the interrelation of the recapitulation is seldom literal—any more
these two descriptions. Interrelated they must be, than it is in the biological case. We do not teach
for enough has been learned of gene-enzyme the phlogiston theory in chemistry in order later
mechanisms to show that these play a major role to correct it. (I am not sure I could not cite exam-
in development as in cell metabolism. The single ples in other subjects where we do exactly that.)
clue we obtain from our earlier discussion is that But curriculum revisions that rid us of the accu-
the description may itself be hierarchical, or mulations of the past are infrequent and painful.
nearly decomposable, in structure, the lower lev- Nor are they always desirable—partial recapitula-
els governing the fast, “high-frequency” dynam- tion may, in many instances, provide the most
ics of the individual cell, the higher-level interac- expeditious route to advanced knowledge.
tions governing the slow, “low-frequency” dy-
namics of the developing multicellular organism. SUMMARY: THE DESCRIPTION OF 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
is compatible with this notion.25 To the extent world are enormously redundant, and we can use
that we can differentiate the genetic information this redundancy to simplify their description. But
that governs cell metabolism from the genetic to use it, to achieve the simplification, we must
information that governs the development of dif- find the right representation.
ferentiated cells in the multicellular organization, The notion of substituting a process description
we simplify enormously—as we have already for a state description of nature has played a cen-
seen—our task of theoretical description. But 1 tral role in the development of modern science.
have 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, we might expect ontogeny partially to simple description of the complex. In the preced-
recapitulate phylogeny has applications outside ing paragraphs I have tried to show that this
the realm of biology. It can be applied as readily, characteristic of scientific inquiry is not acciden-
for example, to the transmission of knowledge in tal or superficial. The correlation between state
description and process description is basic to the
25
There is considerable evidence that successive genes functioning of any adaptive organism, to its ca-
along a chromosome often determine enzymes control- pacity for acting purposefully upon its environ-
ling successive stages of protein syntheses. For a review ment. Our present-day understanding of genetic
of some of this evidence, see P. E. Hartman, Transduc- mechanisms suggests that even in describing it-
tion: a comparative review, in W. D. McElroy and B. self the multi-cellular organism finds a process
Glass (eds.), The chemical basis of heredity, Baltimore,
Johns Hopkins Press, 1957, at pp. 442-454. Evidence for description—a genetically encoded program—to
differential activity of genes in different tissues and at be the parsimonious and useful representation.
different stages of development is discussed by J. G.
Gall, Chromosomal differentiation, in W. D. McElroy CONCLUSION
and B. Glass (eds.), The chemical basis of development,
Baltimore, Johns Hopkins Press, 1958, at pp. 103-135. Our speculations have carried us over a rather
Finally, a model very like that proposed here has been alarming array of topics, but that is the price we
independently, and far more fully, outlined by J. R. Platt,
A ‘book model’ of genetic information transfer in cells must pay if we wish to seek properties common
and tissues, in M. Kasha and B. Pullman (eds.), Horizons to many sorts of complex systems. My thesis has
in biochemistry, New York, Academic Press, forthcom- been that one path to the construction of a non-
ing. Of course, this kind of mechanism is not the only trivial theory of complex systems is by way of a
one in which development could be controlled by a proc-
ess description. Induction, in the form envisaged in Spe- theory of hierarchy. Empirically, a large propor-
mann’s organizer theory, is based on process description, tion of the complex systems we observe in nature
in which metabolites in already formed tissue control the exhibit hierarchic structure. On theoretical
next stages of development.
VOL. 106, NO. 6, 1962] THE ARCHITECTURE OF COMPLEXITY 482

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, hierar- for synthesizing and analyzing complexity than it
chies have a property, near decompos-ability, 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 If this popularity is to be more than a fad, neces-
complex system and makes it easier to under- sity will have to mother invention and provide
stand how the information needed for the devel- substance to go with the name. The explorations
opment or reproduction of the system can be reviewed here represent one particular direction
stored in reasonable compass. of search for such substance.
In both science and engineering, the study of