Complexity and the Economy
Page 27
alterations and innovations are very often discrete and well-marked, so that
in these contexts we can define and observe increases in complexity more
easily. And many are computer based. Thus, they can provide “laboratories”
for the real-time measurement and replication of changes in complexity in
the course of evolution.
In discussing complexity and evolution in this chapter, I will draw examples
from the economy and from several of the other contexts mentioned above, as
well as from biology. I will be interested in “complexity” seen simply as complication. Exactly what “complication” means will vary from context to context;
but it will become clear, I hope, in the mechanisms as they are discussed. And I will use the term “evolution” often in its phylogenetic sense, as development in a system with a clear lineage of inherited structures that may change over
time. Thus, we can talk about the evolution of a language, or of a technology, without having to assume that these necessarily reproduce in a population of
languages or technologies.
GROWTH IN COEVOLUTIONARY DIVERSITY
The first mechanism whereby complexity increases as evolution takes place,
I will call growth in coevolutionary diversity. It applies in systems where the individuals or entities or species or organisms coexist together in an interacting population, with some forming substrates or niches that allow the existence
of others. We may, therefore, think of such coevolving systems as organized
into loose hierarchies or “food webs” of dependence, with individuals further
on t He evolu t ion of comPlexi t y [ 145 ]
down a hierarchy depending for their existence on the existence of more fundamental ones nearer the base of the hierarchy.
When the individuals (and their multiple possibilities in interaction) in
such systems create a variety of niches that are not closed off to further newly generated individuals, diversity tends to grow in a self-reinforcing way. New
individuals that enter the population may provide new substrates, new niches.
This provides new possibilities to be filled or exploited by further new enti-
ties. The appearance of these, in turn, may provide further new niches and
substrates. And so on. By this means, complexity in the form of greater diver-
sity and a more intricate web of interactions tends to bootstrap itself upward over time. Growth in coevolutionary diversity may be slow and halting at first, as when the new individuals merely replace uncompetitive, preexisting ones.
But over time, with entities providing niches and niches making possible new
entities, it may feed upon itself; so that diversity itself provides the fuel for further diversity.
Growth in coevolutionary diversity can be seen in the economy in the
way specialized products and processes within the computer industry have
proliferated in the last two decades. As modern microprocessors came into
existence, they created niches for devices such as memory systems, screen
monitors, and bus interfaces that could be connected with them to form use-
ful hardware—computing devices. These, in turn, created a need, or niche,
for new operating system software and programming languages, and for
software applications. The existence of such hardware and software, in turn,
made possible desktop publishing, computer-aided design and manufactur-
ing, electronic mail, shared computer networks, and so on. This created niches for laser printers, engineering-design software and hardware, network servers, modems, and transmission systems. These new devices, in turn, called
forth further new microprocessors and system software to drive them. And
so, in about two decades, the computer industry has undergone an explosive
increase in diversity: from a small number of devices and software to a very
large number, as new devices make possible further new devices, and new
software products make possible new functions for computers, and these, in
turn, call forth further new devices and new software.
Of course, we should not forget that as new computer products and func-
tions for computers appear, they are often replacing something else in the
economy. Computer-aided design may eventually replace standard drawing
board and T-square design. And so the increase in diversity in one part of
a system may be partially offset by loss of diversity elsewhere. Occasionally, in a coevolving system, this replacement of an existing function can cause
a reversal in the growth of coevolutionary diversity. This happens when the new entity replaces a more fundamental one in the system and the niches
dependent on this disappear. In the economy of the last century, for exam-
ple, there was a steady increase in the numbers of specialized, interconnected
[ 146 ] Complexity and the Economy
“niche firms” in the horse-drawn transportation industry; so that by the end of the century very many different types of coach builders, harness makers,
smithy shops, and horse breeders coexisted. The appearance of the automobile
caused all this to collapse, to be replaced, in turn, by a slow-growing network of interconnected niche manufacturers dependent on gasoline technology, oil
exploration and refining, and the internal combustion engine. Thus, complex-
ity—diversity in this case—may, indeed, tend to grow in coevolving systems,
but it may also fluctuate greatly over time.
Growth in diversity can be observed in several artificial evolution con-
texts: for example, Tom Ray’s Tierra system,14 John Holland’s ECHO system,6
and Stuart Kauffman’s various chemical evolution systems.7 To take the Tierra
example, Ray sets up an artificial world in which computer programs compete
for processor time and memory space in a virtual computer. He begins with
a single “organism” in the form of a set of self-replicating machine language
instructions that can occasionally mutate. This forms a niche or substrate for the appearance of parasitic organisms that use part of its code to replicate—
that “feed” on its instructions. Further organisms appear that are immune
to the parasites. The parasites in turn form a substrate for hyper parasites
that feed on them. Hyper-hyper parasites appear. And so on. New “organ-
isms” continually appear and disappear, in a rich ecosystem of symbiotic and
competing machine-language programs that shows a continual net growth of
diversity. In several days of running this system, Ray found no endpoint to
the growth of diversity. Starting from a single genotype, over 29,000 differ-
ent self-replicating genotypes in 300 size classes (equivalent to species in this system) accumulated in this coevolving computer ecology.
At this point I want to note several things that apply to this mechanism.
First, the appearance of new entities may, in some cases, depend not so
much on the existence of previous entities as on their possibilities in interaction. For example, in the economy, a new technology such as the computer
laser printer mentioned above is possible only if lasers, xerography, and
computers are previously available as technologies. In these cases, symbiotic
clusters of entities—sets of entities whose collective activity or existence is important—may form many of the niches. We could predict that where collective existence is important in forming niches, growth in coevolutionary diver-
sity would be slow at first—with few entities there would be few possibilities in combination and, hen
ce, few niches. But as more single entities enter, we
would see a very rapid increase in niche possibilities, as the number of pos-
sible niche clusters that can be created undergoes a combinatorial explosion.
Second, collapses will be large if replacement by a new entity happens near
the base of the dependency hierarchy; small if near the endpoints. Therefore,
the way in which expansion and collapse of diversity actually work themselves
out in a coevolutionary system is conditioned heavily on the way dependen-
cies are structured.
on t He evolu t ion of comPlexi t y [ 147 ]
Third, two positive feedbacks—circular causalities—are inherent in this mechanism. The generation of new entities may enhance the generation of
new entities, simply because there is new “genetic material” in the system
available for further “adaptive radiation.” And the appearance of new enti-
ties provides niches for the appearance of further, new entities. In turn, these mean that where few new entities are being created, few new entities can
appear; thus, few new niches will be created. And so the system will be largely quiescent. And where new entities are appearing rapidly, there will be a rapid increase in new niches, causing further generation of entities and further
new niches. The system may then undergo a “Cambrian explosion.” Hence, we
would expect that such systems might lie dormant in long periods of relative
quiescence but burst occasionally into periods of rapid increase in complexity.
That is, we would expect them to experience punctuated equilibria.
This mechanism, whereby complexity increases via the generation of new
niches, is familiar to most of us who study complex systems. Certainly Stuart
Kauffman has written extensively on various examples of self-reinforcing
diversity. Yet strangely it is hard to find discussion of it in the traditional biological literature. Bonner’s 1988 book, The Evolution of Complexity, does not mention it, for example, although it devotes a chapter to a discussion of
complexity as diversity. Waddington16 comes somewhat closer when he sug-
gests that niches become more complex as organismal diversity increases. The
more complex niches, he suggests, are then filled by more complex organisms,
which in turn increases niche complexity. But he seems to have in mind an
upward spiral of internal structural complexity, and not of ecological diversity.
An intriguing mention of this mechanism—or something tantalizingly close
to it—comes from Darwin’s notebooks,3 p. 422.
“The enormous number of animals in the world depends, of their varied
structure and complexity . . . hence as the forms became complicated, they
opened fresh means of adding to their complexity.”[1]
But once again this could be read as having to do with internal structural
complexity, rather than ecological diversity.
STRUCTURAL DEEPENING
A second mechanism causing complexity to increase over time I will call structural deepening. This applies to single entities—systems, organisms, species, individuals—that evolve against a background that can be regarded as their
“environment.” Normally, competition exerts strong pressure for such sys-
tems to operate at their limits of performance. But they can break out of these
[1] I am grateful to Dan McShea for pointing out this quotation to me.
[ 148 ] Complexity and the Economy
limits by adding functions or subsystems that allow them to (a) operate in a wider or more extreme range, (b) sense and react to exceptional circumstances, (c) service other systems so that they operate better, and (d) enhance their
reliability. In doing so, they add to their “structural depth” or design sophistication. Of course, such functions or subsystems, once added, may operate at
their limits of performance. Once again they can break through these limits by adding sub-subsystems according to (a)–(d) above. By this process, over time
the original system becomes encrusted with deeper functions and subfunc-
tions. It may improve greatly in its performance and in the range of environ-
ment it can operate in. But in doing so, it becomes internally complex.
The history of the evolution of technology provides many examples of
structural deepening. The original gas-turbine (or jet) aero engine, designed
independently by Frank Whittle and Hans von Ohain in the 1930s, for
example, was simple.2 It compressed intake air, ignited fuel in it, released the exploding mixture through a turbine that drove the compressor, and then
exhausted the air mass at high velocity to provide thrust. Whittle’s original
prototype had one moving part, the compressor-turbine combination. But
over the years, competitive pressures felt by commercial and military inter-
ests led to constant demands for improvement. This forced designers to over-
come limits imposed by extreme stresses and temperatures, and to handle
exceptional situations, sometimes by using better materials, but more often
by adding subsystems.
And so, over time, higher air-compression ratios were achieved by using
not one, but an assembly—a system—of many compressors. Efficiency was
enhanced by a variable position guide-vane control system that admitted
more air at high altitudes and velocities and lowered the possibility of the
engine stalling. A bleed-valve control system was added to permit air to be
bled from critical points in the compressor when pressures reached certain
levels. This also reduced the tendency of the engine to stall. A secondary air-flow system was added to cool the red-hot turbine blades and pressurize sump
cavities to prevent lubrication leakage. Turbine blades were also cooled by a
system that circulated air inside them. To provide additional thrust in mili-
tary air combat conditions, afterburner assemblies were added. To handle the
possibility of engine fires, sophisticated fire-detection systems were added.
To prevent the build up of ice in the intake region, deicing assemblies were
added. Specialized fuel systems, lubrication systems, variable exhaust-nozzle
systems, and engine-starting systems were added.
But all these required further subsystems, to monitor and control them and to enhance their performance when they ran into limitations. These subsystems, in turn, required sub-subsystems to enhance their performance. A mod-
ern, aero gas turbine engine is 30 to 50 times more powerful than Whittle’s
and a great deal more sophisticated. But Whittle’s original simple system is
now encrusted with subsystem upon subsystem in an enormously complicated
on t He evolu t ion of comPlexi t y [ 149 ]
array of interconnected modules and parts. Modern jet engines have upwards of 22,000 parts.[2]
And so, in this mechanism, the steady pressure of competition causes com-
plexity to increase as functions and modifications are added to a system to
break through limitations, to handle exceptional circumstances, or to adapt to an environment itself more complex. It should be evident to the reader after a little thought that this increase of structural sophistication applies not just to technologies, but also to biological organisms, legal systems, tax codes, scientific theories, and even to successive releases of software programs.
One laboratory for observing real-time structural deepening is John
Holland’s genetic algorithm.5 In the course of searching through a space of
feasible candidate “solutions” usin
g the genetic algorithm, a rough ballpark
solution—in Holland’s jargon, a coarse schema—appears at first. This may
perform only somewhat better than its rivals. But as the search continues,
superior solutions begin to appear. These have deeper structures (finer sub-
schemas) that allow them to refine the original solution, handle exceptional
situations, or overcome some limitation of the original solution. The eventual solution-formulation (or schemata combination) arrived at may be structurally “deep” and complicated. Reversals in structural depth can be observed
in the progress of solutions provided by the genetic algorithm. This hap-
pens when a coarse schema that has dominated for some time and has been
considerably elaborated upon is replaced by a newly “discovered,” improved
coarse schema. The hierarchy of subschemas dependent on the original coarse
schema then collapses. The search for good solutions now begins to concen-
trate upon the new schema, which in its turn begins to be elaborated upon.
This may happen several times in the course of the algorithmic search.
John Koza’s genetic programming algorithm, in which algebraic expres-
sions evolve with the purpose of solving a given mathematical problem, pro-
vides a similar laboratory.8 In Koza’s setup, we typically see the algorithmic parse trees that describe the expressions grow more and more branches as
increasing “depth” becomes built into the currently best-performing algebraic
expression.
In Figure 1 I show the growth of structure as the search for good “solu-
tions” progresses in one of Koza’s examples. As we can see, once again this
mechanism is not unidirectional. Reversals in structural depth and sophistica-
tion occur when new symbolic expressions come along that allow the replace-
ment of ones near the “root base” of the original system. On the whole, depth
increases, but with intermittent reversals into relatively simpler structures
along the way.
[2] Personal communication from Michael Bailey, General Electric Aircraft Engines.
[ 150 ] Complexity and the Economy
180
160
Structural Depth