Who Ordered That?

Steven R. Postrel from the February 1996 issue

(Page 2 of 3)

Once we get past the question of how complex systems get started in the first place, and ask how their future evolution occurs, we again face a tension between the historical and natural-law modes of explanation. Historical explanations understand the state of a system primarily by where it has been before, so that accidents of initial conditions or chance events largely determine how things turn out. Natural law explanations seek out underlying forces, present at all times, that shape outcomes. Both modes are usually essential for well-rounded understanding, but the balance between them changes from field to field and sometimes over time in the same field.

Kauffman, with a biology background, sees natural selection as a historical mode of explanation, with random mutations reproduced or not based on their possessors' relative ability to cope with a varying environment. This is not surprising; especially in its modern form, exemplified by Stephen Jay Gould's Wonderful Life, evolutionary theory is relentlessly agnostic about the specific outcomes that are to be expected from natural selection.

The idea of evolution as having tendencies toward "higher" or more complex organisms, for example, is considered, at best a dubious proposition, and at worst a piece of anthropomorphic conceit redolent of religious mysticism and apologetics for social inequality. Even the notion that organisms are "organized" by inherent rather than accidental forces is rejected. As Kauffman notes, "This image fully dominates our current view of life. Chief among the consequences is our conviction that selection is the sole source of order in biology. Without selection, we reason, there could be no order, only chaos."

With this backdrop, At Home in the Universe launches its challenge to the standard dogma, asserting that selection is only part of the story, that "Self-organization may be the precondition of evolvability itself. Only those systems that are able to organize themselves spontaneously may be able to evolve further. How far we have come from a simple picture of selection sifting for fitter variants. Evolution is more subtle and wonderful."

This claim is advanced by a kind of indirect proof: Selection can't possibly do all the marvelous things it would have to to generate the order we see in the world, so what's left must be the result of spontaneous order.

Kauffman argues persuasively based on probabilistic considerations, that as a system's performance depends more and more on the interconnections among its parts, any small change in one of the parts is likely to lead to a catastrophic loss of fitness (the "complexity catastrophe"). As a result, classic, incremental natural selection cannot possibly generate very fit, complex forms with a high number of interconnections compared to the number of parts. He also shows that if there aren't enough interconnections relative to the number of parts and to the rate of mutation, then even if a good solution is found, random mutations will tend to carry the system away from the optimum faster than selection can drag it back (the "error catastrophe").

From a theoretical point of view, then, natural selection can only generate high levels of fitness under a relatively narrow range of circumstances--not too much or too little mutation, too many or too few interconnections. Kauffman suggests that it is spontaneous order that forms the basic "just right" structures from which natural selection is launched; the initial conditions of evolution are not accidental but lawful.

Those ideas may be important in the study of human artifacts and institutions as well as in biology. Kauffman's models show, for example, that the rate of performance improvement for evolving systems slows down as higher performance is achieved. He relates this to the well-known "learning curve" in economics, where the unit cost of production in a plant, firm, or industry declines rapidly at first as production experience is accumulated, but gradually flattens out.

Learning curves and Kauffman's simulations of evolution both display a mathematical form called a power law, where every doubling of cumulative output reduces unit costs by the same percentage. He posits that the similarity in mathematical form is due to a similarity in underlying mechanisms--the gradual exhaustion of opportunities to improve as one reaches higher and higher peaks on a metaphorical "fitness landscape." Kauffman notes that his models of evolution and technological history both show initial progress coming as big changes are made in structure (gasoline versus steam engines in automobiles, for example) while later progress comes from small refinements (power versus manual steering).

Those analogies are suggestive and not implausible, but hardly conclusive; there are lots of processes that might lead to power-law curves, just as lots of different sets of data fit a normal bell curve. For example, the late Allan Newell's Unified Theories of Cognition describes how human skill at simple tasks, such as pushing buttons when certain lights go on (a psychologist's idea of a fun date), follows a power law; Newell then lays out a cognitive theory of learning (called "chunking," to describe the welding together of memories into single units) that generates it. Maybe factory performance follows a power law as practice is accumulated because of the buildup of useful chunks in organizational memory, and not because of diminishing returns to evolutionary search of the problem space. Nevertheless, Kauffman's approach is intriguing because it makes multiple predictions within and across problem domains while its assumptions are quite general.

From the point of view of traditional social science, or indeed of anyone concerned about the role of spontaneous order, competition, and cooperation in regulating human life, At Home in the Universe is most interesting when it attacks the problem of coevolution. Here, Kauffman extends his probability models to multiple populations, each of which constitutes part of the environment of the others, mutating and searching for peaks of fitness. The interesting twist is that when one system mutates, it affects the fitness landscape of all the other systems. Thus, if flies improve their vision and can spot threats at a greater distance, frogs with longer tongues might have an advantage over their less well-endowed cohorts. If one business develops a method of speeding up the delivery of orders to customers, another business may find that it should deliver more slowly (thereby cutting its costs), charge lower prices and attract less time-sensitive customers.

The actual models Kauffman uses are abstract and general, with not a frog tongue or delivery guarantee in sight. He focuses on the question of whether the coevolving populations (or firms, or technologies) will eventually settle down to some kind of steady state, where each population is at an optimum as long as the other ones stay put, or whether they will just keep bouncing around in an endless, chaotic display of one-upsmanship. The results: When the amount of interconnection between the evolving populations is "just right" relative to the internal interconnections among the parts of each system, average fitness for all the populations is highest. Furthermore, in this "just right" zone, the overall system is poised between order and chaos--most of its components most of the time are in a steady state, but every now and then some of them start to evolve in new directions.

Finally, if each population is allowed to evolve its own degree of internal interconnection, holding constant its interconnection to others in the coevolutionary system, each will, as if by an invisible hand, be moved by selection pressure toward the "just right" edge-of-chaos regime. Once again, while the specific details of evolution cannot be predicted, important general features appear to spring up in a lawful manner.

In this case, the models seem to be saying, at least metaphorically, that a freely evolving economy or ecology self-tunes so as to maximize the fitness and survivability of each of its members. I'm not sure just what it would mean in real-world terms for a firm to change the number of internal interconnections it possesses, because I'm not sure exactly what the "parts" of Kauffman's abstract entities should correspond to in an economic context. But if a reasonable interpretation is possible, then the result is of great importance.

Kauffman goes further, investigating how varying degrees of decentralization in a system might affect that system's performance at accomplishing some overall goal. He became interested in this issue after being prodded by management-oriented people in the orbit of the Santa Fe Institute, who wanted theoretical principles to guide the delayering and flattening of organizational hierarchies and the outsourcing and reengineering of work processes.