Economics, Bounded Rationality and the Cognitive Revolution.

Economists do seem incapable of understanding what makes Herbert Simon tick (not a dead metaphor in this instance, as this book illustrates). The participants at the Cowles Commission didn't quite know what to make of him in its Chicago years; and the questions put to him at a Colloquium held at Turin in 1988 (and reproduced here) evoke nothing so much as the proverbial blind men and the elephant. Probably many of the missed connections have to do with the fact that most economists come equipped gratis with the conviction they have something cogent to say about "rationality" (as Robin Marris notes herein). Luckily, a good third of this book is vintage Simon reprinted, so the reader has a better than average chance to work out the relationship of the life's work of Simon to artificial intelligence and economics for themselves. Those who sum up his work with a few catch-phrases about satisficing and bounded rationality are probably in for a shock. Herbert Simon will go down in history books as one of the founders of the burgeoning field of artificial intelligence. Lack of familiarity with the history of this field is the first major source of economists' misapprehensions.(1) The second obstruction to our understanding is lack of familiarity with the problems of justifying the use of mathematics in the social sciences after the bombshell of Godel's theorem in the 1930s.(2) Very crudely, three responses to the crisis of the legitimacy of mathematics in economics were represented at Cowles in the 1950s: the Koopmans/Debreu version of Bourbakist axiomaticization; Von Neumann's attempt to model the structure of the brain, which encompassed game theory; and Simon's move towards artificial intelligence. Although all were fundamentally methodologically individualist, and invested their hopes in the mathematicization of economics as the vehicle to endow it with scientific status, each displayed extremely divergent characterizations of human rationality. The indiscriminate mixing and matching of these clashing strains of theory in part explains the present Tower of Babel quality of economic discussions of rationality.

Simon was opposed to the neoclassical model from a very early stage of his career, and still is, as documented here. What is less well known is that he was also one of the very earliest critics of von Neumann and Morgenstern's magnum opus. How, then, was he accorded any hearing amongst economists, in contrast to so many other disgruntled souls? Because like them, he wants to build mathematico-mechanical models of human choice, or in his own words, "to provide explanations of behavior that are mechanistic without being physiological". For him, the computer program which mimics the behavior of an actor is literally a system of difference equations which constitutes the theory. For him, when a computer program wins at chess or "proves" some theorems from Russell and Whitehead's Principia, this is tantamount to the computer "thinking." Throughout his career, Simon has taken to making hyperbolic claims for computers, the latest being that machines can themselves "discover" scientific laws and theories, like Kepler's Third Law or Joseph Black's laws of temperature equilibrium.(3) What this means in practice is programming about 97% of the research process into a computer, and letting it iterate in on the "final" result, rather like the Borges' story of Pierre Menard wanting to rewrite the Quixote on his own.(4) Such exaggerated claims are endemic to AI; but it is also important to note that Simon's is not the only approach to intelligence in the AI community.

So what has this all to do with economics? Plenty. Simon wants to produce a simulacrum of science without society, just as many economists want to produce a simulacrum of markets without social context. The problem for Simon, though, is that his program suggests that a computer which simulates an actual market is good enough as a theory of society; and indeed, this was the limited aim of the abortive "behavioral theory of the firm" which also emanated from Carnegie-Mellon in the 1960s. For reasons which neither Simon nor his critics have yet to confront, no one was interested in programs which could mimic the behavior of an individual firm (but would be different for each firm), while many remain fascinated with the idea of a computer which could mimic the behavior of an individual human being, a la the Turing Test. Clearly something else is going on here beyond the outdated pre-Kuhnian notions of "science" which pervade Simon's methodological pronouncements. In response, Simon has withdrawn to a Department of Psychology, renouncing the opportunity afforded by his Nobel to make anything other than vague critical statements concerning economics. Enthusiasts of "bounded rationality" should be given some pause by these events.

Since Simon waives away objections based on Godel's Theorems, perhaps he would someday answer the following question: Why doesn't he construct one of his "scientific discovery" programs for uncovering a "law" in economics? It would be interesting to see which law he would choose.

Philip Mirowski University of Notre Dame

1. A very good non-technical introduction is Daniel Crevier, AI, New York: Basic Books, 1993.

2. See Roy Weintraub, "Formalist Mathematics and Economic Theory" paper presented to Veblen Society, Sept. 1993; Philip Mirowski, "Mathematical Epistemology at Mid-Century", History of Economics Review, Summer 1993.

3. See Pat Langley, Herbert Simon et al, Scientific Discovery, Cambridge: MIT Press, 1987.

4. Simon's article in this book "Scientific Discovery as Problem Solving" was also printed in the International Studies in the Philosophy of Science 1992 (6):1; only there it is followed by 13 critiques by distinguished philosophers, historians and cognitive scientists, with a response by Simon. The contrast to this book is enlightening.