Summer 2002; volume 8, Number 2
Games Economists Play
Machine Dreams: Economics Becomes a Cyborg Science by Philip Mirowski (Cambridge University Press, 2002; Xiv + 655 pages)
Robert Nelson tells us in Economics as Religion that modern economics is a branch of theology.1 In a book that shows his immense learning, Philip Mirowski presents an altogether different story of post-World War II economics. As he sees matters, economics faced a crisis at the end of the 1930s. The standard model of economic theory, based heavily on nineteenth-century science, encountered problems it could not solve.
Developments in military research during World War II offered a way out of the crisis. In particular, John von Neumann, a mathematician of surpassing intellect, wished to reconstruct economics on new foundations. At first, he invented a new branch of mathematics, game theory, as a tool with which to accomplish this task. But he quickly moved on to another project.
In his new view, economics was to be subsumed in a general science that investigated the nature of artificial intelligence. The new theory of automata blurred the distinction between human and machine operations—thus the “cyborg science” of the book’s title. (A cyborg has both human and machine properties; sometimes, Mirowski uses the term to describe scientists who study these devices.)
Unfortunately, economists proved unequal to the task von Neumann had set them. Mirowski concentrates his attention in this regard on the Cowles Commission, an influential group of economists that benefited heavily from military funding, and on the RAND Corporation, for which many noted economists worked. Instead of developing a new science along the lines von Neumann suggested, the economists in these two groups attempted to salvage as much as they could of the old general equilibrium model.
Two factors especially limited their progress. Their version of game theory stressed the importance of Nash Equilibrium, a concept that von Neumann dismissed as having little importance. To Mirowski, the progenitor of this concept, the brilliant but mentally unstable mathematician John Nash, is the antihero who has misdirected modern economics. Further, the Cowles Commission and its allies failed adequately to use cyborg science, owing to their bewitchment by the fata morgana of equilibrium.
Not content with being a historian, our author also dons the prophet’s mantle. He sketches a new way of practicing economics that he regards as more responsive to the cyborg revolution. Those inclined to an Austrian perspective will see many things differently from Mirowski; nevertheless, his vast tome has much to teach us.
As already noted, Mirowski sees much of modern economics as dominated by the general equilibrium construct. The elaborate attempt to work out the mathematical details of equilibrium was not simply an attempt to carry out the program of Léon Walras. Instead, a key purpose was to counter Mises’s calculation argument.
Mises maintained that a developed modern economy could not function without economic calculation; and only the price system of capitalism can fulfill this task. If he was right, the entire socialist project lay in ruins: hence the imperative necessity for true believers in socialism to answer him. A popular solution endeavored to enlist the equations of general equilibrium in support of socialism. If socialist planners could solve these equations, Mises had not destroyed socialism—so at any rate it was claimed. To make good their contention, these socialists needed to develop an exact account of equilibrium. “It has yet to be fully appreciated that many of the key figures in the history of the postwar Cowles Commission cut their eyeteeth on this [calculation] controversy. Jacob Marschak’s first published article (1923) was a response to the Mises broadside. Leonard Hurwicz, who joined Cowles in 1942, subsequently constructed a separate interpretation of the controversy” (p. 233).
For some of the Cowles group, the issue between capitalism and socialism became entirely practical: which could come closer to realizing the requirements of equilibrium, a state of perfect efficiency? The distinguished Dutch economist Tjalling Koopmans, later a Nobel laureate, “helped innovate Cowles’s explicit novel response to Hayek, that the Walrasian model actually illustrated the ‘economy of information’ that existed in a perfect market or in decentralized market socialism” (p. 259).
According to Mirowski, an even more famous economist found himself gripped by the need to respond to Mises and Hayek. Kenneth Arrow’s impossibility theorem “was also a direct product of Cowles’s participation in the socialist calculation controversy, although few have seen fit to situate it within that context” (p. 302).
Mises and Hayek said that a socialist economy could not function. Arrow turned the tables on them by claiming that a democratic system of majority rule could not generate a consistent set of social preferences. Only “dictatorial or imposed regimes” could achieve complete logical consistency. “For anyone steeped in the socialist calculation controversies of the 1930s, it is hard to see it [Arrow’s theorem] as anything other than a reprise of the Cowles theme that the Walrasian market is a computer sans commitment to any computational architecture or algorithmic specification; the novel departure came with the assertion that democratic voting is an inferior type of computer for calculating the welfare optima already putatively identified by the Walrasian computer” (pp. 303–04).
I am not sure this account of Arrow will stand up: does a problem with democratic voting show a weakness in the free market? But Mirowski has established his main point: much of the elaboration of general equilibrium by the Cowles economists aimed to reply to Mises. Why, though, should this lead to a crisis in economic theory?
The answer lies ready to hand. These models have no relevance to the real world: they give us an economics “that never was, on sea or land.” Our author might have noted that Mises knew this perfectly well. He found it “a serious mistake to believe that the state of equilibrium could be computed, by means of mathematical operations, on the basis of the knowledge of conditions in a nonequilibrium state. It was no less erroneous to believe that such a knowledge of the conditions under a hypothetical state of equilibrium could be of any use to acting man in the search for the best possible solution of the problems with which he is faced in his daily choices and activities” (Human Action, pp. 710–11).
Of course the Cowles economists did not admit that they had failed to answer Mises’s challenge to socialism. But, even from their own point of view, their models proved unsatisfactory. They were not praxeologists in the style of Mises, who could claim that their results are known a priori to apply to the real world. Rather, they sought empirical validation for their models, but this was not in the offing. Mirowski stresses in this regard the failed attempts of Harold Hotelling and Henry Schultz to derive demand curves on Walrasian principles. “Schultz wrote up the results of his decade-long search . . . in his 1938 book Theory and Measurement of Demand. . . . Schultz bravely reported the empirical debacle in detail, and then produced a litany of excuses why things had not worked out as hoped” (p. 195).
Here then lay the problem for general equilibrium economics as it stood at the onset of World War II. In a vain attempt to refute Mises, the Cowles economists attempted to elaborate detailed equilibrium models. But these proved useless for understanding the world. What, then, was to be done?
Once more, Mises offered a way out. The proper method of economics neither admits of, nor requires, empirical validation: “[Schultz’s results] are, at best, rather questionable and unsatisfactory contributions to various chapters of economic history. They are certainly not steps toward the realization of the confused and contradictory program of quantitative economics. . . . There is in the field of human action no means of dealing with future events other than that provided by understanding” (Human Action, p. 349).
The Cowles group, along with their RAND confreres, chose a different way to respond to the crisis. Perhaps economics, as they practiced it, had been too narrowly conceived; the key to progress was to embed economics within a general science of information.
Our question recurs: How was this to be done? Once more, there was an Austrian answer, this time provided by Hayek. “The need to refute ‘market socialists’ such as Lange thus led directly to the initial landmark revision of the image of market functioning away from static allocation and toward information processing. The clearest statement of this momentous shift can be found in Hayek’s ‘Use of Knowledge in Society’ ” (p. 238).
Once more Mirowski’s economists spurned the Austrian approach: it was not “scientific” enough to suit them. Hayek presumed to write about economics without using high-powered mathematics. “By his own admission, Hayek was incapable of appreciating von Neumann’s mathematical enthusiasms. He may also have been oblivious of the withering skepticism trained in his direction by his former colleague [Oskar] Morgenstern. . . . The Morgenstern diaries are replete with disparaging remarks about Hayek” (p. 238).2
The Cowles economists wanted a general science of information; and, fortunately for them, a mathematical genius offered the tools to proceed. John von Neumann first devised game theory as a new approach to the study of rational action. Not content with this, he shifted his attention to an even more sweeping study: “I [Mirowski] believe the best way of making sense of the evidence from von Neumann’s last decade is to regard game theory as being progressively displaced as a general theory of rationality by the theory of automata” (p. 146).
Such economists as Arrow, Debreu, and Koopmans did not make full use of the opportunity von Neumann’s innovations provided them. Instead, blinded by their unwillingness to cast general equilibrium aside, they inconsistently attempted to combine cyborg science with their old models.
Here the book’s archfiend, John Nash, enters to exert his malign influence. His famous Nash Equilibrium took game theory away from the cooperative solutions that interested von Neumann toward “terminal paranoia” (p. 340). In Nash equilibrium, no player can benefit from a change in strategy, so long as everyone else’s strategy remains unchanged. By a process of reasoning I am unable to fathom, Mirowski considers this idea an expression of Nash’s well-known mental illness (or weirdness, if you take the Szaszian view that mental illness is a myth). Again in a way difficult to understand, our author regards Nash equilibrium as somehow inconsistent with “cyborg science.”
In the midst of his crude psychologistic reductionism directed at this veritable Lucifer, Mirowski makes a valuable historical point. “John von Neumann himself explicitly rejected the Nash equilibrium as a ‘reasonable’ solution concept” (p. 334, emphasis removed). Nevertheless, because Nash’s views fit the strategic war-gaming practiced at RAND, they prevailed. Later exponents of game theory, including Robert Aumann and Kenneth Binmore, have papered over the breach between von Neumann and Nash, in the process virtually expunging von Neumann from the history of game theory.
However gifted such figures as Arrow and Debreu, these economists proved too tied to old ways of thinking. Though they saw the Promised Land of a true machine science, they turned back. If they had been willing to follow von Neumann, they would have demoted game theory to a minor place and devoted full attention to cyborgs.
I cannot think that Mirowski’s bold thesis succeeds. Von Neumann, as our author rightly claims, endeavored to investigate the general properties of automata. But it does not follow from this undoubted fact that he wished to replace economics with this study. Mirowski offers no evidence that von Neumann favored such a replacement, much less that he counseled economists to abandon their discipline for “cyborg science.”
I have delayed long enough; I must here confess abject failure. Despite Mirowski’s best efforts, I have been unable to grasp why the study of cyborgs is supposed to revolutionize economics. The idea does not seem altogether useless; but the relevance of the new discipline is hard to grasp. When we learn, for example, that Alain Lewis has proved that under certain conditions, a Walrasian general equilibrium cannot be programmable on a Turing Machine, how much does this damage the Walrasian system? Is it essential to the Walrasian view that it be thus programmable?
Toward the end of the book, one sees better Mirowski’s aim. “[T]he logical apotheosis of all the various cyborg incursions into economics recounted in this book resides in a formal institutional economics that portrays markets as evolving computational entities” (p. 539). If you think this, then the conditions of equilibrium, and everything else for that matter, had better be programmable. But as to the reasons for adopting the cyborg view, we are left in darkness. Mirowski sees cyborgs as the wave of the future. In his view, economists must ape the progressive science of the day. Would it not be better for them to study human action? And how better to do so than to study Human Action?
1See my review in The Mises Review, vol. 8, no.1, pp. 6 ff.