Wed Jul 17, 2002

Philip Mirowski's Machine Dreams

I spent the weekend and most of Monday reading Philip Mirowski’s Machine Dreams: Economics Becomes a Cyborg Science. It’s a fascinating—though often exasperating—partial history of modern economics. I’ll be very interested to see how (and if) it plays to its different audiences, especially within the Economics profession.

Machine Dreams is 650 pages long. This is partly because of Mirowski’s prose style, which oscillates between the pithy and—- to use one of the author’s favorite words—the perfervid. It is partly because the book incorporates large chunks of archival material (especially correspondence) into the text in order to make its argument. But it is also partly because that argument is complex, and resists compression.

Economics and the Military Mirowski’s argument proceeds in more than one register. The ground bass of the book is that the course taken by post-war economic theorists in the United States was deeply influenced by the military infrastructure of the state that was built up from the 1940s onwards. Many of the preoccupations and much of the theoretical machinery that now dominates economics can be traced to military research carried out in cold war think tanks, especially the RAND organization. Mirowski’s case isn’t a conspiracy theory. He’s not arguing that the economists cooked it all up on purpose and in secret, or that they were entirely in the pockets of the military, or that there were no conflicts. In fact, much of the length of the book comes from tracing the complex development of ideas within and between RAND, the Cowles Commission, MIT’s Rad Lab, and the economics departments at Chicago, MIT and Yale.

Mirowski does persuade the reader both of how deeply implicated in military-funded research most top-flight economists were on the one hand, and the degree to which that connection was effaced or erased in the main academic journals on the other. Often, work that had its origins in some military appropriations, budgeting or strategy problem would morph into a mainstream paper that made no mention of those origins. And of course the substance of modern economic doctrine on the role of the state and the power of the free market seems a world away from the world of military “Command, Control, Communications and Information”. The story of how state-financed economists, largely researching problems of hierarchical organization in military settings, came to kill and eat their mother (as it were—and then forget they ever had a mother) is fascinating.

Someone who knows a lot more about this stuff than me suggested to me that Michael Bernstein’s recent book A Perilous Progress covers much the same ground, perhaps more thoroughly and with less flamboyance than Mirowski. It’s on my reading list. Another book that’s relevant here (and which I also need to read) is Paul Edwards’ The Closed World: Computers and the Politics of Discourse in Cold War America.

What’s a Cyborg Science? The second main theme of Machine Dreams adds a layer of complexity to the argument about the military origins of modern economic theory. The central trope here (borrowed from the science studies literature) is the cyborg, and the cyborg sciences that study them. The term “cyborg”—a contraction of “cybernetic organism”, as you probably know—was originally coined to describe any kind of organism that incorporated chemical, mechanical or especially computer enhancements to better deal with (i.e., control) its environment. More generally, it refers to the meshing of the natural and the artificial. Cyborgs are bound up with Cybernetics, the information-based “general systems theory” that came into vogue in the 1950s through the writings of Norbert Weiner, Ludwig von Bertalanffy and others. The central concept of cybernetics is system maintenance through feedback mechanisms. The central metaphor of cybernetics is the computer and computation generally. The “cyborg sciences”—computer science, AI, operations research, automata theory, sociobiology and game theory, amongst others—all share this central concern with entity/environment interaction via feedback mechanisms. Mirowski lists the core elements to the cyborg ideal:

“[T]he existence of the computer as a paradigm object for everything from metaphors to assistance in research activities to embodiment of research products” (13). A blurred distinction “between the Natural and the Social, the Human and the Inhuman”. A related blurring of boundaries between reality and its simulation. A paradigm case is the way that game theory and wargames gradually moved from taking place outside the computer to happening wholly inside it. First people played games against each other; then they played games or simulations mediated by the computer; and eventually the players themselves were automated programs. “The fourth landmark of the cyborg sciences is their heritage of distinctive notions of order and disorder rooted in the tradition of physical thermodynamics … Whether it be the description of information using the template of entropy, or the description of life as the countermanding of the tendency to entropic degradation, or the understanding of the imposition of order as either threatened or promoted by noise … the cyborg sciences make ample use of the formalisms of phenomenological thermodynamics as a resevoir of inspiration” (16). Fifth, and perhaps most important, is that “terms such as ‘information,’ ‘memory,’ and ‘computation’ become for the first time physical concepts, to be used in explanation in the natural sciences” (16).

What to do with Cyborgs Mirowski argues that economics was simultaneously attracted and repelled by the cyborg ideal, and that the period from 1940-1990 is characterized by its various complicated efforts to incorporate cyborg themes of information-processing, communication, computability and simulation into its existing Walrasian general-equilibrium view of the world. Mirowski thinks this effort has largely been a failure, or at least has left large lacunae or inconsistencies at the heart of microeconomic theory. Some of these are very deep, such as Mirowski’s contention that the ideal of the rational economic actor as information processor and “utility computer” runs afoul of fundamental limits on computability (370-437). Others are more like ironies of greater and lesser degrees of bitterness. Small ones crackle by on every other page, like a passing ad hominem about “the potential embarrassment of an award of a Nobel for the formalization of rationality to a mentally ill individual” (333). Large ones evolve over whole chapters. Mirowski’s discussion of the relationship between economic theory and the concept of the Self is very good. Economics is all about the inviolable Self and its perfectly rational choices. Yet the “age of methodological cyborgism” (443) has undermined the Self many times over, and as it has imported cyborg ideas economics has joined in this process:

If a long-overdue calm reassessment of the Culture Wars should ever materialize, we would eventually come to realize that it was not those wily postmodernists who wrought the most havoc … with their “decentering of the self” and their fragmentations of the body. Rather, postmodernism was itself an effluvium of the iintellectual innovations in the natural sciences … Wherever the computer has cast its allure, there be cyborgs. Indeed, some of the most poignant evidence of this trend is the parade of intellectuals—- Daniel Dennett here again springs to mind, bu also sociobiologists like John Maynard Smith, or Deirdre McCloskey or Kenneth Arrow—- setting themselves up as defenders of Old Time Religion, all the while sowing the seeds of postmodern fragmentation in their wake. (448)

Still other ironies take most of the book to flower. The main one is a reassessment of the role of John von Neumann in economics. Insofar as it was relevant to economics, Mirowski argues, his thinking moved through three phases. The middle phase produced his model of the expanding economy and the Theory of Games and Economic Behavior. The final phase was the cyborg one—von Neumann’s theory of automata and his work on the modern digital computer marked a new phase of his thinking. Thus,

where game theory tended to suppress formal treatment of communication and the role of information, the thoery of automata elevates them to pride of place. The very idea of a roster of possible moves in the game is transmuted into an enumeration of machine states. Strategies, a foundational concept of game theory, become internally stored programs. However, under the tutelage of Godel and Turing, exhaustive prior enumeration of strategies … is abandoned as implausible … Game-theoretic notions of struggle and interdependence soldier on in the theory of automata, but now they are rendered subordinate to the larger project of a computational approach to interaction (147-9)

This program, Mirowski argues, was not followed up on by economists. Instead, they went to some lengths to write von Neumann’s contributions out of the picture (in favor of John Nash’s) (406-415). The irony, he thinks, is that von Neumann was on the right track all along, that microeconomics has wrapped itself into hopeless knots trying to absorb the cyborg program without fundamentally changing itself, and that “von Neumann’s vision for a computational economics” (536) is now returning, perhaps, to save the field:

The problem facing economists seeking to come to grips with the theory of computation has been to work out the relationship of the doctrine to a viable economic theory … [T]hey have been oblivious to the possibility that John von Neumann did not anticipate that his theory be subordinated to explication of the psychological capacities of the rational economic agent, or even to be appended to Nash noncooperative game theory to shore up the salience of the solution concept… As we have argued throughout this volume, von Neumann consistently maintained that his theory of automata shoudl be deployed to assist in the explanation of social institutions … [T]he appropriate way to round out von Neumann’s vision for economics is to construe markets (and, at least provisionally, not memes, not brains, not conventions, not technologies, not firms and not states) as formal automata. In other words, the 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 (537-9).

But What Does it all Mean? Machine Dreams is controversial reinterpretation of the history of post-war American economics, blended with a trenchant critique of formal microeconomic theory, wrapped up in a wide-ranging narrative rich in detail and (it seems to me) filled with score-settling asides, that finally culminates in the triumphant return of the repressed and a theoretical program for the future. Oh, and all presented in a prose style that constantly threatens to spin out of control. The book gets inside your head, but it’s hard to know how to assess it. (I wouldn’t normally write a sentence like the one that opens this paragraph, but Machine Dreams has obviously temporarily infected my writing.)

Mirowski seems on solid ground when writing about the institutional underpinnings of economics research after World War II, though as I mentioned above this may be the least original contribution of the book. The recentering of cybernetics and its offspring is also convincing, as is MIrowski’s insistence of the importance of cognate disciplines like Operations Research to the course of the Economics profession. As for his arguments about the substance of the theories: well, unfortunately it’s been a long time since I took micro. I need a refresher course, or at least a bit more reading (mining the book’s bibliography would be a useful exercise). So I leave that question to people more qualified than me.

Much the same applies to Mirowski’s advocacy of the Computational Economics program, which I hadn’t heard of prior to reading the book. One thing I was intrigued by, however, was his suggestion, as part of this program, that economics needs “to entertain the gestalt switch from a single omniopotent market to a plurailty of markets of varying capacities and complexities” (540). He thinks “the market automata approach does provide the wherewithal” to develop a taxonomy of kinds of market (540). There are suggestive parallels here (in goal, though probably not method) with recent work in Economic Sociology, and particularly with Harrison White’s Markets from Networks. White doesn’t adopt the automata picture, of course, but he does push the vision of a map or taxonomy of possible markets. Markets from Networks is yet another book that I need to digest more fully. Stay tuned.

Overall, I’m not sure what to make of the book. If you can stand the style, it’s an absorbing read. But because I’m not as sure-footed as I’d like to be with the theory that the book wants to dismantle, I’m unwilling to buy into Mirowski’s view without reading a bit more. So it’s off to the bookstore, I think.