Machine Dreams: Economics Becomes a Cyborg Science

The author, Mirowski, is interviewed repeatedly in the great documentary film, "The Trap: What Happened to our Dream of Freedom?" This BBC documentary was broadcast in 3 episodes in 2007. The British filmmaker Adam Curtis is perhaps more famous for other documentaries including _The Century of the Self_ and more recently _The Power of Nightmares_ about the war on terrorism. It is available on the Internet in reduced or compressed formats. As Curtis clearly articulates the main thesis, _The Trap_ is about "how a simplistic model of human beings as self-seeking, almost robotic, creatures led to today's idea of freedom." That is, Mirowski is not only interviewed, but this particular book, _Machine Dreams_ is the key text informing much of Curtis's documentary. I'm not an economist, but I found the film so informative and hair-raising that I purchased Mirowski's book. The book equals the film in brilliance, though it doesn't reach quite as far as Curtis's more political argument about how the economic models got adopted into political programs, and hence led to the trap suggested in his title. Hard to say whether the film helps to comprehend the book more or vice-versa the book helps to deepen the film. I'd say both.

As you can see from the previous reviews, this is a book that provokes strong feelings. As usual that's often more a reflection on the reader than on the book. You'd never guess that this book is (by intention) very funny. But it is. Mirowski has written elsewhere that John von Neumann is the "hero" of this book. Von Neumann thought neoclassical economics was nonsense, and made no secret of that opinion. As a result, many post-war American economists have tried to write him out of history. One fruit of their effort was the beatification of John Nash as the patron saint of game theory, a process that began in the 1980s. According to this book, the irony is that those same economists have "followed the trajectory" of von Neumann's thinking for the last five decades, even if they wouldn't acknowledge it. Through the 1970s or so they relied on fixed-point theorems and other nonconstructive proof techniques (von Neumann in the 1930s). From the 1980s to now, they have relied on game theory (von Neumann for a few years in the 1940s). Recently, they have begun to rely more on computers, particularly to study "agent"-type automata (von Neumann from the mid-1940s to the end of his life). And, as for von Neumann, military funding has been an important factor throughout this development. Actually, this isn't "the" irony, but just one of many. If you've ever had any suspicions that neoclassical economics was kind of a crock, you'll find them well-supported in this impressively well-researched book. (Some highlights include the misplaced aspiration to axiomatize economic theory, the impossibility of computing Nash equilibria, ditto for Walrasian general equilibria, the socialist antecedents of "free market" jingoism, the bounded usefulness of V. Smith's market experiments, and much else.) It may be a bit of a stretch to say that the book reads like a thriller, but the fun of uncovering some additional bit of intellectual dishonesty with each turn of the page did keep my attention. For over 500 pages, this story is told with a sustained, righteous and gleeful sarcasm. Such a tone may sound tiresome, but based on the evidence Mirowski brings forward - much of it the neoclassicals' own words - it struck me as quite justifiable. And I laughed a lot. However, be aware that this book is less self-contained than Mirowski's earlier book, "More Heat Than Light". Even if you've read that book first (which I recommend, especially if you're not an economist), you should have at least a Scientific American-level of acquaintance with theory of computation, a bit more math-intensive experience with game theory (like a few chapters of Myerson -- not that "Machine Dreams" has any equations, but the math is often alluded to), and smidgens of Arrow, Debreu, Herb Simon, Vern Smith and Kahneman & Tversky. You should also know "who" Bourbaki is and have some experience of the Bourbakist style, because it's taken for granted that you already do. There are a few quirks, bu

The suggestion made in the last chapter is to try to identify an automaton that describes a particular market. This program will not work because of lack of uniqueness, as is explained by the work on generating partitions in nonlinear dynamics. Given any sttistical distribution, one can find infinitely-many different automata that can be programmed to generate that distribution. Mirowski's suggestion cannot be carried out in any meaningful sense for that reason. In finance theory we have recently (with Gunaratne) deduced a particular stochastic dynamics from market histograms, and there we also have faced nonuniqueness in identifying the underlying dynamics. The bigger and more immediate problem is to find nonfinancial economic data that are accurate enough to draw any meaningful conclusion from the purely empirical histograms.Now for the irritation. I find it academically irresponsible in this day and age to equate Newtonian mechanics with 'equilibrium'. From the beginning, Newtonian mechanics was about periodic and quasiperiodic orbits. The orbits that were studied prior to 1900 typically have neutral equilibria. To be 'in equilibrium' in such a case, the earth (for example) would have to sit at the center of the sun. Poincare' discovered chaos in Hamiltonian systems around 1900. In a chaotic system all equilibria are unstable but the orbits are bounded. See Ivars Peterson's 'Newton's Clock' for a description of the history of the discovery of chaos in the solar system. Toffoli and Fredkin discovered Turing machine-level complexity in a Newtonian system (constructed of billiard balls) around 1983, and Chris Moore (now at the Santa Fe Institite) showed around 1993 that certain area preserving maps are equivalent to Turing machines. In other words, Newtonian systems can exhibit not merely chaos but maximum complexity as well. The misidentification of Newtonian mechanics with 'equilibrium' or simple mechanics should now be laid to rest once and for all. It would be more accurate to say that the economists borrowed the idea of static equilibrium from Archimedes. Also, take note please that every digital computer is a Newtonian electromechanical system.

I've read about 250 pages and can recommend that anyone with an interest in economics and finance should read this fantastic book. The basis for the text are the contributions of Shannon, Turing, von Neumann, Wiener, Koopmans, Marshak, and Arrow. Mirowski tells us the main story of the interaction of the Cowles Commission with RAND, which Bernstein does not at all hint at in his Capital Ideas. Having praised the book, I will now concentrate mainly on a few points of disagreement. Undecidability should not be confused with noise in stochastic processes. Systems at the transition to chaos can define automata that can perform simple arithmetic. That 'cyborg' has it's origin in the physical sciences seems farfetched (the connection between Turing and physics is supposed to be via Maxwell's demon, but was Turing really motivated by the idea of Maxwell's demon?). Nonlinear dynamics and fractals ('chaos' and fractals) certainly did not evolve from cybernetics or 'system theory' ('system theory' was based at best on an awareness of equilibria and limit cycles of differential equations, and made vague, unjustifiable allusions to holism). Cybernetics cannot really be seen as the midwife of what is now loosely called 'complexity' either, rather, that (still undefined) field grew out of nonlinear dynamics, neural networks, computability theory and molecular biology. Mirowski is right that many scientists confuse simulations with experiment and observations. I have argued against this confusion in papers and books.Mirowski paints an intriguing picture of (Gödel-influenced) von Neumann, RAND, researchers with awareness of information and computability limitations leading to agent-based modelling with some respect for empiricism on the one hand, and then, on the other hand, Arrow, the Cowles Commission and their later rejection of empirics, instead with emphasis on Bourbaki-style existence proofs leading to infinte demands on information requirements on Walrasian agents and noncomputable equilibria. We now know that agent-based modelling can easily lead to fat-tailed price distributions (as observed empirically), whereas in contrast the origin of the systematic head-in the-sand philosophy of the neo-classical economic theorists is made quite clear in this work. One can summarize the neo-clasical economic agent as follows: his dynamics are trivial (equilibrium, including Nash equilibria) but the information demands made on him to interact with other agents and locate an equilibrium point are impossible (noncomputable). Moreover, we now know that financial market statistics point toward the instability of Adam Smith's hand, so that the notion of dynamic equilibrium is complelety uninteresting so far as understanding markets is concerned.