For a brief moment starting in the 1870s, economics was close to becoming a genuine science of human action. Carl Menger in Vienna, William Stanley Jevons in England, and Léon Walras in Lausanne independently (re)discovered what the Spanish Scholastics had only glimpsed centuries earlier: Value is not inherent in objects, forged in the sweat of labor or dictated by cost; value is subjective, existing and emerging in the minds of acting individuals confronting scarcity—individuals who choose, at the margin, between scarce means and infinite wants.
What began as a breakthrough in understanding subjective value gradually became something else. Through the neoclassical synthesis and the mid-century obsession with formalism, economics drifted into a numbers-first enterprise, suspicious of theory (unless dressed up as mathematics), and intoxicated by measurement. “Let the data speak,” say the econometricians, as if data had a voice of its own: Numbers need interpretation; statistics need a view of human action beneath them. Without such foundation, they tell us little about a world shaped by uncertainty, constant change, and the entrepreneurial imagination.
From the marginal revolution to the credibility revolution, economists gathered ever-larger datasets, ran more intricately-specified regressions, and “discovered” more and more economic relationships. The more sophisticated the empirical work, the less we understood the enterprise embarked on. The economics discipline thus spent decades chasing empirical respectability—“physics envy.”
“Jevons strand of marginalism,” says George Mason University professor Tyler Cowen in a short but provocative new book on the marginal revolution, The Marginal Revolution: Rise and Decline, and the Pending AI Revolution, “contained the seeds of its own destruction.”
It’s well-known that the Victorians were obsessed with measuring things, with names like Galton, Pearson, or Fisher, verifiable legends in the field of statistics. Cowen adequately sums up the intellectual world in which Jevons’s, Walras’s, and Menger’s creations emerged:
…the entire historical period saw both the rise of marginalism but also the rise of quantitative and statistical techniques in economics and the other social sciences. (p. 45)
And Jevons specifically:
If you look at the introduction to The Theory of Political Economy, the first page is spent outlining the marginalist idea. Immediately thereafter, Jevons spends pages arguing for quantification and the unity of the social and mechanistic sciences. For Jevons, The Theory of Political Economy really was a plea for two distinct revolutions, not just what we now call the marginal revolution but a quantitative and statistical revolution as well. (p. 47)
It wasn’t inevitable that the discipline of economics would go this way. Cowen makes a big deal of the quantification and mathematization implied in Jevons’s version of marginalism, at least as opposed to Menger’s strand. In a nice touch of modern irony, Cowen released an AI companion together with The Marginal Revolution, which asks why the economics discipline ultimately followed Jevons and not Menger. I’m told that Jevons’s data obsessiveness and measurements of, for example, the weather, made all the difference: “Menger’s approach didn’t contain averageism, so it couldn’t generate the self-undermining dynamic that makes the Jevons story interesting,” says the LLM.
Perhaps that was obvious in hindsight, but for a generation or so after the marginal revolutionaries had published their writings, the distinction wasn’t so clear. John Maynard Keynes, too—in critiquing marginalist economists—referred to them as “classicals,” and, earlier still, as Cowen shows in the book, the reception of their achievements was lukewarm at best. As late as the 1930s, Mises still considered the differences between the revolutionaries to be matters of terminology and emphasis, not core differences. In Epistemological Problems of Economics, Mises wrote:
Within modern subjectivist economics it has become customary to distinguish several schools.… These three schools of thought differ only in their mode of expressing the same fundamental idea and that they are divided more by their terminology and by peculiarities of presentation than by the substance of their teachings.
It was only later, around the time of Mises’s Nationalökonomie (what later became Human Action) where he properly recognized the differences between Jevons’s and Menger’s marginalisms. Another world war and a Keynesian revolution later, the distinct Austrian flavor of economics was so sidestepped that whatever semblance of mainstream respect the old Misesian and Mengerian methods might have once carried, it was too late. The quantitative, statistical, theory-lite empirical economics of Samuelson and others already ruled.
Cowen spells out what in hindsight seems quite obvious: “LLMs don’t carry ‘theory’; they don’t price theory or construct predictions to test. They let their algorithms ‘build the “theory” for us.” The software spits out a result; you don’t know what it did, can’t tell or comprehend what inputs caused the result to display this way. The computer says no.
Data can describe what happened but only sound theory explains why, and LLMs don’t have any.
AIs excel at pattern recognition, correlation spotting, and out-of-sample prediction from vast datasets; they do not deduce the impossibility of economic calculation under socialism, the coordinative power of market prices under private property, or why fiat results in boom-bust cycles or all manner of unproductive financialization schemes.
“What role,” asks Cowen toward the end of the book, “is intuitive microeconomics supposed to play in such a system? Big data, flexibility of estimation, and out of sample prediction are prioritized, not concordance with what an economist, even a brilliant one, is geared to expect or even able to understand.” Making ourselves obsolete and all of that.
Books reassessing the seminal 19th-century mind-shift in economics we call the Marginal Revolution come out now and again: Mr. Cowen’s is no exception, but the lessons he draws from it for the future of AI and the economics profession specifically are profound. The Marginal Revolution lands just shy of saying that the marginal revolution was a mistake, a dead end.
If AI now outperforms economists at their own statistical game, the profession may have to return to the fork in the road it missed in the 1870s.
