John Maynard Keynes

In the [YOU] on AI Field Guide

The cycle that began with [YOU] on AI asks what it means to see the machine clearly and to act wisely in its presence. Keynes is the cycle's theorist of decision under genuine uncertainty—the conditions under which the clearest seeing is not enough because the future is not the kind of thing that can be predicted from any available data. He did not counsel paralysis; he was the least paralyzed of men. He counseled knowing which kind of situation you are in, refusing to mistake a confident number for knowledge when the number rests on nothing, and treating the future as a thing to be decided rather than predicted. Those are the disciplines the AI moment most requires and most consistently evades.

His account of radical uncertainty cuts directly at the operating assumption of machine learning: that the future resembles the past in ways that a model trained on the past can exploit. In the domain Keynes called risk this is true and enormously useful. In the domain he called uncertainty—novel situations, structural breaks, events without precedent in the training data—the method has no purchase. The danger is not that the machine knows it is in the second domain; it cannot tell. It will produce a confident-looking probability for something that admits no probability at all, because that is what it was built to do. Keynes identified this failure mode in 1921, in the context of human reasoning; we have now built a technology that commits it at scale.

His macroeconomics supplies the demand-side analysis of AI's distributional consequences. The gains of automation flow, in the first instance, to owners of the technology and capital; if those gains are not redistributed, the economy can face exactly the configuration Keynes most feared: rising productivity, rising output, and falling demand, because the income has concentrated among people and firms that save large fractions of it. The aggregate demand problem is Keynesian in its structure, and Keynes's prescription—that demand does not take care of itself and must be deliberately managed—is the most consequential available framework for thinking about who actually benefits from the abundance AI promises.

Origin

Born in Cambridge in 1883 to an academic family—his father was the economist and logician John Neville Keynes—he was educated at Eton and King's College, Cambridge, where he read mathematics and philosophy before turning to economics. He joined the India Office in 1906 but returned to Cambridge almost immediately, and for the rest of his life he divided himself between economic theory, public service, and personal investment, making and losing large fortunes in the markets he theorized about. His first major book, the Treatise on Probability, appeared in 1921; his account of the economic consequences of the Versailles peace settlement, which warned with prophetic accuracy that the reparations imposed on Germany would breed catastrophe, had appeared in 1919 and made him briefly famous and lastingly controversial.

The work of his maturity was accomplished under two catastrophes. The first was the Great Depression, which produced the General Theory of Employment, Interest and Money in 1936—the work that demolished the classical assumption that markets automatically tend toward full employment and built the macroeconomics that still frames most governments' fiscal policy. The second was the Second World War, which led to his work designing, at Bretton Woods in 1944, the international monetary institutions that would govern the postwar economy. He died in 1946, two years after Bretton Woods, of a heart condition aggravated by the physical demands of the conference negotiations. He did not live to see the postwar boom whose possibility he had worked to create, or to watch his framework be simplified into a doctrine he would have disputed.

He was wrong about important things. His prediction that affluence would produce a fifteen-hour work week was a forecast about human nature as wrong as any ever made by a careful thinker: he assumed wants would saturate once needs were met and work would be freely chosen away. He underestimated the relative character of human desire, the drive for status and comparison that has no ceiling because it is defined against others rather than against any fixed standard of sufficiency. His errors are instructive rather than embarrassing: they are failures of the same kind as his successes, occurring when he forgot his own deepest insight—that human beings are not rational optimizers—and predicted as if they were.

Key Ideas

Technological unemployment. Keynes named the concept precisely: unemployment caused by our discovery of means of economising human labor outrunning the pace at which we find new uses for it. The historical record, until now, has refuted this fear: new uses kept arriving faster than old ones were eliminated, because automating tools left human cognition untouched and displaced workers moved into the higher-value cognitive work the machines could not do. The wager of the present moment is that large language models are automating the cognitive escape route itself. Keynes named the race; he did not predict its outcome.

Radical uncertainty and the limits of calculation. Keynesian uncertainty is not the same as measurable risk. About some situations—the prospect of a European war, the price of copper twenty years hence, the trajectory of a new technology—there is no scientific basis for any calculable probability whatever. A system built to estimate probabilities from data is a system built to speak the language of risk, and it will speak that language even when the situation is one of genuine uncertainty. The danger is the confident decimal for something that admits no probability at all.

Animal spirits. Animal spirits—the spontaneous urge to action rather than inaction that drives investment decisions when rational calculation is impossible—are not a deviation from rational behavior but a substitute for it in the vast domain where rational behavior is undefined. A model trained to optimize expected value cannot capture what happens when expected value cannot be calculated: the leap of confidence or fear that drives human action in exactly the conditions that matter most.

The Keynesian beauty contest. Professional investment is like a competition to pick not the faces you find prettiest but the faces you think others will think others will find prettiest. The Keynesian beauty contest describes a reflexive system in which the act of predicting changes the thing predicted, and in which prices can detach from fundamentals because no one is rewarded for tracking fundamentals. When trading is dominated by algorithms, the beauty contest becomes a contest among models, and the dynamics can produce cascades no participant intended and no fundamental justifies.

The long run and the cost of waiting. In the long run we are all dead—a methodological manifesto, not a bon mot. The adjustment of labor markets, the equalization of AI's distributional effects, the emergence of countervailing institutions: all of these may be true in the long run while being cold comfort to the people living through the transition. The transition is where actual human lives are lived, and Keynes's contribution is to insist that the transition, not the equilibrium, is the real political and moral problem.