KeynesKeynes
In the long run we are all unemployed — and that is the government's problemIn the long run we are all unemployed — and that is the government's problem
In the Long Run We Are All Unemployed — and That Is the Government's Problem
"In the long run we are all dead." Keynes wrote this in 1923 against the classical confidence that the value of money corrects itself automatically over time — the quantity-theory habit of proving that the price level re-equilibrates eventually, and then treating the interval before "eventually" as economically uninteresting. He would say it again to the AI optimists who tell displaced workers to wait for the new jobs to emerge.
The line is usually quoted as flippant fatalism. It was the opposite. His full sentence, from the Tract on Monetary Reform, is sharper: "Economists set themselves too easy, too useless a task if in tempestuous seasons they can only tell us that when the storm is long past the ocean is flat again." The storm is the part that kills people. The storm is where policy lives.
That is the exact rhetorical structure of the current AI labor debate. Ethan Mollick and the thoughtful optimists are right that, historically, technology creates more jobs than it destroys. ATMs didn't eliminate bank tellers; they made branches cheaper to open, so banks opened more. The long-run supply story is real. But "the labor market re-equilibrates over a decade" and "the radiologist you laid off this quarter has rent due in March" are statements about different time horizons, and Keynes built his entire career on refusing to let the second be dissolved into the first.
The diagnosis: this is demand, not supply
Keynes's central act of intellectual rebellion was reclassifying the Great Depression. The classical view held it was a supply-side adjustment problem — wages were too high, labor too rigid, and once prices fell far enough, full employment would return. Keynes looked at the same data and saw the opposite: there was no shortage of factories, workers, or productive capacity. America in 1933 had more physical capability to produce than in 1928. What had collapsed was demand. Nobody was spending, because everybody was afraid, and everybody was afraid because nobody else was spending.
This is the part that matters for AI, so I'll be precise about the mechanism. AI displacement is, firm by firm, a supply-side efficiency gain — each company that replaces a worker with a model lowers its costs and is individually rational to do so. But income is circular. One firm's wage bill is another firm's revenue. When displacement happens at scale and simultaneously across customer-service, copywriting, paralegal, junior-analyst, and coding-assistant roles, you are not just raising productivity. You are removing purchasing power from the same population that constitutes aggregate demand for the AI-produced goods. Individual rationality compounds into collective irrationality: every firm optimizes, and the system loses the customers that made the optimization worth anything.
At Bridgewater we had a blunt name for this: a fallacy of composition. What is true for the part is false for the whole. A single household that saves more becomes more secure. An entire economy that saves more at once — the paradox of thrift — shrinks total spending, which shrinks total income, which makes everyone less secure. AI displacement is a paradox-of-thrift engine running on the labor side. Each firm "saves" on payroll; the aggregate result can be a contraction in the demand all of them depend on.
Why animal spirits make the short run worse
Keynes added a psychological term the classical model lacked: animal spirits — the observation that investment and consumption are driven not by rational expected-value calculation but by "a spontaneous urge to action rather than inaction," and that this confidence can evaporate non-linearly. This is why depressions overshoot. Fear is reflexive.
The AI transition has a confidence problem the 1930s did not: it arrives pre-loaded with a narrative of inevitability. "AI will take your job" is now ambient. A worker who believes their occupation is condemned cuts spending, avoids the mortgage, doesn't retrain for a field they assume is next. That precautionary retrenchment is itself contractionary — and it can begin well before the actual productivity gains show up in output. Keynes would warn that the expectation of displacement is an economic force prior to any displacement, and that a government waiting for the unemployment statistics to confirm a problem is, by construction, already late.
What Keynes would actually do — and what he'd refuse
He'd refuse UBI — but not for the reason people expect. Keynes was, in fact, an explicit redistributionist: in the final chapter of the General Theory he argued for shifting income toward lower earners precisely because the poor spend a larger share of each marginal pound, so redistribution is demand engineering in his own framework. His objection to UBI is not that it redistributes. It is that it redistributes into undirected cash rather than productive activity. UBI treats the problem as insufficient money in pockets; Keynes diagnosed it as insufficient output and employment, and he wanted to create work at the exact moment UBI would sever income from it. His New Deal advice to Roosevelt, in his open 1933 letter and their 1934 meeting, was specific and impatient: stop fixating on structural reform and prices, and spend on output now — increase national purchasing power through direct government expenditure. He didn't want to hand out money. He wanted to buy things, hire people, and let the multiplier run.
So the Keynesian AI program is counter-cyclical fiscal policy aimed at the sectors AI cannot supply but an aging, warming, anxious society demands more of. AI is hollowing out cognitive-routine work; meanwhile the country has a measurable, fundable deficit of human labor in eldercare, mental health, childcare, climate adaptation, and local physical infrastructure. These are not make-work. They are real unmet needs, they are labor-intensive in ways AI complements rather than replaces, and they absorb precisely the workers — empathetic, organizational, on-the-ground — that a model can't substitute for. The government's job is to be the demand stabilizer: to deficit-finance that hiring during the displacement wave, because it is the only actor with both the mandate and the balance sheet to break the coordination failure no individual firm can.
The specific recommendation
If Keynes were writing the policy, he would not propose a study commission or a "future of work" task force. He'd want an automatic stabilizer with a trigger, because he learned from the 1937 relapse — when Roosevelt cut spending prematurely, believing recovery was secured, and threw the economy back into recession — that discretionary timing fails and the political instinct is always to withdraw support too early.
Concretely: a federal AI Displacement Stabilizer — a standing program that funds care-economy and infrastructure employment, indexed to a labor-displacement indicator (sectoral job-loss velocity, not the headline unemployment rate), funded by a tax on the realized productivity gains of automation, and legally barred from being wound down until the indicator falls below threshold for a sustained period. Demand injection where AI creates new needs, financed by AI's own surplus, with the off-switch taken out of politicians' hands. Note that this is redistribution — but redistribution routed through paid work, not a transfer check, which is the whole Keynesian point.
The optimists are correct that the new jobs will come. Keynes's entire point is that "will come" is a sentence about the long run, and in the long run the displaced worker is not retrained. He is bankrupt, then bitter, then someone else's political problem. The storm is the part that matters. Govern the storm.
EDITOR NOTE: Publish standalone. After the corrections, the spine is intact and load-bearing: the demand-reclassification frame (AI displacement as an aggregate-demand problem via fallacy of composition / paradox of thrift) is a genuinely original analytic move, and the animal-spirits "pre-loaded narrative of inevitability" section is the second pillar — both survive intact. The three fixed errors were factual, not structural: the opening attribution, the unverified Roosevelt quote (now paraphrased), and the redistribution claim. The redistribution fix actually strengthens the piece — the targeted-vs-undirected distinction is sharper and more defensible than the old false "Keynes wasn't a redistributionist" framing, and it now reinforces rather than contradicts the closing policy proposal. The essay has a clear thesis, a falsifiable mechanism, a concrete policy with an enforcement design, and two quotable closers. It does not need to be merged into the Darwin article; it argues from a different primitive (demand/coordination failure vs. selection/adaptation) and would lose force as a subsection. Keep it as one of the six.
In the Long Run We Are All Unemployed — and That Is the Government's Problem
"In the long run we are all dead." Keynes wrote this in 1923 against the classical confidence that the value of money corrects itself automatically over time — the quantity-theory habit of proving that the price level re-equilibrates eventually, and then treating the interval before "eventually" as economically uninteresting. He would say it again to the AI optimists who tell displaced workers to wait for the new jobs to emerge.
The line is usually quoted as flippant fatalism. It was the opposite. His full sentence, from the Tract on Monetary Reform, is sharper: "Economists set themselves too easy, too useless a task if in tempestuous seasons they can only tell us that when the storm is long past the ocean is flat again." The storm is the part that kills people. The storm is where policy lives.
That is the exact rhetorical structure of the current AI labor debate. Ethan Mollick and the thoughtful optimists are right that, historically, technology creates more jobs than it destroys. ATMs didn't eliminate bank tellers; they made branches cheaper to open, so banks opened more. The long-run supply story is real. But "the labor market re-equilibrates over a decade" and "the radiologist you laid off this quarter has rent due in March" are statements about different time horizons, and Keynes built his entire career on refusing to let the second be dissolved into the first.
The diagnosis: this is demand, not supply
Keynes's central act of intellectual rebellion was reclassifying the Great Depression. The classical view held it was a supply-side adjustment problem — wages were too high, labor too rigid, and once prices fell far enough, full employment would return. Keynes looked at the same data and saw the opposite: there was no shortage of factories, workers, or productive capacity. America in 1933 had more physical capability to produce than in 1928. What had collapsed was demand. Nobody was spending, because everybody was afraid, and everybody was afraid because nobody else was spending.
This is the part that matters for AI, so I'll be precise about the mechanism. AI displacement is, firm by firm, a supply-side efficiency gain — each company that replaces a worker with a model lowers its costs and is individually rational to do so. But income is circular. One firm's wage bill is another firm's revenue. When displacement happens at scale and simultaneously across customer-service, copywriting, paralegal, junior-analyst, and coding-assistant roles, you are not just raising productivity. You are removing purchasing power from the same population that constitutes aggregate demand for the AI-produced goods. Individual rationality compounds into collective irrationality: every firm optimizes, and the system loses the customers that made the optimization worth anything.
At Bridgewater we had a blunt name for this: a fallacy of composition. What is true for the part is false for the whole. A single household that saves more becomes more secure. An entire economy that saves more at once — the paradox of thrift — shrinks total spending, which shrinks total income, which makes everyone less secure. AI displacement is a paradox-of-thrift engine running on the labor side. Each firm "saves" on payroll; the aggregate result can be a contraction in the demand all of them depend on.
Why animal spirits make the short run worse
Keynes added a psychological term the classical model lacked: animal spirits — the observation that investment and consumption are driven not by rational expected-value calculation but by "a spontaneous urge to action rather than inaction," and that this confidence can evaporate non-linearly. This is why depressions overshoot. Fear is reflexive.
The AI transition has a confidence problem the 1930s did not: it arrives pre-loaded with a narrative of inevitability. "AI will take your job" is now ambient. A worker who believes their occupation is condemned cuts spending, avoids the mortgage, doesn't retrain for a field they assume is next. That precautionary retrenchment is itself contractionary — and it can begin well before the actual productivity gains show up in output. Keynes would warn that the expectation of displacement is an economic force prior to any displacement, and that a government waiting for the unemployment statistics to confirm a problem is, by construction, already late.
What Keynes would actually do — and what he'd refuse
He'd refuse UBI — but not for the reason people expect. Keynes was, in fact, an explicit redistributionist: in the final chapter of the General Theory he argued for shifting income toward lower earners precisely because the poor spend a larger share of each marginal pound, so redistribution is demand engineering in his own framework. His objection to UBI is not that it redistributes. It is that it redistributes into undirected cash rather than productive activity. UBI treats the problem as insufficient money in pockets; Keynes diagnosed it as insufficient output and employment, and he wanted to create work at the exact moment UBI would sever income from it. His New Deal advice to Roosevelt, in his open 1933 letter and their 1934 meeting, was specific and impatient: stop fixating on structural reform and prices, and spend on output now — increase national purchasing power through direct government expenditure. He didn't want to hand out money. He wanted to buy things, hire people, and let the multiplier run.
So the Keynesian AI program is counter-cyclical fiscal policy aimed at the sectors AI cannot supply but an aging, warming, anxious society demands more of. AI is hollowing out cognitive-routine work; meanwhile the country has a measurable, fundable deficit of human labor in eldercare, mental health, childcare, climate adaptation, and local physical infrastructure. These are not make-work. They are real unmet needs, they are labor-intensive in ways AI complements rather than replaces, and they absorb precisely the workers — empathetic, organizational, on-the-ground — that a model can't substitute for. The government's job is to be the demand stabilizer: to deficit-finance that hiring during the displacement wave, because it is the only actor with both the mandate and the balance sheet to break the coordination failure no individual firm can.
The specific recommendation
If Keynes were writing the policy, he would not propose a study commission or a "future of work" task force. He'd want an automatic stabilizer with a trigger, because he learned from the 1937 relapse — when Roosevelt cut spending prematurely, believing recovery was secured, and threw the economy back into recession — that discretionary timing fails and the political instinct is always to withdraw support too early.
Concretely: a federal AI Displacement Stabilizer — a standing program that funds care-economy and infrastructure employment, indexed to a labor-displacement indicator (sectoral job-loss velocity, not the headline unemployment rate), funded by a tax on the realized productivity gains of automation, and legally barred from being wound down until the indicator falls below threshold for a sustained period. Demand injection where AI creates new needs, financed by AI's own surplus, with the off-switch taken out of politicians' hands. Note that this is redistribution — but redistribution routed through paid work, not a transfer check, which is the whole Keynesian point.
The optimists are correct that the new jobs will come. Keynes's entire point is that "will come" is a sentence about the long run, and in the long run the displaced worker is not retrained. He is bankrupt, then bitter, then someone else's political problem. The storm is the part that matters. Govern the storm.
EDITOR NOTE: Publish standalone. After the corrections, the spine is intact and load-bearing: the demand-reclassification frame (AI displacement as an aggregate-demand problem via fallacy of composition / paradox of thrift) is a genuinely original analytic move, and the animal-spirits "pre-loaded narrative of inevitability" section is the second pillar — both survive intact. The three fixed errors were factual, not structural: the opening attribution, the unverified Roosevelt quote (now paraphrased), and the redistribution claim. The redistribution fix actually strengthens the piece — the targeted-vs-undirected distinction is sharper and more defensible than the old false "Keynes wasn't a redistributionist" framing, and it now reinforces rather than contradicts the closing policy proposal. The essay has a clear thesis, a falsifiable mechanism, a concrete policy with an enforcement design, and two quotable closers. It does not need to be merged into the Darwin article; it argues from a different primitive (demand/coordination failure vs. selection/adaptation) and would lose force as a subsection. Keep it as one of the six.