← All Notes

In 1962, Milton Friedman published Capitalism and Freedom. It became the intellectual foundation for the free-market revolution that followed — deregulation, union decline, tax cuts for capital, the gradual dismantling of worker protections that characterized the next forty years of economic policy.

In the same book, Friedman proposed the negative income tax: a mechanism by which the government sends money to people whose income falls below a threshold, rather than collecting taxes from them. Simple. Efficient. No bureaucracy. No means-testing. No welfare queens. Just a floor.

He called it a market-compatible solution to poverty. He was right about the compatibility. He underestimated when it would become necessary.

The negative income tax is Universal Basic Income. And UBI — or something structurally identical to it — is what AI-driven displacement of the knowledge economy eventually requires to prevent consumer capitalism from collapsing under its own logic.

The free market's most powerful tool demands the free market's antithesis to function. That is the structural irony at the end of the road.

Why the Market Can't Self-Correct

Every time technology displaces workers at scale, economists reach for the same argument: don't worry, new sectors will emerge to absorb them. The industrial revolution destroyed agricultural jobs and created factory jobs. Automation destroyed factory jobs and created service and knowledge jobs. The internet destroyed retail and created e-commerce, logistics, and digital services. The market self-corrects.

The argument has been correct, historically. It has also been applied mechanically, without examining what actually made it work.

New sectors absorbed displaced workers in prior waves because two conditions held: the workers could retrain, and the new sectors needed human cognition. Steel workers could learn to operate service businesses. Clerical workers could learn to work in knowledge industries. The skills required for the new sectors were acquirable through training.

AI automates cognition. The new sector is the one doing the displacing. The "new jobs" that AI creates — AI trainers, prompt engineers, safety researchers, model evaluation specialists — employ hundreds of thousands, not tens of millions. They require deep technical skills, not the retraining-accessible skills that absorbed prior cohorts of displaced workers.

Jobs displaced vs. jobs created — each technology wave

The chart above is the death of the self-correction argument. Each prior wave left a manageable residual: more displaced than created in some categories, eventually absorbed by adjacent growth. The AI wave projects 28 million jobs at risk against perhaps 1.5 million new AI-specific roles — a gap of over 25 million, occurring in a window of 5 to 10 years with no visible absorbing sector.

"New sectors we can't imagine yet will emerge" is the final refuge of the argument. It's theoretically unfalsifiable and practically useless. The people being displaced don't have a decade to wait for unknowable sectors to materialize. Their mortgages are due monthly.

The Irony in Full

Here is what actually happened.

The free-market revolution of the 1980s and 1990s was, at its core, an argument about who deserves the surplus. Productivity gains from technology and globalization — the argument went — should flow to capital, not labor, because capital is what takes the risk that enables productivity. Workers should accept market wages. Unions distort efficient labor markets. Government redistribution creates dependency. The economy works best when it's allowed to sort naturally.

The companies that won this argument — the tech giants, the financial sector, the global manufacturers who offshored production — captured an extraordinary share of the productivity surplus over the next four decades. Labor's share of GDP fell from 68% to under 57%. Corporate profits as a share of GDP more than doubled. The argument worked, on its own terms: capital did very well.

And now capital's most powerful new tool — AI — is about to eliminate the consumer base that capital needs to sell to.

You cannot run a consumer economy without consumers. The middle-class and upper-middle-class workers whose incomes sustained 70% of US GDP — the mortgage payers, the car buyers, the restaurant-goers, the vacation-takers — are the people being displaced. Corporate earnings go up when you eliminate them. Corporate revenues go down when they stop spending. The bull market and the consumer collapse are the same event, separated by the lag between cost reduction and revenue impact.

The only mechanism that breaks this loop is redistribution: taking some of the surplus that AI generates for capital and returning it to the people who were displaced so they can continue to function as consumers. The market produced the problem and cannot solve it. The solution is, structurally, socialism — income transferred from those who have it to those who need it, at a scale determined by the gap between displaced workers and available income.

Sam Altman, the CEO of OpenAI, backs UBI research. Elon Musk has called it inevitable. Andrew Yang built a presidential campaign on it. These are not leftwing politicians. They are capitalists who have thought through the demand-side math and arrived, inescapably, at the same conclusion Friedman did in 1962: the floor has to exist.

What It Would Actually Cost

The intellectual concession is easy. The fiscal reality is not.

The fiscal math: what redistribution at scale actually costs

A full UBI at $15,000 per year for all American adults costs $3.9 trillion annually. The entire current federal budget is $6.5 trillion. Social Security — the largest existing transfer program — costs $1.4 trillion. Medicare and Medicaid together cost $1.8 trillion. Adding a full UBI would increase total federal spending by 60%.

A targeted intervention — bridge income of $8,000 per year for the 25 million projected displaced workers — costs roughly $370 billion. That's significant but manageable: it's larger than the defense discretionary budget but smaller than Medicare. It's the number that economists who take the problem seriously tend to land on as a minimum viable floor.

Where does $370 billion to $3.9 trillion come from? The obvious answer is the companies and shareholders who captured the productivity surplus: corporate income tax, capital gains tax, a wealth tax, an automation tax on firms above a certain scale. The revenue exists in theory. Accessing it is the political problem.

The Three Walls

The political economy of redistribution at scale runs into three walls simultaneously.

The timing wall. Political consensus for major redistribution requires visible crisis. Social Security passed after the Depression. Medicare passed after decades of elderly poverty data. The Affordable Care Act passed after a sustained crisis of uninsurance. The pattern: harm must be widespread, documented, and politically legible before the coalition to address it can be built.

AI displacement doesn't work that way. The displacement is currently invisible to most of its eventual victims — people who still have jobs, who see AI as a productivity tool rather than a replacement, who have not yet been handed a severance package with a note about "operational restructuring." By the time it's visible enough to generate the political coalition for redistribution, the displaced workers have already spent their savings, lost their housing, and lost the political influence that comes with economic security. The people who would vote for redistribution are the ones who get processed first.

The mobility wall. Capital is not national. Any jurisdiction that implements an automation tax above what competitors impose faces capital flight: AI companies restructure operations to lower-tax jurisdictions, the tax base contracts, the redistribution funding shrinks. This is the same dynamic that has prevented effective corporate taxation for thirty years — the race to the bottom on corporate tax rates, played again at higher stakes.

The problem is genuinely global. If the US imposes an automation tax and the EU doesn't, AI operations migrate. If both impose it and China doesn't, Chinese AI companies outcompete both. The global coordination required to prevent this arbitrage is the same coordination that has failed to prevent labor arbitrage, environmental arbitrage, and corporate tax arbitrage for three decades. There is no mechanism that produces it.

The lobbying wall. The companies that would fund redistribution have extraordinary resources to prevent it. Tech industry lobbying has grown from negligible in 2000 to over $100 million annually at the federal level, and these figures don't capture the full picture of soft influence: the revolving door between tech companies and regulatory agencies, the funding of think tanks that produce market-friendly research, the campaign contributions that shape which legislators sit on which committees.

The workers being displaced have diminishing resources as the displacement progresses. Unemployed people donate to campaigns less. People in economic crisis have shorter time horizons and are more susceptible to arguments that blame their condition on immigrants or foreign competition rather than automation. The political coalition that would fund redistribution weakens as the displacement it needs to address grows.

These three walls are not independent. They reinforce each other. The timing wall means the coalition forms late. The mobility wall means the revenue base has already partially fled. The lobbying wall means the legislation that passes is shaped by the interests it was meant to tax.

What Friedman Actually Understood

Friedman's negative income tax proposal has always been politically radioactive because it concedes the necessity of redistribution. If you accept that people below a floor should receive a government transfer, you have accepted that the market, by itself, produces outcomes that are politically unacceptable — that there are losers the system is obligated to cushion.

The free-market right never fully accepted this concession. The policy that emerged instead was means-testing, work requirements, in-kind benefits, programs with constituencies that could be managed and reduced and stigmatized. The result was a welfare state that was expensive to administer, inadequate to the problem, and politically vulnerable to "welfare queen" narratives.

What Friedman understood, and what the next decade will force into the open, is that the choice is not between redistribution and no redistribution. The choice is between redistribution by design and redistribution by crisis. The New Deal was redistribution by crisis — panic-driven, expensive, structurally incoherent, eventually effective. The alternative Friedman proposed was redistribution by design: simple, efficient, funded by growth, politically durable.

The AI displacement is going to produce a crisis. The question is whether the political system responds before the crisis — with a designed floor — or after it, with whatever panicked coalition forms when the demand-side collapse is no longer deniable.

The after-crisis response is historically worse. It's faster, more expensive, less coherent, and generates the political conditions for demagoguery. The Weimar Republic did not end because German politicians couldn't do math. It ended because the crisis came faster than the math could be applied.

That is the structural irony of where we are. The free-market revolution that dismantled worker protections for forty years produced a technology so powerful that its own logic now demands those protections back — at a scale that would have been considered radical in any prior era — and produced a political economy so tilted toward capital that administering them in time is nearly impossible.

Socialism saves capitalism. Whether capitalism will permit that in time to matter is the question Part 4 takes on.

— J.P. Howlett

Next: Tell Me I'm Wrong — the case for hope, the case against it, and what the narrow path actually looks like.

Part 1: When the Levee Breaks — the dam, the leading indicators, and the signal to watch.

Part 2: The Demand Side Doesn't Forgive — forty years of ratchet, three piggy banks, and the escalator that ends in the air.

Also in the series: Transitional Tech — the gig economy was itself transitional technology. The floor disappears.

Sources