Economists said AI wouldn’t take jobs, and some now admit they got it wrong

For many years, economists have been the professionals most likely to tell you to calm down about any technology-related concerns. ATMs have not replaced tellers, Excel has not replaced accountants, and robot vacuums have not replaced maids. There was a consensus on “increase, not replace.”

Well, that consensus is cracking.

A New paper Researchers at the Federal Reserve Bank of Chicago, the Forecasting Research Institute, Yale, Stanford, and the University of Pennsylvania surveyed 69 economists, 52 AI specialists, and 38 superforecasters about how AI will reshape the U.S. economy.

All three groups agree on one thing: faster progress in artificial intelligence means lower labor force participation. That’s the polite way of saying “fewer people working.”

The numbers are amazing. Under what researchers call a “rapid” scenario — in which AI outperforms humans on most cognitive and physical tasks by 2030 — economists expect the U.S. labor force participation rate to decline from the current 62% to 54% by 2050.

Nearly half of this decline, roughly 10 million jobs lost, can be attributed directly to AI rather than to demographics or other trends.

The fast scenario is not science fiction. It is a world where AI can negotiate contracts and assist in any factory or home and replace all the freelance software engineers, paralegals and customer service agents.

Anthropic CEO Dario Amodei Already warned The disruption is accelerating faster than most people expect, and the study’s rapid scenario effectively validates this framework. GDP tells the other half of the story.

Under the same rapid scenario, economists expect annual GDP growth to reach 3.5% by 2045-49, approaching post-World War II prosperity levels. Artificial intelligence experts are more optimistic, as they expect growth of 5.3%. Massive total wealth creation, concentrated at the top, with less labor power to share it. The researchers suggest that under rapid artificial intelligence, the richest 10% of households could own 80% of total wealth by 2050, which is higher than inequality before World War II.

But there is a nuance that is often lost in the debate about AI jobs. The study found that expert disagreement is not mainly about whether strong AI will arrive, but about what happens to the economy once it arrives. This is a meaningful shift. Previous pro-technology arguments assumed that even transformative automation would eventually create new categories of work. A new question that economists are grappling with is whether artificial intelligence, unlike ATMs, automates the task of inventing new tasks.

For now, overall employment data still looks mostly stable. A Yale and Brookings study Since late 2025, no sign of mass unemployment has been found nearly three years after ChatGPT was launched. but research The new report documents a relative employment decline of 13% among workers ages 22 to 25 in occupations most exposed to AI. The macro is stable. The leading edge is not.

On politics, economists and the general public have a heated exchange. Economists favor targeted retraining programs (71.8%) and largely reject job guarantees (13.7%) and universal basic income (37.4%). The general public is more open to structural interventions. The authors of the research point out that the optimal policy depends heavily on the scenario that will be implemented, and at the moment, no one knows which scenario will play out.

So “augment rather than replace” is not dead, but it is now on life support, and the economists running the numbers have enough data to worry about.