For the better part of three years, mainstream economists have waved off warnings about artificial intelligence displacing workers. Rising unemployment among young college graduates? Blame high interest rates. Companies laying off staff and citing AI? That was "AI washing" by executives looking for convenient cover. The prevailing consensus in the economics profession held that AI was overhyped and that the labor market would adapt, as it always had. That consensus is now fracturing. A new working paper published this week, surveying economists about their outlook over the next five and 25 years, reveals a profession quietly reconsidering its assumptions and growing worried that policymakers are not remotely prepared for what might be coming.
The Shift
Daniel Rock, an economist at the Wharton School of the University of Pennsylvania who has studied the economic impact of artificial intelligence for years, captured the emerging mood in an interview with The New York Times. "I don't think A.I. has hit the labor market yet, and I don't think it's radically changed corporate productivity yet, either, but I think it's coming," Rock said.
That careful phrasing matters. Economists are not declaring that mass displacement is happening right now. What they are saying, with growing conviction, is that the window of comfortable denial is closing. The new working paper, coauthored by Ezra Karger, an economist at the Federal Reserve Bank of Chicago, found that most surveyed economists expect the economy to grow somewhat faster as AI improves. But the researchers also explored a scenario they consider unlikely but plausible: rapid AI advancement that produces faster growth alongside greater inequality and the disappearance of millions of jobs.
"Economists are certainly taking A.I. seriously," Karger told The New York Times.
The Data That Changed the Debate
Several converging research threads are driving the shift. In February, the Federal Reserve Bank of Dallas published an analysis showing that employment in the 10 percent of sectors most exposed to AI has declined 1 percent since ChatGPT's release in late 2022, while total U.S. employment grew approximately 2.5 percent over the same period. Employment in the computer systems design sector specifically has dropped 5 percent.
The Dallas Fed analysis, authored by researchers including Stanford University economists Erik Brynjolfsson, Bharat Chandar, and Ruya Chen, found that the employment decline in AI exposed sectors is falling disproportionately on workers under age 25. Employment totals for older workers have not declined. The researchers suggest this pattern reflects AI's ability to automate codified knowledge, the kind of information found in textbooks, while failing to replicate the tacit knowledge that comes only from years of experience.
Wages in AI exposed sectors, however, have not fallen. In fact, they are rising. Nominal average weekly wages in the computer systems design sector have increased 16.7 percent since fall 2022, compared with 7.5 percent nationally. This seemingly contradictory finding, declining employment alongside rising wages, supports the theory that AI is simultaneously replacing entry level workers and augmenting experienced ones. The young workers losing their footholds are being replaced by technology, while the veterans whose tacit knowledge AI cannot replicate are becoming more valuable.
The Gap Between What AI Can Do and What It Is Doing
A March report from Anthropic, the AI company behind the Claude model, introduced a metric called "observed exposure" that compares what AI is theoretically capable of doing with what it is actually doing in professional settings. The results were striking. For computer and math workers, large language models are theoretically capable of handling 94 percent of their tasks. Yet the technology currently covers only 33 percent of those tasks in observed professional use. The same gap exists across office and administrative roles: 90 percent theoretical capability, a fraction of that in actual use.
The Anthropic researchers, Maxim Massenkoff and Peter McCrory, attribute the gap to legal constraints, technical limitations, and the continued need for human review of AI output. But they project those barriers are temporary. As capabilities improve and adoption deepens, they write, the actual usage will expand to fill the theoretical capability. The report names the scenario business leaders should be planning for: a "Great Recession for white collar workers," noting that during the 2007 to 2009 financial crisis, the overall U.S. unemployment rate doubled from 5 percent to 10 percent. A comparable doubling in the top quartile of AI exposed occupations, from 3 percent to 6 percent, would be detectable and significant.
The workers most at risk are not who most people picture. According to the Anthropic analysis, the most AI exposed group is 16 percentage points more likely to be female, earns 47 percent more on average, and is nearly four times as likely to hold a graduate degree compared to the least exposed group. The people facing the greatest theoretical displacement are lawyers, financial analysts, and software developers, not warehouse workers or mechanics.
The Counterargument Is Weakening
There are still credible voices arguing that the AI displacement narrative is premature. An analysis from The Atlantic published this week found that unemployment for young workers has increased the most for those in occupations least exposed to AI, such as construction workers and fitness trainers. Economists Adam Ozimek and Nathan Goldschlag showed that a significant number of young workers without college degrees have simply stopped looking for work altogether, distorting the unemployment rate and creating the appearance that college graduates are uniquely struggling.
"It turns out the labor market for young people, all young people, is even worse than we thought," Goldschlag told The Atlantic. "That makes me doubt that this is an AI story."
Martha Gimbel, the executive director of the Yale Budget Lab, echoed this caution. "I'm very open to the possibility that AI could displace entry level workers," Gimbel said. "But we're just not seeing it show up anywhere in the data."
These arguments were more persuasive a year ago than they are today. The Dallas Fed data showing that employment declines are concentrated specifically in AI exposed sectors, and specifically among young workers in those sectors, is harder to wave away. The fact that wages are rising for experienced workers in those same sectors fits neatly into the theory that AI is replacing book knowledge while complementing experiential expertise. And the Anthropic gap analysis suggests that whatever displacement has occurred so far is only a fraction of what is technically possible.
The Policy Vacuum
What unites both sides of the debate is a shared concern about policy readiness. Economists who believe AI is already affecting the labor market and those who think it will happen soon both agree that existing government programs are not designed to handle what is coming. Workforce retraining programs remain underfunded and poorly targeted. The safety net for displaced workers, designed in an era of manufacturing layoffs and regional plant closures, is not equipped for a scenario in which entire categories of white collar work contract simultaneously across every city in the country.
The working paper from Karger and colleagues emphasizes that policymakers should be acting now to modernize programs that could help displaced workers. That message has so far landed in a political environment consumed by the Iran war, trade disputes, and the approaching midterm elections. Congress has shown little appetite for ambitious labor market legislation, and the administration has been focused on deregulating AI rather than preparing for its potential disruptions.
David Deming, an economist at Harvard, offered a historical frame. "This is the same thing that happened with high school 50, 60 years ago," Deming told The Atlantic. "A high school diploma used to confer huge advantages, but then it became so ubiquitous that the advantages went away." The implication is that a college degree may be following the same trajectory, with AI accelerating a process that was already underway.
What Comes Next
The economics profession's shift on AI is not a reversal. It is a migration from confident dismissal to cautious alarm. Most economists still do not see definitive proof that AI is currently driving mass displacement. But the number who consider it a serious near term possibility has grown substantially, and the data are beginning to point in a direction that the skeptics cannot easily explain away.
The question is no longer whether AI will reshape the labor market. It is whether anyone will be ready when it does.