Economists Who Downplayed AI Job Losses Are Now Rewriting Their Predictions
For decades, mainstream economists were the first to pour cold water on worries that new technology would wipe out jobs. When ATMs appeared, they said bank tellers would simply shift to more customer-facing duties. When Excel spread through offices, they argued bookkeepers would become more productive, not unemployed. Even the arrival of consumer robots-like robotic vacuums-was framed as a convenient helper, not a threat to cleaners or domestic workers.
“Technology augments workers, it doesn’t replace them” became the default narrative. Now that narrative is starting to come apart.
A new multi-institutional study from researchers at the Federal Reserve Bank of Chicago, the Forecasting Research Institute, Yale, Stanford, and the University of Pennsylvania suggests that many experts are quietly but decisively changing their minds. The researchers surveyed three groups:
– 69 professional economists
– 52 AI specialists
– 38 so‑called “superforecasters,” people with an established track record of accurate predictions across complex domains
Despite their very different backgrounds, these groups converged on one uncomfortable conclusion: the faster AI capabilities advance, the fewer people will be working.
In the language of the paper, accelerated AI progress leads to “lower labor force participation.” Stripped of jargon, that means a smaller share of the population will hold jobs or be looking for work. It is, effectively, a forecast of job loss-something many economists previously insisted was unlikely at scale.
The researchers modeled several scenarios for how quickly AI might improve. In their “rapid” scenario, AI systems overtake humans on most cognitive tasks relatively quickly. Under that assumption, the forecasts for employment are grim: a substantial portion of existing roles are expected to become economically obsolete, with overall joblessness projected to rise and a significant fraction of working-age adults pushed out of the labor market entirely.
The “moderate” and “slow” scenarios tell a different, but still sobering, story. When AI progress is assumed to be gradual, the experts expect more time for adjustment: new jobs may emerge, existing roles may evolve, and retraining might work for at least part of the population. But the central pattern remains: the more powerful AI becomes, the more pressure it puts on traditional employment-even if the timeline stretches out.
One of the most striking findings is how much the consensus has shifted among economists themselves. Not long ago, many were adamant that fears about automation were overblown, pointing to historical examples: industrialization, computers, the internet. Those technologies displaced specific tasks and jobs, but overall employment, they would emphasize, kept rising. The market always found somewhere new to put people.
This time, the experts aren’t nearly as confident.
The key difference is the nature of the tasks AI can perform. Previous waves of technology excelled at physical labor or narrow, well‑defined routines. Large language models and advanced machine learning systems, by contrast, are targeting cognitive, white‑collar, and even creative work: drafting legal documents, coding software, generating marketing campaigns, summarizing complex research, and providing customer support at scale.
Many of the economists surveyed now concede that if AI systems can handle a wide spectrum of knowledge work faster, cheaper, and at comparable or better quality than humans, the old historical analogy-“we always create more jobs than we destroy”-may not hold in the same way.
Interestingly, the AI specialists and superforecasters are not more optimistic than the economists; if anything, they tend to be even more cautious about the labor market. AI experts, familiar with current research trajectories, tend to see a clear path toward systems that can automate large swaths of office work. Superforecasters, trained to weigh probabilities rather than hopes, assign nontrivial chances to scenarios where a significant share of the workforce is no longer needed for economic production.
Where the groups diverge is timing and severity. Some economists still expect a relatively drawn‑out transition, where employment slowly declines or fragments over many years. A subset of AI specialists and superforecasters, however, regard a sharper transition as plausible-especially if breakthroughs in reasoning, planning, or robotics arrive sooner than anticipated.
Another important nuance in the paper is how the experts think about “jobs” versus “tasks.” Even in pessimistic scenarios, not every occupation vanishes completely. Instead, most jobs become heavily restructured. A single AI system might take on 60-80% of the routine tasks that used to occupy a professional’s day, leaving a smaller bundle of tasks that still require human judgment, interpersonal skills, or responsibility.
The economic problem is simple: if one worker plus AI can now do what five workers used to do, firms don’t need the other four-even if the job title technically “still exists.” This is why the study’s focus on labor force participation matters. The question isn’t whether the job category survives on paper, but how many people it still meaningfully employs.
The looming issue, then, is distribution. The experts surveyed generally agree that AI will lift productivity and could make societies far richer on paper. But more output with fewer workers raises old, unresolved questions: Who owns the AI? Who captures the gains? And what happens to those who are no longer economically necessary in the traditional sense?
In their responses, participants repeatedly flagged policy as the missing piece. If governments and institutions treat AI like just another productivity tool, they risk sleepwalking into a world where millions of people are structurally unemployed or underemployed. On the other hand, if policymakers anticipate the shift, they can, in theory, cushion the blow:
– By reshaping education around adaptability, digital tools, and human‑centered skills rather than rote knowledge
– By investing heavily in retraining, especially for mid‑career workers facing disappearing roles
– By experimenting with income supports, wage subsidies, or new forms of social insurance for those displaced
Many also raised the likelihood that AI will widen existing inequalities. High‑skill workers who can design, deploy, or own AI systems may see their incomes surge. Routine knowledge workers-clerks, junior analysts, paralegals, basic programmers, content writers-could see their bargaining power erode as firms rely more on automated systems and maintain only a slim layer of human oversight.
This tension is already visible in early corporate deployments. Companies are piloting AI to handle customer service, documentation, compliance checks, and internal communications. Jobs aren’t necessarily being cut overnight, but hiring slows, open roles go unfilled, and teams learn to “do more with fewer people.” Over a decade, such incremental decisions can translate into large changes in how many humans are actually employed.
Another concern highlighted by the shift in expert opinion is geographic and sectoral unevenness. Economists in the study note that AI’s impact will not fall equally across the economy. Certain industries-finance, tech, media, professional services, customer support, and back‑office operations-are likely to feel the pressure first. Regions heavily dependent on those sectors could experience rapid employment shifts, while others, anchored in physical services or on‑site work, might see slower change-at least until AI‑powered robotics catch up.
That raises a second‑order problem: political backlash. If some communities lose a large share of their middle‑class jobs to AI while others prosper as hubs of AI development and capital, the pressure for protectionist or anti‑tech policies will grow. Many of the experts surveyed view this political instability as a serious risk, comparable to, or even more threatening than, the direct economic disruption.
The rethinking among economists also touches on how we measure “work” in the first place. Traditional labor statistics track formal employment, hours worked, and wages earned. But in a future where fewer full‑time roles exist, more people might live on a patchwork of gig work, micro‑tasks, creator income, and government transfers. Official measures might show acceptable unemployment rates while masking deep insecurity and underutilization of human potential.
Some survey respondents argue that this transition could eventually force societies to rethink the relationship between income and employment. Ideas that once seemed fringe-such as universal basic income, guaranteed public employment, or dividends tied to national wealth or automation-are gaining attention precisely because the old expectation of “everyone has a job” may no longer be realistic in a high‑automation equilibrium.
Of course, not all voices in the study are resigned to mass joblessness. A minority of economists still emphasize human comparative advantages: empathy, complex social interaction, ethical reasoning, and the sheer diversity of niche demands in a wealthy society. From this perspective, AI could trigger an explosion of new occupations we simply can’t yet imagine, just as previous technologies did. But even these optimists tend to concede that the transition period could be harsh, especially for current workers trained for tasks that AI will easily absorb.
For individuals, the shifting expert consensus carries clear implications. Workers who treat AI as an adversary to compete against are likely to struggle; those who learn to treat it as a powerful tool-integrated into their daily workflows, used to amplify their abilities-will be better positioned. Skills that complement AI, rather than directly compete with it, gain value:
– Deep domain expertise combined with AI literacy
– Relationship‑building, negotiation, leadership, and care work
– Problem framing, strategy, and responsibility for outcomes, not just task execution
For institutions-companies, governments, schools-the new stance from economists is a warning that “business as usual” is a risky bet. Planning on the assumption that the labor market will automatically self‑correct, as it has in previous technological waves, may be complacent. Active, deliberate adaptation-redesigning roles, training programs, safety nets, and education systems with AI in mind-looks less like a luxury and more like a necessity.
The core message of the study is not technological doomism but a recalibration of confidence. For years, the professional class most inclined to dismiss AI‑driven job fears is now openly entertaining scenarios where those fears are at least partially realized. The old mantra-“technology always creates more jobs than it destroys”-is no longer treated as a law of nature, but as a historical pattern that may or may not repeat under the unique conditions of advanced AI.
As AI systems grow more capable, the real debate is shifting from “Will jobs be lost?” to “How many, how fast, and who will bear the cost?” Economists, AI experts, and forecasters may disagree on the exact numbers and timelines, but the days of easy reassurance are over. The question now is whether societies will use this warning to prepare-or wait until the labor statistics catch up with what the models are already signaling.