By the time artificial intelligence can negotiate a book contract, manage a movie studio and outperform every software engineer on earth, what will have become of the American economy?
That question, once the province of science fiction, has migrated into the forecasting models of central banks, university economics departments and Wall Street research desks. Now a team of researchers has tried to systematically measure what the experts actually believe — and the results offer a sobering counterweight to some of the more apocalyptic and utopian predictions circulating in Silicon Valley.
Read next: Yes, AI is going to take your job
The study, conducted by researchers at the Federal Reserve Bank of Chicago, Yale, Stanford, the University of Pennsylvania and the Forecasting Research Institute, surveyed five groups: academic economists, employees at frontier A.I. companies, policy researchers, elite forecasters known as superforecasters, and members of the general public. The survey ran from October 2025 through February of this year.
The headline finding is deceptively simple: most experts expect significant A.I. progress, but not economic transformation — at least not in the near term.
Across all five groups surveyed, a majority of respondents assigned meaningful probability to substantial advances in A.I. capabilities by 2030. The average economist placed a 61 percent chance on what the researchers called a "moderate or rapid" progress scenario — one in which A.I. systems move well beyond their current role as sophisticated assistants and begin autonomously performing tasks that now require days of skilled human labor.
And yet, when the same economists were asked to give their unconditional forecasts for G.D.P. growth — their all-things-considered best guesses, accounting for everything they know — the median answer was 2.5 percent annualized growth through 2030. That is only marginally above recent baselines, and it exceeds nearly all projections from government agencies and private-sector forecasters by a thin margin.
The gap between those two findings — high confidence in A.I.'s coming capabilities, modest expectations for near-term economic impact — runs through the entire survey and defines its central puzzle.
Economists explained the disconnect in their written rationales. Electrification, the automobile, the personal computer: every transformative general-purpose technology in modern history took decades to register in aggregate productivity statistics. The reasons are familiar — businesses must redesign workflows, workers must acquire new skills, regulators must adapt, and the physical infrastructure to support new technologies must be built. A.I., in the economists' telling, is unlikely to be different.
Geopolitical headwinds compounded their caution. Trade tensions, an aging population, declining immigration and constraints on energy and data center construction all appeared repeatedly as reasons that even highly capable A.I. might not translate quickly into measurable economic gains.
The picture changes substantially when economists are asked to assume the most dramatic scenario: one in which, by 2030, A.I. systems surpass human performance on most cognitive and physical tasks — writing Pulitzer-caliber novels, collapsing years-long research timelines into days, and operating robots capable of navigating any home or factory on earth.
Under that assumption, median forecasts for annualized G.D.P. growth rise to 3.5 percent by 2050 — approaching the pace of the postwar economic boom. The labor force participation rate, currently around 62 percent, falls to 55 percent, with roughly half that decline — equivalent to about 10 million workers — attributable to A.I. rather than to demographics or other trends. Wealth held by the top 10 percent of households rises to 80 percent, a level not seen since the late 1930s, though one the United States has reached before.
Those are large numbers. They are also, the researchers are careful to note, not the numbers some of the most prominent voices in the technology industry have floated. Dario Amodei, the chief executive of Anthropic, has suggested unemployment could reach 10 to 20 percent within five years. Sam Altman of OpenAI has predicted that whole classes of jobs will vanish. The academic literature contains models projecting growth rates of 30 percent or more in a world of fully autonomous A.I.
The economists in this survey, even under the rapid-progress scenario, forecast nothing of the kind.
The researchers suggest one possible explanation: participants in the survey may have anchored too heavily on historical data, which was prominently displayed in the survey interface. Another possibility is that the study's scenario descriptions, while ambitious, may not capture the most extreme end of the capability spectrum imagined by the industry's most bullish observers.
Perhaps the study's most intellectually striking finding concerns what drives disagreement among experts.
A common assumption in debates about A.I.'s economic impact has been that the primary argument is about the technology itself — whether truly transformative A.I. will actually arrive. On this view, once you agree on the pace of progress, the economic consequences more or less follow.
The survey's data challenge that assumption directly. Using a statistical technique called variance decomposition, the researchers found that disagreement about long-run economic outcomes is driven primarily not by different beliefs about the pace of A.I. development, but by different beliefs about what economic impact a given level of capability will actually produce.
In other words, two economists who agree that there is a 20 percent chance of rapid A.I. progress by 2030 may hold radically different views about what that rapid progress would do to G.D.P. or labor force participation. The economic mechanisms — how quickly technology diffuses through the economy, whether new jobs offset displaced ones, how institutions and regulations respond — are where the real uncertainty lives.
U.S. labor force participation rate, historical and forecast to 2050, by A.I. progress scenario. Shaded band shows the range of economist uncertainty under the rapid scenario.
When asked to rank specific occupations by expected employment change through 2030, economists showed a fairly consistent view of who faces the most exposure.
Clerical workers, assemblers, and machine operators ranked near the bottom — occupations characterized by routine cognitive and manual tasks long considered vulnerable to automation. At the top were personal care workers, health professionals and military occupations — roles defined by physical presence, human interaction or security functions that remain difficult to automate.
Blue-collar occupations, as a group, were more likely to be placed in the job-loss category than white-collar ones. Care and service sector jobs, while spanning a wide range, included some of the most optimistic forecasts in the survey.
Conditioning those rankings on the rapid A.I. scenario did not dramatically reorder them — a finding the researchers note may reflect that respondents were already incorporating some expectation of A.I. effects into their unconditional answers.
Occupations ranked by economists' expected employment change between 2025 and 2030 (unconditional scenario). Bars show relative ranking — the paper presents ordinal rankings rather than precise percentage figures for each occupation. The 50% midpoint divides those where a majority of economists expect growth (right) from those where a majority expect decline (left).
The survey also measured something rarely captured in economic forecasting: what should policymakers actually do about it?
Here, the divergence between economists and the general public was stark and, in at least one case, nearly inverted.
Economists strongly favored targeted interventions. Retraining support — a program offering displaced workers up to $25,000 per year in training credits, along with career counseling and relocation assistance — drew 71.8 percent support from economists, the highest of any policy tested. Modernized unemployment insurance, extended to cover workers displaced by automation at 75 percent of prior salary for up to 18 months, drew 62.3 percent support.
By contrast, a federal job guarantee program — offering any adult who wanted one a government-funded job paying at least $15 an hour — drew the support of only 13.7 percent of economists. The same policy was backed by 57.1 percent of the general public.
Universal basic income, funded by a 15 percent value-added tax, found a plurality of economists opposed, at 38.2 percent against. Among the general public, a plurality supported it.
Economists assigned essentially zero probability to any of these policies being enacted by the end of this year.
Share of respondents supporting implementation of each policy to address A.I.'s economic impacts. The gap between expert economists and the general public is widest on the most expansive proposals.
The survey's authors are candid about the limits of what their data can tell us. The scenarios used to elicit conditional forecasts bundled many dimensions of A.I. capability together, meaning two respondents who both chose the "rapid" scenario may have imagined quite different technological worlds. The study also cannot cleanly separate the causal effect of A.I. from the broader economic environment in which it unfolds.
What the data do capture, with unusual precision, is the structure of expert uncertainty. Under normal, unconditional assumptions, economists' forecasts cluster tightly around historical baselines. Under the rapid scenario, the distribution of possible outcomes widens enormously — particularly for labor force participation, where the pooled forecast for 2050 includes a non-trivial probability mass below 45 percent.
That combination — moderate median expectations, fat tails in the scenarios that matter most — may be the most policy-relevant finding in the report. If A.I. progress is slow, existing institutions and incremental reforms may be adequate. If it is rapid, the breadth of possible outcomes means policymakers cannot simply plan for the middle case.
As one economist wrote in their survey rationale: the rapid scenario creates a genuine upside tail, but it also widens uncertainty — because the same world can feature large transitional frictions, costly retooling, and the possibility that measured G.D.P. gains are damped by adjustment costs, even if the underlying capabilities are remarkable.
For advisers helping clients plan for retirement, allocate across asset classes or make decisions about human capital investment, that uncertainty is not an abstraction. It is the environment in which those decisions will have to be made.
A former Alabama investment advisor and ex-Kestra rep has been permanently barred and penalized after clients he promised to protect got caught in a $2.6 million fraud.
As more active strategies get packaged into the ETF wrapper, advisors and investors have to look beyond expense ratios as the benchmark for value.
Survey finds AI widely embedded in research and analysis, but barely touching portfolio construction or trade execution.
Two firms land teams managing more than $1.1 billion in combined assets from Kestra and Edward Jones.
A private partnership, Edward Jones is a giant in the retail brokerage industry with more than 20,000 financial advisors.
Wellington explores how multi strategy hedge funds may enhance diversification
As technical expertise becomes increasingly commoditized, advisors who can integrate strategy, relationships, and specialized expertise into a cohesive client experience will define the next era of wealth management
