A wide-ranging new study shows that academic, industry, and policy experts largely agree that artificial intelligence will reshape the economy, but they differ widely on the extent to which it will do so.
What will the American economy look like by the time artificial intelligence can negotiate book deals, run movie studios, and outperform every software engineer on the planet?
Once the realm of science fiction, this question has moved into the predictive models of central banks, university economics departments, and Wall Street research desks. Now, a team of researchers has set out to systematically measure what experts actually believe. The result is a stark rebuttal to some of the apocalyptic and utopian predictions circulating in Silicon Valley.
Read next: Yes, AI will take your job.
The study, conducted by researchers at the Federal Reserve Bank of Chicago, Yale University, Stanford University, the University of Pennsylvania, and the Forecasting Institute, surveyed five groups: academic economists, employees of frontier AI companies, policy researchers, elite forecasters known as superforecasters, and the general public. The survey was conducted from October 2025 to February this year.
Discovering headlines is deceptively simple. Most experts expect significant advances in AI, but not economic transformation, at least in the short term.
Technology that surprises no one
Across all five groups surveyed, the majority of respondents assigned a meaningful probability that AI capabilities would significantly advance by 2030. The average economist has a 61 percent chance of being in what researchers call a “moderate or rapid” progress scenario, one in which AI systems move far beyond their current role as sophisticated assistants and now begin to autonomously perform tasks that require days of skilled human labor.
And yet, when we asked these same economists for their unconditional forecast for GDP growth (their best overall guess, taking into account everything they know), the median answer was 2.5% annualized to 2030. That’s only slightly above recent benchmarks and narrowly ahead of nearly all forecasts by government and private sector forecasters.
The gap between these two findings—high confidence in AI’s future capabilities and modest expectations for short-term economic benefits—runs through the study and reveals the study’s central puzzle.
Read next: Is it okay to not have an advisor? Launch AI Startup Era File as RIA
Economists explained this disconnect in written rationales. Electrification, automobiles, personal computers. Every innovative general-purpose technology in modern history took decades to register in aggregate productivity statistics. The reason is well known. Businesses will need to redesign their workflows, employees will need to learn new skills, regulators will need to adapt, and physical infrastructure will need to be built to support new technologies. Economists say AI is unlikely to be any different.
Geopolitical headwinds made them wary. Trade tensions, an aging population, declining immigration, and constraints on energy and data center construction have all repeatedly emerged as reasons why even highly capable AI may not immediately translate into tangible economic benefits.
when things move quickly
Things change dramatically when economists are asked to imagine the most dramatic scenarios. By 2030, AI systems will outperform humans at most cognitive and physical tasks, writing Pulitzer-level novels, compressing years of research schedules into days, and operating robots that can move through every home and factory on the planet.
Under that assumption, the projected median annualized GDP growth rate would rise to 3.5% by 2050, approaching the pace of the postwar economic boom. The labor force participation rate, currently about 62 percent, has fallen to 55 percent, and about half of that (representing about 10 million workers) is attributable to AI rather than demographic or other trends. Wealth held by the top 10 percent of households has risen to 80 percent, a level not seen since the late 1930s, although the United States has reached that level before.
Read next: Morgan Stanley says market misunderstands how AI will impact brokerages
That’s a big number. Researchers note that these numbers are not published by the most prominent people in the technology industry. Anthropic CEO Dario Amodei has suggested that unemployment could reach 10-20% within five years. OpenAI’s Sam Altman predicted that jobs at all levels will disappear. Academic literature includes models predicting growth rates of more than 30 percent in a world of fully autonomous AI
Economists in the survey made no such predictions, even under a rapid-development scenario.
The researchers suggest one possible explanation. Survey participants may have placed too much emphasis on historical data that was displayed prominently in the survey interface. Another possibility is that the study’s scenario descriptions, while ambitious, may not capture the most extreme parts of the capability range envisioned by the industry’s most bullish observers.
The real cause of disagreement
Perhaps the most intellectually striking finding of this study concerns what causes disagreement among experts.
A common assumption in discussions about the economic impact of AI is that the main debate is about the technology itself: whether truly transformative AI will actually emerge. According to this view, once we agree on the pace of progress, the economic consequences will more or less follow.
The survey data directly challenges that assumption. Using a statistical technique called variance decomposition, the researchers found that disagreements about long-term economic outcomes are driven primarily by differing beliefs about what economic effects a particular level of capability will actually produce, rather than differing beliefs about the pace of AI development.
In other words, two economists who agree that there is a 20% chance that AI will advance rapidly by 2030 may have fundamentally different views about how that rapid advance will affect GDP and labor participation. The real uncertainty lies in economic mechanisms: how quickly technology will spread across the economy, whether new jobs will replace displaced people, and how institutions and regulations will respond.
Occupations that are at risk and those that are not
When asked to rank specific occupations by expected changes in employment by 2030, economists were fairly consistent about who was most at risk.
Office workers, assemblers, and machine operators rank near the bottom, and these occupations are characterized by routine cognitive and manual tasks and have long been considered vulnerable to automation. At the top are personal care workers, medical professionals, and military occupations, roles defined by physical presence, human interaction, and security functions that remain difficult to automate.
Blue-collar occupations, as a group, were more likely to fall into the unemployed category than white-collar occupations. Employment in the care and services sector is wide-ranging, but the survey also included some of the most optimistic forecasts.
Conditioning these rankings based on the AI’s rapid scenarios did not dramatically reorder the ranks. The researchers note that this result may reflect that respondents already have some expectation of AI’s effectiveness built into their unconditional answers.
(Free) Please show me the money?
The study also measured something that economic forecasts rarely capture: what policymakers should actually do about it.
Here, the divergence between economists and the general public was striking, and in at least some cases nearly reversed.
Economists strongly supported targeted intervention. Retraining assistance, a program that provides up to $25,000 in annual training credits to job leavers along with career counseling and immigration assistance, received 71.8% support from economists, the highest of any policy examined. Modernized unemployment insurance, extended to cover workers displaced by automation at 75% of their previous pay for up to 18 months, received support from 62.3%.
By contrast, the federal job guarantee program, which provides government-funded jobs with a minimum wage of $15 an hour to willing adults, received support from just 13.7% of economists. The same policy was supported by 57.1 percent of the general public.
Several economists opposed a universal basic income funded by a 15% value-added tax, with 38.2% opposed. There was a lot of support for this among the general public.
Economists say the chances of these policies being implemented by the end of this year are almost zero.
The country is divided on how to respond to AI: Economists vs. the general public
Percentage of respondents supporting implementation of each policy to address the economic impact of AI. The gulf between professional economists and the general public is greatest for the most far-reaching proposals.
Economist (% agree)
General (% approval)
uncertain horizon
The study authors are frank about the limitations of what the data can tell us. The scenarios used to derive conditional predictions grouped together different aspects of AI capabilities. In other words, the two respondents who both chose the “quick” scenario may have been imagining completely different technological worlds. This study also cannot clearly separate the causal effects of AI from the broader economic environment in which it is deployed.
What the data capture with unusual precision is the construction of expert uncertainty. Under normal unconditional assumptions, economists’ forecasts cluster around a historical baseline. Under the rapid scenario, the distribution of possible outcomes expands significantly, especially with respect to labor force participation, and the integrated projection for 2050 includes significant probabilities below 45%.
This combination, a moderate median expectation and a fat tail in the most important scenarios, may be the report’s most policy-relevant finding. If progress in AI is slow, existing institutions and gradual reforms may be sufficient. If it is rapid, the wide range of possible outcomes means that policymakers cannot simply plan for intermediate cases.
As one economist wrote in the rationale for the study, a rapid scenario creates real upside, but it also increases uncertainty. This is because even in the same world, large frictions and costly readjustments can occur during periods of transition, and even if fundamental capacities are significant, measured GDP growth can be constrained by adjustment costs.
For advisors who help clients make retirement planning, allocation across asset classes, and human capital investment decisions, uncertainty is not an abstraction. It’s the environment that has to make those decisions.
