Economists Admit They Are Guessing About AI’s Economic Impact
A recent survey of several hundred economists found widespread uncertainty about AI’s economic impact, with respondents largely admitting they are navigating the issue with limited evidence and relying on analogies and assumptions rather than hard data.
How confident are economists about AI’s economic impact?
Confidence is modest to low across the board. The survey asked economists to rate their confidence in forecasts about AI’s impact on productivity, wages, employment levels, and income distribution, and respondents reported modest to low confidence in every category. Many said publicly that they simply do not know what will happen, and a meaningful share said their views could change quickly as new evidence emerges.
Where do economists agree, and where do they disagree?
Despite the uncertainty, some patterns surfaced. A majority of respondents expected AI to raise overall productivity over time, though estimates of how much and how soon varied widely.
Views on the labor market were more divided. Some economists predicted that AI would complement existing workers and lift wages, while others warned that certain job categories, particularly routine cognitive work, could face displacement.
On inequality, opinions diverged further. Several respondents flagged AI’s potential to widen the gap between high-skilled and low-skilled workers, especially if the gains accrue to firms and employees who already have access to advanced tools. Others argued that falling costs and broader access could reduce disparities, though they acknowledged this outcome depends heavily on policy choices.
Why is the picture so unclear?
Economists pointed to several reasons for their hesitation. Historical comparisons, such as the spread of electricity or personal computers, offer rough templates, but AI’s generality makes those analogies imperfect. The technology is improving quickly, adoption patterns vary by industry and country, and the downstream effects on prices, demand, and hiring are difficult to isolate.
There is also a measurement problem. Official productivity statistics lag behind real-world changes, and many firms are still experimenting with AI rather than deploying it at scale. As a result, the macroeconomic data that economists typically rely on has not yet caught up with the technology’s presence in workplaces.
What would help clarify AI’s economic effects?
Respondents suggested a few paths to clearer answers. More granular firm-level studies, better tracking of task-level automation, and faster publication of adoption data would all help. Some also called for governments and research institutions to invest in long-running studies that track AI’s effects over years rather than quarters.
A cautious bottom line
The survey’s clearest message is humility. Even professionals whose job is to model the economy are openly working in the dark on one of the most discussed technological shifts in decades. That uncertainty does not mean AI’s economic impact will be small, but it does mean the confident predictions circulating in boardrooms, news headlines, and policy debates deserve a healthy dose of skepticism until the evidence catches up.
FAQ
What did the survey of economists find about AI’s economic impact?
The survey of several hundred economists found modest to low confidence across all categories they were asked about, including productivity, wages, employment levels, and income distribution. Many respondents said they do not know what will happen.
Do economists agree on whether AI will help or hurt workers?
Economists are divided. Some predicted that AI would complement existing workers and lift wages, while others warned that certain job categories, particularly routine cognitive work, could face displacement.
Why are economists uncertain about AI’s economic impact?
Respondents cited imperfect historical analogies, fast-moving technology, uneven adoption across industries and countries, lagging official productivity statistics, and many firms still experimenting with AI rather than deploying it at scale.