Driving in the Fog: Economists on the Challenge of AI Forecasting
A recent survey of hundreds of economists finds deep uncertainty about how artificial intelligence will shape productivity, employment, and growth, with many describing the current moment as forecasting in a fog. Respondents reported limited confidence in their projections, citing fast-moving technology, uneven adoption, and a lack of reliable historical precedents.
What economists are saying
Three threads run through the responses.
- The technology is changing faster than the data used to model it, leaving forecasters working with outdated inputs.
- Adoption varies sharply by sector, which complicates any economy-wide estimate.
- Past technology waves, from electricity to personal computing, took decades to show up in productivity statistics, and there is little agreement on whether AI will follow that pattern or compress it.
Why is AI forecasting so hard right now?
Economic models tend to assume that inputs change slowly. AI has done the opposite. Capabilities that were considered speculative a few years ago are now in routine use, and the cost of running large models has fallen while access has widened. At the same time, surveys of business investment in AI show a wide range of reported spending, partly because companies are still experimenting rather than standardizing deployments.
There is also a measurement problem. Many of the gains that individuals and firms report from AI tools, such as faster drafting, quicker code reviews, or shorter research cycles, do not appear cleanly in official statistics. That makes it harder for economists to convert anecdote into a forecast they can defend.
What signals are economists tracking?
Respondents pointed to a small set of signals they expect will matter most over the next few years:
- Capital expenditure on compute and data centers.
- The share of tasks within an occupation that AI systems can perform reliably.
- Hiring patterns in roles most exposed to automation.
- The rate at which productivity growth revives in industries that adopt AI earliest.
What questions remain unresolved?
Whether AI leads to broad-based productivity gains or concentrates gains in a narrow set of firms is still debated. The effect on wages is similarly unclear, with some respondents expecting compression and others expecting widening. The role of regulation, energy supply, and the cost of compute in shaping the trajectory was also flagged as uncertain.
The survey is a snapshot of expert opinion rather than a prediction. Its main contribution is to make the uncertainty visible, and to remind readers that confidence intervals on AI forecasts are likely to be wide for some time.
FAQ
What did the survey of economists find about AI’s economic impact?
The survey of hundreds of economists found limited confidence in projections about AI’s near-term effects on productivity, employment, and growth, with respondents citing the rapid pace of model development, uneven adoption across industries, and a lack of reliable historical precedents.
Why is forecasting AI’s economic effects so difficult?
Economic models assume inputs change slowly, but AI capabilities have moved from speculative to routine use, and companies are still experimenting rather than standardizing deployments. Many reported gains, such as faster drafting, quicker code reviews, and shorter research cycles, do not show up cleanly in official statistics.
What signals are economists watching for AI’s economic effects?
Respondents identified four signals: capital expenditure on compute and data centers, the share of tasks within an occupation that AI systems can perform reliably, hiring patterns in roles most exposed to automation, and the rate of productivity growth revival in industries that adopt AI earliest.