Forecasters find top AI experts kept underestimating AI progress
Expert forecasts ran systematically conservative, but the report admits its method favors catching underestimates only.
Original event 2026-09-24
An interim report from the Forecasting Research Institute (FRI) says leading AI experts have consistently underestimated recent AI progress.
The report pools forecasts from 339 experts collected since 2022: AI reached gold-medal level at the International Mathematical Olympiad in July 2025, five years before the median expert forecast, and experts assigned an average probability of only 24.6 percent to the benchmark results that actually happened. Economic forecasts missed wider: the median expert estimate for the highest AI-company annual revenue at end-2026 was $20 billion, while FRI cites roughly $100 billion annualized for Anthropic in September 2026.
The report flags its own bias: underestimates become obvious as soon as reality overtakes a prediction, while overestimates only show once a deadline passes, so an interim report is naturally tilted toward finding forecasters too cautious.