Artificial Intelligence
Study finds experts underestimated AI progress
An interim report from the Forecasting Research Institute said specialists in academia, industry and economics repeatedly predicted slower AI progress than later benchmark results and adoption figures showed.
- The Forecasting Research Institute said expert forecasts on AI progress were often too low in its interim report.
- AI reached gold-medal level at the International Mathematical Olympiad in July 2025, five years ahead of the median expert forecast and ten years ahead of the median superforecaster forecast.
- FRI said AI may have solved a Millennium Prize Problem, while experts in an August and September 2025 survey put the median chance of that by the end of 2027 at 10 per cent.
- FRI cited roughly $100 billion in annual recurring revenue for Anthropic in September 2026, against a median end-2026 forecast of $20 billion from experts.
Since mid-2022, the Forecasting Research Institute had collected AI forecasts across several studies and projects. Its first Longitudinal Expert AI Panel included 339 people, among them computer scientists, industry experts, economists and AI policy specialists.
FRI said the largest gap was in mathematics. AI reached gold-medal level at the International Mathematical Olympiad in July 2025, which it said was five years ahead of the median expert forecast and ten years ahead of the median superforecaster forecast.
The institute also said AI may have solved a Millennium Prize Problem, although it was not yet clear whether the result met the evaluation rules. In virology and cybersecurity studies, FRI said experts had also predicted later progress than what it judged had already happened.
FRI said economic forecasts were also conservative, and it gave Anthropic’s September 2026 annualised revenue as about $100 billion. It said forecasts for autonomous ride-hailing, economic growth, employment and major AI harms could not yet be judged reliably, and it planned to publish updated LLM forecasts alongside human ones.
Named in this story
Companies
- Anthropic
- was cited for about $100 billion in annualised revenue
Organisations
- Forecasting Research Institute
- issued the interim report on AI forecasting
- Longitudinal Expert AI Panel
- provided the first round of expert forecasts
- International Mathematical Olympiad
- the contest where AI reached gold-medal level
Products and systems
- ChatGPT
- marked the launch point after which the forecast pattern still held
- ForecastBench
- was cited as a benchmark where some models matched superforecasters on some question types
How the source tells it
Mostly reportorial and restrained, with mild urgency created by repeated contrasts between forecasts and earlier-than-expected AI milestones.
- urgency the article repeatedly set recent AI results against forecasts made years earlier