The Forecasting Research Institute (FRI) released an interim report analyzing expert predictions on AI development. The study involved 339 specialists, including 76 computer scientists, 76 industry experts, 68 economists, and 119 policy analysts, alongside a group of superforecasters.
AI reached gold-medal performance at the International Mathematical Olympiad in July 2025, five to ten years ahead of expert and superforecaster predictions made in 2022. Experts assigned a 24.6 percent probability to these results, while superforecasters estimated 9.7 percent.
In virology, experts and superforecasters predicted AI would match top teams on troubleshooting benchmarks by 2030 or 2034. FRI data suggests this milestone was likely achieved in April 2025. Similar underestimates were observed in cybersecurity and economic revenue metrics.
Forecasts for the highest annual recurring revenue of an AI company by the end of 2026 ranged from $16 billion to $25 billion. However, reports indicate Anthropic reached $100 billion in September 2026, while OpenAI reached $65 billion in July 2026.
Not all predictions were conservative. Experts overestimated the success rate of language models in biological lab tasks and the market share of autonomous ride-hailing trips in the US for 2027. Respondents are now revising their expectations for AI's societal impact upward.
FRI plans to adjust its methodology by highlighting respondents who anticipate rapid progress through 2040. The institute will also integrate continuously updated LLM forecasts alongside human predictions to improve accuracy as the field evolves.
