Study Finds AI Experts Underestimated Pace of Progress
A study by the Forecasting Research Institute shows that leading AI experts consistently underestimated how quickly the field would advance, with milestones like gold-medal performance at the International Mathematical Olympiad arriving five years earlier than expected.
Leading artificial intelligence experts have consistently underestimated the speed of progress in their own field, according to a study by the Forecasting Research Institute. The findings suggest that even those closest to the technology have failed to anticipate how quickly new capabilities would arrive, raising questions about how society prepares for rapid change.
The study points to several concrete examples where reality outpaced expert predictions. AI systems reached gold-medal level at the International Mathematical Olympiad five years ahead of the median expert forecast. In another striking miss, the annualized revenue of the AI company Anthropic is about five times what experts had predicted. These gaps indicate that expert intuition about the pace of AI development may be systematically biased toward caution.
But the picture is not uniformly one of underestimation. Forecasts for real-world applications such as self-driving cars have painted a more mixed picture, with some predictions proving overly optimistic rather than too conservative. That nuance matters: it suggests that expert misjudgment is not simply a matter of always expecting too little, but of struggling to gauge which capabilities will accelerate and which will stall.
The findings carry implications beyond the research community. If experts themselves cannot reliably forecast the trajectory of AI, then policymakers, educators, and business leaders face even greater uncertainty as they plan for the future. The study does not offer a simple remedy, but it underscores the value of updating expectations frequently rather than relying on long-range predictions made in earlier eras of the technology.
The Forecasting Research Institute’s work adds to a growing body of evidence that the development of AI is not following a smooth, predictable curve. Instead, progress appears to arrive in bursts that defy conventional forecasting methods. For a field that increasingly touches on economics, security, and everyday life, that unpredictability is itself a key finding.
3
