In a September 23, 2026 interim update on four years of studies, Research_FRI says experts—including top economists, computer scientists and biologists—dramatically underestimated AI capability progress each year. Superforecasters underestimated it even more.

The group reports a mixed record on AI diffusion: forecasters substantially underestimated AI revenue, while predictions of AI’s share of electricity use remain on track to be accurate. They overestimated how much mid-2025 models could help amateurs with biorisk-relevant laboratory tasks and how quickly self-driving cars would roll out.

Research_FRI says it is still too early to judge forecasts of broader impacts, including GDP growth, major AI harms and deaths averted from disease.