Expert disagreement drives AI forecasts

A major multi‑author survey finds that disagreement about how fast AI capabilities will advance, not differences in modeled scenarios, explains most of the variation in economic projections; only ~5.2% of forecast variance is tied to scenario choice, implying the single biggest lever is settling capability expectations. That makes efforts to better measure and forecast AI capability growth — not just policy levers — central to credible economic planning. — If forecast divergence mainly reflects uncertainty about AI capabilities, public policy should focus on capability monitoring and contingency planning rather than fixed bets about outcomes.