CONCEPT
AI Forecasting
The discipline of predicting when specific AI capabilities will arrive — a domain where Clarke's First Law applies cleanly: the distinguished elderly scientist who says X is impossible is, on the historical pattern, very probably wrong.
AI forecasting is the attempt to predict, with calibrated confidence, when specific AI capabilities will be achieved and when their societal effects will be felt. It has produced, in the last decade, a body of empirical work (Metaculus forecasts, expert surveys, scaling-law extrapolations, automated-evaluation time-series) that is more rigorous than the informal commentary which preceded it and still less reliable than one would want for decisions of the stakes now being taken. The key finding from the literature is Clarke's: expert forecasts of the impossible are systematically unreliable on the downside, while expert forecasts of the possible are systematically unreliable on the upside of timeline length.
In The You On AI Field Guide
The historical record of AI forecasting is a mixed one. Expert surveys in the 1960s predicted AGI in a decade; surveys in the 1980s, after the second AI winter, predicted it in centuries; surveys in the 2010s converged on a median of thirty to fifty