Subjective-Probability Forecasts of Existential Risk: Initial Results from a Hybrid Persuasion-Forecasting Tournament

This article presents a multi-stage forecasting tournament designed to evaluate how experts and generalist superforecasters assess short- and long-term existential risks to humanity, with a focus on artificial intelligence. The study finds that specialists consistently assign higher probabilities to catastrophic and existential threats than generalists, especially for long-term AI risks. Despite structured debate and incentives for persuasion, participants showed little change in views, suggesting entrenched priors and cognitive biases. The findings underscore both the potential and limitations of forecasting methods in resolving speculative, high-stakes debates and point to the value of anonymity and improved deliberative design in future forecasting efforts.