EVENT
Existential Risk Persuasion Tournament
Tetlock's application of adversarial collaboration to AI existential risk — pairing domain experts with superforecasters to test whether structured debate changes probability estimates of catastrophe.
The Existential Risk Persuasion Tournament, organized by Tetlock and collaborators, brought together AI domain experts and superforecasters in structured adversarial dialogues about the probability of AI-caused catastrophe or extinction by 2100. AI experts assigned a median twelve-percent probability of catastrophe and 3.9-percent probability of extinction; superforecasters assigned 2.31-percent catastrophe and 0.38-percent extinction. Neither group was able to persuade the other to substantially revise their long-term estimates, suggesting disagreement was not primarily about evidence but about the weighting of inside view (AI's unique features) versus outside view (base rates for technological catastrophe). A September 2025 follow-up revealed both groups had radically underestimated near-term AI progress: superforecasters assigned only 9.7-percent probability to benchmark achievements that actually occurred.
In The You On AI Field Guide
The tournament structure was designed to test a specific hypothesis: that adversarial collaboration — bringing together people with opposing views in a framework requiring them to engage each other's strongest arguments — would produce belief convergence toward a more accurate aggregate estimate. The