AI systems post 0.135 Brier score versus 0.154 for top human superforecasters on 400 events

The Economist article documents deployment of proprietary models that ingest structured data feeds and historical resolution records to output calibrated probabilities. These systems outperformed the median superforecaster drawn from the Good Judgment Open leaderboard by margins exceeding prior statistical baselines. Integration of retrieval-augmented generation and ensemble weighting produced the recorded accuracy lift.

Tetlock's 2015 superforecasting cohort and subsequent Metaculus platform data establish the human reference distribution. AI performance exceeded the 80th percentile of that distribution on questions with resolution dates between January and December 2025. The gap widened on high-volatility items where base-rate neglect typically degrades human calibration.

Operational implications include substitution of AI probability outputs for human judgment in supply-chain and policy models within 18 months. Organizations already routing forecasts through Metaculus or similar platforms can substitute model scores without retraining staff. Remaining human advantage persists only on questions requiring novel causal structure absent from training corpora.

Next measurable threshold is sustained sub-0.120 Brier performance on out-of-distribution events through 2027.