AI trading bots now beat human forecasters on prediction markets

Researchers built a system that deploys 50 different AI models to predict binary outcomes on decentralized betting markets, then combines their predictions using statistical methods to estimate probabilities and place trades. The system outperforms human expert forecasters on calibration metrics, which means its confidence levels are better aligned with actual outcomes.

This is the first documented case of AI systems systematically beating human superforecasters at real-money prediction markets using live market data. Until now, AI performance benchmarks lived in controlled settings against synthetic tasks. Here the AI is making actual bets against human experts in a domain where overconfidence is punished with real losses. The implication is straightforward: if AI can beat humans at probability estimation in markets, it can probably beat them at other domains where calibration and risk-adjusted judgment matter — medical diagnosis, safety analysis, strategic forecasting.