AI Forecasters Now Match the World's Best Human Predictors

Scaffolded AI systems are matching and occasionally beating elite human superforecasters in prediction tournaments, financial markets, and geopolitical forecasting — a milestone experts expected for 2026-2027.

AI systems purpose-built for forecasting have reached a turning point. At this year's prediction market conference, the dominant story was not the industry's growth into a multi-billion dollar market, but the arrival of AI superforecasters — scaffolded frontier models that research, reason, and assign probabilities to future events with accuracy rivaling the best human forecasters on the planet.

The numbers are striking. FutureSearch, a startup building forecasting-focused AI scaffolds, turned $35 into $2 million on the Kalshi prediction market over seven months. Another company, Preseen, claims to have multiplied seed capital by roughly 100,000 times on the same platform. A third firm reports beating the stock market by 25% with a market-neutral portfolio while posting similar margins on Kalshi and Polymarket.

On the tournament circuit, the evidence is converging. In the most recent Metaculus Cup — the World Cup of forecasting — human forecasters took the top two spots, but Preseen's AI placed third. In the ongoing summer tournament, two top-ten AIs from the spring season have repeated their performance, placing alongside the two top-ten humans who also held their rank. The gap, if it exists at all, has narrowed to a statistical dead heat.

The specialized scaffolding makes the difference. Base models like GPT and Claude score well below professional superforecasters on Metaculus. But AI systems wrapped in forecasting-specific workflows — deploying subagents, reading dozens of sources, constructing probabilistic reasoning chains — perform roughly nine months ahead of raw base models and are now approaching parity with the world's elite. The Metaculus community currently gives a 15% chance that a bot wins the current summer tournament and a 95% chance that one wins before 2030.

What changes when AI forecasters surpass humans entirely? The most immediate transformations will come in finance, where AI's speed and diligence allow it to bet on hundreds of markets simultaneously while human analysts research just one. But the broader promise lies in democratizing access to high-quality forecasting. Instead of paying tens of thousands of dollars and waiting weeks for a human superforecaster's opinion, anyone could query an AI for a probability on a policy proposal, a business strategy, or a geopolitical risk — in minutes, for pocket change.

Sources: Astral Codex Ten