Prediction markets are still mostly football bets — can AI fix them?

Prediction markets were envisioned by Robin Hanson and Vitalik Buterin as tools to aggregate information and improve democracy. Yet today, Kalshi and Polymarket — the two leading platforms — devote roughly 65% of their volume to sports betting, with crypto and politics trailing at 12% each. Only about 1.2% of trades occur in STEM or other high-value categories. This perverse outcome stems from a supply-and-demand problem: markets need subsidizers to create them and informed traders (sharps) to push prices toward truth. But in practice, gamblers and entertainment-seeking users dominate, turning prediction markets into de facto casinos that regulators are increasingly targeting (e.g., the “Prediction Markets Are Gambling Act”).

The article argues that AI could break this cycle by automating market creation, improving information aggregation, and attracting a new class of automated savers and hedgers. AI agents could act as sharps, analyzing vast datasets to place accurate bets, thereby crowding out uninformed gamblers and making markets more reliable for governance and peer review. If successful, AI could transform prediction markets from a sports-betting gimmick into a genuine “truth machine” for society — but only if the incentive structures are redesigned to reward accuracy over entertainment.