New 'Evidence Markets' Paper Proposes AI-Powered Crowd Wisdom with LLM Verification
Prediction markets get a major upgrade: now submit evidence, not just bets, for smarter forecasts.
Evidence markets, introduced in a new paper, generalize prediction markets to incentivize evidence submission alongside beliefs. Using a logarithmic market scoring rule with dynamic liquidity tied to evidence quality, the system rewards evidence proportional to uncertainty and can resolve endogenously via crowdsourced evidence. Platform loss is bounded, and the mechanism can be implemented via an automated market maker. The paper proves that truthful belief and evidence reporting is always an ε-dominant strategy incentive compatible (DSIC) strategy. It also proposes evidence verification using an LLM-as-a-Judge framework with staking and an asynchronous execution algorithm. LLM evaluations serve as the paper's running example.
- Evidence markets allow participants to submit evidence and reasoning, not just binary bets, unlocking deeper crowd intelligence.
- The system uses a dynamic liquidity parameter that adjusts based on evidence quality, ensuring bounded platform loss and proportional rewards.
- LLM-as-a-Judge with staking verifies submissions, and the mechanism is proven ε-DSIC for truthful reporting.
Why It Matters
Crowdsourced intelligence gets a rigorous, AI-verified upgrade—useful for model evaluation, consensus, and subjective forecasting.