Bayesian Truth Serum

A mechanism that elicits truthful subjective opinions by rewarding predictions that align with unexpectedly common beliefs.

What is Bayesian Truth Serum?

The Bayesian Truth Serum (BTS), developed by Drazen Prelec, is an information elicitation mechanism that encourages truthful reporting of subjective opinions by rewarding participants whose predictions align with beliefs that are “surprisingly popular” compared to what others expect. As Scott Kominers explains in the transcript, BTS asks participants for their own belief (e.g., did you like a movie?) and their estimate of others’ beliefs, using discrepancies to identify truthful answers. If a participant’s belief matches an unexpectedly common response, they are rewarded, assuming they’ve provided unique, accurate insight.

For instance, the transcript cites a study by Prelec, Seung, and McCoy, where BTS outperformed simple crowd polling. When asked the capital of Pennsylvania, most might guess Philadelphia (incorrect) and expect others to agree, but a minority correctly choosing Harrisburg would score higher due to its surprising popularity. This method excels in subjective or low-information settings, where prediction markets may falter due to thin participation or lack of objective outcomes. Unlike prediction markets, BTS doesn’t require trading digital assets or event resolution, making it simpler to implement.

BTS is valuable for market research or community feedback, where subjective data is critical, and can be enhanced by blockchain for transparent, auditable scoring. However, it’s less effective for objective events like elections, where prediction markets’ financial incentives and thick markets provide stronger forecasts, as seen in the 2024 election’s accurate predictions on platforms like Polymarket.