No abstract

No abstract

Making adaptive decisions often requires inferring unobservable states based on unreliable information. Bayesian logic prescribes that individuals form probabilistic beliefs about a state by integrating the likelihood of new evidence with their prior beliefs, but human neuroimaging studies on probability representations have not typically examined this integration process. We developed an inference fMRI task in which participants estimated the posterior probability of a hidden state while we parametrically modulated the prior probability of the state, the likelihood of the supporting evidence, and a monetary penalty for estimation inaccuracy. Consistent with a neural substrate for Bayesian integration, activation in left posterior parietal cortex tracked the estimated posterior probability of the solicited state and its components of prior probability and likelihood, all independently of expected value. This activation further reflected deviations in individual reports from objective probabilities. Thus, this region may provide a neural substrate for humans’ ability to approximate Bayesian inference.