Figure 1.

Illustration where a person intends to predict whether it will snow on Thursday. On Sunday night, the person watches a weather forecast, leading to the formation of a prior belief. During the next 3 d, the person observes factual weather information. On Monday and Tuesday it snows, but Wednesday is sunny. To make a prediction on Wednesday night, it is adaptive to take into account both the subjective prior information gathered from TV and the factual observations. It is unknown how the neural representation of this forecast integrates the prior knowledge with the actual frequency of experienced outcomes. Bayesian probabilities combine prior knowledge and empirical observations in an optimal way, whereas frequencies only depend on empirical observations.