Rational arbitration between statistics and rules in human sequence learning

Detecting and learning temporal regularities is essential to accurately predict the future. Past research indicates that humans are sensitive to two types of sequential regularities: deterministic rules, which afford sure predictions, and statistical biases, which govern the probabilities of individual items and their transitions. How does the human brain arbitrate between those two types? We used finger tracking to continuously monitor the online build-up of evidence, confidence, false alarms and changes-of-mind during sequence learning. All these aspects of behaviour conformed tightly to a hierarchical Bayesian inference model with distinct hypothesis spaces for statistics and rules, yet linked by a single probabilistic currency. Alternative models based either on a single statistical mechanism or on two non-commensurable systems were rejected. Our results indicate that a hierarchical Bayesian inference mechanism, capable of operating over several distinct hypothesis spaces, underlies the human capability to learn both statistics and rules.