How Past Experiences Shape New Beliefs: Generalisation and Adaptation in Human Reinforcement Learning

Humans are adept at learning under uncertainty. However, the way in which we learn must adapt to different environments, and we must determine when to generalise learning strategies across different contexts, making use of prior experience to guide present learning. However, how learning adapts based on prior experience remains poorly understood. Here, we examine how prior exposure to environments with differing levels of uncertainty shapes subsequent learning. A large sample recruited online (N=330) completed a gamified two-armed bandit task comprising two blocks with varying uncertainty, manipulated through the volatility or changeability of outcome contingencies. Computational modelling enabled the extraction of learning rate parameters quantifying the speed with which participants update their expectations. Analyses revealed significant effects of both previously experienced and current block volatility learning rates, and that this combination of generalisation and adaptation was consistent across learning from positive and negative prediction errors. Exploratory regression and factor analyses did not identify any moderating effects of mental health symptoms. Our results demonstrate that prior experience biases belief updating strategies across environments, revealing a mechanism through which learning both adapts and generalises to facilitate effective behaviour.