Computational cognitive mechanism of stereotype: Social learning and generalization

Abstract: Stereotypes—generalized beliefs that members of a social group tend to possess certain traits—profoundly shape interpersonal interactions and intergroup relations, making it important to understand the cognitive mechanisms underlying their formation and maintenance. Traditional theories, however, have largely remained at the deAt the level of social learning, we first describe how Bayesian structure learning enables the brain to infer latent group categories from observable social data. We then distinguish two pathways through which group–trait associations are established: the experiential pathway, which primarily forms automatic associative representations through RL mechanisms, and the linguistic pathway, which primarily conveys probabilistic propositional representations through Bayesian inference. Regarding the updating and consolidation of these associations, prediction error serves as the core signal driving revision, yet prior biases and asymmetric learning rates cause updating to favor existing beliefs. Furthermore, the explore–exploit dilemma reveals that biased information sampling is a key source of stereotype persistence. At the level of social generalization, we analyze how the brain categorizes novel individuals into known groups on the basis of perceptual similarity and functional similarity, and discuss the retrieval mechanisms through which learned associative knowledge is applied. Finally, we propose three directions for future research: relational cues as a generalization pathway beyond feature similarity, social cognitive maps as an integrative framework for multi-cue representation, and large language model as a tool to simulate stereotype consolidation through linguistic transmission.