Reinforcement learning in social interaction: The distinguishing role of trait inference

Abstract

People learn about the world by making choices and experiencing feedback—a process characterized by models of reinforcement learning in which people learn to associate their actions with rewarding outcomes. Although reinforcement models provide compelling accounts of feedback-based learning in nonsocial contexts, social interactions typically involve inferences of others' trait characteristics, which may be independent of their reward value. As a result, people may learn differently about humans and nonhumans through reinforcement. In two experiments (and a pilot study), participants interacted with human partners or slot machines that shared money. Computational modeling of behavior revealed different patterns of learning for humans and non-humans: participants relied more on feedback indicating trait generosity (relative to monetary reward) when learning about humans but relied more on monetary reward (relative to generosity) when learning about slots. Furthermore, this pattern of learning had implications for attitudes: whereas participants preferred generous humans, they preferred rewarding slot machines, relative to their respective counterparts. These findings reveal a distinct role for reinforcement learning in social cognition, showing that humans preferentially form abstract trait inferences about other people through feedback in addition to reward associations.