Identifying social partners through indirect prosociality: A computational account.

The ability to identify people who are prosocial, supportive, and mindful of others is critical for choosing social partners. While past work has emphasized the information value of direct social interactions (such as watching someone help or hinder others), social tendencies can also be inferred from indirect evidence, such as how an agent considers others when making personal choices. Here we present a computational model of this capacity, grounded in a Bayesian framework for action understanding. Across four experiments we show that this model captures how people infer social preferences based on how agents act when their choices indirectly impact others (Experiments 1a, 1b, & 1c), and how people infer what an agent knows about others from knowledge of that agent's social preferences (Experiment 2). Critically, people's patterns of inferences could not be explained by simpler alternatives. These findings illuminate how people can discern potential social partners from indirect evidence of their prosociality, thus deepening our understanding of partner detection, and social cognition more broadly.