Inferring Hidden Motives: Bayesian Models of Preference Learning in Repeated Dictator Games (2511.07825v1)

Abstract: Human cooperation depends on how accurately we infer others' motives--how much they value fairness, generosity, or self-interest from the choices they make. We model that process in binary dictator games, which isolate moral trade-offs between self and other stripped of strategic complexity. Participants observed others' allocation decisions and predicted their future behavior while playing through an exhaustive, randomized payoff space implemented on The Morality Game platform. We formalize social-preference learning as Bayesian belief updating over continuous parameters such as self-interest, altruism, envy, and guilt. The resulting Utility Bayesian Model (UBM) outperformed non-Bayesian alternatives and Bayesian models that categorize others into discrete social types. Because Bayesian updating requires a utility function in its likelihood term, we conducted the largest utility-function comparison to date--476 candidate forms differing in psychologically meaningful properties (e.g., payoff exponents, reference dependence, payoff ratios, and envy-guilt asymmetries). Exploring this joint space of payoffs and models allowed us to identify the function that unifies prior theories and generalizes across payoff conditions. Parameter estimation revealed moderate altruism, strong inequality aversion, and nonlinear payoff valuation (exponent > 1). Altruism and social-comparison motives were largely independent, revealing diverse moral phenotypes from cooperative to competitive or sadistic. Together, these findings provide a computational framework and a map of social motives, clarifying how humans learn whom to trust and offering quantitative foundations for promoting cooperation in social and artificial systems.