Antisocial learning

Abstract I study a model of social learning within a social dilemma. Some agents independently learn how to profit from a socially harmful action, the rest can only infer this information from public behavior. In all equilibria, if the rate of independent learning is sufficiently low, then beliefs do not converge. Yet, this is asymptotically efficient: almost all agents cooperate. In this case, reducing the publicity of actions can increase social learning and reduce cooperation. However, if the rate of independent learning is high, then beliefs converge to the truth and almost nobody cooperates. Here, allowing limited privacy can increase cooperation.