Abstract

To model strategic interactions standard game theory assumes that agents have common knowledge of rationality. This allows agents to use deduction to form expectations on the behavior of their counterparts. In coordination games however, deduction fundamentally fails to prescribe a unique solution to agents, raising a “matching” problem in game theory. The question is, when deduction is of no use, how are agents to match or decouple their choices? The thesis explored here is inspired from the recent finding that humans recruit the same neural structures to reason about themselves and similar but not dissimilar others, and of friends but not strangers; a finding which has led some investigators to speak of self-referential mentalizing. This meshes nicely with the widely-established cross-species observation that social beings usually exhibit a preference for similar others; as well as with the well-known observation that social closeness fosters cooperation. However, as investigated by nearly all previous experiments, cooperation critically required agents to match their choices. My work adds to the experimental literature in several respects: i) it shows that both objective social closeness (friendship) and psychological closeness (perceived lab-induced similarity) can have an opposite effect on strategic interactions, depending on whether they require to match or decouple choices; ii) that this behavior is best explained by synergistic contributions of expected reciprocity and altruism, which we show to be dissociable both in behavior and the brain. From a neural perspective, expected reciprocity relies on the ventromedial prefrontal cortex, an area previously implicated in reward, interpersonal similarity and depth of reasoning; while the temporo-parietal junction is particularly important for altruism. Taken together, our results provide novel insight into the neuro-cognitive mechanisms that facilitate social cohesion.