Abstract
The precise functions of friendship and the information-processing architecture that evolved to navigate friendships remain unresolved. Research assessing which types of benefits friends provide or which benefits people value most is unlikely to enable researchers to discriminate between alternative evolutionary accounts of friendship, such as reciprocal altruism theory and alliance theory, because these theories generate similar predictions about these features of friendship. We suggest that greater progress toward adjudicating between alternative evolutionary accounts of friendship can be achieved through an increased focus on the underlying information-processing architecture that evolved to secure the functions that friendship evolved to serve. An alliance theory perspective suggests an architecture that hierarchically integrates cues: the mind should track resource sharing and other cues to alliance when information about a friend's alliance cannot be observed directly, but should discount those cues once alliance can be directly observed. By contrast, a reciprocal altruism perspective suggests an additive information-processing architecture: the mind should track resource sharing and other types of benefits a friend might deliver both when a friend's alliance cannot be observed directly and—contrasting alliance theory—when it can be. We outline diverse candidate inputs to alliance-detecting systems that have yet to be systematically empirically investigated, as well as candidate outputs that can reveal the underlying algorithms by which different friend behaviors are processed. We hope that, equipped with these tools, researchers can further illuminate the cognitive architecture of friendship and shed new light on the functions that friendship evolved to solve.