Discriminators who have limited tolerance for helping dissimilar others are necessary for the evolution of costly cooperation in a one-shot Prisoner’s Dilemma. Existing research reports that trust in societies decreases when agents copy markers and tolerance from more successful others. Global cooperation is possible only in highly homogenized societies. Emergent cooperative societies are not robust against mutant defectors with tolerably similar markers. Our simulation experiments compare one society where each agent has immutable binary tags at the same dimension (‘caste society’) to the other where such markers are distributed across different dimensions (‘modern society’). In both societies, the majority of population display strong parochialism. However, cooperation is significantly stable in modern societies although members are more tolerant than those in caste societies. Modern societies are characterized by loosely coupled small-sized groups with different cultural markers. In terms of efficiency, they achieve higher levels of cooperation than do societies where agents change markers rapidly before defectors attack cooperative clusters.

Theoretically informed by recent computational and mathematical studies highlighting the importance of signals and networks in the evolution of cooperation, the present research undertakes simulation experiments to investigate socio-psychological bases and structural foundations of cooperation as institutional order. Simulated societies consist of three groups of agents with markers and tolerance: altruists, defectors, and parochial egoists. They simultaneously play a one-shot Prisoner's Dilemma game with neighbors by helping tolerably similar ones and by declining to help otherwise. They either imitate markers and tolerance of more successful neighbors or switch partners by breaking the old ones to out-group neighbors after creating new ties to others if both sides accept each other as in-group.

Our study furthers understanding of the dynamics of cooperation in human societies facing the Prisoner's Dilemma: altruists alone are less likely to defeat defectors regardless of network topology as cooperation becomes costly and the average number of interaction partners increases; under these conditions, parochialism is necessary for the institutionalization of cooperation; agents adapt themselves to increasingly homogenized environments by learning intolerance; emergent cooperative societies in the presence of imitation errors are vulnerable to free-riders with tolerably similar markers; either persistent cultural diversity or highly clustered networks with long paths stabilize cooperation; and even if markers are completely mutable, society-wide and robust cooperation is achievable with increased diversity as hierarchical networks of cultural groups self-organize where less tolerant agents in the periphery shield more tolerant ones in the core from emerging free-riders.

The current study provides sociologists with theoretical and methodological resources for a game-theoretical approach to institutional analysis. Socio-psychological research maintains that group identity enhances cooperation either because human subjects in pre-assigned groups maximize group-level rewards or when they expect continuing benefits from in-group favoritism. We rather demonstrate the evolution of cooperation without expectations of in-group reciprocity. Cultural groups emerge as arbitrary markers become salient together with stratification driven by parochial interactions. Our study also contributes to understanding of how stable markets can evolve out of repeated plays of a social dilemma game among parochial agents locally embedded in on-going exchange networks with limited knowledge about costs and opportunities.