Influence of LLMs on indirect reciprocity, human prosociality, and human–AI cooperation

Determine the influence of large language models on indirect reciprocity mechanisms, human prosocial behavior, and human–AI cooperation, including how LLM-driven reputation judgments affect cooperation rates and punishment patterns.

Background

Indirect reciprocity (IR) relies on observers’ judgments to assign reputations that guide future cooperative decisions. As LLMs increasingly mediate or advise such judgments, their embedded social norms and biases could reshape reputation dynamics and thereby cooperation.

The authors extract second-order social norms from multiple LLMs and analyze their cooperative consequences in populations under public and private reputation regimes. Despite these concrete analyses, they explicitly note that the general influence of LLMs on IR, prosociality, and eventual human–AI cooperation is not yet clear, motivating this open problem.