Updates in Cooperative AI

Indirect reciprocity is a key mechanism for sustaining cooperation in mixed-motive settings. By helping someone, agents can enhance their reputation, increasing the chances that others reciprocate and cooperate back with them in the future. The reputation of individuals depends, in turn, on social norms that define a good or bad action, offering a mathematically appealing way of studying the emergence of moral systems. In this presentation, Fernando will introduce indirect reciprocity models with increasing complexity. First, a model based on evolutionary game theory, to understand cooperation dynamics under different social norms. Second, a model of (independent) reinforcement learning, where agents belong to multiple groups and might learn to reciprocate exclusively with their own group. Third, he will discuss the effects of having reputations assigned by LLM-based agents. He will show that indirect reciprocity norms revealed by LLMs exhibit high variance, within and between models, which significantly impacts the dynamics and prevalence of cooperation.

We’re delighted to host this ninth seminar in our 'Updates in Cooperative AI' series. You're welcome to subscribe to our Google Calendar to stay up-to-date on all upcoming events.