Engineering Optimal Cooperation Levels with Prosocial Autonomous Agents in Hybrid Human-Agent Populations: An Agent-Based Modeling Approach

The evolution of cooperation in social interactions remains a central topic in interdisciplinary research, often drawing debates on altruistic versus self-regarding preferences. Moving beyond these debates, this study investigates how autonomous agents (AAs) with a range of social preferences interact with human players in one-shot, anonymous prisoner’s dilemma games. We explore whether AAs, programmed with preferences that vary from self-interest to other-regarding behavior, can foster increased cooperation among humans. To do this, we have refined the traditional Bush–Mosteller reinforcement learning algorithm to integrate these diverse social preferences, thereby shaping the AAs’ strategic behavior. Our results indicate that even a minority presence of AAs, programmed with a moderate aspiration level, can significantly elevate cooperation levels among human participants in well-mixed situations. However, the structure of the population is a critical factor: we observe increased cooperation in well-mixed populations when imitation strength is weak, whereas networked populations maintain enhanced cooperation irrespective of the strength of imitation. Interestingly, the degree to which AAs promote cooperation is modulated by their social preferences. AAs with pro-social preferences, which balance their own payoffs with those of their opponents, foster the highest levels of cooperation. Conversely, AAs with either extremely altruistic or purely individualistic preferences tend to hinder cooperative dynamics. This research provides valuable insights into the potential of advanced AAs to steer social dilemmas toward more cooperative outcomes. It presents a computational perspective for exploring the complex interplay between social preferences and cooperation, potentially guiding the development of AAs to improve cooperative efforts in human societies.