Enhancing cooperation in dynamic networks through reinforcement-learning-based rewiring strategies

Abstract

Cooperation is a fundamental aspect of social and biological systems, yet achieving and maintaining high levels of cooperation remains a significant challenge. This study investigates the dynamics of cooperation among players engaged in repeated two-player Prisoner’s Dilemma games, utilizing a novel integration of the Bush–Mosteller reinforcement learning model with adaptive network rewiring mechanisms. Each player updates its probability of cooperation and rewires its connections based on the payoffs received from neighbors. Our results demonstrate that incorporating network rewiring guided by reinforcement learning significantly enhances both the level of cooperation and the average payoff across the population. Players that prioritize rewiring over strategy updates are found to form more stable cooperative structures, while those with heightened sensitivity to payoffs and optimal aspiration levels achieve greater cooperation. By identifying and analyzing key parameters that influence cooperative dynamics, our findings provide deep insights into the mechanisms that drive cooperative behavior. This research not only highlights the transformative potential of adaptive network rewiring in promoting cooperation within complex adaptive systems but also offers a framework for designing resilient cooperative networks across diverse domains.

Keywords

• reinforcement learning
• Prisoner’s Dilemma
• cooperation dynamics
• network rewiring
• social networks