The evolution of cooperation in multi-games with reinforcement learning

The multi-games have been shown to facilitate the emergence of cooperative behavior to some extent. However, agents in this model merely imitate their neighbors' strategies, neglecting interactions with the environment. To enhance this interaction, we integrate reinforcement learning techniques into the multi-games. Specifically, each agent's state is defined by the number of neighboring agents that choose to cooperate, and social payoff is incorporated into the Q-table updates. Numerical simulations show that with social payoff included, payoff matrix diversity contributes virtually nothing to boosting cooperation rates. This is because social payoff provides cooperators with extra compensation independent of the “sucker's payoff”, neutralizing its role in facilitating cooperation diffusion and thus becoming the key intrinsic driver of cooperative dominance. We then analyzed in detail the intrinsic mechanism behind the significant improvement in cooperation, aiming to robustly support the synergistic development of theory and practice in this field.