Effects of value-driven social learning on cooperation in the prisoner’s dilemma games

Despite the growing attention and research on the impact of Q-learning-based strategy updating on the evolution of cooperation, the joint role of individual learners and social learners in evolutionary games has seldom been considered. Here, we propose a value-driven social learning model that incorporates a shape parameter, β, to characterize the degree of radicalism or conservatism in social learning. Using the prisoner’s dilemma game on a square lattice as a paradigm, our simulation results show that the cooperation level has a non-trivial dependence of β, density ρ, and dilemma strength b. We find that both β and ρ have nonmonotonic effects on cooperation; specifically, moderate levels of radicalism in social learning can facilitate cooperation remarkably, and when slightly conservative, can form a favorable cooperation region with the appropriate ρ. Moreover, we have demonstrated that social learners play a key role in the formation of network reciprocity, whereas individual learners play a dual role of support and exploitation. Our results reveal a critical balance between individual learning and social learning that can maximize cooperation and provide insights into understanding the collective behavior in multi-agent systems.