Under the framework of evolutionary games, how to update strategies when individuals interact with multiple neighbors at the same time is a key problem to understand the evolution of cooperation. At present, a widely known method is imitation updating. However, it is undeniable that when resources or capabilities are limited, it is difficult for individuals to obtain more comprehensive information about their counterparts, resulting in imitation method no longer applicable. Reinforcement learning algorithms attract attention because players do not need to know information such as the opponent's payoff, and only adjust strategies based on their own experience and expectations. In view of this, we consider both imitation and reinforcement learning as two behavioral adjustment models, and explore the evolution of cooperative behavior in mixed populations of extrinsic imitators and intrinsic learners. Numerous simulations have shown that when the value of temptation to defect is small, there exists an optimal proportion of intrinsic learners bring group cooperation to its peak. However, in interactive environments with larger social dilemmas, intrinsic learners are effective in resisting betrayal and maintaining group behavior compared to traditional extrinsic imitators. The proposed model reflects the characteristics of seeking advantages, avoiding disadvantages, and imitation in biological groups, and provides new ideas for the design of new game models.
Read full abstract