Reinforcement learning in evolutionary game theory: A brief review of recent developments

With the rapid progress of artificial intelligence, the integration of evolutionary game theory and reinforcement learning has become a hot research frontier in the last years. Evolutionary game theory provides a mathematical framework for depicting the strategy interaction among individuals, traditionally based on pre-defined, rule-based strategy update protocols. In contrast, reinforcement learning enables agents to adaptively select optimal actions through trial-and-error learning, hence better reflecting real-world decision-making. These complementary features create the foundation for their convergence. Our paper presents a didactic review of contemporary reinforcement learning applications in evolutionary game theory, focusing on those recently published works which open novel research paths to enrich our understanding of mutualistic cooperation. We summarize major concepts and terms, including the basic problem of collective cooperation, modeling of complex population dynamics, influence of algorithmic parameters, and the combination of deep learning. Finally, we discuss prospects for this interdisciplinary field, emphasizing the importance of intelligent learning through the lens of evolutionary game.