Dilution, diffusion and symbiosis in spatial prisoner’s dilemma with reinforcement learning

Recent studies on spatial prisoner’s dilemma games with reinforcement learning have shown that static agents can learn to cooperate through a variety of mechanisms, including noise injection, different learning algorithms, and access to neighbours’ payoff information. In this work, we use an independent multi-agent Q-learning algorithm to investigate the effects of dilution and mobility in the spatial version of the prisoner’s dilemma. Within this framework, different possible actions for the algorithm are defined, linking our results to those of the classical, non-reinforcement learning spatial prisoner’s dilemma. This highlights the algorithm’s versatility in modelling diverse game-theoretical scenarios and demonstrates its potential as a benchmarking tool. Our findings reveal a range of effects, including evidence that games with fixed update rules can be qualitatively equivalent to those with learned ones. Additionally, we observe the emergence of a symbiotic mutualistic effect between populations when multiple actions are defined.