Abstract

Real-world interactions often occur within a broad spectrum of incentives, many of which are often only partially known. Moreover, individuals can each perceive the outcome of such interactions differently. For example, the same loss can be more detrimental for a retail investor, compared to an institutional investor. In the context of Public Goods Games, we explore the impact of such factors on the emergence of cooperation among reinforcement learning agents. First, we introduce uncertainty about the alignment of agents' incentives with those of others. Second, we integrate models of individual agents' risk attitudes, such as risk aversion or risk inclination, represented as preferences over the games' rewards. While uncertainty substantially lowers the agents' ability to engage in cooperative behaviour, despite it being the best course of action, we also observe that different combinations of individual preferences and environmental uncertainties sustain the emergence of cooperative patterns in non-cooperative environments, while others sustain competitive patterns in cooperative environments.