Aitchison, MatthewBenke, LyndonSweetser, Penny2025-12-172025-12-1797830309177841865-0929ORCID:/0000-0002-6543-557X/work/161119694https://hdl.handle.net/1885/733795932Deception is prevalent in human social settings. However, studies into the effect of deception on reinforcement learning algorithms have been limited to simplistic settings, restricting their applicability to complex real-world problems. This paper addresses this by introducing a new mixed competitive-cooperative multi-agent reinforcement learning (MARL) environment, inspired by popular role-based deception games such as Werewolf, Avalon, and Among Us. The environment’s unique challenge lies in the necessity to cooperate with other agents despite not knowing if they are friend or foe. Furthermore, we introduce a model of deception which we call Bayesian belief manipulation (BBM) and demonstrate its effectiveness at deceiving other agents in this environment, while also increasing the deceiving agent’s performance.Acknowledgements. This initiative was funded by the Department of Defence and the Office of National Intelligence under the AI for Decision Making Program, delivered in partnership with the NSW Defence Innovation Network.21enPublisher Copyright: © 2021, Springer Nature Switzerland AG.Bayesian beliefDeceptionDeep reinforcement learningIntrinsic motivationMachine learning202110.1007/978-3-030-91779-1_585121909505