Emergent Cooperation and Deception in Public Good Games

Communication is a widely used mechanism to promote cooperation in multi-agent systems. In the field of emergent communication agents are usually trained on a particular type of environment: cooperative, competitive, or mixed-motive. Motivated by the idea that real-world settings are characterised by incomplete information and that humans face daily interactions under a wide spectrum of incentives, we hypothesise that emergent communication could be simultaneously exploited in the totality of these scenarios.
In this work we pursue this line of research by focusing on social dilemmas, and develop an extended version of the Public Goods Game which allows us to train independent reinforcement learning agents simultaneously on different scenarios where incentives are aligned (or misaligned) to various extents.
Additionally, we introduce uncertainty regarding the alignment of incentives, and we equip agents with the ability to learn a communication policy, to study the potential of emergent communication for overcoming uncertainty.
We show that in settings where all agents have the same level of uncertainty, communication can help improve the cooperation level of the system, while, when uncertainty is asymmetric, certain agents learn to use communication to deceive and exploit their uncertain peers.