Reputation as a new route to cooperation in multi-agent reinforcement learning
Artificial agents are likely to face the same dilemmas of cooperation that humans evolved to solve. To design efficient systems where collective gains are maximised, artificial agents must learn to forego their self-interest and spend effort to help others. How to design adaptive agents that autonomously learn to cooperate? In this project we study how to use reputation systems to sustain cooperation among agents adapting through trial-and-error (that is, through reinforcement learning). We will explore how reputations should be assigned to sustain cooperation in the long-run, and how to design reputation systems in increasingly complex environments.

NWO M-grants are intended for realising curiosity-driven, fundamental research of high quality and/or scientific urgency. The M-grant offers researchers the possibility to elaborate creative and risky ideas and to realise scientific innovations that can form the basis for the research themes of the future.