Resolving Complex Social Dilemmas by Aligning Preferences with Counterfactual Regret
Keywords: Counterfacutla Regret, Sequential Social Dilemma
Abstract: Social dilemmas are situations where gains from cooperation are possible but misaligned incentives make it hard to find and stabilize prosocial joint behavior. In such situations selfish behaviors may harm the social good. In spatiotemporally complex social dilemmas, the barriers to cooperation that emerge from misaligned incentives interact with obstacles that stem from spatiotemporal complexity. In this paper, we propose a multi-agent reinforcement learning algorithm which aims to find cooperative resolutions for such complex social dilemmas. Agents maximize their own interests while also helping others, regardless of the actions their co-players take. This approach disentangles the causes of selfish reward from the causes of prosocial reward. Empirically, our method outperforms multiple baseline methods in several complex social dilemma environments.
Primary Area: reinforcement learning
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 3056
Loading