Integrated Altruistic and Fairness Preference Induces Advanced Mutual Cooperation in Sequential Social Dilemmas

Jul 16, 2026

Upload by Noah

Inducing cooperation among distributed agents remains a central challenge in multi-agent reinforcement learning, especially in sequential social dilemma settings where self-interest conflicts with collective welfare. Humans achieve cooperation through social preferences, inspiring a new approach that integrates altruistic preferences and fairness preferences into a single AFP utility and reward-sharing mechanism. Comparative experiments against standard RL and inequity aversion agents show AFP achieves mutual cooperation with stronger collective rewards and improved equity. Training analysis further indicates altruism promotes public-good contribution while fairness shapes mutual behavior.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages9

Reading Time23 min

Abstract

Introduction

Cooperation in MARL and social dilemmas

Reward design challenges

Limitations of centralized training

Preference-based mechanisms for cooperation

Integrated Altruistic and Fairness Preference Induces Advanced Mutual Cooperation in Sequential Social Dilemmas Yu Wei The University of Tokyo Tokyo, Japan [weiyu@isi.imi.i.u-tokyo.ac.jp](weiyu@isi.imi.i.u-tokyo.ac.jp) Yukiko Ogura The University of Tokyo Tokyo, Japan [ogura@isi.imi.i.u-tokyo.ac.jp](ogura@isi.imi.i.u-tokyo.ac.jp) Yoshiyuki Ohmura The University of Tokyo Tokyo, Japan [ohmura@isi.imi.i.u-tokyo.ac.jp](ohmura@isi.imi.i.u-tokyo.ac.jp) Ildefons Magrans de Abril The University of Tokyo Tokyo, Japan [ildefons@isi.imi.i.u-tokyo.ac.jp](ildefons@isi.imi.i.u-tokyo.ac.jp) Hoshinori Kanazawa The University of Tokyo Tokyo, Japan [kanazawa@isi.imi.i.u-tokyo.ac.jp](kanazawa@isi.imi.i.u-tokyo.ac.jp) Yasuo Kuniyoshi The University of Tokyo Tokyo, Japan [kuniyosh@isi.imi.i.u-tokyo.ac.jp](kuniyosh@isi.imi.i.u-tokyo.ac.jp) arXiv :2607 .047 10v 1 [ cs .AI] 6 Jul 2026 ABSTRACT Inducing cooperation among distributed agents is still a difficult problem in the field of multi-agent reinforcement learning (MARL), particularly in social dilemma situations. There, individual interests are misaligned with the common good and individual rationality leads to suboptimal group outcomes. In contrast, humans are able to achieve cooperation with one another in such situations. A common explanation for such cooperative behavior is that individuals have social preferences. In order to achieve cooperation in MARL, we design a new utility function integrating altruistic preferences (incentive for other’s reward) and fairness preferences (incentive for equality) from social psychology and behavioral economics, namely, Altruistic and Fairness Preference (AFP), a reward-sharing mechanism which converts one’s own and other’s rewards to incentives for cooperative behavior. We performed comparative experiments with standard RL and inequity aversion agents in two challenging sequential social dilemma games, and showed that AFP agents successfully achieved mutual cooperation with more collective rewardsand higher equity than the baselines. To further understand the progression of AFP during training, we subsequently explore the effects of altruistic preferences and fairness preferences on agents’behavior. The results suggest that altruistic preferences encourage agents to contribute to the public goods, and fairness preferences induce mutual behavior between agents. KEYWORDS multi-agent reinforcement learning, social dilemma, cooperation, altruistic preferences, fairness preferences ACM Reference Format: Yu Wei, Yukiko Ogura, Yoshiyuki Ohmura, Ildefons Magrans deAbril, Hoshinori Kanazawa, and Yasuo Kuniyoshi. 2022. Integrated Altruistic and Fairness Preference Induces Advanced Mutual Cooperation in Sequential Social Dilemmas. In Proc. of the 21st International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2022), Auckland, New Zealand, May 9–13, 2022, IFAAMAS, 9 pages. Proc. of the 21st International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2022), P. Faliszewski, V. Mascardi, C. Pelachaud, M.E. Taylor (eds.), May 9–13, 2022, Auckland, New Zealand. © 2022 International Foundation for Autonomous Agents and Multiagent Systems ([www.ifaamas.org](www.ifaamas.org)). All rights reserved. 1 INTRODUCTION Multi-agent systems can be used to address problems in various applications, such as decentralized control, simulations in economics or sociology, and robotics. Multi-agent reinforcement learning (MARL) is a promising approach for addressing complex tasks without the need for preprogrammed agent behaviors. Recently, the evolution of cooperation among agents has become an important problem in MARL. Due to the difficulty in designing rewards for multi-agent systems, individual rewards are often intertwined with each other and create unexpected social dilemmas at the group level. This occurs because individual actions are taken to obtain higher individual rewards, and this can lead to suboptim

FAQs

What problem does the paper address in multi-agent reinforcement learning?

It addresses how to induce cooperation among distributed agents in sequential social dilemma situations where individual incentives conflict with collective outcomes.

How does the proposed AFP method work?

AFP integrates altruistic preferences (reward incentives for others’ returns) and fairness preferences (incentives for equality) into a unified utility function via a reward-sharing mechanism to promote cooperative behavior.

What do the experiments show about AFP compared with baseline methods?

AFP agents achieve mutual cooperation with higher collective rewards and greater equity than standard RL and inequity aversion agents in challenging sequential social dilemma games.

Jul 16, 2026

Upload by Noah