New AI utility function promotes cooperation in multi-agent systems

Researchers have developed a new utility function called Altruistic and Fairness Preference (AFP) to encourage cooperation in multi-agent reinforcement learning (MARL) systems, particularly in social dilemma scenarios where individual incentives conflict with collective well-being. By integrating preferences for others' rewards and for equitable outcomes, AFP agents demonstrated successful mutual cooperation and higher equity compared to standard reinforcement learning agents in sequential social dilemma games. The study found that altruistic preferences drive contributions to public goods, while fairness preferences foster reciprocal behavior between agents. AI

IMPACT Introduces a novel approach to foster cooperation in multi-agent AI systems, potentially improving coordination in complex environments.

RANK_REASON Academic paper detailing a new utility function for MARL. [lever_c_demoted from research: ic=1 ai=1.0]

Read on Hugging Face Daily Papers →

• Altruistic and Fairness Preference
• Hugging Face
• Multi-agent reinforcement learning
• Social Dilemmas

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →