Emergent Dominance Hierarchies in Reinforcement Learning Agents

Abstract

Modern Reinforcement Learning (RL) algorithms are able to outperform humans in a wide variety of tasks. Multi-agent reinforcement learning (MARL) settings present additional challenges, and successful cooperation in mixed-motive groups of agents depends on a delicate balancing act between individual and group objectives. Social conventions and norms, often inspired by human institutions, are used as tools for striking this balance.In this paper, we examine a fundamental, well-studied social convention that underlies cooperation in both animal and human societies: dominance hierarchies.We adapt the ethological theory of dominance hierarchies to artificial agents, borrowing the established terminology and definitions with as few amendments as possible. We demonstrate that populations of RL agents, operating without explicit programming or intrinsic rewards, can invent, learn, enforce, and transmit a dominance hierarchy to new populations. The dominance hierarchies that emerge have a similar structure to those studied in chickens, mice, fish, and other species.

Bio:

Ram is an AI Safety researcher at Tufts University. Ram is interested in exploring the use of social behavior in MARL environments for AI Safety goals such as AI Interpretability and AI Corrigibility. Ram has spent most of his career working as a software engineer, most recently for Google. Ram is a recurrent contributor to the open-source ecosystem, for which he was recognized as a Fellow by the Python Software Foundation. More information on Ram's research at https://r.rachum.com/.