Navigating the landscape of multiplayer games

Shayegan Omidshafiei,Paul Muller,Francisco C Santos,Daniel Hennes,Wojciech M Czarnecki,Bart De Vylder,Mark Rowland,Julien Pérolat,Audrunas Gruslys,Karl Tuyls,Rémi Munos,Jerome Connor

doi:10.1038/s41467-020-19244-4

Abstract

Multiplayer games have long been used as testbeds in artificial intelligence research, aptly referred to as the Drosophila of artificial intelligence. Traditionally, researchers have focused on using well-known games to build strong agents. This progress, however, can be better informed by characterizing games and their topological landscape. Tackling this latter question can facilitate understanding of agents and help determine what game an agent should target next as part of its training. Here, we show how network measures applied to response graphs of large-scale games enable the creation of a landscape of games, quantifying relationships between games of varying sizes and characteristics. We illustrate our findings in domains ranging from canonical games to complex empirical games capturing the performance of trained agents pitted against one another. Our results culminate in a demonstration leveraging this information to generate new and interesting games, including mixtures of empirical games synthesized from real world games.

Highlights

Multiplayer games have long been used as testbeds in artificial intelligence research, aptly referred to as the Drosophila of artificial intelligence
While the empirical game-theoretic results are subject to the policies used to generate them, we rely on a sampling scheme designed to capture a diverse variety of interactions within each game, and subsequently conduct sensitivity analysis to validate the robustness of the results
Our work demonstrated a means of revealing this topological structure, extending beyond related works investigating this question for small classes of games (e.g., 2 × 2 games[40,85,101]). We believe that such a topological landscape of games can help to identify and generate related games of interest for artificial intelligence (AI) agents to tackle, as targeted by the Problem Problem, hopefully significantly extending the reach of AI system capabilities

Summary

Introduction

Multiplayer games have long been used as testbeds in artificial intelligence research, aptly referred to as the Drosophila of artificial intelligence. Researchers have focused on using well-known games to build strong agents This progress, can be better informed by characterizing games and their topological landscape. AI research has primarily placed emphasis on training of strong agents; we refer to this as the Policy Problem, which entails the search for super human-level AI performance. Despite this progress, the need for a task theory, a framework for taxonomizing, characterizing, and decomposing AI tasks has become increasingly important in recent years[11,12]. Designation of a single measure characterizing games is a non-trivial task

Objectives

Methods

Results

Conclusion