Abstract

Despite impressive advances in Artificial Intelligence (AI)-based gameplaying in recent years, many such systems still lack the capability of robustly dealing with novelty in a sufficiently rich environment. Novelty may be operationally described as the states or situations that violate (implicit or explicit) assumptions about agents, the environment, and agent–agent and agent–environment interactions. In this article, we describe a simulation platform called GNOME (Generating Novelty in Open-World Multi-agent Environments) that was developed to support the development, training and evaluation of AI agents designed to play the strategic four-player board game of Monopoly. Specifically, GNOME evaluates the ability of these agents to detect and react to novelty that is potentially unanticipated by agent developers, and is one of the first gameplaying simulators to treat novelty as a first-class citizen. We hope that its …

Date

November 1, 2021

Authors

Mayank Kejriwal, Shilpa Thomas

Journal

Simulation Modelling Practice and Theory

Volume

112

Pages

102364

Publisher

Elsevier