go back

Quantifying cooperation between artificial agents using information theory

Patricia Wollstadt and Matti Krüger, "Quantifying cooperation between artificial agents using information theory", HHAI2022: Augmenting Human Intellect, vol. 354, pp. 302 - 304, 2022.

Abstract

When designing interactive human-machine systems, it is often assumed that it is desirable for such systems to behave cooperatively towards a human operator, to improve trust, acceptance, and usability, but also to increase task efficiency. To design cooperative HMI systems, we have to be able to define and quantitatively describe cooperative interactions, for example, to control, optimize, or evaluate system behavior. Despite the increased interest in cooperative HMI in recent years, an approach that provides a suitable definition of cooperation as well as a method for its quantification is still missing. In the present work, we therefore develop a novel definition of cooperative behavior in HMI contexts, based on which we propose to quantify cooperative behavior using recent methods from information theory. We define cooperation as joint, coordinated actions that are mutually adapted such as to facilitate the realization of a joint task. Thus, cooperation is characterized by a synergistic effect of joint actions towards a goal. Here, we propose to use the recently introduced partial information decomposition framework from information theory, which proposes measures to quantify the synergistic contributions two inputs have towards a target variable. We propose to apply the synergy measure to two or more input variables describing two agents' agents towards a target variable, describing the current goal state. We provide a first validation of our approach by successfully applying our definition and approach to a model system from multi-agent reinforcement learning.



Download Bibtex file Download PDF

Search