Toward Multi-Agent Algorithmic Recourse

Challenges From a Game-Theoretic Perspective

Abstract

The recent adoption of machine learning as a tool in real-world decision making has spurred interest in potential harm caused by these methods. Providing those negatively impacted by decisions made by   machine learning models with ways to change those decisions is now an important area of machine learning known as algorithmic recourse. Past work has largely focused on the effect algorithmic recourse has on a single agent. In this work, we relax this assumption and examine algorithmic recourse from the multi-agent perspective. We use ideas from game theory to explore challenges from the multi-agent perspective that are unaddressed in the current literature and to propose new criteria to guide future algorithmic recourse research.