Regulation Games for Trustworthy Machine Learning
Mohammad Yaghini, Patty Liu, Franziska Boenisch, Nicolas Papernot
TL;DR
The paper reframes trustworthy ML as a multi-agent, multi-objective regulation problem and introduces SpecGame, a repeated Stackelberg game between a model builder and regulators for fairness and privacy. It then proposes ParetoPlay, an equilibrium-search algorithm that leverages a shared Pareto frontier to coordinate strategies and induce correlated equilibria, enabling efficient policy design. The authors provide theoretical insights (shared Pareto frontier viability, scalarization-derived frontier) and empirical guidance through simulations on gender classification tasks, showing that regulator-initiated penalties and first-mover dynamics can steer outcomes toward compliant, efficient equilibria. The work highlights the inadequacy of single-agent formulations for trustworthy ML, offers practical incentive-design guidance (how to set $C_ extsl{fair}$ and $C_ extsl{priv}$), and discusses limitations and future directions for regulation in ML systems.
Abstract
Existing work on trustworthy machine learning (ML) often concentrates on individual aspects of trust, such as fairness or privacy. Additionally, many techniques overlook the distinction between those who train ML models and those responsible for assessing their trustworthiness. To address these issues, we propose a framework that views trustworthy ML as a multi-objective multi-agent optimization problem. This naturally lends itself to a game-theoretic formulation we call regulation games. We illustrate a particular game instance, the SpecGame in which we model the relationship between an ML model builder and fairness and privacy regulators. Regulators wish to design penalties that enforce compliance with their specification, but do not want to discourage builders from participation. Seeking such socially optimal (i.e., efficient for all agents) solutions to the game, we introduce ParetoPlay. This novel equilibrium search algorithm ensures that agents remain on the Pareto frontier of their objectives and avoids the inefficiencies of other equilibria. Simulating SpecGame through ParetoPlay can provide policy guidance for ML Regulation. For instance, we show that for a gender classification application, regulators can enforce a differential privacy budget that is on average 4.0 lower if they take the initiative to specify their desired guarantee first.