Rethinking Algorithmic Fairness for Human-AI Collaboration

2023-10-05

AI Summary

Existing fairness algorithms assume human decision-makers fully adopt algorithmic recommendations; however, in practice, selective compliance—where humans override certain recommendations—can exacerbate discrimination. Method: This paper introduces “compliance-robust fairness,” a novel paradigm ensuring that overall decision fairness remains no worse than the baseline policy under *any* human adoption rate. We formally define this property and demonstrate that standard fairness constraints may fundamentally degrade the fairness–accuracy trade-off in human–algorithm collaboration. We propose a robust fairness algorithm integrating constrained optimization with causal modeling to account for heterogeneous compliance behavior. Results: Empirical evaluation on Virginia criminal sentencing data shows that our method consistently improves fairness across unknown adoption patterns, whereas conventional fairness algorithms significantly worsen discrimination under partial compliance.

Abstract

Existing approaches to algorithmic fairness aim to ensure equitable outcomes if human decision-makers comply perfectly with algorithmic decisions. However, perfect compliance with the algorithm is rarely a reality or even a desirable outcome in human-AI collaboration. Yet, recent studies have shown that selective compliance with fair algorithms can amplify discrimination relative to the prior human policy. As a consequence, ensuring equitable outcomes requires fundamentally different algorithmic design principles that ensure robustness to the decision-maker's (a priori unknown) compliance pattern. We define the notion of compliance-robustly fair algorithmic recommendations that are guaranteed to (weakly) improve fairness in decisions, regardless of the human's compliance pattern. We propose a simple optimization strategy to identify the best performance-improving compliance-robustly fair policy. However, we show that it may be infeasible to design algorithmic recommendations that are simultaneously fair in isolation, compliance-robustly fair, and more accurate than the human policy; thus, if our goal is to improve the equity and accuracy of human-AI collaboration, it may not be desirable to enforce traditional algorithmic fairness constraints. We illustrate the value of our approach on criminal sentencing data before and after the introduction of an algorithmic risk assessment tool in Virginia.

Problem

• Ensure fairness in human-AI collaboration
• Develop compliance-robust fair algorithms
• Balance equity and accuracy in decisions

Innovation

• Compliance-robustly fair algorithmic recommendations
• Optimization strategy for fair policies
• Challenges traditional algorithmic fairness constraints