An Operational Perspective to Fairness Interventions: Where and How to Intervene

TL;DR

This paper presents a comprehensive framework for evaluating fairness interventions in AI systems, emphasizing where and how to intervene, and demonstrates its application through a case study on predictive parity using extensive empirical analysis.

Contribution

It introduces a novel method for achieving predictive parity without group data at inference time and provides empirical insights from a large benchmarking study across multiple models and datasets.

Findings

Distributionally robust methods improve fairness without inference group data

Plain XGBoost often outperforms neural networks in fairness-accuracy trade-offs

Fairness interventions are more effective when designed with the model's inductive bias in mind

Abstract

As AI-based decision systems proliferate, their successful operationalization requires balancing multiple desiderata: predictive performance, disparity across groups, safeguarding sensitive group attributes (e.g., race), and engineering cost. We present a holistic framework for evaluating and contextualizing fairness interventions with respect to the above desiderata. The two key points of practical consideration are \emph{where} (pre-, in-, post-processing) and \emph{how} (in what way the sensitive group data is used) the intervention is introduced. We demonstrate our framework with a case study on predictive parity. In it, we first propose a novel method for achieving predictive parity fairness without using group data at inference time via distibutionally robust optimization. Then, we showcase the effectiveness of these methods in a benchmarking study of close to 400 variations across two major model types (XGBoost vs. Neural Net), ten datasets, and over twenty unique methodologies. Methodological insights derived from our empirical study inform the practical design of ML workflow with fairness as a central concern. We find predictive parity is difficult to achieve without using group data, and despite requiring group data during model training (but not inference), distributionally robust methods we develop provide significant Pareto improvement. Moreover, a plain XGBoost model often Pareto-dominates neural networks with fairness interventions, highlighting the importance of model inductive bias.