An Intersectional Definition of Fairness

TL;DR

This paper introduces intersectional fairness definitions for machine learning that account for overlapping social identities, providing theoretical guarantees and a practical algorithm validated through case studies.

Contribution

It proposes a novel intersectional fairness framework, with proven guarantees and a new learning algorithm, addressing limitations of existing fairness notions.

Findings

The criteria behave sensibly across protected attribute subsets.

Theoretical guarantees include economic, privacy, and generalization bounds.

Empirical case studies demonstrate practical utility.

Abstract

We propose definitions of fairness in machine learning and artificial intelligence systems that are informed by the framework of intersectionality, a critical lens arising from the Humanities literature which analyzes how interlocking systems of power and oppression affect individuals along overlapping dimensions including gender, race, sexual orientation, class, and disability. We show that our criteria behave sensibly for any subset of the set of protected attributes, and we prove economic, privacy, and generalization guarantees. We provide a learning algorithm which respects our intersectional fairness criteria. Case studies on census data and the COMPAS criminal recidivism dataset demonstrate the utility of our methods.