Factoring the Matrix of Domination: A Critical Review and Reimagination of Intersectionality in AI Fairness

Intersectionality is being oversimplified in AI fairness research: The paper finds that researchers typically reduce intersectionality—a framework designed to examine how multiple social inequalities interact—to simply optimizing fairness metrics across demographic subgroups, missing the framework's deeper purpose of understanding structural power dynamics.

Current AI fairness approaches ignore broader social context: The reviewed papers fail to adequately discuss the social and historical contexts in which AI systems operate, and when they do mention "power," they narrowly focus only on technical aspects within the AI pipeline rather than examining systemic power structures that shape both the technology and society.

There's a critical gap between theory and practice: While intersectionality as a concept appears in AI fairness literature, researchers are not actually implementing it in meaningful ways—creating a disconnect between what intersectionality promises (examining interconnected inequalities) and what AI fairness work actually delivers (demographic parity metrics).

Actionable change requires grounding intersectionality in AI's foundational assumptions: The paper recommends that AI fairness researchers must move beyond surface-level adoption of intersectionality by fundamentally rethinking how AI systems are designed and evaluated, integrating intersectional thinking into the core epistemology (ways of knowing) that underpin AI development.