Toward Substantive Intersectional Algorithmic Fairness: Desiderata for a Feminist Approach

Single demographic categories miss how discrimination compounds: The paper argues that current algorithmic fairness research typically examines discrimination through isolated demographic groups (e.g., women OR racial minorities separately), but real-world discrimination often affects people simultaneously across multiple identities in ways that can't be understood by studying each category alone.

Structural inequalities require more than technical fixes: Rather than treating fairness as a purely computational problem with mathematical solutions, the authors contend that algorithmic systems must address underlying systemic inequities. This means fairness cannot be achieved through technical operationalizations alone and requires understanding the social context in which algorithms operate.

Ten guiding principles replace one-size-fits-all solutions: The paper proposes ten "desiderata" (desired principles) rather than fixed rules, encouraging AI practitioners to actively reflect on assumptions like whether their systems are truly neutral, how they use demographic data, and whether they include multiply marginalized groups in their design and testing.

Multiply marginalized communities are often overlooked: The authors emphasize that people experiencing discrimination across multiple dimensions (e.g., Black women, disabled immigrants) are frequently neglected in fairness research, which tends to focus on larger demographic subgroups and may inadvertently reinforce existing inequalities.

Fairness work requires bridging computer science and social science: The paper calls for AI developers to integrate insights from intersectional feminist theory and social science alongside computational approaches, recognizing that technical expertise alone is insufficient for addressing systemic inequities.