Drawing on her co-authored paper, Toward substantive intersectional algorithmic fairness: desiderata for a feminist approach, Marie Mirsch delivered a compelling AI & Equality Pub Talk, challenging conventional understandings of fairness in algorithmic decision-making, urging AI practitioners to move beyond superficial fixes and formal metrics.
Mirsch began by recounting her realization, spurred by Caroline Criado-Perez’s Invisible Women, that the data she had mastered could tell a profoundly different, often obscured, story. This insight led her to connect the seemingly disparate worlds of philosophy and engineering, and Gender.
There is an enduring problem with many current fairness metrics: their failure to grasp the concept of intersectionality. Mirsch demonstrated how current approaches often fall back on single-axis thinking, evaluating groups like “gender” or “race” distinctly. As Kimberlé Crenshaw argued 35 years ago in her critique, this approach obscures the compounding discrimination faced by those at multiple intersections. A constructed example showed how simple single-axis metrics could register seemingly acceptable error rates for broad groups, yet completely miss 0% accuracy for specific subgroups like Black female or White male. This phenomenon, known as fairness gerrymandering, highlights a critical blind spot in technical evaluation. Mirsch argues that merely adding more subgroups (subgroup fairness) risks collapsing the rich, contextual, and political idea of intersectionality into a purely technical exercise—a concept she terms formal intersectional algorithmic fairness.
Instead, she proposed adopting a substantive intersectional algorithmic fairness. This requires AI practitioners to challenge fundamental assumptions, beginning with epistemological precision. Concepts like intersectionality, she noted, lose their political and contextual power when they travel across disciplines without critical deliberation.
Most crucially, there is also the myth of algorithmic neutrality and objectivity. Mirsch highlighted that many engineers still assume technical systems are objective or error-free. She argued that the prevalent view from nowhere concept of objectivity is impossible, as all knowledge is situated. There needs to be a shift to perspectival suppleness: viewing objectivity not as a binary but as a spectrum achieved through the rigorous scrutiny of diverse epistemic communities. Mirsch’s central call to action is for reflexivity: questioning assumed neutrality, making one’s own positionality explicit, and creating processes that deliberately invite multiple, often marginalized, perspectives to the decision-making table. This commitment to a reflexive, substantive approach is essential if AI is ever to align with true social justice.