Limitations of statistical and causal notions of fairness

December 28, 2020

Statistical and causal notions of fairness are ubiquitous when trying to ensure equal treatment between social groups in algorithmic decision-making. In their FAT* 2020 paper, Lily Hu and CSI-affiliated researcher Issa Kohler-Hausmann argue that these traditional methods for ensuring fairness make a conceptual error by assuming that the membership in a demographic group is separable from the social phenomena that go along with it. Instead, they call for a reimagining of what mathematically formal definitions of fairness can be — and the normative assumptions that underlie them.


Blog post: