Fairness in machine learning: against false positive rate equality as a measure of fairness

False positive rate equality may not actually measure fairness: The paper argues that one of the two most popular fairness metrics in machine learning—equality of false positive rates across groups—doesn't actually capture anything meaningful about fairness, despite its widespread use in the field.

The "fairness tradeoff" assumes both metrics matter equally: Machine learning literature often frames the conflict between calibration and false positive rate equality as an unavoidable tradeoff between two equally important fairness goals, but this paper challenges whether both metrics deserve equal weight in ethical decision-making.

Philosophers have been absent from a crucial technical debate: The paper highlights that while computer scientists have debated these metrics extensively, philosophers and ethicists have not examined whether these statistical measures actually correspond to real ethical concerns—a gap that matters for consequential decisions.

Intuitive appeal doesn't guarantee ethical validity: Just because a fairness metric seems intuitively important doesn't mean it actually tracks something normatively significant; the paper suggests the field has adopted false positive rate equality without sufficient ethical scrutiny.

We need ethical frameworks, not just statistical ones: The paper calls for moving beyond purely technical discussions of metrics to develop explicit ethical frameworks that clarify what fairness actually means before choosing which algorithmic measures to prioritize.