Appearing Unfair in the Wrong Direction

LessWrong explains a statistical paradox that can make a hiring classifier appear fair from the outside while actually disadvantaging the group it’s meant to help. Using a fictional alien planet with two species, the post shows that equal‑error rates can coexist with higher rejection of qualified candidates from the disadvantaged species because the groups have different base rates of ability. The author argues that common outcome‑based fairness measures, such as the proportion of hires that turn out to be mistakes, can therefore be misleading and should not be used as the sole indicator of fairness. Understanding this distinction helps researchers avoid drawing the wrong conclusions about bias in machine‑learning systems.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton