Fairness Has More Than One Definition

Here's something that surprises most teams: there is no single agreed definition of a "fair" AI system. Ask three experts and you'll get three answers — and the math shows you often can't satisfy all of them at once.

One definition says the system should approve the same share of applicants from every group. Another says that among the people it approves, the success rate should be equal across groups. A third says similar individuals should get similar outcomes, regardless of group.

In plain English: fairness can mean equal treatment, equal results, or equal accuracy — and these genuinely conflict. A system can be fair by one definition and unfair by another, at the same time.

This isn't a reason to give up. It's a reason to choose deliberately. Before your team deploys any AI that affects people — hiring, lending, admissions, screening — write down which definition of fairness you're using and why. "We chose this definition because..." is one of the most powerful sentences in AI governance.

You can't satisfy every definition. But you can defend the one you chose, and that's what accountability looks like.

I explain frameworks; I don't give legal counsel.