Course contentsShow
← AI Engineering
Lesson 1565 of 1,88638. Bias, Fairness, and AlignmentFree lesson

Defining Fairness in AI Systems

Understanding different definitions of fairness and why no single metric fits all contexts.

Defining Fairness in AI Systems

You'll learn why fairness isn't a single checkbox, but a context-dependent choice between competing mathematical definitions.

The Fairness Paradox

After detecting bias in your AI system, the natural next step is making it "fair." But here's the challenge: fairness means different things in different contexts, and many definitions are mathematically incompatible with each other.

Think of fairness like "success" in sports—what counts as fair depends on the game. In basketball, equal opportunity means everyone can try out; equal outcome would mean every team wins half their games. Both are valid interpretations, but you can't optimize for both simultaneously.

Core Fairness Definitions

Demographic Parity: Every group receives positive outcomes at equal rates. If 40% of Group A gets loans, 40% of Group B should too.

Equal Opportunity: Among qualified candidates (those who should succeed), every group has equal true positive rates. If 80% of creditworthy applicants from Group A get loans, the same should hold for Group B.

Predictive Parity: Positive predictions are equally accurate across groups. If your model says "approve," it should be correct 90% of the time regardless of group membership.

Individual Fairness: Similar individuals receive similar predictions, regardless of protected attributes.

The catch? Research proves you often cannot satisfy multiple definitions simultaneously—especially demographic parity and predictive parity when base rates differ between groups.

Choosing Your Metric

Your choice depends on your domain and values. Criminal justice might prioritize equal opportunity (don't miss dangerous individuals in any group), while loan approval might emphasize predictive parity (approved individuals should repay at similar rates).

Key Takeaway: No single fairness metric fits all contexts. You must explicitly choose which definition aligns with your application's goals and ethical requirements, understanding that trade-offs are inevitable.