Individual Fairness: Similar Treatment for Similar Individuals
What you'll learn: How to think about fairness at the level of individuals rather than groups, ensuring similar people receive similar predictions.
The Core Idea
While group fairness metrics (like demographic parity or equalized odds) focus on statistical properties across protected groups, individual fairness takes a different approach: similar individuals should receive similar treatment, regardless of group membership.
The intuition is simple: if two loan applicants have nearly identical financial histories, credit scores, and employment status, they should receive similar approval probabilities—even if one is male and one is female, or they differ on other protected attributes.
The Mathematical Intuition
Individual fairness is often formalized as a Lipschitz condition:
If individuals
x₁andx₂are similar (according to some distance metricd), then the model's predictionsf(x₁)andf(x₂)should also be close.
The challenge? Defining what "similar" means. For loan applications, similarity might weigh credit score heavily but age lightly. This requires domain expertise and value judgments that statistical metrics can't provide automatically.
Why It Matters
Unlike group fairness, individual fairness:
- Doesn't rely on pre-defined protected groups
- Can catch unfair treatment within demographic categories
- Aligns with legal principles like "equal treatment under the law"
However, it's much harder to measure and enforce in practice, because you must define a meaningful similarity metric for your specific domain.
Key Takeaway
Individual fairness ensures that similar people get similar predictions by enforcing consistency at the person-to-person level, but requires careful definition of what "similarity" means in your application context.