The Impossibility of Fairness
What you'll learn: Why achieving multiple fairness goals simultaneously in algorithmic systems is mathematically impossible.
The Fundamental Conflict
You've learned various fairness definitions and metrics—but here's the hard truth: you cannot satisfy all fairness criteria at the same time. This isn't a technical limitation or a bug; it's mathematically proven. Different fairness definitions contradict each other, forcing you to choose which notion of fairness matters most for your context.
Why Fairness Criteria Clash
Consider a loan approval system:
- Demographic parity requires equal approval rates across groups (e.g., 60% approval for everyone)
- Equalized odds requires equal false positive and false negative rates across groups
- Predictive parity requires that among all approved applicants, the same proportion from each group actually repays
If groups have different base rates of loan repayment in your data, satisfying one criterion necessarily violates another. It's like trying to draw a triangle where all three angles are 90 degrees—the math simply doesn't work.
The Real-World Implication
This impossibility theorem means:
- There is no "one true fairness"—you must explicitly choose which fairness notion aligns with your values and legal requirements
- Trade-offs are unavoidable—improving one fairness metric may worsen another
- Transparency is critical—document which definition you prioritized and why
- Context matters—criminal justice may demand different fairness criteria than college admissions
The impossibility of fairness doesn't mean we abandon the effort—it means we recognize that fairness is a complex ethical choice, not a simple technical checkbox.
Key Takeaway: Multiple fairness criteria mathematically conflict and cannot all be satisfied simultaneously; choosing a fairness definition is an ethical decision requiring explicit justification and stakeholder input, not a purely technical problem.