The Impossibility Theorem of Fairness
You'll learn why achieving multiple fairness criteria simultaneously is mathematically impossible in most real-world scenarios.
The Core Impossibility
After learning about demographic parity, equalized odds, predictive parity, and calibration, you might hope to satisfy all of them at once. Unfortunately, mathematics proves this is generally impossible. The Impossibility Theorem of Fairness states that except in trivial cases (like when base rates are equal across all protected groups or when the classifier is perfect), you cannot simultaneously satisfy multiple fairness definitions.
For example, you cannot achieve both:
- Calibration (predicted probabilities match true outcome rates)
- Equal opportunity (equal true positive rates across groups)
...unless the prevalence of the positive class is identical in all groups or your model makes no errors.
Why This Matters
Think of fairness criteria like knobs on a mixing board—turning one affects the others. If Group A has a 30% disease rate and Group B has 10%, a calibrated model will naturally predict different rates. Forcing demographic parity (equal positive prediction rates) will break calibration: some groups will receive inflated or deflated risk scores relative to their true risk.
Practical Implications
This theorem doesn't mean fairness is impossible—it means you must:
- Choose which fairness criterion matters most for your context
- Accept tradeoffs explicitly rather than hoping for a perfect solution
- Document which fairness goals you prioritized and why
A loan system might prioritize calibration (accurate risk assessment), while a resume screener might prioritize equal opportunity (ensuring qualified candidates from all groups advance).
Key Takeaway: Multiple fairness criteria are mathematically incompatible in realistic settings; practitioners must make explicit, context-dependent tradeoffs rather than seeking universal fairness.