Knowledge MaxxingFairness trade-off
A fairness trade-off is the mathematical fact that, except in special cases, a model generally cannot simultaneously satisfy every fairness criterion at once, forcing a choice of which to prioritize.
Prerequisite: Calibration
What matters
Understanding this trade-off exists prevents teams from assuming that picking any one fairness metric automatically ensures the system is fair by every possible definition.
Example
A model tuned to satisfy demographic parity may fail to satisfy equalised odds for the same groups, since the two criteria can mathematically conflict.
Common misconception
Choosing one fairness metric to optimize for is not a sign of poor design; it's often an unavoidable, deliberate choice given that multiple fairness definitions can conflict.
Recall question
Why can't a model generally satisfy every fairness definition simultaneously?