SESKStuff Everyone Should Know
SESK: Stuff Everyone Should Know · sesk.info/ai-fairness-metrics-conflict

Can every mathematical fairness condition usually be satisfied at once?

No. Important fairness conditions can be mathematically incompatible.

Start playing free

  1. Yes, with enough data
  2. Yes, with a larger model
  3. No, some conditions conflict ✔

Why: Fairness has multiple definitions. Except in special cases, a classifier cannot simultaneously satisfy every desirable group fairness constraint, so the choice must fit the real context.

Huh, didn't know that: Calibration can conflict with balancing errors across positive and negative classes.

The dad joke
The fairness metrics agreed to disagree, very evenly.
Like this card?
Play the game

References

Answer: No, some conditions conflict. Fairness has multiple definitions. Except in special cases, a classifier cannot simultaneously satisfy every desirable group fairness constraint, so the choice must fit the real context.

  1. NIST SP 1270: Identifying and Managing Bias in AI nvlpubs.nist.gov/nistpubs/SpecialPublications/NIST.SP.1270.pdf
    there is an inherent incompatibility between two conditions: calibration and balancing the positive and negative classes.
    Checked 2026-10-10.

Think this answer is wrong? Tap "Challenge this answer" on the card in the game and tell us why.

Test yourself in the game

Artificial IntelligenceHow do AI, machine learning and deep learning fit together?Artificial IntelligenceWhat did Alan Turing call his 1950 test of machine intelligence?Artificial IntelligenceTuring predicted that by about 2000, judges would ID the machine at most what percent of the time?

That's stuff everyone should know.