Fairness
Also called Algorithmic Fairness
Making sure a model doesn’t systematically treat some groups of people worse than others.
Think of it like
A hiring manager who judges every candidate by the same yardstick, not by their name or neighborhood.
Example
A loan model is audited to check it doesn’t approve one demographic at far lower rates than another for equally qualified applicants.
How it actually works
Fairness turns out to be plural: equal accuracy, equal false-positive rates, and equal outcomes are different definitions that can be mathematically impossible to satisfy at once. Bias usually enters through skewed training data reflecting historical inequities. So fairness work is as much about choosing which definition fits the context as it is about the metric.
For product teams
A legal, ethical, and reputational must — and one where "fair" needs an explicit definition per use case.
For engineers
Multiple fairness criteria (parity, equalized odds, calibration) can conflict provably; pick the one that fits the harm and measure it.
Related
- Transparency — A pillar of responsible AI alongside transparency.
- Tied to accountability for outcomes.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome