The practice and measurement of equitable AI outcomes across demographic groups—operationalized via metrics like disparate impact, equal opportunity, and calibration.
AI fairness requires choosing a fairness definition appropriate to the harm: no single metric captures all fairness concerns (and some are mutually exclusive). Toolkits include IBM AI Fairness 360 and Fairlearn.
See also
Bias and fairness for the broader framing and Responsible AI for program-level practices.