GovernanceCore
Trustworthiness

Fairness

Working definition

Reviewed 30 July 2026

Fairness in AI concerns whether system design, data, performance, and outcomes treat individuals and groups appropriately in context. It is not a single metric: different fairness criteria encode different values and may be mathematically incompatible.

Context

Why it matters

A system can be accurate overall while producing worse errors or access for particular groups. Governance must decide which disparities matter and what remedy is appropriate.

Operating note

What this looks like in practice

  1. 01Define affected groups, plausible harms, and the normative goal before choosing metrics.
  2. 02Measure performance at relevant intersections and include qualitative evidence.
  3. 03Test the full decision process, including data collection, thresholds, human action, and recourse.

Sources & further research

Primary authority anchors the definition. Research links add conceptual or operational depth. External sources may update independently; always verify legal duties against the current official text.

  1. Official source

    The Language of Trustworthy AI: An In-Depth Glossary of Terms

    Catalogs several contextual and mathematical conceptions of bias and fairness.

    NIST
    2023    Open source ↗
  2. Research paper

    Fairness and Abstraction in Sociotechnical Systems

    Explains how decontextualized fairness work can miss the social system in which an algorithm operates.

    Selbst et al.
    2019    Open source ↗