Entry 15 of 44
Fairness
faeger — beautiful, fitting; also fair as in equitable
Old English
From Old English faeger (beautiful, pleasant) and separately from fair (just, equitable). The equity meaning carries: what is owed to each person according to their due.
Established meaning
In law and moral philosophy: just treatment; giving to each what is their due. Procedural fairness: processes applied equally. Substantive fairness: outcomes that do not systematically disadvantage protected groups. Fairness is relational — it concerns what one party owes another.
AI industry deployment
A family of mathematical metrics (demographic parity, equalized odds, calibration, etc.) that cannot all be simultaneously satisfied. 'Fair AI' often means the developer has chosen one metric and optimized for it. Which metric was chosen, and why, is typically not disclosed.
Semantic gap
The normative choice
Choosing a fairness metric is a value judgment about what kind of equity matters most — a political and ethical decision. Industry usage presents this choice as a technical operation, hiding the normative decision inside the metric.