Fairness Needs Visibility, Not Blindness
Explain why fairness assessment often requires sensitive features even when they are not used for prediction.
Fairness blindness can become harm blindness. Inputs Are Not The Whole System Fairlearn describes fairness in terms of harms experienced by groups. Those harms can appear even when sensitive attributes are absent from the model input. Location, education, employment gaps, device type, language, shift pattern, and purchase history can all carry social patterns. A model can learn those patterns without ever seeing a protected column. Audit Data Has A Different Job Sensitive features can be inappropriate for prediction and still necessary for assessment. The audit question is: are outcomes, errors, or service levels comparable across groups that may be harmed?…
Sign up free — one personalized lesson every day, matched to your role and goals.
Already have an account? Sign in