fairness through unawareness
In plain wordsThe simple idea of making a system fair by hiding sensitive traits like race or sex from it; it usually fails, because other information gives those traits away.
The approach of seeking non-discrimination by excluding protected attributesprotected attribute, n.: A personal trait, such as race, sex, religion, age or disability, that the law says must not be used to treat people unfairly. from a model's inputs, so that decisions are formally blind to them. It mirrors the legal prohibition on disparate treatment and is a common default in practice. As a guarantee of fair outcomes it is generally ineffective, because remaining features correlated with the attribute let a model reconstruct it (proxy discriminationproxy discrimination, n.: Discrimination that happens when a system cannot see a trait like race directly but relies on something closely linked to it, such as a postcode, and so treats groups unequally anyway.). It can even make outcomes worse: without the attribute, a model cannot account for group differences in how features relate to outcomes, and disparities become harder to measure. Most fairness methods therefore use the attribute, at least during training or auditing.
Fairness through unawareness failed: with the gender field removed, the résumé model penalized women's colleges instead.
Origin
Named in contrast to 'fairness through awareness', the title of Cynthia Dwork and colleagues' paper (arXiv 2011; ITCS 2012), which criticized 'fairness through blindness' and argued that fairness requires using protected information. The earliest use found is in Moritz Hardt, Eric Price and Nathan Srebro's 'Equality of Opportunity in Supervised Learning' (October 2016); Matt Kusner and colleagues' 'Counterfactual Fairness' (2017) listed it, abbreviated FTU, among existing fairness definitions.
Related
- proxy discriminationproxy discrimination, n.: Discrimination that happens when a system cannot see a trait like race directly but relies on something closely linked to it, such as a postcode, and so treats groups unequally anyway.
- individual fairnessindividual fairness, n.: The idea that a fair system should treat similar people similarly: two applicants who are alike in every way that matters should get similar outcomes.
- protected attributeprotected attribute, n.: A personal trait, such as race, sex, religion, age or disability, that the law says must not be used to treat people unfairly.
- counterfactual fairnesscounterfactual fairness, n.: A fairness test that asks: had this person been of another race or sex, with everything that would have followed from that, would the system have decided the same way?
- algorithmic biasalgorithmic bias, n.: Unfairness in software that judges or serves people: it consistently produces worse results for some groups, such as women, older people or ethnic minorities, often because it learned from skewed or incomplete records.
- individual fairnessindividual fairness, n.: The idea that a fair system should treat similar people similarly: two applicants who are alike in every way that matters should get similar outcomes.
Sources
- Hardt, M., Price, E. & Srebro, N. (2016). Equality of Opportunity in Supervised Learning. Advances in Neural Information Processing Systems 29. arxiv.org/abs/1610.02413
- Kusner, M. J., Loftus, J., Russell, C. & Silva, R. (2017). Counterfactual Fairness. Advances in Neural Information Processing Systems 30.
Mentioned in
protected attributeprotected attribute, n.: A personal trait, such as race, sex, religion, age or disability, that the law says must not be used to treat people unfairly., proxy discriminationproxy discrimination, n.: Discrimination that happens when a system cannot see a trait like race directly but relies on something closely linked to it, such as a postcode, and so treats groups unequally anyway., individual fairnessindividual fairness, n.: The idea that a fair system should treat similar people similarly: two applicants who are alike in every way that matters should get similar outcomes.