Sex, Causation, and Algorithms: Equal Protection in the Age of Machine Learning
54 Pages Posted: 13 Mar 2020
Date Written: March 12, 2020
U.S. constitutional law prohibits the use of sex as a proxy for other traits in most instances. For example, the Virginia Military Institute [VMI] may not use sex as a proxy for having the “will and capacity” to be a successful student. At the same time, sex-based classifications are constitutionally permissible when they track so-called “real differences” between men and women. Women and men at VMI may be subject to different training requirements, for example. Yet, it is surprisingly unclear when and why some sex-based classifications are permissible and others not. This question is especially important to examine now as the use of predictive algorithms, some of which rely on sex-based classifications, is growing increasingly common. If sex is predictive of some trait of interest, may the state – consistent with equal protection – rely on an algorithm that uses a sex-based classification?
This Article presents a new normative principle to guide the analysis. I argue that courts ought to ask why sex is a good proxy for the trait of interest. If prior injustice is likely the reason for the observed correlation, then the use of the sex classification should be presumptively prohibited. This Anti-Compounding Injustice principle both explains and justifies current doctrine better than the hodge-podge of existing rules and concepts and provides a useful lens through which to approach new cases.
Keywords: discrimination, equal protection, algorithms, machine learning, “real differences,” sex discrimination
Suggested Citation: Suggested Citation