Is Algorithmic Affirmative Action Legal?

This Article is the first to comprehensively explore whether algorithmic affirmative action is lawful. It concludes that both statutory and constitutional antidiscrimination law leave room for race-aware affirmative action in the design of fair algorithms. Along the way, the Article recommends some clarifications of current doctrine and proposes the pursuit of formally race-neutral methods to achieve the admittedly race-conscious goals of algorithmic affirmative action. The Article proceeds as follows. Part I introduces algorithmic affirmative action. It begins with a brief review of the bias problem in machine learning and then identifies multiple design options for algorithmic fairness. These designs are presented at a theoretical level, rather than in formal mathematical detail. It also highlights some difficult truths that stakeholders, jurists, and legal scholars must understand about accuracy and fairness trade-offs inherent in fairness solutions. Part II turns to the legality of algorithmic affirmative action, beginning with the statutory challenge under Title VII of the Civil Rights Act. Part II argues that voluntary algorithmic affirmative action ought to survive a disparate treatment challenge under Ricci and under the antirace-norming provision of Title VII. Finally, Part III considers the constitutional challenge to algorithmic affirmative action by state actors. It concludes that at least some forms of algorithmic affirmative action, to the extent they are racial classifications at all, ought to survive strict scrutiny as narrowly tailored solutions designed to mitigate the effects of past discrimination