As a vast and ever-growing body of social-scientific research shows, discrimination remains pervasive in the United States. In education, work, consumer markets, healthcare, criminal justice, and more, Black people fare worse than whites, women worse than men, and so on. Moreover, the evidence now convincingly demonstrates that this inequality is driven by discrimination. Yet solutions are scarce. The best empirical studies find that popular interventions—like diversity seminars and antibias trainings—have little or no effect. And more muscular solutions—like hiring quotas or school busing—are now regularly struck down as illegal. Indeed, in the last thirty years, the Supreme Court has invalidated every such ambitious affirmative action plan that it has reviewed.
This Article proposes a novel solution: Big Data Affirmative Action. Like old-fashioned affirmative action, Big Data Affirmative Action would award benefits to individuals because of their membership in protected groups. Since Black defendants are discriminatorily incarcerated for longer than whites, Big Data Affirmative Action would intervene to reduce their sentences. Since women are paid less than men, it would step in to raise their salaries. But unlike old-fashioned affirmative action, Big Data Affirmative Action would be automated, algorithmic, and precise. Circa 2021, data scientists are already analyzing rich datasets to identify and quantify discriminatory harm. Armed with such quantitative measures, Big Data Affirmative Action algorithms would intervene to automatically adjust flawed human decisions—correcting discriminatory harm but going no further.
Big Data Affirmative Action has two advantages over the alternatives. First, it would actually work. Unlike, say, antibias trainings, Big Data Affirmative Action would operate directly on unfair outcomes, immediately remedying discriminatory harm. Second, Big Data Affirmative Action would be legal, notwithstanding the Supreme Court’s recent case law. As argued here, the Court has not, in fact, recently turned against affirmative action. Rather, it has consistently demanded that affirmative action policies both stand on solid empirical ground and be well tailored to remedying only particularized instances of actual discrimination. The policies that the Court recently rejected have failed to do either. Big Data Affirmative Action can easily do both.
Peter N. Salib,
Big Data Affirmative Action,
Nw. U. L. Rev.