ALGORITHMIC DISCRIMINATION AND INPUT ACCOUNTABILITY UNDER THE CIVIL RIGHTS ACTS

• Published in: Berkeley technology law journal Vol. 36; no. 2; p. 675
• Main Authors: Bartlett, Robert, Morse, Adair, Wallace, Nancy, Stanton, Richard
• Format: Journal Article
• Language: English
• Published: Berkeley University of California, Boalt Hall School of Law 01.10.2021
• Subjects: Accountability; Algorithms; Antidiscrimination; Civil rights; Civil Rights Act 1964-US; COVID-19; Decision analysis; Decision making; Discrimination; Legislatures; Racial inequality; State legislatures; Urgency; United States > US
• ISSN: 1086-3818; 2380-4742
• DOI: 10.15779/Z381N7XN5B

The disproportionate burden of COVID-19 among communities of color and a necessary renewed attention to racial inequalities have lent new urgency to concerns that algorithmic decision-making can lead to unintentional discrimination against members of historically marginalized groups. These concerns are being expressed through Congressional subpoenas, regulatory investigations, and an increasing number of algorithmic accountability bills pending in both state legislatures and Congress. To date, however, prominent efforts to define algorithmic accountability have tended to focus on output-oriented policies that may facilitate illegitimate discrimination or involve fairness corrections unlikely to be legally valid. Worse still, other approaches focus merely on a model's predictive accuracy-an approach at odds with long-standing U.S. anti-discrimination law. We provide a workable definition of algorithmic accountability that is rooted in case law addressing statistical discrimination in the context of Title VII of the Civil Rights Act of 1964. Using instruction from the burden-shifting framework codified to implement Title VII, we formulate a simple statistical test to apply to the design and review of the inputs used in any algorithmic decision-making process. Application of the test, which we label the Input Accountability Test, constitutes a legally viable, deployable tool that can prevent an algorithmic model from systematically penalizing members of protected groups who are otherwise qualified in a legitimate target characteristic of interest.