This working paper examines how automated decision systems reproduce and obscure the forms of discrimination that civil-rights law was built to prevent, and proposes both a vocabulary and an accountability framework for addressing them. Grounded in the disparate-impact doctrine established in Griggs v. Duke Power Co. (1971) and codified across Title VII, the Fair Housing Act, and the Equal Credit Opportunity Act, the paper analyzes three legally significant cases — Mobley v. Workday (employment screening), Louis v. SafeRent Solutions (tenant scoring), and United States v. Meta Platforms (advertising delivery) — each of which extends liability to the operators of algorithmic systems. It distinguishes these genuine algorithmic-discrimination actions from adjacent enforcement matters, such as the 2024 CFPB order against Apple and Goldman Sachs, that are frequently miscategorized as discrimination cases. The paper introduces two analytic contributions: the Communicado / Incommunicado distinction, which names the boundary between what a system discloses to a subject and what it withholds without the subject's knowledge; and the Heritage Schema, a proposal to replace color-based racial classification with nationality-based self-identification in administrative data. It concludes with a tiered accountability framework combining mandatory bias auditing with a graduated, revenue-indexed civil-penalty structure. The aim is neither to indict automation as such nor to defend it, but to specify the conditions under which automated decisions can be made legible, contestable, and accountable.
Edward Callender (Wed,) studied this question.