Choosing Which Fairness to Relax: Algorithm Governance Under Civil Rights Law

Algorithmic decision systems in the world of housing and credit flow through long-standing racial fault lines when it comes to wealth and access. In another paper, I establish a precise impossibility result: no decision process can simultaneously be efficient, formally neutral and robustly minority-protective in groups-triggered differentiation. This Article demonstrates that U.S. civil rights law already negotiates that impossibility. Drawing on parallel doctrines in the Equal Credit Opportunity Act and Fair Housing Act, I suggest that courts and regulators implicitly solve other parts of the fairness trilemma by deciding which principle to soften. Business-necessity defenses emphasize evenhandedness; Special Purpose Credit Programs permit disparate-impact remedies based on fairness considerations of breaking symmetry; segregative-effects claims and consent decrees embed minority representation through the practice of governance. These doctrinal devices are not ad hoc reactions to the problem of algorithmic bias-they are institutional engines of compromise over an inherent tradeoff. In setting forth these choices more deliberately, the Article reconceives algorithmic fairness as an issue of institutional structure rather than discretionary metric choice. The analysis sheds light on recent regulatory move, reveals distributional implications masked by a technical doctrine, and opens up a horizon for more transparency and democratically legitimate algorithm governance.