Keith Swisher (University of Arizona – James E. Rogers College of Law) has posted Classifying and Countering AI Proxy Discrimination, Stanford L. Rev. Online (forthcoming) on SSRN. Here is the abstract:
AI is opaquely producing a civil-rights crisis. AI systems increasingly shape who gets hired, housed, financed, insured, educated, surveilled, and detained, and these systems usually do not classify people by race, sex, or disability in the way civil rights law expects. Instead, they rely on proxies and emergent categories that can replicate prohibited discrimination without ever triggering the increasingly outdated letter of the law. This growing proxy discrimination reveals a critical mismatch between civil-rights doctrine and algorithmic decisionmaking. The protected-class model assumes discrimination relies on existing categories, but machine learning evades that assumption. It classifies on its own terms, often through data patterns so diffuse and technical that existing doctrine fails.
To remedy the mismatch, the Essay proposes three interdisciplinary reforms: an epistemological test for proxy discrimination grounded in mutual information; classification impact assessments, modeled on pre-deployment review in environmental law, conducted before high-stakes decisions; and a fairness-by-design rule making low information entanglement with protected classes the default (rather than the later burden of the victims). These proposals honor civil rights law’s foundational commitments but update the regulatory approach for a world in which discrimination is increasingly statistical, evolving, and obscure to human reviewers.
Recommended!
To receive new posts from Legal Theory Blog by email, get a free subscription to Legal Theory Stack.
Lawrence Solum