Lily Manshel

Volume 29.1 (download PDF)

Abstract

As one-click applications and a competitive job market result in hundreds of applicants per listing, hiring tools that use Artificial Intelligence (“AI”) promise recruiters a convenient way to sort through the crowd and find the “perfect” candidate, all while eliminating human bias. These tools—which include resume screeners, gamified assessments and personality tests, and video interviewing software—are proliferating so rapidly that their ubiquity is positioned as inevitable by both the software companies that produce them and the employers who use them. However, the use of these automated decision-making tools creates a paradox under existing antidiscrimination law: the systems clearly perpetuate and accelerate existing human biases, but the mechanisms of that bias are nearly impossible to prove due to the complexities of machine learning and the proprietary nature of the underlying algorithms. This “black box” problem shields both discriminatory intent and disparate impact from both public scrutiny and legal inquiry, despite ongoing negative effects on minority groups, especially applicants with disabilities.

By examining theoretical approaches to AI regulation and evaluating the effectiveness of recent regulatory efforts at the state and local level, this Note argues that the “disclosure and consent” models of regulation currently in use are largely unenforceable and do not address the power imbalance between employers and applicants. Instead, any regulatory approach to automated hiring tools must include a federally enforced affirmative duty for developers to empirically prove that their tools are unbiased before they are ever deployed in the marketplace.