Composite scenario: the resume-screening algorithm

EDITORIAL NOTE: This is a composite scenario, not a record of a specific person's case. It draws on the EEOC's 2023 guidance ``Select Issues: Assessing Adverse Impact in Software, Algorithms, and Artificial Intelligence Used in Employment Selection Procedures``, and on the 2020 audit by Pymetrics, HireVue, and two academic groups demonstrating disparate impact in resume-screening models.

An employer adopts a resume-screening model trained on the company's own ten-year hiring record. The model learns to predict the likelihood of a successful hire — and, in doing so, learns the demographic patterns of past hires, including who was screened out before reaching the hiring manager.

The model develops weights for variables that correlate with race without being explicitly racial: graduation year, ZIP code of current address, names of high schools or community colleges, membership in affinity-named organizations on the resume. Two applicants with identical credentials but different demographic signals receive different scores.

Cathy O'Neil's ``Weapons of Math Destruction`` (2016) and Safiya Umoja Noble's ``Algorithms of Oppression`` (2018) frame the policy question. The 2023 EEOC guidance treats algorithmic selection as actionable under Title VII's disparate-impact standard, the same standard ``Griggs v. Duke Power`` (1971) established for written employment tests. The mechanism is new; the legal framework is fifty years old.

The use of artificial-intelligence-driven hiring tools has expanded substantially across American employers over the past decade. The principal categories of such tools include automated resume-screening systems, video-interview-analysis systems that score candidates on speech, facial expression, and behavioral patterns, and predictive-analytics systems that use historical hiring data to score candidate-job fit. The Society for Human Resource Management's 2024 survey estimates that approximately one in four American employers uses some form of AI-driven hiring tool, with adoption concentrated among larger employers and in specific industries (technology, financial services, retail, healthcare).

The principal regulatory framework addressing the disparate-impact effects of AI hiring tools is the federal Equal Employment Opportunity Commission's enforcement authority under Title VII of the Civil Rights Act of 1964. The EEOC's 2023 technical assistance document on the use of AI in employment decisions and the EEOC's 2024 settlement with iTutorGroup (the first federal AI-hiring-discrimination settlement) have established the operational framework: AI-driven hiring tools are subject to the same Title VII disparate-impact analysis as traditional hiring tools, and employers using AI tools are required to validate the tools' operational practice against the four-fifths rule and the broader disparate-impact framework.

The Bertrand and Mullainathan resume-audit study (2004) and the subsequent replication literature provides the empirical context for the current AI hiring discussion. The pre-AI callback differential between Black-name and white-name resumes was approximately fifty percent. The Quillian, Pager, Hexel, and Midtboen (PNAS, 2017) meta-analysis found no statistically significant decline in hiring discrimination against Black Americans across the 1989-2015 study period. The pre-AI hiring system was already producing substantial racial-disparate outcomes; the question with respect to AI hiring tools is whether the tools mitigate, replicate, or amplify the existing disparate patterns.

The principal mechanisms by which AI hiring tools can produce racial-disparate outcomes are several. Training data drawn from historical hiring decisions encodes the racial-disparate patterns of the historical hiring decisions. Proxy variables — ZIP code, name, alma mater, gap-year status, language patterns — can produce racial-disparate outcomes even when race is not directly used as a model input. Threshold effects in scoring systems can produce binary-classification differentials that compound across the hiring pipeline. Video-interview-analysis systems have been the subject of particular controversy because the facial-expression-and-speech-pattern features they use have documented racial-demographic correlations.

The state-level regulatory response has expanded substantially. Illinois's Artificial Intelligence Video Interview Act (2019) requires employer disclosure of AI use in video interviews and applicant consent before AI-driven analysis. Maryland's HB 1202 (2020) and New York City's Local Law 144 (2021, effective 2023) impose additional disclosure and bias-audit requirements. The platform's pathways pages cover the principal EEOC and state-agency intake routes for individual employment-discrimination complaints involving AI hiring tools.

The contemporary AI-hiring-discrimination regulatory framework has expanded substantially across recent years. Illinois's Artificial Intelligence Video Interview Act (2019), New York City's Local Law 144 (2021, effective 2023), Maryland's HB 1202 (2020), Colorado's SB 205 (2024), and California's parallel pending legislation establish state-level regulatory frameworks for AI-hiring-tool operation. The cumulative state-level framework has produced documented changes in the operational practice of AI-hiring-tool deployment in the affected jurisdictions. The federal EEOC enforcement framework continues to evolve across successive iterations of agency guidance and enforcement actions.

The principal civil-society organizations addressing AI-hiring-discrimination include the ACLU's Project on Speech, Privacy, and Technology, the Center for Democracy and Technology, the AI Now Institute at New York University, and the broader network of digital-civil-rights organizations. The American Bar Association's 2023 statement on AI and employment discrimination, the parallel academic-society statements, and the successive subsequent guidance documents have shaped the contemporary professional engagement with the question. The platform's pathways pages cover the principal EEOC and state-agency intake routes for individual AI-hiring-discrimination complaints.