Largest study of AI hiring algorithms to date finds 'clear racial disparities'

Key Points:

• A study by Stanford, Chapman, and Northeastern universities analyzed over 4 million job applications screened by Pymetrics’ algorithm, revealing significant racial disparities in hiring outcomes, particularly disadvantaging Black and Asian applicants.
• The research found that assessing bias at the individual job position level, rather than aggregating data across positions, exposes discriminatory impacts masked by broader analyses, with over 10% of jobs showing adverse effects against Black applicants.
• The study identified a "systemic rejection" or "algorithmic blackball" effect, where applicants rejected by one company using Pymetrics are statistically more likely to be rejected by others, as the same algorithmic scores are reused across employers for up to 330 days.
• The concentration of AI hiring tools among a few dominant vendors poses systemic risks, including widespread disruption if a vendor fails or is found discriminatory, affecting thousands of employers simultaneously.
• The authors urge policy changes including position-level bias measurement, enhanced market surveillance, monitoring of vendor concentration risks, and legal access for independent researchers to hiring algorithm data to ensure transparency and fairness.