Stanford researchers find AI hiring algorithm discriminates against Black and Asian job applicants across multiple employers

Stanford researchers analyzed a dataset from pymetrics (a talent platform acquired by Harver in 2022) spanning December 2018 through December 2022, containing 4,197,168 job applications from 3,372,132 applicants to 1,746 positions across 156 employers with a total annual revenue of $225 billion. They found that 26 percent of Black applicants and 15 percent of Asian applicants applied to positions where the AI system discriminated against their racial group, using the US Equal Employment Opportunity Commission's four-fifths rule as the standard.

The pymetrics algorithm recommends on average 58.2 percent of applicants per position for employers to interview. The researchers argue that when the same algorithm is used across multiple employers, job seekers who apply to four positions at different companies using this shared platform face a 10 percent rejection rate from all applications—a pattern that does not appear in hiring without AI algorithms.

If Black and Asian candidates had been advanced at the same rate as the most favored group, about 40,000 more job candidates would move on to the next screening stage. The researchers contend that discrimination occurs job-by-job but can be masked when recommendations are averaged across all positions.