A Stanford-led study of the Pymetrics AI hiring platform, using 4 million applications from 156 employers between late 2018 and late 2022, finds systemic racial biases in screenings with about 10% adverse impact for Black applicants and around 5% for Asian applicants, disproportionately affecting those groups in high-volume roles.

The research notes that candidates applying to multiple jobs screened by the same vendor can be repeatedly rejected across roles; for example, 10% of applicants submitting four applications were rejected from all of them.

Using the EEOC four-fifths rule, the authors estimate roughly 40,000 additional candidates would advance if Black and Asian applicants were treated at the same rate as the most favored group.

The authors acknowledge prior findings that algorithmic systems can reproduce or amplify biases, but contend their results show harms can emerge without explicit demographic data.

They urge accountability for hiring tools and caution that discriminatory AI recommendations can be subtle, arising from biased training data, underscoring the need for oversight and further study.

The study adds to global scrutiny of automated hiring, calling for more careful examination and potential reforms in how AI tools are developed and deployed across industries.

Overall takeaway: AI hiring tools can reproduce or amplify human biases, with increasing attention on regulatory and voluntary safeguards for fairness, explainability, and compliance in high-volume recruitment.

Pymetrics uses game-based assessments to gauge traits like risk-taking and social behavior; the firm is now owned by Harver, with candidates aligning closely to top performers more likely to advance.

The study examined Pymetrics’ neuroscience and machine-learning screening via mini-games and found persistent biases despite claims of objectivity.

Discrimination patterns persist even without explicit demographic data, suggesting the models exploit proxies for demographics such as zip codes or school affiliations.

Related findings include an AI-hiring study showing bias against women and a lawsuit alleging discriminatory screening against candidates, highlighting broader gender and wage biases.

Researchers warn that dependence on a single AI vendor could amplify flaws across the hiring ecosystem, even as platforms vary in methods like resume review or video interviews.