Study finds AI hiring tests can create racial gaps in job screening

In short: A large study found that some AI-based job screening tests reject Black and Asian applicants more often, and the same applicant can be rejected across different companies using the same model.

What's going on

Employers are increasingly using AI tools to sort job applications before a person looks at them. One common approach uses short online games that measure things like reaction speed and willingness to take risks.

A Stanford-led research team studied 4 million job applications submitted through the Pymetrics platform between December 2018 and December 2022. The data covered 156 employers, most with annual revenue above $5 billion. The researchers say this is the largest review of AI hiring systems so far.

The study found “clear racial disparities” in outcomes for some roles. About 1 in 10 positions showed “adverse impact” against Black applicants, and about 1 in 20 did so for Asian applicants. “Adverse impact” is a US legal term that means one group is picked at a much lower rate than the most selected group.

The researchers also found signs of what they called “systemic rejection.” Some employers were using identical screening models. That is like different stores using the same bouncer with the same rulebook. If an applicant failed one company’s test, they could be more likely to fail at another company using the same model, because their score would be identical.

What to watch

The researchers caution that results may not apply to every type of automated hiring, such as resume scanning. Still, the findings add pressure on employers and vendors to check these systems for unfair patterns, and to explain how decisions are made when a computer is screening people out.

Source: Financial Times