Study finds extensive racial bias by AI hiring tools

Employers now overwhelmingly use artificial intelligence (AI) in screening and selecting candidates to hire. But a new study from Stanford revealed that the most popular AI platforms routinely discriminate against racial minorities.

More than 90% of employers use some form of AI screening tool to filter or rank job applications, according to the study; most rely on the same few third-party AI vendors.

“We find substantial evidence of racial disparities in AI-based candidate screening,” the authors of the study wrote.

The study analyzed data for 3.4 million real job applicants who submitted 4 million applications to 156 employers across 11 market sectors. Every application was assessed by algorithms from the same vendor. When the same algorithm is used across multiple employers, the adverse impact of a biased AI system is massively expanded, the study demonstrated.

To measure adverse impact, the study applied the Equal Employment Opportunity Commission (EEOC) “four-fifths rule,” which flags a position when one group is recommended at less than 80% of the rate of the most-recommended group. The findings showed that the AI platform screened out 26% of Black applicants and 15% of Asian applicants, more than meeting the EEOC’s threshold.

Employers are ultimately responsible for all hiring decisions and should:

• Understand the AI algorithm used and monitor results
• Partner with the vendor to test the system for biases and review actual hiring results
• Build in human oversight and checkpoints

Furthermore, when applicants apply to multiple positions, 10% experienced repeated rejection across employers using the same system. This highlights the risk of a single AI algorithm, used by multiple employers, producing discriminatory outcomes across an industry.

Employers need to understand the AI algorithm used and monitor results at the individual job level where bias can be exposed. Employers, who are ultimately responsible for all hiring decisions, should not rely solely on vendors’ claims of fairness; instead, they should partner with the vendor to test the system for biases and review actual hiring results. It is critical to also build in human oversight and checkpoints, especially in high-volume or entry level roles where AI screening is used most often.