MIT Technology Review reports on research examining bias in AI systems used for hiring. As applicants increasingly submit résumés that may be screened by AI before any human reviewer sees them, questions arise about whether these systems evaluate candidates fairly.
Researchers have already established that large language models pick up human biases present in their training data. The new research described here suggests an additional concern: that LLMs can also develop biases of their own, beyond those inherited from the data they were trained on.
Why it matters
If AI tools used in recruitment introduce biases that are not simply reflections of their training data, this complicates efforts to ensure fair evaluation of job applicants. It points to a source of unfairness that may be harder to identify and address.
Who should care
Job seekers whose applications may be screened by AI, as well as employers and hiring platforms relying on such tools, have reason to consider these findings.