Episode Details

A Brookings Institution study found something that single-axis bias audits are built to miss: when identical resumes were screened by AI, Black male candidates were selected zero percent of the time against white male candidates. Not a reduced rate — zero. The problem isn't racial bias or gender bias on their own. It's intersectional: the two compound, creating a discrimination tier that race-only and gender-only audits will never surface. In this episode we walk through three converging studies — Brookings, the FAIRE benchmark, and a Stanford field study of 3.4 million applicants across 150 employers — and explain why aggregate fairness numbers can look fine while the intersection quietly collapses to total exclusion. We also unpack algorithmic monoculture: when the same biased vendor is deployed across an entire sector, a rejected candidate can lose every door at once, with no alternative path. Then we get practical. Under Title VII, employers — not vendors — own the disparate impact of any screening tool they deploy. So we cover a four-part audit toolkit HR leaders can use right now: intersectional subgroup analysis, counterfactual resume testing, the vendor due-diligence questions that actually matter, and how to document your liability exposure. The takeaway isn't that AI is hopeless — audited systems often beat human selection on fairness — it's that you have to test at the intersection, because that's where the failures hide.