Brookings Warns AI Hiring Bias Goes Deeper Than the Tools — It's in the Training Pipeline
The News
HR Brew reports that Nicol Turner Lee, senior fellow and director of the Center for Technology Innovation at the Brookings Institution, addressed AI bias in hiring at the Meltzer Center for Diversity, Inclusion and Belonging at NYU School of Law. Turner Lee argued that bias originates not only in AI tools but in the development pipeline itself: 'these technologies are not necessarily designed by us… These are actually technologies being created and controlled by a handful of companies.' She specifically flagged deep learning models underlying pre-hiring video interview assessment, which analyse facial expressions for traits like smiling and eye contact, and said these systems 'omit and exclude Black and Latina applicants because they're perceived as not being friendly based on the emotional cognition of their face.' Turner Lee framed 'digital invisibility' as extending beyond broadband access to include people ignored in AI development and by AI systems. She concluded: 'It gets more complicated with AI.'
Why It Matters
This matters because the legal and compliance conversation around AI in hiring has focused heavily on disclosure and audit obligations — NYC Local Law 144, the EU AI Act, EEOC guidance. Turner Lee's framing points to a deeper problem: even a fully disclosed, audited tool can produce discriminatory outcomes if the underlying model was trained on data that excludes or misrepresents certain populations. For TA leaders, this means vendor diligence questions need to go beyond 'is this tool compliant?' to 'what data was this model trained on, and whose faces, voices, and patterns are missing from that training set?'
Taking Action
All action is situated in your corporate context. These are starting points, not prescriptions — choose what fits.
Connect
Who to talk with in your enterprise
Your DEI lead: Share Turner Lee's specific concern about video interview analysis and emotional cognition. If you use any video assessment tool, your DEI lead should be part of the vendor evaluation conversation, not informed after the contract is signed. Your HR technology procurement lead: Ask current vendors of any AI-assessment tools for documentation of their training data demographics. If they can't answer, that's your answer.
Inspect
What to take a closer look at
If your organisation uses AI-assisted video interviewing or facial/voice analysis tools, request the vendor's most recent bias audit results — not the marketing summary, the actual methodology and demographic breakdown. Check whether the audit tested for bias against the specific populations Turner Lee identified. If it didn't, the audit is incomplete.
Reflect
What to think about more deeply
You can audit a tool's outputs all day, but if the model was trained on a world that doesn't include your candidates, what exactly is the audit measuring?