Last week, I wrote about bias in AI and how the risk of bias using these tools is exponentially greater than human bias (no matter what that AI vendor is telling you). You can find that article here . There shouldn’t be a fear of using AI – the benefits far outweigh the risks. Today I’ll build on the topic by sharing how bias can enter the system and present a business case for change.
Where Bias Actually Enters the Pipeline
To conduct an effective ethics audit, talent leaders need to understand the specific mechanisms through which bias can enter the system. It rarely enters through a single door.
Training data contamination. This is the most common issue. Resume filters that prioritize candidates who resemble prior hires will perpetuate whatever demographic skews existed in past hires, including skews caused by the very discrimination the tool was meant to eliminate. If your historical hires in engineering roles were predominantly male, the algorithm learns to prefer male-oriented signals.
Proxy variables. Algorithms are clever about circumventing protected characteristics. They learn to use zip codes, commute distances, college names, and even vocabulary patterns as stand-ins for race, age, or socioeconomic status, none of which appear to be protected characteristics on the surface, but all of which can constitute unlawful disparate impact discrimination if they correlate with protected class membership.
Video assessment platforms. Tools like HireVue that analyze facial expressions, vocal cadence, and speech patterns have been found to disadvantage candidates with disabilities, non-native accents, and atypical communication styles. Research indicates these platforms are also less accurate in assessing expressions from individuals who don’t conform to binary gender norms, as the underlying models were trained predominantly on individuals whose gender identity matches the sex they were assigned at birth.
Intersectional blindspots. Perhaps the most insidious issue: most bias audits test for race and gender separately. But research consistently shows that Black women, Black men, women with disabilities, and other overlapping socially-defined groups face discrimination that wouldn’t be visible from single-axis testing. California’s recognition of these groups as legally protected identities in 2024 signals where anti-discrimination enforcement is heading.
“Just because vendors say their AI is fair doesn’t make it so.” – ACLU complaint to the FTC regarding AON Consulting’s hiring tools, May 2024
The Business Case Is Inseparable from the Ethical Case
What I’m writing about isn’t just a compliance risk management task. The strongest argument for getting this right isn’t legal – it’s competitive.
A 2024 Gallup survey found that 93% of Fortune 500 CHROs are integrating AI into business practices. Yet only about one-third of employees knew their employer uses AI tools in hiring or management. That awareness gap is closing rapidly. Candidates are increasingly sophisticated about algorithmic screening and they’re becoming increasingly vocal when they believe they’ve been filtered out unfairly. The Arshon Harper case vs. Sirius XM Radio, in which a candidate applied to approximately 150 IT positions and received 150 rejections, alleging that the AI screening system used zip codes and educational institutions as proxy variables for race, illustrates a pattern that is becoming more visible and more litigated by the month.
The organizations that will win the talent competition in the coming decade are those that build screening processes that are genuinely based on merit, not just defensible in court. Biased filters don’t just expose you to liability. They cost you talented candidates who didn’t match a flawed historical template.
The Recruiting Ethics Audit Checklist
Interested in a free 7-step audit checklist to assess your organization’s readiness for AI-assisted hiring? Email info@estalentsolutions.com with subject “AI Checklist” and it’s headed your way.
I regularly recommend my clients audit their recruiting process every 6 months to ensure they are getting the full value from their technology and team. The AI ethics audit is different. It’s not a six-month checkbox exercise. It is an ongoing commitment to interrogating the systems we’ve built with the same rigor we apply to financial controls or cybersecurity. It’s clear to see the tools we trusted to remove human bias from hiring often do no such thing. What happens next depends entirely on what we choose to do with that knowledge.
The question isn’t whether your AI hiring tool has inherited historical bias. The more useful question is “Are you auditing hard enough to find it?”
Have you seen other examples of bias entering the recruiting process? Share your thoughts in the comments below.
ES Talent Solutions helps organizations navigate the intersection of recruiting strategy and emerging technology. Want to discuss how agentic AI could transform your talent acquisition function? Contact Eddie Stewart at estewart@ESTalentSolutions.com. I’m always happy to talk with fellow leaders about building recruiting functions ready for the future.
0 Comments
Interview Intelligence: AI Watching, Scoring, and Judging Your Candidates
Making the case for AI in the interview process and understanding where and why to use in are important in creating recruiting departments of the future
Say Goodbye to “Figure it Out Yourself” HR. Say Hello to Wowledge CoCreator™
Wowledge's new tool, CoCreator, is a must have for human resources professionals.
The Ethics Audit: How to Ensure Your Automated Screening Tools Aren’t Biased (Part 1)
As automated tools play a larger part of the interview assessment process, companies need to understand the implications of using such tools and how to ensure they're doing what they're supposed to.