Ex-EEOC commissioner: AI in HR could cut bias in hiring
Keith Sonderling, a former U.S. Equal Employment Opportunity Commission member, argues at the HR Technology Conference that properly implemented AI can reduce bias in hiring.
In the ongoing debate on the risk of AI in HR, Keith Sonderling, whose term on the U.S. Equal Employment Opportunity Commission ended in August, believes that issues with bias lie more with humans than with AI. But that's a different take than the EEOC's public communications.
The commission's warnings about AI's potential to exacerbate bias are at the forefront of its messaging to employers. It has described some HR vendor tools as "snake oil." It has teamed up with the Department of Justice to warn about AI's potential for discrimination.
Sonderling, an employment law attorney, isn't denying the potential bias risks of AI in HR. Addressing those risks is a central part of his message. But in an interview and in a presentation at last week's HR Technology Conference, Sonderling also made clear that humans are the biggest source of discrimination -- something AI in HR might help fix.
AI can absolutely remove bias.
Keith SonderlingEmployment attorney, former U.S. Equal Employment Opportunity Commission member
AI that is carefully designed and properly used "can absolutely remove bias," Sonderling said during a conference presentation. Sonderling was appointed to the EEOC by former President Donald Trump and confirmed by the Senate in 2020. His term as commissioner expired in August.
While the EEOC's mission is to enforce federal laws against employment discrimination, differences in perspectives can emerge among its commissioners. Last year, Sonderling criticized the EEOC for holding a hearing on the risks of AI-based systems without including input from AI designers and systems builders. He argued that it's just as important to highlight AI's potential to mitigate unlawful discrimination as it is to warn about its regulatory pitfalls.
If the AI in the HR system is not properly designed or used, "it can scale discrimination to the likes that we have never seen before," Sonderling said. And it's the employer, not the HR systems vendor, that will be held liable for bias under law, he noted.
A tension in HR
There is tension between AI's promise and risks, which was evident at HR Tech, where some presenters noted the caution from their company's legal advisers.
The risk for HR is clear, said Betsy Summers, an analyst at Forrester Research, but the focus for some buyers is on how quickly they can get productivity benefits. HR vendors are pitching an attractive "speed to value," or how quickly a system delivers results.
"I would slow down," Summers said. "I don't think that speed to value is the right selling point here."
HR users have to understand what the AI is doing, how the algorithm was trained and on what data, Summers said. Users might have to bring in auditors once a quarter to look at the data to determine if a system is having disparate impact.
Rebecca Wettemann, an analyst at Valoir, said the technology analyst firm is "seeing real hurdles" to AI adoption around legal and regulatory compliance. Vendors should be able to explain algorithmic data and training questions "to your satisfaction, not to their satisfaction," she said.
Private and academic studies support Sonderling's point about human bias in hiring. For instance, a recent major academic study that involved sending out 84,000 fictitious resumes found that people with white-sounding names were more likely to get a callback than those with Black-sounding names.
Sonderling argued that it might be easier to demonstrate bias by an algorithm than by a human.
"When you talk about the black box of the algorithm, what about the black box of the human brain?" Sonderling said.
Unlike a human brain, a well-designed AI system will "give you a contemporaneous record of what factors went into that algorithm to make that employment decision," he said.
Patrick Thibodeau is an editor at large for TechTarget Editorial who covers HCM and ERP technologies. He's worked for more than two decades as an enterprise IT reporter.
