Audit Annually to Catch Bias in Artificial Intelligence

Some human interaction needed to see if AI has disparate impact on minorities

Allen Smith, J.D. By Allen Smith, J.D. April 27, 2018

​Unchecked artificial intelligence (AI) can lead to the institutionalization of biases that result in less diverse recruitment, experts warn.

HR professionals should conduct periodic reviews of AI-driven recruitment and employee performance platforms to determine if biases result from the technology, according to Kriti Sharma, vice president of AI and ethics with Sage, headquartered in Newcastle, England.

The reviews should be conducted at least annually with an attorney, said Garry Mathiason, an attorney with Littler in San Francisco who is co-chair of its robotics, AI and automation industry group.

AI Is on the Rise

AI is shorthand for computer-science techniques that try to bring human-level intelligence to computers, said Julie Albright, a sociologist specializing in digital culture and communications at the University of Southern California in Los Angeles.

AI exhibits behaviors associated with human intelligence, like planning, learning, problem-solving, understanding speech and even conversing, Sharma noted. "If you've ever used keyword matching for pulling online job applications or predictive search on Google, or asked Siri what the weather is today or requested Alexa to play your getting-ready playlist, then you've used AI," she said.

HR professionals increasingly use AI for hiring, promotions, compensation decisions and employee surveys, Mathiason noted. There's even a new robot developed in Russia, "Vera," which vets resumes and conducts interviews, he said.

It's important to recognize the trend toward the use of AI because of the efficiency it offers, he said, describing HR as a "laggard" when it comes to the use of the technology. Only about 20 percent of HR departments use artificial intelligence. He predicted HR's percentage would be as high as 50 percent by 2025.

"This area of development is probably the most critical in the entire HR spectrum," Mathiason remarked, saying it affects everything that's done in HR.

"We're not at the precipice of a robot takeover," Sharma said. But automation is happening, she added. "Skilled manual-labor industries, as well as service jobs, will be seeing the most change from AI and automation, while jobs that require a higher emotional intelligence and keen soft skills will remain less affected," she said.

[SHRM members-only online discussion platform: SHRM Connect]

Albright cautioned that while AI has existed since the late 1950s, human thinking is difficult to emulate. There are nuances that aren't simple to replicate in a computer model but that people routinely process, she noted. "For example, we read the body language of others to determine their emotional state and assess threats. This kind of data is difficult at this point for computers to assess, though they will eventually get better at it."

Automated Bias

Mathiason said that AI can be less biased than people. But he cautioned that some of coders' biases may creep into AI algorithms.

Coders need to be a diverse group of people, Sharma said. But she added that "AI technology itself is evolving. We're getting closer to the point when AI starts to write code for itself."

This will make it more difficult to identify the sources of biases introduced into it, Mathiason noted.

AI looks for patterns, much like the human brain does, he said. If the technology matches successful employees with a zip code, it correlates a group of applicants with that zip code and puts an emphasis on job applicants in that same zip code, which could have a discriminatory impact based on segregated populations, he said.

Ultimately, there will be a "superartificial intelligence" that will monitor the process and ensure those in protected categories aren't discriminated against. But that's months away, he predicted. And even then, someone needs to look at AI results to ensure the right decisions are being made about hiring and promotions. The next step is a formal audit, he said.

Despite concerns that AI might result in bias, HR professionals need to familiarize themselves with AI, as its efficiency far outweighs its flaws, he emphasized. "If you don't move in this direction, it won't be too long [until] your company is obsolete," Mathiason said.


Was this article useful? SHRM offers thousands of tools, templates and other exclusive member benefits, including compliance updates, sample policies, HR expert advice, education discounts, a growing online member community and much more. Join/Renew Now and let SHRM help you work smarter.



Hire the best HR talent or advance your own career.


HR Daily Newsletter

News, trends and analysis, as well as breaking news alerts, to help HR professionals do their jobs better each business day.