Companies are increasingly using AI to conduct job interviews, and, according to experts in the field, the technology is leading to some impressive results. However, giving candidates the choice between an AI interviewer or a human can create bias that makes landing a job tougher for some people, according to a new report.

AI is now a common part of the job application process. According to the World Economic Forum, around 88% of employers use some form of AI for initial candidate screening such as filtering or ranking job applications. But AI is also being used to conduct interviews. Currently, around 21% of U.S. companies use the technology for initial interviews. 

AI interviewers can give companies an edge when during the hiring process. One study found that candidates who were interviewed by an AI were more likely to land a job than candidates who were sourced by humans screening résumés: 54% of candidates interviewed by AI got the job, compared to about 29% of candidates sourced by a traditional résumé screening.

Still, there is a lot to learn about how utilizing AI interviews impacts both people and firms. Brian Jabarian, a researcher at the University of Chicago Booth School of Business with doctorates in economics and philosophy, recently examined what happens to candidates when they are offered a choice between an AI interviewer and a human interviewer, which he detailed in his paper, Choice as Signal: Designing AI Adoption in Labor Market Screening. The research, which has not been peer reviewed, finds giving candidates a choice between a human and AI interview could also create a new hurdle for low-ability candidates—applicants whose skills are below the firm’s hiring threshold.

Jabarian tells Fast Company that different applicants will automatically be drawn to either AI interviewers or human interviewers based on their strengths. For example, “applicants with strong language skills prefer human interviewers to highlight their English proficiency,” he says. “In contrast, applicants with strong analytical skills choose the AI interviewer to showcase their quantitative strengths.”

But the choice isn’t neutral, like a candidate may expect it to be. “An applicant’s decision to be interviewed by a human or an AI agent can reveal private information about their strengths, weaknesses, or expectations for relative performance,” Jabarian writes in his paper, also pointing out that employees with high abilities benefit because companies can identify them more easily “using both the signal and the selection decision, increasing their probability of being hired.”

However, it also means firms are able to more easily identify low-ability workers. Jabarian writes: “Consequently, low-skilled workers succeed less often in obtaining a job and therefore experience a welfare loss.” Essentially, by interpreting both the choice itself as well as the information from the interview, an employer’s precision increases, which doesn’t serve lower-ability candidates. 

Jabarian says if firms had no insight into the candidate’s choice, then all workers would have the advantage of choosing which interviewer best shows their skill set, but companies would lose out on the advantages of using AI interviewers.

While on the surface giving job candidates choices about how they are interviewed seems like a solid idea, Jabarian says that in practice, it’s not quite so simple. “Before this new paper, I was really rooting for giving this choice to people because I was confused about why everyone was assuming it was just okay to impose a new technology on people in a high-stakes environment when they maybe didn’t want it,” he explains. However, now he believes it’s clear that the choice alone hurts the weakest candidates, and therefore it shouldn’t be one that is routinely offered but rather “on a case-by-case basis.” 

Jabarian says he expects AI interviewing to increase, particularly because it’s good for firms. Still, that doesn’t mean humans as interviewers are a thing of the past or irrelevant. AI interviewers and humans have different strengths: Human recruiters can improvise and are able to vary their interviews, while AI creates a consistent experience and is excellent at garnering information from candidates. That means adopting hybrid techniques—where humans and AI run interviews with opposing purposes—might really be the smartest and fairest way to hire.