A recent study found that the appearance of an AI-powered virtual interviewer can influence job applicants' perceptions of fairness in the hiring process, even when the algorithm makes the same decision for all candidates.

The study showed that candidates whose applications were rejected evaluated the automated interview differently, depending on the gender and skin color of the virtual avatar that conducted the interview, indicating that interface design may be as influential as the algorithm itself.

Experiment with 220 participants

Approximately 220 people took part in virtual job interviews for a fictional customer service position, each interviewed by one of four realistic-looking virtual avatars that differed in gender and skin color.

Although all participants received the same rejection decision, their assessment of the fairness of the process was not identical; rather, it was influenced by the appearance of the interviewer on the screen, according to a report published by digitaltrends and reviewed by Al Arabiya Business.

Researchers argue that auditing algorithms alone may not reveal this issue, because applicants do not interact with code but with a virtual face that asks questions and announces the result.

Partial similarity increased feelings of unfairness

The results showed that participants who shared only one characteristic with the avatar—either gender or skin color—were most likely to consider the hiring process unfair, compared to those who matched both or neither.

The study did not definitively determine the cause, but researchers suggest that partial similarity may raise applicants' expectations of the avatar, making the rejection feel more personal.

This finding indicates that giving a virtual avatar familiar features does not necessarily make applicants perceive it as more neutral.

Trust declines after rejection announcement

Before the result was announced, the study found that trust levels in the AI system were similar among all participants, regardless of the avatar's appearance.

However, eye-tracking data revealed that participants focused longer on faces with a different skin color from their own.

After the rejection announcement, participants became more suspicious of the system's fairness, and the likelihood of perceiving the decision as biased increased when the avatar's skin color differed from the applicant's.

This means the automated decision did not change, but the way it was presented affected how applicants interpreted it.

Results need broader testing

Researchers note that the experiment used a fictional job and a uniform rejection decision, so it does not prove that all real hiring systems would yield the same results.

However, it demonstrates how quickly people's feelings of fairness can change when an automated decision becomes a personal interaction with a virtual avatar.

What should companies do?

The study recommends that companies using AI interviewers should focus not only on the AI model's accuracy but also on the design of the user interface and the virtual avatar representing the system.

It also calls for testing these systems with applicants from different demographic backgrounds, comparing their reactions before and after the results are announced, and evaluating whether less human-like avatars provoke fewer concerns than highly realistic ones.

Researchers conclude that the best design for an AI interviewer may not be the most realistic, but rather the most clear and transparent, helping users understand the system's nature and enhancing their sense of fairness and trust.

Sponsored Content

Sponsored Content