A day after her interview for a part-time job at Target last year, Dana Anthony got an email informing her she didn’t make the cut.
Anthony didn’t know why — a situation common to most job seekers at one point or another. But she also had no sense at all of how the interview had gone, because her interviewer was a computer.
More job-seekers, including some professionals, may soon have to accept impersonal online interviews where they never talk to another human being, or know if behind-the-scenes artificial-intelligence systems are influencing hiring decisions. Demand for online hiring services, which interview job applicants remotely via laptop or phone, mushroomed during the COVID-19 pandemic and remains high amid a perceived worker shortage as the economy opens back up.
These systems claim to save employers money, sidestep hidden biases that can influence human recruiters and expand the range of potential candidates. Many now also use AI to assess candidate skills by analyzing what they say.
Anthony likes to look an interviewer in the eyes, but all she could see was her face reflected in the screen. “I interview better in person because I’m able to develop a connection with the person,” she said.
But experts question whether machines can accurately and fairly judge a person’s character traits and emotional signals. Algorithms tasked to learn who’s the best fit for a job can entrench bias if they’re taking cues from industries where racial and gender disparities are already prevalent.
And when a computer screens out some candidates and elevates others without explanation, it’s harder to know if it’s making fair assessments. Anthony, for instance, couldn’t help wondering if her identity as a Black woman affected the decision.
“If you apply for a job and are rejected because of a biased algorithm, you certainly won’t know,” said Oxford University researcher Aislinn Kelly-Lyth. In a face-to-face interview, by contrast, a job seeker might pick up discriminatory cues from the interviewer, she said.
New rules proposed by the European Union would subject such AI hiring systems to tighter regulation. Advocates have pushed for similar measures in the U.S.
One of the leading companies in the field, Utah-based HireVue, gained notoriety in recent years by using AI technology to assess personality and job skills from an applicant’s facial expressions during the interview. After heated criticism centered on the scientific validity of those claims and the potential for bias, the company announced earlier this year it would end the practice.
But its AI-based assessments, which rank the skills and personalities of applicants to flag the most promising for further review, still consider speech and word choices in its decisions.
The privately owned company helped create a market for “on-demand” video interviews. Its known customers have included retailers like Target and Ikea, major tech companies like Amazon, banks like JP Morgan and Goldman Sachs, oil giants, restaurant chains, supermarkets, airlines, cruise lines and school districts. The Associated Press reached out to numerous brand-name employers that use the technology; most declined to discuss it.
HireVue CEO Kevin Parker says the company has worked hard to ensure its technology won’t discriminate based on factors such as race, gender or regional accents. Its systems, which translate speech to text and sift for clues about team orientation, adaptability, dependability and other job skills, can outperform human interviewers, he said.
“What we’re trying to replace is people’s gut instinct,” he said in — naturally — a video interview.
HireVue says it interviewed more than 5.6 million people around the world in 2020. Supermarket chains used it to screen thousands of applicants a day amid a pandemic-fueled hiring surge for cashiers, stockers and delivery crews, Parker said.
Providers of broader hiring-focused software such as Modern Hire and Outmatch have started offering their own video interviews and AI assessment tools. On its website, Outmatch touts its ability to measure “the must-have soft skills your candidates and employees need to succeed.”
HireVue notes that most customers don’t actually use the company’s AI-based assessments. Atlanta’s school district, for instance, has used HireVue since 2014, but says it relies on 50 human recruiters to score recorded interviews. Target said the pandemic led it to replace in-person interviews with HireVue interviews, but the retail giant told the AP it relies on its own employees — not HireVue’s algorithms — to watch and evaluate prerecorded videos.
None of that was clear to Anthony when she sat down in front of a screen to interview for a seasonal job last year. She dressed for the occasion and settled into a comfortable spot. The only hint of a human presence came in a prerecorded introduction that laid out what to expect — noting, for instance, that she could delete an answer and start over.
But she had no way to know what sort of impression she was creating. “We’re unable to provide specific feedback regarding your candidacy,” Target’s rejection email said. She was rejected again after completing a HireVue interview for a different job in December.
“I understand companies or organizations trying to be more mindful of the time and the finances they spend when it comes to recruitment,” said Anthony, who obtained a master’s degree in strategic communications last year at the University of North Carolina at Chapel Hill. Still, the one-way interviews left her uneasy about who, or what, was evaluating her.
That inscrutability poses one of the biggest concerns about the rapid growth of complex algorithms in recruitment and hiring, Kelly-Lyth said.
In one infamous example, Amazon developed a resume-scanning tool to recruit top talent, but abandoned it after finding it favored men for technical roles — in part because it was comparing job candidates against the company’s own male-dominated tech workforce. A study released in April found that Facebook shows different job ads to women and men in a way that might violate anti-discrimination laws.
Governments across the U.S. and Europe are looking at possible checks on these hiring tools, including requirements for outside audits to ensure they don’t discriminate against women, minorities or people with disabilities. The proposed EU rules, unveiled in April, would force providers of AI systems that screen or evaluate job candidates to meet new requirements for accuracy, transparency and accountability.
HireVue has begun phasing out its face-scanning tool, which analyzed expressions and eye movements and faced derision by academics as “pseudoscience” reminiscent of the discredited and racist 19th century theory of phrenology. The Electronic Privacy Information Center filed a complaint in 2019 with the Federal Trade Commission, citing a HireVue executive who had said 10% to 30% of a candidate’s score was based on facial expressions.
“The value it was adding related to the controversy it was creating wasn’t very much,” Parker told the AP.
HireVue also released portions of a third-party audit that examined fairness and bias issues around its automated tools. A published summary recommended minor changes such as modifying the weight given to the especially short answers disproportionately provided by minority candidates.
Critics welcomed the audit but said it was merely a start.
“I don’t think the science really supports the idea that speech patterns would be a meaningful assessment of someone’s personality,” said Sarah Myers West of New York University’s AI Now Institute, which studies the social implications of AI. For instance, she said, such systems have historically had trouble understanding women’s voices.
Kian Betancourt, a 26-year-old who is pursuing a doctorate in organizational psychology at Hofstra University, also failed a remote HireVue interview for a consulting position earlier this year. He acknowledged that he might have tried too hard to predict how the system would evaluate him for a consultancy job, tailoring his diction to include keywords he thought might boost his score.
While Betancourt is supportive of “structured interviews” involving a standard set of questions, he’s bothered by the opacity of automated systems.
“Tell people exactly how we’re being evaluated, even if it’s something as simple as, ‘This is an AI interview’,” he said. That basic information can affect how people present themselves, he said.