An artificial-intelligence hiring system has become a powerful gatekeeper for some of America’s most prominent employers, reshaping how companies assess their workforce – and how prospective employees prove their worth.
Designed by the recruiting-technology firm HireVue, the system uses candidates’ computer or cellphone cameras to analyze their facial movements, word choice and speaking voice before ranking them against other applicants based on an automatically generated “employability” score.
HireVue‘s “AI-driven assessments” have become so pervasive in some industries, including hospitality and finance, that universities make special efforts to train students on how to look and speak for best results. More than 100 employers now use the system, including Hilton, Unilever and Goldman Sachs, and more than a million job-seekers have been analyzed.
But some AI researchers argue the system is digital snake oil – an unfounded blend of superficial measurements and arbitrary number-crunching, unrooted in scientific fact. Analyzing a human being like this, they argue, could end up penalizing nonnative speakers, visibly nervous interviewees or anyone else who doesn’t fit the model for look and speech.
The system, they argue, will assume a critical role in helping decide a person’s career. But they doubt it even knows what it’s looking for: Just what does the perfect employee look and sound like, anyway?
“It’s a profoundly disturbing development that we have proprietary technology that claims to differentiate between a productive worker and a worker who isn’t fit, based on their facial movements, their tone of voice, their mannerisms,” said Meredith Whittaker, a co-founder of the AI Now Institute, a research center in New York.
“It’s pseudoscience. It’s a license to discriminate,” she added. “And the people whose lives and opportunities are literally being shaped by these systems don’t have any chance to weigh in.”
Loren Larsen, HireVue’s chief technology officer, argues that such criticism is uninformed and that “most AI researchers have a limited understanding” of the psychology behind how workers think and behave.
Larsen compared algorithms’ ability to boost hiring outcomes with medicine’s improvement of health outcomes and said that the science backed him up. The system, he argued, is still more objective than the flawed metrics used by human recruiters, whose thinking he called the “ultimate black box.”
“People are rejected all the time based on how they look, their shoes, how they tucked in their shirts, and how ‘hot’ they are,” he told The Washington Post. “Algorithms eliminate most of that in a way that hasn’t been possible before.”
The AI, he said, doesn’t explain its decisions or give candidates their assessment scores, which he called “not relevant.” But it is “not logical,” he said, to assume that some people might be unfairly eliminated by the automated judge.
“When 1,000 people apply for one job,” he said, “999 people are going to get rejected, whether a company uses AI or not.”
The inscrutable algorithms have forced job-seekers to confront a new kind of interview anxiety. Nicolette Vartuli, a University of Connecticut senior studying math and economics with a 3.5 GPA, said she researched HireVue and did her best to dazzle the job-interview machine. She answered confidently and in the time allotted. She used positive keywords. She smiled, often and wide.
But when she didn’t get the investment-banking job, she couldn’t see how the computer had rated her or ask how she could improve, and she agonized over what she’d missed.
“I feel like that’s maybe one of the reasons I didn’t get it: I spoke a little too naturally,” Vartuli said. “Maybe I didn’t use enough big, fancy words. I used ‘conglomerate’ one time.”
HireVue says its system dissects the tiniest details of candidates’ responses – their facial expressions, their eye contact and perceived “enthusiasm” – and compiles reports companies can use in deciding who to hire or disregard.
Job candidates aren’t told their score or what little things they got wrong, and they can’t ask the machine what they could do better. Human hiring managers can use other factors, beyond the HireVue score, to decide which candidates pass the first-round test.
The system, HireVue says, employs superhuman precision and impartiality to zero in on an ideal employee, picking up on telltale clues a recruiter might miss.
Major employers with lots of high-volume, entry-level openings are increasingly turning to such automated systems to help find candidates, assess resumes and streamline hiring. The Silicon Valley start-up AllyO, for instance, advertises a “recruiting automation bot” that can text-message a candidate, “Are you willing to relocate?” And a HireVue competitor, the “digital recruiter” VCV, offers a similar system for use in phone interviews, during which a candidate’s voice and answers are analyzed by an “automated screening” machine.
But HireVue’s prospects have cemented it as the leading player in the brave new world of semi-automated corporate recruiting. It says it can save employers a fortune on in-person interviews and quickly cull applicants deemed subpar. HireVue says it also allows companies to see candidates from an expanded hiring pool: Anyone with a phone and Internet connection can apply.
Nathan Mondragon, HireVue’s chief industrial-organizational psychologist, told The Post the standard 30-minute HireVue assessment includes half a dozen questions but can yield up to 500,000 data points, all of which become ingredients in the person’s calculated score.
The employer decides the written questions, which HireVue’s system then shows the candidate while recording and analyzing their response. The AI assesses how a person’s face moves to determine, for instance, how excited someone seems about a certain work task, or how they’d behave around angry customers. Those “Facial Action Units,” Mondragon said, can make up 29 percent of a person’s score; the words they say and the “audio features” of their voice, like their tone, make up the rest.
“Humans are inconsistent by nature. They inject their subjectivity into the evaluations,” Mondragon said. “But AI can database what the human processes in an interview, without bias. . . . And humans are now believing in machine-decisions over human feedback.”
To train the system on what to look for and tailor the test to a specific job, the employer’s current workers filling the same job – “the entire spectrum, from high to low achievers” – sit through the AI assessment themselves, Larsen said.
Their responses, Larsen said, are then matched with a “benchmark of success” from those workers’ past job performance, like how well they’d met their sales quotas or how quickly they’d resolved customer calls. The best candidates, in other words, end up looking and sounding like the employees who’d done well before the prospective hires had even applied.
After a new candidate takes the HireVue test, the system generates a report card on their “competencies and behaviors,” including their “willingness to learn,” “conscientiousness & responsibility” and “personal stability,” the latter of which is defined by how well they can cope with “irritable customers or co-workers.”
Those computer-estimated personality traits are then used to group candidates into high, medium and low tiers based on their “likelihood of success.” Employers can still pursue candidates ranked in the bottom tier, but several interviewed by The Post said they mostly focused on the ones the computer system liked best.
HireVue offers only the most limited peek into its interview algorithms, both to protect trade secrets but also because the company doesn’t always know how the system decides on who gets labeled a “future top performer.”
The company has given only vague explanations when defining which words or behaviors offer the best results. For a call-center job, the company says, “supportive” words might be encouraged, while “aggressive” ones might sink one’s score.
HireVue said its board of expert advisers regularly reviews its algorithmic approach, but the company declined to make the system available for an independent audit. The company, Larsen said, is “exploring the use of an independent auditor right now, to see how that could work.”
The company said last month that the private-equity giant Carlyle Group would become its new majority investor, providing an undisclosed sum from an $18.5 billion fund.
At the hotel giant Hilton International, thousands of applicants for reservation-booking, revenue-management and call-center positions have now gone through HireVue’s AI system, and executives credit the automated interviews with shrinking their average hiring time from six weeks to five days.
Sarah Smart, the company’s vice president of global recruitment, said the system has radically redrawn Hilton’s hiring rituals, allowing the company to churn through applicants at lightning speed. Hiring managers inundated with applicants can now just look at who the system ranked highly and filter out the rest.
At the consumer-goods conglomerate Unilever, HireVue is credited with helping save 100,000 hours of interviewing time and roughly $1 million in recruiting costs a year. Leena Nair, the company’s chief human-resource officer, said the system had also helped steer managers away from hiring only “mini-mes” who look and act just like them, boosting the company’s “diversity hires,” as she called them, about 16 percent.
“The more digital we become, the more human we become,” she added.
But Lisa Feldman Barrett, a neuroscientist who studies emotion, said she’s “strongly skeptical” that the system can really comprehend what it’s looking at. She recently led a team of four senior scientists, including an expert in so-called “computer vision” systems, in assessing more than a thousand published research papers studying whether the human face shows universal expressions of emotion, and how well algorithms can understand them.
The systems, they found, have become quite perceptive at detecting facial movements – spotting the difference, say, between a smile and a frown. But they’re still worryingly imprecise in understanding what those movements actually mean, and woefully unprepared for the vast cultural and social distinctions in how people show emotion or personality.
Luke Stark, a researcher at Microsoft’s research lab in Montreal studying emotion and AI – who spoke as an individual, not as a Microsoft employee – was similarly skeptical of HireVue’s ability to predict a worker’s personality from their intonation and turn of phrase.
Systems like HireVue, he said, have become quite skilled at spitting out data points that seem convincing, even when they’re not scientifically backed. And he finds this “charisma of numbers” really troubling, because of the overconfidence employers might lend them while seeking to decide the path of applicants’ careers.
This story was originally published at washingtonpost.com. Read it here.