A look at HireVue's controversial AI-driven face and voice scanning “employability” assessment tool, used by 100+ employers including Hilton and Goldman Sachs
The coverage arc since then runs both directions: universities began training students to impress the screening algorithms, a wave of AI-generated applications and fakes sowed enough distrust that some employers reversed course entirely — with L'Oréal among those returning to in-person assessments — even as McKinsey pilots its own AI tool, Lilli, inside recruiting.
First-order effects
Candidates for roles at HireVue's clients are being ranked on facial and vocal analysis they cannot see or contest, putting brands like Hilton and Goldman Sachs behind an opaque gatekeeping method.
Harver and other pre-hire assessment vendors gain a competitive opening: any credibility hit to video-scoring pushes buyers toward text-based or structured alternatives.
Second-order effects
Career counselors at universities are already adapting curricula to coach students on how to perform for hiring algorithms, turning interview prep into algorithm optimization.
Employer-side skepticism compounds downstream: reporting on AI tools sowing distrust — including a near-miss deepfake candidate — gives HR leaders ammunition to pull back from vendor-scored screening.
Third-order effects
If the pattern holds, high-volume recruiting splits into two camps: fully automated pipelines like HireVue's facing mounting calls for explainability and audits, and premium employers retreating to human-led assessment.
The longer trajectory points toward hiring AI becoming a regulated category, where vendors must prove validity and bias controls before enterprises will attach their brands to them.
The trend: Algorithmic hiring is swinging from rapid adoption of opaque video-scoring toward accountability pressure and selective returns to human judgment, even as tools like McKinsey's Lilli push AI deeper into recruiting from the other direction.
New: More than a million job-seekers have been assessed by HireVue's job-interview AI, which analyzes candidates' faces and voices and spits out an “employability” score. But AI researchers say it's “pseudoscience” and a “license to discriminate” https://www.washingtonpost.com/ .…
If I was an ambulance chasing lawyer, I'd be putting together a million person class action suit against HireVue. If I was a financier, I'd be selling their stock short. As it is, I'm an AI researcher so I'm tweeting what a bad idea this use of computer vision is ... https://twit…
Hey Google, define intellectual dishonesty “The AI doesn't explain its decisions or give candidates their assessment scores, which he called ‘not relevant.’... it's ‘not logical,’ to assume some people might be unfairly eliminated by the automated judge.” https://www.washingtonpo…
It's inherently discriminatory - judging people by their physical features always is and always has been. When gender and racial bias of development journey are factored in; it becomes even more so https://twitter.com/...
This reminds me of many of the DNA/genetic tests that have been available since the 1990s which provide a risk/susceptibility score but are: 1) based on dubious assumptions 2) unregulated How can we apply the lessons from gene/DNA tech to AI? https://twitter.com/...
“But people scowl all the time, she said, 'when they're not angry: when they're concentrating really hard, when they're confused, when they have gas.'” https://www.washingtonpost.com/ ...
I feel like we need a new word for things that are incredibly dystopian-feeling but only because they use computers to lay bare the randomness and inanity that already control most aspects of our society https://www.washingtonpost.com/ ...
This story raises so many terrible aspects of HireVue: Treating data and algorithms as objective Referencing the number of data points as evidence of the system's sophistication Forcing people to adjust their behavior to appeal to the algorithm https://www.washingtonpost.co…
“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” https://www.washingtonpost.com/ ...
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. https://www.washingtonpost.com/ ...
Bigger trend: Software now can gather a lot more data about what makes us us. And tech firms claim they can use it to predict our personality and worthiness. But can they, really? And what happens if they get it wrong? (e.g.: “perfect babysitter” scans: https://wapo.st/2N6LV2r )
That someone could actually say this, on the record, without bursting into flames, should tell us everything we need to know about how much respect for human dignity there currently is in the tech industry: “The more digital we become, the more human we become,” she added. https:…
“It's pseudoscience. It's a license to discriminate ... And the people whose lives and opportunities are literally being shaped by these systems don't have any chance to weigh in.” - @mer__edith https://twitter.com/...
Notice and consent will NOT solve this problem. You can't meaningfully withhold consent if it means no job. Frankly, notice and consent make the problem worse because it legitimizes subjecting people to unfair and deceptive pseudoscience. https://www.shrm.org/... https://twitter.…
A lot of people told me it also just feels weird to be silently judged while talking to a screen. One said it felt “dehumanizing and fairly insulting”: It “sends the message that I am not valued enough to warrant the time of an actual human” https://www.washingtonpost.com/ ...
Can software really know your “willingness to learn” and “personal stability” from a 30-minute webcam chat? The AI researchers and neuroscientist I talked to said: Probably not. Yet it's already so widespread that colleges now teach kids how to interview: https://www.washingtonpo…
There are ~200 types of bias running in our brains. Transferring it to software is not an answer; it is codification of the most discriminatory cues determined by people in positions of power. Tech keeps applying itself to the least imaginative, least transformative purposes http…
These “AI assessments” - in which you “interview” with a computer that quietly judges how you look and talk, rates your personality, and can decide the fate of your career - are driving people a little insane: https://www.washingtonpost.com/ ... https://twitter.com/...