Google sells AI tools to companies as a way to speed up hiring by helping recruiters sort through thousands of job applications and identify the strongest candidates.
But some researchers at Google DeepMind apparently do not want to take that same approach when hiring for their own team.
Google DeepMind’s AGI Safety and Alignment Team, which works on reducing the risks associated with advanced AI, has been asking candidates applying for its open roles to complete an additional form alongside their regular application.
The reason? The team wants to make sure a human actually sees their application.
Why is Google’s AI bot struggling to hire good talent?
According to a document reviewed by Bloomberg, the team warned applicants that Google’s internal application system has a “non-trivial probability” of incorrectly screening out a CV or taking too long to send it to the relevant hiring team.
Candidates were therefore encouraged to fill out the separate form so their application would be reviewed directly by someone on the team.
The document even carried a warning asking people not to share it widely. The message is particularly striking given who is behind it. Google DeepMind’s safety and alignment researchers spend their time thinking about how advanced AI systems can behave reliably and how potential risks can be identified and controlled.
And now, at least for some of their own hiring, they appear to be taking precautions against an AI system making the wrong call.
Google refutes rumours
Google, however, disputes the idea that its hiring systems incorrectly filter out candidates.
A Google DeepMind spokesperson said the company is focused on hiring the “most qualified talent at Google DeepMind” and denied that its systems incorrectly screen applicants.
Still, the episode points to a much bigger problem with using AI in recruitment.
Companies are increasingly turning to automated systems because hiring teams can receive thousands of applications for a single role. An AI system can scan resumes far faster than a person, identify certain qualifications and narrow down the pool before recruiters begin interviewing candidates.
The problem is that an algorithm does not necessarily understand a candidate the way a human does.
A resume might use unconventional wording, describe experience in a different industry or highlight skills that do not neatly match the keywords an automated system is looking for. A strong candidate could therefore be overlooked without ever getting the chance to speak to a recruiter.
That is what makes the DeepMind situation particularly ironic.
The team working on AI safety appears to be saying, in effect, that applicants should not leave their fate entirely to an automated system.
For job seekers, the episode is a useful reminder that getting past the first stage of the hiring process can sometimes be as important as having the right qualifications. And for companies, it raises a difficult question: if AI is supposed to make hiring better and more efficient, why are some of the people building advanced AI systems still looking for a human safety net?
Google may disagree with the idea that its hiring AI gets things wrong. But the fact that DeepMind researchers reportedly created an extra route for applications to reach a person shows that, when it comes to hiring, humans may not be ready to hand over the final call to AI just yet.
