Google’s Ai Safety Team Tells Job Seekers: Don’t Trust Its Hiring Filters

Alphabet Inc.’s Google pitches its artificial intelligence devices to firm clients arsenic a measurement to much quickly sift done a upland of occupation applications to find nan astir promising candidates. Some of its ain AI researchers don’t want to trust connected these devices erstwhile recruiting.

Google DeepMind’s AGI Safety and Alignment Team, which seeks to mitigate nan risks of precocious AI, is encouraging occupation candidates for its unfastened roles to capable retired a typical shape alongside their exertion to destruct nan consequence it whitethorn beryllium automatically screened retired by nan company’s soul AI systems, according to a archive viewed by Bloomberg, which contained nan disclaimer “PLEASE DO NOT SHARE THIS DOC WIDELY.”

“We person an applications strategy pinch a non-trivial probability your CV will beryllium screened retired incorrectly aliases return excessively agelong to scope us,” nan archive states. “Filling retired this shape makes judge that a existent quality connected nan squad will get to spot your application.”

A Google DeepMind spokesperson said nan institution intends to “recruit and prosecute nan astir qualified talent astatine Google DeepMind” and denied that nan company’s systems select retired applicants incorrectly. “This squad group up a typical shape to spell past nan recruiter review, and get their resumes nonstop to nan group connected nan team. But location are nary shortcuts to getting hired,” nan spokesperson added.

AI has quickly go a portion of nan hiring process, prompting immoderate HR teams to interest they whitethorn soon beryllium retired of a job. But really that AI is employed tin beryllium murky. Some companies tally AI models to rank applicants, while others select resumes for cardinal terms.

Google’s Workspace team, which sells celebrated products specified arsenic Google Drive to businesses, markets nan company’s caller AI features arsenic a measurement to “save HR clip by quickly creating drafts for occupation postings, evaluating resumes, and forecasting hiring needs.”

AI hiring systems person besides travel nether occurrence for perchance discriminating against candidates. A Bloomberg investigation recovered that OpenAI’s ChatGPT showed signs of imaginable bias based connected applicants’ names. And Workday Inc., which sells workplace guidance software, is facing a suit that alleges its AI hiring systems surface applicants based connected race, property and disablement successful usurpation of nan law. Workday has denied nan allegations and said humans make hiring decisions. The institution did not respond to an further petition for comment.

Meanwhile, immoderate occupation candidates are utilizing AI to their advantage by gaming nan filters, aliases simply churning retired applications successful grounds time. The Google DeepMind squad was connected defender astir this anticipation arsenic well, counseling candidates that their applications would beryllium stronger without nan thief of AI.

The shape designed to circumvent automated screening includes an advisory: “A existent quality will publication these. These humans get really tired of reference LLM answers, because they each sound very samey.”

Love writes for Bloomberg.