Timnit Gebru leaves Google after a dispute over her paper

Timnit Gebru, co-lead of Google’s Ethical AI team, announced on Twitter on the evening of 2 December 2020 that the company had ended her employment. The two accounts of what happened do not agree. Gebru said she was fired after sending an internal email, to a Google Brain diversity mailing list, criticising the company’s record on hiring and retaining women and telling colleagues that diversity paperwork “doesn’t make a difference.” Google’s head of AI, Jeff Dean, told staff in an internal memo that Gebru had instead resigned, after setting conditions for staying that the company could not meet.

Underlying the email was a dispute over a research paper. Gebru and three co-authors had written “On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?”, surveying the environmental cost, embedded bias and misuse potential of large language models — the kind Google itself was building. Google management asked her to retract the paper or remove her name and those of other Google-affiliated co-authors before an upcoming conference deadline; Dean later said internal reviewers found it “didn’t meet our bar for publication” because it omitted relevant mitigating research, a characterisation Gebru disputed given the paper carried 128 citations. Gebru replied that she would resign unless the company explained the retraction request and committed to changes in its review process; her corporate account was cut off before she returned from a scheduled break, and she said she learned of her termination from her own direct reports rather than through management — which she and supporters took as confirmation of a firing rather than an accepted resignation.

Within two weeks, over 2,600 Google employees and 4,300 people in academia, industry and civil society had signed a petition denouncing the dismissal as “unprecedented research censorship.” Google said it would review its research-approval process, and Margaret Mitchell, who co-led the Ethical AI team with Gebru, was herself dismissed the following February after searching her email for evidence supporting Gebru’s account. The episode became a reference point in wider arguments over how much independence corporate AI ethics research can have from the products it studies.