Timnit Gebru leaves Google over a paper
On 2 December 2020 Timnit Gebru's employment at Google ended, where she was a staff research scientist and co-lead of the ethical AI team. The occasion was management's demand that she withdraw an unpublished paper on the risks of large language models. Google wrote that it was accepting her resignation; Gebru said she had been fired. A petition supporting her was signed by 2,695 Google employees and 4,302 outside signatories.
Why it matters
This is the first record in the atlas in which research into risk inside a company that builds these systems ended in the end of an employment. What matters is not whose account is right but that both sides fell out publicly over whether criticism of a company's own direction can be published from within it.
What is established. Gebru set management conditions on which she would stay and wrote that otherwise she would work on a last date. Google's reply, as the BBC quotes it: we respect your decision to leave Google and we are accepting your resignation, with the addition that the end of her employment should happen faster than her email reflects because certain aspects of the email she had sent the previous night to non-management employees in the Brain group reflect behaviour inconsistent with the expectations of a Google manager. Gebru denied that she had resigned. The petition Standing with Dr. Timnit Gebru, published on 4 December 2020, demanded that Jeff Dean, senior vice-president of research, and Megan Kacholia, vice-president of engineering for Google Brain, explain to the team how the paper had been rejected, and demanded public transparency about the order to withdraw the research on large-scale language models. The record does not settle whether this was a dismissal or a resignation: the parties say different things, and none of the available sources is neutral on the point. The signature counts are what the petition page carries now; it gathered names over several days, and on 3 December the BBC reported nearly two thousand. The paper itself, and whom Gebru was listed as working for on it three months later, are separate records.