• On Wednesday, one of Google's top AI ethics leaders announced she had been fired.
  • Observers were confused by the decision, but inside Google, tensions between Timnit Gebru and senior management had been building for days.
  • Gebru had co-authored a research paper that observed the biases of machine learning, but said she was ultimately fired over an email that criticized the company's treatment of minorities.
  • Are you a current or former Googler with more to share? You can contact this reporter securely using the encrypted messaging app Signal (+1-628-228-1836) or encrypted email (hslangley@protonmail.com). Reach out using a nonwork device.
  • Visit Business Insider's homepage for more stories.

Timnit Gebru, a co-lead on Google's ethical-artificial-intelligence research team, tweeted late Wednesday that she had been ousted from the company. Outside observers were stunned and confused: Why would Google terminate one of its top AI ethics figures who was also a highly respected name in the field?

But inside Google, tensions had been mounting for several days, starting with a research paper co-authored by Gebru, which was submitted to an academic conference and was critical of biases being built into artificial intelligence. According to Gebru and other employees familiar with the matter, management asked Gebru to either retract the paper or remove the names of all co-authors who were Google employees. Gebru later aired her frustrations to an employee listserv for women at Google, criticizing the company's treatment of minority employees.

The next day, she was fired.

It's the latest example of tensions between the company's corporate interests and workers' ethical concerns. The firing also drew the attention of others in the AI research field, while it left many Google employees confused over the seemingly-aggressive response.

On Thursday, Google's AI chief Jeff Dean told employees in an email obtained by Business Insider that Gebru's firing was a "difficult moment."

According to an employee familiar with the situation, Google was unhappy with the research paper, which analyzed the ethical dangers of language models. Gebru subsequently pushed back on Google's request to redact author names or retract the paper entirely, asking management to provide more details on their reasoning.

As Gebru recounted via Twitter on Wednesday: "I said here are the conditions. If you can meet them great I'll take my name off this paper, if not then I can work on a last date. Then she sent an email to my direct reports saying she has accepted my resignation. So that is google for you folks. You saw it happen right here."

The day before she was fired, Gebru sent an email message to an internal Google Brain Women and Allies group on, venting frustrations over her experience and calling on members to find new ways to find "leadership accountability."

According to another employee who is part of the group and asked to remain anonymous, the email group is generally used for mentorship and "enabling members to feel empowered to lean in to the workplace." Since the group had become moderated, they said, conversations had become more restricted. "Discussion and threading is severely limited," they said.

As such, Gebru's message made a big splash.

"What I want to say is stop writing your documents because it doesn't make a difference," wrote Gebru in a message, a copy of which was published by Casey Newton's Platformer. "There is no way more documents or more conversations will achieve anything."

The day after she sent that email, Gebru said in a tweet that she received a message informing her that she was being dismissed. "Thanks for making your conditions clear. We cannot agree to #1 and #2 as you are requesting. We respect your decision to leave Google as a result, and we are accepting your resignation," it read, according to Gebru.

The same message also referenced the email sent to the research group the previous day, described as "inconsistent with the expectations of a Google manager," and used as grounds to expedite her termination.

"We believe the end of your employment should happen faster than your email reflects because certain aspects of the email you sent last night to non-management employees in the brain group reflect behavior that is inconsistent with the expectations of a Google manager," read that email, according to Gebru.

Stochastic parrots

The debacle has raised questions over the nature of the research paper, which Google says was submitted before the company gave it approval.

The paper, titled 'On the dangers of stochastic parrots: Can language models be too big?', examined how major language models trained on large amounts of data bring biases and carry the risk of ethical harms.

In the paper, a copy of which was reviewed by Business Insider, the authors identify a variety of costs and risks associated with what they describe as "the rush for ever larger language models,"

"The paper doesn't say anything surprising to anyone who works with language models," said one employee familiar with the topic who reviewed it. "It basically makes what Google's doing look bad," said another, noting that the conclusions of the paper were largely critical of work being done in AI.

But Google argues that the paper was not approved because it didn't follow the correct procedure and "ignored too much relevant research."

"Unfortunately, this particular paper was only shared with a day's notice before its deadline — we require two weeks for this sort of review — and then instead of awaiting reviewer feedback, it was approved for submission and submitted," said Google's AI lead Jeff Dean in an email to employees sent Thursday, and obtained by Business Insider. 

"A cross functional team then reviewed the paper as part of our regular process and the authors were informed that it didn't meet our bar for publication and were given feedback about why," he added.

The events have caused concern and frustration among many employees. "People are seriously pissed," said one, noting that Dean's email did not seem to justify Gebru's firing, and failed to acknowledge the email that Gebru claims was used as grounds for her dismissal. "People are trying to find out why the reaction was so extreme."

Vijay Chidambaram, an assistant professor at the University of Texas at Austin, tweeted that the stated reason for blocking the paper in Dean's email was "pretty much BS." 

"This is the job of conf reviewers, and not the job of Google," he said.

The debacle has only raised further concerns over Google's treatment of AI ethics, and how its desire to get ahead is clashing with employee values. Last year, Meredith Whittaker, an employee at the time, alleged that Google had pressured her to abandon her work with the AI Now Institute, a research center, cofounded by Whittaker, which is focused on the social implications of artificial intelligence.

"Waking up, still shook by the way Google's treating Timnit, and thinking about how urgently we need to confront the racist history and present of the AI field," Whittaker tweeted on Thursday.

Get the latest Google stock price here.