Google workers say AI researchers haven’t stepped down, regardless of Exec's claims


Jeff Dean, Google Senior Fellow, speaks at an event in China in 2017.

Source: Chris Wong | Google

A group of current and former Google employees who started a petition in support of the late researcher Timnit Gebru deny an executive's report of their departure.

In a blog post titled "Setting the Record Out," Gebru colleagues and petitioners claim she has not resigned and point to inconsistencies in Google AI chief Jeff Dean's statements, including his allegation that Google's approval process research takes two weeks. In fact, they claim, papers are often approved much faster, and Gebru & # 39; s paper has been approved through "standard procedures" known as PubApprove.

"This is a standard that has been applied unevenly and discriminatively," says the Google Walkout for Real Change group.

Google didn't immediately respond to the request for comment on the latest blog post. The company has so far refused to comment on Gebru & # 39; s departure. Gebru did not respond to CNBC's requests for comment.

Gebru, a noted artificial intelligence researcher, was the technical co-lead of Google's Ethical AI team and a vocal critic of tech companies' treatment of black workers. She tweeted last week that Google abruptly fired her after a disagreement over a research paper examining possible biases in large-language AI models.

She reportedly asked for more details on who requested the withdrawal and emailed an internal group describing the incident and including general complaints about Google's treatment of minority workers. After that, she says, the company suddenly cut off its corporate accounts and said it would "accept (her) resignation immediately".

When staff and industry leaders criticized the move, Dean told staff that he pulled the research paper because Gebru failed to follow the protocol for the paper and that it "ignored too much relevant research." He added that Gebru had resigned and that he simply accepted her resignation.

The staff immediately went to Twitter and said that Dean's explanations were inconsistent with their experience.

"Dr. Gebru has not resigned, even though Jeff Dean (senior vice president and head of Google Research) has stated publicly. Dr. Gebru has said this clearly and others have documented it meticulously," the Walkout group stated on the blog on Monday's post . "Dr. Gebru detailed conditions that she hoped could be met."

Gebrus terms, according to the group, were: "Be transparent about who was involved in calling for the paper to be withdrawn, have a series of meetings with the Ethical AI team, and understand the parameters of acceptable research on Google."

She then asked for a longer talk after her vacation, they said.

The group also backed down on Jeff Dean's testimony that Google's research approval process takes two weeks.

"It is not absolutely necessary that papers actually go through this review with a period of two weeks," the group explained in the blog post. "Numerous articles are eligible for publication without meeting this" requirement ", it says.

It also backed off Dean's claim that his leadership team gave Gebru feedback on their paper.

"There was no written feedback from the leadership, the authors were not given an opportunity to communicate the verbalized concerns with all parties involved, and the authors were not given an opportunity to revise the paper in the light of the feedback despite being willing to watch the camera. " The deadline is more than a month away, "the group said.

Employees at Google and other organizations have asked Google for answers on how they have dealt with Gebru's departure in an online petition with more than 3,000 signatures from employees and industry colleagues. Gebru & # 39; s manager Samy Bengio, who leads a team within Google Brain, said in a Facebook post over the weekend that he was "stunned" by Gebru & # 39; s resignation and added: "I stand by you , Timnit. "

CLOCK: The renowned AI researcher says Google abruptly fired her and sparked industry-wide criticism


Katherine Clark