Wed. Sep 28th, 2022
Former Google AI research scientist Timnit Gebru speaks onstage during Day 3 of TechCrunch Disrupt SF 2018 at Moscone Center on September 7, 2018 in San Francisco, California.
enlarge Former Google AI research scientist Timnit Gebru speaks onstage during Day 3 of TechCrunch Disrupt SF 2018 at Moscone Center on September 7, 2018 in San Francisco, California.

Kimberly White | Getty Images

Google struggled Thursday to mitigate the fallout from the departure of a top artificial intelligence researcher after the internet group blocked the publication of a paper on a key AI ethics issue.

Timnit Gebru, co-head of AI ethics at Google, said on Twitter that she was fired after the paper was rejected.

Jeff Dean, Google’s head of AI, defended the decision in an internal email to staff Thursday, saying the paper “didn’t meet our publication bar.” He described the departure of Dr. Gebru also considered a resignation in response to Google’s refusal to give in to unspecified terms she had made to stay with the company.

The dispute threatened to shine a bright light on Google’s handling of internal AI research that could harm its business, as well as the company’s longstanding difficulties in trying to diversify its workforce.

Before leaving, Gebru complained in an email to colleagues that there was “zero liability” within Google over the company’s claims it wants to increase the proportion of women in its ranks. The email, first published on Platformer, also described the decision to block her paper as part of a process to “silence marginalized voices.”

One person who worked closely with Gebru said there had been tensions in the past with Google’s management because of its activism to push for greater diversity. But the immediate trigger for her departure was the company’s decision not to allow the publication of a research paper she co-authored, this person added.

The article looked at the potential bias in large-scale language models, one of the hottest new areas of natural language research. Systems such as OpenAI’s GPT-3 and Google’s own system, Bert, attempt to predict the next word in any sentence or sentence — a method that has been used to produce surprisingly effective automated writing, and which Google uses to better understand complex queries.

The language models have been trained on large amounts of text, mostly sourced from the Internet, leading to warnings that they could break out racial and other biases that appear in the underlying training materials.

“From the outside, it looks like someone at Google has decided that this is detrimental to their interests,” said Emily Bender, a professor of computational linguistics at the University of Washington who co-authored the paper.

“Academic freedom is very important – there are risks when [research] takes place in places that [don’t] have that academic freedom,” giving companies or governments the power to “stop” research they don’t approve, she added.

Bender said the authors hoped to update the paper in time with newer research so that it can be accepted at the conference to which it was already submitted. But she added that it was common for such work to be replaced by newer research, given how fast work is progressing in these areas. “In the research literature, no paper is perfect.”

Julien Cornebise, a former AI researcher at DeepMind, the London-based AI group owned by Google’s parent company Alphabet, said the dispute “shows the risks of concentrating AI and machine learning research in the few hands of powerful industrial actors, as it allows censorship of the field by deciding what is or is not published.”

He added that Gebru was “extremely talented — we need researchers of her caliber, not filters, for these issues.” Gebru did not immediately respond to requests for comment.

Dean said the paper, written with three other Google researchers and outside collaborators, “didn’t take into account recent research to reduce the risk of bias.” He added that the article “talked about the environmental impact of large models, but ignored further research that showed much greater efficiencies.”

© 2020 The Financial Times Ltd. All rights reserved. May not be redistributed, copied or modified in any way.

By akfire1

Leave a Reply

Your email address will not be published.