Posted on by

Google Fired Another AI Ethics Researcher for a Paper Critical of Bias, Energy Use, in AI

Reading Time: 9 minutes.
A photo of Dr. Timnit Gebru

Dr. Timnit Gebru. Photo: Kimberly White/Getty

Last week, Google fired an AI researcher, Dr. Timnit Gebru. She wasn’t the first employee Google went after for calling out practices in artificial intelligence that could be harmful. She wasn’t the only person Google fired for speaking up about problems at the company either, especially their inability (or refusal) to hire and retain women, BIPOC, or other marginalized groups. Dr. Gebru was critical of Google’s treatment of Black and female employees. Her departure silences a necessary and outspoken voice at Google. Despite having tens of thousands of employees all over the globe, Black employees make up only 3.7% of Google’s employees, and 2.6% of leadership. Leadership is 65.9% white at Google. Furthermore, Google’s female population is only 32%, with only 26.7% in leadership roles. This is your reminder that women make up half of the population.

Only 1.6% of Google’s employees, worldwide, are Black women.

1.6%.

Black and female employees at Google have been blowing the whistle for years over the racist and sexist bias they see at Google. AI researchers, especially, have called out issues in Google’s application of AI and the effects it will have on a more diverse population than Google’s accounting for. Rather than learn from this criticism, Google’s simply firing their most outspoken employees. More often than not, those are marginalized groups trying to ensure that as Google builds the future, it doesn’t only consider the needs of straight, white, men.

AI research demands diverse voices to avoid prejudice. But Google continually silences the voices telling them they need to change.

Dr. Timnit Gebru: Longtime Critic

I’m going to reveal something. When you have an employee who is critical of the company, that’s not someone who’s disloyal. In software, especially, there’s a longstanding culture of jumping jobs frequently. Many tech employees stay at jobs less than a year, either by choice or because the startup they worked at couldn’t stay up. When an employee makes their opinion known instead of jumping ship, they’re trying to help. They’re trying to save the work they love. With that in mind, Dr. Gebru certainly tried with Google.

Dr. Gebru published a paper some time ago calling out racial bias and gender bias in facial recognition. It’s often cited as one of the core pieces of research that proved the problem to the AI community. Thanks to that research, we now know just how important it is to not only have diverse data sets, but a team of diverse engineers who will keep issues like these in mind. Because it was mostly white men working on AI in the U.S., much of our facial recognition carries racial and gender-based bias. Dr. Gebru built a diverse team at Google to ensure they could tackle the issues in AI and help shape the future of AI towards one that works for and with everyone.

After George Floyd was murdered by police officers, Dr. Gebru’s boss, Jeff Dean, a white man and senior vice president of AI at Google, made some dismissive comments about policing. Even if you do hold the belief that most or many police officers are good people, voicing that as a reaction to “a few bad apples” killing a man send a different message. It’s then about protecting wrongdoing, and, in this case, racial bias. No one was shouting “all lives matter” until someone tried to tell them that Black lives matter, and this is in the same vein. Dr. Gebru was willing to call out her own boss on his racially insensitive remarks, but while it didn’t lead to her firing then, it contributed to hostility from upper management towards Dr. Gebru.

Dr. Gebru’s Paper, Google’s Blockade

A chart showing lbs of CO2 released. About 36,156/year for the average American, 626,155lbs for a single AI process, a 213M parameter transformer w/ neural architecture search. Sources: Emma Strubell et al

Dr. Timnit Gebru’s latest paper was about problems with large data sets, including bias that originates from collecting data from online discussions. Many versions of AI have used conversations found on the internet to form topic recognition, sentence structure, and natural language processing. It’s what allows tools like Google Translate to produce more realistic translations. It does have its negative sides, like when Microsoft’s Tay chatbot began spouting racist and sexist language due to an influx of bad input from users looking to corrupt the chatbot.

Dr. Gebru also pointed out that this will have a bias towards people who have and use the internet. If you’re sourcing your input solely through online discussions and content, you’re cutting out people in poorer areas of the world, those without easy or cheap internet access, and older generations who simply don’t use the internet much, if at all, for content creation.

The paper also referenced the findings of a previous paper by Emma Strubell, which examined the environmental impact of collecting such large amounts of data. While the language produced may sound natural, it comes from unfathomable amounts of data. The processing can create more CO2 emissions than multiple cars over their entire lifetime, just to run through one dataset once. AI researchers often make many tweaks and run their datasets through processing multiple times, and a company may run the same task many times as well. The energy use, especially in countries that do not sustainably produce electricity, releases carbon dioxide and methane into the atmosphere, two greenhouse gases that have lead to global warming and climate change.

Google’s Not Happy

Google has been combing the internet for decades to provide search results. Scooping up that data for language processing was simple. They’re collecting tons of data on their users, what’s posted on websites, and what people are sharing on social networks. Google’s doing the harmful practices that Dr. Gebru shared in her paper. The paper had multiple co-authors, including other authors from within Google, as well as external sources. While Google initially approved the paper, they then came at both the content and how it was presented. Eventually, they’d ask Dr. Gebru to pull her paper or remove all of the Google employees’ names from the paper. The latter would be dishonest and potentially illegal, to pull credit from researchers. The former was censorship.

Neither sat well with Dr. Gebru, and she finally gave on up Google being anything but evil. Pushing for ethics at a company that does not respect them was futile. She stated that she would discuss resignation timelines upon her return from a vacation.

She wouldn’t get that opportunity.

Email That Google Supposedly Fired Her For

Dr. Gebru sent out an email to Google’s “Brain Women and Allies Group.” The group had been discussing ways to improve Google for women and other marginalized groups at the company. However, despite their documentation, suggestions, meetings, all-hands presentations, and emails, nothing was changing within Google. Furthermore, as Dr. Gebru learned the hard way, speaking up about racial bias or gender-based bias can land you in hot water, with extra scrutiny over your work and censorship, preventing her from doing research or even her job at Google. She shared that frustration with the group.

“What I want to say is stop writing your documents because it doesn’t make a difference. … there is zero accountability. There is no incentive to hire 39% women: your life gets worse when you start advocating for underrepresented people.”

– From Dr. Timnit Gebru’s Email to Brain Women and Allies Group

She discussed Google’s lack of effort on hiring. Google has made little improvement on diversity over the past three decades of its existence. Google simply doesn’t take diversity seriously, and has cut back on projects to improve its hiring and retention of diverse employees. 2019 was defined by Google taking care of sexual abusers at the company, one of the largest employee walkouts in history, and retaliation against employees who spoke up about issues of racism and sexism within Google. The U.S. government is actually investigating Google for that bias and retaliation currently.

After Dr. Gebru sent out her email to the group, Google fired her, claiming she resigned.

Google Reacts

Dr. Gebru summed up Google’s reaction to the news of her firing best.

“You write emails, they get ignored. You write documents, and they get ignored. Then you discuss how it’s being done and then they talk about you as if you’re like some angry Black woman who needs to be contained.”

– Dr. Timnit Gebru in her interview with VentureBeat

Keeping that in mind, here is Google’s reaction. Considering Dr. Gebru is a leading voice in AI ethics, that she has documented proof that much of this is not true, and their tone over her criticism, and it’s easy to see how Google’s leaning on prejudice to salvage their brand. It won’t work. You can read Jeff Dean’s full email to Google employees here, but here’s a summary.

“I’m sure many of you have seen that Timnit Gebru [sic] is no longer working at Google. … Because there’s been a lot of speculation and misunderstanding on social media, I wanted to share more context about how this came to pass… we require two weeks for this sort of review — and then instead of awaiting reviewer feedback, it was approved for submission and submitted.”

– Jeff Dean in his email to Google employees

Dr. Timnit Gebru says her paper was shared for publication before Google’s arbitrarily enforced deadline. Even Dean’s email seems to point out that Google approved the paper before trying to redact it.

“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. It ignored too much relevant research …”

– Jeff Dean in his email to Google employees

https://twitter.com/le_roux_nicolas/status/1334601960972906496?s=20

For those uninvolved in the academic or research world, this is extremely uncommon. Companies don’t get to step in and force “corrections” on papers. They don’t edit for content like this. Furthermore, according to one of the paper’s co-authors, Emily M. Bender, the paper had six collaborators and a large citation list of 128 references. She stated the paper was, “the sort of work that no individual or even pair of authors can pull off.”

https://twitter.com/william_fitz/status/1335004771573354496?s=20

“We acknowledge that the authors were extremely disappointed with the decision that Megan and I ultimately made, especially as they’d already submitted the paper. … Timnit wrote that if we didn’t meet these demands, she would leave Google and work on an end date. We accept and respect her decision to resign from Google.”

– Jeff Dean in his email to Google employees

Note that here, Dean is again dishonest. They never discussed an end date with Dr. Gebru, and she was fired during a vacation. This was in no way a resignation due to the timing and lack of a discussion with Dr. Gebru and Google.

“I also feel badly that hundreds of you received an email just this week from Timnit telling you to stop work on critical DEI programs. Please don’t. I understand the frustration about the pace of progress, but we have important work ahead and we need to keep at it.”

– Jeff Dean in his email to Google employees

The irony here is that Dean was a large part in silencing Dr. Gebru’s work, and defended police immediately after officers murdered George Floyd. His insensitivity about race and asking people to accept Google’s lack of “pace of progress” is gaslighting. Google has no “pace of progress” on racial issues, and people just like Jeff Dean have been holding the company back by silencing its loudest voices.

Dean’s response was tone policing and gaslighting. Coming from a white man who tried to tone down the criticisms of employees, it’s salt in the wound.

Sundar Pichai, the CEO of Alphabet, Google’s parent company, has since apologized for her “exit.” He did not admit wrongdoing or that Dr. Gebru was fired, not “resigned.” The “apology” was lacking any actual apology, stating that they’d investigate what went wrong. Of course, this comes with no quantitative measures for improvement.

Tech Industry Speaks Out

The group of Google employees who protested Google’s rewarding of sexual abusers and silencing of victims, as well as their retaliation, Google Walkout For Real Change, shared an article on Medium. It’s an open letter in support of Dr. Temnit Gebru, and condemnation of Google’s actions. As of now, it has the signatures of over 2,300 Google employees and 3,700 industry experts. If you were to read it all, Medium estimates that it would take you 134 minutes.

The group has three demands of Google’s leadership. The full demands are below, but to summarize, they want Google’s leadership to explain their firing of Dr. Gebru, as seemingly she committed no fireable offense. They want more transparency, and they want Google to commit to AI ethics.

  1. We demand that Jeff Dean (Google Senior Fellow and Senior Vice-President of Research), Megan Kacholia (Vice-President of Engineering for the Google Brain organization), and those who were involved with the decision to censor Dr. Gebru’s paper meet with the Ethical AI team to explain the process by which the paper was unilaterally rejected by leadership.

A suggestion from Dr. Gebru pointed out that we need legislation to protect ethics in AI researchers, much in the way we protect whistleblowers. They often have to do their research as part of a company, and often the companies want to silence anyone who may speak out. As such, it can defeat the purpose of doing ethics in AI research, and could send us down a very dark path with prejudice AI.

Google: Don’t Be Evil

Protestor holds a sign reading "Not Ok Google #DontBeEvil"

From the Google employee walkout over sexual harassment in 2018. Photo: Stephen Lam/Reuters

Google abandoned their “Don’t Be Evil” mantra a long time ago, when it was clear their way forward was significantly evil. Internally, Google has treated their employees with the same disregard that they’ve shown their customer’s privacy. For women, Black, and Latino employees, it’s been even worse.

Like a parasite or virus, Google has woven its way into every aspect of our lives. From their ads to their data, search engines and shopping. Living a Google-free life is almost impossible. Still, where there are alternatives to Google, consider using them. Ecosia or DuckDuckGo for web searches. Apple Maps or OpenStreetMaps instead of Google Maps. Apple Pay instead of Google Pay. iOS instead of Android. Vimeo instead of YouTube. Where you can find alternatives, take them.

Google has a longstanding history of silencing marginalized voices. They’ve tried to keep people quiet, push victims of sexual harassment into forced arbitration, and they’ve retaliated against whistleblowers and critics. It’s bad enough that the National Labor Relations Bord is investigating Google for unlawful surveillance, interrogation, or suspension of workers. Recently, Google has done more work to silence marginalized voices than innovate. They’re charging forward, knowing they’re doing wrong, but not considering it because, so far, it’s been profitable.

 

“I’ve seen how my expertise has been completely dismissed. But now there’s an additional layer saying any privileged person can decide that they don’t want your paper out with zero conversation.”

– From Dr. Timnit Gebru’s Email to Brain Women and Allies Group

Sources: