As I understand this saga, Gebru wrote a paper portraying language models and Google negatively. They asked her to change it, she refused to (assuming this), Gebru wrote an email to a wide audience with a series of unprofessional complaints and made an ultimatum to her higher-ups, the ultimatum was declined and rather than work on an end date Gebru was terminated immediately. Unless I'm missing something it hardly seems outrageous.
That's... not especially consistent with the facts that have come to light.
* Timnit pointed out that language models have unavoidable flaws, including reflecting racial biases in their training data. This is sort of an obvious claim if you ask me but google stands to make a lot of money if it isn't true.
* They demanded she take her name off the paper entirely. No changes requested. Research careers advance largely by publishing papers, and this is several months of her work they want disowned.
* She agreed to take her name off provided she could have an explanation of what she was and was not permitted to say. IMO being a researcher and having publications yanked at the last moment based on unspecified criteria is not a tenable position.
On one side, we have an unemployed researcher whose future credibility & employ-ability depend on not being caught lying (the easiest way to do this is not to lie, but who knows).
On the other side, we have a billion-dollar company who have not been shy about spending significant sums of money to push their PR lines.
Obviously either could be telling the truth but one seems markedly more likely than the other to me.
Perhaps I'm wrong about Google asking her to change it versus just telling her it's rejected. That's just how I read Jeff Dean's document posted here the other day.
Assuming I am wrong about that, I'm not sure it changes much about my comment. Presumably Google has the authority to ask for a publication to be pulled, and responding with an unprofessional email to a wide list is not ideal. On the matter of not agreeing on an end date, Gebru raised resignation and Google picked the date.
I just put it in context by thinking about myself. If I acted the way Gebru did when I was upset by a professional setback, would I expect to be fired? Yes, unequivocally.
You mention her having a research career - but she's not in research. She works at Google. Her job, however they try to fancy it up, is to sell ads. Explaining to people that language models are racist and energy inefficient is not a good way to sell more ads, which is why Google wanted her paper pulled (I assume). If Gebru wants to be an academic she should work at a university. If she wants to work at Google she should realize her job is to sell ads. If her paper made Google look good (here's what Google is doing to combat the racism latent in our competitor's language models) then I have zero doubt Google would've been pleased by her work.
I don't disagree with most of you say, except for this:
> You mention her having a research career - but she's not in research. She works at Google. Her job, however they try to fancy it up, is to sell ads. Explaining to people that language models are racist and energy inefficient is not a good way to sell more ads,
She was hired to do research, and her job was explicitly related to ethics, so this is very much in line with her work. My company has a research arm, and people who are hired into it are very much expected to publish papers and will be judged based on their ability to do so. Having disagreements about what the company does in public, and sending sensational emails to coworkers would not be tolerated here either, though.
I would expect that if a researcher at your company came up with a paper that concluded "Our work at this company is sexist, racist, and bad for the environment and we should abandon it" that your company would prefer it not be published, especially not published with the name of the company or the company's researchers attached. I think that's basically Google's stand.
> I would expect that if a researcher at your company came up with a paper that concluded "Our work at this company is sexist, racist, and bad for the environment and we should abandon it" that your company would prefer it not be published
I'm not really sure what your point is. At a very basic level: Of course they would not prefer it published. Just like they would prefer not to have transparency, pay competitive market salaries, give free lunches, have an HR department, etc. We all prefer things if we can get them.
I would, however, expect that if a company hired a researcher to study whether their work is ethical, then it's silly to get in that researcher's way. If my company hired someone to ensure their supply chain for manufacturing does not involve sweatshops, then it certainly wouldn't look good if my company prevents him from saying so. If we aren't prepared for uncomfortable truths, we should not hire people to look into them.
I honestly think there are other reasonable reasons to get rid of this researcher than what you are positing.
I'm not saying that they fired Gebru because of this reason. I'm saying that's why they wanted the paper pulled. Dean says as much in his statement - that the paper lacked the context that explained new efficiencies and what Google was doing to solve these problems.
I think Gebru's reaction, the ultimatum and the email, were what led to her firing. Curious what you think the reasons to fire her were.
She might have been getting managed out over her twitter spat with Yann LeCun, or maybe they just wanted her name off research or the research toned-down to not attract attention.
What she definitely did wrong was misunderstand that a researcher on "Ethical AI" is a PR role, and job is to make Google look good. That doesn't mean she can't ask hard questions, she just has to do it the right way.
It seemed to me that Google handled their reaction to her paper in a very unscientific way. They refused to give her any real feedback on why they wanted the paper pulled and her ultimatum was about how to handle the situation. Maybe it doesn't rise to outrageousness but Google handled it in a very concerning way for employees who publish papers.
What I think is Google's worst sin here is publishing the reason for her termination. I've not seen employers do that before. In my opinion it should be a matter between Gebru and Google.
I get that they wanted to publish their side because they want to avoid being seen as racist, but it strikes me as unprofessional.
That depends on the size of your network and who backs you.
Traditional Hierarchies are floundering about because more and more employees are hooked into networks of influence, that in many cases are larger than the networks of their bosses.
Presumably it would be framed as "I will get my affairs in order and transition my projects to colleagues. This should take about X weeks" (where X gets you past the payout date).
How does the bonus structure work at Google? In my company, you get your prorated share of the annual bonus if you leave during the year, so this is not a factor here.
Gebru [did something Google didn't like] and was terminated immediately.
You got that part right.
Unless I'm missing something it hardly seems outrageous.
What you're missing is the central point of contention here: that Google is claiming she "resigned" when quite plainly she was terminated. Along with the secondary point that Sundar is now trying to make it sort of look like he understands what went wrong and it apologizing for it. Except he doesn't and he isn't, really.
Gebru wrote an email to a wide audience with a series of unprofessional complaints
Would you care to enumerate what these were, specifically?
The difference between resigned and fired doesn't seem that meaningful to me. Is there an agreed on definition of the terms?
As I understand it Gebru said something like I need X, Y, and Z or I can work on an end date. Google then "accepted her resignation" or fired her. However you prefer to say it.
I can see that mattering if she needs to collect unemployment or something, but then it seems like a legal discussion which I'm not equipped for. In the vernacular, it seems like she threatened to resign and Google accepted it. Either way, I don't see why a vocabulary dispute is a big deal.
Finally, regarding Pichai's comments, I've not read them so I don't know what he did or didn't say. You seem to be assuming he should apologize and complaining that he didn't though. I'd start by asking why he should apologize? I'm not really convinced Google did anything wrong here.
My colloquial understanding is that resign is when you end it and fired is when they do. In this situation it seems half and half, which may be why the difference between the terms doesn't matter to much to me.
Using your definition, yes, it seems clear this was a firing, but again, so what? Does a disagreement on terms really necessitate the drama (news articles, petition, protests)? If she was being denied money somehow because of the distinction then I could see it being a legal case, but I haven't heard anything like that. Google said she resigned, she said she didn't, okay.
If anything, saying she resigned sounds a bit nicer to me than saying she was fired. Probably best if Google had not commented on it at all.
She wasn't asked to change it, and she didn't refuse. She was summoned to a meeting where she was informed that the paper must be retracted by her, due to substantive feedback that was provided anonymously, that would not be provided to her. She ask to be given the feedback and a chance to address it so that the publication process could continue. After some discussion, she was told her manager would be allowed to read her the feedback, but she wouldn't be provided a copy; regardless, the paper must be retracted. She was not given an opportunity to edit the paper based on the feedback.
This was the cause of her email to the list. Google hasn't disputed any of these details.
Jeff Dean's document makes it sound like they aren't opposed to Gebru's paper in principle, but are opposed to it without what they consider missing context.
"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 — for example, it talked about the environmental impact of large models, but disregarded subsequent research showing much greater efficiencies. Similarly, it raised concerns about bias in language models, but didn’t take into account recent research to mitigate these issues. 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. "
And:
"But the paper itself had some important gaps that prevented us from being comfortable putting Google affiliation on it. For example, it didn’t include important findings on how models can be made more efficient and actually reduce overall environmental impact, and it didn’t take into account some recent work at Google and elsewhere on mitigating bias in language models. Highlighting risks without pointing out methods for researchers and developers to understand and mitigate those risks misses the mark on helping with these problems. As always, feedback on paper drafts generally makes them stronger when they ultimately appear."
I'd be surprised, having read this, if Gebru and coauthors made the requested adjustments, the paper wouldn't have been approved by Google. That's my intuition at least, maybe I'm off.
This is in line with my entire problem with the whole Twitter fiasco: screaming about problems but not offering any solutions.
I don't know the lady well, but given her background I assume she's intelligent. So, this isn't to degrade her but that trait in general - in what job can your role be to just tell people that they're wrong? How long would a roofer last that just stood on the ground and yelled up they were doing it wrong...even if the roofers really were? Or a programmer who denies your PR because it's wrong? Communication is an invaluable life skill. Handling disagreements with tact and real suggestions in hand will always be better received.
You ALWAYS must provide a constructive suggestion. Only criticizing shuts people down. However, rotten leadership can't handle anything less than full loyalty, and this is rampant in companies, and now also politics.
Dean’s explanation makes sense if it is based on a standard that is consistently applied to all research papers coming out of Google. But there are public statements from other Googlers implying that it’s not, as noted in this article:
Gebru's account is that she was first refused the feedback, and then offered only a recitation of the feedback, not the written account. She offered to rewrite to account for the feedback, and was denied. Dean doesn't dispute this.
What do they have to gain by insisting they were right even if they were? Do you think it would change anyone's minds or just make this whole thing last longer in the press?
Even if they could definitively prove their side of the story, it would just make them look like even more of a bully to target an already let go employee.
