> Timnit co-authored a paper with four fellow Googlers as well as some external collaborators that needed to go through our review process (as is the case with all externally submitted papers). We’ve approved dozens of papers that Timnit and/or the other Googlers have authored and then published, but as you know, papers often require changes during the internal review process (or are even deemed unsuitable for submission). 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.
> 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.
> Timnit responded with an email requiring that a number of conditions be met in order for her to continue working at Google, including revealing the identities of every person who Megan and I had spoken to and consulted as part of the review of the paper and the exact feedback. 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.
Big red flag. Why are the identities of reviewers important here? Did she plan to take those reviewers to the court of public opinions for a trial/exposure?
A lesson is delivered in time IMO. Those entitled people need to be called out.
Google went through their standard process, and the paper was rejected.
If they have actually approved dozens of Timnit's papers before, then the likelihood that this rejection is due to discrimination seems low.
There is not a single software team on planet earth that would not want Jeff as their technical leader.
It is like comparing a high school athlete to Usain Bolt.
If the paper was bad it should have been rejected by the conference to which it was submitted. If the paper was critical of things Google does and it didn't like that, that's pure suppression. And, as you say, academic norms provide a period of rebuttal, it's unclear how much was allowed here.
Secondly there is value of having anonymous review in an academic setting, however it sounds like they tried to enforce that anonymous review for the company, which again doesn't serve the same purpose. People should be told exactly why decisions are being made so they can correct their mistakes in the future. If you aren't willing to invite someone into a decision making process, when a large portion of their research is in biases in institutional decision making, then you have got to know you're in for a bad time.
If a company publishes only papers that clearly advance its interests, but the research is sound and independently reproducible, does that bias really lessen the perceived value and quality of the research?
Fair academic criticism: We shouldn't publish this paper because it has flaws X,Y,Z. (usually followed by an opportunity for the researchers to address X,Y,Z).
Unfair and/or non-academic criticisms: We shouldn't publish this paper because: I don't like the author/authors institution. I don't like this approach, approach X is better. I don't like the conclusions.
The last is potentially relevant here. And yes, clearly the quality and value of research is reduced by filtering based on results you want to promote. Which well may be in a companies best interest, but there are sometimes difficult lines to draw - when exactly do you move from "promoting the companies interests" to "propaganda intended to mislead the public"?
The done thing is generally for the reviewers to suggest additional references to improve the paper. If any mention of related work is entirely missing, then that's another matter, but a reviewer thinking that the authors should cite a particular piece of work (e.g. the reviewer's own paper) is not normally something that could lead to outright rejection.
Why did they not fight to keep her, if she was so singularly brilliant? IMO, it's because she wasn't an employee they wanted to keep around.
People don't get fired for asking for opportunities to respond to criticisms. My guess is there were previous incidences regarding her behavior and, in this situation, she threatened to quit and they accepted.
Industrial researchers and non-PIs (IE, research analyst working in a PI's lab) don't have the same freedom and integrity rights as a PI. To me this seems like a reasonable (and probably the only reasonable) policy for industry.
I don't really think it's dystopian- if a paper I wrote reached the point of needing a retraction it means some serious messup occurred or there's a major PR fiasco.
Note that in this case I am not sure if there was a retraction, or an desubmission- the paper hadn't been published yet so it's not really a candidate for retraction. The term for this is 'withdraw submission'
It's just as if you'd want to open source some code that you wrote for your job, and suddenly the boss says "hey, don't do that" - there are many reasons why they might want to not publish it, and it's not your code to publish as you want.
In this world, where there are bills to pay and you can't disadvantage yourself if you want to be ok. Having bills to pay is, I agree, just mad.
To some extent, you have to realize that one day you will fall on the wrong side on the fence and you'll be the victim of your own creation.
The original article has been updated with Jeff Dean's response, while a member of Gebru's team (who still works at Google Brain) states that Dean's email is misleading and inaccurate: https://twitter.com/alexhanna/status/1334579764573691904
As I understand it: The paper had already been submitted to a conference, and the "cross functional team" Dean speaks of basically reviewed it like peer reviewers and said you have to retract it. It sounds like Gebru and her co-authors were not given an opportunity to respond to the criticisms to stave off retraction. If you've experienced peer review, you know that such reviews are by no means a definitive analysis of a paper, and one usually gets the opportunity to respond. Here, it sounds like anonymous internal peer review was used as a hammer to retract it. Dean's descriptions of the criticisms sounds rather pedestrian to me, as a researcher, and I'm sure Gebru and her co-authors would push back on these characterizations.
Honestly, it sounds like the paper was attacking some sacred cows internally. And Gebru's work probably threatened other researchers on other teams. Not unlike traditional peer review mind you, but at least one can usually post in on arXiv without worrying about this.