It sounds to me like Google has a policy that papers must be internally reviewed before they can be "published". She understood publish in the academic sense, while Google views sending the paper out for conference review as publishing. The paper ends up failing internal review, so per policy it must be promptly retracted. This is confusing to the academic who expects to get access to the raw review responses so that the paper can be fixed. After all in her mind it is not published yet, and updates can be submitted to the conference to fix the issues.

The underlying reason for the paper failing internal review looks to be that it basically condemns Google's whole approach to AI (big models) while failing to take into account some relevant research that may favor google's approach. (I have no way of guessing if it would have passed review if it did take into account those, but still condemned the approach).

The massive amount of miscommunication makes her feel like she was being oppressed, due to race and or gender. Hence her aggressive tone in her emails and ultimatum.

Looks like a fairly classic case of a major blowup due to poor communication.

While there hasn't been enough evidence to fully condemn either side, I disagree that this is just poor communication.

Unless Google deliberately changed the enforcement of the policy just to mess with her, she should have known the policy. It doesn't seem to be a complicated process, and 2 weeks is a reasonably short time to wait.

On the other side, Google has been in this game long enough, that they must know a paper can be updated in this case. So there wouldn't be a misunderstanding there, either.

What exactly happened is unclear so far. But I doubt it will come down to just a series of misunderstanding.

On a side note, I suspect that working research for a company with a conflicting interest in the same area is generally a bad idea. At some point, the dilemma of either leaving or stopping caring about it would present itself.

I will say that having many processes, selectively enforced is a common cover for disparate treatment.

That’s a good point. How many other Google researchers have submitted papers to conferences without waiting for the 2 week internal review to complete? What happened in those cases if the paper failed the internal review? Without more context it’s impossible to tell if this was a case of discrimination or just a case a researcher not following Google’s internal policies and suffering the consequences.

I worked as a run-of-the-mill engineer at Google and was involved in publishing a paper externally. We couldn't do it the first time around because we finished only 3 days before the external deadline and there wasn't enough time to get it approved internally. This seemed to be very standard around me. People were actually submitting well ahead of time and to me it seems like that Gebru expected the rules to be bent for her if she realistically believed a single day would be enough. Let me just point out that this is a standard practice even in academic research at universities when larger teams are involved.

Having worked at three large academic institutions and been part of the submission process for about a dozen manuscripts, I have never seen or heard of a requirement for an internal administrative approval.

IRB approval for study designs, yes, admin submission of grants, yes, but never an authorization that the results and manuscript could be submitted to a journal or conference

Are you serious? Every large collaboration in e.g. physics has those. You simple cannot create a splinter cell within the organization and publish whatever you want. You have to coordinate with everybody because the work is never truly yours. In case of e.g. CERN, literally thousands of people helped you in some way and they do not want you to potentially tarnish their reputation. Industry AI labs are like that as well. I know this for a fact at least in the organizations I worked at.

It sounds like we’re confusing two things.

What you’re describing in this comment sounds like a consortium. These often an embargo on publication of independent manuscripts until the primary paper(s) have been published. However, once the data’s out, each investigator is free to publish whatever secondary analysis they like.

I understood the Google process to be an _administrative_ approval, based on the conclusions of the paper and unrelated to the quality of the science. This I have never heard of at an academic institution. The final call that a manuscript is ready for publication has always been made by the corresponding/senior authors, not admin.

I don't have stats, but when I tried to do this it was made quite clear that I needed to submit 2 weeks early because nobody who reviews can drop their day work to do a priority review for anything less than a superpaper. Most of the people doing paper reviews (just like in academia) are busy and doing them in their "spare time" (evenings, weekends).

I don't think we even need data on this. One or two day prior is unthinkable, just per common sense. Although data points would be valuable and appreciated but we don't really need data to understand the one or two day is just not enough time for anyone or anything.

It's not like there are a bunch of dedicated reviewers, it's always other researchers who review. I don't think some of the most valued papers like MapReduce or BigTable would be reviewed in just a day or two. It's just not a reasonable amount of time in any condition in terms of reviewing research papers.

In any large company there are lots of process rules and internal deadlines that are unrealistic for whatever reason and don't get followed in practice, leading to many last-minute exceptions. I don't know about the specific case of research papers, but if would not surprise me at all if "one day before the deadline" were a common submission timeframe at Google. Of course that not ideal and probably sucks for the team in charge of reviewing these things, but in that context it wouldn't be such a "damning" piece of evidence against her and the other researchers.

Submitting for an internal review a day before the deadline is not a common thing in Google Research at all. I worked there and actually was not able to submit a paper to an external conference because the internal deadline was only 3 days away and it was likely we wouldn't be able to get the internal approval ahead of time. The rules were clear and as far as I understood it, people did follow them. If there was any selective enforcement, it would be in her favor if she expected a single day to be enough.

A lot of process rules are unrealistic, but that doesn't mean this one is, and that can't be used to conclude that all internal processes are unrealistic and wouldn't be followed. It's an quite unrealistic statement to make. Most places rely on internal processes to work, and it would be extremely rare for any case of exception.

> Most places rely on internal processes to work, and it would be extremely rare for any case of exception.

That's a pretty broad statement with no supporting evidence. In fact, I'd argue that most large organizations rely on internal processes to be ignored or bypassed just as often as they rely on other processes to be followed.

I've been on teams that required "every feature" to go through a design review. Guess what? Suddenly a lot of what we were already working on stopped being "features" and started being "small bug fixes". Approvals for launches are often sought from the last few approvers just before launch because the list of required approvers is so long, and the VP of engineering isn't going to approve until everyone else has already approved etc... This kind of shit is par for the course at a large company because the alternative is basically a "work to rule" strike.

Me thinks it's a common misconception for strongly opinionated people to believe that companies want to hire them in order for them to "fix things and make the company a better place".

In reality, what every company/team/management wants is someone who will give them some constructive criticism on small mistakes here and there that are easy to fix, but ultimately act as a supportive shield for the company against /other/ criticisms. The more credible that someone is, the better, and the more sophisticated that act of supportive shielding is, the better.

In exchange, the company provides support, financial or otherwise, for that someone's personal career ambitions in becoming rich and famous themselves.

Anyone else see it this way? Or am I just too jaded these days?

That's pretty jaded IMHO. I would think most management wants positive change, not just to be shielded from criticism.

But reasonable people can disagree about what the status quo is, how good or bad that status quo is, what counts as positive change, what timescale positive change needs to happen at, and what costs/tradeoffs they are willing to accept to accomplish positive change.

Probably Timnit and the management disagreed on "what timescale positive change needs to happen at, and what costs/tradeoffs they are willing to accept to accomplish positive change".

Not jaded. I think that's actually a very insightful way to put it. Heck I wouldn't even mind if a boss I had put it like that, as it's at least honest and not really unreasonable. I mean, if you go into an organisation and decide the entire premise of the organisation is irredeemably flawed then at some point your feedback is going to no longer be useful, even if some of it is on point, because "this entire enterprise is worthless" is not really an actionable view most of the time.

Sums it up pretty well. Same reason consultants are hired.

Quite jaded but not wrong per se.

Usually is a reason they hire consultants and contractors to do say the unpleasant truths. They can be ignored, reframed, bribed to say whatever, and have the luxury of saying ugly things and then fucking off not long after.

I'd even expect that we only know the tip of the iceberg. You don't leave employment that you are otherwise happy with over such a thing.

There were clearly conflicts brewing below the lid. This was just taken as the final straw. On both sides, I'd reasonably expect. She was dissatisfied with a bunch of things that she couldn't makr public, so she boiled over on this one. That's why many people are surprised this is such a big deal.

And then she gave a ultimatum, which was great for both sides. For the employer since they could use it to get rid of her right away and make it look it's because of this (and "technically" they are right). And for her because she can use it for maximum outrage. It was a calculated provocation. Come on, if you really want to be part of a process of change, that's not how to achieve that. She is smart enough to understand this.

I don't think it's miscommunication in anyway. Timnit had submitted dozens of papers before and unless the process requirements are completely new, she would be well aware of them. This seems more of a case of selective enforcement, where she had been allowed to go around the process in the past (or felt that she has enough leverage to be able to do so now) and then not being allowed to do so this time.

This feels like a classic privileged tantrum which has blown up in the face because Google refused to bend this time.

> This seems more of a case of selective enforcement, where she had been allowed to go around the process in the past (or felt that she has enough leverage to be able to do so now) and then not being allowed to do so this time.

Jeff's email explicitly says the paper was "approved" then submitted, albeit perhaps in one day. The two-week thing might be a policy but would be immaterial if it ended up approved anyway within whatever timeframe. It sounds like a red-herring to emphasize the time-in-advance; in any case the blame, if any, is to be on whoever flipped the approval bit and did not wait for other reviewers (possibly her manager?)

She sends for approval 1 day before the conference deadline, proceeds to submit the paper with conditional approval, and waits for the actual approvals. It is common practice with one side effect: if reviewers don't like certain parts of the paper, they can ask you to withdraw it (since you don't have an option to update the paper at this stage). If the paper was submitted for approval before submitting to the conference though, then they would have some room for back-and-forth engagement with updates on the paper.

Nowhere in the account of the story from the other party anything is specified as "conditional" nor whoever issued the initial approval is mentioned at all. More weirdly, in her account of the story, the "back-and-forth" came from, tada... HR department, which is super weird.

Look, I am not necessarily picking her side in the more general story, but it is apparent this paper submission story in isolation is super weird. It seems feasible that she was in the queue for getting the axe having engaged in prior fights with the company (some mention of threat of lawsuit, etc). I also acknowledge the account of story from her side was quite cherry-picked and actively strategic, if not deceitful, and it does not even seem that Jeff was a direct player in this saga. Looks the highest level person in the loop was Megan, not Jeff, and Jeff is simply dragged into the mix for the opportunity to throw a punch at his reputation. That, however, does not minimize the weirdness of putting the whole thing on this two-week paper submission policy.

I agree with you for almost everything you said. This case is definitely unusual. What I mean by conditional is that the person who approved it initially is not an actual reviewer.

For your last sentence, I think Jeff uses that 2-week policy to state that his position is "technically" right.

I agree -- the passive voice here (intentionally?) makes it impossible to tell whether the external submission was done in good faith:

> 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.

If (to the best of her knowledge) she got approval through the correct process, and then an unidentified party opaquely removed the approval after the fact, I think her questions are justified.

I found the quoted sentence to be very unclear, which is surprising because it is the crux of the matter. From it, I can't tell the timeline between the order of when things happened and who did what. Who approved and submitted what and when?

> where she had been allowed to go around the process in the past

I'm curious, is there anything in the posted information above that points to the fact they were able to go around the 2-week limit in the past? Or is that just a guess?

As a side note, I worked there as a run-of-the-mill engineer and the 2 week deadline was respected as far as I can tell. I once couldn't submit a paper to an external conference exactly because there was not enough time to get the internal review and approval.

It could still be miscommunication. It is quite plausibly that she did not have any previous papers fail internal review, and almost certain that she did not have any fail internal review while already submitted to a conference (and thus "published" in Google's POV).

Not understanding Google's policy of retract it now, possibly re-"publish" (resubmit) later after addressing Google's concerns, could very well escalate things, especially since it sounds like this all occurred right as a bunch of people were taking vacation, so ability to get answers to questions/concerns was likely greatly hampered.

> Timnit had submitted dozens of papers before and unless the process requirements are completely new, she would be well aware of them. This seems more of a case of selective enforcement, where she had been allowed to go around the process in the past (or felt that she has enough leverage to be able to do so now) and then not being allowed to do so this time.

If the established practice differs from the nominal process, a no-notice switch to rigorously enforcing the nominal process is for most practical purposes no different than a no-notice change of process. Out of date documentation is at least as common in human processes as software; the real process is what is actually recurrently executed, including how that recurrent practice takes into account position of particular participants.

Also, the description of the official process (much less what is typically done in practice) seems to be in dispute:

https://twitter.com/timnitGebru/status/1335017524937756672?s...

> classic privileged tantrum

just... wow. Can anyone point to a man ever being labelled in such a way on HN?

Edit: comparing the number of comment hits for " his tantrum" vs " her tantrum" to the relative numbers for " his " and " her " shows that "tantrum" is, indeed, used 2.5 times as often to describe women as it is for men.

It's not automatically a given that women don't do things that get labelled as tantrums at a higher rate. To demonstrate this is sexist, you would have to show that people judge the same behavior as tantrum or not tantrum at a different rate depending on the specified gender. Which may well be the case, in which case please link the study.

To turn it around: an interesting self-experiment is to try flipping the pronouns in the story and seeing if your internal reaction to it changes.

FWIW, searching for "musk tantrum" gives 24 comments. (Musk is a known nutcase, though, so it may be an unfair comparison.)

The price suspect I think about when I hear "tantrum" is Trump, so I don't know...

> the academic who expects to get access to the raw review responses so that the paper can be fixed.

Asking for the review notes is one thing, it seems like the main demand Jeff refused was giving the names of every reviewer. Why would the names of the reviewers be relevant to the academic, unless they specifically wanted to use the reviewers identity in order to undermine the notes?

Gebru has a history of using her considerable social media following to bully people she has minor disagreements with. This alone would practically warrant anonymity to make sure the reviewers do not get doxed by her.

Blind peer review should be pretty obvious to anyone from an academic background anyway. Asking for reviewers names from a publisher would be an outrageous request.

Was this a Google internal blind peer review? It sounds more like management pushing back for management specific reasons. I'm not sure it's the same thing, or that management should expect to be able to roadblock publication at the last minute without working with the authors directly.

> This is confusing to the academic who expects to get access to the raw review responses so that the paper can be fixed.

At most CS conferences, there is no second round of reviews (apart from a few PL ones that have been experimenting lately). If you get rejected, it's with the expectation that you make whatever changes you see fit and then submit anew to a different conference. In addition, there is generally no way to update a submitted paper after the submission deadline in such a way that the changes can still be taken into account in the single round of reviews, and making major changes after the accepting venue's review stage is very much frowned upon (as those changes would then not be reviewed). So her options would have been:

1. retract the paper;

2. get the paper rejected from the conference, and resubmit it later;

3. get the paper accepted and have it published without the major changes requested by Google, thus violating Google's policy/interests;

4. get the paper accepted and perform the major changes requested by Google at the camera-ready stage, thus violating academic convention and possibly even specific rules set out by the accepting conference.

2 is mostly 1 with more of everyone's time wasted, and 3 and 4 both violate some obligation or another. From Google's point of view, she might also well have gone for 3 without keeping them in the loop to present them with a fait accompli, and her enormous amount of social capital may have made it easy to get away with. Demanding 1 therefore seems like the most reasonable choice if they want to retain a credible norm of subjecting publications of people who work for them to prior review.

It seems like she has been working at Google for a fair amount of time. If so, I’m surprised that there was still confusion over when a paper needs to be taken through the internal review process. Outside of that, I found this to be a good summary of the situation. It omits her widely distributed email denigrating the entire diversity approach taken by Google, though.

Looks like a fairly classic case of a person who injects race and gender into every topic... injecting race and gender into a topic.

Obviously it's hard to pass a judgement without knowing the paper contents. Reading the emails carefully, it seems Jeff acknowledges that the initial review culminated in an "approval" despite some other "reviewers" still in pending state. That is a curious detail. Presumably someone with authority must have flipped the Approval bit within that 1 day. Why that individual did not object? The two-week excuse sounds like a made-up excuse post-facto based on the written policy but clearly that alone does not seem to be the reason this blew up.

This is pure speculation, but it looks like there may have been some additional backchannel from the conference reviewers to their buddies at Google that caused a follow-on sensitivity or objection later from other functions at Google, namely PR & Policy (based on her tweets), which may have been why Timnit was super curious about the identity of the feedback authors.

The back channel theory sounds like a good point to me. Anonymous peer review is largely a charade, I'd be more surprised if Google didn't have fingers on the scales and flies on the wall at top ML venues than if they did. It's very possible that the leak/breach of anonymity is what kicked off her demands to get the names of people who filed complaints that led to the paper's retraction.

> The massive amount of miscommunication makes her feel like she was being oppressed, due to race and or gender.

That seams a big jump. Isn't it more likely that the treatment is due to pushing for ethics in an organization whose business model is antithetical to ethics?

Google is investing in coming up with smaller models aswell i.e "TensorFlow Lite". I think it is unjustified of her to condemn approaches at this stage, all research starts humble and almost always inefficient. People were doing Resnet-50 not so long ago and then processes improved. People came up with better configs with lesser layers YOLO for instance. So big models in and off itself isnt a problem. I dont see this as a good enough reason to resign.

That sounds like a horrible use of human hours, especially at Google salary level.

Paper writing is painstaking and mentally tiring. It would be much more efficient if they reviewed results and an abstract and then decided based on that "Go ahead and write the paper and publish it" or "Don't waste time writing it up".

Money is not an issue at all. It's about protecting the perception of quality that comes from having Google co-authors, and ultimately the Google brand.

Otherwise, everyone and their mother would milk the Google name by submitting papers in every direction.

Sure, then they should have a 2-step process. The first step should be approving writing the paper, and any high selectivity should happen at that point.

The second step should be approving the final contents, and that process should not reject a high fraction of submissions and should allow resubmissions with changes.

It would suck to spend 200 hours figuring out how to fight to position figures in LaTeX only to be rejected by some dudes who think the figure is confidential.

That wouldn't be effective though. Most of the time these reviews are to make sure company IP isn't being published, which could be outside the main results or abstract.

I think it's worth highlighting that Google's internal review of papers is probably significantly different than what we'd expect in an academic setting. I'm not big corporate entity expert, but I wouldn't be shocked to find out Google's internal review heavily weighs impacts on Google from both a public relations and earnings aspect.

What I really cannot understand is how someone who works on ethics can think that revealing names of anonymous reviewer is OK?

Why would they need those names anyway? I'm also unable to think of any reason they might need the identities of the reviewers.

My guess would be that they took the issue as a personal offence, and they believe that someone higher in the power structure has some personal beef against them as a person, rather than against parts of their work. They want to confront that person.

Maybe they want a good old fashion public Twitter judgement for the reviewers. She has a crowd.

I'm curious about this "submitted paper for review 1 day before publication" assertion. From my reading, it sounded like work on this paper had been going on for a while. The email quoted on "Platformer" makes it sound like communication with HR (which seems weird, honestly, for a research paper) had been going on for at least two months.[0]

> Imagine this: You’ve sent a paper for feedback to 30+ researchers, you’re awaiting feedback from PR & Policy who you gave a heads up before you even wrote the work saying “we’re thinking of doing this”, working on a revision plan figuring out how to address different feedback from people, haven’t heard from PR & Policy besides them asking you for updates (in 2 months).

The NY Times reported that four of the researchers working on the paper also worked at Google.[1]

> In an interview with The Times, Dr. Gebru said her exasperation stemmed from the company’s treatment of a research paper she had written with six other researchers, four of them at Google.

It seems to me Google management should have been aware of this paper prior to the day before publication.

----

[0]: https://www.platformer.news/p/the-withering-email-that-got-a...

[1]: https://www.nytimes.com/2020/12/03/technology/google-researc...

I don't think google are saying they weren't aware of the paper prior to the day-before, just that they hadn't seen the actual paper itself before then.

Out of curiosity do people think that a company either should or would publish self-funded research that's detrimental to their PR? I'm not sure that this is a realistic expectation with our current methods of company governance (which should probably be changed). I'm assuming the research isn't overly safety oriented and is more like "overuse of google leads to decreased overall well-being" rather that "use of the cigarettes that we sell leads to significantly reduced life expectancy". Of course that statement begs the question of where the line between the two lies.

I did find a Reddit post with the abstract and some discussion from a person who claims to be a reviewer.

https://www.reddit.com/r/MachineLearning/comments/k69eq0/n_t...

In my opinion, it seemed pretty low key. But I've only read the abstract.

In terms of Google's expectations and what should we consider reasonable, I'm really not sure... I do think that when they hired this person to be "technical co-lead of the Ethical Artificial Intelligence Team at Google", they probably should've guessed that there could, possibly, be a conflict between "make loads of money" and "ethical".

I think one in important point is that she submitted the paper the day before the deadline when 2 weeks is usually required for review. It was approved then, but after the review process worked out she was asked to retract her paper.

I think I covered that in #1 & #2, unless there's something subtle I missed.

I think the subtle part might be "It was approved then", I read that somewhere too and it was not clear whether it was approved by google despite it not following the normal 2 week process or if it was talking about some other form of approval.

From Dean's email:

>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.

Not sure what to make of that - maybe that Gebru submitted the paper and then immediately clicked the "approve" button herself?

Most definitely not. I think that is a critical detail. You can be 100% sure if she clicked on Approve herself, Jeff would have either omitted that fact entirely or definitely explicitly mentioned it was a self-approval. Some other person clicked on Approve for sure and they are somehow not putting any responsibility on that person in this public narrative.

Yeah, I had a hard time making heads and tails of what that bit meant, exactly.

it is sort of a placeholder approval. The actual reviews may be done by a legal team, HR/PR, other researchers in the field etc.

Do you have actual information or just randomly speculating about the nature of that "approval" under a newly created anonymous HackerNews account?

A "placeholder approval" is also known as an "approval". If they didn't want to approve it, they shouldn't have done it.

what you refer to as "they" are two different entities.

If the company doesn't even have a clear indication who is authorized to approve things, that seems much more a Google problem.

> Timnit Gebru demanded access to all of the internal feedback, along with the author of the feedback. If Google wouldn't provide this access, she said that she intended to resign and would set a final day.

Is the part that makes me cynical about Google’s motive. What sort of process calls for XFN feedback on your work but withholds that feedback and its sources from you? This is very odd.

Depends on how many followers the author has on Twitter and her history of Twitter use. After what I've seen, I think it was a good call.

There is one additional detail that is relevant I'm aware of; Timnit also sent an internal email to staff as reported here:

https://www.platformer.news/p/the-withering-email-that-got-a...

One can judge the content for themselves, but it was clearly intended to ratchet up pressure on management.