Ethical AI researcher Timnit Gebru fired from Google
twitter.com
twitter.com
Without knowing more context, it sounds like she tried to do some hardball negotiating by threatening to leave, and got her bluff called.
Quoting the tweets for posterity, link at the bottom:
> I need to be very careful what I say so let me be clear. They can come after me. No one told me that I was fired. You know legal speak, given that we're seeing who we're dealing with. This is the exact email I received from Megan who reports to Jeff
> Who I can't imagine would do this without consulting and clearing with him of course. So this is what is written in the email:
> Thanks for making your conditions clear. We cannot agree to #1 and #2 as you are requesting. We respect your decision to leave Google as a result, and we are accepting your resignation.
> However, we believe the end of your employment should happen faster than your email reflects because certain aspects of the email you sent last night to non-management employees in the brain group reflect behavior that is inconsistent with the expectations of a Google manager.
> As a result, we are accepting your resignation immediately, effective today. We will send your final paycheck to your address in Workday. When you return from your vacation, PeopleOps will reach out to you to coordinate the return of Google devices and assets.
[1]: https://twitter.com/timnitGebru/status/1334364732480958467
You cannot fire someone for membership in a protected class (race, religion, sexual orientation, etc) or participation in protected activities like reporting an employer’s violations of the law, but it’s difficult to prove intent in such cases.
There are exceptions for unionized workplaces with collective bargaining agreements. Employers in those situations cannot fire employees without “just cause.” Google R&D is not unionized, however.
I assume they wanted her out and she accidentally gave them immunity, so they jumped on it.
She made some pretty serious demands, not having been met, with such politicization there's no reasonable way to accommodate a longer handoff, and she knew that. Waaay too much risk there.
She set her own terms, they accepted, that's it.
She should have been more transparent in this, because in her Tweets she wasn't forthcoming that it was her terms.
There are a ton of people who are sympathetic to those who are oppressed, but misrepresentations, cancelling etc. on the wrong terms are going to lose general sympathy very quickly.
None of this should be construed as a measure of whether or not Google is 'Good or Bad' or whatever, all these things are individual situations, with their own unique circumstances.
All we know is that she gave a hardball ultimatum, and Google accepted option 2. I don’t know enough about what’s going on, but if you say “Do this, or I quit,” you have to be prepared for them to show you the door immediately. That’s how at-will employment works.
"“Do this, or I quit,” you have to be prepared for them to show you the door immediately. That’s how at-will employment works."
That's pretty much it.
She must be smart enough to know better and possibly she was looking for an exit of some kind.
> In the letter she criticised the use of pre trained language models in Google's products (e.g., BERT is now used for most searches, machine translation, etc.). Apparently, the two conditions she mentioned concern how Google goes forward in deploying these models despite the warning of (their own) AI ethics researchers about biases that are manifested in these models.
Is it though? There is bias in everything.
There is bias in Googles original search - it crawled content. That content will be biased.
Google search has been biased since the start. It has nothing to do with AI.
Biased is also contextual: what does it mean for all of us non-Americans to see tons of American content in everything - the 'American bias' is overwhelmingly the strongest bias, where are the concerns about that?
And how does bias imply a 'lack of ethics'?
The entire consideration is ridiculous:
1) AI is not special and odes not deserve it's own ethical oversight. Every social tech has issues and it all needs to be thought about.
2) Someone with no real training in the issue may very well merely be injecting their own politics into the situation, and possibly overstating issuses.
3) There's nothing objective about morality or ethnics, so it's really hard to even find such at thing as 'objective'.
What the company needs is a clean, comprehensive framework for the issue, and probably some independent oversight that can given them private assessments of where there are red flags.
Not individuals who want to make a name for themselves on the issue publicly, and who might be part of some kind of ideological movement.
Your disagreement seems less to do with that and more to do that you disagree with where her ethics rest. I think the reason why she pushes for marginalized groups is that those are the groups that are less likely to be able to stand for themselves and thus become the victims of the bias. A bias against corporate or powerful interests will likely be rectified relatively quickly.
But hiring activists, and particularly those with a focus on a very specific technology ... I think is a double-wrong.
'Ethics' is all encompassing, every company has to deal with it ... it needs probably specialization and some cold calculating.
If there were some gigantic loophole whereupon marginalized people were truly thrown under the bus here, then maybe there's a case, but I seriously doubt that.
It's also pretty common in roles like sales and PR to be shown the door the day you give notice, especially if you're going to a competitor.
If you make an ultimatum to your employer, and they don't want to accept, assume that it's over, right then and there.
Unless you're Michael Jordan, or have some kind of really existential terms with the company.
But at the very least, I'd guess that this will make any potential future employer think twice if they really want to be her manager.
And lastly, I remember the name Jeff Dean from reading about advanced AI research, so if someone publicly says "Jeff Bean says XY is unbearable", I'd be inclined to just believe it.
All in all, this looks to me like Timnit Gebru could improve her career chances by being a bit more diplomatic with her public announcements.
He's also a rather mild mannered and very nice man. When I worked there I never heard anyone say anything bad about Dean. He rarely expressed strong opinions even on engineering topics; the sort of guy who prefers to win debates by simply coding up the best solution.
The name Timnit Gebru also rings bells for me. Last time this name came up she was busy chasing Yann LeCun off Twitter with pitchforks. LeCun! One of the other major luminaries of AI research. In fact she stirred up so much trouble and made so many outrageous accusations he quit Twitter altogether. She appears to be a serious troublemaker, far more interested in identity politics than AI. She was an embarrassment to Google before and it's good for them they finally canned her, but seriously, why did it take this long? And who will hire her next? I'm sure someone will be dumb enough to do that but hardly any companies have openings for "ethical AI researchers", and she already burned her bridges at Facebook.
Because public division creates sides, and taking one of those sides gives you favour with one group.
A pragmatic, mild mannered person who works on such issues behind the scenes is not going to be popular.
But division makes enemies in one camp, allies in the other.
Now we all know someone's name whereas we wouldn't have before.
So one can make a career out championing said allies.
I think a lot of people have this instinct, just expressed in different ways.
that just reminded how Marisa resigned - if i remember correctly she just called on Monday morning to notify that she is already at Yahoo. So the "at-will" works both way.
>> certain aspects of the email you sent last night to non-management employees in the brain group reflect behavior that is inconsistent with the expectations of a Google manager.
as my friends managers explained to me - a manager represents the company to employees, i.e. when a manager speaks it is basically the company speaks and thus [until the company reacts quickly by disowning the manager's words] the company bears the burdens and responsibilities that the manager made promises/representations about - similar to that relationship between Pope and God pictured in the Dogma - so managers are trained to be very accurate with their words, and understandably a manager going off-rails is like a fire emergency to be dealt with immediately.
So looks like Timnit sent an e-mail to the group which wasn't OK in the eyes of the higher management, then sent a strong mail about her demands to the same management and used "I'll resign otherwise" card to force her way.
The management, agitated from the first e-mail, didn't buy it and sent her home. The way they did is neither elegant nor polite but, she pushed a lot to get this response it seems.
This is my understanding of the issue and I may be completely wrong but, this is what I was able to decipher.
https://www.platformer.news/p/the-withering-email-that-got-a...
It also sounds like she did lawyer-rammed the Google a year before, so they were tip-topping around her strictly by the book, and with such clear an offense as that email she just threw the HR a freebie, as another commenter put it, so they could now get rid of her strictly by the book.
> Have you ever heard of someone getting “feedback” on a paper through a privileged and confidential document to HR? Does that sound like a standard procedure to you or does it just happen to people like me who are constantly dehumanized?
Taking a dispute with your management chain re: feedback that said management chain felt they had to take actions like a private, hr-involved feedback session, and broadcasting that dispute to a mailing list of peers, downlevels, etc.
> But now there’s an additional layer saying any privileged person can decide that they don’t want your paper out with zero conversation.
This was, from the perspective of google management, probably very much not a reasonable characterization of the aforementioned circumstances. Managers represent the company, including publicly. Timnit was obviously in an awkward position where she seems to have been a player coach, ie a manager also doing IC level work (in this case, research). It feels like she expected her actions as an IC not to be viewed as actions that also came from a manager. That's a very hard dual role to have.
Third, maybe shouldn't get you fired in a perfect world, but prolly will in ours:
Public internal criticism of DEI efforts on the part of your management chain. Imagine how that reads to a line-level employee at Google. Particularly an URM.
Fourth, same vein -- auguring on an internal mailing list to get the CBC to criticize your employer
> I believe that the Congressional Black Caucus is the entity that started forcing tech companies to report their diversity numbers
I'm only paying attention because I'm curious if Timnit was hired to -- if you will excuse the phrase -- whitewash google's policies, and she didn't understand what Google was buying; or if there genuinely was some conflict here. I suspect both from different players in the executive ranks.
This is the world we live in. Twitter is a god awful pitchfork paradise. I cannot stand it.
Is the assumption here that Google is somehow better behaved than other corporations? I don't think there's any factual basis for such thinking. In my experience, corporations are laser focused on money, quarter-by-quarter, with little to no concern for the human cost. Google has been asked for comment and has consistently declined, in my opinion we should give much more weight to the real person here.
We all have much more in common with the AI researcher than the large, faceless corporate entity. If we are at all principled, it's easy to imagine ourselves in a similar situation.
Why the hand-wringing over this email they sent, the suspicion that it was all "hardball negotiating" and "bluffing"? The email to the mailing list pretty clearly outlines their concerns and reaction to Google, it seems a pretty safe assumption that their email to Google management covers much the same information.[0]
This idea that they are bluffing strikes me as insulting: is there any reason to suspect they are bluffing? I don't see that in any of the coverage, by my reading this person felt this back-and-forth over the publication of this paper at the last minute was the last straw, their exasperation and frustration with Google is clear. It seems to me they were entirely willing to resign.
[0]: https://www.platformer.news/p/the-withering-email-that-got-a...
She's even tweeted that she suspects there was a team of people behind the decision to accept her resignation. All those people don't put Google first. The hard truth is that she was probably extremely unpleasant to work with it (this is my reading of the above link and from what I've witnessed on Twitter) and they had just reached their limits. Not Google, but the people that had to work with her (I'm clearly not referring to her subordinates who are all apparently "indebted" to her which is a whole other issue onto itself). She comes across as very entitled and quite frankly, although I hate to say it as it's a bit of a cliche at this point, narcissistic.
These are all based on what I've been able to find online. I've never met her in person. I'm trying very hard to not see her negatively but it's also hard finding evidence that's she's not hostile to those that don't 100% accept and agree with her. I'm definitely open to seeing her in a different light if anyone wants to share links. However, just because you research ethics doesn't mean you practice what you preach nor that you're a saint who can do no wrong.
Would that be somehow bad? I'm not even entirely sure of what does that even mean. If I submit a resignation letter, I don't even expect it to be rejected, engineers and researchers are not some kind of ministers. At most some people can consult each others and ponder if its worth to attempt to retain her, but even then the concept is not entirely one of "I/we don't accept your resignation letter"...
I'd suggest given Timnit is choosing to only tell a very small part of the story here, and Google is obviously not going to publicly comment on an HR matter, especially one this sensitive, that it's best to withhold judgment.
Does anyone else remember when she picked a fight with Yann LeCunn by deliberately misinterpreting his remarks to cast him as a racist?
Seems reasonable to assume she picked another fight. Reading the Twitter threat, it reads like she made demands in a way visible to a large number of employees (something about authorship on a paper, to skim the Twitter thread) and perhaps attacked Google while being a manager there (the emphasis on talking to non-management employees sounds like this). I assume the email will leak, this being Google we are talking about.
LeCunn made repeated attempts to explain his position, only to be virtually shouted over about he wasn’t listening to people of color because he pointed out that a model trained on White people learned to produce images of White people, and that if it had been trained on Black people it would produce images of Black people. It was a research paper, not a production system, and Gebru and her acolytes were much more interested in scoring cheap points than in having a serious conversation about either ML fairness or the merits of this particular paper.
Seriously, the primary sources can be read by anyone.
But this claim is certainly not necessarily correct, and he shouldn't have been so confident about it. Any part of a system can contribute to bias and that includes the model design, not just the data. If they want the system to work then they need to actually demonstrate it works, not just say it could with X change without testing that.
Though I remember people arguing it didn't work properly at the same time they were saying it was evil (because it contributed to surveillance). Which is odd because if you don't want it to exist, you shouldn't want it to work either.
The most obvious and easiest way to clear this up would be for the original researches to train the model on a "Senegal" dataset as mentioned by LeCunn and see what the results are.
While it would be a symmetric situation a vacuum, we do not live in a vacuum. And acknowledging dataset bias by itself doesn't address the bias meaningfully. In practice, we often treat the bias as an exogenous factor when it is not, moving it outside the scope of our responsibility. But it is very much the product of our work, a reflection of our choices, values, and beliefs about what to prioritize. We can't abdicate our responsibility for it; the stakes are too high.
I am not sure why the wording around this has to be so abstract.
As you say, it's a reflection of our beliefs about what to prioritize. If you're interested in developing methods of image upscaling that generalize better and preserve facial properties that were underrepresented in the training data, that's an interesting area of research, and you're welcome to work on it, but you don't get to demand that people working on something else prioritize this instead, they get to choose their own priorities unless you're paying them for a particular direction of research (e.g. like Google should be able to direct Dr. Gebru while she was working for them).
There's a responsibility to correct for bias when implementing models in production (e.g. if you'd be actually deploying it in Senegal, then there would be a responsibility to use a Senegal-appropriate dataset), there's a responsibility to acknowledge the bias of a particular algorithm if one exists - for example if it highly relies on contrast values which would be different for different types of faces, then that's relevant, because it's a statement about the generalization of the algorithm; but there's no proactive duty to "address the bias meaningfully", that's nice thing to do, but it's like charity - a voluntary choice to factilitate a social goal, but not a requirement or responsibility to do that. There's a responsibility to correct harms you caused, there is no responsibility to correct harms caused by others, that's a good thing to do, but not a moral duty.
Someone who invests a lot into charity work or addressing bias is doing a good thing, but it doesn't mean that people who are doing other things are abdicating their responsibility - it's never was their responsibility in the first place to fix social issues in the wider society. You can't simply point at random people and declare that they're going to be responsible for something that they didn't personally cause and where they did nothing wrong, that accusatory behavior is unacceptable, so naturally there's a backlash to people who try to assert personal responsibility of others without a basis to do so.
This would be called “research”. That is what she was hired to do at Google. For everybody of every race ethnicity orientation able status and background that would be honored to give their best effort at the opportunity to do research at google, acceptance of resignation was the right thing.
My only point around her feud with LeCunn is that she threw a fit because people including LeCunn pointed out that it is not necessarily a race thing, and she wanted to give it a race spin without confirming that it was actually the case.
And I am not at all commenting on all this drama around her resignation/being fired without getting a look at the contents of the two emails: the one she sent to Brain Women and the one with her demands. But I know this: if I were to issue an ultimatum to my employer, I am treading dangerous waters and should be able to digest the outcome of that, including being fired.
This is not surprising. Black faces and White faces are not the same data manifold. This is like training a network to upsample oranges, then running it on an apple and being surprised when the result is an unusual color and texture for an apple.
I’m not sure what else there is possibly wrong here. Is the width of their convolutions racist? Their choice to work on super resolution? The fact that they released their work for reproducibility?
Not that parameter specifically, but if you say the data is wrong then that means any other parameter might be wrong too. More data might mean the model is too small to fit it, or the hyperparameters might be wrong to train it, or you now have the wrong ratio of other phenotypes (let's say that instead of races…) in the training set and their results regress.
Also, if you're adding people with darker skin, that of course means the pixel values are lower. That matters for image processing code, things like SAD thresholds or noise reduction will work differently.
> Their choice to work on super resolution?
Superresolution is only useful as a toy and they should have mentioned that when they put up a live colab, yes. They added a disclaimer later on and it was good - there was a big issue where people were convinced this was somehow a surveillance technology because they watched too many TV shows, even though it literally can't work that way!
in particular because features contrast (dynamic range) is lower for darker faces.
>Is the width of their convolutions racist?
Not width. As a result of the above mentioned lower contrast, the racist here is the sensitivity of the resulting Gabor filters produced by the training in the first convolution layers and the Gauss filters in the next layer's. I suspect to deal with that problem one would have to normalize dynamic range of the faces, i.e. it would look something like kind of lightening of the dark faces and/or darkening of the white ones.
Since everyone else is just saying "you don't get it" without explaining what "it" is, I will provide some brief avenues of exploration. Because they are brief, they will be coarse and imperfect, and aiming to give you directional assistance on the topic.
So with that massive preamble because I'm not seeking to argue, here you go:
* Pre-trained models already encode much of this bias and if you don't use them you won't get very far very fast
* Large available datasets also reflect these biases
* AI model performance is generally measured against this dataset, further entrenching the bias. If you're better on Indian face generation it will give you barely any benefit on most scoring methods
* Saying "the outcome is only biased because the data is biased" misses the fact that the performance is on the biased data
There's a little more to it but I believe that's the meat of it. Anyway, not too keen on arguing this. Just sharing because it took me some work to figure out what they were talking about and I wish someone had explained it to me so doing so here.
For instance, if I were to champion metric A which purports to measure performance on human faces but it really only rewards performance on Senegalese then models that do better in general may not be recognized for being better.
In an isolated sense this is not a problem. However if the mainstream is that all metrics that are taken seriously are dataset-biased then we'll have an environment where the models will be trained on biased datasets in order to be successful.
For instance if everyone uses LFW to determine how good facial recognition is, then Senegalese fine-tuned facial recognition tech will not be recognized as being good at facial recognition.
So the argument is that dataset bias is built-in to our approach to the problem. I, personally, think that this isn't malice. We need benchmarks to judge approaches against each other. Benchmarks always have a first mover advantage and a massive path dependence issue. The first benchmarks aiming for things on humans do not reflect humanity accurately. These benchmarks became standard among the community. To be taken seriously you have to do well on benchmarks that are standard in the community. Dataset bias is then natural in newer approaches because the approaches are judged against how good they are against the inaccurate (if you will) benchmarks.
So no one need be racist or anything for the end result of the field to end up being discriminatory.
I don't think a successful approach is to call people racist over this. After all, it isn't malice that guides them. The discrimination comes from the sort of historical accident that has North facing up on a map. And no individual is really racist. It's sort of like the Bechdel test: no movie is crappy simply because it doesn't have two girls talking to each other, but if very few movies have two girls talking to each other about something other than boys, then it makes you think "hmmm, why's that the case".
That's, like, not even culture war, it's just basic correctness of the reference datasets. If a fruit classifier was missing oranges, we'd just fix it and move on.
Like, for instance, HN has people who will bring up privacy violations of big tech constantly. They see their role as making sure the conversation is happening. Not justifying this. Just aiming to understand it.
For my part, I prefer to take the approach you're talking about because I, too, think that the fastest path to this is getting the photos, labeling the photos, and then lobbying for inclusion. Ultimately, I think it's okay if things optimize fast for growth and then we fix up issues afterwards. So the people building the benchmark sets weren't able to get a set that's representative of humanity. Should they have waited till they could have done that? IMHO, no. Rapid release moves the state of the art forward and then we can put in all of these corrections as we move.
Then there's the question of whether all-humans dataset is a good thing or if instead a thing that is white-humans and another that is black-humans is better. Anyway, all said, my personal approach to this problem (if I cared about it a lot, which I don't) would be to say "Current benchmarks and training data available bias towards certain races. I'd like to build X/Y to solve that. Here's what I have so far" etc. etc. I think positive engagement like that yields better results because the vast majority of scientists actually aren't weird race supremacists and the vast majority of AI researchers will gobble up any more data you give them which is segmented and labeled differently, the greedy bastards :D
Some challenges that I can still see:
* Getting the data. Might not actually exist.
* Labeling the data. Probably needs some work.
* Getting it into the benchmarks. It'll invalidate old scores, so there's just a product adoption problem here. I don't know how the community handles newer releases.
As a last aside, I suspect that this conversation ended up the way it did because:
a. It's charged. It's race-based differing outcomes. That's a sensitive subject.
b. People feel unheard. This is natural. Like, this is not an 'interesting' problem. It's literally just a data error so the luminaries in the techniques part of the field aren't really that interested in it. And the techniques part is where the sexy is.
c. This sort of thing has a tendency to escalate. One side says "You're not listening to what I say" and the other side says "I'm not racist. I don't get why you're calling me that" and before you know it it becomes "You have to be racist to be ignoring me" and whatnot and de-escalation becomes impossible. Especially because everyone rewards the loudest on each side.
Honestly, I think it's quite interesting to observe and to understand as just a view into the human condition but we use AI models professionally in the GIS space and professionally we just don't go near this at all. No part of me finds it interesting to solve or to interact with the discussion in any way. I only sort of participated in this here because I think I managed some insight into what it is and I wanted to write that down because I wish someone else could have accelerated me into it.
Anyway, I think that's all the insight I have on the subject, so I'm going to just leave it there. Any more and I'll be ass-pulling.
Do something productive and uncontroversial, or something unproductive and very controversial?
Am I missing anything?
I think it’s clear Gebru was acting in bad faith and LeCunn was baffled and trying to deescalate. Gebru responds to each attempt to deescalate by a reply designed to further rile up Twitter, without actually talking to any of the points LeCunn makes. Virtually all of her replies are some variant of “you are wrong, but I won’t say why” and “listen to me because I am Black”.
The same is true of AI research Twitter.
I'd say she deserved to be fired for she is a racist who's damning to a normal society.
Whom did Gebru discriminate against, based on their race? No one.
That episode is unfortunate imho, but she only suspected racist rationales... Calling out racism does not make you a racist, just like calling something "fishy" does not make you grow fins and gills
Which she doesn't.
Most notably, people here have been raising concern about the discussion with Yann Lecun, but that was also just a civil, if tense, discussion.
You're better off not twisting the meaning of words to favor your argument, though. It'll just create more problems down the line
And her paper was rejected and her immediate reaction is to demand the company to reveal the identity of every reviewer? Yeah, right.
> Virtually all of her replies are some variant of “you are wrong, but I won’t say why” and “listen to me because I am Black”.
I don't see that in the summary you posted.
(BTW This is actually the fun part of the Damore story: the guy think he is smart by explaining how "evolution" makes gender different but misses the ABSOLUTE main point of evolution which is that "to survive to an environement you must fit in it". He didn't fit in the google environement, he got fired.)
I also saw the tweets at Yann LeCun. I'm not his biggest fan, but that wasn't right. She's hurting minorities for her own agenda.
Edit: removed info about myself.
Unless you still work at Google, how do you know she actually broke any rules? As far as I know, her e-mail to the Brain list is not public?
Yes, it is a bit absurd. And if you're the first minority hire on a team that hires 1 person a year, you will be paid under the team average for a while, just as if you weren't.
> And at what level are we talking about?
Entry level employees at FAANG mostly. For PhDs managing people, such as the person in the article, I expect there's a much wider band of possibilities.
You can absolutely get paid that much more. Companies make exceptions all the time and justify two people doing the same job by paying more if someone is more experienced even if the technical skills performed and responsibilities by both people are equal. Instead of being rewarded for working twice the hours to learn something faster due to whatever reason someone can still be paid more for basically accomplishing the same amount of progress in their career but taking more time to do it.
Trust me there are all kinds of reasons you can get paid three times more.
What does this mean? That you don't feel like the race you actually are because you don't play the race card?
I hope I misread that and there is a more charitable explanation.
https://www.platformer.news/p/the-withering-email-that-got-a...
The most relevant parts (in terms of back story) should be this:
> A week before you go out on vacation, you see a meeting pop up at 4:30pm PST on your calendar (this popped up at around 2pm). No one would tell you what the meeting was about in advance. Then in that meeting your manager’s manager tells you “it has been decided” that you need to retract this paper by next week, Nov. 27, the week when almost everyone would be out (and a date which has nothing to do with the conference process). You are not worth having any conversations about this, since you are not someone whose humanity (let alone expertise recognized by journalists, governments, scientists, civic organizations such as the electronic frontiers foundation etc) is acknowledged or valued in this company.
> Then, you ask for more information. What specific feedback exists? Who is it coming from? Why now? Why not before? Can you go back and forth with anyone? Can you understand what exactly is problematic and what can be changed?
> And you are told after a while, that your manager can read you a privileged and confidential document and you’re not supposed to even know who contributed to this document, who wrote this feedback, what process was followed or anything. You write a detailed document discussing whatever pieces of feedback you can find, asking for questions and clarifications, and it is completely ignored. And you’re met with, once again, an order to retract the paper with no engagement whatsoever.
I'm just very confused now. What kind of "privileged and confidential document to HR" would be relevant to retracting a research paper? She has some very legitimate frustrations with the process here, but she doesn't seem to be disputing that the document was relevant. In a more normal context, my default assumption is that "privileged and confidential document to HR" means "someone's raised a serious issue and we'll be sued if we don't fix this", but I don't see how that fits here.
> 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.
By virtue of their research area, her and her team no doubt push against this limit very often. Based on this, I’m not sure it’s possible to really pursue many lines of research inquiry into AI ethics and fairness within a place like Google.
Yes. [1]
> Had she accomplished something of great distinction in this field prior to google?
Yes. [1]
[1] https://scholar.google.ca/scholar?hl=en&as_sdt=0%2C5&q=timni...
I see many support her (from a justified? knee jerk reaction?) that being fired from a corporation for questioning it's bottom line is due to the power dynamics. The support to Timnit Gebru (provided her demands are provably good, but only hurts google's bottom line) would be more powerful/rational if the same people will vocally support Damore.
Even if supporting a worker being fired without complete information is a justified position to take based on corporate dynamics, some have started making this into ugly race issue. Some are `supporting` her in the lines of "She was black and hence was fired". If people are judging others as racist just based on tweets now, Jeff Dean seems the opposite of such smear. Even if Dean was a corporate zombie pulling the trigger to save the company bottomline, calling him racist for that without evidence for the latter is insane.
There seems to be a sick trend in progressive politics in US to move away from supporting workers and start making everything into a race issue with no justification.
Gebru was an ethicist who was willing to resign over some deep concerns. Presumably ethical concerns.
We don't know much more about Gebru's situation, but they're different enough that it's not a useful comparison.
> whose continued employment would open the company up to a risk of lawsuit
If protecting google from a lawsuit is your main concern, aren't you for firing her? The memo of James isn't as much a threat to bottom line or a lawsuit as unveiling some unethical practices (which she allegedly is up to due to her ethical concerns).
My point is precisely you can't have it both ways and be consistent.
As recently as last week, while she was still a Google employee, Gebru publicly accused her “privileged White men” bosses of racial bias:
>Nothing like a bunch of privileged White men trying to squash research by marginalized communities for marginalized communities by ordering them to STOP with ZERO conversation. The amount of disrespect is incredible. Every time I think about it my blood starts boiling again.
https://mobile.twitter.com/timnitGebru/status/13317576299961...
>But they are doing exactly the job of squashing marginalized people's voices which is their job. I at least hope no one tries to talk DEI and other things
https://mobile.twitter.com/timnitGebru/status/13317729006936...
“Opinions my own” in her Twitter bio or not, as a manager, she should know better than to publicly accuse coworkers of bias against members of a protected class, particularly if she’s unwilling to make those same accusations in court. Public statements like that can lead to serious legal consequences for the company, not to mention repetutational damage for the accused, and should not be made lightly or without evidence.
Really? I never heard this, and I followed the Damore drama very closely at the time. It's an interesting detail; do you have a source?
https://www.bizjournals.com/sanjose/news/2018/02/22/lawsuit-...
Just look at how she immediately starts jumping to more sinister conclusions just because "white man":
[1] https://twitter.com/negar_rz/status/1334369747241218050 [2] https://twitter.com/dylnbkr/status/1334395186705702913 [3] https://twitter.com/L_badikho/status/1334393782310227970 [4] https://twitter.com/alexhanna/status/1334348137616568321 [5] https://twitter.com/dylnbkr/status/1334372430437994500
AIs were accused of being biased so Google may have hired her as a defense mechanism. Her job was to justify Googles actions, basically a PR job and not a control organ.
Dunno where you've worked but people who are disruptive to team functioning can (and should!) be terminated. Nobody should have to put up with a hostile work environment.
That does not create a hostile work environment. It's healthy and normal. I would expect an ethicist to fall under a similar category.
Have you considered that she has to remain vague due to the potential litigation coming down the road? In fact didn't she leave some tweets hinting at that ? ("Everything I say will be used against me ..." ?)
This seems to be it though, and I quoted some (I think) relevant parts of what she's criticizing:
https://www.platformer.news/p/the-withering-email-that-got-a...
(2) I don't understand how someone can make public accusations without providing the full picture. Maybe she is legally constrained, but then she shouldn't have said anything and handle the whole situation legally first and then write what she wants to write about it. Google might be wrong here, but we definitely cannot see that.
P.S. It seems to me that the only person who handled "firing" professionally was the most ridiculously dressed person on the planet who goes by the pseudonym Dr. Disrespect. So much drama going on around nowadays.
However, the main problem with Timnit and her colleagues is that they come into every situation assuming racism even when the situation is ambiguous. Likewise, if you disagree with her or want to have a nuanced discussion, she'll accuse you of imposing intellectual labor on her and in some sense you are racist for forcing her to engage in discussion.
Again, so while I appreciate how she is purportedly working to make my life better in tech, there needs to be a real awakening as far as giving people the benefit of the doubt and allowing for discussion without shaming. Google has accumulated many people with this type of attitude who have created a culture of intimidation. You can see all of this happening in real time as her colleagues come to her defense knowing little about the situation and assuming that she was fired because she was black.
The casualness to make this now about race/gender as well when that wasn't even her inital argument.
I can't imagine working in this kind of environment. It must be exhausting, always having to tread on eggshells or end up on a twitter firing squad. Not being able to argue points, without it ending up being about race/gender.
Do we really have to pretend that someone blessed to be working on AI at google with the greatest minds on something so exciting is disadvantaged compared to the rest of us?
I don't see how people can deny that people take you more seriously on tech stuff if you're a man. It's pretty obvious to me, honestly.
Equally obvious, you're not going to become a top AI researcher without being immensely talented.
It always was (also) about race/gender!
That said, I don't know Timnit, and I don't know her specific circumstances... I believe that Google is full of people who wants to solve these problems, but somehow leadership fails to capitalize on that. It might only be a few bad apples in the leadership, but each one wields a huge amount of power (hiring and firing people)...
These really exist? How do you know when you've reached perfect diversity? I'm honestly curious.
Is it based on population distributions? What would be an example diversity OKR
https://www.blog.google/perspectives/melonie-parker/how-rete...
Many Amlaw 500 firms have agreed to the Mansfield rule[0].
> Now in its third iteration, the Mansfield Rule Certification measures whether law firms have affirmatively considered at least 30 percent women, attorneys of color, LGBTQ+ and lawyers with disabilities for leadership and governance roles, equity partner promotions, formal client pitch opportunities, and senior lateral positions.
Mansfield rule certification is independently audited on a biannual basis.
It's also common for clients to have their own diversity requirements[1].
[0] https://www.diversitylab.com/pilot-projects/mansfield-rule-3...
[1] https://www.diversitylab.com/knowledge-sharing/clients-push-...
The VP that fired her/"accepted resignation" wrote to her reports about it:
https://mobile.twitter.com/alexhanna/status/1334348137616568...
It's frustrating to realize, but unsurprising: the people higher up in the hierarchy will always have more power than those below.
Hopefully some clarification comes out about what her demands were and the internet drops their pitchfork about something that happened long ago
https://twitter.com/timnitGebru/status/1334341991795142667
> Apparently my manager’s manager sent an email my direct reports saying she accepted my resignation. I hadn’t resigned—I had asked for simple conditions first and said I would respond when I’m back from vacation. But I guess she decided for me :) that’s the lawyer speak.
> I said here are the conditions. If you can meet them great I’ll take my name off this paper, if not then I can work on a last date. Then she sent an email to my direct reports saying she has accepted my resignation. So that is google for you folks. You saw it happen right here.
In the email she posted, it seems that they "accelerated the timeline" for her termination because "certain aspects of the email you sent last night to non-management employees in the brain group reflect behavior that is inconsistent with the expectations of a Google manager", i.e. a separate email from her conditions email.
It seems like she was ready to leave if they didn't meet her conditions, and that this would have been a reasonable outcome for her. However, it bothers me that they wouldn't even come to the table about the non-condition email.
Like OK, "inconsistent with the expectations of a Google manager". That's some weak shit. It had better be something pretty fucking severe for them to fire her immediately and not, I don't know, do some planning around transitioning her projects? Work out what to do with the team she herself managed? Be given the grace of leaving on a good note that an engineer of her caliber deserves?
She's a leader of their ethical AI group. What happens to that group and all of their research and direction now? Who do they hire to replace her? This isn't the same level of impact as firing a cog in the wheel IC who gets too uppity with organizing.
This stinks like a convenient excuse to shoo an AI ethics researcher asking some inconvenient questions out the door.
https://twitter.com/mjg59/status/1334382463498391557
Neither of us are on her team at Google. (I'm certainly not and I doubt you are.) We have no idea how easy it is to work with her.
Certainly if I were working with an AI ethics researcher I would expect them to ask difficult and uncomfortable questions. I think that's part of the job description.
That said, Garrett makes clear his ideological sympathies on Twitter. Undoubtably comments like this one don't endear him to his management chain:
When @computerfemme was fired, the VP of security at Google Cloud sent out an email to the security organisation containing one absolutely false statement and one misleading statement, both of which portrayed her acts in a more negative light.
Pretty serious stuff. He doesn't elaborate on what those claims were. The woman in question was fired from the security team for violating IT security policies: in her words, "All I did was make a popup to share the labor notice Google has to share with its workers", something that Google seems to have recurrent problems with staff doing. In other words she was rewriting the contents of web pages to meet her own political objectives.
If I were the VP of Security at Google I would have serious concerns at that point about insider threats from sympathisers on the security team. You can't be publicly slamming your own VP as dishonest for firing someone who subverted IT policies if you actually work in security.
And yet, somehow, he’s not enough of an “insider threat” to be dismissed.
This paragraph that you just wrote is one of those claims that are wrong and/or misleading
The Google security team is frankly deeply worrying at this point, and I say that as someone who used to work there on security related topics (but consumer account security, not internal). A small number of them have privileged access to mandatory Chrome extensions that are used for various security objectives, yet they seem to have no qualms about abusing them to advance their own political agendas. Moreover they don't recognise that this is a problem.
Controlling Chrome extensions that can inject JS into arbitrary websites equals root at google.com because everything they do is admin'd via web UIs.
What if you're a law firm who advises companies faced with union action? Is it safe for you to use GSuite or whatever it's called today? My guess is no, because the people who ultimately control Google's IT infrastructure are (a) willing to break the most basic of IT security rules and then boast about it publicly, and (b) many of them are politically extreme by the standards of most of the world.
Most astoundingly of all, I don't believe this is the first time Google has fired people for injecting JS into websites using internal security mechanisms to advance political agendas. In any normal company there'd be a mile of controls, processes and guard dogs surrounding any mechanism that could do this by now as it's been abused before. Apparently Google Security do not have their act together on this.
Does not mean rewriting the content of webpages, or having "root at google.com". It's better if you'd stop such speculation.
The reality is much more pedestrian than what you describe: the person that was retaliated against only wrote simple, inert, strings of plain text.
That text was not injected anywhere, and there was no arbitrary code involved.
In fact, everything was safe, and went through the usual security controls.
The only "danger" in having that, is that people can read it.
Ultimately, it's laughable how you think that informing other coworkers about the NLRB is some kind of politically extreme agenda. It's clear that you have an axe to grind
Please do explain how someone makes a message appear in your browser when you visit a particular website, without having access to a highly privileged browser extension? That can at minimum monitor the URLs you're visiting?
I honestly really hope you aren't on the Google security team, because you seem to be missing the issue here. Someone who had a high level of access was fired for violating basic trust, apparently with the collaboration of her coworkers. What other things could this person have done to advance her political goals by subverting her access for things it wasn't meant for? Outsiders can't know. All they can see is that the company doesn't have a grip on its own workforce. That is not confidence inspiring.
Yes, that's crazy... And that's exactly why people have been angry at the retaliation
> Please do explain...
To get more details, the appropriate thing to do is to draft a post that explains how these things are accomplished at Google and get it approved for publication by the legal review team. I've never done it and, while I don't exclude that I might do it, it probably won't happen soon.
Hiring has apparently failed to meet goals with zero consequences suggesting a lack of concern with minorities.
I don't think someone misinterpreting disregard for ethical ai research as disregard for minorities is slanderous, heck it might be charitable as the later is known.
Be weary of "rockstar" scientists. Science is for the most part about incremental insights, and shouldn't make you famous.
Trying to get context, I see big names professors from Stanford, Caltech, Berkeley, Cornell all commenting/retweeting on how bad this is, so it's a big deal for sure
> I was fired by @JeffDean for my email to Brain women and Allies.
In a later tweet, says she would repost the email publicly but she no longer has access to it. I'm guessing it will leak soon.
Google Brain Women and Allies is a mailing list within the team.
I don't work at Google; I work at Amazon. I can only speak about my experience, but it's likely the same at Google.
You don't have to be a part of an affinity group to be a member of the mailing list (that's where the ..and Allies portion comes into play).
The 'Google Brain Women and Allies' mailing list is probably full of women from the Google Brain team, non-binary and men from in from the Google Brain team, and others who aren't on the Google Brain team but are interested in its work.
Again, I don't know. But this is educated conjecture based on my experience at another FAANG.
At the hazard of becoming 'political' and triggering people, affinity groups are typically created for groups that have traditionally been underserved, marginalized, or have special considerations.
As a cis male, I don't really see the need for a 'Google Brain Men and Allies' distro list because the issues that men face in the workplace are generally the same issues everyone faces. Women (and nonbinary people) can be confronted with issues that men simply aren't.
---
At Amazon, there are groups for Black employees, Latinx employees, LGBTQA+ employees, transgender employees, women, disabled employees, etc.
Sometimes these groups talk about work; but sometimes they talk about just life things. For example, often cisgendered employees will have children that come out as trans. So they'll reach out to the transgender community for help and support as to how to be the best advocate for their child.
Or, as a personal example, I moved to a new city to take the job at Amazon. I was looking for a therapist, and it was important that the therapist be supportive of LGBTQA+ issues and ethical non-monogamy. So I reached out to the LGBTQA+ group for therapist recommendations.
But when I hear about a 'Google Brain Women and Allies' distro, I'm more thinking about peers which discuss things which impact them. I wouldn't get too hung up about the 'Google Brain' portion of the distro; it's likely just a way to self-select a group of peers which share a common interest.
Gender equality activism has been part of US politics since at least 1776: http://www.thelizlibrary.org/suffrage/abigail.htm (Abigail Adams advocating to the constitutional convention)
Is there an affinity group for the overwhelming majority of "Latinx" people who don't use or like this term?
In my opinion, this leads to division and contempt/hatred for the other group. I live in a super progressive US state, and the university near me got a lot of backlash for hosting White affinity groups.
https://www.wired.com/story/prominent-ai-ethics-researcher-s...
Have to say that being asked to retract a paper because her manager didn't like what she had to say sounds pretty objectionable to me. Academic researchers thinking about joining Google beware.
2) She sent them right after an email blast that apparently rubbed execs the wrong way.
3) Then she threatened to leave and they called her bluff.
If I was a manager and an otherwise well-respected BUT irate employee who made unreasonable demands threatened to leave right after sending out a controversial email blast, I think I'd let them leave, as well. Negativity doesn't produce positive solutions.
> 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.
Source: https://www.platformer.news/p/the-withering-email-that-got-a...
I do not respect that kind of person and will not take a stance on this. I am disappointed that the twitter crowd doesn't care about the email, but not surprised.
Just trying to keep in mind that the smart reasonable people tend to be silent.
It's a bad sign from Google that this is their argument. If they were good on diversity/ethical ai they'd fight on that not on tone. If the facts are bad fight on tone, etc
This latest event gave them an opportunity, since, she essentially gave them an ultimatum. Perhaps after last year's legal success she felt she would have the upper-hand and didn't expect them to actually decline her conditions.
Honestly, the focus on the paper is coming more from her and her supporters trying to spin it (although I'm sure many truly believe it) to make it sound like she was fired for trying to uncover ethical flaws at Google. When in reality, it sounds like she was difficult to work with and burnt a lot of bridges
[1] https://www.reddit.com/r/MachineLearning/comments/k5ryva/d_e...
I doubt the wording used here will matter when it comes to legal brass tacks. Whatever contracts Google and Gebru actually sign will be authoritative there, not the public wording.
Right now the only way to disagree with unethical actions by your employer is leaving; easily replaced.
I'd rather they kept their powder dry to protect employees fired for things like union organisation, and protecting the workforce during large layoffs/reorgs, individuals discriminated against for their race/religion/sexuality, being compelled to take actions that are immoral/illegal etc. None of which seem to apply here.
(If they're not public then I imagine they'll be leaked eventually if this gets enough media attention. Are we witnessing the opening act of Google's next enormous Damore-style PR clusterfuck? Grab your popcorn.)
I personally have ever seen the 'do x or I quit' move work exactly once. The second time they tried it they took them up on the offer. They were even very nice to work with and were part of the 'do not fire this person or it is chaos' team. When he did it the first time he said 'I can only do this once if I try again they will fire me on the spot'. He did it a second time just to see if they would match his external offer (they let him ride the 2 weeks at least). You usually only get one shot at that sort of move. If you do it all the time, managers start looking for ways to offload you. You will be seen as a threat to org stability as you come off as unreliable. No matter how key you are.
Who ends that sentence with a smiley?
I respond to stressful and/or tragic situations with sardonic/dark/derisive humor. It's a coping mechanism for me. Some people cry. I crack jokes.
Just yesterday I said in a meeting with my boss 'No one wants to touch that project, because the customer gives out negative feedback like cheap halloween candy.'
If I were fired from my job in a way I felt was unfair, I'd probably do something similar.
This move makes it clear that Google has no commitment to Ethical AI as a field; they want rhetorical cover for their ongoing bad actions.
Such an inspiration!
That's the deeper question ML research/industry has to grapple with. When these models are deployed increasingly quickly and at scale in ways that can potentially cause massive harm, why is it okay to keep doing the same exact harmful things?
LeCun mentioned that there would have been opposite problem if it were trained on a dataset from Senegal. But why wasn't it trained on a dataset from Senegal? Why do we always see these errors where white-centric datasets produce white-centric results?
It's obvious that while it would be a symmetric situation a vacuum, we do not live in a vacuum. We live in a world with deep sociocultural biases in favor and against various racial groups. And it is unjust to let AI perpetuate and entrench these biases by acting at with this bias at scale.
Acknowledging dataset bias by itself doesn't address the bias meaningfully. In practice, we often treat the bias as an exogenous factor when it is not, moving it outside the scope of our responsibility. But it is very much the product of our work, a reflection of our choices, values, and beliefs about what to prioritize. We can't abdicate our responsibility for it (even if we choose not to prioritize it.)
That said, how should a person like Yann LeCun argue his case? Namely that biased models are the result of bad datasets more so than the result of bad algorithms.
I don't think Yann would disagree that biased datasets is a large systemic issue. How should a person like Yann make his point?
It doesn't seem to me that Yann and Timnit disagree all that much. They both agree biased datasets is a problem. They both agree it's a systemic problem. They both agree it does tremendous harm when these biased models are deployed.
I'm very confused as to why there even could be debate between these two people. I cannot find any meaningful difference in their views.
Probably because it was a research AI not a production AI. Having a very diverse dataset at that stage doesn’t help with your research, so it is fine to use whatever is easily available.
At that stage, you are trying to show that your approach can work in some cases. Once you've got that, it is time to expand the research with a wider range of inputs to find out what the limits of your approach are.
For example, if I were trying to make a US English speech to text transcription system, I might start with recordings of assorted NPR programs, because NRP often makes the recording available online along with they transcripts.
That would be great for determining if my basic approach has promise. Once I have determined that, so know that the whole endeavor is not just a waste of time, I could go looking for data that includes speech that has characteristics that would be missing from the NPR data, such as heavy regional accents.
Thanks the comment and this explication, you clarified the conflict that incident for me.
While I think the sibling comment is correct that LeCun and Gebru would agree on the proximate causes and mitigations of the adverse outcomes of that particular super-resolution model, the issue was (seemingly?) that focusing on the proximate causes can be read as absolving researchers of their responsibility for those outcomes, and avoids discussion of that responsibility in any instance.
Which is a entirely fair criticism of the field as a whole, though it may have been a bit lost in translation to the particular avatars in that instance.