Firestorm over Google termination of AI research group employee

Registered users do not see ads, it's free and easy. Register Here

lizkat

Thread Starter
Elite Member
Site Donor
Top Poster Of Month
Posts
2,515
Reaction score
5,725
Whether it had to come to a resignation or firing, who knows, but this gig certainly sounds like there should have been (and should be) a better way to handle what happened. And a far better way of making it clear to the public what has happened. Especially since the matter involves critical research on ethics and bias in language, a field in which Google has strong representation, and not least because the researcher in question is a Black Ethiopian-American female who is considered a prominent scholar in that field of AI.

Anyway Dr.Timnit Gebru is now apparently an ex-Google employee and the manner in which her most recent research offering was handled is at issue, along with (yet again) more general concerns about Google's workplace culture and some perceived gaps in what they say versus how behave regarding diversity of not only gender and race but opinion and freedom to express it.



More than 1,200 Google employees and more than 1,500 academic researchers are speaking out in protest after a prominent Black scientist studying the ethics of artificial intelligence said she was fired by Google after the company attempted to suppress her research and she criticized its diversity efforts.

Timnit Gebru, who was the technical co-lead of Google’s Ethical AI team, wrote on Twitter on Wednesday that she had been fired after sending an email to an internal group for women and allies working in the company’s AI unit.

The email, which was first published by the tech newsletter Platformer, referenced a dispute over a research paper, but more broadly expressed frustration at Google’s diversity programs.
.

Gebru told Wired that she attempted to negotiate with Google, offering to remove her name from the paper in exchange for a full explanation of the company’s objections, as well as a discussion of a better process for handling such matters in the future. If the company declined, she would arrange to leave at a later date.

Google rejected her request, and – after Gebru sent the frustrated email about diversity efforts – the company emailed her team to say it had accepted her resignation. They also cut off her access to company email, Gebru said. Google has maintained that Gebru resigned.

Upshot: professional reaction in the AI field appears to lean towards Dr. Gebru and against a once practically revered software engineer and now leader of the AI research unit at Google:

Jeff Dean, head of Google’s AI unit, told staff in an email obtained by Platformer that the research paper “didn’t meet our bar for publication”. Dean, who is white and male, also addressed Gebru’s email to the women and allies group, writing: “I also feel badly that hundreds of you received an email just this week from Timnit telling you to stop work on critical DEI programs. Please don’t.”

On Friday, Dean released a longer statement about Gebru’s termination, which he continued to call a resignation. In it, he reiterated his argument that the paper did not include enough references to efforts Google has made to lessen the environmental impacts of computing or its attempts to mitigate bias in AI. Google declined to comment beyond Dean’s statement.

Surely this is not over... and the National Labor Relations Board will probably have another case on its hands, not the first notable one stemming from a fracas at Alphabet / Google.
 
Registered users do not see ads, it's free and easy. Register Here

lizkat

Thread Starter
Elite Member
Site Donor
Top Poster Of Month
Posts
2,515
Reaction score
5,725
Yeah see this thing is opening all kinds of Pandora's boxes now, a move that Google would now likely rather their own guy hadn't launched by trying to stifle an inconvenient research paper... regardless of whether he had any valid points in wanting to put the brakes on it or seek revisions.

The optics of this "review" and Timnit Gebru's dismissal (or an overbearing conversion of her qualified offer to resign into an immediate firing) are are only part of it and the rest of it will not be pretty either.

And what a weird coincidence that the NLRB happens to rule adversely to Google on a matter of illegal firings on the same day. The rest of Big Tech can't be too happy to see one of its behemoths owning itself in this totally klutzy manner either, even if they're happy it's not them (yet) burning their own house down. Google is basically screaming to Congress ask us how we're doing again because we prob'ly got a lot more data now.

 

Chew Toy McCoy

Elite Member
Site Donor
Posts
1,464
Reaction score
2,576
So many words and I still have no clue what any of this is about. Are we supposed to believe she got canned for not being a white male?

I'm not saying there aren't workplace issues in regards to gender and ethnicity but I also feel the "not a white male" card is being played a little fast and loose as of late. I had a friend who did the hiring for his company for years and for whatever reason was a big second chances guy, meaning hiring people who had a record. A lot of these second chance hires weren't white, just stating this fact because pretty much without fail when firing these employees for justifiable reasons they'd go straight to the labor board claiming racism. Um yeah, was it also racism when the same guy hired them in the first place?
 

Huntn

Site Champ
Site Donor
Posts
745
Reaction score
1,120
"fired by Google after the company attempted to suppress her research and she criticized its diversity efforts."
Well... I realize you maybe in jeopardy if you criticize your company publicly, but there are several ifs there. Was the criticism based on the research she was doing or was it a separate issue? Was it public or about to be made public? Was she working within the company to increase diversity, or did she bring up the issue and was told to stuff it, and then threatened to go public or where they asking her to remove conclusions from her research?
 

lizkat

Thread Starter
Elite Member
Site Donor
Top Poster Of Month
Posts
2,515
Reaction score
5,725
This was or is "supposed to be about" suppression via internal rejection of a research paper that among other things criticized Google's environmental approach (energy use) to training machines in how people write things so that machines can be taught to write more like people do.

The approach initially was basically to keep feeding the learning machines "the internet" --so of course including what passes for fact, fiction and of course social commentary with all their political and other sorts of biases.

So, involving petabytes on petabytes of data with associated massive use of computing power.

The other criticism of Google in the rejected research paper was related to Google's approach to machine instruction in not deciding up front how to address language biases by applying proper filters as part of the learning process. One may wish to produce a biased piece of writing but in use of AI, presumably one wishes the bot to know not just how but when it's doing that.

So let's call the language learning setup under review at Google a kind of mega-version of what went wrong when Amazon wrote some algorithms related to screening resumés of job applicants: the learning included learning how to express bias without necessarily knowing it as bias.


Google, in its initial public defense of its action in terminating Gebru, opined that the rejected research paper from her team was NG from the get go currently, because it had not been revised to address how Google (he asserted) had recently modified its all-the-data-you-can-eat approach to use less energy... leaving aside what data it was still eating and how machines might be taught to consider issues of biased language.

Problematic for Google: The manner of "review" of the paper and of Gebru's value to the corporation were not per usual protocols --not least their cutting out participation of her own manager-- and were apparently undertaken in negative reaction to the attempts of Gebru to negotiate new ways of handling both the paper and her ongoing presence in the AI research team.

Some of it appears to be the optics, some of it head-butting on protocols, and some of it about the inbuilt bias of Google as a subsidiary of Alphabet Inc to protect itself from external criticism, since one option that Gebru proposed was to let her exit and take the paper elsewhere for review and revision. Might have looked a little better if Google itself didn't look so bad in the paper they decided to squash.

Gebru's problem: There's an old saying about "if you strike the King, best kill him," right? Well that does operate metaphorically in cases where an employee decides to go over or around immediate management. It looks a little bit like Gebru is blowing a whistle only now, or rather began to do so, once she realized the paper was being squelched in an unusual way. But her dismissal, almost simultaneous to that discovery, muddies the waters and it's not clear what G's whistleblowing process entails.

Blowing a whistle irrespective of a safety net's existence is always an option, and there are times maybe nothing else seems morally acceptable... one does have to look in the mirror of a morning... but every action does have consequence and it's foolish to imagine they might not include dismissal in that sort of scenario.

On the other hand whistleblowing is usually afforded protection nowadays, particularly in large corporations. Questions are whether the whistleblowing process in that area of Google is formal and whether it actually protects against retaliation.

What Google was suppressing besides harm to its reputation was far more problematic considering the weight that company can throw around in the field of AI. Its approach to a machine learning more about "how to write like a human" includes petabytes of unfiltered how to write like a biased human, so eventually gaining ability to write unwittingly as a biased human.

How does it help to risk teaching an ignorant machine more ignorance (without labeling it ignorance) is another way to put one question that arises when one thinks about these issues. That seems to have been an underlying concern of Dr. Gebru, and the AI research community, including many employees inside Google, does support looking into that far more carefully.
 

lizkat

Thread Starter
Elite Member
Site Donor
Top Poster Of Month
Posts
2,515
Reaction score
5,725
Welp... Google's CEO has weighed in. He was going to have to say something at some point.

The cited piece contains complete text of the memo Pichai emailed to Google employees on Wednesday. He did not apologize to Gebru but to all employees about the effect of a mishandled set of events.


Couple excerpts from the Pichai memo

First - we need to assess the circumstances that led up to Dr. Gebru's departure, examining where we could have improved and led a more respectful process. We will begin a review of what happened to identify all the points where we can learn -- considering everything from de-escalation strategies to new processes we can put in place. Jeff and I have spoken and are fully committed to doing this. One of the best aspects of Google's engineering culture is our sincere desire to understand where things go wrong and how we can improve.

Second - we need to accept responsibility for the fact that a prominent Black, female leader with immense talent left Google unhappily. This loss has had a ripple effect through some of our least represented communities, who saw themselves and some of their experiences reflected in Dr. Gebru's. It was also keenly felt because Dr. Gebru is an expert in an important area of AI Ethics that we must continue to make progress on -- progress that depends on our ability to ask ourselves challenging questions.
 

Huntn

Site Champ
Site Donor
Posts
745
Reaction score
1,120
Welp... Google's CEO has weighed in. He was going to have to say something at some point.

The cited piece contains complete text of the memo Pichai emailed to Google employees on Wednesday. He did not apologize to Gebru but to all employees about the effect of a mishandled set of events.


Couple excerpts from the Pichai memo
A lot depends on did she quit or was fired. Google will try to put the fire out, but the bottom line is that words without action make little difference.
 

lizkat

Thread Starter
Elite Member
Site Donor
Top Poster Of Month
Posts
2,515
Reaction score
5,725
What are a few things that depend on whether she resigned or was terminated?

It's a mess no matter how that sorts out really. But Pichai is right in going after better ways to deal with situations with potential to blow up. That one needed higher level oversight than by a once revered software engineer who had more or less switched ladders and acquired midlevel management credentials [that's a difficult shift sometimes], and it sure needed bringing Gebru and her immediate manager back into the picture.

It's not wise to try to cut people out of the picture that you think will give you a hard time or make you look bad, because the hard time will erupt later anyway and travel upstairs and then you will definitely look bad where it counts. Better off taking it a few flights upstairs yourself. Worst case they think "what you can't just handle this?" and then when they get involved they bring in some other folks as well and in the end they're on their knees with "thank god he didn't just handle this."

Sure can always try to suppress something but how you do it matters and how this got done was the opposite of suppression in the end, it was a springboard to the internet. This thing blew up like there were no walls in or around the entirety of Google at all. That's almost comical since the whole point of what went on in a corner of the AI research group was just to wall off one paper they didn't want out there without at least some revision.

I'll say this though: not easy to be a white guy needing to reject a research paper (or, get it revised) that suggests bias is not being handled right in process or data either, when the lead researcher is an expert in bias, is female and not white, is calling out the company on a bunch of stuff including bias... and is correct... and so a company looks bad in the paper for good reason and the company's process and data treatment need to be fixed.

So it's not just "the paper" ... or the researcher(s). There really are issues affecting the larger company at stake. But to me seems unimaginable that it could work to cut out the research lead and her boss and start talking about what the hell to do with that paper. Haul them in and some higher level help to keep things civil and have it all on the table... inside a conference room. Keep bringing people in if the researchers are talking about being suppressed or oppressed. Bring in whoever they want to be included, plus whoever higher management figures is also needed. Legal counsel, HR... Sure the researchers are vested in their work but they're employees not upper management. It takes some skills to manage a conflict of interests where bias is the subject of the work and also a concern of the employee.

Bottom line: management can always win a battle today, but how they do it can cost more than was worth and becomes part of what the employees end up experiencing as a war. Management has other fish to fry this afternoon, maybe there's a bond issue being floated. They move on. A group of employees though will rehash the battle just because it was on their hill whether they perceive that they won or lost.

Gotta love reading between the lines in Pichai's "We will begin a review of what happened to identify all the points where we can learn -- considering everything from de-escalation strategies to new processes we can put in place. Jeff and I have spoken and are fully committed to doing this."

Yep I bet they are both committed to not having this or something like this go sideways again.
 

Huntn

Site Champ
Site Donor
Posts
745
Reaction score
1,120
What are a few things that depend on whether she resigned or was terminated?
The “a lot” I referenced boils down specifically to was she fired or quit? And what specifically was she terminated or quit over? Based on what I’ve read being terminated makes the company look worse because they took punitive action to silence adverse publicity. Although if the the injured empoyee quit and goes public, it could turn out to be as much of an Internal and public relations hit, if the critique leveled by the employee was valid, and the company expressed no interest in addressing it (unverified). None of these are verified by myself. How about:
  • an employee being pressured to alter research to protect the image of the company and saying I’m fed up and quit.
  • an employee introducing un-associated critiques directed at their employer into their research or contingent on their research.
  • being threatened with and subsequently fired if they don’t alter their research results when those results are accurate.
  • variations not mentioned.
Being terminated is the worst if what the employee alleges is true, and harmful to the company’s reputation. Consideration might be given if the company said keep this quiet but we will fix it. No clue if that happened.
 
Last edited:

lizkat

Thread Starter
Elite Member
Site Donor
Top Poster Of Month
Posts
2,515
Reaction score
5,725
variations not mentioned.


She was apparently trying to negotiate an exit with rights to the paper and publish elsewhere, or else get Google to change some of the ways they operate. She didn't want to stay without getting some of those changes made. They "in effect" anyway did sack her by saying they accepted her resignation [but not her terms]. She was offsite and so they just removed her remote access. There could be a legal question on that process, but their intent was clearly that she become an ex employee at that point.

Companies don't like being told how to run their business by employees in other than a forum they establish for inviting that input, and certainly not in some sort of "This has to happen or else I'm leaving" format.

They pretty much tend to hear just the "or else I'm leaving" part and if you're on site will probably escort you out with your personal stuff to be sent home later.
 

Huntn

Site Champ
Site Donor
Posts
745
Reaction score
1,120
She was apparently trying to negotiate an exit with rights to the paper and publish elsewhere, or else get Google to change some of the ways they operate. She didn't want to stay without getting some of those changes made. They "in effect" anyway did sack her by saying they accepted her resignation [but not her terms]. She was offsite and so they just removed her remote access. There could be a legal question on that process, but their intent was clearly that she become an ex employee at that point.

Companies don't like being told how to run their business by employees in other than a forum they establish for inviting that input, and certainly not in some sort of "This has to happen or else I'm leaving" format.

They pretty much tend to hear just the "or else I'm leaving" part and if you're on site will probably escort you out with your personal stuff to be sent home later.
If she made demands of the company or else I quit, imo this is not as bad as being forcibly terminated and yet the controversial issues may still arise to hurt the company.
 
Registered users do not see ads, it's free and easy. Register Here
Top Bottom