> News flash: if even Jeff Dean thinks you're an unbearable asshole, you got some issues.

Wait, does Jeff have a reputation for being an "asshole" or something?

No, the opposite. Jeff Dean is known for exceptional kindness and humility. He is the kind of person that wouldn't have a problem with a researcher standing up for ethics.

Therefore, the implication is that anybody Jeff Dean cannot bear is an unbearable asshole.

Oh, thought so :)

>Therefore, the implication is that anybody Jeff Dean cannot bear is an unbearable asshole.

And yet Internet / News Media seems to take it another way. I will have to wait for Google's PR machine to do their work.

Does Google have a particularly effective PR machine? I think people like Google because of things like search, Maps, Gmail, YouTube, Android, etc. - not because they are some PR wizard.

They dont :), but unless someone is willing to take the risk themselves and stand up to defend Google, what is reported in the press will then be simply known as "truth". And it is their PR's job to do something about it. Although arguably not doing something about it is also a variable option.

No, quite the opposite.

You’re conflating the current person who was fired and did the public callouts with the person who has a Doctorate from Stanford and was fired or resigned because Google wouldn’t let her publish a research paper criticizing large language models. Let’s instead just not trash people we don’t know, and don’t have any firsthand knowledge of their motivation.

To me the story is that it turns out when you fire a fairly accomplished ethicist in a way that people perceive as unethically, a bunch of ethicists they work with don’t take it well - and in this case it looks like they went pretty extreme on it.

> Funny way of phrasing "paper wasn't good enough yet to pass peer review."

Peer review happens after it's submitted, in a blinded review process, by other researchers who don't have a conflict of interest - i.e., people who are at other institutions. It does not happen internally by people hand-picked by management.

You can disagree about the merits if you want, but it's important that we're on the same page about the facts. The review process Jeff Dean got mad about was an internal review process (for IP etc.), not a peer review process. It also was approved in the internal review process, which Jeff Dean confirmed in his letter. He was unhappy that the reviewer chose to review it and approve it within a day (which the reviewer was allowed to) instead of sitting around for two weeks to give management a chance to stick their fingers into the process.

Gebru was a very bright individual who could have gotten a job in academia, but instead chose to work for Google. Part of that arrangement was Google got discretion on what she could print. They wanted her to change what was printed because they believed the paper was unfairly critical of technologies Google had a stake in.

She disagreed, threatened to resign, wrote an unprofessional email and then Google accepted her resignation.

Where's the scandal?

Very bright in some aspects, maybe. But also apparently insensitive, abusive, narcissistic, myopic, and apparently clueless about corporate realities, and either delusional or dishonest with extreme bias in many communications.

Not my idea of bright ..

Oh, that's not the scandal at all, and I don't know why people think it is.

The scandal is that Google, already one of the most powerful entities in the world, is building an AGI and doesn't even want the slightest veneer of accountability for it. Nobody who isn't on Gebru's old team has any realistic access to know what they're really building. All we know is that the AI keeps outsmarting humans.

And now they've fired the manager in charge of the "AI ethics" team.

What are they doing such that they felt obligated to hire an AI ethics team and then ensure that team could not criticize Google?

If you worked in this field you would know how ridiculous your claims are. Dr Gebru wasn’t trying to stave off the robot apocalypse, she was interrogating biases exhibited by some neural networks due to the data they are trained on and how they are trained, which is something people have been looking at since statistical learning and statistics has existed. And when her colleagues raised reasonable scientific objections to her claims she went on the warpath and started calling her colleagues a bunch of dumb racist and sexist dirt bags. And it is actually fairly well known what Google is working on because they publish most of their work, and anyone in the field can go out and pull up all their research and have a pretty reasonable idea of what they’re working on, there’s not much mystery there. And despite claims to the contrary, the “AI ethics team” isn’t some grand overarching team, they’re a small team in one small part of the Google hierarchy—the leads of the AI Ethics are/were essentially first line managers, which is hardly the grandiose mandate that everyone seems to be attaching to it.

People are attaching far more significance to this whole flap than is remotely reasonable to anyone who works in this area or has any experience in the corporate world.

If the AI ethics team is a small team subservient to the AI practitioners, and they don't have the ability to raise concerns about small and apparently well-known problems, how could they ever hope to raise concerns about big problems?

And whether I work in the field or not is irrelevant (though, as it happens, I was fighting with deep learning GPU drivers just yesterday). The whole problem is the idea that nobody has the right to direct the AI beyond the people who are running the AI, that if practitioners have been "looking at" a specific concern, then it's illegitimate for anyone else to ask questions.

The paper in question's big claim was that language generators could repeat phrases like "both genders".

There are actual ethical concerns here, you mention some bigger ones, why die on that hill?

> Peer review happens after it's submitted, in a blinded review process, by other researchers who don't have a conflict of interest - i.e., people who are at other institutions. It does not happen internally by people hand-picked by management.

This is absolutely untrue. Your confusing peer review of journal with peer review prior to submitting to the journal. It’s super common to have papers reviewed prior to submission anywhere, and that review is usually done by peers or even superiors.

She's also not publishing this as just herself, she works for Google and is publishing on their behalf. Of course they have veto power over what she publishes. You take issue with that; quit. Which, I believe she did (she didn't get fired AFAIK)

Yes, there exist review processes by peers, and yes, your institution can choose to exercise veto power. The scientific process of "peer review" is a very specific thing. (My code is not "peer reviewed" in the scientific sense simply because a teammate, a peer, reviewed my pull request.) So saying the paper was not good enough for peer review is misleading.

And yes, Google can veto any paper. That's not the issue. The issue is they chose to use their veto power to block a paper that did not sufficiently praise Google. That means no paper coming out of Google can be trusted, because it's not the output of researchers but also management. Does Spanner work? Maybe there were some negative examples that were deemed too embarrassing to Google and removed.

I don't know, maybe you already acted as if no paper coming out of Google could be trusted, and you were smarter than the rest of is in that regard. But a lot of people act as if their papers are trustworthy and not propaganda (e.g., https://blog.acolyer.org/2015/01/08/spanner-googles-globally... takes the Spanner paper as if it were a legitimate scientific paper). Want to tell them they need to stop?

My organization includes a peer review process before it is submitted to journals. It makes sense because how would an organization be able to do this without peers.

There are multiple types of peer review, not just from formal journals. Your statement of a narrowly scoped peer review being the only kind is incorrect.

I’m not talking about Google specifically, but I imagine any institution creating scientific knowledge will have multiple phases of peer review before the literature.

Separately your logic of “no paper coming out of Google can be trusted” is faulty. It’s not that it can’t be trusted at all, it just can’t be trusted for some things. Peer review outside Google should identify if Spanner has bugs or whatnot. I think it’s safe to say that work of Google should not be estimated to only exist based on what’s published.

So while Google published things should be true since they are reviewed outside of Google, we’ll never know the things not published.

Also this is sort of the point of science right, you shouldn’t blindly trust things based on institution, but need to review critically based on your own experience and expertise.