Google AI scientist Samy Bengio resigns
reuters.com
reuters.com
Any newspapers that don’t have this kind of gossip?
It's kind of incredible that at least 5 people at Reuters worked on this story, yet nobody stopped to think that this would be a confusing headline.
Or, if you want to be cynical, maybe they knew it would be confusing, but also that it would get more clicks...
And even if they did, why would expect a non trivial portion of readers to know?
And even if they didn't expect many readers to know who Yoshua Bengio is, he's still more well-known than Samy.
But the email excerpts quoted in article do not support that conclusion.
> Gebru has said the company wanted to suppress her criticism of its products and its efforts to increase workforce diversity.
> Bengio had defended the pair, who co-led a team of about a dozen researching ethical issues related to AI software.
From his email:
> "how difficult yet important it is to organize a large team of researchers so as to promote long term ambitious research, exploration, rigor, diversity and inclusion"
No need for sarcasm.
This article says that "He did not mention their firings in his farewell note.".
Keep in mind that there's more to the sentence you're quoting...
>"Though he did not mention the firings in his farewell note, they influenced his decision to resign, people familiar with the matter said, speaking on condition of anonymity."
No, that's a different sentence, that replaced the one I read. The article was edited after I wrote my comment.
When I wrote my comment, that sentence said "He did not mention their firings in his farewell note."
You can see this original sentence here: https://web.archive.org/web/20210406202729/https://www.reute...
Were they trying to avoid negative publicity? That seems to have worked out well.
Or were they afraid of being sued?
My org has a big focus on scientific integrity and doesn’t want to put out inaccurate information. So the upside is reputational to have high quality scientific output.
I’d guess they may also want some sort of review for IP or trade secrets to make sure info is appropriate for release.
Finally, I’ve worked places where there is comms and branding review where the org wants all the right product names and stuff to line up.
Sundar Pichai, for instance, I believe, was chosen not just because of his browser endeavours' success, his background etc, but because he could help break into the indian market (remember that these tech companies have been eager to get into and dominate the Next Billion Users market [India, China, et al]). It was actually a smart move, IMHO.
Mark Zuckerberg is going to be a tough one to get rid of because he completely controls Facebook. But the media have been doing everything they can to get at him (privacy issues, etc). I mean, it's not that the media aren't right (about some of the issues I'm aware of), but it's the best play they can do at the moment. But this Facebook, or rather, social media and media fight is a whole 'nother and different story despite sharing some variables with this topic.
Today the answer is about 70%. So yeah, we are trending away from 100%. And within certain groups of white men, this has created a lot of anxiety about losing their privileged position in society.
Some other evidence that the shift away from 100% white men is not some concerted agenda, but rather people who deserve it being put into positions of power.
Companies with female leadership tend to outperform the market.
https://www.consultancy.com.au/news/2362/companies-with-wome...
You're claiming that Sundar was chosen for his skin color with no evidence other than speculation. Chrome is one of the most important products for Google. Who else was passed up for his job? How are you so sure?
I don't see this concerted agenda to remove white men from positions of power. Today the answer is about 70%. So yeah, we are trending away from 100%. Companies with female leadership tend to outperform the market. You're claiming that Sundar was chosen for his skin color with no evidence other than speculation.In fact, this latter one not only for one reason but for _two_ reasons: because he belongs to a minority group (which from a PR perspective would look like "see? no more white men! Larry and Sergey are gone!") _and_ because, from a business perspective, it makes sense to pick him to help Google break into and establish itself in the indian market. It's actually a genius move.
Chrome is one of the most important products for Google. Who else was passed up for his job? How are you so sure?Ah, yes, and, finally, the constant barrage of negative news about this issue will most likely help Jeff leave.
The repeated assertion that something is obvious, combined with the repeated choice to not provide any evidence of it, makes it hard for me to take this argument seriously.
You mean: directly attributed to him and Megan Kacholia.
But it's not just me who figures that people will be less inclined to want to work there including this Google employee https://twitter.com/vickitardif/status/1334499064457650176 and this person who co-authored a paper with Timnit https://twitter.com/emilymbender/status/1370463074692472834
Edit: oh and that time their sponsorship of the FAccT was cancelled. https://venturebeat.com/2021/03/02/ai-ethics-research-confer...
Furthermore, Samy Bengio was Timnit's direct manager, but was completely blindsided by Timnit's firing, saying "I was stunned by what just happened to Timnit Gebru, who was in my team until a few days ago ... I stand by you, Timnit". Timnit was notified of her termination by Megan Kacholia, the Vice President of Google Research, who has since been removed from overseeing the remaining members of the Ethical AI team.
Regarding the publication which precipitated Timnit's ousting, "On the Dangers of Stochastic Parrots ...", Samy Bengio granted publication approval for the submission. It was only after this approval that the Google Ethics AI researchers were told to rescind their names from the paper (they could not withdraw the paper entirely, since it was a collaboration with external academic researchers).
Mitchell was fired shortly thereafter for allegedly violating the company's code of conduct and security policies.
Now, only a few months after Gebru and Mitchell's firings, Samy Bengio has also resigned.
Considering the sequence of events reviewed above, it seems clear to me that Samy Bengio's resignation is directly the result of Google's mishandling of this situation. If Timnit had not been fired while away on vacation, and Google leadership had instead found a way to address her concerns, or even accept her resignation more tactfully, then this debacle would have never unfolded in this manner, and Samy Bengio would still be at Google.
I am not particularly surprised by Google's mishandling of this situation though, as it seems completely predictable and even cliche that such corporate behemoths often handle these situations in the most ham-handed ways.
What I do find astonishing is the willingness of some Hacker News commenters to read into Timnit's actions and motivations, often with the most negative interpretations, but when it comes to Samy Bengio's resignation, they seem completely unable to read between the lines, and instead raise their burden of proof, so that they can deny any connection between this chain of events. Resignation letters and company wide farewell emails almost never disclose the real reasons behind a departure. Just because Samy Bengio has not explicitly said something like, "I am resigning because I stand in solidarity with Timnit Gebru", does not mean that these events are not linked, despite the assertions of some comments in this thread. In fact, it's seems to me that one would need to willfully ignore the evidence so far to conclude that these events are not linked, at least to some degree.
"He did not mention their firings in his farewell note."
It may very well be the case, that his leaving is related to his colleagues being fired, but there's no evidence of it in the article. So Reuters should not be strongly implying — to the extent that no reader would likely draw any other conclusion — that it is. I mean, it probably is, let's be honest, but I kind of expect better from Reuters.
The article mentions multiple links and things he said.
That's not quite correct, but true that the evidence presented is weak (e.g. anonymous "sources familiar")
I find that people tend to trust anonymous sources on issues that they agree with, and distrust on issues they don't.
And I mistrust the process of reporting anonymous sources like it, not Reuters per se. That is, I would mistrust this no matter who it is saying it because the process itself is untrustworthy and unverifiable. If I were writing such an article (and I have written quite a few news articles, mostly about scientific or legal matters), I would leave this information out entirely.
A more trustworthy way to do this is to name a source and give their basis for claiming to know that. E.G. Person X heard Bengio say XYZ or what have you. But if your source isn't willing to go on the record, I would leave it out to avoid reporting rumors.
>"Though he did not mention the firings in his farewell note, they influenced his decision to resign, people familiar with the matter said, speaking on condition of anonymity."
A farewell note is a formality for many people. I'd wager, given that this article discusses his previous statements on the matter, the firings did influence his decision. Having said that, it seems that he's departing as professionally as he possibly can, without blowing things up on the way out. Prospective employers won't have to be afraid of him making a scene even if he disagrees with them.
So assuming that it’s true is not a good idea since there’s no evidence that it’s true.
To me, that it is true seems much more likely than the reverse.
And "anonymous presidential aide says X" is always a political axe-grinding act. Those people don't just say things without an agenda. Historically, it's either someone testing a messaging line with prez approval or an act of factional warfare within the administration to try and shape the narrative.
So I am one of these people you are saying do not exist.
I suspect that actually the vast majority of people who know who Yoshua Bengio is at all are in this gray area with respect to X = yoshuabengio. That is, the people you are saying do not exist vastly outnumber the people who are confident that they are up to date on where Yoshua Bengio works.
I still haven't seen any improvement in AI ethics, even if the article tries to suggest it: we're getting much better models without any guarantees that they won't be misused to gain control of the world. Finding model bias is the easy part, as data is always biased.
This headlines are hairpulling the issue in every news.