Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Google fires another AI researcher who reportedly challenged findings

engadget.com

Post reply on HN
Print view View original

171–180 of 238 posts

Re: Google fires another AI researcher who reportedly challenged findings

#171

Post by oh_sigh »

lupire wrote: post #163
oh_sigh wrote: post #126

Earlier quoted context omitted.

I don't think you can quantify any of the claims that you made in your post. Is your argument that the "Computers can design chip components better than humans" paper is so fundamental to Google that they're silencing any critics? Conspiracy theories work better if there is even a kernel of truth to them. Maybe I could accept this if Chatterjee was rebutting advertising as a whole or something, but who exactly is so…

If you write a paper, as happened here, claiming that your boss's greatest achievement is bogus, your employment might be in danger. Not because Sundar and the Board and large shareholders care, but because your boss cares.

Your boss can't just fire you at Google(unless you report directly to Sundar I suppose). Especially not "with cause". Whose greatest achievement was "Computers can design some chip components better than humans" that you are accusing of having fired Chatterjee?

Re: Google fires another AI researcher who reportedly challenged findings

#172

Post by chaps »

yodon wrote: post #167
chaps wrote: post #161

Earlier quoted context omitted.

The fact that you've heard someone in the media say these things doesn't make the aspersions true. Good one. I'm a journalist who's done extensive research on this exact subject in Chicago and have spent hundreds of hours spent looking through leaked documents, FOIA request documents, network graphs, etc etc. I'm not going to write a long post about the subject for you, but I can point you in the right direction if y…

I'm not seeing the smoking gun here in your documents. I'm aware of many academics who work with city governments, including at least one researcher not from Chicago who works closely with the city of Chicago. He does so precisely because there are big problems there. The city of Chicago may even contribute to the funding of the work he does (it's helpful for the city to have skin in the game), but they certainly don…

The point I'm hoping to show isn't a "smoking gun" (frankly, I doubt any single document, or set of documents, would meet your criteria), but to show that research is deeply motivated by funding rather than for "pure science". Bad results that don't fit the narrative of the funders or political body's interest either don't get published, or are downplayed to the extent that they're effectively unpublished. Anything remotely positive is exploded through the roof.

I recommend checking out Robert Vargas on twitter. He's a tenured professor at University of Chicago who's written a lot about this subject. https://twitter.com/robvargas21

This post describes a lot of the problems well: https://crimethoughts.com/year/04/24/readi-chicago-and-incon...

Re: Google fires another AI researcher who reportedly challenged findings

#173

Post by t_mann »

wth? a "for cause" termination is a pretty big deal (in some larger organizations practically an automatic application killer), and to say it in a press statement is furthering its damaging impact. why do they think they need to do that?

even if we take for granted that there's probably a lot more to this story than what's reported in the article, this leaves a bitter aftertaste. there are so many ways this could have been handled more gracefully. and the optics of doing that in the context of a critical study are poor, to say the least.

Re: Google fires another AI researcher who reportedly challenged findings

#174

Post by muh_gradle »

dekhn wrote: post #75
muh_gradle wrote: post #7

Earlier quoted context omitted.

Which papers are you referring to?

https://www.techrxiv.org/articles/preprint/The_Carbon_Footpr... cites Bender, E.M., et al., 2021, On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?. ACM Conference on Fairness, Accountability, and Transparency. In short, Jeff Dean (and others at Google) cited the finally published (not by Google) version of the Parrots paper. The citation [20] is in the context of the paper: "The opening quote in…

Ahh thank you for the context. I was familiar with the Timnit Gebru incident, but I haven't followed up on it since it happened.

Re: Google fires another AI researcher who reportedly challenged findings

#175

Post by heavyset_go »

petters wrote: post #135
throwawaygh wrote: post #77

Earlier quoted context omitted.

The AGI safety stuff is weird and silly. AFAICT that community is mostly filled with philosophy types anthropomorphizing ML models that they don't understand. It's a waste of time and money. But then, I feel the same way about like 80% of what NSF CISE funds and there's plenty of toxic types in academia. At least the AGI safety stuff is all private money. My whole reaction to that community is mostly "weird and not w…

"anthropomorphizing ML models" seems like the complete opposite of what AI safety people do. They often point out that just because you are intelligent, you don't have to seem human. I think they call it the "orthogonality thesis".

Anthropomorphism isn't just about directly comparing things to people, it's also about assigning qualities that are related to people, like motivations, goals, opinions, concepts like good and bad, etc, even if the specific qualities would be alien to most people.

In that sense, the AI safety crowd has been anthropomorphizing hypothetical AI for close to 20 years now.

Re: Google fires another AI researcher who reportedly challenged findings

#176

Post by yodon »

chaps wrote: post #172
yodon wrote: post #167

Earlier quoted context omitted.

I'm not seeing the smoking gun here in your documents. I'm aware of many academics who work with city governments, including at least one researcher not from Chicago who works closely with the city of Chicago. He does so precisely because there are big problems there. The city of Chicago may even contribute to the funding of the work he does (it's helpful for the city to have skin in the game), but they certainly don…

The point I'm hoping to show isn't a "smoking gun" (frankly, I doubt any single document, or set of documents, would meet your criteria), but to show that research is deeply motivated by funding rather than for "pure science". Bad results that don't fit the narrative of the funders or political body's interest either don't get published, or are downplayed to the extent that they're effectively unpublished. Anything r…

Reading your second set of links, it sounds like much more a case of "the media does a bad job of reporting on science" than "the research is corrupt." If you want to improve the quality of science reporting, that's probably better handled by criticizing the reporters than by criticizing the researchers (most researchers already work hard to keep their work out of the popular press, precisely because the popular press does such a bad job of reporting on research).

Among the first tweets in the twitter stream you provided is a University of Chicago researcher actively criticizing the actions of the Chicago PD. That doesn't seem to support your narrative that the U of C research is corrupt and in the pocket of city government. I certainly can believe research is being poorly and inaccurately reported in the press, but again I think if you want to affect change you'll be more effective if you focus your efforts on the reporting press who seem by your own description to be the ones at fault.

Re: Google fires another AI researcher who reportedly challenged findings

#177

Post by brian_cloutier »

heavyset_go wrote: post #175
petters wrote: post #135

Earlier quoted context omitted.

"anthropomorphizing ML models" seems like the complete opposite of what AI safety people do. They often point out that just because you are intelligent, you don't have to seem human. I think they call it the "orthogonality thesis".

Anthropomorphism isn't just about directly comparing things to people, it's also about assigning qualities that are related to people, like motivations, goals, opinions, concepts like good and bad, etc, even if the specific qualities would be alien to most people. In that sense, the AI safety crowd has been anthropomorphizing hypothetical AI for close to 20 years now.

Do you have an example of someone in the community making this mistake?

My understanding is that if we knew that all future AGIs would have human-like motivations, goals, opinions, and concepts of good and bad then that crowd would be much less concerned. Smarter humans are not the concern. Intelligences which are _not_ human-like are the explicit concern.

Re: Google fires another AI researcher who reportedly challenged findings

#178

Post by 21723 »

khazhoux wrote: post #131
21723 wrote: post #111

Earlier quoted context omitted.

That wasn't my intent, and his username appears to be a real name (although I don't know). Apologies if it seemed that was what I was trying to do. I wasn't. I assumed that he was a real-name account who wanted to be known by his real identity (an identity that was maliciously attacked by an ex-CEO). I wouldn't guess at the identity of an account obviously intended to be anonymous.

I think doxxing is not just uncovering an anonymous user's real name, but any form of tying someone's real-world presence into online discussion, when it's not directly related to the discussion. Here, you were referring to an unpleasant event in the other user's real life, which they did not bring up themselves. That crosses the boundary. That user may not want that connection established here. But I'm not the HN po…

Fair point. I thought I was being helpful but I won't do it again.

Re: Google fires another AI researcher who reportedly challenged findings

#179

Post by godelski »

Manuel_D wrote: post #170
godelski wrote: post #140

Earlier quoted context omitted.

Are we talking about the LeCun fight where they were discussing Pulse (the white Obama super-resolution paper)? Didn't Pulse update their paper and show that it was in fact a dataset bias? A lot of these datasets have bias and it is frequently hard to train an unbiased model on a biased dataset. It also isn't the goal for many, though of course there is work on this. Bias is everywhere in a ML system, from the datase…

The "white Obama" pictures didn't even seem biased. They took a brightly lit photo of Obama and the output face has the same skin tone. Compare Obama's skin tone in this photo: https://video-images.vice.com/test-uploads/articles/5ef216e9... :* With other photos. E.g: https://www.obama.org/wp-content/uploads/ar_obama_sq.jpg

Bias is an overloaded word. To be clear, there is bias (there's always bias). Every model always has bias, we just want to understand that to understand the limitations of it. But yes, I agree that the reconstruction is decently faithful. I'm sure an L2 loss isn't that high when comparing the images.

Part of what I was trying to say is that there is a clear human bias when evaluating the reconstruction. The problem here is our prior. We humans already know that the compressed image is Obama because he is a well known face. So we're judging based on that prior. So there is some irony when talking about model bias when you don't recognize the prior bias in evaluation. We should analyze the bias of the model (I mean there's problems with the reconstruction other than the race issue, like the teeth), but we have to also be careful when analyzing because we also have a bias (which we may or may not want to impart upon the model).

The thing that gets me about these bias people in ML is that it is easy to recognize and identify bias. It is much harder to solve them. There seems to be two camps. Those that just want to identify and those that work on resolving them. The former typically say they work in bias and the latter typically says they work on long tail data (or sometimes few/low shot learning, or even metalearning). Yeah, we should recognize biases and discuss them, but just yelling at people doesn't help develop the necessary mathematical frameworks to resolve the issues.

Re: Google fires another AI researcher who reportedly challenged findings

#180

Post by sjg007 »

spoonjim wrote: post #125

Why does Google even fund this AI safety research internally? These conflicts are guaranteed to happen.

Checks a box that outside regulators can look at. Otherwise outside regulators would likely try to investigate and regulate. Also Google's own research can be used to counteract AI ethics claims that are existential to Google. These are likely going to be Big Data type claims. Big ominous future threats like the GDPR for example.

We know the oil and gas industry produce copious amounts of CO2 causes global warming -- they knew because they funded the research. But publicly their position was "uncertainty".

Post reply on HN