Google AI Research Manager Quits After Two Ousted from Group(bloomberg.com)
bloomberg.com
Google AI Research Manager Quits After Two Ousted from Group
https://www.bloomberg.com/news/articles/2021-04-06/google-ai-research-manager-samy-bengio-resigns-in-email-to-staff
13 comments
Sammy Bengio is a good researcher, but Google really needs to align its team around company's business. The AI "Wokeness" team was really an embarrassment for the company. As the "woke" people are leaving, my respect for Google gradually keeps growing back (although, admittedly it's a long road back to the actual respect). I hope Google gets back to engineering.
I don't know if they were really that "woke". Apparently the paper covered these issues:
> Gebru had co-authored a paper on the risks of very large language models, regarding their environmental and financial costs, inscrutability leading to unknown dangerous biases, inability of the models to understand the concepts underlying what they learn, and the potential for using them to deceive people.
Environmental/financial costs are pretty "woke" I guess (in that it's stupid to complain about them; they're just a fact). Maybe the deception issue is "woke" too because it's pretty much inevitable and therefore uninteresting.
But hidden biases and lack of real understanding are definitely legit problems.
In any case she was fired for not highlighting new research into solutions for biases, not for being too woke.
There's definitely more to the story than we've heard though. I generally believe the company doing the firing in these instances because they normally can't give their side of the story so everything you hear is biased towards the person that was fired. That and firing people is really hard in most civilised countries. You have to really screw up for them to resort to firing.
> Gebru had co-authored a paper on the risks of very large language models, regarding their environmental and financial costs, inscrutability leading to unknown dangerous biases, inability of the models to understand the concepts underlying what they learn, and the potential for using them to deceive people.
Environmental/financial costs are pretty "woke" I guess (in that it's stupid to complain about them; they're just a fact). Maybe the deception issue is "woke" too because it's pretty much inevitable and therefore uninteresting.
But hidden biases and lack of real understanding are definitely legit problems.
In any case she was fired for not highlighting new research into solutions for biases, not for being too woke.
There's definitely more to the story than we've heard though. I generally believe the company doing the firing in these instances because they normally can't give their side of the story so everything you hear is biased towards the person that was fired. That and firing people is really hard in most civilised countries. You have to really screw up for them to resort to firing.
Out of curiosity, what is an uncivilized country?
I have to imagine that it is now MUCH harder to reach diversity targets now than it was before—because it is so fraught and oppositional.
You can take a chance on someone and find out they need a lot more training, or you can take a chance on someone only to discover they are trying to crash the system.
You can take a chance on someone and find out they need a lot more training, or you can take a chance on someone only to discover they are trying to crash the system.
What would be wrong with just hiring the best qualified person for the job?
It doesn't always maximize long term benefits. For instance, let's say you tend to only hire men because they are (by a hypothetical measure) better. If you have a bro work culture, that may make it harder to recruit and retain talented women. It may also make it harder to be sensitive to the needs of women users. The same goes for racial and cultural diversity.
Can you elaborate on the "wokeness" part of the Google AI team?
It is believed that having a set of strong ethical guidelines and representation can often lead to a high functioning research team, compared to a traditional monoculture team of researchers and developers.
Happy to be proved wrong though.
It is believed that having a set of strong ethical guidelines and representation can often lead to a high functioning research team, compared to a traditional monoculture team of researchers and developers.
Happy to be proved wrong though.
Google Brain team is great and Google DeepMind London team is truly amazing. They both care about AI ethics very much and I am very interested to hear what they've got to say on the subject. It's the Timnit Gebru's "AI Ethics" team that is very problematic in my view. Getting into a public spat with Yann LeCunn, demonstrating that she has no idea about AI to begin with. Demonstrating totally politicized, ideological views.
Ethics in AI are already an ill-defined and gigantic field, and while I agree with your point on diversity, Google is a business that wants to create value. How do you create value with ethics? By positioning your product as being less biased than the next.
There is a lot (really a lot!) of discoveries that are needed to identify and "fix" biases in DL models. The current method actually used is to test the obvious. i.e. "does our model work as well on all skin colors" which is far from being a robust and systematic method.
That's what Google actually needs and there is still a lot of fundamental research to do before getting there. Having ethics researchers writing papers on BERT generating too much CO2 is not only something an accountant could tell you, it's also not actionable as they are already trying to use zero-carbon electricity. Having researchers focus on inequalities in the US is good but it's again not actionable without the fundamental methods that need to be developed.
TLDR: Google needs researchers that are trying to solve the hard problems of the DL algorithms, not that get side-tracked by problems that are a matter of public policy.
There is a lot (really a lot!) of discoveries that are needed to identify and "fix" biases in DL models. The current method actually used is to test the obvious. i.e. "does our model work as well on all skin colors" which is far from being a robust and systematic method.
That's what Google actually needs and there is still a lot of fundamental research to do before getting there. Having ethics researchers writing papers on BERT generating too much CO2 is not only something an accountant could tell you, it's also not actionable as they are already trying to use zero-carbon electricity. Having researchers focus on inequalities in the US is good but it's again not actionable without the fundamental methods that need to be developed.
TLDR: Google needs researchers that are trying to solve the hard problems of the DL algorithms, not that get side-tracked by problems that are a matter of public policy.
Unfortunately with the recently NLRB ruling about the work activists at Amazon, I think it’ll be much harder to oust woke people who distract the company with personal ideological battles and politics. Meanwhile these institutions become weaponized for their causes...
so you have more respect for google if they don't care about ethics of AI?
He didn't even mention the two who were "ousted," but Bloomberg feels compelled to align with its ongoing anti-AI narrative by using a deceptively-worded headline (sequencing trying to imply causation).
I would like to learn more of this, but this article lacked a lot of substance.