That said, if Google is "investigating" her because she was trying to find evidence of discrimination or bad treatment of Dr. Gebru, that seems borderline criminal behavior. (Assuming her attempts at trying to find evidence did not involve exfiltrating secret information).
This seems like a contradictory criticism to make. The Less Wrong folk, in my perception, mostly dabble in thought experiments (such as Roko’s basilisk) that aren’t terribly relevant to real world AI work. Timnit’s work on identifying racial and gendered inaccuracies in facial recognition [1] seems much more actionable.
[1] https://www.technologyreview.com/2020/06/12/1003482/amazon-s...
https://news.mit.edu/2018/study-finds-gender-skin-type-bias-...
Miri is mostly focused on the AGI control problem and thinks that AGI is closer than others believe. If true, all other problems are pretty irrelevant.
There is real work in ML bias and fairness to be done too, but it seems overrun by toxic personalities and partisan politics.
That’s fair, I haven’t visited LW in a while. I did thoroughly enjoy HPMOR, but thought that most of Eliezer’s other work fell closer to speculation than practice.
As other commenters have mentioned, these AI algorithms are already being implemented in the real world with real consequences. Their associated ethical concerns therefore seem more urgent, and they can be acted upon.
[1] https://www.technologyreview.com/2020/12/04/1013294/google-a...
> “It is past time for researchers to prioritize energy efficiency and cost to reduce negative environmental impact and inequitable access to resources,”
What? She should read previous work on the topic first - quantization, sparsification, distillation, fine-tuning pre-trained models, etc. I couldn't find her actionable ideas because she doesn't have any original ones. It's a hard topic which requires a complete rethinking of both algorithm and hardware. And she doesn't recognize that Google is investing in both to do that - TPUs on the hardware side, and improved algorithms such as the Reformer - a variant of the Transformer, reducing complexity from O(n^2) to O(n).
Then regarding dataset bias - I couldn't find their actionable ideas. I mean, except telling people to be careful about how they select training data, but nothing about how to replace the model they criticize with an unbiased model that works just as well in the general case.
It's easy to point out problems while not providing any solutions and making yourself the critic of those who are doing the hard work. See how she treated Yann LeCun on Twitter - she basically told him to go educate himself on her papers and fuck off because she doesn't have time for debating him, that after denouncing his tweets.
Hopefully, if it was just normal conversations and after identities had been redacted, their access is restored. Still, analysing this information on their work machine itself and then sending out a summary seems like a better course of action than this.
——
Separately, from the drama I’ve seen around the recent AI ethics stuff, I’d bet money Google is on the right side of this and the activists are not.
Why is your username 'foss user' when you're showing you are for intellectual property rights by calling this person a crook?
Also, if you're into the LessWrong mindset and are susceptible to think that that which can be destroyed by the truth should be, you might find this tidbit interesting:
> Buried in the recent trillion parameter language model paper is how the dataset to train it was created. Any page that contained one of these words was excluded: https://github.com/LDNOOBW/List-of-Dirty-Naughty-Obscene-and... Two sample banned words: "twink" and "sex" https://twitter.com/willie_agnew/status/1350551463718621184