"And I used to say, I'm trying to pave a path forward using myself as the pavement." -- @mmitchell_ai regarding her trailblazing. I am in awe of the work that she and @timnitGebru and their team carried out, despite the adverse environment (as detailed in this article). https://twitter.com/dinabass/status/1384915124083904515

Dean has "also taken a more pragmatic approach to the AI ethics group, telling them that when they raise issues, they should also offer solutions, rather than just focusing on benefits and harms": neatly sums up the problem when engineering mentality is overrepresented in power.

You can't know to look for solutions if the problems aren't well explored. If you tell those with the expertise to surface and characterise problems to shut up until they also have solutions, the problems will never get solved.

Here's @mmitchell_ai again: "“We were like, ‘let us help you, please let us help you.’”"

Because finding and solving these problems is _also_ in Google's long term interests.

Also: "during a performance review in spring 2020, Dean helped Gebru get a promotion. “He had one comment for improvement, and that's to help those interested in developing large language models, to work with them in a way that's consistent with our AI principles,” Gebru said."

I hadn't known that particular detail, but surely the research in our paper can only be understood as groundwork for exactly that. @timnitGebru and @mmitchell_ai got fired for doing their jobs, and given the nature of their jobs, that looks TERRIBLE for Google.

"“We have hundreds of people working on responsible AI, with 200+ publications in the last year alone,” a Google spokesman said. “This research is incredibly important and we’re continuing to expand our work in this area in keeping with our AI principles.”" -- empty words.

Also, I propose a game: try replacing "AI" with "pattern recognition" or "pattern recognition at scale" and see how things read:

https://twitter.com/emilymbender/status/1384953488145326080

What would "responsible pattern recognition" entail?

I also find it really troubling that Ed Lazowska is quoted with a paragraph-long character reference for Jeff Dean. The story here isn't about whether or not anyone is a good person. It's about what actions Dean and others have taken, as Google managers.

Any given individual's personal qualities are only relevant insofar as the story is about how the incentive structures (at Google, at other tech cos, at big companies) can lead "even good people" to make the kinds of harmful decisions that management at Google made.

But the Lazowska quote comes off as centering Dean as if he is somehow the victim in this story who needs defense against attacks on his reputation. Classic DARVO.

To finish this unwieldy thread: go read the whole article. @NicoAGrant @dinabass and @josheidelson did excellent reporting https://www.bloomberg.com/news/articles/2021-04-21/google-ethical-ai-group-s-turmoil-began-long-before-public-unraveling