The demand for retraction was initially given while refusing to state what the reason(s) for the demand were. Only after the meeting where they demanded she retract the paper did they belatedly agree to read her the feedback that they stated as the reason. They wouldn't even let her read it first hand.
Ultimately it seems like Google could have plainly stated their issues with the paper from the start, and that said issues were not so serious that they merited a demand for retraction.
There are a number of things that just don't add up. Even if you take it at face value that Gebru's ultimatum demands were unreasonable, there are several problems with how the situation got to that point in the first place.
> Timnit co-authored a paper with four fellow Googlers as well as some external collaborators that needed to go through our review process (as is the case with all externally submitted papers). We’ve approved dozens of papers that Timnit and/or the other Googlers have authored and then published, but as you know, papers often require changes during the internal review process (or are even deemed unsuitable for submission). Unfortunately, this particular paper was only shared with a day’s notice before its deadline — we require two weeks for this sort of review — and then instead of awaiting reviewer feedback, it was approved for submission and submitted.
> A cross functional team then reviewed the paper as part of our regular process and the authors were informed that it didn’t meet our bar for publication and were given feedback about why. It ignored too much relevant research — for example, it talked about the environmental impact of large models, but disregarded subsequent research showing much greater efficiencies. Similarly, it raised concerns about bias in language models, but didn’t take into account recent research to mitigate these issues. We acknowledge that the authors were extremely disappointed with the decision that Megan and I ultimately made, especially as they’d already submitted the paper.
I'm curious: why is ignoring too much relevant research not a good reason to demand retraction? Isn't this one of the most common embarrassing mistakes that inexperienced researchers tend to make? I can fully understand if Google does not want to associate itself with work that might be viewed by the wider research community as shoddy.
For starters if the academic content of the paper were shoddy, that is what peer review is for. Secondly, if it was really about the paper lacking some new context, why did Google demand retraction while refusing to share why? That's what doesn't make sense.
It seems like there's some missing context in the situation because on the surface Google's actions don't line up with their stated reasoning.
Notice how in Jeff Dean's email he made no reference to the fact that they refused to share the reasons for the retraction with Gebru in that first meeting. He also skipped over where Gebru asked for the reasoning and a manager read it to her rather than letting her read for herself.
I just don't understand the secrecy and cageyness if the only problem with the paper was some academic omissions.
The timing of the submission for review, and the deadline for submission to publication seems like a simple factual statement that should be easily demonstrated as true or false.
Even with the efficiency benefits that it already has, we all know ML still stands to be further optimized, and a lot of people are working on it.
Comparing the exponential increase of "intelligence" of computers versus our own, computers will be as smart as humans with similar energy efficiency in around 50 years, and they require no cars, food, etc. It will be hard to criticize the efficiency of ML when it is already moreso than our biological selves, but it will be done, and we will be wholly outsmarted.
To be clear, reading that reason in a negative review can mean at least two things, and I'm not judging which of the two happened here:
1. The paper did fail to take into account the results in some highly relevant and widely cited papers. (I've rejected papers claiming new results myself before now with "this problem was in fact already solved by X, published in Y".)
2. The paper was fine, but it didn't cite every paper published recently on the topic ever, and someone used that to shoot it down for whatever BS reason. Matt Might calls these reviewers snipers: http://matt.might.net/articles/peer-fortress/