Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

Her research does cast Google in a bad light. She writes about how language models (like Google's BERT which influences somewhere between 10% and 100% of searches) are environmentally unfriendly and encode hegemonic language making them subtly racist and sexist. Her solutions are to use smaller language models and more curated data sets, which are probably not palatable for Google.

Dr. Gebru was told the paper shouldn't be published with Google's name and some of the feedback on it that she received had to do with the paper not mentioning how Google was combating or leading on the problems with language models that she called out. In other words, if her paper had made Google look good, they probably would've been okay with her publishing it. Given that her paper made Google look bad, they wanted her to change it or not publish it. When she got that feedback she sent her intemperate email (in which she included her ultimatum to resign) and was terminated shortly after.

I think it's debatable, given that she did offer an ultimatum, whether she was fired or resigned - but I know she prefers the former and I think that's at least a coherent way to describe what happened (Google accepted one side of her ultimatum by firing her).



>In other words, if her paper had made Google look good

A more charitable assumption might be that they wanted her paper to address feedback from the reviewers in lieu of being published as is. The feedback that she was presenting a biased interpretation by leaving unmentioned the existing work being done to combat bias in the models.


Firing someone who gave notice is still firing in California.[1] And Gebru didn't even give notice. She proposed discussing a resignation date.

[1] https://www.edd.ca.gov/uibdg/Voluntary_Quit_VQ_135.htm




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: