re: #15 Dopamine Fish

Computers are designed and built to perform repetitive tasks at an extremely high speed. They are extremely efficient at this. However, when you break it down, the complex interactions of humans are not repetitive, they are constantly changing. Trying to manage hate speech with an algorithm is nothing more than a cheap cop-out to avoid paying the army of people they’d need to do the job properly. I’m sure that eventually, it’ll be possible to build AIs that are efficient at evolving and adapting to changes in patterns, but it’ll be a very long time before then, and even longer before they approach the ability of humans to do the job, if ever.

I think that AI is already there for 95%+ of the issues. If anything, the problem would be with false positives, which could then be checked by humans once someone flags it. A standard training set for AI sentiment detection is looking at the text of IMDb user movie reviews and determining whether the review was positive or negative. Detecting whether or not a post is racist or extremist or X is just a matter of putting together the appropriate training set.
My guess is that FB is weary of actually getting rid of racist and extremist content because it would cost them too much money. I would be shocked if they actually didn’t cripple their AI setup to not be too good at it.