They do it to save their own ass. Since they publish it, they are responsible. Facebook should either review the all content they publish, or they should stop being a publisher.
> Do we honestly believe that between ML and a real person Facebook is validating and approving every comment on Facebook?
I never said that; here's my point: if, say, the New Yorker replaced their human editors with AI ones, then the outcome is the same and they are still publishers. Alternatively, say they made human-run personal curations---each subscriber gets an individually-tailored set of articles which are suggested---then they are obviously still publishers. So if they replace these human curators with AI, are they still publishers? Because that would be what Facebook is today.
And on your Gmail point, Google is not a publisher, it acts like a phone service. You send your email, the recipient receives it, unconditionally, without modification. Facebook actively shadow-bans, time-delays and puts fact-check warnings on posts. Since that show they disapprove of some posts, it kind of implies that they approve of all the others.