The cult around ML here is pretty amusing. As another poster said, AI is currently staggeringly bad at this task, and I'm not sure it's going to be good enough in the near future either.
I think ML could be used to flag certain content as potentially problematic (I'm hoping to experiment with this in the future), but yeah, it's definitely not going to be doing the final verdicts anytime soon.
Machine learning is incredibly bad at this kind of stuff. There are far too many false positives and special cases to leave to ML. And the stakes are fairly high for Facebook: if they block content that shouldn't be blocked, it's newsworthy.