Well, if you can't robo-ban then that's kind of the problem. As you say, identifying likely bad accounts isn't a problem, and they definitely can do that. But what do you do next?
Let's suppose that you identify just (on FB scale) 10 million likely fake accounts. There are only three options - apply further scrutiny, which would take hundreds of workers and cost lots of money, which isn't worth the (minimal) benefit to FB; robo-ban them which inevitably means false positives and would mean lots and lots of people contesting the ban, and real bot accounts are just as likely to contest than normal users, costing time in manual review; or do nothing unless the account is reported for something horrible, which seems to be the only cost-effective option.
Let's suppose that you identify just (on FB scale) 10 million likely fake accounts. There are only three options - apply further scrutiny, which would take hundreds of workers and cost lots of money, which isn't worth the (minimal) benefit to FB; robo-ban them which inevitably means false positives and would mean lots and lots of people contesting the ban, and real bot accounts are just as likely to contest than normal users, costing time in manual review; or do nothing unless the account is reported for something horrible, which seems to be the only cost-effective option.