Facebook also has lots of business profiles (not pages) that aren't trying very hard to pass as people, as well as spam pages, forums and the like also using human profiles

With the rise of organized government trolling ops, this issue is becoming much more important.

You'd think that these leading lights would have figured out how to flag bad accounts more effectively than they do, and to train ML models to find similar accounts to those that are closed after manual reporting. But none of this seems to be the case.

Could it be that they simply lack the ability to do so? And does that tell us anything about the state of AI and Big Data in 2016?

*The classic meaning of "troll" not the generecized insult