Cathy O’Neil:

That leaves hiring humans to filter everything that emerges from the firehose of meaningless updates, cat pictures and lies, possibly with an automatically generated list of ranked things to worry about (which, to be clear, is not AI, it’s just an automatically generated list of things to worry about). Yet there are major problems here, too. For instance:

If you use people, you’re admitting that you have a policy of censorship, and users don’t like that, especially when they get censored. (Of course you already had censorship-by-algorithm before, but it was nice — and very inexpensive! — to imagine that it wasn’t censorship, when it totally was.)
You have to create a censorship policy, but you’re a rich nerd boss and you have no idea how to do that, which doesn’t feel great.
It’s expensive to hire people to do censorship. And if you try to save money by hiring people in other countries who can’t possibly understand the context, you’re asking for trouble.

Cathy O’Neil wrote “Weapons of Math Destruction”.