alt Hacker NewsRegarding your last paragraph (And as long as...) Bill Gates convincingly argues that the traits that make AI do good things are not easily separable from those that make it do bad things. So it may not be possible to explicitly avoid an AI that does bad things.
Gates is talking about dual-use capability (e.g. designing a medicine requiring same capabilities that allow it to design a pathogen). This is a real threat, but not a threat to the human role in the economy, which is one of being an independent strategic actor.