unmanned machines on the land or in the air, that can decide to kill a human, are one of the legitimate existential fears around AI advances right now. source: G7 white robe guy this week
I had a buddy in 2004 from the sister lab (doing strong AI), who was acquired by one of the big military contractors to implement "ATAK" (automatic target acquire kill). The task was mostly solved, as far as the US was concerned, as collateral (civilian) damage wasn't considered important for the deployment scenarios. (Think: total war.)
Not a very good analogy, we spent thousands of years creating dogs that are very unlikely to attack people. Just imagine the world in which all dogs turn into wolves overnight.
