Giving the quote its most charitable reading, it sounds like they mean it's possible technologically to create a machine that: * identifies a potential target

Giving the quote its most charitable reading, it sounds like they mean it's possible technologically to create a machine that: * identifies a potential target * chooses to kill them * kills them And so the problem isn't whether those are possible but entirely on the ethical component of choosing who to kill (or building such a system at all in the first place.) And obviously a portion of that ethics is the adversarial component you mentioned which makes the closed loop choosing more difficult.