this is a different question than i looked into, but the google algorithm threw it at me (after i blocked a million young turks network videos trying to tell me how trump was DESTROYED by natalie harp scandal, or whatnot. this is a better suggestion, google. thank you for improving under human direction.) and i think it's worth posting here. the issue i was looking into was the question of misalignment and i determined the potential threat comes from poor programming and insufficient testing. it was much like the nuclear threat in the cold war. of course nuclear war was a serious threat, but there were efficient systems in place (including the mutual use of game theory, which is something the americans abandoned under reagan) to minimize it. the biggest threat was always that somebody would make a mistake, make a bad decision, and then we're all dead. 

once again, the issue being exposed is that ai is a legitimate threat due to the potential of human error, rather than the machines evolving beyond us. the latter is not something grounded in science. the former requires some careful regulation to ensure that ai systems are rigorously tested and have sufficient error correction before they are put into use and the evidence i've seen suggests that the private sector is failing at that task (where is the profit in that?) and needs governments to do the regulatory work for them.