This film is merely bad science fiction, nothing more. Sure, we could build such bots, but the scenario is absurd. Would banning any kind of weapon control system prevent terrorists from using it? That itself is a basic error of this film. It says nothing useful, except for attacking a straw man argument about autonomous military drones, that autonomous AI agents would make far more effective, humane, or preferable weapons -- which I do not believe anyone has made in earnest. In other words, it is an attempt to associate AI with "unethical weapons". We know that weapons are unethical, already. Thus, saying they could be used for an evil purpose changes nothing. The film wishes to go one step further, making AI-equipped military drones a class of weapons that are analogous to nuclear, chemical or biological weapons, weapons that should be illegal under Geneva conventions or equivalent, and hoping to make a big fuss about it all. And further drawing a false dichotomy saying that anyone who does not agree with our scenario is defending these terrible, evil AI's that we imagined, as is typical of their world-saving charade.
Who cares zilch about that? Military, terrorists, criminals, the kind of people who do use such means. AI does not quite improve the destructive capabilities of such weapons. If you as a terrorist or a military commander wished to kill a massive amount of people, you would not care about collateral damage; you would launch a Hellfire missile from a drone like the US army does. Therefore, the US army is far more "terrorizing" using technologically more advanced drones, because they kill indiscriminately. This counter-example defeats the main purpose of the film. I know that this is Stuart Russell's favorite example about the supposedly unavoidable dangers of AI, but it is a ridiculous scenario that explains no new or useful ideas at the same time, wasting everyone's brain power. Some people without the requisite imagination may wish to stick with their field of expertise. Let James Cameron shoot science-fiction/horror scenarios involving robots, and Russell might want to stick with computer science.
The trouble is, this is merely a science fiction/horror scenario, it is by definition AI scare. Well done, Prof. Russell. Unfortunately, your artistic effort at fiction here is rather pointless and misguided. I can understand that you would like to demonize AI by suggesting that it could be used for weapon technology, but in case you have not noticed, AI is a general-purpose technology, it could be used in anything. And such autonomous weapons already exist; guided missiles, for instance are autonomous weapons, though it seems that you believe just because you called a face recognition circuit an "AI" it suddenly becomes evil. However, Cruise missiles are just perfectly moral, right? Because they are not tainted with "AI". They are used by good people, our heroic, God-fearing militaries. That is how sensible your scenario is, it is detached from reality, like everything else published by AI doomsayers as usual. It simply makes no sense, and the saddest part is that your dearest scaremongers will never realize or admit why it does not.
Your severe cognitive dissonance forces you to craft and then superstitiously live by straw man scenarios. That sort of superstition is of course essential to the pseudo-scientific AI doomsayer cults that zany conmen like Anders Sandberg rally around. Alas, I digress. Bad science fiction fails to inform anything about actual risks of AI technology. Perhaps, you would like to stop pretending that when we make robots autonomous, they suddenly become much more "evil" than their RC or slightly more primitive counterparts with a previous generation of control software. We already do know that military drones are dangerous. That is precisely why they are being built, to terrorize people, and to destroy lives and property. That is the very purpose of military, to kill, and to destroy; in case you have not noticed it, the military has seldom made ethical choices about human lives. It is the military technologists, military companies, military commanders, and militaries that are evil, not AI technology. There, the problem is, nobody believes that adding more AI routines to already highly dangerous and destructive drones is going to make them "right", but this is apparently your only chance at making it seem like adding AI routines to already highly dangerous military drones would make them a morally reprehensible class of weapons when they were not without AI.
I tried to clearly explain this so that everyone would understand the reductio ad absurdum and wrong assumptions. It seems to present nothing of relevance, because you would find it very hard to explain why you approve of any kind of military drones in the first place, in any intellectual debate which is not organized by your AI doomsayer cultists. You can say anything in a film, that's the beauty of fiction, but it is also no substitute for an intellectual thesis. You are wrong precisely because you have run out of intellectual options, and thus instead are seeking to create a speculative "Public Relations" moment using the idiocratic mechanism of Youtube, and paid publicists. I suppose the vernacular term is "beating a dead horse".