Autonomous weapons are coming, and in some degree already exist. There are CV systems on machineguns in korea protecting the border. There's the TrackingPoint rifle scope that allows one to delay a shot until the rifle is line up perfectly, similar to how tank cannons account for the wobble down their length to ensure a straight shot).

It would be ridiculously easy right now for a person to attach a firearm to a quadrotor with a CV system, program it to identify human shaped objects, and fire at them. It's not a miracle that such a thing hasn't happened, there are very few people in this world with both the desire and skills to do these things.

I'm reminded of a short story I read a few years back [1]. It got me thinking about the possibility of a single depressed individual hitting the big red button and destroying all humanity. I think it's a real possibility, and when you think it through with the assumption that these super-weapons will be readily available much in the same way that AK-47s are ubiquitous now, you can start to wrap your head around ways to perhaps prevent this.

First, mental health is a huge issue. We're seeing the results of neglecting mental health in America now with these spree/ego shootings, where disturbed individuals either radicalize or "snap" and go kill people in malls, churches, theaters, schools, and other public places. Semi-automatic firearms and explosives are enabling these things to happen, but weapons are a Pandora's box which can never be closed once opened. The way that nuclear weapons restrictions are handled now is probably best. Restrict the ingredients to make the weapons, and you can prevent them from falling into the wrong hands.

Another thing that we as a global society can do to prevent these weapons from becoming widespread is to sabotage them. Include vulnerabilities and kill switches that would allow us to neutralize their effectiveness. China did this with the ASICs used in some military machines built by the US [2]. This is a valid option, but requires a motivation on the part of those creating the devices. One could argue that a nation-state could create devices like this lacking intentional back-doors.

Engineers can also work on counter-measures and disseminate them widely, and that seems like it will be the most likely outcome. Humans are clever, AI is very specialized, and that will make the big difference. Once a general AI better than humans is created, our desire to not be murdered by it won't really matter anyway.

I think our best hope is to build these AI weapons, but do it so shittily that they fail more than they work. I also like the American technique of building it better than everyone else and sabotaging the work of any potential rival so that they can develop them but not realize they don't work correctly until it's too late.

1: http://www.fullmoon.nu/articles/art.php?id=tal 2: http://www.scribd.com/doc/95282643/Backdoors-Embedded-in-DoD...