> I find it quite extraordinary how many on here are dismissing that there is any risk at all.
The fear over AI is a displaced fear of unaccountable social structures with extinction-power that currently exist and we allow to continually exist. Without these structures AI is harmless to the species, even superintelligence.
Your (reasonable) counter-argument might be that somebody (like, say, my dumb self) accidentally mixes their computers just right and creates an intelligence that escapes into the wild. The plot of Ex Machina is a reasonable stand-in for such an event. I am also going to assume the intelligence would desire to kill all humans. Either the AI would have to find already existing extinction-power in society, or it would need to build it. In either case the argument is against building extinction-power in the first place.
My (admittedly cynical) take about this round of regulation is about several first-movers in AI to write legislation that is favorable to them and prevents any meaningful competition.
...
Ok, enough cynicism. Lets talk some solutions. Nuclear weapons are an instructive case of both handling (or not) of extinction-power and the international diplomacy the world can engage to manage such a power.
One example is the Outer Space Weapons Ban treaty - we can have a similar ban of AI in militaries. Politically one can reap benefits of deescalation and peaceful development, while logistically it prevents single-points-of-failure in a combat situation. Those points-of-failures sure are juicy targets for the opponent!
As a consequence of these bans and treaties, institutions arose that monitor and regulate trans-national nuclear programs. AI can likewise have similar institutions. The promotion and sharing of information would prevent any country from gaining an advantage, and the inspections would deter their military application.
This is only what I could come up with off the top of my head, but I hope it shows a window into the possibilities of meaningful political commitments towards AI.