I really enjoyed TWIT's Triangulation podcast #152 entitled: AI - Our Final Invention. (http://twit.tv/show/triangulation/152)
Author James Barrat talked about whether emerging AI technologies may someday threaten humankind and what we could do about it.
I still have arguments with friends over whether computers will one day be 'smarter' than humans, but I think that anyone who has seriously studied the issue takes this for granted.
According to Barrat's sources, AI could match human intelligence by around 2030 and a Superintelligent AI could emerge by around 2040 that would be much more intelligent than us meatsacks.
And the pace could quicken as countries rush to be the first one to produce a superintelligent AI, much like the space race or the race to build the atom bomb. It could be built in America, China, Russia or many other places.
Just don't give it the ability to replicate.