The always interesting Edward Tenner has some thoughts at The American on "Could Computers Get Too Smart?" He writes in response to James Barrat's Our Final Invention, which I reviewed recently, and the gist of Tenner's piece is that there are some reasons you shouldn't worry too much about the dangers posed by out-of-control AI. He makes the intriguing observation that the vulnerability of computers to hacking is a point against the grim scenarios Barrat and others have outlined. Now, one could take a darker view there, saying that hackers getting into advanced AI is the sort of thing we ought to worry about; but it's important to recognize that the news isn't all bad.
UPDATE: For more on the future of technology, see these responses to a recent paper by economist Robert Gordon, in which bloggers take Gordon to task for underestimating what automation is going to achieve in the next few decades. I find not particularly compelling Matthew Yglesias's deployment of exponential growth, as illustrated by the old Persian chessboard story, to argue for massive imminent advances. That chessboard was an abstraction on which you could put an incredible amount of wheat on a square. In the real world, even if you can keep doubling your wheat production as you pile it on one chess square after another, your towering pile of wheat will fall over. You'll also run out of chess squares. The real world has limits, and the interesting, unknown question in computing is what they are.
UPDATE 2: "How the rise of smart machines will affect the US economy and jobs: A Q&A with Erik Brynjolfsson and Andrew McAfee." I long assumed technology would always produce more jobs than it eliminated, which is one of many longtime ideological assumptions I've come to doubt in the last years. However, if as they suggest is plausible, technology drives the workforce toward more flexible and part-time work arrangements, my technophilia might get back on the upswing.