Karo, regarding speech recognition being one of the least accurate areas, what does it say that this is the one field in which Kurzweil is an acknowledged expert? It ties into Tetlock's work where he found this same effect, at least among so-called "hedgehogs", people who focus on one single style of cognitive analysis and prediction (as compared to bringing in a wide range of viewpoints and strategies to get a handle on complex data). I don't know if Kurzweil really matches the hedgehog personality but it does seem that he tries to get a lot of mileage out of exponential curve fitting at the expense of other tools.
This seems worth posting around now... As I've previously observed, futuristic visions are produced as entertainment, sold today and consumed today. A TV station interviewing an economic or diplomatic pundit doesn't bother to show what that pundit predicted three years ago and how the predictions turned out. Why would they? Futurism Isn't About Prediction.
But someone on the ImmInst forum actually went and compiled a list of Ray Kurzweil's predictions in 1999 for the years 2000-2009. We're not out of 2009 yet, but right now it's not looking good...
Now, just to be clear, I don't want you to look at all that and think, "Gee, the future goes more slowly than expected - technological progress must be naturally slow."
More like, "Where are you pulling all these burdensome details from, anyway?"
If you looked at all that and said, "Ha ha, how wrong; now I have my own amazing prediction for what the future will be like, it won't be like that," then you're really missing the whole "you have to work a whole lot harder to produce veridical beliefs about the future, and often the info you want is simply not obtainable" business.