Funny, I just watched a documentary on Wilson the other day. Hearing them explain how hard it is for a computer to understand the little nuances of speech was interesting. They said at one point humans who were on Jeopardy usually answered the questions correctly 90–95% of the time. Wilson was getting 10–15% correct. I guess it's something I would have figured a computer would kill a person at. But obviously there is much more involved.
I would love to work on projects like that. I think working on something cutting edge or something really hard would be very rewarding. You wouldn't have any constraints as long as you could find a way to get the computer to cooperate.
And something that I think about a lot is binary. How a computer only speaks in on or off and is able to do what it does is flat out unbelievable.
I was watching Dara O Brian's Science Club last night, and episode of the brain. Some clever scientist type people put a detector on the surface of a woman's brain and used it to detect neurons firing when she spoke/thought of speaking. They were very on/off signals like binary.
Actually it was someone who was paralysed imagining moving their hand that had the neuron detector. The other person saying ooohhh and aaahhhh had a different sensor to measure rain waves or something.
I think it's important not to make assumptions about what's actually being discovered in this kind of experiment.
It's a little like trying to work out what 'going to work' is and coming up with the observation that we see a shadow being moved from point a to point b every time. Interesting, true, and you don't get one without the other (dark days aside), but I could go off on all sorts of wasted investigations if I assumed that it was a useful observation in answering the question.