It wasn't that long ago that technologists considered the ability to play chess as a goal for computers that would indicate a dramatic accomplishment. That goal has since been passed and practically forgotten.
Moore's Law predicts an exponential growth in computer attributes such as speed, memory size, and compactness with the attribute doubling or halving every two years.
Assuming that Watson requires a room which is ten by ten by twenty feet, how many doublings of computer density would be required to fit it into a package of a six inch cube?
My rought calculation suggests this could happen in 35 years; a mere eye-blink in recorded history. The cost could be expected to be below $1000. I probably won't live to see it, but my daughters will. Life will be very different then.
Watson's achievement is impressive, all the more so in that it comes in 2011 and not some years hence.
A computer is now the world champion of Jeopardy, a FAR more difficult undertaking than chess, which less than 3 years ago was likened (elsewhere on the web) to a "canary in the coal mine." A an interesting quote from that article:
Let’s recall what the respected AI theorist Douglas Hofstadter said about chess in his famous book Godel, Escher, Bach, published in 1979:
Question: Will there be chess programs that can beat anyone?
Speculation: No. There may be programs which can beat anyone at chess, but they will not be exclusively chess players. They will be programs of general intelligence, and they will be just as temperamental as people.
Clearly Hofstadter was way off the mark here, in light of Deep Blue’s defeat of Garry Kasparov in 1997 and the situation today, where software such as Rybka, running on a desktop PC, can defeat the strongest players in the world.
It took only 18 years to make "respected AI theorist Douglas Hofstadter" look like an idiot. It took only 14 years to go from chess to Jeopardy.
Of course, being able to play Jeopardy is nothing like general intelligence. The same program, asked to chitchat about the weather, would draw an utter blank. But this is a huge milestone on the way to a generalized AI.
Throw in speech recognition, and you're getting close to the computer from Star Trek.
I agree with you that the hardware will rapidly shrink in size, but what I think will make the most difference is the software (and firmware).
We will learn to write programs to make better use of parallel processors, and reduce the need for so much hardware.