-snip-
First of all, I said nothing about improvisation or being spontaneous. Spontaneity is stupid and easy, even in humans. I was talking about
understanding.
Watson's goal is to interpret questions, OK. Well let's look at its resources:
$3 million in hardware.
The assumption that each question is a Jeopardy question, which is a VERY specific context
Thousands upon thousands of questions from exactly that same context
Massive portions of the internet, including all of wikipedia, which effectively means it has the combined memory of a million humans
Multiple programmers working for years to get it to do this one thing correctly.
Watson is using brute force. Its very impressive, but that doesn't change that fact that its merely imitating something humans can do using many many times the resources humans use to do it.
Secondly, I wasn't saying that you had to reprogram the computer each time, I was saying that A even if it learns its going to learn in a way specifically dictated by its programming and B the computer doesn't actually understand what its doing, its just doing what its told. What I'm saying is that its not a sliding scale from calculator to human, its a sliding scale from calculator to much much much faster calculator, because that's fundamentally all a computer is doing. Performing simple actions ultra-fast. You can have those simple actions change the code that dictates which simple actions it does, but that is a fundamentally different beast than actually thinking.
Going to back to game AI, massive application of data is a fundamentally different question than taking the same amount of data, and using it
well. That's why Watson isn't a step forward, because winning a Trivia contest with all of wikipedia at your fingertips is liking plugging an aimbot into an oldstyle shooting gallery game. Its, if not cheating, winning using inhuman capabilities that have nothing to do with intelligence.
Look, what is one thing all good chess programs do? They weight the AI towards putting its pieces in specific places. Why do they do this? Because the AI does not understand the point of the game. It doesn't understand anything. It literally plays by going through every possible combination of move, every possible combination of enemy move, ect. (with some pruning of course) and weighting the outcome based on pre-set value judgements.
A computer cannot think. It cannot understand. It can only act, because its a tool. Its like a hammer, but a billion times more complex. Now, maybe you could take electronics and metal and build something that does think. Maybe you could even build a sci-fi sentient machine. But it would be a fundamentally different beast than modern hardware.
Finally, because it needs to be said an arbitrarily large number of time: computers are fucking stupid.