Monday 3 February 2003

Pic of the day: I guess it's kinda pathetic that a box like this can beat me hands down in a strategy game. (Look, it's laughing at me!) I still think I write a better journal than it could, though.

Artificial intelligence?

As I'm struggling against a patient computer program trying to teach me kindergarten Go, I find a certain solace in the latest issue of The Economist. Here the editors comment on the idea that a chess-playing computer would be a case of artificial intelligence, and that computers would eventually surpass and replace us. The fact that computers now are the world leaders in chess – albeit barely – has little to do with any advance in AI over the last decades, but depends on the plummeting prices of processing hardware. The computers are no closer than before to understanding what happens on the chess board. They are just insanely powerful and can compute all possible moves from any given situation onward to a valid checkmate. They are programmed by people who have assigned values to the various pieces and constellations on the board; the computers just follow the programming.

Indeed, computers have so far failed at playing Go on the large (19x19) board, though they do well enough on a kiddie board of 9x9. Given sufficient cheap hardware, I suppose even the supposedly divine mastery of Go will fall to computers eventually. (The animated series I'm watching these days is about a ghost returning to modern Japan in order to obtain the Hand of God, supposedly the ultimate goal of Go or Igo, as it is called over there. The series is full of cosmic references, a kid compares the laying of stones on the board with creating the universe.)

***

In reality, computers are struggling at most tasks that insects excel at. Not to mention humans. Precious few robots are able to walk yet, not to mention walk stairs or find their way to the shop and back. And while computers can make entertaining conversations, the entertainment is mainly in the obvious lack of understanding they display.

It is still not possible to get a good answer from Internet search engines if you ask a common question. The engines will simply extract web pages that contain the same words you used, or (in case of the highly advanced Google) web pages referenced using the same words. So for instance if I link to Al Schroeder's Nova Notes calling it an angsty teen journal, then it will eventually come up on your list when you search for angsty teen journals. Never mind that it is neither angsty nor written by a teen. The computer would understand this just by reading it, if it had artificial intelligence. But it doesn't even understand what I'm writing here. The more often I used the words "angsty", "teen" and "journal", the more likely that both Nova Notes and my own Chaos Node will show up in that category. Which is rather entertaining, I think, but not very useful. Except as an example.

Some people believe that it is all a question of processing capacity. Once the computer reaches the same processing power as the brain with its billions of neurons, the computer will overtake us in intelligence and understanding. I doubt this very much. It is not a question of processing power, but of design. Human brains and the operating system that runs on them, are designed for certain tasks. Some of these tasks form part of what we call intelligence or understanding. Our ability to recognize patterns, for instance. We are really good at it, while computers have to be programmed on a case by case basis.

***

So, my computer will probably be able to beat me at Go for the foreseeable future, unless I quit my job and my hobbies and dedicate myself to the noble game. But unlike the computer, I understand what Go is about. I can see the stones as military units deployed in a highly simplified battlefield, the crossing lines as roads that can be blocked or kept open depending on whose units control them. Once you see what it is about, it makes sense that Go only has two rules. It really need not have rules at all. It is this kind of understanding that I believe a computer will never gain ... and so, no matter how fast it can calculate, it will never truly obtain the Hand of God. For the goal of mastering Go is to master life. (Though I think I'll try to find another way to do that, personally.)