Artificial intelligence researchers from around the world have come to Pittsburgh this week to play Chinese checkers.

They've also been playing -- or, actually, having computers play -- a version of the board game Othello called Nothello, where the goal is to lose, not gain, as many chips as possible. And they've had computers playing other games you've never played and, more importantly, that the computers have never played.

That was the whole point of the first General Game Playing Competition -- to design a computer program smart enough to play any previously unknown game when given only the rules. The contest at the American Association for Artificial Intelligence meeting concluded yesterday, with Jim Clune of UCLA beating David Kaiser of Florida International University in a game called "Racetrack Corridor" to win the $10,000 prize.

More than 1,000 scientists are attending this week's AI meeting, which concludes tomorrow at the Westin Convention Center.

Playing games, such as chess, has served as a test of a computer's ability to mimic the thinking of a human ever since the field of artificial intelligence was invented almost 50 years ago.

"The idea is if you're smart, you should be able to play games well," said Michael Genesereth, a Stanford University computer scientist. This seemingly culminated in 1997, when IBM's Deep Blue program beat world chess champion Garry Kasparov.

But single-game computer programs, such as Deep Blue, tend to give the computer a recipe, rather than thinking through a problem, Genesereth said.

In truth, even Deep Blue wasn't all that sophisticated, AI pioneer Marvin Minsky of the Massachusetts Institute of Technology said yesterday during his keynote address. The IBM program appeared to be largely the same as an earlier chess-playing program developed at Carnegie Mellon University, he said, and wasn't much different than a circa 1970 program by MIT's Richard Greenblatt.

"Moore's Law should get credit for that" victory, Minsky said, referring to the "law" that computing power doubles roughly every 18 months. Deep Blue was able to outplay Kasparov, he suggested, because of its brute computing strength.

The general game playing programs, by contrast, have no data banks full of moves and strategies, but have to play the games cold.

"It's the program that has to do the thinking," said Genesereth, who ran the competition. "It's like opening a box of Scrabble and saying, 'Here's the rule book. Play.' "

"Making legal moves is not that difficult," said Clune, a graduate student at UCLA whose game-playing program was in the inaugural competition. "But [making] good moves can be very difficult."

Clune, who began work on his program in September, said some of the procedures he wrote into his program worked better with some games than others; perfecting the program has meant replacing those procedures with new ones that apply more generally.

Seven teams competed in the event. Besides UCLA and Florida International, they represented the universities of Texas, North Carolina State, York (England) and Dresden (Germany) and one entrant from a private firm, Sensym.

Because this was a new contest, Genesereth said he expected most of the programs to be fairly simple, focused primarily on making legal moves. "I'm very impressed by what's happened this year," he said, noting that some programs managed to incorporate some strategies developed on the fly.

Even preliminary rounds of Chinese checkers generated occasionally raucous reactions, as the games were projected on large screens for spectators to watch. But many of the entrants emphasized it was the programming, not the games themselves, that they found most appealing.

"Go [the game] can be very challenging," Clune said. "But I'm not that interested in Go."