Advertisement

The Region

Programmer's 'ChessBrain' Is Learning All the Right Moves

July 22, 2002|STEVE CHAWKINS | TIMES STAFF WRITER

As Bugzilla and Venomous slug it out, an intense but soft-spoken computer programmer watches from his apartment in Newbury Park.

Carlos Justiniano has nine computers arrayed on a table in front of his four-poster bed. Sleeping just four hours a night, he pours himself into his fast-growing pet project--a globe-girdling network of computers that he hopes will one day play killer chess.

He says it will be the largest chess computer network ever. But more than that, it will be a resource for universities, a playground for researchers in artificial intelligence. Like IBM's storied Deep Blue, it will take on the world's best chess players; unlike Deep Blue, Justiniano says, it will learn from its mistakes.

The task is daunting. The total of possible moves in a typical chess game is the kind of glaze-inducing number ordinarily used to describe distances between solar systems--10 followed by 43 zeros, according to some mathematicians. In 1997, Deep Blue succeeded in humbling chess grandmaster Garry Kasparov, but, despite a number of efforts, no dedicated band of brainy volunteers has created a network that can rival it.

"It's just huge," said Justiniano, 37. "The only way to do it is to find the time."

During the day, he works as chief software architect at a Moorpark industrial controls firm. At night he dines with his wife, Karla, plays with 5-year-old daughter Kaitlyn and in the wee hours fires up his computers for work on his burgeoning network, ChessBrain (www.chessbrain.net).

Via the Internet, more than 300 enthusiasts in 25 countries have volunteered to help, each downloading a program that dedicates a tiny fraction of their computing power to the cause. While the "node operators" are sending e-mails to their friends or looking up recipes on the Internet, a small subconscious piece of their computers silently drifts to an Internet chess site, where it observes games in progress.

While Deep Blue was a single, blazing-fast computer, ChessBrain is an example of "distributed computing"--a system that draws power from a network of machines. It is being used in efforts as varied as solving math puzzles and scanning the universe for signs of life.

One recent day, the player who calls himself Bugzilla was wiping out the one who goes by Venomous. Justiniano's network focused on their game, analyzing moves smart and stupid, strategies grand and pathetic. Like minuscule bits of brain food, some 1 million positions are observed and ingested by ChessBrain in a typical day. Since the project started a few months ago, more than 39 million positions have been processed in a central computer.

If all goes well, ChessBrain will itself start to play later this year, taking on the likes of Bugzilla with the help of a sophisticated chess program. Eventually, Justiniano envisions researchers conducting experiments on the network and developing programs capable of instantly scrutinizing their past mistakes.

Meanwhile, a virtual storehouse will hold all the moves from the games ChessBrain will have played with both fallible humans and clever machines.

How well it plays isn't even the point, said Justiniano. As the hundreds of ChessBrain computers grow into thousands of ever-more-powerful machines, the play will improve. But more importantly, he said, so will the network's ability to mull over its losses and pull winning strategies from the wreckage.

"Deep Blue had no idea what it was doing," he said. "It had an immense volume of stored information. It was fed every game Kasparov ever played in tournaments. It had grandmasters and chess consultants saying: 'This is how you can beat Kasparov.' This is very different from a machine that learns as it plays."

Justiniano's goal is audacious, according to experts.

"It's a wonderful idea, but it poses some real technical challenges," said Andy Prince, a spokesman for United Devices, an Austin, Texas, company that harnesses distributed computing systems for complex problems in fields like finance and drug research.

Situations in chess are so complex that they may not be broken down into parts simple enough for hundreds or thousands of computers to handle efficiently, he said.

And machine learning is still a relatively new field.

"Deep Blue definitely didn't learn, but I'd be skeptical that his [network] will be learning anything but some trivial things," said Monty Newborn, a computer science professor at McGill University in Montreal who organized the first match between Kasparov and Deep Blue.

"People have tried to get computers to learn, but with only mediocre success," Newborn said.

Robert Hyatt, creator of a program that twice won world computer chess championships, agreed that getting a computer to learn the way even the dullest student does--much less "an electronic Bobby Fischer"--is a long way off. However, he said, a network like ChessBrain, with its potentially formidable computing power, could be a useful tool for researchers.

"The field is wide open," said Hyatt, who teaches computer science at the University of Alabama. "There could be a lot of simultaneous experiments on it that deal with machine learning."

That, Justiniano said, is exactly what he wants.

So far he has poured more than $30,000 into the project--money that could have been used toward a down payment on a family home. He said he doesn't intend to make a dime on ChessBrain, but hopes it will enhance his professional reputation.

"I could get a house and do what normal people do," he said, "or take an incredible risk. I'm shooting for something very large."

Advertisement
Los Angeles Times Articles
|
|
|