Volume 29 - Number 16 - Thursday, May 8, 1997

On the scene with Deep Blue II

by Sylvain Comeau
Computer science professor Monty Newborn is officiating at this year's Kasparov/Deep Blue rematch


Deep inside its awesome silicon brain, maybe Deep Blue has been itching for a grudge match. Last February, a creature of flesh and blood, namely world chess champion Garry Kasparov, defeated IBM's chess-playing supercomputer by a score of 4-2 in a six-game match.

The IBM programming team behind Deep Blue has since been preparing a new and improved contender for the rematch currently under way in New York City.

McGill professor of computer science Monty Newborn, organizer of last year's match, has been chosen as officiator of the rematch with responsibility for the umpires, the appeals board, and a team of three expert commentators.

In an interview before leaving for New York, he said that despite Kasparov's victory, Deep Blue's two points last year were a tremendous achievement.

"Very few people in the world can win one game and draw two in a six-game match against Kasparov. Maybe only the top 10 chess players in the world could do it.

"Most chess players thought (Deep Blue) would do much worse," he recalls, "and the chess world went into a state of shock when Kasparov lost the first game. It was a very tough match."

Kasparov rebounded to win games two, five and six, but Newborn points out that Deep Blue could have salvaged one of the losses if its handlers had allowed it.

"Kasparov offered a draw halfway through the fifth game, and that would have meant that both sides would have had two and a half points going into game six. You can't get closer than that." In chess, a draw or tie game is worth a half point to both opponents; a win is worth one point to the victor.

Deep Blue itself thought that it was losing the game, but the IBM team decided against it. "They were interested in doing well in the match, but they're also scientists and they wanted to see how well Deep Blue would do. The moral of the story is that the computer was already smarter (at chess) than anyone on its side of the match!"

Newborn said that both opponents will be even tougher and more competitive this year. He points out that top U.S. chess grandmaster Joel Benjamin is helping the Deep Blue team expand the computer's understanding of chess theory and its repertoire of openings, which often set the tone for the rest of a chess game.

In addition, Deep Blue is now twice as fast as last year, a crucial factor in a game in which both players must perform under strict time controls.

"Doubling its speed probably adds at least 60 rating points to its playing strength. Based on last year's match, the computer was rated at 2700, while Kasparov is rated around 2800." Ratings are assigned to players based on their tournament and match performances. A rating of 2600 and above is considered "super grandmaster" level, the very best in the world.

Newborn says Kasparov will be tougher, too, based on the world champion's recent results.

"He's playing the best chess of his life; he's won several very strong tournaments recently, and he'll learn from last year. He's at the top of his game, and many people have been saying for years that he's the best player in history."

Although Newborn is not on the Deep Blue design team, he feels that he has indirectly contributed to its chess mastery. He is a computer chess pioneer, chair of the computer chess committee for the Association for Computing Machinery (ACM), and creator of an early chess program called Ostrich, which came in second in the 1974 Computer Chess World Championship.

He has written several books about computer chess, including Kasparov versus Deep Blue: Computer Chess Comes of Age, published in January.

He has helped to lay down the fundamental principles of computer chess, upon which machines like Deep Blue are built. They work using mathematical calculations of the possible moves in the chess board's 64 squares and 32 chess pieces.

"In some ways, playing chess has been reduced to a bunch of multiplications, additions and subtractions," says Newborn. "Computers start from scratch with every move. They grow a 'tree' of all possible directions in which the game could go.

"From the current position, there are usually 30 moves, 30 replies, and 30 replies to the replies, and so on. The best programs today carry that process to 14 levels, seven moves for one opponent and seven for the other. Then they try to aim for a position very deep in the tree which is favourable to them, with the understanding that their opponent will try to prevent that outcome. In the process, they look at about 200 million positions a second."

Newborn compares it to looking at the income tax records of every person in the United States in one second, and then trying to come to some conclusion from that information.

Ironically, despite the impressive skill of Deep Blue and other strong chess programs, Newborn says that their actual understanding of chess is hardly up to grandmaster standards.

"Programmers over the years have worked at improving chess programs' knowledge of the game, but they still understand much less than a human being. Humans have a much more sophisticated understanding, but computers are so meticulous in their calculations that it's like playing an octopus. The 'octopus' has eight tentacles waiting to take one of your pieces at all times, so you constantly have to make sure that it isn't about to snatch one of them."

For chess players, that represents a fascinating challenge: understanding and human intuition versus the brute force of millions of positions calculated per second. Newborn himself is a former class A (near expert) level tournament player who has been humbled in a few informal games against Deep Blue.

But for computer scientists, the applications of the technology behind computer chess go beyond pawn grabbing and checkmates. For example, Newborn uses what he learned in computer chess to prove mathematics theorems.

Computer chess also addresses one of the most intriguing questions of the 20th century: Can machines think? Newborn says that computer chess was first undertaken to see if that question can be answered, but he feels the jury is still out.

"We've reached the point at which computers play grandmaster level chess, but does that mean they think or have intelligence? That depends on how we define intelligence."

Newborn suggests that, if machines think, they do it in a primitive way compared to the human brain. "In my book (about last year's match) I compared it to the first time that man tried to fly. The birds were not too impressed. Now that man does fly, in airplanes, they still aren't impressed, because planes can't fly with the same elegance as birds."

But this week, Newborn is more concerned with the thrill of competition than with philosophical musings. In an interview from his hotel in New York on Monday night, he said that the seesaw battle between the two opponents has everyone guessing as to the final outcome.

"Kasparov crushed the computer in the first game, so it looked like Deep Blue wouldn't do well at all. Then the computer crushed Kasparov in game 2, so it looks like it's going to be a tremendous match." Newborn noted that both victories were scored with the white pieces. White always plays first, which gives that player an edge.

The media has swarmed to the match. According to Newborn, 150 journalists cover the event daily, accompanying about 450 rapt chess fans in the audience.

"The games are very dramatic, and the audience is captivated. Nobody seems to leave their seats, even to go to the bathroom. They watch for six hours, just riveted by the struggle."

We will know on Sunday, when the final game in this battle of man against machine is played, what kind of brain can lay claim to world chess supremacy. At least until the next rematch.

URO Central

Front Page

Contact us

Back issues