ROANOKE TIMES
Copyright (c) 1996, Roanoke Times
DATE: Tuesday, February 6, 1996 TAG: 9602060033
SECTION: EDITORIAL PAGE: A-5 EDITION: METRO
SOURCE: JIM GLANVILLE
RUSSIAN Gary Kasparov is still the chess champion of the world. But this month comes a dangerous threat to his tenure: a six-game match, at a convention of computer researchers and experts in Philadelphia, against the IBM computer called Deep Blue.
The event promises to be the long-awaited human-machine chess showdown. And with $500,000 on the line, the match is serious business. Computer specialists and chess insiders figure that, for the first time in history, a computer has at least a fighting chance against the human champion under conventional tournament conditions.
Since the early 1950s, chess programs have served to test the power of computers. Beating the human world champion remains the holy grail for computer designers and programmers. For this reason, the Philadelphia match is eagerly awaited by computer professionals. Human chess players, on the other hand, are nervous.
Humans and programs earn chess ratings by playing competitively. A rating is a single number that summarizes competitive performance. Beginners are typically rated 1,000 or less, average players around 1,500, experts around 2,000, and masters 2,200 to 2,400. Grandmasters typically rate above 2,500, and the world champion soars to 2,808. Deep Blue, some speculate, will perform at over 3,000. If so, Kasparov will surely have a fight on his hands.
In 1980, a computer with a 1,400 rating was a strong computer, though regarded with contempt by human chess experts. The intervening 15 years and the ongoing computer revolution have radically changed the picture. In 1996, a half-dozen programs are rated at 2,400 or higher when operating on fairly ordinary personal computers. For an investment of about $1,000 in a personal computer and $100 in a program, anyone can own a chess machine capable of outplaying all but the best few hundred chess players in the world.
Blitz chess is played with short time limits - five minutes per player for the entire game is the commonest. Under conventional tournament conditions, such as those of the world championship, a single game might last up to six hours. At blitz chess, computers dominate human opponents more than at longer time limits. That's because most chess games are decided by one or other player making a mistake. Humans become much more mistake-prone under time pressure.
In the past two or three years, innovations on the Internet have enabled chess players routinely to play chess with one another. A host computer programmed to enable distant opponents to play chess is called a "chess server." About a dozen chess servers are currently operational, and blitz chess is their most popular feature.
Games by chess-playing programs provide a significant part of the action on the chess servers. As recently as two years ago, only one or two programs had reached the "top 20" list. Now, more than half the top 20 "players" are programs, and the very highest rated "player" is a program. These days, at blitz chess on the 'Net, the average computer is better than the average grandmaster!
Kasparov has himself been bloodied by one of these computer programs. In I994, he came up against the Genius program playing on a Pentium-based machine. It's all on videotape - and painful to watch. The world's mightiest human agonizes as he gets squeezed to death and knocked out of the competition by a silicon monster.
But that, we say in human consolation, is not regulation chess. Under conventional tournament conditions, will not Kasparov surely uphold humankind's honor? Maybe, maybe not. Enter Deep Blue. According to IBM's press release it's a "massively parallel, special-purpose system with an overall processing speed capable of analyzing nearly one billion chess positions per second ... using custom accelerator processor chips running in parallel."
Why does IBM spend its money playing chess? "Computer chess," says IBM, "is largely dependent on fast searching and many fundamental algorithmic methods have been emerging from the work." Hmmm ... better living through better algorithmic methods. It's the 40-year old story: A computer that plays better chess is a better computer.
Deep Blue's predecessors, Deep Thought and Deep Thought II, played grandmaster-level chess but were obviously not world-championship caliber. Kasparov easily beat Deep Thought in a short match about five years ago. With Deep Blue, IBM sounds as though it is claiming to have replaced the low-orbital Atlas booster with the Apollo moon launcher. The general principle remains the same, but there's a lot bigger machine.
I pick Kasparov to win, but expect he'll have a tough battle. I predict Deep Blue will win at least one game. I'm not sure I'd pick any human other than Kasparov to win. Incredibly fierce, Kasparov seems to relish his role as defender of human intelligence against the machines.
How good is my prediction? Well, about 25 years ago I confidently asserted that chess is so difficult that I didn't expect computers to give humans any trouble during my Iifetime.
Jim Glanville, who teaches chemistry at Virginia Tech, has been playing chess almost exactly as long as have computers.
LENGTH: Medium: 91 linesby CNB