Highlights This Issue

Greetings from the President

Dear Chess Enthusiast,

The Internet Chess Club is proud to release the first issue of The King's Messenger: the newsletter for online chess. This "e-zine" is edited by the staff of the ICC, and contains contributions by our members. It will be published four times per year via email and the web.

This newsletter was necessitated by the staggering growth of the ICC: Now over 25,000 games are played every day, and the population reaches over 800 on an average night. The ICC has become one of the more significant organizations in world chess.

This issue starts with an article by Michigan philosophy professor (and ICC admin) Tim McGrew. He gives his perspective on the most exciting chess event in years: Kasparov versus Deep Blue. He describes how a group of ICC members discovered the drawing lines that Kasparov missed in game 2 of his match with Deep Blue.

I met CB ("Crazy Bird") Hsu in the mid 80s, soon after he arrived at Carnegie Mellon University, where I teach computer science. He had soon built "Chip Test", his first chess machine. (I now have the historic Chip Test circuit board in my office.) Working with several others at Carnegie Mellon, CB went on to build Deep Thought. After getting his PhD, CB was hired by IBM to develop Deep Blue.

Some have criticized Deep Blue as being "mere brute force". Yes, the brute force is there, but that's not all. To effectively make use of the powerful hardware, CB, Murray Campbell, Joel Benjamin, and the rest of the Deep Blue team developed new statistical training algorithms, and new ways to incorporate chess knowledge into the search. This is demonstrated by the fact that PC programs don't make Deep Blue's moves, even after running for weeks. Kasparov found the computer's play incomprehensible -- its play simply did not fit into the framework he had developed by studying other chess programs. In short, Deep Blue is a breakthrough. It's the result of brilliant scientists working with dogged determination for years.

On the other hand, I'm skeptical of claims that this has widespread applications outside of chess. Consider, for example, that the techniques developed for chess don't even work for the game of go, which is another perfect-information two-player game. How then can these techniques be expected to solve problems with even less similarity in structure?

In any event, I'm confident that there will be more exciting Deep Blue matches, which will of course be covered live on the ICC.

I hope you enjoy the rest of the issue. I'll see you online.

Daniel "Darooha" Sleator

P.S. If you don't currently have an account on ICC, feel free to register for a free 7-day trial on the Internet Chess Club web site.

Internet Spectators Find A Draw for Kasparov

by Tim McGrew

Sunday, May 4th, well over two thousand people watched online as the human race took a drubbing. The Internet Chess Club put a lot of effort into the coverage of the widely-publicized match between Garry Kasparov and the IBM supercomputer Deep Blue, sending a representative to New York with a laptop and a modem so that chess players from all over the world could watch. They even brought on board a raft of internationally titled players to give live commentary on the games and kept a log of the comments for each game as it unfolded. Though I am not a GM or an IM, I teach chess online and have given some lectures here. As a perk, the ICC gave me a special _Ersatz_ title for the match so that people who wanted to hear "just the titled players" (filtering out the deafening cross-chatter from the assembled fans) would be able to hear my comments as well. 

Now I almost wished they hadn't. Kasparov, widely acknowledged to be the best chess player in the world, had made some bad decisions early in the second game and was being pushed back across the board as the machine outplayed him in astonishingly "human" fashion, building up positional pressure with iron control. Black's position seemed to grow more desperate with each passing move until finally both the Grandmasters watching and the microcomputers we had online to provide move-by-move analysis indicated that Black was lost.

 As we contemplated the position after White's 45th move, Matej Guid, the 18 year old Slovenian Junior Champion, suggested that Black should send his Queen deep into White's territory in a desperate bid for a perpetual check -- a maneuver that, if successful, would turn defeat into a draw. An IM commenting on the game immediately dismissed it, and Matej gave up the idea. But it seemed to me that the concept had value, so I followed up on it and questioned the IM's off-the-cuff dismissal: after Black's Queen penetrates and White snaps off an undefended Bishop, Black should, paradoxically, play another quiet move. {See the analysis below.}

Here are the highlights of the analysis generated by the discussion on the ICC. After 45...Qe3:

a) 46 Qxd6 Re8 47 Bf3 loses the Bf3 to stop the perpetual, and then after Black takes the Bf3 he still threatens a perpetual.

b) 46 Qxd6 Re8 47 Qc5 keeps the game alive, but 47... Qxe4 48 d6 Qb1+ looks tenable for black.

c) 46 Qd7+ Kg8 (46...Be7 47 Qe6+ eventually wins for White) 47 Qxd6 Rf8 is similar to 46 Qxd6 Re8.

 The central idea is not the sort of thing that a chess computer, even one that can evaluate 200 million positions per second, will come up with. It runs something like this: Given the pawn structure, the board is essentially divided into two independent parts. In one of those parts White is trying to checkmate Black, but at the price of cutting off his pieces from defense of his own King; in the other, Black is trying to snag a perpetual check. White has more force on hand but he can't put a Rook on a7 -- the ideal square from which to try to force checkmate --because Black's Queen looks back down the diagonal at that square. After 46...Re8, any series of moves by White that ends without a threat of immediate mate will give Black a free hand to go after White for a perpetual. In effect, this sorts out the horrendous multiplicity of variations into a set of equivalence classes. One doesn't have to worry exactly how one gets to a quiescent position; one merely needs to get to one or another of them. Then it will be Black's move, and he can pursue his perpetual essentially unhindered.

 The real situation is slightly more complicated since White has a few ways to avoid the complete bifurcation of the position (Qc5, or Ra1), but that was the essential insight that struck me as Garry was thinking over his 45th move. It is the sort of thing that one would have expected Garry himself to come up with -- he is, after all, vastly superior to the rest of us as a chess player and is well known for his incredible resourcefulness in tough positions. It is also a "human" approach to the position; rather than sorting through countless variations, most of which is utter rubbish; a human master tries to find ideas that lead through the labyrinth to the desired outcome.  

Instead of giving the Queen move a shot, Garry resigned. I was baffled and frustrated, and I vented these feelings on the ICC with a string of question marks and comments to the effect that I didn't see a win. At first, most of the strong players online were dismissive: one very strong Grandmaster went so far as to say that the whole draw idea was "nonsense." But I persisted in my heresy and managed to pique the interest of a few other players. Within half a minute of Kasparov's resignation we had begun an impromptu online analysis session on the idea. Sometime early in this process John Hartmann, a foresighted ICC member, realized that we were serious and began to log the analysis as it unfolded. 

Ironically, the first people to give the idea a fair hearing were for the most part not the famous Grandmasters but rather people who know something about the strengths and weaknesses of chess computers. Bruce Moreland, the programmer of Ferret (the world microcomputer chess champion), brought his machine to bear on the analysis of my idea. Though Ferret was initially giving evaluations favorable to White, Bruce knew enough to find this unconvincing. "I've seen too many positions," he said, "where the computer initially thought it was up by two pawns but on closer examination it turned out that the other side could give a perpetual." Bruce waded into the analysis session and soon, instead of two-pawn advantages for Deep Blue, Ferret was finding perpetual check buried deep in line after line. Grandmaster Jonathan Tisdall may have been the first GM to be won over to the idea; he stuck with the analysis session for hours as other strong players came, scoffed, tried to find a win, and were unable to do so.  

I couldn't stay long: I had already spent too much time online and was late for dinner. But when I returned some hours later the analysis was still going strong and the entire mood had changed. Most of the strong players thought that a draw really was there for the taking, and skeptics were the outsiders. The entire session lasted some six hours. When it was over, John Hartmann edited the log he had been making for readability and fired it off to a couple of key people on the net --an analysis package that bounced across the Atlantic and then back to New York where, at about 2 a.m., it landed in the laps of Kasparov's astonished seconds.

 It is impossible to overestimate the impact of the Internet on chess. Just a few years ago I would have had to wait for the arrival of the New York Times just to get the moves of the game. Like the majority of chess fans, I can't afford a trip to New York City for the privilege of seeing the event live, and apart from the ICC there would be no question of my hearing live Grandmaster commentary. The Internet not only enabled us to have live coverage and instant analysis: it also made possible the transfer of our discovery around the world in just a few hours.

But there is another and potentially more significant way in which the Internet affected this event. Most chess clubs are fortunate if they have one or two masters showing up on a good night. A marathon analysis session like ours required a certain critical mass: three or four enthusiastic players would have been unlikely to have the staying power to keep going for six straight hours, verifying line after line until the outcome was beyond dispute. The ICC brought together nine Grandmasters, as many International Masters, some of the world's strongest microcomputer chess programs, and a host of other interested participants all focussed on the same problem, fresh minds taking over when others had to leave, computers working out the details when the position became too obscure for human intuition to be sure of itself. The collaborative potential when that many experts are brought into real-time contact is overwhelming. It may have been the first time in history that a group of armchair quarterbacks in any field outsmarted both the world's greatest living authority and the world's fastest computer -- and received confirmation within 12 hours from the experts themselves. 

IBM got a tremendous publicity boost out of the match because of its unexpected outcome, and this was not entirely unwarranted. The machine does a remarkable job of playing chess at a level that used to require human understanding. If one accepts the Turing test, then Deep Blue is, at least on the chessboard, intelligent