AI
Tuesday, November 18, 2003
  A few thoughts and wild-a**-guesses after watching game 4 of Kasparov v. Fritz --

- GK played for a draw again, although I can hardly blame him. The game was sharp and white was ahead on development. Not to state the obvious, since GK took longer on this move than all the others, but taking 13 pxd5 rather than nxd5, while it reduced his attacking options and made one of the commentators look bad, dulled the game and avoided the dreaded queen sacrifice. Might have been a draw either way, but GK would have sweated through one of his elbow-patched sports jackets if he accomplished it. Taking with the bishop on the next move created a "bad bishop," covering white square pawns with a white square bishop, but it developed the bishop regardless, consolidated his defense, connected the queen and rook, removed the bishop from the scope of the invading white queen, reinforced his center and generally made things a lot more tenable and calm for the flustered GK. Half the variations went out the window right there. When Qxf6 a few moves later, suddenly the defense is solid, white's forward queen, while still causing havoc, is now also a liability, and all the weaknesses that GK used to have are now limited to a nominally weak bishop (although it was protected by exhange) and a pawn that, while also nominally weak, was centrally located, passed, and had the force of all of black's heavy armor watching over it. And, while the Queen is at first blush both blocked and stuck guarding lesser pieces, the diagonal is about to open up, and the connected rooks provide almost immediate attacking possibilities, switching from defense to offense quickly when the black Q opens the lines with qxp. Don't care what Yasser Seirawan says, black did the smart thing, not least of all because it got white out of it's opening book by move 14 - still late, but not as late as it could have been. Take the queen sacrifice, and it goes into the twenties, with Kasparov in deep waters and having already shot his proverbial psychological wad while the computer is merrily making moves every 5 seconds on a meticulously chosen line. Of course, it wouldn't have been merry, but it would have seemed that way to the sweaty, hives-ridden Kasparov and the chortling commentators if it pursued an obviously prepared line that late. (Mistakenly in my opinion - who cares if it's still book? Book means tenable for both sides, or else everyone would play the line or it would disappear.) Speaking of which:

- Funny how White's final game starts with a QP opening, of all things, leading directly to an opening line which happens to exactly mimic a QGA line from a game with who? Kramnik. Which, the first time around resulted in what? GK sacrificing the queen, which would be ludicrously risky against the machine. Here's the way I see the Fritz team's strategy - and of course, this is all human strategy, has little to do with artificial intelligence, and demonstrates why we still haven't seen a true man v. machine test:

Let's play for the draw on black and win on white. Meaning, let's cram in full, broad, 25 move opening books on a few of the sharp lines Kasparov likes to play, that are suicidal against a computer, like, oh, QGA, special, narrow packages for when we have white. We only play white twice, so we can concentrate on only two. If we spend a little less time on the other opening lines or on fine tuning the evaluation engine, who cares? As long as the payoff is that we come up with two lines out of all the lines out there that we like and that we know he'll follow us into, we'll likely win (unless he finds a way to blunt the lines and slow the game back down - see 13 pxd5 above). We'll let the standard engine more or less truck away on the black side and hope for a draw, but on white we'll custom tailor aggressive lines from obscure history on sharp openings that lead him right into our strengths - like for instance a line from a speed chess games he played against the nominal world champ. No mindf'ing there. GK could have shat when b-b3. He could smell Kramnik's cologne.

- As an appendix to the first paragraph - not to pick on Yasser, but taking with the pawn on f7 and ruining the pawn structure in front of his king would have been a serious mistake against a computer. In my humble 1500-ish opinion, taking with the queen instead was the second most important move of the game, and Seirawan missed it totally. (He made up for it by spotting black's mating possibilities many moves ahead though, and broadcasting them, despite Maurice's discomfort at the lengthy discourse.)

- ESPN obviously decided Jeremy Schaap (the MC from the Deep Junior match) was completely unnecessary. They were right on. I heard they were in intense negotiations until the last minute to get Maurice Ashley, and he did fine. Schaap tried to be enthusiastic, but he brought a spirit of bemused befuddlement that any ESPN anchor would likely have brought to chess, and it didn't play particularly well. Maurice handled the point guard duties just fine, although it was a little silly to hear one grandmaster ask another, "Can you show me some of the lines that Garry is facing here?" when he clearly knew them at least as well. Still, it certainly was better to have someone who knew the issues at hand and took it seriously be the one to dish out the straight lines, even if he would rather have been the one showing off and delivering the punch lines.

- I'll take credit for foreseeing the frustration Kasparov would have with the interface in crunch time. What I didn't fully appreciate, however, was that he's also the greatest living blind-chess player. When he got tired of the interface, instead of having to put up with it and losing his cool, he basically eschewed that riduculously expensive board setup altogether, taking the glasses off (!) for all of his serious thinking. Next generation virtual reality, ha - they could have sent him the moves via teletype for all the difference it made to him. Kudos to him for realizing that he might do better blind (with the help of prodding questions from the press after game 3) and moving seamlessly into his own head with minimum distraction to his personal carbon-based electrochemical Fritz in game 4.

- A star is born! Joel Laurien is the next big chess glamour boy. The crowd ate him up. He also will have better luck against machines, with his stoic demeanor. And he likes Fischer and speaks english like a native. Joel Laurien's even a good chess name.

- Kasparov, while he is by far the biggest star and is the only man alive who would have made this event what it was, objectively speaking, is precisely the wrong person to be defending the honor of mankind against an impassive mechanical opponent. He relies on agression, putting pressure on an opponent, going for the kill, playing mind games, and generally impeding the thought processes of his opponent by distracting him with other concerns - what Garry is doing, how much pressure there is on the king, why am I sweating down my collar, it's only move 15, why are his rooks next to my queen, I can't believe I'm playing this guy, is everyone watching me, why am I always on the defensive, and, oh yeah, what if I lose in spectacular fashion? This is exactly what the computer does to people. Not only does the computer not react at all to his ploys, which gradually makes him panic, but it uses his own tricks on him. Some of his best weapons rely on the emotion of (in this case) an emotionless opponent,or the imperfect lines of thought of (in this case) a perfect thinker. The computer plays like Kasparov about as well as Kasparov does, and the irony is that he is particularly vulnerable to his own brand of chess. Witness the ridiculousness of him reacting so emotionally to a move made by a computer chip! He looked like he wanted to cry. If you want to cry, do it after the match. It's only going to distract you now. Put your demons back in their boxes, and look at the board! It affects his play, and especially his endurance, and it always has.

Basically, I believe the best person to play a computer is the best person to play Kasparov, and vice versa. That person is not GK. It very well may be VK.

- Human chessplaying against machine opponents in 5 years will be a matter of frantically pruning trees, narrowing possibilities and hoping against hope for a draw. Assuming, that is, that there are still people willing to pay millions to produce a chess engine, and ESPN is still interested. If not, it may take another few years before it's reduced to that level of absurd defensiveness in the face of superior firepower. You simply must innovate, or play for a draw, if you are going against an extension of all existing chess knowledge.  
AI

ARCHIVES
08/01/2003 - 09/01/2003 / 09/01/2003 - 10/01/2003 / 11/01/2003 - 12/01/2003 / 12/01/2003 - 01/01/2004 / 03/01/2004 - 04/01/2004 /


Powered by Blogger