I agree that a good player doesn’t play a dubious move hoping that their opponent won’t be able to find the best move. In fact, in searching for the best move one must look for the best moves on both sides, given the constraints of the clock. If one sees an inferior move that might trap the opponent if he fails to find the best move, but instead will trap you if he does, then the answer is clear: the player who’s about to move must always assume that his opponent will be able to find the best move.
However, when you are losing anyway, it might be worth a shot. If the opponent is in time-trouble and you’re reasoning that it would be hard to find it in a few seconds, you might go for it, if, say, you need a whole point to win a tournament, and a draw wouldn’t change anything.
Also, it depends on other factors. When Kasparov played the best combination of his life, by his own judgment, the one everyone knows about, the 15 move-ahead combination against Topalov, Kasparov simply plunged in. But if Topalov had instead turned down the sacrifice, Kasparov would have reached an unfavorable position. But Topalov’s curiosity was peaked and he went for it, which was not the best move, analysis showed.
In general, Kasparov always played the man, not the board, being keenly aware of the psychological factors. At one time he was playing exclusively 6. Bc4 in the Naidorf. But against Velimirovic, he chose a different move, not wanting to go toe to toe with a specialist of that line. Even though he was a specialist as well. But he knew his limits. It is this versatility and, of course, new insights into many openings that made him the best. He played a novelty once that Tal felt was especially concocted for him!
That was his lifelong strategy and his research and memory allowed him to be successful with it. But versus Deep Blue he was faced with a new problem. They couldn’t get any printouts, due to that clause that allows printouts only of the official games, which weren’t any. Then, like you said, they went on to assume that the new machine was similar to the first one, which proved to be wrong.
But still, there are differences between machines and humans that one has to take into account. In the opening, they have an advantage and trying to get it ‘out of the book’ as soon as possible might have been worthwhile. He was not afraid of trying that one.
In an open position, with lots of tactical possibilities, the machine is superior. So it is logical to try to veer the game into closed positions.
There are also other factors that affected his thinking process, because he was not facing a human player. For instance, against a human player he might have found a successful sacrifice which ended the game quickly. But against a strong computer, you can assume that if you found a successful sacrifice, then it is not successful at all, because a machine would not allow such strong play to happen.
And yet the same ‘stronger and improved version’ of Deep Blue commited a couple of mistakes in the end of the first game, where Kasparov resigned prematurely in what turned out to be a drawn position. Although years later, new analysis with stronger engines changed that evaluation again. That was a big blow, for a strong player to resign in a drawn position.
Going back to the Caro-Kann, it is true that he played a Caro before in their games, but it is also true that he never played the 4...Nd7 line. I said previously that he played 7...Bd6 against Karpov, but I meant to say that Karpov played that move against him. He only chose 5. Ng5 in the aforementioned line. So he never plays 4...Nd7 before and on the morning of the last game Illescas introduces this very specific sacrifice in case Kasparov who, once again, has never played anything other than the classical line, makes this obscure, premature pawn move...And with IBM hiring two spies...Do we see a connection between the spies and the last minute tweak?
He didn’t do anything out of the ordinary. He’d played the opponent before, at least when it comes to openings, but not only (as I have explained above), so now he correctly reasoned that on its own the machine would not sacrifice on e6 without concrete investments. Since he’d never played 4...Nd7 before, he also tried to surprise the machine (and the GMs behind it) with this ‘reversal of moves’ (pawn before bishop, instead of bishop before pawn), reasoning that the GMs would not think of this line which he never played and specifically of an inferior move like 7...h6? He was perfectly aware that after a capture on e6, Black would be in dire straights, not only against a computer of Deep Blue caliber, but also against any strong GM.
And he almost got away with it. I mean, on that morning Illescas to tweak the code is highly improbable because there would be no last-minute reason to alter a line he’s never played. Unless those two spies had turned out last-minute info about Kasparov’s plan in the opening.
Now, I can see this in hindsight. But at the time, from Kasparov’s vantage point, he placed all of his eggs in the basket of never having played that line before. Which, despite being logical, is not foolproof.
...maybe not.
btickler wrote:
As for repertoire, you should understand that in order for what you said to be true, the opening book would have to be changed, or, even more drastically, they would have added some direct code that calls for that knight sacrifice in that exact position, which would be unheard of. Every account I have ever read says that they did not do this, but rather just tweaked the valuations for aggressive knight sacrifices in this general type of position, rather as if they had changed valuations for a speculative bishop sac on h7 knowing that such sacs are beyond the engine's horizons. This was actually changing the way the engine plays chess and improving it, not adding a specific move/opening to it's repertoire.
You are wrong! Here’s what Miguel Illescas said in a 2009 interview:
On this same morning we also introduced the move Knight takes e6 in the Caro-Kann, on the same day that Kasparov played it. That very morning we told Deep Blue, if Garry plays h6, take on e6 and don’t check the database. Just play, don’t think...This was his bet, that the machine would never like this piece sacrifice for a pawn. And indeed, if we had given freedom to Deep Blue to choose, it would have never played it.’
In response, Kasparov reasons:
I’m no Nate Silver, but the odds of winning the lottery are quite attractive in comparison to those of the Deep Blue team entering a specific variation I had never played before in my life into the computer’s book on the very same day it appeared on the board in the final game. And not only preparing for the 4...Nd7 Caro-Kann—even during my brief dalliance with the Caro-Kann as a fifteen-year-old I played the 4...Bf5 line exclusively—but also forcing it to play 8. Nxe6 and doing this despite generally giving Deep Blue ‘a lot of freedom to play’, in Illescas’s own words.
And again, Deep Blue instantly took on e6, as an actual fact.
There were other factors as well. The machine was programmed to ‘self-terminate itself’ from time to time, if efficiency was running too low, which would make it impossible to understand future moves, as ‘the move timing changes, the hash tables change, who knows what else?’, said Shay Bushinsky, one of the creators of Deep Junior. In addition, the Deep Blue team would not provide the printouts, even when challenged, although it is not clear whether they were trying to hide something or simply toy with Kasparov’s mind.
Moreover, Illescas also admitted that IBM had hired two Russian-speaking spies, a fact that, when coupled with that mysterious specific change in Deep Blue’s opening repertoire, all of a sudden casts doubt over the whole thing. Let us not forget that IBM benefited financially after that game. And let us not forget what most people seem to forget: this was the second match, the rematch with Deep Blue, and so ‘humanity vs machines’ had already been decided in their first encounter, but somehow the publicity of the second match reached amazing levels which overshadowed the results of the first match.
Let’s also not forget that IBM mysteriously adamantly refused to give Kasparov a rematch, even though the score was even, and to make sure that would never happen—a potential loss in a third, deciding game could have brought on financial losses as well—they simply dismantled Deep Blue...Kasparov took for granted the idea of a rematch.
Regardless, it was Kasparov's team that shot themselves in the foot by playing those openings at all. The implication of your book reference is that the IBM team had no reason to change that particular variation, but Kasparov had just played the Caro Kann in his previous game as black. Obviously, they must have added a number of Caro Kann variations, and, if they tweaked the code for that particular knight sacrifice, so be it...this is nothing that a human opponent prepping would not also have done between games. No need for "spies" when your opponent telegraphs his openings choice. Deep Blue played each game as it had been programmed to, on its own.
If Carlsen's team told him that Caruana had missed a specific 4 move deep combination winning a piece in one of his opening lines 100% of the time, and Carlsen played into that line and Caruana won the piece, whose fault is it for losing the game? It's Carlsen and team for playing a knowingly inferior move based on an assumption that the opponent will "fall" for it. If you decide to play sub-optimal chess for an advantage you are assuming based on what you know of your opponent, you can't cry about it when your assumption turns out to be incorrect. If you played in a giant rock-paper-scissors tournament and beat 50 people on the way to final using a logical algorithm for selecting rock vs. paper vs. scissors, and then before the last round some schmoe tells you "your opponent always chooses paper first round" and you lost because of following this advice, it's your fault. Stick with the best play that got you to the match in the first place. In Kasparov's case, that translates to "play the chess that got you to world's best player". It's insecure and self-destructive to do otherwise, and that's what Kasparov learned when he imploded in this match: play the board, not the man. The most basic of chess platitudes that went forgotten because the opponent wasn't a man and some fear of the march of machines infected Kasparov's decision-making.
The mistakes lie here:
Game 3: "The third game was interesting because Kasparov chose to use an irregular opening, the Mieses Opening. He believed that by playing an esoteric opening, the computer would get out of its opening book and play the opening worse than it would have done using the book."
Game 6: "As Kasparov later recounts, he chose to play a dubious opening in an effort to put Deep Blue out of its comfort zone."
That's two games out of 6 where Kasparov removed his own winning chances rather than Deep Blue actually outplaying his best game. That's only the two most obvious examples, but numerous times Kasparov and team made references to how their prepared lines were going to exploit assumed weaknesses that Deep Blue V2 never had...a perfect example of "past performance may not indicate future earnings".
Kasparov and team fucked up, period. History shows us that he would have lost in another few years anyway to much weaker hardware, but he could have actually won this match, if he hadn't underestimated the advances Deep Blue was capable of and stopped trying to play Deep Blue "like a computer". If Carlsen walked into a match and played the Danish Gambit against Caruana and lost because of it, people would be all over him, but somehow the whole "human vs. machine" aspect provides "forgiveness" for Kasparov (a) assuming he should win/was entitled to win before even sitting down at the board, and (b) continuing to play "anti-computer" lines even after Deep Blue proved it was beyond the closed game rook-shuffling of engines up to that point, and (c) accusing IBM of necessarily cheating because he lost, as if it were inconceivable he could lose any other way.
They decided to get tricky, and they didn't have the leeway they thought they had given Kasparov's relatively easy win of the first match (after the opening shocker). That's their fault. They failed to understand, because, to them the Deep Blue they played against the second match was just an iteration of the first match...it was not. It was a new engine running on new super computer hardware designed to be efficient for board evaluations, and was dismantled because those processor boards were fucking expensive...IBM supercomputers at the time were tens of millions of dollars, not 10s of thousands like some super-TCEC rig might be.