Commentary for white move 34:

MIKE VALVO: There's less to look at.

GK MOVE: 34 Nh4

YASSER SEIRAWAN: And Susan's move, Nh4, she correctly guessed. She correctly guessed Garry's move.

MIKE VALVO: Well, the reason is that concepts apply much more in the endgame than anywhere else, where you say, "Well, my king is over here but let's keep everything else the same, but put the king over here, well, it takes a lot of moves to get over there, maybe as many as 20 ply. Well, computers don't think 20 ply, usually. So that's why beyond the scope of its ability. So if you can get an endgame where concepts apply where it's just a question of reposition your pieces in the different parts of the board computers can't do that because they're just thinking positions ahead at a time. They can't follow that deeply. It's much more evident in the endgame than any other phase.

YASSER SEIRAWAN: So what you're saying is that for the human player he gets an endgame or she gets an ending --

MIKE VALVO: We have to say "she" now.

YASSER SEIRAWAN: We do. And you say to yourself "Okay, what I want to do is I want to man you've my -- maneuver my king up the board, activate my rook, and I've got to stop that, and you're looking at concepts that are so deep, the computer which is calculating move by move by move by move is actually an inferior --

MIKE VALVO: Right. And it's not deep in the sense of complicated, it's deep in the sense of involved. There's more moved involved. But it's not more complicated. There's not a lot of pieces on the board. For a human it's very easy to see. It's just a question of "Oh, let's leave everything the same but just shift one piece over or maybe two pieces but that's way beyond the ken of the computer.

YASSER SEIRAWAN: One of the things that caught us by surprise was when Garry had played the move c4 and b3, keeping the knight out, we thought that it was in Deep Blue's interests to trade queens. But we had anticipated the queen trade coming on the d3 square and not on the g6 square. The g6 square allowed Kasparov to double Deep Blue's pawns. Now, in Deep Blue's algorithm, we had anticipated just -- I think that this is move 29, Garry had just played the move c4. We've anticipated Qd3. Instead the computer shows this very surprising move, Qg6, allowing Garry to double pawns. Why would the computer do that?

MIKE VALVO: Well, certainly it sees it gets a tempo on the f2 pawn. That might have been part of the problem, and it didn't follow the whole thing through. It seems quite obvious to a human that Qd3 it's just naturally better and simpler and doesn't have any problems, and it's sort of contorted to make a move like queen g6.

Real-time text commentary is made possible by LiveNote, Inc. and Vincent Varallo Associates