A Step into the Wayback Machine: How Kasparov Could Have Beaten Deep(er) Blue
If you were a chessplayer in 1997, you were following the match between Garry Kasparov and Deep Blue (or rather, Deeper Blue, as the upgraded version was nicknamed at the time). The year before, Kasparov won by a 4-2 score after some early difficulties, but in the rematch things were much tighter. Kasparov won game 1, but famously lost game 2 he resigned in a drawn position. The computer had played excellently throughout, but at the end, oddly, it made tactical errors in a strategically winning position that could have let Kasparov escape. Had Kasparov faced a human he very likely would have sought and found his escape, but trusting Deep Blue's tactical prowess he resigned at the moment when salvation was available. This discovery didn't happen years later, but very soon after the game, and when Kasparov learned of this he was shocked and confused. How was this possible?
After draws in games 3 and 4 the score was knotted at 2-2, just as in the first match, when Kasparov won the last two games. This year, things didn't go as smoothly. Kasparov's meltdown in game 6 is well-known, when he chose a very dubious opening variation, played it badly, and resigned in disgust after just 19 moves. But this didn't occur in a vacuum. He was extremely upset about game 2 in particular, and about IBM's conduct, and a host of other things, including his failure to win in game 5. He was pressing very nicely, but the computer found an incredible draw that impressed everyone, from grandmasters on down. (I know I was impressed - Deep Blue's last-second counter-attack looked like a marvel of active defense.) Had Kasparov won game 5, he would have been in a better mood, could have played more safely, and may have had a better sense of the machine's limitations. Instead, he was at his wit's end and collapsed in the last round.
So a great save by Deep Blue in game 5, right? It turns out that this is not the case! It took longer to discover, but just as in game 2 Deep Blue was not tactically infallible, but made a slip. This was one of the things I discovered from Kasparov's new book Deep Thinking. Sure enough, I turned on the engine at the critical moment, and voila! he's right. And that, by the way, is very interesting: I have a decent computer running the latest engines, but they don't even calculate 30 million positions per second, let alone the (up to) 300 million positions per second Deep(er) Blue was capable of. And yet my engines identify the right move as the right move, the move that could and should have won for Kasparov, almost instantly, and recognize that it gives White a large-to-winning advantage in fewer than ten seconds. DB had minutes to find it, but couldn't. So hats off to today's programmers, who have not only greatly increased the computer's chess "wisdom", but even their tactical skill to a colossal degree.
Here is that game, with some of my comments. The critical moments are on moves 43 and 44. DB's 43...Nd2 was a big mistake, rather than the start of the miracle counter-attack, and Kasparov could have won with 44.Rg7+. It isn't a trivial variation, but it wouldn't have been impossible for an in-form Kasparov to find, either.
Errare computerum est etiam?