Carlsen was in a difficult position but found a defensive fortress idea that worked to hold the draw. Along the way, there was one moment where computers show that Caruana had a way to break the fortress and force a win. But the win is extremely counterintuitive and essentially impossible for a human to find and play.
Kasparov said on Twitter that he’d never have seen it in a real game, and would only have solved it as a chess problem by process of elimination, since there’s no obvious logic to the starting moves.
So, the game did actually reach a point where the computers were able to see a completely forced win? That is indeed interesting, and the fact that the best human players in the world couldn’t see it is yet another sign of just how much we meatbags have to learn about the game.
It’s a forced win (mate in 36) for Caruana after Carlsen played 68.Bc4. And playing through the computer’s moves (see here) you can see what’s happening - black dances around with his bishop until white eventually is forced to leave the defence of his h-pawn. But the start of the winning procedure might be literally the last maneuver human players would consider. Putting your knight on g1 with the opponent’s bishop completely trapping it from g4 is exactly the kind of thing you normally want to avoid.
I did use to wonder whether having access to computers would render the game meaningless but…definitely not.
Perversely, having the ability to instantly say definitively that a winning positions exists actually adds to the tension. Will either of them see it? How close can either of them come to the quasi-perfection of the computer? And of course, there is a face-to-face visceral and mental element to a human contest that silicon can’t replicate. A laptop is never of the verge of a mental breakdown so how much fun is there in that?
Wait, mate in thirty-six!? That’s how far out the computers were seeing? Even allowing for Moore’s Law, I would have expected that the best computers could completely plot out maybe 10 moves into the future.
It helps that the amount of material is so reduced. There are 7-piece tablebases available, and the position we’re discussing has 8 pieces, so the computer knows the evaluation of any position where a capture happens without having to calculate it. Also, if I understand correctly, the machine isn’t saying that 36 is the shortest forced mate, only that it’s possible to force mate at least that quickly.
OK, 8 pieces makes it reasonable that the computers would see it. But then again, it’s also few enough that I would have expected that the human ability to aggressively prune the move tree would also enable a human to see it. Except that, from what folks are saying, the winning move was one that would have been very quickly pruned.
I expect that the top human players of 50 years hence will be leaps and bounds beyond where they are now, due to all of the lessons we’ll learn (and more importantly, unlearn) from computers.
It seems to be a common view that if the match goes to speed chess, his chances decline. So you’d imagine he’d be interested in cranking up his attack a bit.
This is why I can’t even comprehend the skill of top-level players. On the Chess.com stream after that pawn to h3 move was made Hikaru Nakamura immediately called it a terrible move. But to an awful player like me it looked totally innocuous, and maybe even good.
From what I can tell the guys doing the commentary are pretty explicit about when they are looking at a computer. That clip I linked to (and almost all of the “real time” analysis on the feed from what I can tell) is just the commentators opining and calculating without computer assistance (other than the computer board they are using, of course - no engine assistance is what I mean).
You may be right, of course, but if they are using an engine to come up with their suggested lines they do a very good job of faking it.
