Ignore this message.
Generally, when we say a game is “solved”, it means that there is a known strategy that one player can employ such that that player will win regardless of the other player’s actions. If both players know the strategy, then either one player will win based on turn order, or the game will be a draw. So, yes, the computer could lose, but only if his opponent plays the same strategy.
There’s an interesting - and troubling - corollary to the fact that chess computers now reliably beat humans: The time is now in view when human-human chess matches can’t be reliably fair.
This is because technology for undetectable communication is proceeding apace. All you need is an assistant who uses a good computer program to analyze your game and has a means of passing the recommended moves to you. The players at the board will become proxies for the machines behind the scene.
That’s not quite right. The opponent doesn’t have to play the “same strategy” or anything. There are various senses of a game being “solved”; the weakest would probably be that a computer can avoid losing if it gets to choose which player to be. For any two-player turn-based game of perfect information with win, lose, draw outcomes, such a program exists, though not necessarily efficiently. A step up would be that the computer can avoid losing no matter which player it is. This is where checkers is at (and for various simple games like Tic-Tac-Toe, it’s the best possible), but for some games, no such program can exist, since for some games, one of the players can force a win. Speaking of which, another notion of solved game would be that a computer can force a win if it gets to choose which player to be. This is the same as the weakest notion in a game that has no draws (like, say, Hex), but otherwise, it can be different. Clearly, this can’t be strengthened to the notion of a computer being able to force a win no matter what player it is, because of the possibility of making the computer play against itself; either one player can force a win or the other, but not both. Then there are other notions, such as “The computer can avoid losing if it steps in for any player in a situation in which loss can still be forcefully avoided”, and such things. The strongest consistent notion of solving a game would be a computer that played “perfectly”, in the sense that, no matter what position it is plopped down in, as either player, it will avoid a loss if it’s possible to forcefully avoid one, and it will force a win if it’s possible to do so. (I.e., it will reach the best outcome it can force).
Anyway, I don’t see any reasonable interpretation on which a game being “solved” means an opponent can do well if and only if he knows the strategy of the relevant computer program. I’d say more but I’ve got to hastily go.
Actually Turing wrote a famous paper in 1946 on machine intelligence where he used chess as an example of an application. (One of many Turing firsts.)
On chess vs. humans. Cars can out-race people. So what. People still watch humans racing. Checkers with humans has lost a lot of interest even before the computer solution. Just a glorified tic-tac-toe to many. They had to introduce initial move cards that players randomly drew in order to have some variety in games.
Though Turing did go on to do further significant work in computer chess, trying to track this particular 1946 paper down, the best I found was Turing’s report “Proposed Electronic Calculator” in which he only spoke of chess in passing in a paragraph, noting that perhaps a computer could be made to play good chess, but giving no significant indication as to how one might do so, and was notably reluctant to say much about machine intelligence. I might as well give the whole quote:
This is substantially less than what Shannon did in the 1950 paper, beginning to outline actual algorithms by which a computer could be made to play chess efficiently. Have I perhaps tracked down the wrong reference?
This is obviously incorrect; it’s easy to think of trivial examples where this does not hold. I don’t know who you’re talking about when you say “we”, but when we computer scientists say a game is “solved”, we mean that given a starting position, and assuming perfect play on both sides, we can prove whether the next player to move will win, lose, or draw.
On preview: What Indistinguishable said.
Well, let me add, now that I’m back, all the discussion I gave above was about senses of “solved” in which an efficient algorithm to “carry out the solution” is known. But I should have mentioned the even weaker sense, which is honestly the first one which always pops to my mind when I hear people talking about a game being “solved”, which psychonaut gives, in which a solution simply consists of knowledge of the final outcome of “perfect play” (either from various kinds of configurations, or, in the weakest possible sense, just from the start of the game), without necessarily having any knowledge of how to carry out perfect play for either side. Hex, for example, is solved in this sense: it is fairly simple to prove that the first player can force a win (it’s beautifully easy to prove the first player can force a non-loss; the tricky thing is showing that no Hex game ends in a draw), but for sufficiently large board sizes, efficient algorithms to force a first-player win aren’t known.
It may be helpful for those interested to refer to the Wikipedia article “Solved game” which helpfully reviews the various definitions of the term as used by game theorists. The article basically says what Indistinguishable and I have been trying to explain, except with helpful examples, references, and hyperlinks.
Most perfect move by move progression could be determined by taking the w/d/l percentages everytime each color did something other than the main line. If this is the best line, at least up to move 5, all other moves should make the player that veered from the path do a certain percentage worse.
The Najdorf is the most popular opening in GM play right now, that and the Petroff, which I completely agree is the great drawing weapon of choice. But I think there’s something to the Najdorf when they’re are “Najdorf Tournaments: 2000+ elo” going on. That’s not proof of it at all, but it should make you raise an eyebrow ( :dubious: )
Chess? Pfft! Let’s see the computer programmers wrap their heads around Go 
What is Chess 4.0? I can’t find mention of it in the article you linked to.
-FrL-
Chess 4.0 was a chess program from way back. I have a reference to it in an article about computer chess by David Levy from the 1970s. (He made a bet in about 1968 that there would not be a computer within ten years that could beat him under contest conditions. It was a very good bet, but today, he would lose his money.)
/mildly astonished that psychonaut would waste electrons pontificating in GQ about a game that he admits in the same breath he knows nothing about.
I don’t see psychonaut saying anything about chess in this thread. He’s only “pontificated” about concepts from game theory.
-FrL-
but maybe the pontification on chess is located in the same post in which he says he doesn’t know anything about chess, because I don’t see that either.
Sorry, here I understood him to be saying something about checkers, and then saying he didn’t know the first thing about the game. My reading comprehension skills must be not too sharp today, or something.
Well, the subject of this thread is about whether “chess” is a “solved game” in a game-theoretic sense. No one seems to be disputing the meaning of the word “chess”, but so far at least one poster has come up with quite an odd definition of “solved game”, which was corrected by another poster and myself. It’s not possible to address the question in the OP if we can’t agree on the terminology.
I am puzzled by the accuracy of this assertion.
Using Chessbase statistics, some offbeat openings, for example, have high % scores. However it turns out that **there are few games involved ** and the winner is a specialist in the opening.
Do you include wins from Open tournaments where one player is rated several hundred ELO points above their opponent? How much did the opening matter?
Similarly, are you counting an opening as good for a win if the opponent gets a better position early on and then blunders?
Also openings go through bursts of popularity with GM’s - witness Kramnik holding Kasparov in 2000 with the Berlin Defence, which had previously been almost completely ignored.
Do you have any real statistics to back this up?
There have been opening tournaments on all sorts of openings (usually gambits). It doesn’t mean GM’s are playing them.
One good example of that might be Smith, when he first started using the Morra Gambit. He slew many a dragon before the equalizing moves emerged.
Isn’t it possible that before any moves are made, white is in zungzwang?
Yes, I know. I was defending you against an accusation that you were “pontificating” about a subject (chess) you had said you know nothing about. You weren’t–because you haven’t actually been talking specifically about chess. Just about games in general.
-FrL-