Slashdot Mirror
Rybka Solves the King's Gambit Chess Opening
New submitter smarq2 writes "Chessbase reports that chess programmer IM Vasik Rajlich has solved the King's Gambit chess opening with technical means. 3000 processor cores, running for over four months, exhaustively analyzed all lines that follow after 1.e4 e5 2.f4 exf4 and came to some extraordinary conclusions."
Update: 04/02 22:11 GMT by U L : Skuto points out that this is the same person who was found guilty of plagiarizing GNU Chess and Crafty.
There is only one move in which White can force a draw - and to find out what it is, you'll have to RTFA.
Nah, I'm just pulling your leg, here you go..
Anybody really interested in the details will still RTFA anyway and the rest of us won't be left hanging with a teaser.
"Whenever Rybka evaluates a position with a score of +/– 5.12 we don't need to search any further, we have our proof that in the continuation there is going to be a win or loss, and there is a forced mate somewhere deep down in the tree. We tested a random sampling of positions of varying levels of difficulty that were evaluated at above 5.12, and we never saw a solution fail. So it is safe to use this assumption generally in the search."
Hmmm... Really? The whole "solved" thing hinges on this assumption.
They didn't calculate all possible moves, but skipped every branch where analysation showed an advantage high enough for one party to be "absolutely sure" to win. So while the algorithm is very sophisticated, it technically didn't solve King's Gambit.
Slashdot even covered that; http://games.slashdot.org/story/11/06/29/1824253/worlds-best-chess-engine-outlawed-and-disqualified
Looks like the editors have a short memory.
Chess programs usually score a position in "pawn equivalents". Having one pawn more is a +1, unless your opponent has compensation in position. Having one less would be a -1. Other examples are:
-a knight or bishop is worth roughly 3 points
-a rook is worth roughly 5 points
In practice, skilled players will win a +5 position reliably. A +3 is usually enough as well. So even if Rybka's evaluation is a bit off, I would not see much chances to win the match from the inferior position.
C - the footgun of programming languages
For the benefit of those reading this who are not familiar with the Rybka situation, I want to point out that the author of Rybka did do a lot of original work on it too -- it wasn't a "complete clone" like some of the more blatant plagarism cases in the computer-chess world have been. But he still did plagarize certain parts of Crafty and Fruit which gave him a significant advantage over the other competitors in the WCCC and other tournaments. These tournaments are really competitions between programmers to see who can make the best-playing engine. And in order for them to be fair, each team entering the tournament must write their engine entirely by themselves, and disclose the origins of any third-party code used in their engine. Rybka versions that contained third-party code from Crafty and Fruit were entered into several of these tournaments without declaring that this code was used, and without getting the permission of the authors of Crafty and Fruit (either explicitly, or via some sort of license grant). In fact, Rybka's use of this code violated both Crafty's license (a non-commercial-use license which also has a clause prohibiting use of Crafty code in a tournament without permission from Crafty's author, Dr. Bob Hyatt) and Fruit's license (GPL v2).
First, the King's Gambit has not technically been "solved", for the most rigorous definition of "solved". Unlike, say, checkers, there are still lines (i.e. series of moves) within the King's Gambit that have not formally been examined.
Second, we are strictly speaking about the King's Gambit Accepted. That is, white begins with e4 (King's pawn forward two spaces), black replies classically with e5 (King's pawn up two spaces), white then gambits the f-pawn (King's bishop's pawn up two spaces), and black captures the f-pawn, accepting the gambit. As TFA mentions, the King's Gambit Declined has not been examined nearly as thoroughly.
Third, all of this is only somewhat relevant to actual chess playing, and only at the very highest levels of play; the average FIDE Master (i.e. a well above average tournament player, though nowhere near being among the 1,000 best players in the world) need not remove the King's Gambit from his repertoire because it has been "solved". This has, historically, been one of the most dynamic openings in chess, with tons of opportunities for tactical tomfoolery and psychological pressure. When we talk about "perfect play", or "near perfect play", we're already reaching beyond the level of world champions.
Fourth, while not every line has been thoroughly analysed, the ones that haven't are irrelevant. An advantage, in chess, is calculated on the basis of a difference of pawns. So, if the black player has all the same pieces as his opponent, save for an extra pawn, all other things being equal, we evaluate the position as -1 (i.e. from the perspective of white, the position is minus one pawn). Pieces other than pawns are weighed differently, even when we are solely looking at material differences. Traditionally, knights/bishops are said to be worth three pawns, rooks are worth five pawns, and the queen is worth nine pawns. However, the actual position of the pieces affects their worth; a knight very near the centre of the board is, often, worth more than a rook (i.e. A knight near the centre can have up to eight possible moves, whereas a knight in a corner can only have two possible moves). Thus, a position that has been evaluated as +/- 5.12 means that one player has more than a rook's worth of advantages over his opponent. Even in low level tournament play, it is very reasonable to assume that the advantaged player will win the game; at grandmaster level, this is so certain that it is considered impolite, even downright offensive, if the disadvantaged player refuses to resign.
Fifth, while different computer chess engines do evaluate positions differently, I have yet to come across a position about which the analyses of different engines have diverged by more than 2 pawns. An evaluation of +/- 5.12 by a top-notch engine can safely be assumed to be conclusive, since since most of what I said in the above paragraph also applies to an evaluation of +/- 3.0. Whatever else it may be, Rybka is certainly a top-notch engine.
Finally, it is true that Rybka's having reached its current strength relies on what are at best described as questionable appropriations of others' source code and algorithms. Nonetheless, the presented findings have an intrinsic value that is not dependent or reliant on notions of intellectual property or publicity. I am frankly ashamed by posters who have suggested that this article ought not have been publicized by slashdot because of its source. Knowledge is knowledge, period, and while it is both sensible and necessary to place ethical restrictions on scientific methodology, it is simply insane to deprive oneself and others of data that has, for better or worse, already been gathered.
Trust me, kids; don't drink and post.