Back in college, my friend and I got into an argument about chess. This was during the Kasparov vs Deep Blue experiment and my friend argued that chess was pointless because there’s a finite number of positions and computers will solve the game. Besides, the results with Deep Blue suggested that computers were already well under way at doing just that.
Of course, I argued that those results showed nothing of the sort as Deep Blue had the unfair advantage of learning every published game of Kasparov whereas Kasparov couldn’t do the same with Deep Blue. Further, it is possible that one day in the future we’ll solve chess but for now it’s an enigma. Humans play imperfectly yet they still play better than computers. I still believe I’m right.
Nevertheless, the day will come. And that day will come soon. I say that within 20 years we’ll have gotten quantum computing to a state where we can start plugging in real world calculations and solving for chess positions. Couple that with memory storage boosts and I’m going to predict that chess will be solved in the next 50 years. Every move mapped out. Every opening analyzed from start through the endgame.
At that point, chess becomes a complex game of tic-tac-toe, right? What happens after that?
Not much, actually, since this has already been done with checkers/draughts and I’d suppose that the majority of checkers players don’t even know this. It’s still an interesting game for humans, and it’s still a challenge to build a computer AI that is plays more like a human than a computer.
I should note that there are basically three different senses in which a game can be solved. The first is to prove simply whether the game is a first-person win/loss/draw. The second is to find one winning or drawing strategy–if the computer is playing from the beginning, that’s all you need. The third sense is what you implied, with “every move mapped out.” Only very simple games have been solved in this way; checkers is not one of them.
I think we’ll have practically unbeatable computer chess players long before we have provably unbeatable players.
I still think the two problems with the experiment are:
humans have done a lot of the opening analysis already
starting from the original position wastes a lot of computing time looking at inferior moves (and it’s hard to tell they are inferior until a long time later in the game)
However thanks for providing both the idea and interesting links as I’ve learnt a lot about modern computing research.
One thing that surprises me is that the opening being played in Glee v Mosier, the fried liver, has not been proved to be sound or not. Or has it? The opening is so tactical it just feels to me like low hanging fruit. Do any GMs play the fried liver?
As I understand it, “Fried Liver” refers to, after, 1.e4 e5 2.Nf3 Nc6 3.Bc4 Nf6 4.Ng5 d5 5.exd5 Nxd5 the move 6.Nxf7 which is thought to be inferior to 6.d4. I don’t know what the state of the art is regarding 6.Nxf7.
As the game went, with 5…Na5, and up to move 9, that is a fairly well traversed area of theory, and is thought to be OK for both sides. GMs do play this opening. If you want to look up openings by position (and rating), ChessLab and ChessBites are a couple of options.
Ops, yeah I was referring to the fried liver, not the actual game being played (it was one of the possible branches the game could have taken). I’m curious whether Nxf7 is sound – I’m guessing it’s been more or less proven unsound, but that it’s still often played below GM-level by aficionados because in order to win against it you have to not make any mistakes.
You won’t see Nxf7 much at a high level because you won’t see Nxd5. New In Chess, volume 94 has Dan Heisman making some comments about (his) recent analysis of Nxf7. The indication there is that the state of the art is that white gets the normal advantage in 6.Nxf7, and might also only get such an advantage in 6.d4 (which is the important point, for his purposes).
I can at least suggest a potential algorithm for a chess computer that would get around some of the limitations of current algorithms.
First, take some simple, low-power algorithm (call it A0), such as might have been used by a consumer chess program in the early 80s. With as much faster as computers are than they were then, a modern computer using such an algorithm should be blazingly fast.
Now, we use that simple algorithm to construct a new, more sophisticated algorithm (call it A1), as such: In any given game position, we look more-or-less exhaustively a couple of moves in advance, to get a bunch of potential positions, and then we evaluate each of those positions by using the blazingly-fast algorithm A0: Starting from that position, the computer plays the entire rest of the game against itself using algorithm A0, and simply sees which side wins, and how quickly. If the side the computer is actually playing wins quickly from that position, then that position is rated highly, and if it loses quickly from that position, then that position is rated lowly, with slow wins or losses ranked in between.
This way, the computer will never be caught in a trap of not quite looking far enough, since it’s always looking all the way to the end of the game. It could, of course, fail by virtue of not looking broadly enough, since it gains that depth at the expense of breadth. But it’d at least provide a different way of looking at the game, which might turn up something new and interesting.
I just wanted to say how neat this database is (particularly chessbites, where for any position you can see the winning percentage). It would be fascinating to look at a similar database for computer vs computer matches. Is there one?
There are both databases with some computer games in them, and ones that only have computer games in them. But I don’t think there are any online (i.e. in browser) database programs that will allow the degree of functionality needed (being to sort out the computer games; or being able to use your own database file). In terms of downloadable databases; there are the CEGT databases (these were mentioned upthread), there are probably more dedicated computer game databases, but offhand I don’t know where to find them. You would also need a database program. I had a quick look for free ones, Scid seems capable of generating statistics on positions.
As an aside, ICC and Playchess (online chess serves) both keep archives of all the games played there, and the archives number in the many millions. And a decent proportion of those games, I would guess, are computer vs computer games. There isn’t any way I know of downloading their databases, and even if you could you would have the problem of figuring out which games are computer vs comptuer. I mention this just because, in terms of quantity, which it seems some people are interested in, I am guessing there are actually millions of computers vs computer games out there; but they might not be easly accessable.
ETA-Maybe not all their games; but regardless they number in the millions.
Does anyone know why chessbites shows ~48000 games starting with “1. a4” but only ~80 games with responses to “1. a4”? Is it such a frightening opening that over 99% of players resign immediately when confronted by it?
I imagine these games are ones where there’s a forfeit or the actual moves of the game are not known. Maybe there’s a requirement in the PGN file standard that games must have at least one move, or maybe some chess software forces this.
When I looked, on the first couple of pages of results, there were a number of draws and black wins, too. I checked a couple of the games against another database, and the other database had zero move games. I think it is a quirk of chessbites.
I think this is right.
PGN files are in text format, and there is no requirement in the file format that there be any moves. No program I have used has had any trouble with games with no moves in a pgn file. It could be a quirk of whatever file format(s) chessbites does use for its databases.
Yes, nobody would play 1. a4 (= a complete waste of time as an opening move), so it could safely be used as some sort of default option for forfeits or similar.
When playing with intelligent chess boards that are used to transmit tournament games live to the Internet, the end of a game is signalled by putting the two Kings on adjacent central squares (e4 + d5 if memory serves.)
Heh. I’d seen that episode earlier. It’s quite cool that someone from the SDMB is actually in there playing. A question - I was actually(for once) able to figure out how Derren Brown did that trick while watching. Did you also cotton on while playing him or figure it out later or what?
When we chessplayers arrived, the arrangements were briefly explained and then the camera crew finished setting up.
While they did that, the players confirmed that none of us had heard of Derren as a player. We’d already checked the English Chess Federation grading list and anyway chess players love to gossip about chess-playing celebrities! (A mate of mine gave chess lessons to Guy Ritchie and Madonna, so you see how that works…)
We agreed there were just 3 ways for Derren to beat a bunch of top players:
offer us money (sadly this didn’t happen)
hire Gary Kasparov or a use a top computer program (unlikely)
play mirror chess (the obvious ploy)
Mirror chess is very specific.
You need to keep the players from seeing each other’s games and also wait for one player to move before making your move on the paired board.
So when we were seated in a huge circle facing outwards we all knew what was going on.
**I want to stress at this point that playing 8 mirror chess games is very difficult. ** You have to remember groups of 4 moves and replay them perfectly whilst simultaneously remembering another set of 4 moves.
And repeat this process over 70 times with no break (the games lasted about 35 moves.)
I could do it, but I’m an internationally rated chess player.
For an amateur like Derren, this was an extraordinary feat of memory. And he did the whole thing in one continuous 3 hour take!
(If you want any more info on the event, I’m happy to talk!)
We all knew each other well apart from the weakest player (who none of us had met.) Apparently he was the secretary of a University chess club.
That game was played next door to mine. Having watched it with interest, I would say that Derren was being advised by a 2300 player (there was one listed as the chess adviser to the TV program.)
I really have no idea how Derren was fed the moves. I suppose the game could have been prepared in advance, but that’s a strain on Derren’s memory (considering all the other moves he had to remember…)
