As if just to spite FIDE, the Israeli program Deep Junior just won the 12th World Computer Chess Championship. Israelis were banned from the human WCh, but this one was held in Israel. It was a narrow win over Shredder to which I credit the home (magnetic) field advantage. Junior is programmed by Amir Ban, Shay Bushinsky and opening book trainer GM Boris Alterman. Kudos to Amir and my former Herzlia comrades Shay and Boris.
The consistent dominance of very few names shows the primacy of programming, but as Ban himself recently said, book training is critical. From a recent message board post:
This is unavoidable in a competition situation in which program tweaks could be disastrous but the book must be tuned for each opponent. With hundreds of millions of positions in a database custom-designed by a strong Grandmaster, the line between human and machine play is blurred beyond recognition. I wonder how far this will go until such databases are either limited or banned altogether to put the focus back on artificial intelligence. The way it is now many games don't really start until move 20 or beyond and many of the decisive games are already evaluated as a winning advantage by the time both programs are out of book!
Of course this is the case in human-human chess as well. But study (and what to study) and memorization are part of the human challenge of chess. Adding a few more gigs of databases is not machine chess. When Deep Boris prepares an opening trap for Fritz, that's not computer chess, it's a hybrid. Junior must play what it is told by the human. (Just using Junior as an example here; all the programs are like this.)
Using opening ballots like checkers seems to be the unavoidable future. The starting position of each game (or even all the games in each round) would be drawn at random from a long list of positions. In a perfect world they would play each position with both colors.
Mig says: This is unavoidable in a competition situation in which program tweaks could be disastrous but the book must be tuned for each opponent. With hundreds of millions of positions in a database custom-designed by a strong Grandmaster, the line between human and machine play is blurred beyond recognition. I wonder how far this will go until such databases are either limited or banned altogether to put the focus back on artificial intelligence.
Steve responds: I think the quest for artificial intelligence is a complete failure. Do you see any great gains being made? The focus today seems to be stretching the opening books preparation, increasing the number of pieces available for end game database, and lengthening the search function. If you consider a 20 move book, a search function of 12 moves, and a game that gets down to 7 piece databases at 50 moves, then the computer only plays chess for 18 moves.
Mig says: The way it is now many games don't really start until move 20 or beyond and many of the decisive games are already evaluated as a winning advantage by the time both programs are out of book!
Steve responds: Yes, and that's hardly the awe inspiring view of computer chess the average person on the street has. When I tell people that the computers would get killed if not for the preprogrammed openings, they are always disappointed. Reminds me of the old Reti joke about "A Difficult Position", where he shows the opening setup. Yes, it's so hard the best computer programs in the world can't play it well.
I have always wondered why a computer is allowed an opening book. I think the GM's of years ago were too persuaded that the computer programs would never be competitive and casually agreed to the books. Now it's the accepted practice.
Mig says: Of course this is the case in human-human chess as well. But study (and what to study) and memorization are part of the human challenge of chess. Adding a few more gigs of databases is not machine chess. When Deep Boris prepares an opening trap for Fritz, that's not computer chess, it's a hybrid. Junior must play what it is told by the human. (Just using Junior as an example here; all the programs are like this.)
Steve responds: I agree. Computers don't play chess. They are programmed to play their first 15 moves or so, and they end game data bases play their last moves. What's left? The middle game, where computers are admittedly at their best. So they don't play chess at all, but just one of the three primary components of the game. I've always wondered how a 2300 IM would fare against Kaspy if he could use Junior's books and end game databases.
Mig says: Using opening ballots like checkers seems to be the unavoidable future. The starting position of each game (or even all the games in each round) would be drawn at random from a long list of positions. In a perfect world they would play each position with both colors.
Steve responds: I recently had the same idea while reading a bridge column about a duplciate tournament. I think the above suggestion would also be a good idea in human-computer matches also.
One fact of the world computer chess championships stands out above the rest. The order of the results does not reflect the strength of the programs, rather it reflects the speed of the hardware. Quad-processor opteron systems won over single processor AMD 64 sytsems which beat Intel Pentium systems. Simple as that. Until the WCCC has a machine formula (like formula 1) ie a level playing field, which program "won" is meaningless.
The focus in my opinion is that the aim of computer vs computer games is not related to an idea of "sport" but to the best use of all the available technical resources (the biggest dataBases, the quickest cpu, the best algorithms)to produce nearly perfect chess games.
Nothing of this is related to the Artificial Intelligence concepts.
Moves in the opening book of a program I imagine were stored by teams according with a previous program evalutation (like human preparation no? program analize position for hours to fix his knowledge at the best, in games he can use only few minutes to choose moves)...and a lack in this could be considered a fault of the program and of its team.
I mean if a program in a game ends his book and find itself in a lost position this is its fault (or better its team) exactly as a grandmaster when his variation receive confutation from
"at home" better prepared or "at board" very intuitive opponent.
I do not see the sense to limit computer knowledge, why looking at computer games where each move is the fruit of a few minutes analisys of a single CPU?
When instead is easily possible to see moves that are the best summary of hours and hours of many powered CPU's work?
What an innatural boring pity to see a Pentium in a game to think minutes and minutes in a well known position (even the startposition) and then performing the move just like 10000 times prev
I think the only interest a chess player could have in following computer games is in discovering very good moves, the best moves tecnically available, and not to see a sport fight like a football supporter: sport I think is reserved in chess to human tournaments, computer tournaments is only science.
For this last reason I do not agree also with Mig's idea to select a random startposition in computer game: again chess player will see the tournament games with the spirit of supporters for one or other program, not with the spirit of discovering the best moves.
"Using opening ballots like checkers seems to be the unavoidable future. The starting position of each game (or even all the games in each round) would be drawn at random from a long list of positions. In a perfect world they would play each position with both colors."
This is taking us away from the topic of computer chess, but I just wanted to comment on Mig's quote above.
The list of positions mentioned doesn't necessarily have to be long (for human-to-human play, at least). Some advantage would be gained simply by flipping a coin at the start of play, and on "heads" white must play 1.e4, and on "tails" he can play anything except e4. This would be in keeping with the current statistics on opening moves -- in which 1.e4 is played almost exactly half the time -- and would force both white and black to be a little bit more flexible in their preparation.
An improved version of this one-ply system would be to roll a die before each game, and on 1 to 3, white plays 1.e4, on a 4 or 5, he plays 1.d4, and on a 6 he plays something other than e4 or d4. This is also in keeping with the current statistics on opening moves.
As a final improvement, the above technique could be expanded and made "non-biased" by also including black's common first replies to each of whites forced moves, creating a two-ply system. On 1.e4, black rolls a die, and on a 1 to 3 he plays 1...c5, on a 4 or 5 he plays 1...e5, and on a 6 he plays something other than c5 or e5. On 1.d4, black rolls a die, and on a 1 to 3 he plays 1...Nf6, on a 4 or 5 he plays 1...d5, and on a 6 he plays something other than Nf6 or d5. On a white first move other than e4 or d4, black is free to play anything.
There are two main advantages to this system over other proposed ballot systems. First, it is *simple*, in both breadth and depth of choices. In checkers (draughts), a game with many fewer permutations, and hence a much higher risk of "draw death", the random openings only go three-ply deep, so anything above two-ply for chess may be more of a "solution" than necessary.
The second main advantage is that this system is *open*. By allowing a "none of the above" choice for each ply, we are not forced to account for *every* possible viable opening permutation -- remember, this is a much higher number than in checkers (draughts) -- but only need to account for the most commonly played ones.
A further advantage to the "open-ended two-ply ballot" system is its stability. These basic ratios of opening first moves are not likely to change significantly in the coming years, or even decades (or possibly even centuries). This means no panel of judges will need to review the viability of the ballot list and make continual changes, changes which would be extensive given the vast fluidity of current opening theory at deeper depths in the move tree. Also, this panel would have a great deal of power, and anything that avoids further power struggles in chess is a good think, I think.
I would welcome further comment from the chess community on these ideas, both here or in the message boards, or by private e-mail.
On chess openings, there is a good site called chessopeningsdatabase.net