Author: Dann Corbit
Date: 13:31:51 07/27/00
Go up one level in this thread
Computers don't play like people at all. They formulate no plans, and have no goals except: snack, snack, snack, snack ... win (by accident!) Tactially, computers are better than any human on the planet. Maybe an ELO of 3000 or higher. Once you get out of the opening book, tactical blunders will be incredibly rare. Certainly GM's will make tactical blunders at ten times the rate of a very strong computer program (just a guess -- it might be hundreds or thousands of times higher). Strategically (positionally) computers are club players. Some programmers have added some board position knowlege, but they obviously don't play like GM's. Considering Deep Junior's play at Dortmund, I think open games at 40/2 are no longer the exclusive property of GM's -- as Junior excelled. Closed positions did much better, but computer programmers can fight to achieve an open position. Crafty [for example] has code especially for that purpose (to prevent rows of locked pawns). What the computer does, is search very deeply for the most expensive target it can snack on without repercussion. The strenght in this operation is phenomenal, with a full 8 moves being possible at correspondence time controls for many types of position. At such types of calculations, the program will be relatively flawless. What fools a computer, and why? A locked position fools a computer. The reason is that it can take a long time for something to happen. This means you are *forced* to plan, which a computer is unable to do. If no big goal can be located in 8 full moves, it's play will be completely aimless. An overpowering massed force attack on the kingside sometimes fools computers. That's because you may have to lose a lot of material to achieve the goal. The computer will see you throwing away pieces in your attack and decide "Surely he won't do that..." by means of the alpha-beta search (and possibly by NULL move). As long as the actual checkmate is too deep for the computer to see when the attack is initiated, the computer won't see it. Computers can be fooled by pawn races when the board is sparse (but not too sparse!). If you have paired pawns and a king or trouser pawns and a long way to go, the opposing computer may not see it (the horizon effect). I have seen this effect at 40/2 games, but I don't know if a computer can be fooled by this at correspondence time controls. Once you are down to 6 or 7 chessmen, though, the computer will play *flawlessly*. This does not mean it will win! If it is in a losing position, it may still lose. But if you slip and any win for the opponent is possible it will definitely find it because of the endgame tablebase files. If there is a long series of forced or nearly forced exchanges that are possible, it may be able to use effectively the tablebase files sooner than that. I strongly suggest to all human players who are not familiar with how computers play chess to go and read Raphale Vasques' Anti-Computer Chess page. There are lots of games demonstrating the weaknesses of computers. However, none of the games is at correspondence time controls. To make sure that the techniques work at that level, they must be tested.
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.