I have written about how traditional chess engines are constructed and about how they go about figuring out what moves to make. I have speculated on how Rybka, a new and powerful chess engine is constructed. I have written a software specification for a new way to estimate the value of a chess position, or a new “evaluation function”.
My task now is to convince everyone that my speculation is sound. Although I lack time, I am slowly starting to create software to test these ideas. I suspect that someone else will have some code togther to experiment before I will, but nevertheless it is worth a try to see if these ideas have any merit.
It would sound like a good project for a grad student or someone who has time to give it a try. —John Jerz —RandomVisitor (chessgames.com)
My big brother has been blogging on chessgames.com under the name “RandomVisitor.” He just recently revealed his identity there.
My son, who’s eight, can already whip my butt in chess if I don’t pay close, close attention. Until recently he has been too willing to sacrifice his queen for little gain, and after he does that I can often win. But not always. At any rate, he’s far better than I am at openings and the endgame, because those are easy to practice on a computer.
Every so often, after he beats me, he asks, “Am I ready to beat Uncle John yet?”
I tell him, “Keep practicing.”
Joshua S. – yes, I agree with you. The standard “evaluation function” that has been used for many years rewards pieces based on roughly how close they are to the center of the board. My idea rewards pieces based on what they can do – that is, specifically how many moves does it take for the piece to attack other pieces, perform defensive moves or other objectives. This is an active area of reasearch.
The problem, as I recall, is that the “value” of a chess position is a very hard thing to compute, and many of the estimations of that value (which depend on the number of pieces left on each side and the values of certain squares on the board) have been shown to be wrong for many classic games of chess. The better player won, even from the strategically inferior position. This is wny end-game puzzles (which study forcing an opponent to make a certain move) are so popular, rather than whole game predictions. The difficulty of these algorithms is a big reason why the companies keep such a tight lid on their methods.
Who? Me? I admit Dennis trained me well to work with web design and usability, but I never plan on writing or publishing about those topics. I want to use them for professional development and teaching.
Sounds like this guy has some great ideas. It seems like anyone who is interested in computer chess should read what he has written and give him some feedback :)
I remember spending a year or two working on my Chess skills, armed with a tournament sized board and weighted pieces, being a member of U.S. Chess…and then academia happened for me. Now my skills are probably too rusty to be significant. I still have that board and pieces in my closet, though.