Author: Don Dailey
Date: 11:02:08 07/09/98
Go up one level in this thread
On July 09, 1998 at 04:23:10, Steffen Jakob wrote: >Hi Bruce! > >On July 01, 1998 at 13:42:27, Bruce Cleaver wrote: > >>What do the gurus such as Dr.Bob, Don Dailey, and the rest think of alternative >>search methods such as 'Best Play for Imperfect Players' (each node contains not > >Have a look at Andreas Junghanns' paper "Are there practical alternatives to >alpha-beta?" available at his page >http://www.cs.ualberta.ca/~andreas/Papers/publications.html. There you will find >a list of alternatives to the alpha/beta-search with a short description of >their advantages and disadvantages. > >Greetings, >Steffen. > >P.S.: Sorry for the late answer... I started to read CCC yesterday :) I strive to make Cilkchess just play the best chess since this is difficult enough as it is. Humans play a horrible kind of chess and it is full of flaws, imperfections, biases and superstitions. On the average, computers play much stronger chess that humans but also play a "horrible kind of chess." Not to mentions that some of our flaws, imperfections and biases are inherited by our chess programs! I don't believe humans play anywhere near perfection so this gives us some exciting possibilites to take advantage of their many flaws and biases. If someone ever figures out how to do this well, we will see an increase in strength relative to humans in these programs. I suspect a properly designed computer using anti-human principles (on much faster hardware) would be capable of winning almost every game against a player like Kasparov. As far as I know, no one has figured out how to play anti-human chess so this is all of course very speculative. But it is my belief that humans are a few hundred rating points from perfection in chess despite the awesome strength of players like Kasparov. Unfortunately, so are computer programs and compared to Kasparov are still ugly chess players! I haven't looked at the web page you mention but I am very open to new possibilities in computer chess. I believe (for the moment) that the biggest breakthrough needs to be in selective searching techniques, I think there is much to be gained without having to add lot's of knowledge. Most will disagree with me but I don't think there are huge gains to be had with adding knowledge, at least the kind defined by a static evaluation funtion. There is something between pure static evaluation and search that is missing, the program needs the ability to "figure out" if it has a correct evaluation or not. Tweaking weights and adding terms is not going to do it in my opinion. Still, I am not one of those who are disatisfied with our progress. We have engineered incredibly strong programs and they keep improving. Being human, we will always yearn to do better and will remain painfully aware of the limitations of whatever we are doing. And there are some issues in computer chess where we are far behind even weak human players. But we will keep innovating and finding better ways. As these are found, most of us will not be bashful about adapting techniques that others have pioneered to make their programs play stronger. The best programs will integrate these approaches to produce strong chess playing programs. - Don
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.