Author: Uri Blass
Date: 14:50:12 03/28/04
Go up one level in this thread
On March 28, 2004 at 17:37:55, Artem Pyatakov wrote: >Hi Uri, > >Thanks for the reply. I guess the same people who were active 3 years ago are >still around :-) > >>>At the same time, any AI >>>work has to compare itself with chess engine filled with excellent >>>human-generated tricks, so it seems to perform poorly. >>>Some examples: >>>*ordering captures first during move ordering >>>*check extensions >>>*futility pruning >> >>I think futility pruning can be easily generalized for other games. >>I also think that history based pruning and using history tables and killer >>moves can be generalized for other games. > >Perhaps with Futility pruning you have a point, although I was talking more >about the decision at which point to prune (in qsearch OR search and at what >futility ranking OR maybe only when no checks are involved or whatever - these >would be the tricks). Perhaps I was not very clear, but actually the history >heuristic and the killer heuristic are examples of things that CAN be >generalized to other games, which is why I focus my research on improving things >such as the History heuristic and the Killer heuristic (I will post details in a >different part of this threads under DETAILS) > >>I see no reason to replace alphabeta with something totally different. >>Humans use alpha beta in every game and I see no reason to tell computers not to >>use alphabeta and if they find that a move is bad to spend a lot of time on >>trying to evaluate exactly how bad it is. > >Where are you getting the information that humans use Alpha-Beta in every game? >There is some evidence of forward search in GM analysis, but the majority of the >decisions are made on the pattern-recognition level, not with search. I have >never seen any evidence of anything as sophisticated as alpha-beta being used by >humans to play a game, and indeed many AI students find it very difficult to >grasp at first exposure (including me). The reason I advocate keeping Alpha-Beta >in my approach is simply because I think we are not ready to discard the whole >framework and replace it with one giant simulation of a brain (neural network) >or some giant genetic algorithm. > >More comments are always welcome, as they help me think about the problem. > >Artem I do not say that humans use brute force with no pruning and of course they use pattern recognition to decide which moves to prune but they basically use the idea of the alphabeta with a smaller tree. If they consider more than one move and see that move A lead to better position for them then they may reject move B because of one move of the opponent that lead to equality without investigating other moves of the opponent. Uri
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.