Author: David Dory
Date: 19:07:39 02/11/04
Go up one level in this thread
On February 11, 2004 at 20:36:38, Paul Doire wrote: >Hi Bob, > >My two cents for what it is worth. It would seem clearly if this objective could >be obtained...it would have been obtained....coulda woulda??!! >Some say Junior emulates human play, some say others emulate human play. >AI is a key ingredient, and IMHO true AI would be aware of the environment, >the setting, the pressure...just like a human would be...it would not be >impervious to the surroundings like the pile of silicon that it is. As you >already know, the strength of the programs is in sheer calculations. >GM's select candidate move through a learned process...eliminating what appears >to be futile "trees". They certainly do this far better than computers. >That is the weakness of computers, as I am also sure you already know. >It appears to me that even the newest batch of programs "newfound strength" >comes from an ability to be more selective in its tree..i.e. they are gaining >knowledge. >But, unfortunately they cannot think, and are at the mercy of the current >"state of the art" in the best way to mathematically eliminate "wrong moves". >It is a start most certainly, but it is still based on calculations...and raw >processing power still rules..i.e. Crafty in CCT-6. Some are smarter than >others due to things that I will not pretend to understand fully...null moves, >futility pruning, selectivity... and much more beyond my grasp. >I am not a chess programmer, just a chess enthusiast who loves to test those >engines and their progress. I have been watching and playing these engines for a >little while and they do not understand anything except to play their books and >to follow up with a mathematical examination of what is appropriate for the >situation. They will still make the same mistake over and over...albeit >"learning" has certainly helped that from being so obnoxiously obvious. >We are still so far from this goal it is almost scary. Geez we are the ones in >charge ...right? Right now to this enthusiast it doesn't appear on the horizon >for your wish to become a reality. It sort of reminds me of when Professor Hyatt >states time and time again just how strong a human GM really is. Computers make >up for what they lack in "humanness" by brute force. There has even been talk >about when "brute force calculating power is strong enough ..hardware wise," >that it won't be necessary to have to be so selective in the searches...just >CRUNCH and we will win. That is not human play...we are so far away...and I am >rambling now. >I enjoy your posts, and your exuberance for your beliefs..thank you for your >contributions to CCC. >Regards, >Paul Claude Shannon discussed selective and brute force methods of searching the game tree in his article in 1950. Selective search has been tried, but the success of Northwestern's CHESS4.x showed the pitfalls - it took lots of computation to make selective searching work at all, and all too often, the selection process discarded the brilliant move, along with the dumb moves. As you know, if a pawn is moved just one square sometimes, an amazing move can be reduced to sheer stupidity. Trying to teach a computer to recognize and exploit long range planning, while still tactically playing strong chess, just hasn't been possible, so far. Peasant could find reasonable goals (and moves), for it's pawn playing program, but never could play the whole game, strongly. Edward's Symbolic is the only program I know of trying to integrate planning into it's coding, rather than strictly tactics. It will be very interesting to see what Steven Edward's can do with it. Slate and Atkin's (CHESS's authors), both wanted to create a program that would really understand chess, using a language embedded with chess terms, but that didn't (and still doesn't) exist. They felt the lack of more sophisticated tools (languge) was the biggest reason their program had to settle for being just a tactical monster with a general knowledge of long term goals which stretched beyond it's search horizon. They knew they could improve their evaluation, and make it appear more "human like" by just refining it with countless hours of testing and revision, but were put off by the sheer work required. I'm pleased that our brains are not so easily imitated. :) David
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.