Author: Bob Durrett
Date: 12:20:31 01/28/04
Go up one level in this thread
On January 28, 2004 at 13:20:03, Matthew McKnight wrote: >On January 28, 2004 at 11:45:47, Bob Durrett wrote: > >> >>This general topic has been discussed considerably in prior threads and I do not >>wish to revisit those earlier discussions. >> >>Generally, the research done by humans over the centuries [!!!] should not be >>ignored because of the value of the findings. Therein lies the merit of opening >>books. [It would be nice to also incorporate middlegame research findings too!] >> >>Nevertheless, it may be desired to produce an engine [using tablebases] which >>could do a good job of finding best moves in ANY legal chess position. This is >>at least of academic interest if not for practical purposes. >> >>It has been customary to classify positions as being "opening positions," >>"middlegame positions," and "endgame positions." Transitions between these >>might also add more categories. This customary way of treating positions is >>largely based in the history of development of the game of chess and not >>necessarily for any fundamental reason. >> >>The "opening principles," "middlegame principles," and "endgame principals" and >>corresponding strategies [opening, middlegame, endgame, etc.] are largely >>artificial, it seems to me. The ideal chess engine would look at a new position >>and identify characteristics in the position which would suggest a way to >>proceed from that position. Position evaluation and search algorithms should be >>adaptable in the sense that they would automatically self-optimize on the spot >>[in real time, instantaneously] based on the needs of the given position. I >>doubt that this degree of real-time adaptation is being done in modern engines, >>but cannot know for sure. >> >>The "perfect" chess-playing program should be able to play well from any >>position even if access to historical findings were disabled. Or, at least, >>that would be "nice." >> >>: ) >> >>Bob D. > >Let's say you could have a database of middle-game positions, and let's say it's >a HUGE database (on the order of terabytes) Chances are, you would NEVER reach a >middlegame position in actual play for which information could be found in the >database. And if you did, there is no guarantee that the engine will understand >the move, and just end up losing the game later. Therefore, this huge amount of >data would be useless. Now onto the even more pressing point. Who is going to >create these huge databases of nearly useless information? Remember, there are >more possible chess positions than atoms in the universe. > >Matt Perhaps you underestimate the innovativeness of others! The idea that a huge collection of middlegame positions is the only way to represent "middlegame knowledge" is perhaps unfair. I could just as easily postulate that middlegame knowledge, learned by humans over the past several centuries, could be CONDENSED into a manageable dataset or set of rules [with exceptions] rather than positions. Truly, position evaluation functions already contain much of this information. The issue is whether or not these modern position evaluation functions contain MOST of the learned information. Chess programmers will complain that the size of the software and time required to run will become unmanageable if too much middlegame knowledge is incorporated. It is this concern which suggests that some of this knowledge could be stored and called up as needed. Anyway, that's the way I see it right now. : ) Bob D.
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.