Author: Dave Gomboc
Date: 21:14:43 10/29/99
Go up one level in this thread
I have several deadlines next week, which is why I'm delaying responding to Ratko's posts in another thread. However, I would like to make a timely contribution to this thread, because of an experience today. You may or may not be aware that there is a GAMES group at the U of A. Jonathan Schaeffer, Tony Marsland, and 10 or so Ph.D. and M.Sc. students are involved. Meetings are held periodically, as the situation warrants. Jonathan and Jack van Ryswyck arranged for a visitor to come and speak today. Dr. Vadim Ansheleyevich was a scientific researcher in Moscow before moving to North America and entering industry. He recently wrote Hexy, which plays Hex, at the level of a reasonably strong human. I am the first to agree that GAMES group members are not the finest classic game developers to grace the planet. I hope that you will grant that we are not exactly slouches at it either, though. Jack has written QueenBee, a strong Hex program in its own right, and when he told us he had discovered someone with a program even slightly better than his, we were quite curious to find out more about it. So, Vadim was invited to come to the U of A and talk, and for several hours, that is what he did. He introduced some basic premises on which the assessment of a Hex position should be based, and explained why he thought this was so. He presented an early proposal by -- Claude Shannon! -- and discussed what was good about it, and what could be improved. He provided a mathematical foundation for his static evaluation algorithm, the intuition for which Jonathan described as "brilliant" -- and Jonathan doesn't throw that word around loosely. He commented about constraints for his specific implementation. He generalized experimental results that he had obtained regarding the tradeoff between the quality of static evaluation and the depth of the game-tree search. In short, when I walked out of the meeting room I could have sat down at a computer and wrote an extremely strong hex program. Later, it was Jack's turn, and Jack described in minute detail what he was doing in QueenBee. Ideas, evaluation and search algorithms, data structures, implementation tips, tricks, and gotchas, the whole ball of wax. I didn't stick around for the entire presentation, because I've heard Jack talk about QueenBee before, but the level of detail was similar. Vadim could probably sit down at a computer and implement QueenBee. Vadim and Jack had discovered some similar properties of good Hex programs, but in truth their programs are far more different than similar. Both of them really had their eyes opened by today's revelations. It might be a little rash, but I predict that the next versions of their programs will beat any human opposition: the combination of their respective insights is that powerful. I recounted this experience because I would like to make two observations: 1) Co-operation between parties accelerates progress. Furthermore, my experience with the GAMES group tells me that the co-operation of n parties is much more than (n-1) times more effective than the co-operation of two parties. 2) What academics consider to be "disclosure of information" is significantly different from what commercial software developers consider to be "disclosure of information". I appreciate that some people make a living from computer chess, and therefore do not feel they can afford to "tell all". I also appreciate that even so, some of these people are willing to say something from time to time. Ed is correct when he says that while he might not spell it all out, it only takes a half-word for a sharp listener to pick up the trail and sniff it out. Bob is correct when he says that, in general, commercial programmers do not divulge information. He is even more correct when we add "publicly". These statements are not contradictory: they are different viewpoints of the same reality. And, without intending to offend anyone (my supervisor comes to mind!), it's my opinion that accusations that the publication standard of the ICCA Journal is too low are well-founded. Of course, the flip side is also true: if commercial developers submitted articles, this wouldn't be the case. I don't think that academics will stop publishing papers; it's germane to their jobs. I would like to encourage commercial developers to continue -- or to start -- to contribute as much as they feel comfortable with, in CCC, if not in print. Contributing your ideas and experimental data only increases the rate of progress in computer chess, something that I think is exciting for all of us. Dave
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.