Author: Ratko V Tomic
Date: 08:00:56 10/09/99
Go up one level in this thread
> If you do not want to play the same opening over and over you can start > from the opening that you want to play by playing the first moves > against yourself. With "leaqrning" on Fritz ended up as black, on my 1. e4 always playing Nc6 (Nimzowitsch opening). The rest of hundreds of thousands of positions they brag about in their promotional materials got weighed out of its repertoire, just because the first game it picked that line and it won it, second game it tried it again, won it again, then tried something else, but came back to Nimzo, and ended up stuck in it. The reason was probably since it had very few games in that line, the few it played against me weighted it as "sure win". The whole scheme was obviously thought up and tuned for unsupervised (by a thinking human mind) machine-machine play where Fritz (or others from CB engine set) can win identical game against non-learning program tens of times in a raw, and it all counts into its "rating". I have no problem with companies trying cheap gimmicks to get ahead, you just need to watch tv commercials to know the mentality of marketing folks. I have problem when that mentality takes over and overrides their product quality and design department altogether and they start treating their customer as a dumb machine playing automated match against their program. Why exactly does remembering past evaluations (which could save great deal of customer's time) has to reduce probability of entire openings it didn't play before or entire openings in which it lost some games? It doesn't have to, it's not some prgramming difficulty or logical necessity. It's a pure gimmick devised to win automated games, which can't be turned off when something else than machine plays against the program. The way it works now, if computer wins in some line, then it will throw away bulk of its book and play almost exclusivly just that single line (especally if that was a rare line before since a single win will make it "conclude" the line has, say, 80% success). If it loses a game, it throws away virtually entire opening, be whole Sicilian or whole French, even though it lost for unrelated mistake far away from the opening. Now, their opening book options have some supposed sliding bars, where you could adjust how much you wish learning to affect later choices & variety, but with Fritz it didn't work, I had to reinstall the fresh opening book from CD. Then I adjusted the setting to maximize variety and minimize effects of results on opening choice, but two days later, it again played only Nimzo on 1. e4. Well, I checked their T-notes, and they suggest you make opening books read only files. That worked. Now that's ridiculous. Obviously someone thought it a "clever" idea to ignore "learning off" setting so they can win extra few identical lines against programs which respect the user settings in an automated and essentially unchecked (by common sense) machine-machine play. > I think that customers usually are using the programs for analyzing > and not for playing. Well, they're good only for complementary analysis to a high quality GM analysis, to fill in the missing short term tactical shots (which may be obvious to a GM but not to a regular player). You can't use it to find a good plan of game. Now, the program may guess the right move in a given position, which matches the plan recomended by GMs, but it will guess it for the wrong reason, i.e. it will not pursue the plan but will play toward whatever looks good 10-12 plies from the current position. In any case, I don't think analysis is the what they're used mostly for. Usable analysis feature started appearing only in recenet few years, and some well selling programs don't still have it. If you're a chess player who plays in competitions, you may use it for analysis, but you will also use it for sparring partner and spend more time playing than analyzing (that's more fun anyway; nobody started chess so they can learn to analyze games, but we all started it because of the fun of playing the game). If you're not any more (or never did) competing, analyzing is even less often. Yes, I may check a new program against few positions I checked earlier programs against, but that is really a minor activity. I think most people who buy these chess programs do have a day job, and play for fun of playing. They also have limited time they can spend playing against the program, and it bothers me to have to waste what little time I can put into it, in sitting there in front of the screen waiting for Fritz etc to perform the identical calculation it did yesterday or the day before, to come up with the identical result. Alternatively it will let me play Nimzo opening all day. Well, great, thanks. It doesn't seem that whoever thinks up features of these programs (especially CB UI designers) ever plays a real game against them. If one were even, for the sake of discussion, to grant you that analysis of games/positions is the main purpose of chess programs, unless you wish to claim that programs should not play regular games against its customers at all, it still wouldn't justify hard-wiring the "learning" feature for (in effect) cheating in machine-machine play (by dropping the the entire openings, or playing exclusively one line, "killer line") or alternatively wasting customer's time by making him wait for the same evaluation to be computed over and over. This form of "learning" (automating the comp-comp killer lines) should be an option, while the learning which saves customer's time while not locking the customer into a single line should be the normal learning mode (in non-autoplay). I think by being involved with one of the programs (Junior) you have somewhat different focus, and see automated play & its results (along with the analysis benchmarks on test sets; or whatever Kasparov or Kramnik claim to use chess programs for) as the center of action, while most customers will never play a single machine against machine game.
This page took 0.01 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.