Author: martin fierz
Date: 20:42:06 10/24/02
Go up one level in this thread
On October 24, 2002 at 20:14:04, Russell Reagan wrote: >On October 24, 2002 at 17:33:05, martin fierz wrote: > >>>Why do you ask? No human world champion calibre player has ever lost 10-0, or >>>even come close to losing 10-0, much less 20-0 or 40-0, so what's the point of >>>even asking? >> >>umm, you are young my friend :-) >>too young to remember the candidate's matches which the one and only bobby >>played and won 6-0, against players who were also "candidates", meaning world >>champion calibre players. > >What I meant to say was that no world class human player has ever lost horribly >to a computer. yes, that is right. what i meant to say is that world-class players have lost matches horribly though, when someone came along who was years ahead of his time - like fischer was then. kasparov once smashed miles in a match 5.5-0.5. it does happen sometimes. i also think that this might happen again in the future when computers are even faster and mainly smarter. personally i think it's not too much about processing speed any more now. computers are outcalculating their GM opponents easily, and an extra ply won't make such a difference any more. but moves like h4 by DF in game 1 just have to disappear! david omid seems to have implemented this kind of knowledge in his program, and once somebody shows it can be done, the rest will follow. programmers often think something can't be done, but when a guy comes along and does it, they start thinking about it more seriously and realize they can do it too. >happened, the norm is that it's usually about even when you have top humans >against top computer chess programs. yes, right now. but what if for example hsu did build his trillion-node-per-second machine? or if we just waited 10 years like bob suggested and hoped that moore's law continues to hold? i believe that the day will come when the machines will first start beating the humans in matches, and then dominate them. i'm not looking forward to it though :-( >I'm not sure if we will ever reach a time when the best computer can beat the >best human 10-0. i think for that, the human would have to "cooperate" in the sense that he would have to play for a win too. >I suspect that even if we have computers announcing mate in 342 >after move 4 of the game, that doesn't mean that a world class GM won't be able >to find those moves to get him to the draw. but if the computer announces a mate, then it is mate - how would the world class GM still get a draw then?? >You know more about checkers than I, so perhaps you can answer this better. >Aren't checkers programs better when compared to humans than chess programs vs. >humans? absolutely. there was a match between one of the commercial programs and one of the world champions recently, about 80 games, and the computer won with something like 16-3. i should add in defence of the human that they played something like 10 games per day in the evening when the champ was playing a checkers tournament in england during the day. and, the champ being russian, i also hear there were large quantities of vodka involved. it was definitely not a serious match. in defence of the computers i should add that i believe that the program which played there is nowhere near the best checkers programs, and that the champ would not win a single game against one of the top programs. >I think I remember Shaeffer saying that Chinook pretty much played >perfect checkers, and even so it was not able to beat the world's best human, that is BS. it's like the DF team saying DF is "much better than DB" and hsu saying "DB is much much much better than DF"... chinook at it's prime was running at something like 100kN/s, according to the latest info i have from dave gomboc. my program is running 10x faster these days on decent hardware, but it is *far* from perfect. checkers programs rely very heavily on opening books to steer them through the maze of the opening. checkers is much more forced than chess - you can make an error as early as move 2 and just lose the game. the programs simply can't see through all this. the programs can play close to perfection as long as they manage to stay in book long enough (assuming the book is perfect!) - then they often can see right into the endgame database and see a draw right after the end of the book (happened quite often in vegas). but if you catch them out of book in a tough position, there goes your perfection! BTW, chinook lost a game with a relatively simple mistake against lafferty, after tinsley forfeited their match. so much for perfect... >think humans have shown the ability, at least some of the time, to play close >enough to perfect to reach a draw. > >What do you think? checkers is a relatively simple game. humans can get much closer to perfection in checkers than in chess because of it's simplicity (extreme example - you can play tic-tac-toe against a perfect opponent 1000 times and never lose a game because it's simple enough). among top humans, the drawing percentage is very high too. i think in chess, you can "stir things up" much more easily than in checkers - but i'm no expert on the checkers stuff... aloha martin
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.