Author: Bruce Moreland
Date: 15:23:09 07/13/98
Go up one level in this thread
On July 13, 1998 at 15:11:48, Don Dailey wrote: >I agree with your logic completely. I also believe Fritz is close to >Grandmaster strength, but probably not quite there. The 3 game match >does not disprove this in any way. > >I'm sure there will now be many posts saying Fritz is nowhere near this, >and that no micro is anywhere near it. I think this is just not >logical though. If Fritz had won the match, I think a lot of people >would be rationalizing this victory away, now they will say the match >"proves" man's superiority. > >Your point is well taken that Grandmasters vary a lot in strength. If >Fritz were to get into the system to achieve it's 3 grandmaster norms >it would probably have a difficult time, but would have a chance given >enough attempts. I think it is in part an issue of when to award a human title to a machine, and in part an issue of whethere there is enough data to make a particular conclusion. Humans do some things very well, and computers do some things very well, but they aren't the same things. The notion of a human grandmaster is that of someone who does a certain set of things very well, and there is no computer that can do all of these things as well as a grandmaster. Set up and endgame fortress position and a computer won't evaluate it properly, they may not conduct attacks well, they may not evaluate positional nuances well, and the mistakes they make on individual moves are often referred to as beginner mistakes. It may sound like I'm trying to set up a permanent semantic wall, but I am not. There is a mechanism we can use to assess both computers and grandmasters, and that method is Elo rating. The problem is that this isn't being used. If we really did want to assign these machines the title of grandmaster, it would be a simple matter to make them all memebers of FIDE, and get them playing in rated events, along side of grandmasters and other players that grandmasters play against. This hasn't been done, in large part because the humans don't want to do it, so we are left with guessing about the machines' strength, unless we want to apply some unreachable human standards against the machines, and declare that since they can't reach this standard they aren't grandmasters, as I expressed in my second paragraph. I think that I can easily claim that the stronger micro programs are all of grandmaster strength when they play traditional (fast) blitz chess on the servers. You can find many games where the humans play their strengths against the weakness of the computers, resulting in games that are embarassing for the computers, but the simple fact is that in the vast majority of cases it is the grandmaster who ends up resigning. Everyone who runs a computer that the grandmasters play against will be able to point to many little matches in excess of four games where the program scored between 75 and 100 percent, and will be able to point to very few cases where the reverse is true. Likewise, if you look at lifetime totals between any of the computers and any of the GM's, it will be very rare to see a case where the grandmaster has scored more than 50%. I would expect that at least 75% by the computer would be the most likely case. I think there is way more than enough data to conclude that they are of grandmaster strength at blitz, and in fact there is almost no contradictory data. So in my opinion, the question in blitz chess as manifested on the servers, the question is not whether the computers are GM strength, the question is whether they would be in the top ten. And I personally expect that the answer is yes. I there there is a time control beyond which grandmasters must be stronger than computers. I don't know what it is, but it makes sense to expect that it is somewhere in between blitz and a tournament game. It only makes sense that the crossover point will continue to move closer to tournament time control as we go along. Personally I don't think we are there yet. bruce
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.