Author: blass uri
Date: 00:24:11 07/16/98
Go up one level in this thread
On July 16, 1998 at 01:03:45, Don Dailey wrote: >>>The only interesting question to me is how close is Fritz playing >>>to weak grandmaster chess? (or IS Fritz playing GM chess?) >>>WE cannot reasonably argue about whether Fritz could do it, any >>>player can, we would have to get into statistical analysis of what >>>it's chances are based on how many tournaments it played in a year. >>>And then the argument gets more ambiguous than it already is. That >>>seems to be what we do best though. Endlessly arguing points that >>>cannot be resolved using as ambiguous a language as possible and >>>never defining any terms. >>> >>>Bob posted to me that he was not exaggerating in his belief that >>>micro's were not even close to weak GM strength. Maybe we could >>>take a look at the Aegon games and do a performance rating against >>>all the Grandmasters. Perhaps only the top micro's should be >>>considered in this formula since I believe there is a huge spread >>>among micros. Is this data on the web? We could consider this >>>an upper bound on the strength of top Micro's as of 1 year ago. >>>I say upper bound because the games were not played at long time >>>controls. >>> >>>After we did this then we could have a big argument about how much >>>difference the time control makes. >>> >>>- Don >> >> >>someone did this last year. If you "cherry pick" you will always have a >>couple of "GM" programs. But if you take all the reasonable programs and >>average their TPR's, the result is quite different. And probably much more >>accurate. IE I've had the following results in the past year: Crafty played >>in a game/30 tournament with 4 computers and 4 GM's (Crafty, ferret, two >>others). All the computers finished above the highest-finishing GM. Crafty >>played IGM Walter Browne 4 games at 5/14 (5 minutes +14 seconds increment) >>and it won 3 and lost 1. Unfortunately, I know how it plays, and what it knows, >>and I saw it win one nicely played endgame, saw it tactically out-wizard Walter >>twice, and saw him anti-computer it once and roll it up into a small wad. >> >>Ergo I've had plenty of results good enough to suggest that Crafty is a GM, >>and I could make that claim. But I *know* it isn't. I know *many* of the >>things it can't do... >> >>A GM is something that most here don't understand, but the best word to describe >>them is "balanced". They have no obvious weaknesses, they are strong >>tactically, positionally, know their openings, know their endings well, and so >>forth. Computers have too many holes at present... but they are getting >>better. > >We keep beating around the bush. We both seem happy to give very >strong opinions on where they are at without actually giving an >estimated ELO rating of the TOP micro. So I'll start with a lower >bound and you give me an upper bound. From there we will see how >far off we are from each other and whether it's enough to argue >about. I will continue to argue as long as it remains civil and >you are willing (and if you really disagree by as much as it sounds >like you do.) > >In my humble opinion, the strongest micro is within 100 rating >points of the weak grandmaster level so I'll say that it's no >weaker than 2400 ELO rating points. > >Here is a conversation between Crafty and Cilkchess I overheard >the other day: > >Cilkchess: I got beat by a human the other day. > >Crafty: What? How did it happen? > >Cilkchess: It was one of those damn pattern recognition things > again, I have to give it to them, sometimes that works > in their favor. It's such an ugly way to play the game > but it really throws me off. Maybe that pattern > recoginition thing makes them better? > >Crafty: The only thing that matters is that we see so more than > they ever will, *MUCH* more. They have so many holes > in their game it's completely ridiculous. > >Cilkchess: I think they do ok. You are underestimating their > strengths. > >Crafty: What strengths? They have no strengths that really matter. > >Cilkchess: What about pattern recognition which gives them a > powerful selective search? > >Crafty: That's not important. It will never make up for the fact > that we have bigger books that we never forget, we search > orders of magnitude more nodes that they do, we *NEVER* > make careless mistakes or get tired. Humans do *ALL* of > these things. They have so many holes in their game > you can drive a Mack truck through. The other day I > chewed one up into tiny pieces and spit him back out. > >Cilkchess: But they have a unique way of integrating search with > knowledge that we don't seem to have. Also they seem to > be able to reason things out without using a search. > What about that? > >Crafty: So they have a couple unimportant minor advantages. They > have too many weakness that we can exploit. I've seen > and played a lot of games between humans and BELIEVE ME, I > can tell you from experience that they are constantly > getting hammered by oversights, surprise moves (to them) > and time pressure. Until they learn to overcome these > things they will *NEVER* be as good as us. > > >From Crafty's point of view, we are the ones with weaknesses, >not them. It can see it's own strengths but does not really know >that much about OUR strengths and in fact doesn't consider them >to be very important. > >What I'm saying is that computers can still overtake us without >being better in every single way. This is already the case with >me personally. I am not very strong as a chess player and have >never been over 2000 USCF, but I can still see my program make >errors that I would not make. It's getting much rarer now but >it still happens. And yet I am forced to concede that my >program is much better than I am. > > >- Don I think that if you have enough time (for example 72 hours per move) you will be better than your program. Uri
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.