Author: Mark Young
Date: 04:49:16 07/10/01
Go up one level in this thread
On July 10, 2001 at 06:47:59, Christophe Theron wrote: >On July 10, 2001 at 06:07:21, odell hall wrote: > >>On July 10, 2001 at 05:49:39, Christophe Theron wrote: >> >>>On July 10, 2001 at 05:02:35, Vincent Lejeune wrote: >>> >>>> >>>>>BTW the first round of the match is completed now. ChessTiger 14 reached very >>>>>good 61.5 %. Could someone >>>>>please calculate, how much ELO difference 61.5 % mean compared to the average >>>>>ELO of the opponents? >>>>>Thanks a lot! >>>> >>>>That means a +92 Elo difference >>> >>> >>>Sorry Vincent, but this is not right. >>> >>>A 61.5% winning percentage means approximately a 80.5 Elo points difference. >>> >>>The rule of thumb is very simple: take the winning percentage, substract 50, >>>then multiply by 7. >>> >>>This rule is an approximation (but a very good one) that works only in the range >>>20% to 80%. >>> >>> >>> >>> Christophe >> >> >> Hi Christophe >> >> >>Congradulations on your enourmous successs, I believe i, and others would like >>you to comment on the result of Chess Tiger in Argentina, Please no one will >>think your bragging if you do, you maade computer chess history and i think you >>deserve to bathe in the limelight. Did you expect this result for tiger? Do you >>believe Computers are Grandmasters? I know this is a very controversial subject, >>but please express your honest opinion, Several programmers have already taken a >>stand , what is yours? > > > >Thanks Odell. > >First I would apologize for not answering on this subject earlier. > >Somebody (was it you?) has posted a message "to Christophe Théron", and I have >unfortunately been unable to answer. > >The reason is that I was working very hard on Chess Tiger for Palm (now >released). > >About the result in Argentina I must say that I been caught by surprise. I did >not expect much from this event because I expected much more resistance from the >human side! Chess Tiger has been used, when I had recommended them to use Chess >Tiger (but apparently it was not possible to change at the last minute), and I >expected the human players to have less problems against the more quiet playing >style of Chess Tiger. Well... "quiet"... as long as you do not give it an >opportunity to break into your defenses... > >Maybe I did not expect much because I have slowly been intoxicated by people who >say that current chess program are not at the GM level. I guess I was about to >believe it, and I had very little data from my program against very strong human >players. > >But what we have all discovered is that GMs have weaknesses too and that a >commercially available chess program can easily point these weaknesses out. > >Now to be absolutely honest I must say that the GMs in Argentina have probably >been caught by surprise as well. I think that the tournament rating of 2788 is >probably too high (by 10 or 20 points... no, just kidding :). > >I believe it is indeed possible for a human player to increase his performance >against computers by learning their weaknesses, but I also believe that there >are limits to this. It will be interesting to see how Deep Fritz and Fritz 7 perform in their matches against stronger GM's under match conditions. This may give us an idea of what kind of limits we are talking about. I hate to lump all top computer program results together, I would rather treat each program as its own player. But the data pool would be too small. I know many here think all programs with a simular rating in computer vs computer games should perform about same against humans, "as all computer programs have simular weaknesses". The problem I have with the game data and results of any new program or untested program aganist humans is how true in fact does computer vs computer results correlate with computer vs human. This is only anecdotal evidence, and this is what got me wondering about this question again. I have seen some bad GambitTiger results against humans, even though GambitTiger is rated higer in computer vs computer game then ChessTiger. The question then is is it really better to have a risky style of play such as Junior 7 and GambitTiger, then the more solid style of a ChessTiger or Fritz 7 when playing humans or computers in terms of results. > >If you listen to some people here, they say that with the right anticomputer >techniques it is possible for a 2100 elo player to stand a top chess program. > >This is wrong. And do not forget that chess programmers have only started to >program anti-human strategies. Chess Tiger has a "antihuman" mode, but which has >not taken a lot of time to develop. I mean, if beating human players was my only >priority, there are several very important techniques I can think about that I >would have added to the program, and nobody would dream anymore about a 2100 elo >player beating a top chess program. I think that all my colleagues would be able >to do the same. > >Specializing in anticomputer strategies might allow one to inflate its rating >against computers by, say, 100 elo points, but that's all. > >But the interesting point here is that if it is applied by really strong human >players, then the human side still have some good years. And I like this, >because it means the competition between man and machine is not over. > >I think the SSDF elo rating is comparable to human FIDE rating, if you consider >that the human players have not specifically prepared for computers. So in a >sense the result of Chess Tiger in Argentina confirms this. > >But as I said it is possible, maybe, for a human player to inflate by 100 its >rating by preparing very hard for computers. So if you take the elo of the best >programs on the SSDF, which is approximately 2650, it means that in order to >equal a top program under tournament conditions you need to be 2550 elo yourself >at least (if the program is running on K6-2 450). > >There are lots of 2550+ players, so it leaves a lot of opportunities for the >part of the human kind which uses only his brain to beat the part of the human >kind who has decided to enhance his brain abilities with external tools. > >Yeah... Don't forget that it's not man vs computer actually. Who can really >believe that? It's man vs man. > > > > Christophe
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.