Author: Chris Carson
Date: 12:37:30 04/17/02
Go up one level in this thread
On April 17, 2002 at 14:54:58, Roy Eassa wrote: >On April 17, 2002 at 14:19:02, Chris Carson wrote: > >>feel free to disagree with me. I am often wrong and people catch me at >>it. :) I get frustrated too with the hype and carried away at times. >> > > >Chris, > >I like think my view is quite moderate and is not hype; do you agree? : > >I do not think we can be SURE whether or not PCs are truly 2650+ FIDE. I think >that there is definitely SOME evidence that they are, but that there has simply >not been a sufficient number of games against GMs who are relatively >well-prepared and motivated. The evidence that exists does not come close to >passing the typical requirements of a decent scientific experiment in which >extraneous factors are well-controlled. (BTW, I DO think the top PC programs on >the fastest PCs are more than 50% likely to be over 2500 FIDE.) Older GMs in >particular have not spent a sufficient percentage of their in the presence of >strong computers to be well prepared psychologically, much less to have >developed very good anti-computer techniques. Ofcourse the comps have scored better than 2700 and on slower hw. The ratings are based on games against rated players. Other factors such as age, preparation, motivation do not influence a rating calculation. These may be legitimate factors for discussion, however, they do not have an effect on calculating an ELO rating. Ratings are the measure of strength in chess. If you accept that a GM is 2700 ELO based on games from rated players, then a computer can achieve an ELO by also playing rated players. No additional requirements. People just play the game and get a rating, the program/hw is no different. Factors such as well prepared, motivated, psychological factors, physical factors are not part of ratings calculation. If you sit down to play the game, then you are fit to play. Conners (a great US tennis player) use to say that when you step on the tennis court all excuses a re left in the locker room. Nicholas (a great US golf player) use to make similiar statements. > >You have a PhD and obviously know what it means to really, intensely study some >highly difficult topic for months and months (even years). One's knowledge and >skill-set in that field grow enormously as a result of such work. I just don't >think _any_ GMs have done substantial amounts of intense work thus far on >anti-computer chess techniques. I think it's highly likely that at least a few >younger GMs WILL undertake such an effort in the next 5 or 10 years, especially >if there is any good motivation to do so (money, for example). Also, very young >GMs will have spent a good chunk of their careers in the presence of strong PCs. > This is not a scientific study, I am doing ratings calculations, that is the measure of strength. Human vs Human ratings are calculated the same way, a human is not involved in a research study to get a rating or a GM title, it is a prescribed calculation and for the GM title a defined set of parameters. Computers will never be GM's, however their ratings can easily be calculated, no research study required, just moderated games against rated players. All the human vs comp games in my ratings list meet moderated match/tournament conditions for 40/2 games. That is all that is required. There are plenty of games, this is not a valid argument. Plenty for some programs to have a rating and enough amoung all the programs to be statistically significant. No human needs hundreds of games to get a rating, comps are no different. >Even though computers are getting faster all the time and the software is >improving in its anti-human methods, I still think the scenario described above >(SOME but certainly not ALL GMs really cracking down and becoming much better at >anti-computer techniques) leads to a greater-than-50% likelihood that there will >be at least SOME GMs (or strong IMs) in the next 10 years who will consistently >outscore the top programs running on the fastest PCs. > >I think it is more in the style of hype when somebody says that humans will >NEVER learn to do something (I am not saying that this is what YOU claim, but >that some claim this). > > -Roy. Sure, some GM's will perform better than others for different reasons against the comps. The same is true for GM vs GM performance. GM B. Larsen had a very good (positive) score against GM B. Fisher in during the 60's and early 70's. Computers will score better in human GM tournaments than in match play against GM's. Preparation, adaptability, knowledge and experience are all virtues of humanity. So I agree that finding holes in programs (or other peoples games) is one thing we humans do very well. The programs have some very strong advantages as well, that is why they are GM strength, but not perfect or invincible.
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.