Author: Rolf Tueschen
Date: 11:09:50 02/05/03
Go up one level in this thread
On February 05, 2003 at 13:20:56, Chris Carson wrote: >On February 05, 2003 at 12:02:31, Rolf Tueschen wrote: > >>On February 04, 2003 at 19:13:22, Andreas Guettinger wrote: >> >>>On February 04, 2003 at 16:49:01, Rolf Tueschen wrote: >>> >>>>On February 04, 2003 at 15:54:32, Andreas Guettinger wrote: >>>> >>>>>It can. With tactics, even weaker comps can beat GMs. See Kramnik and Kasparov. >>>>>Be careful, I studied many hours statistics at university. But you're right, it >>>>>is statistics. >>>>> >>>>>Your anti-comp strategy system IS a myth. I laugh always when I see this >>>>>argument. If one traines anti-computer, then he manages to get the computer to >>>>>look really silly in ONE game, but the 50 games he lost until he got this game >>>>>he never shows. Your anti-computer strategy is unsuitable for tournament play!! >>>>> >>>>> >>>>>But maybe you manage to win a 24 game tournament against a top program with your >>>>>anti-computer strategy? Show us! :) >>>>> >>>>>regards >>>>>Andreas >>>> >>>>Please say just a few words about the phenomenon that weak players could beat >>>>2700 comps. Would you believe if I claimed the same for a match against human >>>>GM? I think that is the reason why we should discuss the whole question a bit >>>>deeper. >>> >>>I would not believe you, if you claimed that. :) >>> >>>It is not my intension to deny that there are positions that computers just are >>>not good in. Their knowledge is just not sufficient (maybe about kingsafety or >>>whatever) and they make mistakes. Because the knowledge of the human >>>counterplayer is almost always bigger, such mistakes are often very easy to see >>>and take advantage of it. >>>But what I wanted to point out, that also humans have "weak" points. The >>>knowledge of a GM is extrordinary, but sometimes he just doesen't see a hidden >>>response and makes an error. Although he calculates at least as deep or deeper >>>as the computer, he cannot take into consideration every possible response (in >>>time). Computer can easily do so, indeed the calculate also the most sensless or >>>maybe hidden threats. Also the performance of a human always oscillates because >>>of personal fitness. Therefore, the computer gets its chance against the strong >>>players to win Elo. >>> >>> >>>Let's say a 2600 Elo rated computer loses 1 or 2 games in 20 against a 2100 Elo >>>player, but wins 11 : 9 against a 2600 human player. On the other side, a 2600 >>>Elo GM loses 0 games against a 2100 opponent, but loses 9 : 11 against another >>>2600 GM player. >>> >>>Computer: >>>1. match: E = 2600 + 10*(19-20*0.98) = 2594 >>>2. match: E = 2594 + 10*(11-20*0.5) = 2604 >>> >>> >>>1. match: E = 2600 + 10*(20-20*0.98) = 2604 >>>2. match: E = 2604 + 10*(9- 20*0.5) = 2594 >>> >>>So, through losing 1 out of 20 games against the 2100 player the computer loses >>>nearly no Elo, but it wins Elo by managing a victory by defeating the 2600 >>>player. In the end it ends up with more Elo than the 2600 Elo GM losing against >>>his GM friend. >>> >>>Only because the computers lose some games very clearly because of lack of >>>knowledge, I would not rate them as very weak. >>> >>> >>>>Just anothother point. If you have studies stats you must know that your former >>>>sentence is wrong. That if I put the progs on 2400 I must also put the best >>>>humans on 2400. This is nonsense. >>>> >>> >>>Maybe there was a misunderstanding here. I wanted to point out that both humans >>>and computers have their "drawbacks", and it's not justified to just "downgrade" >>>one of the two groups to the 2400 level. >>>Actually, if a human gets outpowered by a computers tactics, its not right to >>>say he played like "2400" because a 2400 player would have lost more than ever. >>>And similarly, if the comp doesen't handle a position correctly, it's not right >>>to say he played like a "2400", because most often even a 2100 would have done >>>better. Let's just say the played not at the best. Computers can very easily be >>>improved, but can humans? :) >>> >>>regards >>>Andreas >> >>I must admit that I am slow. With your reply you gave me a good lesson for a >>phenomenon I underestimated. Now I know why it's so difficult for inventors or >>scientists with new theories or whole systems have so many dufficulties to >>pesuade collegues and people. >> >>I put it into a short sentence. While I am talking about the future and perhaps >>a better system in comp-human games with view on tournament chess NOT new shows, >>you take every existing possibility to misunderstand me in my views on the >>possible future with mainly claiming the factual existence of the present. I see >>not a single idea that _you_ could add to my ideas. All what you have in mind is >>that I must be wrong because I seek a different reality than the present. >> >>Here in your answer you make a basic fault (about the possible development of a >>different chess attempt when it goes against comps) and then of course you are >>correct. What is that fault? >> >>You see the actual stats and take it at face value. What does that mean? You >>pretend that the actual play (also by GM) is the final reply on CC. But exactly >>this is wrong! Anticomp strategies are just a start in the direction if a >>"fierce competition" would begin some day in future. Of course you are correct >>without such a perspective. You see what I mean? >> >>And then: you did not answer my question! >> >>I asked you: Could you explain why a human like a 2100 player could beat a 2700 >>comp IF a 2100 human player could never win against a GM with 2700 in a serious >>game! Please answer that question. Just do me a favour. >> >>Rolf Tueschen > >Player X is rated 2100, Player Y is rated 2700, that is a 600 point difference. > Given the rating difference, Player Y (2700) should win a large percentage of >the games, but not all the games. > >If Player X (2100) could get Player Y (2700) to play 1000 games, then Player X >will win/draw a "very small" percentage. Also, by the end of the 1000 games, >Player X will have learned a lot (hopefully) and will be a lot stronger player >(higher rating). The problem is that Player X (2100) never plays Player Y >(2700), so the chance of Player X (2100) winning/drawing a single game or >improving from playeing Player Y (2700) is nil. :) That was my point. I want to assure you of my respect for your apparent good understanding. The point is that normal people simply don't understand statistics. The assumtion that a 2100 could EVER beat a 2700 is completely foolish. That is the prob with maths. It's very important to define the range of reality where you want to claim validity. But again, if completely drunken or being corrupted with money even a 2700 human player might lose a SERIOUS championships game against Rolf! So not in a simul! But I didn't find out yet to get into such a serious competition. Could I disguise into a woman? How many chances I had with that specific task? Rolf Tueschen > >However, Player X (2100) can buy a (2700) program/computer and play it all the >time, learn weakness, get stronger against this one opponent and then reset the >program to initial settings. Then Player X (2100) can play a match and score >higher (200 to 300 points) than the delta (600 point difference) rating >predicts. > >If Player X (2100) is smart, then buy the (2700) computer/program and improve >your game with it, after a time you will be a better overall player and earn >those extra points. Just my opinion. :) > >(2700) program/computer = DF, DJ, CT, Hiarcs or Shredder on a 2Ghz Intel or AMD. > >Best Regards, >Chris Carson
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.