Author: Rolf Tueschen
Date: 10:04:01 07/22/02
Go up one level in this thread
On July 22, 2002 at 10:22:42, Robert Hyatt wrote: >On July 22, 2002 at 09:42:50, Rolf Tueschen wrote: > >>On July 21, 2002 at 23:45:54, Robert Hyatt wrote: >> >>>On July 21, 2002 at 14:40:35, Ed Schröder wrote: >>> >>>>On July 21, 2002 at 08:05:33, Uri Blass wrote: >>>> >>>>>On July 21, 2002 at 07:32:32, Geo Disher wrote: >>>>> >>>>>>OK after 56 hours the evals are exactly the same .88 for axb5 and Qb6. >>>>>>Hopefully in another few days axb5 will surpass Qb6. >>>>> >>>>>It is not clear if axb5 is better than Qb6. >>>>>I believe that it is not better. >>>> >>>>Exactly the reason why Kasparov became suspicious. >>>> >>>>Ed >>>> >>> >>>I suspect we can find _many_ positions where Kasparov made a move that >>>was inferior. I saw him do it several times in match 1 against DB in >>>fact. So I don't quite understand why _he_ thinks that his analysis/ >>>opinion is so infallible that because _he_ believes Qb6 was better, it >>>actually was. >> >>This is again something for the private tutoring. Lesson 60. >> >>Bob's logic says that both make mistakes. DB2 _and_ Kasparov. And therefore >>Kasparov has no right or extra-right or simply the status to declare or pretend >>that he has a higher position to judge about chess variations. Although Kasparov >>is the best player, actually. > > > >No, it has _nothing_ to do with judging positions. It has to do with the >question "can a computer choose axb5 over Qb6?" The answer seems to be "yes", >whether that is the right move or not being 100% irrelevant. My program is >getting closer and closer to changing its mind, iteration by iteration. After >a very deep search Fritz says the two moves are _identical_ in score. That is >all that is needed here to answer that question. If the two moves are equal >at some deeper depth, then serendipity could cause _either_ to be played. And >since Qb6 started off _higher_ and slowly dropped, while axb5 started off >_lower_ and slowly climbed, then it doesn't take much to conclude that if the >experiment is continued, it is likely that axb5 gets better than Qb6. So, chess is basically about serendipity and one-dimensional de- or increase? What if after even a deeper chapter the whole trend is reversed? What then? (NB that we could only know this in 40 years when computers are able to go that deep!) > >Note that this is _exactly_ what the DB2 log showed too, although it didn't show >the score for axb5 until the last minute, it showed Qb6 dropping each iteration >until axb5 finally popped out as just a _little_ better. You know what? If DB2 were not deconstructed Rolf would test a few positions to decide whether DB2 really was capable of to deny the present of three pawns. See our teacher Ed! > >That _certainly_ suggests that axb5 would be expected from a computer as fast >as deep blue, since others would play it too given enough time. No conspiracy >or mystery there, except for those that _want_ a mystery... The air is a bit hot though... (To the young readers: Hot air is a metaphor for starting wild guesses or fantasies.) > > > > > > >> >>True logic: we must at first introduce the parameter of overall chess strength. >>Here Kasparov is leading the ranking lists. So, there is a direct connection >>between chess strength and the quality of judgements about moves or lines in >>chess. Now let's take a look at DB2. Except the 6 games from 1997 we have not a >>single gamescore of the practice of the machine. The first game of the show >>event reveiled that DB2 was as weak as typical machines. Some moves were >>absolutely nonsense. The main line leading to its loss wasn't foreseen, which is >>typical for machines. > >Please point out a move in game 1 that was "nonsense". Kasparov never found >one and mentioned it. He was, in my opinion, quite lucky to win that game, >as DB _did_ win material. Give me some minutes, I will check that in my bases. > > > >> >>Verdict. Kasparov is the far better player than DB2. While DB2 is or better was >>a good calculator, so that no amateurs were able to play it successfully, but >>since its understanding of chess is infantile a good GM with eidetics and good >>calculation is far better. If it comes to "judging lines" Kasparov is of >>outstanding class compared with the idiot savant DB2 who must rely on the >>telephone book like databases, features called 'forbidden' in human chess. >>Chess is more than calculating till the point of definite blindness. Chess is >>knowledge and experience. "Eidetics helps, but without the chess genius eidetics >>is simply computerchess" (Tueschen July 22th, 2002). > >And your verdict means _nothing_ when it is about tactics. Examples: Who >resigned in a tactically drawn position? Who overlooked tactical drawing >facilities in game three, thinking he had it won? GMs make as many tactical >mistakes as anyone else, and given the choice of taking a GM's opinion or a >computer's (after a long search) I will tend to go with the computer's unless >I see some odd characteristic of the position that might make the computer >error... Your "lecture" about who is the best is pointless. Does Kasparov >_ever_ lose a game vs a lower-rated player? (hint: Krmanik for one, there >have been others recently). Therefore, better != perfect, by _any_ stretch. You are right, but you must think tactics in the overall context of a game. There's no tactics isolated as in the puzzle books. The reason for errors is of multiple character. > > > > > >> >> >>> >>>Fritz seems to be exposing that as false. Had I run Crafty long enough >>>it would also probably have liked axb5 since the scores of the two moves >>>were heading in opposite directions, albiet a bit slowly. >> >>For all I know, you would have run this test _if_ it had been sure the result >>would have shown what you expected... Concerning our general addiction *time* >>should be no obstacle! > > >"for all you know" doesn't cover very much ground. Why do you think many of >us spent > 24 hours verifying that Kasparov resigned in a drawn position? >Because we simply wanted the _truth_. And how comes that you didn't think about Rolf's Law of the non-comparability of such different machines. Or did I miss something you yourself explained, that DB2 wasn't just commercial progs plus speed. So, excuse me, I always thought it was Bob's Law. Now you've a problem again. Why suddenly your Law is no longer true? Because you want to imitate DB2 with commercial progs? To minimalize the ugliness of the deconstruction of DB2? >Ditto for qb6 vs axb5. And the truth >seems to be what we all (except for a few) expected. It was logical, >predictable, and repeatable by other programs... But I repeat the question why you hadn't time enough to wait for final answers? (Hot air in my eyes.) > >If it hadn't been, we would have needed to search for _other_ reasons why DB2 >might have chosen Qb6. But, we don't need to now, because we _know_ why it >played that move. Objection as before. You can't compare bananas with asparagus! > > > > >> >>BTW do you know where the game scores are from the 10:0 tests of DB2jr or sen >>against some commercial progs? Just take your time for the search. Thanks. >> >>Rolf Tueschen > >This has been answered by them. They are in the same place as the logs for >the games I played between Crafty and Cray Blitz last year. I didn't keep >them because at the _time_ I didn't consider them important... Neither did >they... I see, Hsu or them read CCC but they do not post, so that you post for them, and they and you as their speaker didn't see the questions teacher Ed had? Is this your understanding of serious and true science? Istn't this the confession that it was about something different than the truth? Rolf Tueschen > > > > > >> >> >> >>> >>> >>>> >>>>>Uri
This page took 0.02 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.