Author: Robert Hyatt
Date: 12:02:38 01/31/03
Go up one level in this thread
On January 31, 2003 at 14:22:48, Rolf Tueschen wrote: >On January 31, 2003 at 11:05:56, Robert Hyatt wrote: > >>On January 31, 2003 at 07:56:58, Rolf Tueschen wrote: >> >>>As a careful scientist I can present the following results. The details of my >>>method must remain secret, but you are invited to read CTFfor example. >>> >>>The actual program against Kasparov for the first time in history played for all >>>the psyche of a concrete human opponent. We know that Kasparov believes in >>>magic. Numbers are very important for him as symbols for something coming from a >>>hidden world. So in consequence Kasparov believes in the super-natural of chess. >>>Now what DEEP JUNIOR has done in game three is giving Kasparov the perception of >>>a position that is completely lost for the computer side. In front of a castled >>>King Kasparov saw two Knights on f6 and h6. Not enough, he had an open g-file >>>against such a configuration! And his own King could still castle to the Queen's >>>side! Three officers were directed against Black's King-side. Queen and two >>>Bishops! The black King might have felt like Israel in front of the Arab World. >>> >>>But did Kasparov EVER have such a winning position against a human opponent? Of >>>course not because only patzers would play like that. And against patzers you >>>don't need your best chess. Here is the secret of the actual design of the >>>Israeli computer program. What would happen if Kasparov had to win such a won >>>position against precise calculations on the border of the allowed and possible >>>in chess? Is he prepared for such a challenge? Of course not! >> >> >>You are making one assumption that may turn out to be faulty: "The position >>was winning for white after g4 Nxg4". >> >>It looked dangerous for black. But "looks" don't win against a computer. >>Against a human, black might well have "folded". Just as surely as Kasparov >>folded near the end of the game. But a computer generally won't, and during the >>game no computer ever thought white was up by as much as a whole pawn. So it >>might just be a case of something looking dangerous but not really being >>dangerous. >> >>Computers are known for their ability to handle such positions very well, and >>the inherent problem in such positions is that quite often, there is a very >>fine line to walk as the position is played by both sides. Anytime you put >>a human in a position where he has _one_ good choice, and _lots_ of fair to >>bad choices, for many moves, the probability of a single mistake goes way up, >>and what we saw in game three happens. > > > > > >How could I contradict you here? You are right. But I think that this is a >general computer advantage. Only - in this game here White is better, almost won >(among humans). Ok,please help analyse the position Michael gave here below. If >you also doubt that human GM had winning chances in won positions, then what >could they possibly achieve in drawn or worse situations? Ok, it seems as if you >took the show event at face value, which I would never do, but you should NOT >contradict your own statements, namely that comps are still not GM, or at least >super GM strength. A couple of things here. 1. I _am_ rethinking my "position". But I am beginning to lean toward a _totally_ different conclusion than you might expect. Namely that "super-gm players" are _not_ really super-gm, they are simply able to bluff/out-think the lower-rated players. But when it comes to a computer, it can play multiple moves that are very ugly looking, but the human can't deliver the fatal blow, unless he does it _quickly_. The longer the game lasts, the greater the probability of an error by the human, and the game goes south. It is clear that for a general "chess skill" the computers are _nowhere_ near the top GM players. At least in terms of analyzing a position to produce the best move. But fatigue is a bigger issue that I would have guessed, as can be seen from the Kramnik match and now the Kasparov match. The human seems unable to keep up the mental sharpness needed for 3-4 hours at a time, and one daydream later the game is over, as we have seen. In game 1, Kasparov blew it out quickly before tiring. In game 2, he was in a worse position but found a deep tactical plan and out-thought the machine to a draw when he might have lost. In game 3, he created a good position that the machine had no idea about what was happening, but when there was no quick kill, "meat makes mistakes" returned to haunt him and turn a simple draw into a complicated loss that he didn't even want to fight to a conclusion... So I am perhaps redefining my definition of "GM strength" to have a "weariness" component. Since a computer has no such problem, the longer the game, and the more complex it is, even if the machine is losing, it will probably draw or win due to the "weariness factor". Something I had not considered before. We knew it was a factor for long matches... ie karpov vs kasparov for the WC years ago. But now it is becoming a problem in a _single_ game... that is interesting. >And we two at least should agree here that Elo is nothing of >"real" strength but only performance result. However if that is true then you >must consider that show events do not reveil the maximal strength of human >players against comps. That could only be tested under tournament conditions and >that is exactly what your collegues don't appreciate. They _know_ and you too >know that comps would be torn into pieces if humans could only take some >continual periods of training=preparing. Against that desaster you have the >tradition of always producing new versions, of twisting and tweaking. Ok! What I >am talking about is what rests at the base beyond all these tricks or better >traditions in CC. Let me ask you a simple question. Would you bet a fortune that >you could confuse all human GM _all_ the time and _forever_ right now or would >you admit that human GM would kill the actual programs in the longer run? I think "the longer the run" the worse the human is going to do. Based on the "weariness factor" that has become more apparent. > I am >only interested in that question. Because the meaning is my theory that all the >success in CC against human players results mainly from inexperience of the >human side. However that could be changed. But what you couldn't change is the >actual weakness of your creations, at least for the coming three decades. Of >course that will change in the end. I never doubted it. But not yet today. So, >wouldn't it be more honest to renounce all imposterdom? Itlooks so naive. I >don't mean you in person, but >CC in general... > >Rolf Tueschen > The programs will continue to get better. To the point that a GM can't even match wits with them on a single position to find the best move. But it seems that sitting down and playing for four hours, against a strong program, is getting to be more than the human mind can handle... > >> >>Ng6+ was a solid drawing move, but Kasparov either (a) missed it (which seems >>unlikely) or (b) he thought the rook move gave him winning chances, without >>enough time to really analyze carefully. Whichever reason really doesn't >>matter that much. If you are the world's best "minesweeper" you still take >>a chance every time you walk on to a minefield... >> >> >> >> >>> >>>So this is the answer how David could still beat Goliath. Big super powers have >>>to control a huge traffic of their own while little David must only concentrate >>>on the strategically weakest spaces and entities of the enemy. Perhaps we have >>>seen the birth of a new chess pattern. After the famous Nf8 position that often >>>can defend the whole Kingside for Black we have now the Nh6 position. This is >>>chess of the third thousand. It is worth more than three times Las Vegas. >>> >>>Rolf Tueschen
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.