Author: Robert Hyatt
Date: 10:18:48 11/15/98
Go up one level in this thread
On November 15, 1998 at 09:48:39, Robert Hyatt wrote: >On November 14, 1998 at 13:25:02, Amir Ban wrote: > >>On November 12, 1998 at 16:33:46, Bruce Moreland wrote: >> >>> >>>On November 12, 1998 at 08:30:20, Robert Hyatt wrote: >>> >>>>As I said before, you simply *DON'T* accuse someone of cheating unless you are >>>>*certain*. And if you yourself have been caught cheating in the past (can you >>>>spell Polgar?) then you *really* ought to be careful with the accusation. The >>>>Be4/Qb6 issue *could* have been explained if he had investigated... Because >>>>both Bruce and I played with this and found that Qb6 doesn't win a pawn as he >>>>so boldly and forcefully proclaimed. And if he had invested that time, none of >>>>this would have happened. Or if he had simply kept his mouth shut *until* he >>>>had time to properly investigate, he would also have not said anything. But he >>>>just opened his mouth and inserted his foot promptly... >>> >>>I looked at some of these positions. I don't remember which ones. I know for >>>sure that I looked at that position where DB played Be4. Mine didn't want to >>>play Be4 in a day or two, but it was moving up the list of candidates, and the >>>score for the move it wanted to play was at or near zero. >>> >>>I don't really care if the move won a pawn or didn't win a pawn or lost the >>>game, or whatever, other than minimally. What I was interested in determining, >>>and what has been determined to my satisfaction is that a program could be less >>>optimistic about these positions than all of the people who were watching >>>thought it must have been. >>> >>>It is my contention that during that point of that game, DB spent a lot of time >>>because it had been optimistic up until then, but was beginning to realize that >>>it was going to be a draw. I have no evidence to support this other than it >>>took a long to move, there have been references made to "panic time" in the >>>newspaper, and that mine experienced a score drop when set to think for a long >>>time. >>> >>>I think that DB made a human-looking move out of sheer frustration. >>> >>>I think that a cheating accusation could come out of this is insane, and that it >>>could be supported by anyone in the computer chess community is even more >>>insane. To suggest this is not just to say that one person suffered a serious >>>moral lapse, it's to suggest that the whole team was morally bankrupt, since I >>>assume they would have had to conspire to do this, and in order to execute such >>>a stunt in real-time they may have even had to organize this beforehand. >>> >>>That is an incredibly serious thing to suggest based upon, 1) incomprehensible >>>output (everyone raise your hand who has written a program that produces output >>>only understandable by you); 2) extra time added to a search, which we all do >>>for one reason or another, and it's easy to determine that there could be a >>>reason to do so here; 3) the accusations of the opponent, who is not a technical >>>person, and who has no way of knowing what the thing was capable of *without* >>>cheating, although he might think he does. >>> >>>If DB every plays again I would hope that they would make their output more >>>readable in case this happens again. >>> >>>bruce >> >> >>You are on the wrong page here. You are not even answering the right questions. >>You may have been influenced by Bob, who's trying to portray this as a primitive >>accusation by people who can't understand simple facts or draw easy conclusions. >> >>Briefly: >> >>- You discuss the wrong move 37.Be4, while the problem is centered around >>36.axb5. In both these move Qb6 is the apparently right move, but in move 37 it >>doesn't win a pawn, so Be4 is a plausible choice. In move 36 Qb6 definitely wins >>at least a pawn (and axb5 doesn't), so the question is why would a computer not >>play it. You will find, I guess, that yours like any other program wants to play >>36.Qb6 and can't be persuaded not to. If you find otherwise, tell us because >>that is news. I had to go back a bit to check this one... but as I thought, about the time of that match "crafty" would also play axb5... but it was because of what I considered (at the time and even now) as an overly excessive king-safety evaluation. This was a warning I should have studied or I would probably not have done so badly in Paris... because Crafty was analyzing during that game and it *expected* axb5... so when they played it I didn't give it a moment's thought... Later I retried this (when we were doing the "crafty goes deep" stuff) and found crafty wouldn't touch axb5 after the king safety was brought back down and under control again. Hsu memtioned in the 10 game match that most of the games were tactical busts because they seemed to evaluate king safety much better than the micros they were playing against (this reported by someone that went to one of his or Murray's talks). So it is certainly "possible" that they made this positionally, whether it is right or wrong is certainly debatable. It probably leads to a draw, since it was played in the game and the game definitely was drawn at the end, which is not far off from that move... >> >>- Making this as a "human" over-valuation of a positional advantage, which is >>straightened out by a program like Ferret running overnight doesn't do justice >>to chess intelligence of the people involved. The only possible reason NOT to >>play 36.Qb6 is that black can toss 3 pawns for a speculative king attack. >>Kasparov was aware of this during the game, and was of the opinion that this is >>beyond the understanding of today's computer chess. So far he's been proven >>right on this for every program except DB. >> >>- I don't know where you got the impression that the suspicion was based on not >>being able to read the printouts. I'm only aware of the case that I made out of >>the printouts, so I guess you are talking about me. The printouts are a mess, >>but I can read them. The fact is that reading them doesn't remove the >>difficulties, but only raises others. The answer to these is definitely not >>within the printouts. >> >>- Same about the extra time: neither I nor anyone else ever said that the fact >>that DB spent a lot of time on some moves indicates cheating. There is no extra >>time in move 37 that you discussed, so what you say about it is not relevant. >>There is panic time in move 36, but that's not what is suspicious, but the fact >>that there are anomalies in the timing behavior, which cannot be explained by >>the printout alone. >> >>- The part about a "non technical" person: Since I'm behind most of the case, I >>think I am qualified, but I would argue that Kasparov alone is qualified enough, >>and you are really assuming more naivete in him than you actually can. In any >>case, the way he suggested of settling this, of trying this for days and weeks >>on every other program, is a perfectly valid suggestion that could have been >>made by an expert. I would guess you and Bob, being programmers, consider >>yourself more qualified than Kasparov to make such complaints, but since both of >>you have so far failed to get the chess-related problem involved or even to get >>the move right, that may not be so. > > >I believe that I caused the "move confusion". Because move 37 was the move >that apparently sent Kasparov into a "tizzy"... (no computer would play that >move... I believe that only a few top GM players like Kramnik...). The output >I saw did in fact come from move 36 and not 37 after going back to look at it. >So I take responsibility for the confusion about "panic time"... I will start >a *long* test on this position as well, searching both moves iteration by >iteration to see what happens to the evaluation of both. But I have absolutely >no doubt that either Qb6 will drop steadily to the point where axb looks >better, *or* as in the game between Nuchess and Cray Blitz, at the 1984 ACM >event in Los Angeles, where we played Nb8 for absolutely no discernable reason, >we attribute this to a simple multiple-cpu anomaly. In that game, Nb8 allowed >Nuchess to take the "wrong" pawn and allowed CB to trade off all pieces on the >board and end up with an uncatchable pawn... But *only* if Nuchess took the >wrong pawn. We were *never* able to reproduce that move again, and we ran it >30+ times to see. When we were accused of cheating in 1986 we were quite >fortunate that this didn't happen in *that* game, because a parallel search is >most definitely non-deterministic in nature. But in any case, *either* of those >explanations is *highly* plausible... And since I *know* of such happenings, I >would certainly say that I am far more technically qualified than Kasparov or >yourself, since I've been working with a parallel search for almost 20 years now >and have seen such funny behavior frequently. "Cheating" is certainly not the >*first* explanation I would turn to. It would likely not even be the "last" one >as there is simply *nothing* to suggest cheating, whether you like their output >or not. > >But two moves in a row that Kasparov didn't "like" and can't reproduce with >another "micro-based" program is not exactly "evidence"... And it isn't enough >to produce or justify the vitrol in Shay's "article"... > > > > >> >>This is brief. I have no intention of detailing all the case here. I've done >>this in the past, several times, and by now I've more or less posted all the >>information I have. Anyone who was sufficiently interested must have been paying >>attention, and anyone who was not paying attention, must not have been very >>interested. >> >>So are we insane ? Kasaprov raised this, and he believed there was cheating, >>based on chess issues and personal impressions. I was asked to give opinion on a >>technical document (the printouts). I would have told him that the printouts >>don't support his theory if that would have been the case (say if the issue was >>really 37.Be4), but I couldn't. My interest here is to settle this issue either >>way, and I don't care which way. What I find surprising and inappropriate is >>that the story ended here. > > >what I find *most* inappropriate is the "guilty until proven innocent" approach >you seem to take. I'm used to the opposite. > > > >> >>Was there intervention ? This question is going nowhere because IBM is not >>cooperating. I think anyone who saw what I was asking would have agreed that if >>a good answer exists, it could have easily been provided by IBM, and the case >>would have been closed. This did not happen, and instead the question has >>become: why shouldn't this be diuscussed and settled on its factual merits ? > >Perhaps for one move out of maybe 300 total, there is *no* explanation? I >certainly point to the 1984 game above as but *one* such case we saw. There >were many others. Look up the 1987 or 88 ACM event in Orlando and check out >the Cray Blitz vs Deep Thought game. At one point Hsu was expected Ne6 with >a big fail-low on his side. We had predicted Ne6 at the previous move and were >positive in our eval. Right at the end, we went from Ne6 which was winning, to >Na4 (or Na5, don't recall exactly, that long ago) that ended up losing. We ran >this search a couple of dozen times after the game, and could *never* get it to >play anything but Ne6. Cheating? To lose the game? Or just *another* parallel >search anomaly? If we cheated we were certainly stupid to override a +1.9 move >with a +.002 move... I should add that we "failed high" on Na4, but after playing it and starting our ponder search the score died... > > > > >> >>I think IBM showed here zero accountability and responsibility. Whether or not >>intervention took place, I'm disturbed by this: In hindsight they certainly >>could cheat AND get away with it without sweat. >> >>A few days ago I went back and reread the text of the post-match press >>conference and commentary. A year and a half later, this makes interesting >>reading. I was surprised that the printouts were discussed so much, and I >>realized that in fact everyone expected publishing the printouts to resolve >>this. Garry says so. C.J.Tan implies that the printouts will show this. Then he >>backs off and says IBM has still to decide which printouts to publish (so as not >>to confuse the public :)) Garry protests immediately, and asks specifically for >>move 35 and 45 of game 2 (he never got them). In any case, C.J. says that anyway >>all posible information will soon be in journals and books (so why worry about >>some printouts now ?), and of course IBM is busy doing science and do not want >>to compromise itself for future matches. It should come across that IBM was >>trying hard not to disclose any information, and the second best thing they >>could hope for was that the information would prove to be unreadable. You have >>to take this into account when considering what Hsu and you said about not being >>able to read the printouts. >> >>If you consider this in the timeframe in which it was said, you realize that all >>the parties involved including Kasparov did not know that this was absolutely >>the last time these issues could be raised. IBM were not around for further >>discussions after this, and they just closed the shop and went home. >> > > >This is certainly how I would have handled it. Remember the old proverb "you >can catch more flies with honey than with vinegar"?? Kasparov only offered >vinegar... And got no flies as a result. Now why he can't figure out "why?" >is anybody's guess. But it was predictable... > > > >>By the way, in the commentary part I can see that the commentators understood >>the chess-related issue Garry was raising. Mike Valvo is perfectly right when he >>says king-safety evaluation may explain it, but he doesn't yet know yet how >>much. >> >>Reading this, C.J.Tan strikes me as insincere, evasive and hypocritical. He was >>certainly running this show, not Hsu. Hsu, judging from the interview by Keith >>Ian Price, sounds a bit out of the loop, and hardly interested. If all this >>depended on character assessment of C.J.Tan and Garry Kasparov, I think I would >>choose Garry. >> >>Amir > > >I wouldn't go that far. This was a "big deal" even to IBM. Not because of that >silly "their stock prices went up 20%" nonsense, but because it *did* produce >a *lot* of good P/R at no costs. From CNN to network TV to every newspaper in >the world just about, to every major technical publication like Scientific >American and so forth. I suspect the results were worth at least a couple of >hundred million dollars, not in stock increases or increased sales so much as >the simple "free publicity on a grand scale..." And when something gets "that >big" then higher-ups within the company very likely took over to maximize the >benefit. And I doubt very much they'd be interested in releasing info to >Kasparov after he acted as he did...
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.