Author: Hans Gerber
Date: 17:27:54 05/10/00
Go up one level in this thread
On May 10, 2000 at 18:46:29, Robert Hyatt wrote: >On May 10, 2000 at 15:54:34, Hans Gerber wrote: > >>On May 09, 2000 at 21:56:15, Robert Hyatt wrote: >> >> >>>That is painting with a very wide brush. As I have pointed out on multiple >>>occasions, what you are asking for is impossible to provide. Within the realm >>>of practical solutions. You _could_ lock DB, the operator, and Kasparov inside >>>a steel vault, with no windows, no external cables, sufficient shielding to >>>make certain no form of magnetic radiation could penetrate to the interior. >>> >>>_then_ you could be reasonable sure no outside influence was being used. But >>>I don't think _anyone_ would consent to playing under such conditions. You >>>could not have light in the room, unless it was battery powered, otherwise >>>some sort of modulation could be used... >>> >>>So preventing cheating was not doable. >>> >>>Providing some sort of 'audit' capability is therefore not doable, by >>>simple induction. If a human could influence the game, it could influence >>>whatever 'log' was produced at the same time, to make the influence >>>undetectable. >>> >>>I don't see any way to prevent such. Nor do I see any reason to spend a lot >>>of time trying to prevent such. If you don't trust your opponent, don't play >>>him. After all, it _is_ a "game"... >>> >> >>You're talking about something I didn't write. >> >>Sure it's a game, but the moment we're talking about a machine as the best >>player of the world, we should at least have some certainty that the machine was >>playing on its own. No matter how difficult that could be "doable". Perhaps it >>might _not_ be doable... but then can't you see the consequences? >> >>Kasparov explained that one single intervention from outside could change the >>character of a whole game. Hence such a thing must be made impossible. >> >> >> >>> >>> >>> >> >>> >>>We will just have to agree to disagree. One classic case from my background. >>>In 1984, we were playing "NuChess" at the 1984 ACM event that year, using a >>>4 cpu Cray XMP. We were almost lost, and played a very odd/ugly looking knight >>>move. Our opponent immediately thought he could win either of two pawns, an >>>outside passed a pawn, or a center pawn that would wreck our pawn structure. >>>He went for the center pawn, and that knight on B8 was used to take his last >>>bishop (a trade) leaving us with a passed pawn that could not be stopped by the >>>opponent's king. >>> >>>We were _never_ able to reproduce Nb8 again. And we ran it literally hundreds >>>of times. So did the program play that move on its own? Or did it have help? >>> >>>_I_ know the answer, as does Harry and Bert as we were all sitting there and >>>watching, and we _all_ criticized the move as silly. Unless you assume the >>>opponent doesn't know about 'the square of the pawn' and goes for the center >>>pawn rather than the more dangerous a-pawn. >> >> >>Do you expect me to doubt your own experience? No, I have a lot of respect for >>you. But I must also say that I don't understand what your example has to do >>with our question of control. >> >>I think I understand your logic of reasoning. You want to say that such an >>extreme execption would destroy all attempts of an objective control because >>reproducability is most important to prove non-existence of outer influence. It >>seems as if you are blinded by your own great experience. >> >>I'm not on your height as far as that class is concerned. But from a scientific >>view this doesn't matter. The basics of science have to be respected everywhere. >>If your machine's output, which is the inside process as you had explained, is >>sufficiently controlled then it doesn't matter if a certain move can't be >>reproduced. All that is important is that the logfiles look "a bit" different in >>that case of the exception. >> >>Just for the general understanding, would you agree that the logfiles had to >>contain some data that could prove that the move Nb8 was "somewhere" in the >>thought process of the machine? > > The program liked the same move from ply 1, until ply 10. Right at >the last minute, we saw a line like this: > > 10 >>0 Nb8!! > >my move: Nb8 > >the >>0 is a simple fail high notation we used. Nb8 showed up _nowhere_ else >in the output. And crafty behaves the same in many positions. It changes its >mind at the last second to a brand new move, with no prior mention of that move >at all. How do you prove it wasn't a human's suggested move if the program >won't reproduce it? Ok, I'll play the game. How about the suspicion that you have implemented such code that in certain (lost?) positions you go for an _idea_, sort of last chance. _If_ the opponent falls you'll have a lucky punch. Something like that? Frankly, I don't see why this move should speak for cheating. Are you sure that the whole code stood under your control? Not possible that one of you three had tried something extraordinary? Question to you, was it possible, if a masterplayer was somewhere, to influence CB? (Sorry, you already gave the answer. Of course it could have been possible. So, how about Harry or Bert? Excuse me if these names don't say too much to me. The other members of your team? Or you and the opposing team? And final question, honestly, did you go through some bad moments of suspicion after that game?) > > > >> Or do you want to say that such a move might >>have come "out of the blue sky"? In that case let me give you a hint; just >>implement a code that says "machine in the state of genius" and you have saved >>the orthodoxe situation again. :) >> >>> >>>How do you 'audit' that? I have seen several _other_ cases where a move was >>>never reproduced. That one one for us. In 1987 in Orlando, we had deep thought >>>beat, but rather than playing the move that absolutely crushed him (he had >>>already failed low pondering that move and we had failed high searching the >>>right move) CB switched to a new move at the last second, one it apparently >>>thought was better. It lost quickly. We could _never_ reproduce that move. >>>And I ran it over 1,000 times during the next month, burning up a couple of >>>Crays every night trying to make it fail again... >>> >> >>You are certainly a case for interdisciplinary studies. I don't think that you >>have to believe in the supernatural... I don't understand why you have such a >>big problem with that incidence. This has nothing to do with cheating of course. >>It seems to be a philosophical problem. Deterministic expectations and such. > > >Nothing to do with cheating, but everything to do with showing that you can't >reproduce normal behavior multiple times. So how could you _ever_ validate an >experiment as you might in chemistry or physics? > >Computer Chess is different... I must totally disagree. In social sciences you still have instruments of control although you can't repeat the experiment. If you have implemented the possibility of a "lucky punch move instead of resigning" I don't see why you should not have control over your output. I'm too little experienced in the technology of 1984 but perhaps it had something to do with ... ... over-heating? :) Excuse me, I don't know the answer. > > > > >> >>> >>>There is a long distance between "in theory/principle" and "in reality"... >> >>Yes, you'll have to find a practical solution to guarantee a pratically doable >>control. >> >>> >>> >>> >>> >>> >>>> >>>> >>>>My point is that preventing 'crime' is _impossible_. Otherwise, after a couple >>>>of thousand years, banks would no longer be robbed. Web sites wouldn't be >>>>broken into. Computers wouldn't be vandalized. >>>> >>>>There are some things you can _not_ prevent. >>> >>> >>>Sorry for following up my own post. I had to make a quick exit earlier to go >>>fix a computer problem... >> >>This is another such problem. If we could find a certain consense between our >>formerly so different positions, it could help as such. I take for granted that >>we here don't want to see the other man busted, the way Fischer wanted to see >>his opponents ego crush. We are dealing with a most important problem for >>computerchess in general. >> >>(BTW in the case of the aftermath of the second game Hsu surely should have >>given K. the prints. So he had given him something that hadn't proven anything >>but a friendly connection. In that case K. might have won some game more, if he >>could have taken advantage of some details, but he overall research would be >>still going on now! In the long run computers will be stronger than man anyway. >>So why hurry up? Hsu did a really bad job. He "won" and at the same time has >>lost his field of research! He must even face attacks of being a cheater. >>Justified or not.)
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.