Author: Ed Schröder
Date: 00:50:49 04/14/98
Go up one level in this thread
>Posted by Amir Ban on April 13, 1998 at 19:47:57: >In Reply to: Re: Proposal: New testing methods for SSDF (1) posted by Ed >Schröder on April 13, 1998 at 15:12:46: >On April 13, 1998 at 15:12:46, Ed Schröder wrote: >>>The autoplayer is just a part of your program that implements an >>>interface, the auto232 protocol. It's your own code. There's no such >>>thing as an autoplayer that plays a game between two programs and is >>>independent of them. >>>Much of the debate over this is because this point is misunderstood. >>>Amir >>Because of all the excitement about the Fritz autoplayer I did the >>following experiment last week. When being in permanent brain I sent >>rubbish to the opponent. I used Hiarcs6 and Genius5 as an example. >>Both Hiarcs and Genius crashed. >>Two points for Rebel ??? >Neither the SSDF nor anyone else that I know counted a game that was not >completed because of a crash. >In any case, if the opponents of a certain program are in the habit of >crashing, the testers will be aware of that. >>Recognize and cheat the opponent even more.... >>---------------------------------------------- >>Didn't succeed yet because I can't find the auto232 manual anymore but >>I think this is all possible......... Now pay attention............ >>How to cook Rebel and other chess programs.... >>- Being in permanent brain, after 5 seconds send ALT_M (is force move) >>to the TILL THEN unknown opponent. >>- Do that 3-5 times in a row. >>- If you immediately get back a move: >>a) you *KNOW* your opponent is Rebel (ALT_M is proven). If the ALT_M >>trick fails try the "force move" trick for Genius, CSTAL, Mchess, >>Hiarcs etc. by sending their specific force move commands. >>b) you forced Rebel 3-5 moves on 5 seconds level which is a clear cheat. >>Do such tricks ONLY at 04:00 (GET_TIME) when everybody is sleeping and >>nobody will ever notice the cheat. >I've been in contact with Thoralf Karlsson about this. He asked me some >questions about autoplayers: >1. Can an autoplayer affect the style or level of the opponent ? >2. Can it affect the learning function or thinking on the opponent time >of the opponent ? >3. What else can an autoplayer do that will affect the result ? >I answered: >1. No way. An autoplayer can only do something that is in the >command-set of the opponent's autoplayer, and this does not include >changing levels, styles, memory sizes, thinking times, or anything of >this sort. >2. No, for the same reason as 1, but with qualifications below. >3. I thought about this, and gave him a list of 6 methods. Your >complaint is no. 4 in my list. Here it is: >- Lie about the result, or "resign" for the opponent. >- Lie about the clock. The programs count time for themselves, and the >opponent cannot enforce a correct clock. >- Pretend to hang in a bad position so that the game will be timed out >and ignored. >- Send "play-now" to the opponent to force the move, making it play >faster, and perhaps inviting a blunder. I told Thoralf several programs >disregard this command. Your post confirms this. So it's only Rebel ? >Actually Junior 4.0 would also honor this command. >- Don't send "Save-game" (this is up to the master in the auto232 >protocol). This should not have any direct effect, but as you say, in >Rebel it will disable learning of the current game. >- Send "Take-back" followed "Move-now", making the opponent recalculate >the move. If the opponent originally was thinking on the opponent's >time, on the second time it doesn't have this advantage. >The first three of these are not really connected to the autoplayer, >they could be done by any program in any game with automatic testing. >The last three are autoplayer tricks. I find them a bit far-fetched as >intentional behavior. Methods 4 & 5 only seem to work against Rebel. >To make any difference, the methods should be used on a large scale. >The problem with such tricks, if anyone is stupid enough to try them, is >that you get caught. If the testers somehow fail to see that this, the >autoplayer is in the hands of the SSDF (and Enrique, and others), and >they, if they choose to, can watch it in action and expose it. >All we need to know from the SSDF is that they have not seen this kind >of behavior. If you want, we can ask them to watch a match for several >hours and report if they saw anything. As I said, if anyone is stupid >enough to do such things, he has supplied a piece of evidence that will >repeat its crime in court in front of the jury. >You posted several times that you demand to decide yourself on this >issue. I don't think you have this right. We have the respected >authority of the SSDF, and other testers, to give us an answer on this. >If every result the SSDF posts for Rebel must have your approval, then I >for one would stop considering them objective and start disregarding >those results. I think with all respect to the SSDF and other testers only programmers are able to judge if an autoplayer is clean. I have given the example of sending a "force move" to the opponent doing that at 04:00 at night when every is sleeping. This is just one example. You are a programmer yourself and it is not so difficult to come up with more tricks to make sure nobody will ever notice. Right? I simply want to have to RIGHT to check MYSELF. I have agreed to the SSDF testing for Rebel. I want to know if my program is treated fair. I have a name to keep and so do others. Is that an unfair request? I don't think so. >>Amir, do we need a new auto232? >What we have now is workable. I posted before that we need to define the >auto232 protocol by consensus, and I invited Dr. Donninger to submit the >original draft. I would, for example, omit Take-back which is not >needed, specify that move-now can only be received as the first or the >last command of a game, and so on. I very much want to see "resign" and >"draw-offer" messages added to eliminate this primitive timeout, and I >want to add identification of the opponent to the protocol, since in >high-level chess you know who you are playing against. Identification of the opponent will lead to pre-cooked books tuned on that specific opponent by playing 300-400 games first. I think this identification proposal will only make things worse. >I suggested to Thoralf that they adopt these "transparency" practices: >- Publish all games. >- Allow programmers to produce log files, that will be kept by the SSDF >as a matter of record, and to make these logs available to the >programmer if a dispute arises, or on a periodical basis. This is a good idea! I plan this myself. Every suspect thing that might happen in a AUTO232 game such as strange keyboard input (other then normal moves) written to a LOG file. >Besides, who needs auto232 or the SSDF ? We already have a new list on >Ossie Weiner's web site. It's 100% pure and above suspicion. An Ossie >Weiner program tops the list, Fritz is demoted by a 100 points or so by >some procedure that was not specified. They even have Junior there, >though without a version number, and with a rating probably extrapolated >based on guesswork and set to be suitably below in-house programs. Who >needs auto232 ? Who needs games at all ? They seem to hold you in high >regard, Ed, so surely you will not withdraw from *that* list. Thorsten, >I don't remember if they list CSTal, but if you ask them they surely >will. No need to send a program. If they like you, they will find a high >rating for you, but don't expect to be no. 1. All well said Amir :) The list on Ossi's web site is one big laugh as it is based on NOTHING. What do you want me to do? Ask Ossi to remove Rebel from that list? Would it help? And what if he refuses? So Ossi if you read this posting please remove Rebel from the list on your web site? >More than a new auto232, we need to have our good sense back. The only >thing I understood from this debate is that no one will trust anyone >else, and many are willing to say and do anything to discredit the >opposition. Maybe it is different? When AUTO232 was introduced by Chrilly every programmer personally was asked to join by the mouth of Helmut Weigel. We (the programmers) all checked the new software. We (the programmers) all agreed. We (the programmers) all supported AUTO232. We (the programmers) all released the software in public, free for the public and for ourselves (the programmers) to check each other if we stick to the rules of fairness. Until now every programmer has stick to these rules. Now Chessbase suddenly sets a new trend I can not agree on. Why should Chessbase be excluded from this obligation? >Enough was said already was said about Ossie Weiner's >statements, I find it discouraging that people like you or Thorsten >decided to join in, when you should have better sense. Despite all your >disclaimers, it's hard to distinguish between what you are saying. If >anyone reading this newsgroup concludes that Ossie Weiner and Ed >Schroder are engaged in a vicious attack on the SSDF, you have only >yourself to blame. So it would be better I keep my mouth shut? I know this "openness" will backfire on me. So yes from this point of view I can better shut up. Concerning a possible engagement between Ossi Weiner an Ed Schroder I say there is NONE. I was asked by Ossi to sign the open letter to SSDF signed by Richard, Chrilly and Stefan. I refused to sign it because I disagreed on a few points and certainly on the style of the letter which was my main reason not to sign. I have chosen to deal with the raised problem in my own way because what is happening right now is unacceptable. For the moment that means new released Rebel's will not compete in SSDF any longer until this case is solved. Which means EQUAL RULES FOR EVERY PROGRAM. For a moment forget about the style Ossi Weiner has made his points. You remain with 4 recognizable programmers who officially have complained. All 4 programmers are competing in the top of the SSDF. This is the first time this happens in public. All 4 of them are wrong? Does Fritz5 deserves the gold medal? Probably yes. But not without a normal doping control which is called "public available" in our case. We the competitors (programmers) have the right to check. I just read Rochade Europa, an interview with Peter Schreiner and Matthias. Matthias said, in no way the Chessbase Autoplayer must fall in the hands of competitors. Everybody who wins the GOLD medal has to pass the doping control. And the current gold medal winner *REFUSES* to pass the doping control. Openness is the only way that will take away all suspicion. In the meantime the BASIC trust from my side is gone because the autoplayer is kept hidden by all means. Here it comes again what bothers me the most.... When AUTO232 was introduced by Chrilly every programmer personally was asked to join by the mouth of Helmut Weigel. We (the programmers) all checked the new software. We (the programmers) all agreed. We (the programmers) all supported AUTO232. We (the programmers) all released the software in public, free for the public and for ourselves (the programmers) to check each other if we stick to the rules of fairness. Until now every programmer has stick to these rules. Why should Chessbase be excluded from this obligation? - Ed - >Amir
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.