Author: Robert Hyatt
Date: 12:34:42 09/09/99
Go up one level in this thread
On September 09, 1999 at 04:45:53, Micheal Cummings wrote: >On September 09, 1999 at 00:55:59, Mark Young wrote: > >>On September 08, 1999 at 18:08:56, Dann Corbit wrote: >> >>>On September 08, 1999 at 17:36:38, José de Jesús García Ruvalcaba wrote: >>> >>>>Should the game Rebel-Hoffman be taken in account to calculate the performance >>>>rating of Rebel in the GM challenge? >>>> >>>>-Yes 1 >>>>-No >>>>-Abstain >>> >>>Why such an absurd position as to claim it should count despite machine crashes? >>> >>>Because it does not matter *why* the outcome was as it was. >>> >>>If a GM defaults because he does not feel well enough to continue, does that >>>invalidate the loss? No. And neither should this one. >>> >>>No cries of "mess-up... do-over" should be allowed. >> >> >>If you feel this way then by all means count this game. I am more interested in >>Rebels true strength playing GM's. Rebel is not a human, it is a computer >>program. This loss was due to a program error or hardware error. We know this >>because the person checking over the moves of Rebel could not reproduce some of >>the moves. This tells me it was not the fault of the program, and not some >>excuse by Ed. We have to few games in this match to pollute it with such a >>obvious faulty game. Unless new info comes to light saying this is how Rebel >>would have played in a correct working condition. > >for the offical record it must count. I have seen many GM's stuff up a game >which they should not have. For example WCCC99 Fritz Vs GM, he totally stuufed >up the opening and thus humiliated himself. I say they should be treated like >any human player. Win loss or draw, nothing else. > >Everybody was willing to jump all over kasparov when he lost to deep blue, stuff >his condition or mental state. > >Well it is the same stuff the Rbel software or hardware. It lost due to an >error. Just like a human player it will have to prepare better next time or >expect to stuff up again. There are two different issues at stake here: (1) computers vs humans. _certainly_ there will be hardware failures in some games vs humans. And that obviously gets factored into the overall performance result. Because machines will fail. I had more than my fair share of hardware glitches with Cray, because I almost always ran on a prototype of the next generation machine, as that was the fastest thing they would have. (2) are computers at GM level yet. We have so few games that I believe it unreasonable to include an occasional game where something bad went wrong with hardware. IE I have lost games against humans when power knocked my computer out and the clock kept running. And those games counted in its USCF rating (Cray Blitz many years ago). But to answer the GM-level question, with so few games, an unexpected hardware glitch exerts way more influence on the rating than it really ought to. Ed made a mistake. He wanted to run on the fastest thing he could. I have made that mistake many times, and had it pay off (I won the 1983 WCCC event on serial number 1 of the XMP-2 dual processor from Cray. When we played round 1, that program had _never_ played a game. When I left Hattiesburg driving to New York (about 1200 miles away) the program was not running. Harry was up at Cray debugging a hardware problem like mad. We went thru 5 rounds with nary a glitch, which was amazing.) I have also had it blow up in my face on several occasions just as it did for Ed. Leading edge often becomes bleeding edge at inopportune times. Some are complaining because they don't like the result, they just "knew" that computers were playing at those inflated 2600-level ratings they see on the SSDF. Others don't like the results because they think the computer ought to lose every game quickly and miserably. But the bottom line is that the match is going on, it is offering some very important data. And we only need to use a little sense in interpreting the data. 1 loss and 2 draws is not bad so far. the second loss really had nothing to do with whether the program was better than the human or worse than the human, it had to do with propogation delays and settling times of gates on a processor that is being run well beyond the manufacturer's specifications. I'm simply sitting back and waiting for round 4 (or 5 should you choose to count the hardware failure as a game). Bob
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.