Author: Robert Hyatt
Date: 12:01:03 07/22/98
Go up one level in this thread
On July 21, 1998 at 22:58:52, Don Dailey wrote: >On July 21, 1998 at 07:01:20, Amir Ban wrote: > >>On July 20, 1998 at 21:34:57, Don Dailey wrote: >> >> >>>Although I don't have the same respect for Deep Thought that Bob Hyatt >>>does, I still have a lot of respect for it. It has beaten my program >>>twice, and both times probably prevented my program from becoming >>>world champion. Two games is a small sample to be sure, but common >>>sense makes me believe this was no fluke. >>> >> >>I still wonder sometime why you played 14...Kh8? against it in Hong-Kong. In the >>opening at least, DB made every effort to lose this game. Didn't you want to win >>it ? >> >> >>>I don't think Deep Thought (or Deep Blue) is invincible and you would >>>see many wins and draws against it but I have no doubt it would >>>still dominate the micro's. Murray estimated his odds of winning >>>that championship at about 50/50. He recognized that Deep Blue would >>>be a heavy favorite in any single game but that dodging 5 bullets >>>is a harder task. I was more optimistic than even they were, >>>thinking Deep Thought had about a 70% chance of winning. As it >>>turned out they drew a game and lost a game and this put them out >>>of it. >> >>That's exactly what happened to me, in the same tournament. So I figure that I, >>too, must have had 70% chance of winning. Right ? >> >>I've heard these sort of arguments before, and they are nonsense. The same sort >>of arguments were used to show that Fritz, who won, had a 10% or so chance of >>winning the tournament. If you listen to these arguments carefully, you will be >>convinced that this tournament not only does not weaken Deep-Blue's standing as >>the best chess computer, but actually reinforces it. >> >>The thesis behind these arguments is approximately this: Deep-Blue is >>self-evidently the strongest, it scored this result in this tournament, so let's >>compute the probability for the result. Interestingly enough, the people making >>this argument do not seem interested in revising or qualifying their premise >>that Deep-Blue is strongest. You would think that holding a tournament is done >>precisely with the aim of getting a better estimate of who is stronger and who >>is weaker. >> >>We'll never know if Deep-Blue was unlucky to score only 3.5 points, or lucky to >>score so many. The default assumption is that it scored just what it deserved, >>and there is no other scientific conclusion of any significance. The >>unspectacular quality of its games in this tournament also more or less agrees >>with the result. >> >>Amir > >You probably didn't understand what I was saying here. I didn't >make this guess AFTER the tournament, I made it BEFORE the tournament. > >Campbell did the same thing and used the same line of reasoning I did. > >I was mainly surprised because he was estimating the superiority of >Deep Blue over the others less than I was, to come up with the 50% >number. > >If you want to say this was foolish speculation, I cannot argue with >you, there is not enough data to make a reasonable statement of odds. > >You are correct to say that based on the results of that tournament >you cannot claim a superiority for them, but you are incorrect in >your observation that I was saying this. This is the second time >you have done the same exact thing to me. I am not very good with >words but your english is very good indeed and I didn't think what >I said was that ambigious. > > >- Don What number were you using? to get a probability of 50% winning chances, you simply solve for .5=X^5 because there were 5 rounds (^5) and you assume a probability of X for winning a particular round. Some sample numbers: X=.9 (90%) gives a 59% probability of winning all 5 games. If you allow for a draw in the games, X=.8 gives a 38% probability of winning all 5 games. When I talked to Hsu after the tournament, he told me that they thought that they had a > 90% probability of winning any single game there, which was the same number I used myself. Because >90% probability is over a 400 point rating difference, which I thought was about right. The last time I did this for Cray Blitz, I assumed X=.75, which *really* looks bad when you know you need to win 5 rounds (Yes, when we won the 83 WCCC we won 4 and drew 1, and when we won the 86 WCCC we won 4 and *lost* 1) but .75 gives a really poor chance for winning all 5 rounds. You really should have a variable X too, because for the first two-three rounds, I'd bet DT's X value was almost 1.0, and only starts dropping toward 90% in the last two rounds after the top of the field has been narrowed down to the best 4-8 programs...
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.