On Wed, 2008-08-27 at 13:20 -0700, Bob Hearn wrote: > In principle MoGo ought to be about a stone (or slightly more) weaker > with 1/5 the processing power, which is consistent with 2-3d against > Kim and 1-2d against the 6d. > > I watched both games, and MoGo did seem stronger to me against Kim... > but then, I knew in advance the processing power in each case.
I thought a doubling was worth about a stone, but of course this depends a lot on the particular program and setup. Maybe 5X is closer to 1 stones as you say and not 2 stones. Concerning the ability to "judge" the quality of moves, I have anecdotal evidence from computer chess a couple of decades ago. Two master friends of mine played games all night long with 2 different chess computers that I had and they had no knowledge of the respective strength of the machines. They believed the machine that was 300-400 ELO weaker was the stronger machine. The weaker machine had a more exciting and aggressive playing style and this is partially what fooled them. They didn't believe me, so we played several games between the 2 computers and the stronger machine won every game and they were surprised. I was surprised too, that they couldn't make this judgment call. I have no real evidence other than my own subjective observations, but I have seen this type of thing happen over and over again, players not being able to judge the ability of others (machine or human) based on just "judging" the quality of the moves. This is true in tennis too. You can watch a player playing someone else, and not have a clue how much trouble they will give you. I have had easy matches of players that really looked smart and good and then have been given all kinds of trouble from the "scrappers" who looked just awful subjectively. There was some top pro who played a while back who looked really sloppy but I think made the top 5 or 10 in the world. I also saw Goran Ivanisevic practicing at a tournament when he was a top player and he looked just awful, missing and spraying balls all over the place. I remember thinking in my mind how ludicrous that was, because he really looked quite beatable - even by a club player. But I knew that was just an illusion. - Don > > > On Aug 27, 2008, at 1:08 PM, Robert Waite wrote: > > > * You really can't conclude much about any mogo strength improvement > > from just > > * one game. > > > > It is true that you can't make a conclusion.. but you can draw some > > information > > from two games. I would think it is statistically unlikely that a > > person who is > > > > 10kyu will beat a 3dan in an even game. So far... both games have > > pointed to > > around 2dan. I would agree that you can't really know.. for > > example.. what if > > both human players didn't play seriously.. then you wouldn't gain > > much at all. > > > > But then... even if there were 30 games.. the same thing could happen. > > > > I just find it interesting that Mogo played at approximately the > > same strength > > with 1/5th the processing power. Not really a conclusion.. just an > > observation. > > > > _______________________________________________ > > computer-go mailing list > > [email protected] > > http://www.computer-go.org/mailman/listinfo/computer-go/ > > _______________________________________________ > computer-go mailing list > [email protected] > http://www.computer-go.org/mailman/listinfo/computer-go/ _______________________________________________ computer-go mailing list [email protected] http://www.computer-go.org/mailman/listinfo/computer-go/
