> Why are you comparing humans to computers? It's ridiculous to measure > progress by comparing to the top human players. What we care about is how > much progress we can make from year to year. >
i think that the use of the word "ridiculous" might be a bit strong. measuring against humans is only silly once you've crushed them and can easily crush them at will. for instance, if we developed robots to do the running long jump, and it turned out that they could only jump 5 feet, a natural and exciting milestone would be when they could reach nearly as far as average humans, then as far as the best humans. once they could jump 5 times as far as the best of us ever could, there'd be no more need to compare them to humans anymore because nobody would have much interest in human-robot competitions. so if you set out to design a computer chess program back at the dawn of computers with the sole goal to make a computer chess program which could beat all other computer chess programs, and you never once stopped to compare it against human play, it would seem a little strange. especially since you were teaching a computer to play a game that humans invented, with human rules. computers probably wouldn't invent chess as their own favorite game to play if we left them to their own devices, right? so it's not all that unnatural. it's also a real-world metric that the layman can get his hands on. Here is how we can judge our progress. Go back 30 years and get the best > go program available and compare it to the best GO program of today. Play > 1000 games and tell us what the score is. Then let's have a discussion on > whether we are making progress or not. > i'm pretty sure that we have made progress, and a lot of it. but go is simply a slightly more complicated game to analyze. not to compare against previous go-playing programs, but to analyze as a game. that was really the only point i was trying to make in my original posting. the reason computer programs aren't currently crushing humans by some simple adaptation of chess-playing programs, for instance, is because it is seemingly a more difficult game to make computers good at. not because it is some totally different game that the best people didn't bother to write code for just yet. i say seemingly, but i really mean that combinatorially, it's larger, and there doesn't appear to be any major collapsing of the game tree that immediately reduces it to a simpler game to analyze. I think what is happening here is that we are so shortsighted we don't look > more than 2 weeks ago and think that is representative of what is actually > going on. In Dave's case we took a look 40 years ago and forgot to look > again. > i think we are running the risk of drifting from the original question, grinding axes and getting on soapboxes. s.
_______________________________________________ Computer-go mailing list [email protected] http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
