On Tue, Sep 7, 2010 at 2:42 PM, Dave Dyer <[email protected]> wrote: > > > But you have never (to my knowledge) layed out what way that is. > > You're quite right here. I'm not advocating a specific change, just > pointing out that all the effort going into building faster monte carlo > engines may be irrelevant, because the programs actually need better > steering. > > I know we disagree on this point, but I believe chess has reached it's > current state of success MOSTLY because of Moore's law. >
We disagree, but not just for the reason you think. I don't believe that computer chess has been that successful. I recently posted about this on a chess forum. Yes, the best programs are already considerably better than the best human players but I think that is a foolish way to measure success (unless that is your goal of course.) My reasoning is that at the top levels of chess (which now is computers) there are lots of losses (for both colors.) That means the program are playing far from optimally even after many decades of progress. On the other hand, something you may or may not know, it is generally believed that chess software has advanced just as much as hardware - in other words the programs are as good as they are because of software AND hardware. A few years ago I argued that it's almost all hardware - but I have had to retract that statement. In fact I am now testing that concept with a 16 year old program, one of the very best programs of it's day, against one of the best programs of TODAY. BOTH programs are playing on modern hardware and I've tried to make everything as equal as possible. After 4 games I realized that this test was foolish so I am now giving a 24 to 1 time handicap to the new modern program and it's looking like that is not nearly enough. I have to emphasize that this "old" program was raved over in it's day as a very fine chess program and the author of this program dominated computer chess for several years with his programs. I want you to realize this because you have commented many times about computer chess, implying that it's played out or practically solved or that it's now a simple engineering problem and all that is required to write a good chess program is to add up the material on the board, throw in a few positional heuristics and you are done. That is an embarrassingly naive way to look at modern chess programs. Also, the amazing software advances in chess refutes what you just said about the state of art being where it is because of Moores law. To be fair, that is approximately HALF the picture, but it's not clear to me if it's the bigger half or the smaller half. Several hundred ELO are due to software improvements and several hundred to hardware advances. I'll also give you one huge reason why I like the direction computer Go is going in. In my opinion this is the first time we have had a truly scalable algorithm. I know that some authors claim to have had this for a while (even before MC) and perhaps this is true to an extent, but this is an idea that should evolve for a long time to come, just like computer chess. I think you will see that in 10 years the best GO programs will be significantly stronger (even ignoring hardware) and my guess is that it will have evolved from the current MC programs even if they evolve into something that does not resemble todays program very much. The primary difference in todays chess programs and yesterdays, is that they are MORE scalable. It isn't just that they are scalable. Alpha/beta brute force pruning always gives you something that is scalable as does MCTS, but you have to go much farther than that. I think MCTS will. > It always was believed that Go was would have to be solved by other means, > perhaps even (gasp!) understanding the game. Monte carlo has given some > credibility to the theory that Moores law may be enough after all. I'm > arguing not. > > _______________________________________________ > Computer-go mailing list > [email protected] > http://dvandva.org/cgi-bin/mailman/listinfo/computer-go >
_______________________________________________ Computer-go mailing list [email protected] http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
