>> I'm curious what you guys think about the scalability of monte carlo with UCT.
The MCTS technique appears to be extremely scalable. The theoretical papers about it claim that it scales up to perfect play in theory. My feeling is that as you scale up in power, certain things will improve relative to humans faster than other things as you imply. This happened in computer chess and to this day computers are inferior to human players in many ways, and yet the computers are superior in playing the game overall. At some point computer go programs will start doing some thing better than the top players, while other things will remain behind. So they will still lose. The things the computers do not do well will still continue to slowly improve, helping the situation. The things computers do better will become overwhelming and it's only a question of when the combined effect is enough to beat the top players. I don't believe there are any solid barriers as you imply. There are just some things that are harder than others and improvements in those areas will come slower (when compared to humans.) I personally believe that what computers do better will give humans a hard time more than the reverse. It may be that once computers start doing a few things better, humans will have a difficult time. In chess this manifested itself in 2 areas - tactics and consistency. They do "consistency" better than any human will. A human will make a silly oversight that is "uncharacteristic" of him. A computer may also make errors, but not "uncharacteristically." A chess computer will never miss a 3 move checkmate for instance. This is a wonderful strength that should not be underestimated. In tennis it is hard to beat even a mediocre player who's only strength is that he doesn't make errors. If the ball comes near to him, it will always come back and you will never get a free point. It puts the pressure on you to deliver and you must never miss. Because of the ability to calculate accurately, chess programs quickly became known for their tactical ability. Even though they made positional errors frequently, it got to the point where you still had to work hard to beat them and they never let down. So then players had to actually change their style in order to win, which in turn is a constraint on the way you play and makes you weaker. It will be harder in GO than in Chess. The weakness are more glaring and the strengths are less useful in GO, but I think it will eventually happen. - Don On Sat, 2008-08-09 at 14:54 -0400, Robert Waite wrote: > I'm curious what you guys think about the scalability of monte carlo > with UCT. Let's say we took a cluster like that which was used for the > Mogo vs. Kim game. Then lets say we made 128 of these clusters and > connected them together efficiently. Putting aside implementation and > latency issues... what kind of stones-strength increase would you > imagine? > > Its a pretty arbitrary guess.. but do you think one stone > improvement... or that this would alone be enough to beat a pro even? > > I am wondering because there could be a weakness or limit in MC w/ > UCT. I am only now learning about the UCT addition... but there are > vast numbers of possible games that are never visited during the monte > carlo simulations. The random stone simulations are pretty random > aren't they? I am reading some of the papers on the UCT addition... > and that does seem to show certain branches to be "better" and worth > more time. Pro players may have a counter-strategy that might come out > as Mogo is tested at higher levels of play. Perhaps there will be a > need to combine MCwUCT with heuristics or more knowledge based play. > Going the route of heuristics seems unpleasant and the promise of > using a more computational method would be great. However... if MC > techniques alone have a diminishing return... the route of heuristics > might come back (or perhaps a whole new paradigm for game algorithms). > > I am still secretly on the side of human go beating the machine.. but > the recent match really changed my view on topic and really showed the > value of statistical analysis. I am just wondering about what kind of > roadblocks might show up for the monte carlo techniques. > > _______________________________________________ > 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/
