Mark Boon wrote: > > On 3-mrt-08, at 18:43, Don Dailey wrote: > >> I base that logic on my observations that once the score goes below 10% >> for Lazarus, it is losing. It's extremely rare that it salvages a game >> once the score goes below even 20%. > > In which case I could argue that attempts at winning by playing > 'silly' moves are not working either. It hardly seems like an argument > supporting your position. It may in fact make things worse. This is a > feeling I get when I see MC programs play, as soon as one falls behind > the loss is made irreversible by some extremely silly moves. But this > is more the case early in the game, not by the end of the game. I really believe the source of peoples confusion on this is believing that the program starts playing "ugly random" moves as soon as it is down a little. But in fact, when it gets into "ugly" mode it is because the score is very close to 0.0 or in some programs like Lazarus -1.0
And when the score is zero, it basically means that the winning side can win the game with random moves - the tree search cannot even find lines of play that win for the defender even when the winning side is playing randomly. (This is a slight simplification, but it is essentially what is happening.) If the score is slightly higher than 0.0, it means the tree search combined with nearly random play from the play-outs occasionally finds a way for the losing side to win. But you can be sure that it's extremely unlikely that a skilled opponent will stumble into one of these rare lines. Just to make it clear, the case we want to "fix" is the case where many bots are programmed to resign. Lazarus will resign when the score is below 1% (and has remained so for a couple of moves in a row which is probably just a superstition on my part to delay it.) What's really screwy about this whole discussion is that we have previously argued on this list about how silly it is for programs to play out the game, how anti-go culture it is, how unmannerly it is and how rude it is to play on and on hoping for a blunder by the opponent. Well here is our chance to honor those principles. If the program KNOWS it is losing then it has the obligation as a cultured gentleman-like player to graciously resign the game. If you want to talk about what is ugly and what isn't, it's pretty ugly to play out a hopeless game that even a random player can win in a position where even near-random play cannot go wrong. It may be that we have watched games on KGS with these bots, in what seemed like really close positions and imagined that the game really was close when it wasn't. Probably because we didn't have access to the opinion of the program. We might think differently if we saw that both programs have already recognized what we don't. I think humans, probably do perceive a close game (in points) as not being a true loss, but a close game. I'll give you one anecdote from chess: I remember watching a game between a strong master and a fairly weak player in a tournament that lasted over 40 moves. The ELO difference was enormous, perhaps 1000 or more points. After the game was over, the weak player was quite proud of himself for holding out for so long. His impression of the game was that he kept in really "close" and made the master work for the victory. The truth of the matter, is that he had a dead lost game after the 5th or 6th move and he was down a pawn without compensation. The win was a matter of course and was simple and basic and the master took the "long way around" which they often do which means they do not take any chances. The weaker players tend to equate "how many moves" it takes to win as a measure of how well someone played relative to another. In reality, winning quick entails risk and the sophisticated players avoid risk if it in any way threatens their winning chances. > > To be honest, I'm not sure who is correct in this argument. I do know > that as a human player, when I'm behind by 5 points or less I > generally don't make any desparate attempts to win because I know from > experience I stand a good chance of winning by just being patient, > even against very strong players. But I bet that you still have realistic practical chances to win or you would resign or pass. Being behind 5 points in one position may not be the same as being behind 5 points in some other position. If you are behind 5 points when the game is over, you are lost, but if you are down 5 points in a complicated position, then you have every reason to continue to play carefully. In the positions you are talking about no good MC programs would start playing "randomly." > But this may only work for humans as many players feel a pychological > pressure when they're ahead that increases their propensity of making > errors. > > Mark > ------------------------------------------------------------------------ > > _______________________________________________ > computer-go mailing list > computer-go@computer-go.org > http://www.computer-go.org/mailman/listinfo/computer-go/ _______________________________________________ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
