Hi,
I did a simple experiment in which I let the machine play against itself 1800 games in $ mode (unlimited). Actually I played the first 20 games before letting the machine to play against itself using the "end game" command. This is an important point. The machine was set as computer against human, in settings>players and I had to click "end game" every single game. I noticed that the advantage of gnubg steadily increased. That is why I kept playing 1800 games (it took me 2 and 1/2 hours)! Intrigued, I did a some analysis and I found a statistically significant difference in the probability of winning a gammon between gnubg and the human player. Apparently gnubg has a probability nearly 3% higher. I have attached a PDF file with my results. I find difficult
to believe that gnubg cheats and i plan to do more testing in the future (when I will have some time to kill, that is never i am afraid). In any event, I wanted to share with you these results, perhaps someone knows better. Thanks!
FS