I've been playing against gnu-bg for a while, and every now and then I find myself struggling to understand how it judges some of my moves when I look back at them with the analysis. I must say I learned quite a bit by playing against it and I can realistically see an improvement when playing against other people. I am well aware that intuition and "human way" of looking at probabilistic things sometimes go against the hard numbers and so I generally try to keep an open mind, thinking the machine played millions of games, it must have it right. I can't, however, being a human after all, bring myself to stop questioning it, and today I got this one position where for the life of me I can't see how it makes its call. This one is simple enough for me to do the math (so, really simple ;) )
Position (game attached) is trivial, bearing off it's the machine's turn and it has two pieces on slot 2 and has the turn to roll. I have a single piece on slot 1, so I finish the game next move (if there is one). gnubg will win in the next roll unless it rolls a one not accompanied by another one, so 5 out of 36, right? Anything else makes it bear off the last two pieces and win. That gives me 0.13888 for gnubg to lose. gnubg doubles before rolling, I decline and in the analysis gnubg says I should've accepted and that my winning odds were 0.278.
If the calculation is right I would love to understand why. What am I missing? The cube has no further say here since if it does roll a non-double one it is 100% lost and if I roll or double it'll be the same thing, win single.
Anyway, not sure how the list works, I just subscribed for this so if I am violating some policy I apologize. I hope this message finds someone willing to look at it.
Thank you for a great piece of software (even if it gets on my nerves by beating me more often than not ;) )