Re: computer-go: A problem with understanding lookahead

[Tapani Raiko <praiko@xxxxxxxxx>]

> -- The bad news is that the mathematical framework for dealing
> with this kind of reasoning doesn't exist.

It does exist. It is called Bayesian framework. I would recommend a paper
by Eric Baum and Warren Smith: "Best play for imperfect players and game
tree search"
http://www.neci.nj.nec.com/homepages/eric/bpip1.ps

Instead of having one estimate of the point situation you could have an
estimated probability for each point situation. So if you have a 100 stone
group, which lives with a probability of 90% you would have two larger
spikes 200 points apart. And probability here is not the frequentistical
one, but a bayesian "subjective" probability.

Best move to make would be the one which moves most of the mass from the
losing side to the winning side instead of largest expected value of point
situation.

Of course this creates some extra work, but the framework is there. Has
anyone implemented something like this?

-- 
 Tapani Raiko, <tapani.raiko@xxxxxxxxxxxxxxxxx>, 050 5225 750,
 http://www.hut.fi/u/praiko