Re: computer-go: Mathematic Go

[Thomas Raffill <raff@xxxxxxxxxxxx>]

>'magic formular' for winning Go games.

As far as I understand, their algorithm for computing
the value of a game search tree is NP-complete, meaning
it still requires an exponential search procedure as
far as we currently know. But one could try to use
the theory for good approximations. 

Also, as far as I understand, the reason it breaks
down in the middle game is not just the large number
of combinations, but more importantly because we 
can't treat the various battles over the board as
separated and independent units -- they interact with
each other too much. It would be interesting to have
a mathematical theory to get good, quick approximations
for these interacting battles (if anyone knows of such,
please announce to this group).

Thomas Raffill