Re: computer-go: minimax and go

[Heikki Levanto <heikki@xxxxxx>]

myriam <mabramso@xxxxxxxxxxxxxxxxx> wrote:
> I need to know (succintly) what are the arguments that a brute-force
> minimax approach to Go won't work assuming increased computer power 
> just like it worked for Chess? 

To put it simply:
  a) The branching factor in Go is about ten times higher.
  b) There is no simple-minded way to evaluate the position (like 
     counting pieces in chess), nor is there a clear-cut winning position
     (like chessmate).
  c) the game is much longer, and the sequences one has to read out
     can be extremely long as well. A ladder across the board can easily 
     be 40 moves (20 ply), and reading it only halfway gives you absolutely
     no idea if it makes sense to play it or not. Much longer sequences can
     be constructed, although they do not occur often in real games.
  d) A typical middle-game position consists of several loosely separated
     battles, which anyway have an effect on each other. Isolating them and
     checking for interaction is pretty difficult. Reading all combinations
     for all positions is extremely wasteful.
     
The last point is not so much an argument against minimax, but more an
explanation why humans play so good Go.


-- 
Heikki Levanto     LSD Levanto Software Development   heikki@xxxxxxxxxxxxxxxxx
               "In Murphy we Turst"