[computer-go] Example for Morphys law in Evaluation

[Jacques Basaldúa]

Chrilly wrote:

 > Rule 1: If a group is evaluated dead, look if there is an adjacent dead
 > opponent group which is even weaker ("more dead"). If yes, the group 
is not
 > dead, but just weak.
 > Rule 2: If a groups is evaluated weak, look for a weaker adjacent 
opponent
 > group. If yes, the group is not weak but okay.

I think this kind of rules reflect the present state of computer go.
Would you agree to say, computer go solves tsumego problems at dan 
level, but
when we have to make decisions "between dragons" we adopt simple rules, that
work (say, in the best case) 95% of the times. Unfortunately, that is not
enough because strong players blunder much less than 5%. And as Magnus 
stated,
the decision must be made mostly when the position is still unclear. When
its clear, it is usually too late.

I don't know if everybody knows this paper:

http://cs.nyu.edu/web/Research/Theses/klinger_tamir.pdf

I have my own ideas on something related with this, but, i hope, more 
feasible.
Even if the author (almost) assumes that implementing his ideas failed, how
many of you think that an ambitious step in a similar direction could 
make a
quantum leap in computer go performance?


Magnus Persson wrote:

 > Monte Carlo evaluation was sent from heaven.

Remember Einstein: Der herr Gott würfelt nicht. (God doesn't play dice.) :-)


Jacques.