computer-go: A.I. design, goals and consqeuences

["Jeff Massung" <jmassung@xxxxxxxxxxx>]

One of the threads running around has been about programming common sense 
into a computer program (in general) and I assume how that could help GO 
programs to play better.

Perhaps something to think about when designing AI (common sense), is 
sensory input/output.  I would like to compare the Sony (semi-new) robotic 
dog against that of a person for a moment (here we go again):

If you take the common sense example of walking into a wall.  People  and 
the dog learn quickly that walking into a wall is bad, but for different 
reasons.  People realize that it hurts, this is a negative thing, and the 
feeling of pain compounds this to make a person not want to do it again.  
The dog however, only realizes that it bars its path, and like the standard 
program of Maze, tries to find an alternate route.  There is no sense of how 
hard it his the wall, possible sensory damage, and therefore (other than not 
being able to reach a goal) nothing to prevent the dog from trying again.

An AI program must in some way have a sense of "consequences of actions".  
This is done in chess and in other games via search depth, and trying to 
figure out the worst damage that can be done on either side.  And from many 
other threads, that simply (at least right now) is not an option for GO 
(which is why in my opinion GO must be goal oriented).

Has anyone written (or tried to write) a GO program that is goal oriented on 
any level (owning corners, capturing groups, or as general as winning the 
game)?  Because (even though I haven't been on the group long) it seems that 
most GO programs out there do simple board analysis and move choice based on 
it (like chess).

Any and all comments welcome! :)

Jeff

______________________________________________________
Get Your Private, Free Email at http://www.hotmail.com