Re: computer-go: 5x5 Go is solved?

[Dave Dyer <ddyer@xxxxxxxxxxx>]

For the purpose of solving the whole game (of whatever size), you
can't possible take in account all the possible paths through which
a position can be reached at the same time as you consider the
abstract value of the position, independant of the path through
which it is reached.

Starting from the endgame:
 Terminal positions have a simple value.  
 Positions which contain a single Ko have two possible values
 Positions which contain two Kos have 4 values.
 Positions that have been seen but not solved are "unsolved"
 etc.

When you encounter a previously encountered position, you stop
searching and return the value of the position.

The big problem is for positions where a big capture occurs.  You
can't solve these immediately, because you jump too far back up
the tree.  Leaving these "unsolved" positions may rapidly pollute
the whole tree with unsolveds, or it may not, because the unsolved
nodes will be strongly avoided by alpha-beta because.

Once you have the whole database classified with some mix of
simple and nonsimple values, a second program can be written
to analyze the outcome in light of the nonsimple values and the
precise Ko rules employed.