Re: computer-go: question

[Daniel Bump <bump@xxxxxxxxxxxxxxxxxx>]

Tim Klinger wrote:

> Definitely not.  If you really think that the problem is so simple (and
> even if you don't) I would like to propose the following research project:
> 
> Write (or adapt) a go playing program to take life-and-death information
> from a human expert in every position.  The expert will point to each block
> on the board, identify it as alive or dead and provide moves to save or
> kill on request.  The program will take that information and decide on a
> move.  If you are right, then it should play somewhere around the expert's
> level. 

Efficiently using your hypothetical oracle might involve invoking it
thousands of times per move, at deeper levels of the move tree. One
of the characteristics you presumed for it is that it is fast. Such
an oracle would greatly speed evaluation of positions. One would
still need to estimate territory but this is fast (using Bouzy's 5/21
algorithm, for example) compared with the overhead of determining
which groups are alive and dead. With such an oracle, brute force
searching of the move tree (like chess programs) would not seem
so impossible.

But such a use could not be mimicked with the human expert.

Daniel Bump