[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: Re[2]: computer-go: perfect play
Mark Boon wrote:
>Think only about Berlekamps endgame theories. These are very stable and
>settled positions late in the game, still it's almost beyond any humans
>comprehension to play these positions perfectly. And one mistake loses
>instantly. Now think about how complex it gets if you have to be just as
>cautious in the opening and middle game.
>
>My guess is that todays top professionals would probably still lose a few
>points in the average endgame against the perfect player. Maybe even as many
>as five.
>
I think five is a bit high for an average endgame. Mathematical endgame
theory has now advanced to the point where we can analyze a typical endgame
quite a long way back. We see quite a few small mistakes, on the order of
half a point or less, but big errors (3-4 points) seem to happen mainly
when they miss a move in a complex local fight. Amongst many other things,
professionals are good at counting, too! As a case in point, see the games
of 'environmental Go' that were played between Jujo Jiang and Rui Naiwei.
I don't know if Bill Spight or Bill Fraser are lurking on this list, but
they could give more information.
Or see
http://msri.org/publications/ln/msri/2000/gametheory/spight/1/index.html
http://msri.org/publications/ln/msri/2000/gametheory/fraser/1/index.html
for their talks this summer at the MSRI workshop on combinatorial games.
Martin