Re: computer-go: alpha-beta cautions

[Dave Dyer <ddyer@xxxxxxxxxxx>]

At 12:27 AM 11/17/99 +0200, Mika Kojo wrote:
>
>Dave Dyer writes:
>>  There is a very small area where alpha-beta can help; where seeing a few
>> moves deeper allows you to see that "something dramatic" actually happen;
>> but if your evaluator makes dramatic pronouncements, and is WRONG, 
>> alpha-beta will distil your bad evalutions into bizarrely wrong moves.
>
>I cannot see how alpha-beta can be blamed for this. As alpha-beta is
>not a heuristic it should not weaken your program. 
>

The problem is that alpha-beta leads you to examine many strange 
situations, and evaluate them as though they were normal.  This
tends to lead the program down strange paths, and in effect make
it weaker.  Time and again working with my problem solver I found
that errors were due to misevaluation of absurd positions deep in
the search.  The second prong of the argument is that Go is a game
of percipice-walking. You only have to stumble once to fall
off the cliff.  If you examine more positions, you are more likely
to err spectacularly at some point.