Re: Strong players only?

[Mousheng Xu <xu@xxxxxxxxxx>]

At 07:11 PM 11/24/98 EST, Compgo123@xxxxxxxxxxxxxxxxx wrote:
>Darren Cook wrote:
>
>depends on the computer speed. If a computer can search 361 ply deep, all
>knowledge it needs is the Go rules. Ok, at present time Go programs have to

Hi, Dan,
	You raised an interesting question: how much knowledge do we need to
reduce the calculation to an extent that a modern computer can handle?
	DO WE HAVE ANY SOLID GO KNOWLEDGE WHICH WE CAN USE FOR SURE to reduce the
calculation? 
	1. Pattern matching -- there is always some "pattern" that a pattern
database cannot catch. After we search through the whole pattern database,
we cannot say for sure that "these are the only cases I need to consider".
We may be able to say something like "95% cases are included in this result
set", but that 5% is always critical. :-)
	2. Eye detection -- what percentage of eyes which can be easily detected
by human can be effectively detected by a computer? DO WE HAVE AN SOLID
(not heuristic) ALGORITHM FOR EYE DETECTION?
	3. Survive or die judgement -- without an effective eye detection
algorithm, this issue can not be resolved.
	4. Board value estimation -- ... that's the key ... but ...
	
	As I mentioned early on this list, we need to DO GO RESEARCH. We are
probably fooled by the oriental term "potential","moyo", ... we need maths
or physics statements. :-)
	
	Thanks.

-- Mousheng Xu

P.S.
[Raising my hand]: What are the problems of a context-sensitive grammar for
pattern matching?