Re: computer-go: Insight of a human continued

[Heikki Levanto <heikki@xxxxxx>]

Christian Nentwich <c.nentwich@xxxxxxxxxxxxxxxxx> wrote:

> Local sequences in Go reach a certain point where a local continuation
> is not necessary. 

>   - Does anyone have a cutoff mechanism for reading which detects a
>     "natural point for tenuki" instead of reaching a tree-depth cutoff ?

I think this is pretty clear: Estimate the value of other big points onthe
board, and when ever you can gain more from playing one of them than from
continuing the sequence, take the big point.  The only problem is to come
with these estimates. Also note, that this may require circular logic: To
extimate the value of a play in this corner I need to know the values of
other big moves, which in turn depend on the value in this corner.

And all this was for gote moves only, and for situations that can be
analyzed independently.


>   - Does anyone have any ideas about how to judge a thickness/territory 
>     tradeoff using an algorithm ?

Huh, that's a hard one. Various distance- based or gravity-like influence
algorithms have been proposed, none of them seems to reflect human thinking
very well. 

Analysing the territory is the "easy" part. The value of influence must
necessarily depend on what else is on the board. A small living group in the
next corner may make your influence almost useless - or horribly valuable.
Simplifying again: The value of the influence is some function (simplify:
the sum of) the uses it can be put to.

- Heikki

P.S. I agree with you that patterns do have their uses. But they have to be
seen in the higher-level context. A brilliant tesuji to capture a worthless
stone...

-- 
Heikki Levanto     LSD Levanto Software Development   heikki@xxxxxxxxxxxxxxxxx
               "In Murphy we Turst"