Re: computer-go: unmake move?

["Mark Boon" <tesuji@xxxxxxxxx>]

You're right that abstract reasoning is often a much better approach than
reading. However, before you can make abstractions, you'll need information
to abstract from. For example, in order to have an abstract representation
of a 'group', you first need to generate a lot of information on which you
then base the abstraction. Once you manage to do that, you can do abstract
reasoning on a group level. As an example you can then implement some
reasoning that having two weak groups is bad and open to a split attack.
This will then make it easier to find good candidate moves because you'll
only consider moves that make one of the groups safe instead of going for
another invasion. But the basis of the abstraction has to be generated
information. This can sometimes be generated by some algorithms or by
reading (tactical or not) or by a combination of both. For the group example
I think you'll find it impossible to make a good 'group' abstraction without
tactical reading. Because if what looks like two groups at first is in fact
separated by an opponents chain that can't escape even when it moves first,
you'll obviously want to treat them as one single group.

    Mark

----- Original Message -----
From: Heikki Levanto <heikki@xxxxxxxxxxxxxxxxx>
To: <computer-go@xxxxxxxxxxxxxxxxx>
Sent: Friday, June 09, 2000 7:07 PM
Subject: Re: computer-go: unmake move?


> Yes, this is possible, but not what I intended to mean. My ideal is to aim
> at a much higher-level reasoning of the board, where I can drop the
> irrelevant detailsof stones and liberties as soon as possible, and to
> describe everything in more abstract terms. This ought to enable me to
make
> more qualified decisions about where to play, at least on the level of
which
> grou to attack/defend etc. I still admit that tactical reading is
necessary
> for finding the exact move (and for evaluating the starting position), but
I
> feel interesting stuff lies beyond that.
>