[computer-go] Explanation to MoGo paper wanted.

[David Doshay]

On 9, Jul 2007, at 5:12 PM, Don Dailey wrote:

>  I'm just thinking as a
> purist.    Any human generated GO knowledge, in some idealistic  
> sense of
> the word is WRONG.  In other words, unless you can formulate perfect
> rules, you are introducing prejudice to the search engine.
>
> So knowledge will help until you get to some point that it gets out of
> focus, starts conflicting with itself.

I disagree. The methods that are being used to the greatest degree
today are MC. The knowledge that humans have about Go virtually
always is conditional: this shape tends to be bad, but then again we
know that sometimes making the exact same shape is the best move.
I contend that what we know about playing good Go is in perfect
alignment with what we are now using for search.

What we need is a way to be able to search known good shapes
with a reasonable distribution that keeps us in what I keep calling
the "relevant space" of a reasonable game.

I also think that it is not a bad thing to have our knowledge set
"start conflicting" with itself. That is the nature of life and the
nature of Go. It is well known in learning theory that a child builds
models of the world until they have too many inconsistencies,
and then spends some time pretty confused and less adept than
when just a little younger, but then reformulates something in
their heads and then abruptly become more adept. I do not see
that our programs should be different.

The hard part is that we have a very strong tendency to take all
of that new info out of our knowledge base as soon as we see
the drop in playing strength. I think that it is the programmer
who needs to rethink this stage in their program's learning.


Cheers,
David