Re: computer-go: Pattern matching

[Patricia Hughes and David Elsdon <babel17@xxxxxxxxxx>]

"P.J.L.Cuijpers" wrote:

> Patricia Hughes and David Elsdon wrote:
> >
> > Give me an example of knowledge that cannot be represented on a > machine.
>
> In a course on expert-systems I was told, and I now believe it,
> that "common sence" is very very very very very very hard to represent
> and maybe even impossible.
> But that is because you don't know what you know so maybe I am
> cheating when I use common sence as an example.

It is indeed very hard to represent common sense knowledge and it is because we
are not very good at articulating such knowledge, but then Go knowledge is
pretty hard too. If I am able to articulate my Go knowledge completely down to
the last nut and bolt then I will be able to use it to build a "better than
2kyu" program. I am 2kyu. My program would be better than me because:

1. We are all stronger when critiquing a game than when we are playing.
2. The program would not have an off day, or get tired or bored.
3. It would not mis-read a ladder the way I sometimes do.

It does not suprise me that building strong Go programs has not succeeded in
breaking the 1 Dan barrier yet. Strong Go players, like all other experts I have
worked with, are not very good at introspecting on their own Go knowledge. They
need to work with a person skilled in the "extraction" of knowledge from an
expert, i.e. a knowledge engineer. Once the knowledge has been codified, writing
the program is a "relatively" straight forward task.

The representation of common sense knowledge would be very time consuming and
there is little motivation to do it. But it could be done.

Regards

David