RE: computer-go:life and death

["Mark Boon" <tesuji@xxxxxxxxx>]

I wrote my Go program while at university, which is not the same as working
on a thesis. A thesis is generally a 6 to 12 month project, whereas I worked
on Goliath full-time for most of six years (skipping most of the classes).

I think that the way the Go knowledge was formalized was very interesting
but the claims in the article are quite overrated. This is not necessary,
since the results seem pretty solid by themselves. But I'm not completely
convinced whether it's possible to maintain it practically when you try to
solve more complicated problems. I don't know the Kano problems by heart,
but I seem to remember they're relatively trivial (as shown by the examples
in the paper). The system is of course only really worth something if it
scales well to more complicated situations. For example the failure to solve
diagram 4.5 (the one with all the false eyes) is a more "real-life" problem
and would be solved easily by most Go programs.

In my opinion David is right that a lot of the things described are actually
already done by existing programs, just in a less formalized way. I think
that both the statement that 'such an analysis is beyond current go-playing
programs' and later that your system solves the problem in only a few nodes
'very much smaller than big-search computer solutions' are false. You have
solved the problem using a different approach and representation than that
of existing programs, achieving similar results in a less efficient manner.
You still have to prove that your system offers advantages that in the long
run will lead to better or faster results than the traditional approach.

(My god, this Ghostscript is a program from the middle-ages. Please make it
available in a more friendly format, like PDF, next time you publish
something on the web.)

> -----Original Message-----
> From: owner-computer-go@xxxxxxxxxxxxxxxxx
> [mailto:owner-computer-go@xxxxxxxxxxxxxxxxx]On Behalf Of Tristan
> Cazenave
>
> I believe Mark Boon succeeded in writing a world champion computer go
> program
> during his thesis. Am I wrong ?
> Is Gnugo considered academic effort or not ?
>
> > >I don't know of any such results except (I think) for Wilmott's thesis
> and
> > >the subsequent paper by Willmott, Richardson, Bundy, and Levine
> (mentioned
> > >in an earlier post).
> >
> > Unfortunately, the strong programs aren't publishing much :(  I would
> > publish more, but it
> > is too time consuming.
> >
>
> Please David, please, write a paper.
> By the way, I would be very interested in a paper by the gnugo team also,
> as I have problem analyzing and understanding their code.
> Maybe giving the good ideas underlying a program is as important as giving
> the code.
>
> Tristan.
>
>
>