Re: computer-go: Computer-Human Comparisons

[david elsdon <david_elsdon@xxxxxxxxxxx>]

At 09/12/02 16:47, you wrote:
<snip intro stuff>
>  1) How much do computer programs play like human Go players?
>         1a) And how far are they designed to play like human players rather
> than computer players?
I designed GL7 to play like I do so GL7 just looks around the board and 
makes the first decent looking move it sees; i.e. it does no explicit 
evaluation of the board. However, I'm about 4 kyu and my program is at the 
bottom of the 9x9 ladder. It is clear that when I look around the board and 
make the first decent looking move I see, I am doing something different to 
GL7. It is very hard to use introspection to work out what I do, but I am 
still working on it.

>  2) Is it desirable to have computer Go programs playing like human players?
>         i.e. Will we be able to create a Go program that excels at Go using
> methods that computers are made for (look at    Chess) or are the more
> abstract human techniques essential?
My feeling is that when we do finally crack the problems of computer go and 
the best programs are giving 9 dan professionals 9 stones and beating them, 
the programs will be looking at the game in a very different way to us 
humans. But until we suss out that "very different way", the human way is 
the best we've got, although maybe the neural nets and genetic algorithms 
used to play go are already "looking" at the game in a different way !!

I have just started to investigate the way people learn to play go. This is 
a much more interesting problem. I am going for a high level symolic 
approach here rather than the NN or GA approach.

I'll keep you posted, but don't hold your breath !!

Cheers

David