Re: [gnugo-devel] WinHonte and AI

[Arend Bayer]

On Mon, 3 Feb 2003, Måns Ullerstam wrote:

> I am sorry if I offended anyone regarding WinHonte. I was only trying to
> make a point of testing some interesting machine learning techniques in
> GNU Go. My plan is to try out a few different machine learning
> algorithms in GNU Go and I was hoping that someone else on this list was
> interested in the same area.

I certainly wasn't offended. I find WinHonte a very interesting program.
Dan posted two GNU Go vs WinHonte games previously
(http://mail.gnu.org/archive/html/gnugo-devel/2002-12/msg00013.html
and http://mail.gnu.org/archive/html/gnugo-devel/2002-12/msg00030.html).

I was really impressed by WinHonte's very natural looking play in
the first game. But it certainly still has some weaknesses too, see
how it fails to deal with GNU Go's incursion at J1 in the second game.

Your suggestions about using neural networks, machine learning
within/with GNU Go are certainly interesting. Especially replacing some
of the large pattern databases with a neural network based approach, or a
(semi-)automatically generated database, ... is something that should be
on our longterm agenda, at least as experiments.

As far as I can judge about other games, computer programs have never
been very successful when humans tried to tell them very explicit what
to do. E.g. telling them which moves to try in chess (as in early
non-brute force approaches to chess), or heuristically rule-based
approaches to bridge card play. The more explicit input we, as humans,
provide, the more chances we have to introduce bugs, inconsistencies,
etc...

Our 10000 patterns that we currently have, are a whole load of such
explicit input. And the parts of GNU Go that I personally consider
successful, reading.c and readconnectc.c, don't need a lot of such
input. (Of course, that is mostly because they have easier problems to
solve than owl.c or value_moves.c etc.)

> I love the work you have done so far and my goal is to start
> contributing again a lot more. I think it is a pity that there are so
> many other small projects around the globe trying to solve different
> issues with go, developing their own programs and user interfaces, etc,
> instead of using GNU Go as the foundation and instead add a move
> generator to GNU Go with their favourite theory applied. I want to
> encourage people to work with GNU Go instead.

I understand your point here, but people write these programs to have
fun, and they may have reasons to think their own little program will be
more fun than contributing to GNU Go.
The important thing is that we try to keep the code as simple and as
easy to understand as possible. That lowers the bar for potential
contributors.

Arend