Re: [gnugo-devel] GNU GO and GO AI

[Daniel Bump]

> I am writing a paper on AI programs that are designed to play the game of GO,
> and I would be grateful if you have time to answer a few questions.
> 
> What was the hardest part about writing the GNU GO Program?

This is fairly hard to answer since GNU Go is the work of many people.
One thing that is quite hard is that the move tree in Go branches
much more than in Chess. This means that you have to carefully
prune your search at every node. GNU Go accomplishes this
pruning by use of carefully tuned databases.

> How would you describe your program's method of computation (i.e. Chess
> programs often use minimax methods)?

The program's algorithms are described in detail in our
on-line documentation:

http://www.gnu.org/software/gnugo/gnugo_toc.html

See Chapters 4 and 6-13 for description of the algorithms.

> What challenges do you face in your continuing attempts to 
> improve GNU GO?

GNU Go development is slow right now. If GNU Go is developed
further, I think the biggest challenge would be to
incorporate Monte Carlo algorithms since these have proven
very successful. Another desirable but difficult goal would
be to parallelize the code.

Daniel Bump