computer-go: On rules for comp-go

[Joan Pons Semelis <joan@xxxxxxxxxxxxxxxx>]

( another wee has passed, my message bounced, I repost it )

After a week of wandering I've found my mail full of messages on the
topic; after
reading them all (even if superficially) I've found my self switching
from lurking
mode to posting mode, so here some scattered comments go:

On atari-go and stating correctly the nitpicking rules:
  A little girl I was teaching once saw one of her groups in atari, she
looked
  for a solution as she knew that the first to capture any of the
opponents stones
  was going to win, so she filled the last liberty to avoid capture.
When I pointed
  out that she was commiting suicide she said 'Ah, well.', removed the
dead stones
  and waited for her opponent to keep on trying to capture something.

  Little children do not make many assumptions, they still have little
knowledge to
  try to guess things and do stick very well to what have been literally
said. They
  are a good test for correct wording.

On my prefered computer-go ruleset:
  Usual stone capturing method.
    \ It has to be go after all
  Suicide allowed.
    \ Just to have more ko threats, I'm always short of them :-)
  No repetition of full board position with the same player to move
allowed.
    \ This has been shown as easy to implement, far more easy than even
the capturing rules,     \ I would say
  Most stones aboard (or on board :-) + eyes wins. 6 points komi
        \ This allows for very weak programs to play against each other
as there is
        \ no need of negotiating human-like the end of the game, which
is an added
        \ difficulty for the programmer. If your program knows well
enough when the
        \ game has ended you can set thresholds to begin to pass:
           0 - knowledge : till only necessary eyes remain
           1 - knowledge : all dead stones have been actually killed (
maybe a little more)
           2 - knowledge : all dame points have been filled. ( L&D/Score
disagreement ?)
           3 - knowledge : 2*n gote dames remain ( just to make japanese
happy )

So 2 programs should agree (or a TD should decide) about at wich
threshold they are playing,
whether suicide is allowed or not, what to do with cycles longer than 2
(forbidden, jigo,
replay game).

I would like to point that threshold 1 can be usefull with human
beginners that have
found that go software buried in a 'games pack' and are trying to play
that unknown
game in a small board ( 5x5 ? greater problem with beginners is when and
why a game has finished, well understood atari-go helps at this, but
there's still a step to make, maybe go programs are a long way of
helping Lee Chang-ho to improve it's game the way chess programs are
used by Kasparov or Krammnik, but they should be able to help beginners
learn ) OTOH, only a small handfull of programs can actually play
confidently at threshold 2, and that at human beginners level, so I
think it makes sense to have computer tournament games at threshold 0 to
encourage new programs. Even an atari-go computer tournament could be
interesting!

As it has been seen, most humans don't play accordingly to a logical
ruleset, but do
follow 'standard practice' as most strange cases only appear once out of
thousands of
games, thus, when playing with humans, a program can follow the same
approach, and use
its own internal logic that will go unnoticed by the human !!
Your program can play at threshold 2, substract 1 from blacks score if
black made last
move ( or an equivalent but more complex method at threshold 3 ) count
komi and have
your japanese customers happy. They do not need to know that it was
thinking in chinese :-)

-- 
Joan Pons i Semelis
 joan@xxxxxxxxxxxxxxxxx