[computer-go] Some observations regarding KGS tournaments

[Gunnar Farnebäck <gunnar@xxxxxxxxxxxxxx>]

I missed out on most of the debate while it was going on but I'd like
to add some observations.

1. There is no clear and obvious definition of when derivative works
   are "different enough". But what this really is about is to help
   Nick design the KGS tournaments to be as successful as possible.
   Exactly what success means here is up to Nick to decide as he is
   running the tournaments but I'll hazard a guess that important
   parameters may be to get many strong programs to enter, make the
   tournaments interesting events, and help further the state of the
   art in computer go (however indirectly). I think Nick's proposal to
   split the tournaments into two divisions is worth trying.

2. Remember that this discussion is specifically about the rules for
   the KGS tournaments. If someone thinks there will be an interest in
   a tournament where, to take a somewhat ridiculous example, anyone
   who owns a copy of a commercial program can enter it, you are of
   course welcome to try to organize one.

3. I don't agree with the assertion that programs which don't know how
   to play to remove all dead opponent stones are defective. I would
   be most impressed if, for example, someone with a
   learning-from-pro-games approach managed to make a program which
   could play at 5 kyu up to normal game end and correctly assess the
   status of the stones even if it didn't have a clue in what order to
   play the liberties of the dead stones without them springing back
   to life.

4. Just because GNU Go knows how to play to remove dead stones doesn't
   mean that it's easy. With experience from implementing it (in GNU
   Go) I would say that it's a very interesting problem, from a
   research and algorithm development point of view, but easy - no. Of
   course, if you don't care about losing points now and then there
   are no difficulties, even Brown
   (http://www.lysator.liu.se/~gunnar/gtp/brown-1.0.tar.gz) knows how
   to do that. But if you want to be able to take a group which is
   dead due to a bent four shape off the board you had better have a
   plan for how to first eliminate all ko threats. Seriously, how
   many existing programs have implemented that?

5. There are more alternatives than playing out until all dead stones
   are removed or be prepared to rely on manual intervention with the
   scoring. One alternative is to play until all stones, except seki
   stones, are unconditionally settled in a Benson sense. Not that
   this is very different from removing all dead stones, but it is an
   alternative. Another alternative is to let GNU Go, or some other
   engine, score the games. Programmers who don't trust the scoring
   engine to get it right should play until all dead opponent stones
   have been removed, others may stop whenever they think the position
   is sufficiently settled.

/Gunnar
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/