Re: computer-go: Computer Go Tournament Program

[Don Dailey <drd@xxxxxxxxxxxxxxxxxxxxx>]

   From: Robert Jasiek <jasiek@xxxxxxxxxxxxxxxxx>

   Don Dailey wrote:
   >    How not to give any score?
   > I  don't understand  your question.   When   you pass,  you  give your
   > opinion  of what the final score  will be, or   you choose not to give
   > this opinion.

   Exactly. Does the program state "I, ahem, do >eh> choose not to
   give my opinion!"? How is the program required to state this?

I purposely left this out because I  wasn't trying to describe the low
level  protocol  (in GMP  it  might be a byte   code of some  kind for
instance.)

But the semantics is either:   I pass and believe the final score is n,  or
I pass and choose not to predict the final score. 

   > John Tromp suggests    the semantical simplification  of   viewing all
   > passes as coming with a default score  whether the passer provides one
   > or not. 

   So we have
   - pass
   - default score
   - score optionally stated by black program
   - ...by white
   - optional non-score opinion

   Which entity states what how for which reasons?
   No, I don't want to make it more complex, I want to have it
   so simple that even I understand it:)

   > You don't interpret scores at all.   The players can  agree on a wrong
   > score if they want to.

   Who can interpret and what are the consequences?
   The programs
   The programmers
   The referee
   The public


Tromp/Taylor has a way to accurately calculate a CURRENT score, it does not
try to predict if it's safe to stop playing.  So scores can have 2 different
contexts:
             1. Tromp/Taylor for a given position (easily verifiable and agreeable)
             2. Someones opinion of what it will end up as.

So it ends up being that the players can either agree  on the score by
stating their opinion and hoping the other has the same opinion, or in
the end falling back on Tromp/Taylor.  Presumably no one will disagree
on  this, but if they  do, the arbiter program  or human has the final
say.   If there is  no arbiter, then it's  hopless anyway, because you
and me can play a game and refuse to agree on the  final score as easy
as we can both disagree on what 2+2 is.   

One issue not recently discussed is whether each program has access to
the others estimation of  the score.  I don't  have strong feelings on
this but it has been suggested that it  is slightly fairer to only let
the arbiter  have access to these scores.   But this protocol can work
between 2 reasonable players that  are competent enough to  understand
Tromp/Taylor (it works if both players come  away from a game with the
same notion of who won and by how much.)


   > If they don't agree, you use Tromp/Taylor territory calculations which
   > can  be  easily calculated  without error  by  the human   or computer
   > arbiter.

   "or"? IMO, there should be only one of them or a defined
   hierarchy.

The only   issue is what  is   the score  if  they  fail  to agree  in
consecutive passes.  In this case, the arbiter gives  the score and if
this protocol is implemented  between  2 programs without an  arbiter,
then  we have the  notion of a correct  score (the actual Tromp/Taylor
score) and 2  players walking away  with a different  idea of what the
final score  is.  This is no  different semantically than  the players
disagreeing on the rules, in which case  they have no business playing
each other!



   --
   robert jasiek