Re: [computer-go] 2nd KGS Computer Go Tournament

[Don Dailey <drd@xxxxxxx>]

Hi Nick,

> I have never seen a case where any judgement was involved.

There is little or no judgement about what the actual result SHOULD
be, but you are implicitly making the judgment that the program will
play it out correctly without making it prove it.


Here is a scenario that happens in my autotest games once in a while:

1. Program A is losing because it correctly realizes that it has a dead
   group.  Since the group is dead and it is losing, it passes.  

2. Program B incorrectly thinks the group is alive and thinks it has no chance
   to win, so it passes.

3. You award the win to program B, since it's technically the correct result.

4. HOWEVER, it might not be the ACTUAL result if the game were played out.  Program
   A might be able to win if B is going to keep passing.   A may only need 1 or 2
   moves (while B keeps passing) to actually make the group live.


Nick, I can actually take advantage of this.  All I have to do is make
my program pass a little early when it thinks it is losing and because
it is sometimes wrong, I might be awarded the win.

It's almost bizarre that I can manipulate things such that I can cause
you to occasionally give me an undeserved win!

So I am going to turn off genmove_cleanup, there is really no need for
me to use it if you are willing to do the work for me and it actually
could hurt me once in a while.

- Don





   Date: Tue, 10 May 2005 23:53:34 +0100
   From: Nick Wedd <nick@xxxxxxxxxxxxxxxxx>
   X-Spam-Score: -4.9
   X-Spam-Flag: NO
   X-Scanned-By: MIMEDefang 2.42

   In message <200505102229.j4AMT8t9012417@xxxxxxxxxxxxxxxxx>, Don Dailey 
   <drd@xxxxxxxxxxxxxxxxx> writes
   >
   >One more thing to add.
   >
   >It's true that some programs haven't yet implemented the protocol and KNOW
   >the stones are dead but will not clean them up.
   >
   >But by scoring for the program you have to make a judgement that this
   >really is the case.  If you are wrong, you are rewarding the ignorant
   >program and in some cases actually changing the result.  So the
   >situation becomes MORE than just a scoring situation.
   >
   >William early talked about the separation between "playing" and
   >"scoring."  There should not be a separation because if you score for the
   >program, you could be changing the result.  You can't base your
   >decision on the belief that a given program "should" be able to win or
   >even that you "know" it can, because then you would be adjudicating
   >based on your own personal judgement of a program's abilities and
   >that's biased.

   I have never seen a case where any judgement was involved.

   In a position like this

     . . O # .    Top right corner of board
     . . O # O
     O O O O #
     # # # # #
     . . . . .

   I agree that deducing the status of the two black stones would require 
   judgement.  But, with Chinese rules, bots don't leave things like this 
   on the board, they play them out.

   I am willing to say that if I ever do find a case where judgement is 
   involved, I won't try to use any, I shall instead rule (as far as the 
   stones in question are concerned) in favour of the bot that does 
   implement final_status_list.

   Nick

   >With chinese scoring, there is no good reason for trying to separate
   >the scoring phase from the playing phase when it's not possible in the
   >first place, and when it's so easy to do it right anyway.
   >
   >- Don
   >
   >
   >
   >
   >>  Nick,
   >>
   >>  I assure you that kgs-genmove_cleanup and "final_status_list dead" are
   >>  EASY to implement.
   >>
   >>  kgs-genmove_cleanup just means to play the game out to the bitter end.
   >>
   >>  final_status_list dead is really simple,  it just means "give me a list of
   >>  all the stones you think are dead for both colors."
   >>
   >>  Believe me,  if a program is good enough to pass before killing the 
   >>opponents
   >>  dead stones,  it KNOWS the stones are dead and "final_status_list dead" can
   >>  be implemented easily.
   >>
   >>  If the program DOESN'T know this, then it shouldn't be passing.  If it
   >>  passes anyway and you end the game by scoring it, you are being unfair
   >>  to the opponent because you very well might be aribitrarily changing
   >>  the result of the game.  If the program doesn't know the stones are
   >>  dead, the opponent may be able to save them.
   >>
   >>  In fact, you might as well stop the game even earlier,  when YOU judge
   >>  the game to be decided.
   >>
   >>
   >>  - Don
   >>
   >_______________________________________________
   >computer-go mailing list
   >computer-go@xxxxxxxxxxxxxxxxx
   >http://www.computer-go.org/mailman/listinfo/computer-go/
   >

   -- 
   Nick Wedd    nick@xxxxxxxxxxxxxxxxx

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