[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [computer-go] Protocol B



Okay, two more cents:

I still think part of the problem is that we are approaching the situation
with guidelines developed from typical human play.  The protocol (and
possibly even the rules) needs to be designed to accomodate the simplest
application of the rules.  Ideally they should work for every case from the
worst possible players up through the best possible players.  Currently, in
most rulesets, the game is finished after two consecutive passes.  Does
this mean that the game is scoreable?  In typical human play, yes.  For all
possible play, certainly not.

In my opinion, the game should be played out to "the bitter end" by
computer players.  By "the bitter end" I mean until all dead stones are
removed (seki situations would be a an exception.)   Why?  Otherwise, you
are taking the gameplay of both players and substituting a) human judgement
or b) perfect play or c) gnugo scoring.  Instead of letting the players
decide the contest between themselves, you are letting them initiate the
position, and then assigning a given solution to the go problem they have
created.

Why do humans stop when they do?  Because we have learned to read the
shorthand of "dead stones on the board" and find it boring to play it out
to "the bitter end" and have developed methods of bringing each other into
agreement about the shorthand.  

For computer players, I don't think the game should be considered over
until both players agree on the outcome of the game, or that there are no
more moves left to either player to help clarify that outcome (in which
case outside help is needed to judge the outcome).

As David states in his protocol, play should resume after two passes and a
disagreement, unless neither player feels they can clarify the game with
further play.  

I think the initial agreement question after two passes should be "Who
won?" rather "What is the score?" as this is simpler and bridges both area
and territory scoring styles.

(pass - pass)
?_Who won_?
if agree
	no problem, game over (report score for posterity, if available)
if disagree
	resolve-disagreement-genmove
	if reply is pass
		resolve-disagreement-genmove
		if reply is pass
			game over, third party judgement needed
		if reply is not pass
			continue game with genmove
	if reply is not pass
		continue game with genmove

Asking the programs to agree on dead stones (which is what humans do) is
trickier because what is dead varies by level of analysis, which is the
whole point of playing in the first place.

Does the protocol enable a program to "cheat" and force another program to
lose on time?  I don't think so.  If the cheating program is playing
worthless moves, the refutations should be obvious and not require a
significant amount of time to play.

Will humans consider some of these moves ugly?  Probably, but until the
computers see them as ugly, we should allow them to play each other until
"the bitter end."

Ciao,
Ben Shoemaker.

__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com 
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/