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

Re: [computer-go] I know we disagree,but I choose to do nothing about it.

On 25, Jul 2005, at 5:20 PM, drd@xxxxxxxxxxxxxxxxx wrote:
William,

Your post made something clear to me that I wasn't articulating in my other
posts on the subject:

Although Nick Wedd is willing to fix-up results at the end of the game, that
only addresses tournaments that he runs. And that's not my biggest concern
because I know that those tournaments will get run correctly and Nick is
doing a good job in that regard.

But having some protocol to allow computers to resolve disputes without
needing a human is an important step for Computer Go in general, whether
it's used in any particular tournament or not.

And that's probably why I'm such a champion of the idea. Your post reminded
me that I don't really care that much about whether Nick does it manually or
not, as long as the protocol (or any protocol for this) surfaces and becomes
widely accepted.

Don
We must be getting close, because I am in complete agreement with
everything Don has said. ;^)

I find myself also feeling that perhaps I have not been clear enough
about my problem with the proposed protocol, and the posting from
Ben Shoemaker makes it clear to me what I have not said before.

My problem is that after the kgs-genmove-cleanup the first pass can
carry a heavy penalty: all of your opponent's stones are alive if they
pass. So, after one disagreement about status, it is as if the rules
have changed.

This not only lets, but it encourages a bot with a large number of
dead stones behind enemy lines to respond to one pass with a
pass, disagree about status, and then have the chance that the
otherwise winning bot will pass again. And because there are
points to be made this way, it will happen.

The natural result of this is that after all the dame are played each
player has to start filling up their own territory until all living groups
have nothing but a number of isolated single point eyes (otherwise
the other player can play a single stone in and claim a point for the
stone on the board). This kind of play cannot be good for computer
Go.

I doubt that this kind of a protocol will be universally adopted, and I
agree with Don that we should be moving towards a protocol that
can be universally adopted.

What Ben suggests is:

All games should be played out until both sides agree upon the state of the
game. As Don has pointed out, the computer is _always_ trying to dispute a
loss and play moves which illustrate a win from the current position. When
the opponent, with better ability, prevents this from happening, the
computer exhausts all its potential moves until it concedes the win.

The way I see it, for a game to be complete, the two players have to agree
to one of two things (or both):
1. The final status of every stone on the board.
2. Who won the game.

In the case of a tournament, disputes arise only if the two players can't
agree on either of those. Play should be resumed until an agreement can be
made. In rare cases, players will reach an impasse because they both
assume that correct play is to pass. At that point, the referee should be
able to decide the game.

Even if two players of extremely poor ability are completely insane in
their assessment of the status of each stone on the board, if they agree on
who won the game, there should be no dispute. If the ruleset does not
agree with their determination, the referee can overrule the result.

In practical protocol terms, I guess this would mean adding a request for
"game result" in addition to "final status list" in order to determine if
both players agree on the current state of the game.
I agree completely with this. The extra communication of asking for agreement
again after two passes should not really take that long, and in this case there
is no special case where that first pass carries the possibility of the penalty
of a large dead group suddenly being declared alive. Instead, each time in
the loop some progress is made towards resolution, and in the case of an
absolute impasse a human is called. I agree that it will be great when our
programs get to the point that they can score properly, but I don't think we
are there yet. But if we do allow the loop to continue it may indeed impact the
overall time for the game, and we would have to figure out how to deal with
the time limits in this "dispute resolution" stage.

Another thing I like about Ben's suggestion is that it allows for Tom Cooper's
desire, with which I also agree:

I think it is nice to have the required entry standard into these tournaments
as low as possible, it allows more people to enter, and to enter earlier in
the development of their programs.
I hope this helps clarify my view.

David

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