On 1, Jan 2007, at 12:15 PM, Jacques Basaldúa wrote:
And now remember how this discussion started: There was a proposal
to penalize pass moves made by Lukasz Lew.
If that proposal is implemented, Japanese programs will no longer
loose one or two points against a better ruleset adapted bot, but
they would loose dozens of points. They will frequently loose won
games. Maybe some programs can easily switch from Chinese to
Japanese, but some others may not. Anyway, outside computer go,
people understands go as Japanese. Beginners find it more complicated,
but when they understand, they see its just concentrating on the only
interesting part. A natural evolution of the game. When they are 10kyu
or better they normally agree what is alive and what is not. If they
don't, its probably worth playing out.
I still think Chinese rules are better today for computer tournaments!
But, of course, without penalizing pass moves. I hope that the day
when
computers evolve to Japanese rules as humans did, is near, but that
cannot be forced. It is required that all programs agree when scoring
games. At least: *when* nothing more can be won and what is *alive*
and what is not at that moment.
and
On 1, Jan 2007, at 1:08 PM, Don Dailey wrote:
By far, Chinese is more intuitive and natural. Japanese rules are
based on some very non-intuitive concepts - that it really does
come out the same as Chinese scoring (within a point or two) appears
to be magic to the uninitiated.
The proposal made by Lukaz is the same as AGA rules. The purpose
is to assure that when a player thinking Japanese is playing against
one thinking Chinese both come to the exact same conclusion.
But I think that while this is an advantage, it also completely ruins a
primary emphasis of the Japanese rules: efficiency, and efficiency all
the way to the end of the game.
At one point in this lengthy ongoing discussion, it was noted that it
is not polite to keep playing after the result is already determined.
The Japanese rules do penalize these moves by one player as long
as the other player is knowledgeable enough to see the situation
correctly and simply pass, thereby picking up a point.
To address Don's point, I respectfully disagree. I reason, with liberal
use of analogy, thus:
The Chinese rules acknowledge that it takes two eyes to live, and
in a way that to me is similar to the thinking of a military occupation,
sees no value in any more space than that. If the rest of the group
has extra open spaces or if those possible open spaces are filled with
stones (or people) is of no consequence. Perhaps this is a consequence
of living in a society where it is considered the norm for people to be
packed tightly together.
The Japanese rules also come down to "it takes two eyes ..." but give
credit for the extra open spaces. To me, this is analogous to living in
a city with more parks, or living in a village with more farmland and a
less dense population, and I know that I would take the option with
less crowding and more food production. It mirrors a quality of life
issue very well.
To bring this back to computer Go and what it implies about the level
of understanding of the game we can attribute to the programs, I will
point to the last round of the recent KGS slow tournament. Look at
the game between SlugGo and MoGo. While I am not trying to say
anything about who won, because the rules were clearly stated to be
Chinese, note that SlugGo started passing, indicating that it saw no
purpose in any more moves, at move 239. Here, the boundaries are
clear, the dead stones are clear to a human, and the winner is plenty
clear enough. But the game continued to move 526! All in "invasions"
that were not "reasonable" by human standards, but which are not
costly under Chinese rules. By Chinese rules MoGo wins by 2.5, by
Japanese rules SlugGo wins at move 526 by almost 120. This difference
is due entirely to SlugGo being compensated for Mogo's continuing
the game past the optimal point, and SlugGo recognizing that by
continuing to pass except when required to play on the board to
save something.
Again, MoGo wins the game and the tournament, but I will ask: to a
human Go player of most any rank above 15k, which program seems
to understand the game better? The emphasis must be upon the word
"seems" because there are not too many changes to endgame details
for MoGo to have passed earlier, so I am not trying to discount that
MoGo took good advantage of several SlugGo blunders in the game.
MoGo earned the win. I am only talking about the perception that a
bystander will have after watching all 526 moves.
I do think that it is worth noting that a large number of the SlugGo
moves (which at the late stage of the game are really pure GNU Go
moves, so they deserve the credit) are ones that maximize the extra
points that can be had under Japanese rules instead of making moves
that clarify the situation instantly but allow for more throw-in moves
to follow. To me, this exhibits a very subtile understanding of the
game. It also demonstrates that while GNU Go has a switch to set it
to play under Chinese rules, the engine is still "thinking" Japanese
until final counting.
Where I do agree with Don is that it is much easier to teach the game
to beginners with Chinese rules. So, I will agree with the comment that
it makes sense to think of the Japanese rules as a evolutionary step
from the Chinese rules. Under Chinese rules efficiency is important in
the game until the value of moves goes to zero, and then many moves
on the board continue to have zero value. Under Japanese rules there
is the additional consideration that many of the still legal moves on
the
board have negative value because an aware player should see that this
game is over and it is time to move on to the next game.
I believe that at some point the computer Go community will also have
to accept that programs which understand the details of Japanese
endgame should get credit for that against programs that do not know
when to just win a won game by passing.
Cheers,
David_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
