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

Re: [computer-go] Super Ko on KGS ignores player to move

drd@xxxxxxxxxxxxxxxxx wrote:
Personally, I don't like PSK as set in the Tromp/Taylor rules either. In my personal opinion, and it's just that I admit, it detracts slightly from the elegance of TT rules because it subtracts an element (color to move) from what a position really is.

If given an arbitrary Go board configuration to play, would you care which color you are assigned? Of course you would! There is an advantage to having one over the other given any particular configuation. The same configuration with WHITE to move is a different "position" than the identical configuration with BLACK to move.
Tromp/Taylor tries to simplify by introducing the concept of board configuration, which is an addition. When you play a game, you never make 2 moves in a row (unless the opponent passes), because you understand that color to move is part of the "position." With PSK there has been added a new concept to the game which did not need to exist and detracts slightly from the elegance of the rules.

I understand of course that this call be interpreted differently. I know the Tromp/Taylor designers think this is actually a simplification and they are smarter than I am.
Dear Don,

Let me explain in greater detail why I believe there are very
good reasons to prefer PSK over SSK.

Your basic argument against PSK is that the simple notion of
position doesn't entail everything that matters for the future.

While I agree that is correct, I want to argue that the notion
of position should *shouldn't* entail additional information that
matters for the future.

Why? because it cannot be taken to its logical extreme.

Consider the position

 X . X
 O X X
 . X X

with white to move. I ask you, do you care which color you are assigned?
Then you say, well, that depends on whether O is allowed to capture the ko.

Reflecting on this, we can see that in fact the entire history matters
as to which color we prefer.

At the same time it is entirely obvious that a SUPERsuperko rule that takes
the whole history into account is useless.

The conclusion we must draw as that trying to include information about
the future into the notion of position is inherently flawed. I consider
any solution that includes some information while excluding other information
to be a compromise at best.

Choosing the simplest notion that does what we want in practice seems the
best solution then.

It is both easier to describe and easier to implement (at least conceptually;
there is one set of forbidden stone configurations, rather than two separate


computer-go mailing list