Nick's remarks about teaching computer programs which they understand rang a
bell.
Recently, I operated a beta version of MFG12 at the Cotsen tournament. It
appears to have a very strong tendency to stake out a large center territory.
If the players permit this to be solidified, MFG wins. But in the pursuit of
this huge central territory, MFG make lots of hanes (diagonal plays) which
leave cutting points behind. When players aggressively attack the weak points,
they win.
Qualitatively, subject to the caveat that I only have a sample of five games to
work with, this would be in instance of not knowing how to follow up on
objective - it stakes out a large territory but doesn't know how to keep it
stable, and the root appears to be not so much a lack of tactical skill as a
lack in what Go players refer to as "good shape" - good shapes (patterns) are
easily defended; bad shapes are not.
Such "shape" patterns require a bit of light tactical reading to be effective.
Yilun Yang, published by Slate and Shell, has a nice short presentation of good
shape. To cite the most obvious blunder: a double hane is sometimes effective,
sometimes not. A bit of local tactical reading is required to discern the
difference.
When using a joseki, a MC-based program will need to have enough knowledge to
read out whether it works or not. Perhaps a directed expansion of the top-level
tree, of all the reasonable branches, including the known semeais, trick plays,
and blunders and their refutations? If a ladder is involved, this directed
expansion might add some interesting ladder breakers to the tree - even trying
them in advance of playing the joseki, in order to constrain the opponent's
plays to its advantage.
Terry McIntyre <[EMAIL PROTECTED]>
We must stop dressing up the slaughter of foreigners as a great national cause.
-- Sheldon Richman
> My impression of "Joseki for computers" was that it was really "Joseki
> for 5-kyus".
>
> Suppose you want to teach joseki to a 5-kyu, with the objective of
> making him into a 4-kyu. Assume that you have no higher objective such
> as his one day becoming 1-dan. It is sensible to teach him to play
> josekis in which he understands what he has achieved (such as making
> third-line territory), so he won't screw up later. You should teach him
> to avoid josekis which are sound in the hands of a strong player
> (because of say the central influence they give), but which the 5-kyu
> won't know how to follow up on.
>
> Nick
> --
> Nick Wedd [EMAIL PROTECTED]
_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
