Rich Heimlich wrote:
We actually helped develop one of the few iterations that went beyond
that
approach. It required us to literally rate and weight every single word
in
the Scrabble dictionary. It took months to do. I'll never forget it,
but
the results were amazing. However, the next publisher that got the
rights
went right back to the old way because it was cheaper and faster.
This is very interesting.
The analogous problem in backgammon is to determine what a "difficult"
decision for a human being is. Until now I've thought that this was too
hard to program into GNU, but maybe it isn't. Perhaps it doesn't take
that much hand-marked data to train the AI what sort of decisions are
hard for a human and what sorts aren't. Someone ought to try this.
It's a little surprising to me that the next publisher went back to the
old way...unless you're saying that they acquired the rights only to the
Scrabble name, and not the database of hand-crafted ratings that you
built?
Tim
_______________________________________________
Bug-gnubg mailing list
[email protected]
https://lists.gnu.org/mailman/listinfo/bug-gnubg
