About the convergence proof to go:

As far as I can understand Álvaro's proof sounds correct,
but that possibly applies to most games, but not go as
humans play it. Because what we play is "MCTS go" an almost
identical game where filling your own eyes is not allowed.

All MCTS programs finish their simulations because
the don't fill their own eyes. You can create a position
(I did it years ago and posted it in this list) where
filling your own eye is the only good move. Such position
is clearly a counterexample for this convergence. If you
don't fill your own eye, you get the eval wrong, but if
you do, when does the game end? It probably never does.
Nobody has proposed anything that works other than not
filling own eyes (there is probably no need anyway) and
that is enough to find counterexamples.

Of course you can find other mechanisms too. Paradoxically,
the better it plays the least it converges to minimax
evaluation. That's the trouble with minimax value, you can
prove it exists, but if you can compute it, the game is
not worth playing. ;-)


Jacques.







_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go

Reply via email to