[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: computer-go: minimax and go
Don Dailey <drd@xxxxxxxxxxxxxxxxx> wrote:
> Please note that I am taking a bit of a philosophical approach,
Ok, so I try to be philosophical in my reply ;-)
> At some point, we will have to start searching more than just local
> string battles and we will have to do it really intelligently.
Yes, of course we need some sort of searching. Probably more than "just"
local string battles, but I believe that a global search alone will not be
practically possible in the next few decades or centuries.
Or rather I believe that if there will be a global search, it will not be
directly related to the board image, but happen on a much higher level. For
example considering an invasion here, the search might jump directly (via
some patterns, one local search, or other means" to the seven most likely
results of that invasion, and only if it decides to make that invasion, it
will read it out in more details, and check that those fit in the global
"search". I guess this sort of search would better be called planning.
I believe this sort of approach is much more reasonable in go than in chess,
because the local subproblems are so much less interconnected.
No, I have no guaranteed path to prove that this approach (or any other)
will eventually provide a perfect player. I only have a strong gut feeling
that such a thing will not be seen in the next few centuries...
Have you calculated on your model: How much more computing power would you
need to get a perfect chess player? Assuming Moore's "law" holds, how long
would *that* take? And how long until we can have your perfect go player,
that searches the whole game? If we assume that there are some theoretical
limits on the size and speed of a processor (a switching unit the size of
electron, not working faster than light goes through it, perhaps), how fast
can this thing of yours ever play? How many parallel processors will it
need? Will there be enough material and time in the universe to play the
perfect game?
-H
--
Heikki Levanto LSD Levanto Software Development heikki@xxxxxxxxxxxxxxxxx
"In Murphy we Turst"