[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: computer-go: RE: beat 9d before 2010 - no, size does matter
At 01:37 AM 11/15/99 +0100, you wrote:
Ray Tayek wrote:
> > > >I think it's impossible for a computer to beat a pro 9d
> >...
> >... is difficult to say a computer can beat a pro in 10 years. ...
>
> its actually real easy to say (the answer is no,...
>
> a little arithmetic will make it clear. ... 361! , 3^361 ...
>
BUT I am saying that we can build a Go playing program that does not come up
against these BIG numbers at all.
it will come up against these numbers for any future that i can imagine.
I am saying that we can capture the knowledge
that a 9 Dan Professional has in his head. ...
well, i have been hanging around sensei yang (7p) for a number of years.
most of my "knowledge" is in the form of heuristic rules. some are from the
proverbs. tactical stuff like "always hane at the head of 2 stones" or "the
empty triangle usualy sucks". and global stuff like "don't play near
thickness". some of the rules are from him. for example when your'e
involved in a squeeze (hasami) joseki, the priorities are normally 1) get
out, 2) get the corner 3) complete the shape. some rules like: "play in the
biggest/most important area" are sometimes simple (count the empty spaces)
and sometimes very vague (which is the important area?). i probably have a
hundred or so running around in my head. the problem is always "which one
is the most important now?".
my experience with mr. yang is that he is unusual in that he can
communicates in simple heuristics most of the time. most pro's will tend
show you a complicated set of sequences when asked a question. i suspect he
has than a hundred. local good tactical moves (tesuji) are maybe a hundred
at most and these are local patterns.
global information can skew things. in the fuseki and early midgame, if
your opponent has most of their stones on the third line, you have to play
very different than of they have most of them on the fourth line.
The BIG numbers that concern me are things like:
1. How much knowledge do I need to represent.
not much for each group, but some of it will be hard to come by. "how much
territory to i have here if i am alive?" is easy. "how alive am i?" is
hard. "am i probably connected to this other group" may be hard.
2. How long will it take to elicit and structure.
3. How powerful is the human brain compared to my PC.
i suspect you could do well with a few weighted hundred rules where. but
this may not be what you mean by knowledge.
AND I think the answers are - VERY approximately:
1. Not more than 100,000 rules
hmmm, we must have very different ideas about what rules are. i would guess
the number to be a few hundred (at least that is what floats around in my
head when i play).
2. About 5 years of full time effort for a team of two + good access to a
9 Dan Professional
maybe access to a bunch of any pro dan's. the difference between 1-dan pro
and 9-dan pro is only about 4 stones.
3. About the same to within an order of magnitude.
this is a hard question. this may be true of you are counting neurons and
speed of impulses, but the connectivity and software are not even close.
thanks
Ray (will hack java for food) http://home.pacbell.net/rtayek/
hate spam? http://www.blighty.com/products/spade/