On Fri, 2008-08-08 at 17:19 -0400, Robert Waite wrote:
> If you mean that beating all human opponents would be solving go...
> then I think it is certain that we will.

I would think the distance between perfect play and top human play is
quite far off.    Beating the best human players is a good goal, but in
the grand scheme of things it's only a signpost along the way to a much
more distant goal.  

There may come a day when the best humans have no chance against the
best computers - and still computers will have a long way to go.

This is exactly what is happening in computer chess now.  The previously
best program, Rybka, has just been upgraded to version 3.0 and is
showing a whopping 100 ELO improvement.   This is quite remarkable when
you consider that it was already the best, and that it has already
surpassed the best human players.   

It was not long ago that the idea of even getting up to expert strength
(2000 ELO) was ridiculed as being an unattainable barrier.   Much the
same as in GO, where 10 -15 years ago the idea of Dan level play was so
far off it was considered completely unattainable by pessimists, and
even optimists viewed it as a century away.

It doesn't matter whether GO is pspace-complete or not unless the goal
is to completely solve the game (proving what the best move and score
is.)   We can still continue to attack the problem and make progress as
we have recently seen.    We will do it on 2 fronts,  software and
hardware.   We saw how Mogo threw a lot of hardware at the problem, but
they also threw a lot of software advances at the problem too.   

In the future sometime it will happen that we cannot make much more
improvement.  What will prevent the improvement is not any particular
technological barrier it will be simply that we will already be close to
perfect play.  We may not have solved the game by then, but for all
practical purposes the program will make perfect moves the vast majority
of the time.

This HAS (or is) happening in checkers.   The best programs have only
tiny room for improvement.  Play 100 games to get a score of 2 wins, 1
loss 97 draws (or something like that.)   A major improvement is being
able to win 1 more game in 100.   It's so bad that the strategy is more
about how tricky the moves are than their game theoretic value.
Imagine that happening in Go!  Always playing the best move in GO would
produce an unimaginably strong program,  but imagine that computer GO
got to the state that all programs played perfect moves 99.99% of the
time and the only strategy left was to find a position complicated
enough that your opponent has a slight chance to make a mistake.

- Don




_______________________________________________
computer-go mailing list
[email protected]
http://www.computer-go.org/mailman/listinfo/computer-go/

Reply via email to