Re: computer-go: unmake move?

[Don Dailey <drd@xxxxxxxxxxxxxxxxxxxxx>]

I tend to believe that the best way to do things by computer might not
mimick the  way humans do  it.   But this  is pretty philosophical and
it's just an idea.

However your point is well taken.  I would  think that if humans can't
tell if a  string is safe without  calculating, then it's probably not
too likely that a  computer  can either.   But  I'm open minded  about
this.

My gut  feeling is that there  could some day  be a good  synthesis of
both  methods.  That IS  what humans do so maybe   I'm refuting my own
statement!  

Humans seem to be better at pruning away irrelevant lines of play, but
I'm not sure it always has to be this  way.  In chess, computers often
find lots of surprise moves because they didn't prune, and so it's not
completely one sided in favor of humans.  That's why I believe we will
try  to take the best  of what humans  do, to  the  extent that we can
implement it and combine it with the strengths of the computer.

There is a trap that  computer go is more   likely to succomb to  than
computer chess.  Sometimes humans are elevated  to the status of Gods.
In go humans are so  much superior that humans   do seem like Gods  in
comparison.  And as  humans   we like  to exaggerate  this  even  more
because we prefer black and white  to gray.  Haven't  you ever heard a
player say, "I beat him every time?"  It usually  turns out that every
time is just 2 times.  It used to be this way in chess, because of the
superiority of humans.  It was believed  that it didn't matter how far
you looked ahead, or how good your evaluation function was, a computer
would never  be really good unless it  reasoned like a human  does and
had intuition.  This was hammered home every time a grandmaster beat a
computer by means of a  pretty sacrafice.  Now computer routinely beat
grandmasters  using their boring  unimaginative techniques.   And they
look good doing it!

So it may turn  out, many years from now  that we learn how to program
computers  to play really   strong.  I hope we  don't  give up  simply
because we are intimidated by the Gods.

Enough philosophy!

Don



   Date: Fri, 09 Jun 2000 09:10:53 -0400
   From: David Mechner <mechner@xxxxxxxxxxxxxxxxx>
   Sender: owner-computer-go@xxxxxxxxxxxxxxxxx


   Another kind of "evidence" is what humans do. We have a computers with
   staggering computational horsepower and unmatched pattern matching ability, but
   we still do tactical search to find if a string can be captured (albeit much
   more efficiently than how existing programs do it) and lookahead to figure out
   which move is the best.

   If there were a way to play well without doing these things, wouldn't some
   smart, lazy human have figured it out? 

   Heikki wrote:
   > Is there any good argument why go programs "have to" be based on reading, or
   > is it "just" that the most successfull ones have been so far?

   -David