Re: computer-go: Complexity & SW

[Don Dailey <drd@xxxxxxxxxxxxxxxxxxxxx>]

> What matters is that chess has attracted hundreds of man-years 
> more development time than go, and that advantage has proved 
> overwhelming, as one would expect it would.  

I don't believe this  has much to do  with  why chess programs  are so
strong compared to go programs.

Yes, we have continually  refined the software, but  I would  also say
that MOST of the progress  has been due  to faster computers with more
memory for bigger hash tables.  In the late  70's early 80's there was
a  program  called Chess 4.x  by Slate  and Atkin that  really was the
basic model for modern chess programs.  No sucessful program that I am
aware of does  anything   radically different.  Probably   the biggest
advance since  then has  been  an understanding   of how  to  properly
implement   a selective  search, but  even  that   has not  added much
compared to the enormous contribution  of  ever faster hardware.   And
don't  forget, Deep  Blue   is better than  all  the  current programs
without  even using a selective search,  showing  that the speed's the
thing!

The question that is so interesting to me is whether Go can benefit in
a big way from better hardware.   HOWEVER, I strongly believe that for
now, the biggest  gains  will be in software  because  I think  we are
where Chess was before it was figured out how  to write good programs.
Until that is done in Go, hardware might not be very important.

As an example, in chess, when  it was discovered that check extensions
were quite cheap and yet so powerful, this simple algorithm could make
a   chess program  find combinations that    it  would otherwise  take
computers running hundreds or thousands of times  faster to find!  But
once all  programs are fairly close to  optimal in this area, then the
one running on the fastest hardware is the one that  will tend to find
checking combinations the fastest.

Don