Re: [computer-go] Computer Go tournament at EGF

[Don Dailey <drd@xxxxxxx>]

Hi Rich,

I think you are getting hung up on the analogy I used and not getting
the deeper point.  I think most of the group understood it was about the
relationship between computing power and GO playing strength.

You see, the computer is purely hypothetical.  It is not something you
are supposed to try to build.  Think of it like you might think of a
"plane" in geometry, not a real physical object, just a mathematical
concept.

When I was in high school the teacher made the mistake of comparing a
"plane" to a "very large flat sheet of paper" and several of the
children couldn't get past the physical properties of a piece of
paper!

In computer science, there is a device called a "Turing machine."  A
turing machine is a hypothetic gadget that can never be built.  A pure
turning machine has infinite memory and this is probably the point
that probably would confuse some of us.  It is like the "plane" in the
sense that it is an abstraction, not an actual device or physical
object.

But nevertheless, it is an incredibly useful little gadget.  It is
used much as I tried to use the hypothetical Go playing engine, to
prove things via the use of thought experiements.

So anything you described about the physical properties of computers,
the age of the universe, the speed of light, number of grains of sand
on the seashore, isn't really very interesting or relevant, these are
facts everyone on this group probably are aware of and has nothing to
do with my argument.

Let me try to dumb this down in a way that requires less imagination:

  1. Imagine a real machine that consists of a network of workstations.

  2. Imagine taking GnuGo and running a copy of it on each workstation.

  3. Imagine software that builds a big networked search based on the 
     move generation and evaluation algorithms of GnuGo on this network.

  4. Try to envision the possiblity that such a beast would play much
     better than GnuGo run by itself.   Perhaps even being able to give
     GnuGo a several stone handicap.

  5. Don't forget that such a machine cannot work without consuming 
     significantly more resources than GnuGo running on a single computer.

In my opinion, such a machine would be a wonderful invention.  Would
you say that such a machine demonstrates the relationship between
computing power and Go playing strength?

If I haven't stretched anyones imagination too much, could it be
possible that a far bigger machine might be capable of playing even
better?  It doesn't have to be so big it can't fit in our physical
universe, I know this concept would be too much for some of us to deal
with.



- Don






   Don Dailey wrote:

   > The point he is making is that you can't isolate the execution time of
   > the algorithm from the playing strength.  It IS in fact easy to write
   > a perfect go program given no time constraints (just search until
   > solution.)  The issue isn't how GOOD your program is, it is how good
   > it is in a given amount of time.

   I've heard that assertion many times before, that "It IS in fact easy
   to write a perfect go program given no time constraints (just search
   until solution.)"

   But there is an _inherent_ time constraint:  It's the age (and the
   expected age) of the universe.  It has been shown that it would take
   longer than the duration of the universe to perform the search, even
   if there were enough storage.

   Furthermore, the number of go positions one would have to store is
   larger than the (finite) number of atomic particles in the universe,
   so even if there were an _infinite_ amount of time (which there isn't,
   because the universe does have an age) there isn't enough storage
   space to keep track of the results.

   Similarly, even if there were an _infinite_ amount of storage, (which
   there isn't, because the number of atomic particles in the universe
   is finite) there isn't enough time to perform the search!

   Such problems -- those that require a longer time than the age of the
   universe, or that require more storage than the number of atomic
   particles in the universe are called "inherently intractable".

   And that's why _not_even_ brute-force will work.  It is _NOT_ possible.

   The assertion that it is "easy to write a perfect go program given no
   time constraints" is simply not true.

   Not even theoretically.

   That said, I _still_ think that a program that can play a perfect
   game given a week is of more intrinsic value than one that can play
   a flawed game in an hour.  It all depends upon what is considered
   truly important for the program to be able to do.

   Thus we see that the issue is entirely subjective.

   What is _not_ subject to interpretation, is whether it is theoretically
   possible to write a perfect program that uses brute force, "given enough
   time" or "given enough storage".  It absolutely is not possible, let
   alone "easy".

   -- 
   Rich

_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/