Re: computer-go: minimax and go

[Christian Nentwich <c.nentwich@xxxxxxxxxxxx>]

> The  universe is far too small,  light is too slow  to ever search the
> entire game  tree within  a few   billion years.   But my  argument is
> theoretical.  The practical goal is not even to play  good Go, it's to
> play "better" Go than we did yesterday with a computer.

Your argument up to here has been the following, as far as I can tell:

  1) Given infinite resources you could search the whole game tree
  2) Searching the whole game tree yields a perfect program

and then

  3) There are no infinite resources
  4) The distance to searching the full game tree is proportional
     to the program's playing strength

Is it worth pointing out that 4) is complete speculation ? There is
plenty of evidence of experimental Go programs getting weaker with
increased search depth, because their evaluation functions break under
the amount of trade-off to be done.

Of course it is easy to argue that programs should be developed so that
their strength increases as the search depth increases, but that is
equivalent to arguing that "people should try to write better programs".

In general, if you use heuristics that require trade-off decisions, then
by searching more deeply you will be rewarded by an exponential growth
of trade-offs to be made.

Christian