[FOM] Re: Chess as science?

Timothy Y. Chow tchow at alum.mit.edu
Fri Feb 6 10:36:28 EST 2004


Harvey Friedman wrote:

> 1. There could well be a reasonably coherent theory that is able to
> single out certain chess positions - constructed or naturally occurring
> - and asserts that they are advantageous. And this may be highly
> nonobvious, maybe paradoxical.

This is certainly possible; I said as much in my first posting.  I think
all we disagree on is whether there is any hope of doing so any time soon.
In my mere opinion, solved chess endgames are the only arena where this is
conceivably within reach.  If anyone demonstrates serious progress for
"real" chess positions I will stand corrected.

> The dispute between Silman and probably almost all GM's, versus
> Berliner, it at a much deeper and much more interesting level than that.

It doesn't seem particularly deep to me, but maybe you understand the
debate better than I do, and why it's different from all the other
similar-sounding debates where grandmasters spout off opinions without
being able to prove them.

> >I can't think offhand of a practical example that
> > is like the solved endgames---where both sides maneuver *inexplicably* in
> > an apparently level, quiet position for dozens of moves with no
> > discernible mistakes but gradually the position becomes hopeless for the
> > losing side.
>
> Are you sure about this? My impression was just the opposite. I have seen
> quotes from great players that they find the machinations of other great
> players as inexplicable.

Quotes are potentially misleading; can you show me games that have been
analyzed to death and where the win still remains inexplicable?  That
would be more convincing than an offhand quote that might be designed to
enhance the mystique of a particular admired player or of chess in
general.

> Super GM's are not going to give away great insights - assuming they have
> them - for free. They won't even play serious chess for free (smile).

Yes, I'm fully aware of this.  They certainly keep their opening surprises
secret (but eventually have to reveal them in order to make any use of
them).  They play secret matches.  And it's always possibly to theorize
that they are keeping amazing strategical insights secret, since there is
no way to disprove such a theory.  What I'm saying is that their games and
their observable behavior are much more consistent with the assumption
that they do not have such a secret theory of chess.  For example:

> CONJECTURE: sometime this Century someone will have such great insights
> into the TRUTH that they will crush their opponents (in matches with or
> without computers allowed) without spending any serious time on opening
> analysis.

The fact that this hasn't happened yet is, in my mere opinion, evidence
that the super GM's don't have these alleged great insights into the
TRUTH.

Or look at the 1999 Kasparov-versus-the-World chess match, where Kasparov
got seriously worried about the outcome of the game and enlisted the help
of a top-notch computer program and a computer operator.  Why would he do
that if his intrinsic chess knowledge was so much superior that he could
beat the rest of the world on that basis?

Here is another positive suggestion.  A potentially fruitful direction for
foundational research is the area of go computer programs.  This might
seem to contradict what I said before, because if anything, go is farther
from being solved than chess is.  What of my much-vaunted claim that one
needs a definitive evaluator of positions?  My response is that even if,
as I suspect, top go players are duffers compared to a perfect player,
they are so much stronger than computer programs that for the limited
purpose of creating stronger go programs, professional go players can
serve as "definitive evaluators."

Go players know the importance of shape, of gote/sente, of life and death,
and all sorts of other intuitive concepts.  However, these concepts are
still so poorly understood that they have not been successfully
implemented in computer programs.  The best programs can easily be
defeated by even a moderately strong amateur.  Successful clarification
of these concepts should lead to much stronger programs (and the people
who develop such programs will probably make a lot of money!).

Tim



More information about the FOM mailing list