On Feb 24, 4:36 am, "Krismoor" wrote:
[...]...if it is not possible to determine
"perfect play", how can it be possible to programme computers to play
the perfect game ?

There's no difficulty writing a program that *in principle*
will win any won chess position and draw any drawn position; getting
that program to produce answers within a decent time limit is another
matter! But not really the point [especially as I was addressing the
current situation and its difficulties, not some hypothetical future
in which a computer can analyse chess fully in real time]. ...

Will it be right to say that all lines leading to a
"sure win" can be treated as perfect play? Whether a player exchanges
pieces to arrive at a won end game, or goes for a middle game win,
both should be treated as "perfect play".

... Won positions aren't "the problem". If the initial position
turns out to be a forced win for White [or indeed for Black], then
it's
possible that we will see some brilliant manoeuvres and go "Wow!";
but
it's much more likely that the win will be no more comprehensible than
those longest wins in KBBvsKN. But suppose that, as seems more
likely,
the initial position is drawn; and further that the drawing margin
is,
in some sense, quite wide -- each player has to commit a perceptible
error to lose. What then?

Well, one consequence is that, as *any* draw-preserving move is
as good as any other, a game such as 1 c3 Na6 2 g4 f6 3 Nh3 Kf7 ... is
just as perfect as the sort of game that we currently expect of GMs.
As
long as neither side strays into a forced loss, no error has been
made.
Positions are won, drawn or lost; perfect chess has no concept of
"White
has a small advantage" or "the pawn structure is weak, but there is
some
compensation because ...". Chess programs tend to assume that the
other
side sees the same things and evaluates them the same way; the
concept
that *this* position is a "dead" draw whereas *that* one may be drawn
with best play but a fallible opponent is likely not to find that play
is
very human. Ideas such as "only moves" can help to provide a program
with
that concept, but they're only a tiny part of it.

Note that in such a case, the "best play", even in chess, has
become probabilistic and psychological. The best move against a
Karpov
may be different from the best move against a Fischer or a Kasparov or
a
Kramnik. Chess at the very highest levels becomes, in a sense, no
longer
a game of skill [for all the top players have that skill in
abundance],
but a game of chance. A perfect program knows that moves A, B and C
draw
but D, E and F lose; so what are the prospects that an imperfect
player
will miss A, B, C and play D instead?

--
Andy Walker, School of MathSci., Univ. of Nott'm, UK.