From: Michael Vassar (michaelvassar@hotmail.com)
Date: Tue Aug 23 2005 - 12:25:54 MDT
It seems to me that the claim that the prior probability that exploits are
possible is about equal to the probability that the are impossible is not
reasonable. We should look at formal systems with complexity comparable to
that of the computer hardware and ask how frequently such systems have
historically turned out to have "exploits", where "exploits" = sufficient
complexity to accomplish a particular appearently impossible goal. In daily
life, we call such systems "games". Exploits are equivalent to "ways of
winning from an inferior position". I would assert that analytically
intractible games essentially Always contain exploits. That is to say, I
hypothesize that an experienced player of an arbitrarily structured,
sufficiently long in duration, and analytically intractible game will be
able to reliably beat an inexperienced player of no greater intelligence. I
haven't tested this hypothesis on arbitrarily structured games. It seems
fair to me to state that the hardware of an AI cannot practically be as
simple as an analytically tractable game. The first turn victories that are
possible in Magic the Gathering seem to me to be a particularly exploit-like
example from historically existing games. The ability of skilled players to
put together effective decks out of appearently "hopeless" cards would be
importantly parallel to the ability of an UFAI to use exploits.
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:52 MDT