on the HAL-9000

RE01 Rice Brian T. EM2 BRice@vinson.navy.mil
Sat, 10 Oct 1998 18:56:28 -0700

Perhaps I should explain the relationships I see between the HAL-9000 of
sci-fi movie fame (2001: A Space Odyssey and 2010: Odyssey Two) and the
TUNES system.  Obviously, there are aspects which are simple to see.

	the HAL-9000 was a speaking/thinking/seeing/hearing supercomputer
embedded into a ship's engineering system. therefore, it had a set
hardware specification. the system, as far as we know was not
distributed, or was simply symmetrical and/or static. it did not have to
solve the problems associated with general networked systems. it's
low-level format was set statically, for the most part.

	what the hal-9000 DID have was a fully general reflective
linear-programming system, probably based on LISP. HAL could play chess
as well as it could this is mostly what accounts for its famous
behavior. for those unfamiliar with the story, here it is:

the government discovers an alien artifact buried in dust on the moon.
it is perfectly featureless and sends a radio signal on contact with the
humans to a larger similar object near Jupiter. the movie is set on a
manned expedition to Jupiter with HAL managing the mission with two
humans. (six others are in cryostatic hibernation). on arriving near
Jupiter, the computer claims that the high-gain antenna will fail soon,
but the astronauts find no problem. the HAL-9000 claims it is human
error, since it is designed to be 'perfect and incapable of error'. when
the astronauts become worried, they attempt to disable HAL, but HAL
manages to kill one of them in EVA as well as the sleeping six. the
other survives and makes it back inside the ship and de-activates HAL's
brain. he THEN learns of the ship's true mission to rendezvous with the
artifact, and attempts to do so himself.

the point is that HAL was told by the U.S. government to lie about the
mission as a prime directive, which HAL, being a Lisp-machine, evaluated
linearly. in other words, HAL did not malfunction, it merely solved the
'central' problem of the lying issue in the most direct manner:
eliminate the human subject to the lying directive, in order to complete
the mission alone. this, i believe, displays the limits of
linear-programming rather elegantly. i believe that a computational
system which acts on the meta-protocol that 'it has all the facts' in
any situation will not act intuitively, and would exhibit
characteristics which the average person would fail to account for in
dealing with it. more later...