Re: Questions for Eliezer, SIAI and consciousness

From: Mitchell Porter (
Date: Tue Oct 26 2004 - 18:37:52 MDT

sam kayley wrote:
>What probability do you give for consciousness involving strange physics?

I think it's more enlightening and more important to ask whether
consciousness involves strange *meta*physics, and whether an AI
can be expected to notice this.

In the 20th century, physics was elevated above metaphysics; indeed,
metaphysics was variously attacked as unknowable, meaningless, or
irrelevant. Within academic philosophy, metaphysics has made a quiet
comeback, but it continues very much to play second fiddle to physics.
Metaphysics today (at least in the English-speaking world) cringes before
physics; it is an exercise in clarification and interpretation. Cause,
substance, space, time, being - these were the concerns of metaphysicians,
before mathematical physicists ever came along. Now we have the new
phenomenon of tenth-decimal-place agreement between experiment
and equation, but an equation by itself is not a theory. Even a set of rules
relating experimental measurements to an equation is not much of a theory.
A "theory of everything" is a statement about existence. It should say what
exists, what sort of things exist, why they exist, it should even say
about what existence itself "is". By that criterion, we are nowhere near
having a theory of everything. We have an empirically adequate "standard
model", we have a variety of mathematical conceptions touted as the next
or final theory, and then we have a variety of conflicting metaphysical or
anti-metaphysical opinions regarding the meaning of the equations, which
form a new chapter in the long history of philosophy's inability to reach
consensus on anything.

Do we have any idea how to make an AI that can answer metaphysical
questions? We don't. At best we know how to get it to work within a
pre-ordained philosophical framework - and most likely that will be a
framework derived from the 20th-century storehouse of anti-metaphysical
systems (pragmatism, positivism, linguistic philosophy). If aspiring AIers
want a concrete benchmark, I'd suggest Kant's *Critique of Pure Reason*.
If you can produce a cognitive-scientific account of what Kant was doing
when he wrote that book, you may be making progress towards AI that
is philosophically capable.

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:49 MDT