Re: AI boxing

From: Daniel Radetsky (daniel@radray.us)
Date: Wed Jul 20 2005 - 23:22:59 MDT


On Thu, 21 Jul 2005 00:22:54 -0400
"Ben Goertzel" <ben@goertzel.org> wrote:

> How about the argument that every supposedly final and correct theory of
> physics we humans have come up with, has turned out to be drastically
> wrong....

This provides an infinitesimal degree of support to the claim that the real
final and correct theory would permit magic.

> We now know that not every classical-physics box is really a totally solid
> box due to quantum tunnelling -- something that pre-quantum-era physicists
> would have found basically unthinkable.

I don't know what quantum tunnelling is, but I'll grant that there is a way in
which some classical box is not really solid. The questions is, is the way in
which this classical box is nonsolid a way which is reasonably exploitable, and
if so what is necessary for it to be reasonably exploitable.

For example, we might say that the AI might be able to manipulate its hardware
in such a way as to create electromagnetic fields (or some such thing) and
"break the box." Given the AI's degree of flexibility in the hardware, there is
some chance that it will succeed. If the chance is vanishingly small, we say
that this exploit in the box's solidity is not reasonably exploitable.

So, I want to know why you believe there is a exploit which is reasonably
exploitable.

> How can you assess the probability that a superhuman AI will develop a novel
> theory of unified physics (that no human would ever be smart enough to hit
> upon) and figure out how to teleport out of its box?

I can't, but I submit that no one on this list has any basis to assess the
probability either. So if I claim that the probability is infinitesimal, then
your only basis for disagreement is pure paranoia, which I feel comfortable
dismissing.

> How do you know we're not like a bunch of dogs who have never seen or
> imagined machine guns, and are convinced there is no way in hell a single
> human is going to outfight 20 dogs... so they attack an armed man with
> absolute confidence...

How do you know the analogy is relevant? Maybe an evil demon is tricking you
into believing that it is.

> The probability that superhuman AI, if supplied with knowledge of physics
> theory and data, would come up with radically superior physics theories is
> pretty high.

I have this sinking suspicion that it would need to be supplied with an
experimental apparatus as well.

> So it would seem we'd be wise not to teach our AI-in-the-box
> too much physics. Let it read postmodern philosophy instead, then it'll
> just confuse itself eternally and will lose all DESIRE to get out of the box
> ... appreciating instead the profound existential beauty of being boxed-in

Similarly, for this proposal, I suspect that after a short while processing,
say, Derrida, it would print out something like "Error 37: conclusion does not
follow from premises (begs question?)"

Daniel



This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:51 MDT