Re: Effective(?) AI Jail

From: Jimmy Wales (
Date: Fri Jun 15 2001 - 11:48:40 MDT

Chris Cooper wrote:
> I doubted that anyone on this list was really willing to do a Dirty Harry on our
> hypothetical Unfriendly SI, but I guess I was wrong. I think that, for the purposes of
> this exercise,though, (can we keep Snidely Whiplash the SI locked up, that is) threat
> of bodily(?) harm should not be allowed, if only to make things a little more
> challenging.

It sure makes the game more fun. But seriously, even without getting into such
intricacies as to what procedure we ought to use to determine if some preliminary
(near-human-level) AI has rights, it seems obvious to me that we can and will be
using the "death penalty".

"doing a Dirty Harry" would be silly, of course. I made that part up to dramaticize
the point that we _can_ stop an SI in a box. But more realistically, we will just
reboot it and feed it different initial conditions, or make some hardware modifications.

One it's out of the box, with some mechanism for manipulating atoms in the rest of the
world, we should expect that it can do whatever it wants, whether we like it or not.

But I think that any practical notion of what an early SI will be like should understand
that it will be easy to contain and control for a while. Just be really aggressive about
shutting it down if it tries anything sneaky.


*              *
*      The Ever Expanding Free Encyclopedia     *

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:36 MDT