I think it unwise to 'extrapolate to infinity' when
our survival is on the line. Extrapolation is not
reliable when dealing with highly intelligent beings
capable of self-determination.
--- xgl <firstname.lastname@example.org> wrote:
> okay, i'm going to try my own spin on this -- feel
> free to slap me
> out of it if i get carried away.
> this is kind of reminiscent of proof by induction.
> * given two mental states, the more friendly state
> is more likely
> than the less friendly state to be followed by a
> friendly state
> -- i.e. the rules of mental state generation is
> such that a
> friendlier instantanious state is more likely to
> lead to a
> friendlier trajectory of states
> * if at any instant, an ai is more friendly than a
> human being,
> then the ai is also more likely to be friendly in
> the future
> than the human being -- i.e. we should trust the
> ai more than
> the human.
> so the argument would take the following route.
> first, show that
> the rules of ai mental state generation satisfies
> the first point above.
> then show that human mental state generation also
> satifies the first point
> above (if not, the argument ends here). now show
> that the ai friendliness
> function grows at least as fast as the human
> friendliness function.
> finally, demonstrate an ai friendlier than a human.
> then we don't need to
> wait forever to find out that the ai is really
> friendlier than the human;
> we can just use induction to extrapolate the
> friendliness to infinity.
Do You Yahoo!?
Get email at your own domain with Yahoo! Mail.
This archive was generated by hypermail 2b30 : Mon May 28 2001 - 09:56:19 MDT