The "AI Box" experiment

From: Fabio Mascarenhas (mascarenhas@acm.org)
Date: Fri Mar 08 2002 - 18:00:55 MST


The only way for Eliezer not to win is if you decide to keep locked *all*
SIs no matter what they say, but then you'll also keep the Friendly ones
locked, so why bother creating one at all?
 
Classifying an SI as Friendly/Unfriendly just by talking with it is
undecidable. If you try to guess if it's Friendly then Unfriendly SIs just
have to pretend they're Friendly for long enough that you let them out.
 
Fabio.
 
> ----- Original Message -----
> From: "Thom Lawrence" <thom@tavenue.co.uk>
> To: <sl4@sysopmind.com>
> Sent: Friday, March 08, 2002 11:23 AM
> Subject: Re: The "AI Box" experiment
>
>
> > Nathan said...
> >
> > > I decided to let Eliezer out.
> >
> > Can anyone else play? :)
> >
> >
> > --Thom.
>



This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:37 MDT