From: Stathis Papaioannou (firstname.lastname@example.org)
Date: Sat Dec 01 2007 - 06:32:49 MST
On 01/12/2007, Aleksi Liimatainen <email@example.com> wrote:
> Severe failures seem extremely rare in light of the usual messiness of
> human behavior. I suspect it's because parents' love for their children
> is about as close to an explicit supergoal as humans get. "Friendliness"
> really wasn't the right term but I do find the parallels interesting,
> especially given recent discussion.
Society has mechanisms to deal with parenting failure: children can be
removed from abusive parents. This is because *no-one* is to be
completed trusted; they could go mad or be corrupted in some way, or
cause terrible harm when they honestly intend to do good. I don't see
why an AI should be trusted more than a child's parents. We have to
expect that, even if the fault ultimately lies with the programmers,
AI's will at some point kill humans, just as most of our other
machines capable of killing have at some point killed.
-- Stathis Papaioannou
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:01:01 MDT