Re: How to make a slave

From: Harry Chesley (
Date: Fri Nov 23 2007 - 10:23:57 MST

On 11/23/2007 8:44 AM, John K Clark wrote:
> On Fri, 23 Nov 2007 "Stathis Papaioannou" Wrote:
> > If the AI starts off with "the aim of life is X", then it will do
> > everything it can to further X. It doesn't matter what X is, or how
> > many iterations the AI goes through. […] humans do allow their
> > supergoals to vary
> So nothing can change their “super goal” except for a human being
> because they are special, they are made of meat and only meat can
> contain the secret sauce.

I think you're misinterpreting what people are saying here. I have yet
to see a post to this list that said or implied that meat is somehow
special. Maybe I should let the actual posters explain themselves, but I
really haven't seen that.

Goals will necessarily come from one of two places: pre-programmed or
random. This is true of people as well as machine GAIs. Random top-level
goals are arguably a design flaw. Pre-programmed goals come from us in
the case of machine GAIs, from evolution in the case of people.

So, are you arguing that it's impossible to eliminate the presence of
random top-level goals? Or that it's impossible to predict the actual
result of pre-programmed top-level goals? Or that new non-random
top-level goals will get introduced from thin air?

However, let me add one other point: In humans, I do not believe the
goal set is strictly hierarchical. This means there are no permanent
top-level goals, though there are persistent goals. This often results
in people doing things to achieve local goals that detract from
accomplishing what you'd think were their top-level goals. The current
top-level goals vary based on current circumstances -- i.e., randomly
based on external inputs. Computer GAIs might have the same problem, but
don't have to.

[BTW, the condescending tone of many of your posts does not further your
arguments. It just makes the posts unpleasant to read. Same comment to
others who take a similar tone.]

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:01:00 MDT