Re: "friendly" humans?

From: Samantha Atkins (samantha@objectent.com)
Date: Fri Jan 02 2004 - 19:21:48 MST


On Fri, 2 Jan 2004 14:37:44 -0800 (PST)
Lawrence Foard <entropy@farviolet.com> wrote:

> On Fri, 2 Jan 2004, Perry E. Metzger wrote:
>
> > There is no definition of "The Good", so it would be rather difficult
> > to transfer the desire to do it to a creature when we can't really
> > determine what "it" is ourselves.
> >
> > (I strongly argue that there is no such thing as "the good" outside of
> > the context of a subjective set of desires on the part of one desirer,
> > but that's another story.)
>
> Actually the second paragraph is very important. A 'Good' AI would have to
> be one thats sufficiently human to share human desires. What we consider
> pain and suffering is just a certain set of nerve impulses in a certain
> configuration. There is nothing universal about this that implies that
> this pattern of nerve firing is 'Bad'.
>
I disagree. I also disagree that the "good" is more or less synonymous with what we consider pleasurable (versus painful) or even desirable. A sufficiently intelligent non-human should be able to come to a reasonable grasp of what is "desirable" by humans though.

> You might be able to make a general statement. That which an organism
> acts so as to continue experiencing is good, that which an organism seeks
> to avoid is bad. But it has some problems already.
>

Starting with "organism", continuing to habitual behavior and leaving out the goals, dreams and aspirations, highest potential and so on that get all wrapped up in the word "good".

> Of course that tells you that heroin is good and work is bad. So your AI
> goes and ships everyone off to the "work free drug place" after making
> some adjustments to there opiate systems so they never gain a tolerance,
> or die from massive doses of all the best drugs a super smart being can
> whip up ]:)
>

Well, if we were designing Friendly AI by your notion of "good" above that would be a danger. :-)

- samantha



This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:43 MDT