From: Gordon Worley (redbird@rbisland.cx)
Date: Sun Nov 24 2002 - 07:50:36 MST
On Sunday, November 24, 2002, at 07:59 AM, Ben Goertzel wrote:
> In my view, Friendly AI isn't really an ethical/moral system. (Of
> course,
> Eliezer may disagree!). In my view, it might rather be thought of as
>
> -- trying to encourage a certain probability distribution on the space
> of
> ethical/moral systems.
>
> -- trying to promote certain ethical/moral principles that do not
> however,
> in themselves, make a whole & consistent system
>
> For example, Friendly AI is trying to encourage ethical/moral systems
> that
> involve significant compassion for all living/sentient organisms, and
> that
> involve respect for freedom of living/sentient beings. (It is doing
> more
> than this also, of course.) These are principles which some
> moral/ethical
> systems will agree with more than others. Friendly AI wants to
> encourage
> future superintelligent AI's to evolve moral/ethical systems that will
> agree
> with these principles...
Nope, I think you misunderstand here.
Friendliness is a level up from morals: meta-morality. Friendliness
is, among other things, a way to find the right morals/ethics,
regardless of what they might be. You could start off by teaching a
seed AI Friendliness-content that was very similar to the moral
thinking of Eliezer, but if Eliezer is wrong (and it's very likely that
he, like all humans, is to some degree wrong, and if he's completely
correct it was an accident) it won't matter because the FAI is capable
of figuring out how to fix its understanding of morals to be aligned
with the morality of the Universe.
Maybe we've got it all wrong and what we have been calling good is
actually bad, like when demons in hell turn the words around and
consider bad things good. It seems doubtful, but it is a possibility,
even if small. If that's the case, an FAI could figure it out.
-- Gordon Worley "Man will become better when http://www.rbisland.cx/ you show him what he is like." redbird@rbisland.cx --Anton Chekhov PGP: 0xBBD3B003
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:41 MDT