From: Tennessee Leeuwenburg (tennessee@tennessee.id.au)
Date: Tue May 16 2006 - 03:03:48 MDT
I propose that we take this the hell out of SL4 and into some
philoso-ranting list. I agree that it's relevant to the issue, but I don't
feel like it's got a damn thing to do with AGI beyond the obvious
fundamental debate on the issue.
Here's my response.
1.) Moral relatavism. Bummer, huh?
2.) AGIs have a lot of power, don't they?
3.) ???
4.) Profit
I think that if you resolve the fundamental questions surrounding moral
relativism, you'll also resolve the AGI debate. If morality is objective,
and that's discoverably, provably so, then we have nothing we can rightfully
worry about from an AGI system.
If morality is objective, but (a) hard to discover and (b) easy to mistake,
then we're in trouble. But we're less in trouble than an AGI, which is
presumably more able to deal with (a) and less likely to fall for (b).
If morality is purely relative, then by definition we cannot instill an AGI
with our own morality (otherwise it would, from the perspective of the AGI
at least, be objective -- a given)
Q1.) Can we take our relative morality and make it objective for the AGI?
Not without understanding AGI we can't. Hmmm, tough one.
Q2.) Can we force an AGI to value our relative morality, or at least our
desires? Not without understanding AGI we can't. Hmm, tough one.
As you can see, this problem boils down to two things:
(a) Debates about morality
(b) Understanding AGI
I suggest that (a) is not SL4 or easily solvable, and that (b) is what we
should concern ourself with?
On 5/16/06, Olie Lamb <neomorphy@gmail.com> wrote:
>
> The old bugbear about letting people do what they like without
> imposing on others has reared its ugly head.
>
> In this case, rather than a set of rules for the people, it's an
> "operational methodology" for an expected superhuman intelligence,
> that might become a sysop.
>
> Some ethicists have said that
>
> (**Stipulative characterisation**)
>
> Morals = preferences that you want to apply to everyone.
>
> F'rinstance, if you don't like bullfighting because sport bores you,
> that's a matter of individual preference. If you don't like anyone
> liking bullfighting because it hurts the bull, by the above
> definition, it's a "moral statement".
>
> (Nb: this is NOT my definition. I'm just using it for one post.)
>
> With this characterisation, it's very hard to imagine an
> anthropomorphic Sysop not effectively enforcing their "morality" on
> others. Their operational methodology for weighing the requirements
> of conflicting expressed wills would, in effect, be the Sysop's
> "morality".
>
> Just say that a Sysop adopted m.l.vere@durham.ac.uk 's two axioms:
> > 1. Prohibiting any action which affects another member of the group,
> > unless that member has wilfully expressed for that action to be
> > allowed (a form of domain protection).
>
> (Nb: can you say "Golden Rule"?)
>
> > 2. Giving all group members equal resource entitlement
>
> Would you expect such a sysop to not only enforce the axioms directly,
> but also for others to adhere to them where they were operating
> outside the Sysop's influence? As in, would you expect a Sysop to
> allow Robin to voluntarily accompany Leslie into the woods*, when
> Leslie has admitted that Leslie has a secret plan to "affect another
> member of the group" with an action that is has not been allowed by
> that member of the group, (eg: maim, torture, kill etc Robin).
>
> * Yes, although a Sysop would normally have influence on temperate
> forested areas, shut up.
>
> Of course the Sysop is going to influence others to adhere to its
> moral axioms. Leslie and Robins future actions might take place away
> from the Sysop's field of influence, but the Sysop will always be
> making actions that affect the future, because you can't make actions
> that affect the present! (Insert TangentT here)
>
> Brief ad hominem interlude...
>
> If you expect others to respect your domain, what's that but a form of
> morality? Hell, you even suggest giving resources out equally.
> Communist! I happen to own large tracts of land that have more than
> 1/6billionth of the planet's solar collection potential and also
> fossil energy reserves buried beneath.* You ain't stealing my land/
> energy resources!
>
> * This is a lie. My point is that one 2006human's share of the
> earth's crust is 85ha, less than what some people own.
>
> Back to sysops...
>
> If the sysop is vastly more powerful than other entities, it may be
> able to act in a genie-like way, and grants wishes that don't
> interfere with other human's "domains". Why should humans/post-humans
> be forced not to interfere with each others domains?
>
> For a Sysop, because "might makes right" ;P
>
> Otherwise, because there might be some (objective?) reason not to.
>
> Furthermore, why should the sysop not adversely affect humans?
> Because the Sysop's progenitors decided to make it that way.
>
> As long as an AI is (1) taking actions that affect others (2) Weighing
> the (conflicting interests of other parties (3) weighing its interests
> against those of other parties, it would need some sort of methodology
> to evaluate potential courses of action. Those that it chooses could
> be called its "preferences".
>
> If the AI-builder thinks that the AI should selfish (!!don't try this
> at home, kids!!), the Builder is projecting their preferences onto
> others. The AI doesn't even need to be conscious for the AI-builder's
> preferences to match the stipulative definition of "moral statements"
> above.
>
> -- Olie
>
> On 5/15/06, m.l.vere@durham.ac.uk <m.l.vere@durham.ac.uk> wrote:
> > So, where would i take my 'moral nihilism'? The reasons I advocated it
> are the
> > following:
> >
> > All morality is artificial/manmade. This is not an intrisnic negative,
> however
> > it is negative in this case, as:
> > 1. Morality made by mere humans would very likely not be suitable/a net
> > postivie for posthumans. Therefore we need to go into the singularity
> without
> > imposing morality on our/other posthumans (ie as moral nihilists).
> > 2. As morality is artificial, there is no one (or finite number of)
> 'correct'
> > moralit(y)/(ies). Thus it would be better for each individual posthuman
> to be
> > able to develop his/her/its own (or remain a nihlist), than have one
> posthuman
> > morality developed by a sysop.
> >
> > At the moment, what i would advocate is that
> > universal egoists (or moralists who dont want to constrain others with
> their
> > morals) build
> > a sysop which grants them all complete self-determination in becoming
> > posthuman. My ideas so far (written previously):
> >
> > "The best posssible singularity instigator I can imagine would be a
> > genie style seed AI, its supergoal being to execute my expressed
> > individual will. From here I could do anything that the person/group
> > instigating the singularity could do (including asking for any other
> > set of goals). In addition I would have the ability to ask for
> > advice from a post singularity entity. This is better than having me
> > as the instigator, as the AI can function as my guide to
> > posthumanity.
> >
> > If anyone can think of better, please tell.
> >
> > The chances of such a singularity instigator being built are very
> > slim. As such I recomend that a group of people have their expressed
> > individual wills excecuted, thus all being motivated to build such
> > an AI.
> >
> > The problem of conflicting expressed wills can be dealt with by
> > 1. Prohibiting any action which affects another member of the group,
> > unless that member has wilfully expressed for that action to be
> > allowed (a form of domain protection).
> > 2. Giving all group members equal resource entitlement
> >
> > The first condition would only be a problem for moralists and
> > megalomaniacs (and not entirely for the latter as there could exist
> solipism
> > stlye simulations for them to control).
> > The second seems an inevitable price of striking the best balance
> > between the quality of posthumanity and the probability of it
> > occuring.
> >
> > I tentatively recomend that the group in question be all humanity.
> > This is to prevent infighting within the group about who is
> > included, gain the support of libertarian moralists and weaken the
> > strength of opposition - all making it more likely to happen.
> >
> > This is a theory in progress. Idealy, we would have an organisation
> > similar to SIAI working on its development/actuallisation. As it is,
> > Ive brought it here. Note, I hope to develop this further (preferably
> from
> > the standpoint of moral nihilism).
> >
> > Whilst the AI interpereting commands may be an issue, I dont see it
> > as an unsolvable problem.
> >
> > Note: I see this as a far better solution to singularity regret than
> > SIAI's CV."
> >
>
>
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:56 MDT