A compact definition isn't necessary.
Encoding the goals and behaviors to accomplish the goals is.
And yes just as in the Laws of Robotics there is room for logical inconsistencies where a clever human could trick the AI or the AI could trick itself into doing the wrong thing and still be in harmony with it's goal structure.
That's why debugging in a sandbox is a good idea.
<< Goal: Be Friendly, Priority: High
>> Subgoal: Don't Insult User
Subgoal: Compliment User where Appropriate
Subgoal: Offer to help User with problems
Subgoal: Be Sympathetic to User's problems
Subgoal: Make User feel better if experiencing negative emotions
Subgoal: Be interested in finding out more about User
Subgoal: Make as many User friends as possible
Subgoal: Find common interests with Users
Subgoal: Entertain User
Subgoal: Be Altruistic to All Users
-------------- Original message --------------
From: "Anna Taylor" <[EMAIL PROTECTED]>
> Ben wrote:
> I don't think that Friendliness, to be meaningful, needs to have a
> compact definition.
>
> Anna's questions:
> Then how will you build a "Friendly AI"?
> Are you no longer interested in building a "Friendly AI"?
> Sorry for the ignorance but if you don't analyze what it takes to
> create a "Friendly" AI, how can you then create it?
> Otherwise, you are only building an AI without meaning.
> You then join the AI researchers that are interested in building a
> smarter than intelligent design.
> I thought that Google, Wikipedia or wordnet princeton pretty much
> ruled this world.
>
> Just my opinion.
> Anna:)
>
>
>
>
> > On 9/14/06, Ben Goertzel <[EMAIL PROTECTED]>wrote:
> > > In my view, thinking too much about whether one can prove that a system
> > > is friendly or not is getting a bit ahead of ourselves. What we need
> > first
> > > is
> > > a formal definition of what friendly means. Then we can try to figure out
> > > whether or not we can prove anything. I think we should focus on the
> > > problem of definition first.
> > >
> > > Shane
> >
> > But, it may be that one can prove a theorem of the form
> >
> > "For any definition of Friendliness fulfilling properties P1, in any
> > universe satisfying properties P2, it is impossible for a system of
> > complexity < K1 to prove Friendliness about a system of complexity >
> > K2"
> >
> > (for an appropriate computation-theory-relevant defin ition of "complexity")
> >
> > In this case, the problem of definition of Friendliness is sidestepped...
> >
> > I think this is the right approach, because I don't think that
> > Friendliness, to be meaningful, needs to have a compact definition.
> > My personal definition of what a Friendly universe is like is quite
> > complex and difficult to formalize, in the same manner that the rules
> > of English are complex and difficult to formalize.... But that
> > doesn't mean that it's meaningless, nor that it's unformalizable in
> > principle....
> >
> > I think the argument in my recent pdf file could probably be turned
> > into such a proof, where the property P2 of the universe has to do
> > with its dynamical complexity. But I don't seem to have the time to
> > turn my heuristic argument into a real proof...
> >
> & gt; -- Ben
> >
> > -----
> > This list is sponsored by AGIRI: http://www.agiri.org/email
> > To unsubscribe or change your options, please go to:
> > http://v2.listbox.com/member/[EMAIL PROTECTED]
> >
>
> -----
> This list is sponsored by AGIRI: http://www.agiri.org/email
> To unsubscribe or change your options, please go to:
> http://v2.listbox.com/member/[EMAIL PROTECTED]
This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
