[ExI] Unfrendly AI is a mistaken idea.

[Lee Corbin]

Stathis writes

> [Lee wrote]
> 
> > Still, right now the environment favors---in strictly population biologic
> > terms---those who have as many children as they can and who allow 
> > the state to raise them (e.g., they pay exactly zero for medical care,
> > relying only on ER services).  That's the path of the immediate future.
> 
> But this isn't happening as a plan: most men would be horrified at the
> idea that they might have all these illegitimate children running around,

Then most men are relatively "unfit".  But nature fixes that sooner or later.
"Bad" memes are partly responsible.

> and societal pressures are to minimise teenage pregnancies (which anyway
> are higher in countries and states with less welfare spending).

The higher reproducing males will just learn (through selection) how to avoid
those societal pressures.  And it doesn't surprise me that teenage pregnancies
are higher in countries and states with less welfare spending.

> On the other hand, great effort is expended by people acquiring wealth,
> or breeding flowers, or any of a million other things which at best could
> be seen as only indirectly contributing to evolutionary fitness,

That's because it *used* to contribute to evolutionary fitness.  Nature
cannot respond overnight. It takes the genetic frequences a long time
to respond to a changed environment.

> If your argument is that an AI might spontaneously arise that wants to
> dominate everyone, which would then dominate all the other AI's and
> lesser life forms, the same argument would apply to human reproduction,
> yet it just isn't something that anyone seriously worries about.

But they should, pace Singularity and GE and other changes coming soon.
Right now, we're in a dysgenic death spiral.  (Again, I'm not too worried
because weird and unpredictable things will probably happen sooner.) 
But we already have socialized medicine in that the most biologically
viable lifestyle is to have many children and avoid paying for them.

> Such a dominant AI would be held in check by at least as many factors
> as a human's dominance is held in check: pressure from society in the
> form of initial programming, reward for human-friendly behaviour,

You seem as absolutely sure that the first programmers will succeed
with Friendliness as John Clark is absolutely sure that the AI will
spontaneously ignore all its early influence.  We just don't know, we
cannot know. Aren't there many reasonable scenarios where you're
just wrong?  I.e., some very bright Chinese kids keep plugging away
at a seed-AI, and take no care whatsoever that it's Friendly. They
succeed, and bam!  the world's taken over.

> and the censure of other AI's (which would be just as capable,
> more numerous, and more likely to be following human-friendly
> programs). 

But how do you *know* or how are you so confident that *one*
AI may suddenly be a breakthrough, and start making improvements
to itself every few hours, and then simply take over everything?

Lee