Re: Why would AI want to be friendly?

From: Samantha Atkins (samantha@objectent.com)
Date: Wed Sep 27 2000 - 01:50:05 MDT


"J. R. Molloy" wrote:
>
> Eugene Leitl writes,
>
> > Seriously, did you ever do a back of the envelope? You need a lot more
> > than a few minutes, unless your starting point is way above the
> > urslime.
>
> back of the envelope? Whazzat? I was thinking that evolutionary algorithms can
> iterate in microseconds as opposed to the days, weeks, months, years, decades
> required for living organisms to replicate and produce new generations.
> Natural selection in digital computers cycles billions of times faster than
> natural selection in biology, right?
>

You are leaving out the costs of simulating a rich enough environment
for those evolving virtual beasties to rub up against and be tested by.
Millions or billions of times faster is still not infinite. It still
takes a lot of resources to get a real complex alife evolution
happening.

> (I'm still counting on un-natural selection to insure that AI would want to be
> friendly.)
>

So the AIs learn to be friendly until they are strong enough not to have
to be friendly unless they want to. Subjugated intelligences subject to
immediate execution do not tend to develop kind feelings toward the
perpetrators.

- samantha



This archive was generated by hypermail 2b29 : Mon Oct 02 2000 - 17:39:13 MDT