Re: Why would AI want to be friendly?

From: Barbara Lamar (
Date: Sun Oct 01 2000 - 14:11:28 MDT

On Sun, 1 Oct 2000 02:10:17 -0700 (PDT) Eugene Leitl
<> writes:
> Eliezer S. Yudkowsky writes:
> Friendly is not observer-invariant, goddamit. If you want to impose
> your particular concept on the rest of us via (thankfully
> hypothetical) AI proxy, then please call a spade a spade. Call it
> Eliezer-Friendly, or something.

I brought this same point up towards the beginning of this thread,
addressed specifically to Eliezer, and there was no response.

Two days ago I posted a URL for a recent paper by a couple of people who
are currently working in the AI field--"Cooperative Coevolution: An
Architecture for Evolving Coadapted Subcomponents" by Mitchell A. Potter
(Navy Center for Applied Research in Artificial Intelligence) and
Kenneth A. DeJong (Computer Science Dept. George Mason University).
Published in *Evolutionary Computation* 8(1) (2000)

Again there was no response. As an outsider to the AI field, I would
nevertheless like to have a better understanding of it. Has anyone else
read Potter's and DeJong's paper?

Some questions I have include:

1. What difference would cooperative coevolution make with respect to the
relationship between humans and highly evolved AI?

2. Are the capabilities of AI severely limited in the absence of
algorithms for cooperative coevolution?

3. Potter's and DeJong's EA's call for destroying unsuccessful
"species" and maintaining "genetic" diversity by introducing new species
whenever stagnation is detected in the system. Stagnation is detected by
monitoring the quality of interspeices collaborations via checking each
collaboration for the improvement it provides in the functioning of the
"ecosystem" as a whole. Is this methodology troubling with respect to
possible future interaction between humans and AI? Why or why not? What
are some other methods of maintaining diversity?

4. Do the "species" in this research correspond more closely to
biological genes or to biological species?

> And, of course, it is clearly obvious what is best
> mirroring reality and what not, without being given a chance of
> being
> evaluated by reality, since that wouldn't truncate the
> particular-metric-Unfriendly behaviour space region.

Oh, I see now! This makes it all perfectly clear.


Juno now offers FREE Internet Access!
Try it today - there's no risk! For your FREE software, visit:

This archive was generated by hypermail 2b30 : Mon May 28 2001 - 09:50:14 MDT