Re: Why would AI want to be friendly?

From: Eugene Leitl (eugene.leitl@lrz.uni-muenchen.de)
Date: Fri Sep 29 2000 - 03:39:30 MDT


J. R. Molloy writes:

> Please be patient with me. What do you mean by mature uploads?

An immature upload is a full-detail emulation of a human's low-level
neuronal processes (at, say, compartmental emulation level) based on a
data set more or less imediately derived from digitized
neuroanatomy. Apart from being potentially immortal and having
a radically expanded control of itself and it's (rendered)
surroundings it is very human.

A mature upload is something which has made extensive edits to the
above dataset (or is derived from above original dataset in a
darwinian fashion), and is encoded in a fashion which runs more
efficiently on the (computronium) hardware (both the hardware and the
encoding will have to undergo several optimization steps for this to
succeed). It is much faster than an immature upload (operates on a
time scale 10^3..10^6 faster than we), and requires a fraction of its
resources to operate.

Another definition is that a population of mature AIs can't get
snuffed out by AIs created by evolutionary algorithms. They can
persist in a stable equilibrium with them (nice trick, that).

There is supposedly a traversable development trajectory from immature
to mature. Since it's unlikely any of us will be a player in that
game, we will be essentially dependant on their freely flowing milk of
posthuman kindness for 1) protecting us from AIs 2) not snuff us out
instead of AIs

> (And what's wrong with letting the Amish go extinct?)
 
In principle, nothing. However, being human, I would object.

They should be given a choice, those who reject the choice I would
still upload by force (now things do become ethically iffy), erasing
their short-term memory of what just happened and given an illusion of
subjective continuity of existance. As long as I can spare the
resources, they would be free to pursue what they did before, of
course being given opportunities to leave their self-imposed seclusion
at any step of the game, and certainly if/when I can't afford the
resources.

> So we should speed the occurrence of mature uploads.

Yes. I think in 10-15 years we should be able to make individually
realistic computational models of nematodes, provided we establish and
fund a large project now. After that, you have to scale this up to
higher animals in a series of steps. (E.g. C. elegans,
D. melanogaster, <some suitable intermediate here>, M. musculus,
H. sapiens).

Once we can simulate nematodes, perhaps people will start seeing value
in developing means for vitrifying brains of terminal and freshly dead
people right in the here and now. Meanwhile, several thousands of
people per hour will be continuing dying irreversibly while we will
continue waffling about AI, SI, nanotechnology, and properties of
metallic xenon and unicorns, and whether their horns are really
Mohs>10.

> (I need to know what a mature upload is first.)



This archive was generated by hypermail 2b29 : Mon Oct 02 2000 - 17:39:22 MDT