Re: Goo prophylaxis (was: Hanson antiproliferation method?)

Anders Sandberg (
Mon, 25 Aug 1997 15:31:34 +0200 (MET DST)

On Sun, 24 Aug 1997, Nicholas Bostrom wrote:

> I think there is something that should be pointed out clearly here.
> What you seem to mean is that genetic algorithms will play an
> important role in designing the immune system. However, what you
> often say is that *evolution* will do that. Now, that is not wrong,
> but I might mislead some people. Although genetic algorithms can be
> said to describe some form of evolution, there is a world of a
> difference between natural, blind, Darwinian evolution and specially
> designed genetic algorithms that can be Darwinian or Laplacian with
> any number of sexes and variable sizes of the property chunks that
> are inherited, where all the parameters can be played with by an
> insightful experimenter who a guids the process with his knowledge
> and overseeing intelligence. These are two separate things, but many
> people tend to confuse them.

Good points. I have a skeleton for an essay at home where I discuss
the 'P' word ("progress") and why we seldom use it (short answer:
positivism is no longer fashionable in these postmodern times),
instead saying "evolution" and "development", mixing things up a

> > Design is good at jumping over deserts in the fitness landscape,
> > while evolution is good at searching it
> This would seem to lead to the prediction that the more that is known
> the less useful will evolutionary computing be.

Yes and no. Suppose you are a near-omniscient jupiter brain. You can
see the entire fitness landscape (or more correctly, the entire game
matrix) and find the best strategy if there is one; you might
realize that your immune system cannot defend against nanite X, so
your equally smart enemy will defeat you in any case since it also
knows this. So in this case you will either survive, be destroyed or
face a somewhat uncertain future (when both players play mixed

But can even a jupiter brain see the entire fitness landscape and
find the global optimum? The space of possible designs is very, very
large (I would argue that it is at least of cardinality aleph zero
for physical designs, and the space of possible shapes is of
cardinality aleph two in a continous world). Even a planet-sized
brain would not be able to contain the combinatorial explosion, and
there might be few general rules as designs include self-organization
and deliberate chaos; since good designs are rare in this space, they
will be valuable and not always easy to find (Moravec has argued
something similar when discussing the economy of Exes).

So even a very bright individual might discern the global fitness
maximum in a subspace of immune designs, but can not handle the full
space. That suggests that genetic methods can be useful to find
unexpected new possibilities and search through it for new maxima not
amenable to intelligent search.

> > As an example, assume the worst scenario happens and an escaped badly
> > programmed dishwashing nanite starts to turn all organic life
> Why just organic life? Why not dead organic substances, earth, etc.
> And is there any good reason why it could not change the earth crust
> into something with a higher binding energy?

Energy is the problem here. While I think a workable nanite based on
silicates could be made, the biosphere is the major source of
high-energy chemicals on this planet. So it is the easiest target and
most useful; eating rocks would require so much energy that the
growth would be slow and the threat fairly minimal. And the earth's
crust appears to be at a very deep energy minimum, I don't see how
you could get further without tremendous amounts of energy.

> > into
> > more of itself. It will spread with the speed of an bacterial
> > infection, and be quite deadly.
> Why couldn't it spread much faster? Bacteria are limited to some
> specifid kinds of hosts, the nanites could attack any organic
> material and many inorganic ones too.

I based this on Drexler's calculations of replication. Bacteria are
actually quite good replicators, thermodynamics seems to place some
limits to replication speed at a certain energy level (and you cannot
get much more than 1000 W/m^2 on the surface of the earth).

> And if they were deliberately designed, they could transform
> themself to missiles after they had eaten enough, and then swoosh
> accross the seven sees in a very short time.

Yes, yes. But I'm trying to discuss the immune system problem here,
not the deliberate weapons use problem (as I pointed out in my last

> They might have to go there pretty quickly, like after a nuclear
> alert. They will have to make sure that not a single little nanite
> finds a way in. They will have to hope that the nanite doesn't eat
> rocks and cement. They will have a limited time to figure out how to
> use their very limited resources to eliminate a enemy that already
> forms a think deadly layer over the whole earth. They have to hope
> that the nanites weren't deliberately designed to pile up explosives
> on top of their bunker and blow it all away. --Yes, they *could* make
> it, at least in a Hollywood movie...

Hmm, who is talking about Hollywood movies here? :-)

> I'm sorry, but it does seem to me a bit like wishful thinking (and
> reading to much SF?). I think I will call this the
> go-hide-in-your-basement solution to the antiproliferation problem.

Again, we seem to be arguing about two different things. This isn't a
solution to proliferation (although I like your term for it), this is
a scenario explaining why intelligence is likely to survive at least
a bit into a genetic takeover and likely influence the new ecology.

> >, and then it would just evolve in an ordinary
> > way
> (Unless it was designed not to evolve.)

Good point. Most nanites would definitely have adekvate safeguards
against evolution, it is so bad for business.

> The remarks you made seem predicated on the assumption that the
> nanites will be comparable to a particularly virulent biological
> plague. Suppose that this isn't true. Then the only method for
> avoiding disaster in a society where there are many independet
> individuals with full technological access is to have some kind of
> active nanotech immune system.

Yes, that is entirely true. Once you have parasites, you need an
immune system.

> The first is if all the material were already very close to its
> lowest energy state, so that no more reactions were economical. Does
> anyone have a good design for a computer that would work under those
> circumstances (we would all be uploads then).

Well, I think the rod-logic of drexler would do quite fine here. One
could imagine silicate nanocomputers covering the earth with a thick
and inert surface.

> The second is to have the immune system quickly eliminating any
> plagues, and it could use the fact that it has access to more energy.
> A good design for this?

I have been working on a system, but it is not yet written up.
Basically, it will depend on what you want to defend.

> Aha, I just thought of a third way. The independent folks could all
> live in a virtual reality that were designed so they could do no
> major harm. They would have no access to the real reality, which
> would be ruled by a single entity.

And how do you trust the independents to not figure out how to
subvert reality in some way, or the entity to wield its power well?

Anders Sandberg Towards Ascension!
GCS/M/S/O d++ -p+ c++++ !l u+ e++ m++ s+/+ n--- h+/* f+ g+ w++ t+ r+ !y