poly: New paper on Grey Goo problem

From: Eugene Leitl <eugene.leitl@lrz.uni-muenchen.de>
Date: Fri May 19 2000 - 20:31:42 PDT

hal@finney.org writes:

> I was amazed by the potential threat of unconstrained replicators.

Despite the nice paper, putting things into a single place for the
public to see, hasn't this been already exhaustively addressed on
sci.nano and elsewhere, albeit not in a quantitative fashion? Within
the extropian/transhumanist circles the threat model is fairly widely
known for about a decade, at least (extremely far-sighted people saw
the threat in the early 1960s). I'm thus amazed that you're amazed...

Also, it is much too early to put lower limits on metabolic energetics
of biovorous self-replicators. Large-amplitude high-frequency cycles
of relatively massive machine-phase systems wielding high-energy
density moieties synthesized from scratch seems a remarkably cludgey
way to deposit structures, the design space seems to allow for more
continous-phase system, moieties which can be created from
intermediates of lysed biopolymers (we don't have to reduce everything
to C1 feedstock, nor do we need to deposit gem-grade cubic carbon peu
a peu, do we?), and systems which allow for more parallelism, thus
reducing replication time of the individual agent, while exploiting
their lower thermal signature/footprint.

Active camouflaging (operation pattern and shape (coating oneself with
debris of the the terminated organism), including periods of
latency/stasis and active means of dispersal can obviously address at
least some of the kinetical issue. If a billion of world-wide
distributed dormant nasties caches synchronously revive, there is not
much ground to cover until growth fronts ran into each other. Similiar
goes for seeding/dispersal synchronously deorbited recursive cluster

Moreover, knowing about the potential is not enough, if there's no
infrastructure 1) to detect 2) to act upon detection, reducing the
number of the bad guys below critical density, while not nuking what
we set out to protect. Equilibrium between olive/blue won't quite cut
it, because I don't care much for a theatre solely populated by
multihued players, all higher wet life being extinct (infinitesimal
differences in fitness result in almost instant extinction of the less
fit player).

"Such biovores can emerge only after a lengthy period of purposeful
focused effort, or as a result of deliberate experiments aimed at
creating general-purpose artificial life, perhaps by employing genetic
algorithms, and are highly unlikely to arise solely by accident."

Who's talking about accidents? I guess it's time to don my black hat.

"Homeostatic Resistance to Ecophagy?" Uh, the color is clearly
olive. This is not about industrial accidents, which can be made
arbitrarily improbable (though the current sorry state of
infoprocessing system securety indicates that protection from idiots
(=blaming the luser plus marketplace) at a large scale is nonviable,
at least if the technology is widely available). It is there to solve
the minimax between short-term kill (disrupting vital systems) and
long-term kill (high amplification factor). You can only homeostate if
something is not eating holes in the vascular/nervous systems,
plugging up capilliaries, and flooding (especially) the host and the
environment with neurotoxins, apoptosis inducers and autoimmune

Also, the effectivity of the countermeasure assume that the
countermeasure is immune to attacks, and that the target is
static/readily recognizable. In a rapidly mutating coevolving
postbiology, this is certainly not a safe assumption. The depicted
scenarios are far from worst case, which would be the prudent stance
to take.

Ah, I see there's "8.4 Malicious Ecophagy". At last.


is not a prudent assumption. This assumes an environment ripe with
cruising snoopware, which will take its time to be developed, deployed
and debugged. It is obvious that a self-replicator deployed before the
countermeasures are developed can essentially operate unchecked. Here,
even the crudest, buggy designs can create devastating damage. Imo,
there is a fundamental assymetry in the ease of development of olive,
than an effective blue.

2. Efficiency

is not a prudent assumption, if the kill is a side effect of
operation. Recognition demands constant surface features. Only blue
can't avoid constant, thus recognizable surface features, unless it
uses even more expensive design. Predators have to be more complex and
advanced than their prey to survive. Predators not allowed to
replicate and evolve are thus at a disadvantage vs. their prey who
have no such restrictions. Machines designed not by an evolutionary
process are known to be suboptimal and brittle. Once again, the idea
is not to only to obtain a stable equilibrium between predator (blue)
and olive (pray), we have the theatre populated with wet squishies to
protect (two species fighting on agar substrate, achieving dynamic
equilibrium, while agar itself is depleted in the course of the

3. Leverage

If machines (evolving machines?) are so easy to disable, why do we
have increasing trouble eliminating pests and pathogens globally,
especially the small ones? (insects, bacteria, viruses). Because even
molecular countermeasures cannot be dispersed globally without side
effects, and make the target develop resistancy? Does the kill ease
scale into the micro realm? Is there perhaps a reason why the killing
agent is destroyed in the kill process, be it a bomb, a torpedo, or
antibiotic/toxin? 3. Is another assumption, which is not prudent.

Let's see, "9.0 Conclusions and Public Policy Recommendations".

Again, the detectability of the thermal signature assumes a certain,
fixed replication model. We don't know what the lowest thermal
dissipation level is, and how to tell it from life thermal signatures,
especially is all you have is eye in the sky. Monitoring global
temperature as sign of world infection is a joke, right?

Ecophagy which is slow but olive needs only trace mass ratio to become
lethal to all higher organisms, and is equally terminal, if blue is
losing slowly, but continuously (i.e. if it can't be checked at well
below trace level).

I think thermal monitoring alone is a red herring, and census sampling
(which is the only thing with a slight chance to work) implies a
protective infrastructure in place, which will create artefacts due to
operation, provided that the olive release does not occur prior to
this infrastructure being in place.

Moratorium on ALife is imo not enforcible, and would fall into
deliberate relinquishment, which just doesn't happen. Consider the
enforcibility a worldwide moratorium on script kiddies and virus
writers. Now if dissemination of tools for sofware development was
regulated at an early stage in the game, and the threshold for
reinventing them was high enough, while everybody in the possession of
them would be researching the threat space like crazy...

Eye in the sky for monitoring hot spots is not sensible, if there is
no hot spot, and if you can't contain it if you found a bald patch in
jungle, and the autoreplicators are already spread halfway across the
globe by atmospheric streams.

"3. initiating a long-term research program" is something which I back
up absolutely. This is the only scenario which can possibly save our
ass. Establishing a black/white hat culture in the virtual realm well
before the physical, studying all aspects of the problem, well before
we have progressed towards the stage where we can begin to implement
them. Metaphorically, this is the time of Zuse and ENIAC. We still
have time to sample both the threat and the countermeasure space
before our hand is forced by the demands real world.

Unfortunately, the primate mentality is known to ignore the dangers
until they literally burn our hands. The reaction of the policy makers
to recently intensified system exploits have so far not been extremely
rational (to put it mildly).

The public reaction to http://www.foresight.org/NanoRev/Ecophagy.html
is still 1) well below the visibility of policy makers 2) is
considered loony science who're aware of it at all.

Hopefully, escalation of containable but sufficiently nasty bioweapon
releases in the course of next decades will sensitize the
public/policy makers about the dangers of molecular autoreplicators,
and will translate into effective policies.
Received on Fri May 19 20:35:51 2000

This archive was generated by hypermail 2.1.8 : Tue Mar 07 2006 - 14:45:30 PST