Re: Leading nanotech experts put 'grey goo' in perspective

From: Chris Phoenix (cphoenix_at_best.com)
Date: 06/24/04 

Date: 24 Jun 2004 00:40:05 GMT

Jim Logajan wrote:
> My concern was not merely that useful autoreproductive systems would be
> prohibited, but that Foresight and CRNano were now endorsing prior
> restraint on an entire line of research. It is good to see that is not what
> you intended, but unfortunately it is no longer clear to me (from this
> dialogue) what the original statement was intended to mean. It would be
> helpful to me personally if you could clarify what you meant by
> "prohibition".

We have two choices here. One is to go around in circles getting more
and more pedantic and focused on words. The other is to back off and
look at intent. The intent was to say: there's no reason anymore for
building anything like a dangerous self-replicator, and so anyone who
does it should be considered uncivil / destructive. Whether this is
embodied in law, in the Foresight Guidelines, or simply in common-sense
agreements among scientists, is a policy detail: which will work best to
minimize the number of stupid and/or malicious attempts?

I'll anticipate your argument: no, I don't think such things have to be
constructed for any legitimate research purpose, even to learn how to
fight them.

> Based on this description, it looks to me like the system at the root of
> the control tree doesn't get any input from its subordinates and so doesn't
> need to do anything more that transmit information. At least at that "root
> node," it doesn't really require much intelligence and is strictly open-
> loop control. That's all I meant - nothing more.

Yep.

>>Yep. The idea is that if the mechanosynthesis is reliable enough, you
>>don't need to verify that it's worked. The paper assumes that
>>sufficiently reliable reactions will be found.
>
> That's a big assumption, IMHO. Also, once a designer has access to a
> working feedback control mechanism at the nanoscale, there is no reason not
> to use it since it will sometimes simplify other aspects of the design.

I have nothing against complicating one part of the design to simplify
another part. Designing at a high level, I had to make each part as
simple as I could. I'm sure the actual nanofactory will have better
architecture in several ways.

There are two kinds of "unreliable." One is where you might have to
retry or error-correct, but you'll eventually get it right with very
high probability.

The second kind of "unreliable" is where the chemistry might go so wrong
that it can't be fixed--the machinery simply jams.

An architecture that handles everything with local computers can deal
with the first kind of unreliability just fine. This requires only
local control--no information has to be passed up the chain. (In
theory, not even an "I'm done", though that can be condensed in
sub-nodes so it doesn't matter.) So it doesn't make a significant
difference to the architecture. Nothing can deal with the second kind
of unreliability.

> Well, I think we're in violent agreement here in regards to the persistence
> of the "misunderstanding" among critics. I'd add though that, IMHO, gray
> goo is related to self-replicating assemblers in the same way that
> carnivores are related to animals.

No. Carnivores evolved by blind selection. Gray goo couldn't do that.
  At least it couldn't evolve from a nanofactory, or from any of its
bootstrapping stages. And I think it would be harder to design a
product that would evolve into gray goo than it would be to design a
gray goo.

> I disagree - construction assemblers don't need to navigate or move. At
> least, not in any but the most trivial of ways that a nanofactory must also
> "navigate" or locomote finished products. I happen to believe that the most
> useful and simplest kind of self-replicating assemblers are those that
> remain semi-attached to each other. Any mobility they have is only in
> relation to their peers (i.e. move an assembler in or out of a place in the
> matrix of assemblers using a bucket-brigade mechanism.) Things like
> communication are easier, among other advantages.

"One to hold the bulb and six to turn the ladder."

You might be right that this can be efficient enough to be worth looking
at; but I really doubt it. Especially since a design with
self-contained replicators would require each replicator to drag along a
computer.

> The genome for e-coli has ~4.6 million base pairs; the entire human genome
> has ~3 billion base pairs [1]. Since there are a lot of genomes that code
> nothing, the real information encoded is less than those values (I believe
> only about 3% of human DNA encodes anything useful). A complete nanorobot
> design shouldn't be more complex than a human. For these reasons I think
> your gigabyte estimate is at least a couple orders of magnitude too high
> and I still think my 200k estimate is closer to the mark for a minimum.

Either 1) lots of software and hardware today is more complex than a
human or 2) engineering leads to very inefficient description techniques
that we don't know how to compress. It might be possible to build a
nanomachine description language that could represent a
bonding/placement operation with less than 0.0002 byte per atom (for
intricate nanosystems), but I sure wouldn't put money on it. But I
admit I can't rule it out. I saw an impressively small and opaque C
program once that would print "The twelve days of Christmas".

> I note that there is some discussion on CRNano.org about that issue, and
> there has been similar discussions in the past on this group. Based on what
> I've read of history, arms races are a modern phenomenon. And while there
> is some debate whether an arms race caused WWI, there appear to be no other
> cases where an arms race led to a war or that they are inherently unstahle
[2].

The stability of an arms race depends on the technology involved. MNT
races would be a lot less stable than nuclear arms races, for many
reasons. In no particular order:
1) Cheaper to develop and test
2) Less horrifying; less political pressure against use
3) Less collateral and post-attack damage (more controllable)
4) Much faster obsolescence (orders of magnitude)
5) Much cheaper to build
6) Far more diverse weaponry
7) Almost everything is dual-use

Nukes are easier to stockpile than to use; MNT-built weapons are easier
to use than to stockpile.

> Further, many of the participants in these nanotech arms-race discussions
> seem to me to make a lot of (unsupported and unnatural, IMHO) assumptions
> on which they base their arguments that an arms race will not only happen,
> but that it will lead to disaster. I happen to disagree with quite a few of
> the very foundations on which these discussions are predicated! If you're
> interested in reading my opinionated opinion, I'll open a new thread on the
> subject.

Please do.

Chris 

-- 
Chris Phoenix                                  cphoenix@CRNano.org
Director of Research
Center for Responsible Nanotechnology          http://CRNano.org