[extropy-chat] Re: Social Implications of Nanotech

[Chris Phoenix]

Before the technical stuff, a meta-issue:
It's a common effect of human psychology that two people will annoy each
other in symmetrical ways.  We seem to have run into that here.  

I thought I was explaining in straightforward logic why certain system
configurations were unlikely in practice.  You asked me "not to assume
everyone who disagree with you is an idiot".

Having read this complaint and remembered the psychology feature, I now
suspect you were trying to explain in straightforward logic why my
definition of "unlimited-sum transaction" was not usable.  But at the
time, I read your example of million-dollar water as
condescending/snide.

Let's see if this identification of (what may be) the problem is
sufficient to fix it.

On Wed, 12 Nov 2003 (in two emails), Robin Hanson wrote:
> On 11/12/2003, Chris Phoenix wrote:
> > ....
> >Is there any reason such a thing would not be capable of producing a
> >copy of itself?  ... Is there any reason such a thing would be slow?
> 
> The whole point here is to be able to do economic analysis on this topic
> without having to settle all of the technical disputes that still rage.
> You are arguing that there's really only one interesting scenario to
> consider, but clearly many other people believe that ....

It's an interesting question, how much pruning to apply to the range of
scenarios.  If you become convinced that a scenario is implausible, is
it your duty to prune it ASAP?  Or to talk about it anyway in order to
cover the space, and let others prune it later?  That depends on the
evaluation process in your field, which I don't know.

Is it harmful to talk about implausible scenarios as though they were
plausible?  It wastes time, but that's just a tradeoff.  Logically, a
scenario based on a contradiction can lead absolutely anywhere.  If it's
merely based on an implausibility, it will probably only lead to an
unlikelihood.  And of course they can always be criticized and discarded
later--but may corrupt the discussion in the meantime.

Since the scenarios I was arguing against are based on a less-likely
(IMO) state between two plausible states, naive expectation says that
their outcome will also be in the plausible range.  I'm not sure that
follows, since the outcome is very multidimensional.

But if it's appropriate for you at this stage to refrain from
considering plausibility, then you can ignore a large fraction of what I
said in my earlier messages.  I hadn't considered that this might be the
case.

> .... many other people believe that creating a
> self-reproducing system is damn hard and many other nanotech abilities
> will appear and have interesting consequences before then.  I'm not trying
> to settle such disputes; I'm just trying to identify the key scenarios
> being considered to support some economic analysis.

As far as I know, there are two main technological scenarios.  The first
one is what the NNI is promoting.  Nanoscale technologies (~what the NNI
is funding today) have incremental but noteworthy effects by creating
new materials and components.  This enables some new products and
applications.  Eventually (~4 decades?), when we learn how to handle
self-assembly, this gives us a handle on more complex components and
some system integration.  But I haven't seen any proposals for NNI-type
nanotech, either short-range nanoscale tech or longer-range bionanotech,
to produce an integrated general-purpose production system.

The NNI seems to be inconsistent in their claims for this technology. 
They seem to be claiming most of the benefits of the Feynman/Drexler
vision, and some of the underlying capabilities (manipulating and
fabricating matter atom by atom), but deny the possibility of
translating conventional machines and direct manipulation to the
nanoscale.  It could be viewed as conservative: by the time we get
biomimetics working, it probably won't be that big a deal (IMO, at
least).

BTW, I just today discovered some *fascinating* work on ATP synthase. 
It seems that a mechanical model was constructed, not including thermal
noise, and it worked just the same as the jittery protein.  A major
argument of the pro-biomimetic people is that thermal noise is helpful
to biotech but harmful to MNT.  MNT people have been saying that it's
not harmful; now it appears it's not necessarily helpful to biotech
either.
http://www.santafe.edu/sfi/publications/Bulletins/bulletin-summer98/scimodel.html

Anyway, the second technological scenario is the MNT one, and as far as
I'm aware, it always assumes autoproduction (= self-replication, but
without the "life" connotations) very early in the game (due to
technological necessity), and frequently also assumes a fast takeoff
from "assembler breakthrough" to "flood of products."  Some smart people
like Ted Kaehler (in _Nano_) have argued that design problems will slow
the flood to a trickle.  This was a major motivation for my Nanofactory
paper, which I think is a very strong counter to that argument.

This MNT scenario generally focuses on the effects of a sudden advance
in manufacturing, i.e. cheap general-purpose manufacturing of products
with nanoscale features.  The nanoscale features give useful properties
like very efficient operation and direct interaction with other
nanoscale objects like chemicals and cell parts.  But for the most part,
this scenario ignores the nanoscale technologies--I suspect a major
reason for this is that it predates the whole field of nanoscale
technology-type "nanotechnology".

If you project general technology trends, you may find that something
equivalent to MNT happens naturally around (at a guess) 2050 +/- 15
years.  At this point, it would not be a big deal.  But if it happens
around 2010-2015, in flood-of-products mode, it would be a very big
deal.  A trickle-of-products may still be a big deal, because computers
are fairly simple products, and are worth pouring lots of design effort
into.  In general, an autoproductive nanoscale system should also be
capable of building a very powerful computer, perhaps nine orders of
magnitude ahead of today's, which is, what, 60 Moore's Law years?  

So if you want a range of scenarios that have been proposed, you have
two variables: When does MNT develop (2008 to 2060, with ultra-powerful
computers 3 months to 3 years later), and how long after that do we get
general-purpose manufacturing (3 months to [at a guess] 10 years later).

> >... I don't see any point in talking about MNT systems duplicating
> >themselves in a year--which is a time scale you mentioned.
> 
> I said "within a year", meaning less some time less than a year.

OK.  I wasn't sure whether you recognized that one day is a quite
plausible replication time, once we accept replication at all.  

> > >>>> ... A robust morphogenetic code, an evolutionary system, ...
> > > I'm much more concerned about the robustness of the code.
> >Well, then "lots" is no harder than one.  And we can certainly do one.
> 
> We have one robust morphogenetic code?  I wasn't thinking DNA was anywhere
> near robust enough for many plausible values of the computronium available.

I said "we can certainly do one", not "we have."  I meant that once we
develop one, we can copy it to as many processors as we want.  I think I
misunderstood what you meant by "robust"; I thought you were asking
whether we could deal with the fault-tolerant computing necessary to run
a hugely parallel system.  I don't know what makes a morphogenetic code
robust, so I wasn't trying to assert anything about that.

> > > >"unlimited-sum transaction" ... the benefit to one (or both) of 
> > > >the parties is much higher than the cost, and is not correlated
> > > > with the cost. 
> > >
> > > The liquids I drink over the next few days are worth millions to
> > > me, as I'd die without them.
> >
> >No, they're not worth millions.  ... The
> >value to you of the liquids is no more than the cost of replacing
> >them from a different source.
> 
> You mean the benefit relative to the closest available substitute sold?
> Is that the way you define cost as well?  If so we could find a different
> set of examples that probably aren't what you have in mind.

Um.  I don't quite have the mental vocabulary to think in economics.  I
hope you can help me out a bit.  As I ponder on the difference between
value and benefit, and the significance of multiple diverse users
(diversity implied by benefit not proportional to cost), I wonder if
what I'm trying to say is that for these resources, arbitrage destroys
value.  There's no way to set a market without excluding lots of
potential users.

Can you figure out what I have in mind?  Can you suggest a definition
for it?  

> >Let's take another example.  The largest cause of infant death in
> >Venezuela is diarrhea.  This could generally be prevented by a few
> >pennies worth of salt and sugar in clean water, and very simple
> >instructions.  The value of this resource to the parents of those babies
> >would be immense.
> 
> Er, by analogy here aren't you supposed to be comparing the value of
> one particular source of salt/sugar/water to another, and computing only
> the additional value of one source relative to another?

I assume that if there were *any* source of salt/sugar/water/information
(don't ignore information in this example!) then the babies would not be
dying.  So there aren't two sources to compare.  So any source should be
able to corner the market and, so to speak, make a killing, at least for
a while until the information spread.  The incentives appear to be high.

But I'm not sure we need to follow this example.  You asked whether
markets could fail to supply you with water.  I answer that in some
circumstances they can fail to supply people with low-cost lifesaving
supplies, even when a supplier would make a huge profit.  But the
question of whether the market could fail to supply you with water is
probably incidental.  I'd actually rather hear what you think about the
canteen in the desert.  I'm not questioning whether it can be viewed in
terms of economics; but can it be viewed in terms of market?

> .... compared to many claims made for
> discontinuities induced by nanotech, computers were a pretty smooth > transition. .... overall
> computer tech progress has been relatively steady.

The underlying hardware, yes; the applications have been somewhat more
punctuated.  Look how quickly the web grew.  Or the difference
spreadsheets made to the business world.  

> > .... and most consumer goods are produced locally
> > on PGMDs, via downloaded designs and a few general feedstocks.   A
> > variation on this position posits that PGMDs can produce more PGMDs
> > relatively quickly.  And a refinement of this position posits that such
> > self-reproducing PGMDs dramatically lowers costs relative to technology
> > available just prior to this point.
>
> >I doubt that a
> >non-autoproductive MBPGMD is technically  plausible, much less
> >economically plausible, since you probably can't build a human-scale
> >MBPGMD without bootstrapping it from a much smaller one.
> 
> The Royal Society of London just came out with a report saying nanotech
> is possible, but that most of their advisors say self-reproduction is not.

By nanotech, did they mean the use of nanostructured materials to build
complete macro-scale products?  (I doubt it, but I'm not sure.)

I'm pretty confident in saying that their advisors are quite uninformed
about self-reproduction.  I doubt that most of them have looked beyond
one or two shallow and easily-answered "proofs" that dry nanotech can't
work as described.  The criticism of MNT theory has been of such low
quality, and people are so willing to believe blanket assertions without
testing them for plausibility or relevance, that by now I don't worry
too much about panels of experts.  I hope this doesn't sound arrogant,
because my process is one of open-mindedness.  I have paid close
attention to every anti-MNT argument and thought process I've come
across--and have been able to find a fatal flaw in every one so far.  

>The influential novel "Diamond Age" describes a world
> where most people do not have access to self-reproducing devices,
>though they do have access to capable PGMDs.

A matter of policy, not technology; remember "feed vs. seed"?

> Drexler and the Foresight Institute
>have been trying to downplay the role of self-replication for some time >now.

This is because of the unfortunate connotations of the term.  It doesn't
change the underlying technical proposal: a nanoscale machine which,
provided suitable information and materials, can duplicate its physical
structure.  

> I am struggling to clarify and identify the differing assumptions 
> that different people are making, and place them in economic terms 
> to support economic analysis.  *I'M OPEN TO SUGGESTIONS*, though 
> I'm running out of time for this round.

Sorry for the delayed answer; I didn't see this till a few hours ago.

Above, in the discussion leading to the two dimensions of variability
for scenarios, was my best understanding of the differing *technical*
assumptions that people are making--and how to unify them, by
recognizing that something similar to MNT will happen eventually whether
or not dry nanomachines work, but if it's far enough out it will blend
in.  Hm... Just thought of a third dimension: the cost of fabrication. 
I expect that even early devices will come in under $100/kg, and quite
possibly less than $1/kg, but it's conceivable that it could be much
higher.

In addition to technical assumptions, there are quite a few policy
options, ranging from first-strike global suppressive empire (enabled by
early-flood-of-products) to nano-anarchy (no enforceable policy).  This
is far too big a can of worms to get into tonight; I don't even know
which policies can work for what lengths of time, much less which
policies are plausible given the players, or how commercial and military
policies and actions will interact.  

Near the bottom of our paper "Three Systems of Action", there's a 2x2
matrix that may provide some framework for policy options. 
"Distribution & Access" is one axis, and "Technical Restrictions"
(meaning, securely built into the nanofactories) is the other; the
values for each are "Tight" and "Loose".  Several consequences of each
combination are listed.  
"http://CRNano.org/systems.htm#Table 3" (yes, the URL currently needs
Table, space, 3; I just noticed that; we'll change it to
http://CRNano.org/systems.htm#Table_3 ASAP.  It's about 3/5 of the way
down the page.)

> > > 2.  A big question is by what factor general manufacturing devices > are less
> > > efficient than specialized manufacturing devices, either in terms of
> > > production time, material waste, or final product quality.
> >
> >Don't compare apples and oranges.  That's like asking by what factor
> >digital computers are less efficient than analog computers.  Digital
> >computers can do things that analog computers simply can't.
> 
> My point was just that if digital computers had been too inefficient in
> emulating analog ones, then we'd still be making analog ones.

They *are* extremely inefficient at emulating analog ones, but the ease
of programming them makes up for it somewhat.  But a much bigger point
(and my point is that it *is* much bigger) is that digital computers
don't only emulate analog ones; they also run programming languages and
handle symbols.  Nothing else in the world can handle symbols like a
digital computer.  Likewise, I think that nothing but dry MNT will be
able to (e.g.) build diamondoid spacecraft, or a thousand other
important product families.  And I think this is a bigger point than how
efficiently it will be able to do it.

> > > While many manufacturing plants today
> > > are highly automated, it may cost quite a lot to produce designs for > fully
> > > automated production processes.  So design costs may be a lot > higher.
> >
> >Or it may cost quite a bit less.  You don't have to retrain your
> >workers.  You don't have to pay workers to retool your machines. ...
> 
> Those are not design costs.

No, but they are costs of setting up production of a new product.  I was
thinking that in the life cycle of a product, design and ramp-up can be
lumped together for many purposes.

And actual design costs for a fully automated MNT system may be lower
than current design costs as well.  The design space will be incredibly
huge, allowing lots of levels of abstraction (both conceptual and
physical-in-the-product) to be used.  

> >Um, why are you not counting the labor cost as a saving?
> >And what about transportation costs, warehousing, the costs of
> >compensating for uncertain and delayed supply chains?  ...
> 
> When I said "just lowering those ... costs" I meant just the costs I was
> focusing on there, the cost of physical capital for manufacturing.

This is the benefit everyone talks about, but many other benefits come
from the selfsame source: self-contained general manufacturing.  I don't
understand why you'd consider one benefit, the physical capital of
manufacturing, without considering the others.

For example, if factories are frequently run to capacity, then it often
makes sense to ship goods from a factory that's farther away but more
lightly loaded.  (Likewise if some factories are cheaper than others.) 
But if you almost always have a local up-to-date factory sitting idle,
you'll almost never need to ship anything.  So I suspect (but I don't
pretend to teach you your job) that for very little extra work, you can
consider the effect of to-capacity vs. sitting-idle factories on
shipping costs.  And lowered shipping costs make sitting-idle factories
more worthwhile.  

Basically, I don't understand how you could estimate the benefits of
factories sitting idle without considering all the costs I listed--all
of which depend heavily on the amount of factory idleness.  It looks
like a feedback loop (self-referential equation?  Sorry, I don't know
the term), and if you don't analyze it, I think you're very likely to
build an implausible scenario.

Which brings this email full-circle, and it's 3AM, so I'll sign off.  I
hope this was more helpful and less annoying than the previous ones.

Chris

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