From: Robert J. Bradbury (bradbury@ilr.genebee.msu.su)
Date: Thu Nov 11 1999 - 10:54:00 MST
> Eliezer S. Yudkowsky <sentience@pobox.com> wrote:
> > Doug Jones wrote:
> > > and a bunch of design software (petaflops and then some).
> >
> Glad to hear it, but I don't think I'll be able to rely on it.
I'm not sure whether you mean the software or horsepower.
The software we are pretty close to now. We can do the quantum
simulations on small scales, we can do the molecular modeling
on larger scales, I've got a package from Algor that does the
finite state element analysis for stresses, etc (though not at
the molecular level), Autocad & similar packages handle the
macro scale stuff and I've seen prognostications by people
familiar with M.E. that more and more of this is going to
get fairly automated. I have to believe that M.E. is developing
standard "libraries" just as the E.E.'s have done. What *is*
needed is standard libraries at the atomic level!
As far as the horsepower goes, I'm pretty confident that individuals
will have that as well. Things don't get difficult until after
2010 and by then we should have desktop workstations with 10-20 10+ GIP
processors. So a 10 machine cluster gets you a couple of Teraflops.
I think Doug may be overestimating his processing requirements a bit.
The thing about most macroscale objects (rockets, cars, etc.) is that
they don't require nanoscale structure. Relatively defect-free assembly
of materials you cannot currently assemble is what you really want.
Most current materials science focuses on how to make the grain
bondaries smaller and prevent crack migration. Those problems
go away and the materials get much stronger when you have atomic
assembly. Because you can consider blocks of perfect crystals in the
macro scale analysis, your computer simulation requirements have some
nice aggregation & simplification properties that let you test the design
without having to go to the atomic level that would require petaflops.
You have a very different situation from having to simulate a single
engine with trillions(^??) of atoms and a large amount of "fluid" to
simulating 1 engine of 10^6 fewer atoms that you are going to use
10^6 copies of. Its interesting if you compare say American & Russian
rocket efforts. The American's approach built a huge infrastructure to
manufacture a few large engines while the Russian infrastructure built
more on previous efforts by simply adding on more small boosters to get
the required thrust. I think Boeing may be moving in that direction with
the Delta program as well. (Others please correct if I've got any of this
wrong.)
Nanoengineering takes that much further because of the parallelism,
building many small things is much faster than building single large
things. So we don't have to build a few "perfect" large things, we
only have to build millions or billions of small things with relatively
low failure rates. Our rocket folks might want to comment on how
their designs and testing requirements get impacted when they have
redundancy factors of millions.
>
> How clever does the software have to be? I mean, what exactly are you
> envisioning it doing? (I'm assuming that we're using design-ahead to
> build nanocomputers that run Linux so that the software can be tested,
> in miniature, on existing computers.)
Most of the software already exists today. There are some hurdles
that have to be gone over for things that don't come into play at
the macro-scale level that are important at the nano-scale level.
These scaling-law impacts are slowly being incorporated into MEMS
design packages. The hurdle we need to go over is the merging
of the MEMS design packages with the molecular modeling packages.
There isn't much "cleverness" required except when (a) you crossover
between scales (there are currently maybe 3 of these in molecular
modeling) and in finding efficient ways to distribute large
calculations over distributed computing architectures.
>
> What kind of "design-ahead"? And how much do you need to know about the
> exact nature of the assembler breakthrough beforehand?
You might have to know a fair amount. It doesn't do you any good
to design in a silicon pressure sensor if you don't have any chemistry
to stick Si atoms there. What you want to do is the macro-scale
designs using some assumptions of material scale & properties (equivalent
to electronic circuit design). Then below that, the part libraries and
circuit layout tools take over (correspinding to the low level process
specific translations in semiconductor manufacturing) doing the grunt work.
This low level stuff is going to be very chemistry & assembler specific.
Single crystal Fe2O3 (hematite) doesn't have much worse material properties
than Al2O3 (saphire) which is slightly worse than diamondoid but the
assembly chemistries & feedstocks are all very different.
[BTW, sapphire interlaced with diamondoid heat conductors might be your
best rocket engine material because of the higher melting temp.]
> Suppose the drextech is soft instead of hard?
Doesn't work for space if it is water-based. Soft for space-suits,
ballons or solar sails is fine (even good), but for most of the things
we want to do would require rigid structures. I can think of some
ways using disulfide bonds to make proteins somewhat stronger I suspect
there are limits. Even of you go to pseudo-biocompatible crystalline
materials (e.g. sucrose or salt), I don't think you want to be building
space ships out of them.
> How does the complexity required to build a hard-nano spaceship in a
> vat compare to building soft-nano grey goo that reproduces in the wild?
Soft-nano assembly *will* be here much sooner, but the design issues are
probably much more complex. There is engineering science about how to
design beams, valves, screws, etc. Taking this down to the atomic level
is pretty straight forward. I can assemble organic "disassemblers"
fairly easily (though expensively) now, but I don't know how to
give them robust defenses or easily provide them with capabilities that
I can't "rip-off" from nature. The easy engineering of general purpose
bio-catalytic machinery to "effect" a grey-goo scenario is likely
to require more computer horsepower than Doug's rocket engine
simulations because I *do* have to go to atomic scale.
I could design a diamondoid slicing & dicing apparatus *much* easier
[there is some discussion of this in NM] than I could design enzymes
to disolve humans (much less concrete). Our M.E. skills are way above
our B.E. skills at this point.
Mind you, the design an enhanced flesh-eating streptococcus (if all you
wanted to do was disolve people), seems straight forward, but I suspect
that our ability to block multiple pathways in bacteria will be taking a
leap forward over the next 5 years as big Pharma rolls out a whole suite
of antibiotics (enabled by the genome disassembly of most pathogens).
Given the developing defense prospects and the difficulty of engineering
new pathways that avoid the new antibiotics is difficult, I'm optimistic
that bio-defense will trump bio-offence for the next 10-15 years.
It is worth commenting that I'm not sure that the "Grey Goo" scenario
has had a serious examination by people qualified to judge its development
timeline and/or possible defenses. Given the difficulty of engineering
concrete disolving enzymes and the fact that even diamondoid slicers &
dicers "wear", and the fact that whether there are enzymes or diamond cutters,
for extended operation, they do require energy supplies that can be cut off,
I wonder whether the "Grey Goo" nightmare is *really* that significant.
It might be feasible, only for say an alien civilization to drop on
a civilization totally unprepared to defend itself against such an
attack, but might not ever work in an environment where you can see
it coming and take precautions.
>
> How much power do you need for making fuel? Can the fuel be
> manufactured in advance? If not, is there some way you can use
> nanotechnology to get around the problem by concentrating existing
> resources, at least on a once-off basis? How about fusion drives?
This is a no-op, given growing arrays of nano-solar cells. Plenty
of electricty to split H2O.
Fusion is a very long shot. We don't have working fusion reactors
now so fusion is very unlikely. You may also be paying a nasty
weight penalty for the reactor and/or shielding. I think the most
efficient (low radiation?) reactions also require 3He out of the
lunar topsoil or Jupiter's atmosphere.
>
> How fast does that scale up?
Its limited by your power scale-up which probably is design limited.
> Could you evacuate cities, or at least provide the evacuation vehicles
> to do so?
If you give everyone SUV-AirCars evacuation isn't a problem.
> Could you fire cities, or at least buildings, directly into orbit?
Difficult, these probably aren't engineered for the G forces
and certainly have steering problems. If you wanted to lift
a city up from the bedrock (on rockets), you could do so, but the
power requirements would probably have to include large solar power
satellites. I suspect a better way to go would be hydraulic
lifting of the cities on diamondoid towers since the power
delivery can take place over a much longer period of time.
Robert
This archive was generated by hypermail 2.1.5 : Fri Nov 01 2002 - 15:05:44 MST