Re: No nanotech before AI

From: Robert J. Bradbury (
Date: Wed Jul 12 2000 - 13:52:38 MDT

On Wed, 12 Jul 2000, Eliezer S. Yudkowsky wrote:

> Another possibility that hasn't been discussed is that nanotech might
> not be entirely automated! It could be that for anything more
> complicated than raw diamond, you'll need a group of nanoengineers
> painstakingly assembling the object and watching for errors, with
> automation of the process coming later in the game.

Eliezer I think is correct in that you are going to need a slew
of nanoengineers. If we have diamondoid nanoassembly by 2010
as Von Err has predicted, then those engineers are entering high
school around now. I doubt very much that there are thousands
of young people see themselves on career paths to nanoengineering.
The few nanoengineers that are produced by a few select schools
are likely to find themselves in the high-demand positions of the
software engineers today.

Not to say the software engineers won't have it pretty good as
well (you gotta program those assemblers, verify the designs and
automate as much of the process as possible to work around the
fact that you didn't train enough people during this decade).

So its likely we will have the technology and run *smack* into
the wall of consequences of people never believing we would.

> All things considered, nanomaterials will probably start out at
> thousands or millions of dollars per gram, too expensive for anything
> but nanocomputers, with prices dropping by a factor of ten every year or
> so.

Well, even gold and biotech enzymes aren't *millions* of dollars a gram.
There are nanoscale materials being fabricated in bulk quantities today
and they aren't that expensive. To get the suggested prices, you would
have to be manufacturing something that really cannot be done using
larger scale technology. Nanocomputers don't do it because I can get
the same computational capacity using a large bunch of non-nanoscale
computers. So nanotech solutions will not be able to be much more expensive
than the non-nano equivalents.

> This is still much more gradual than "Surprise! Everything is a
> dollar a kilo!" It looks like nanomanufacturing will hit the economy
> gradually, rather than sharply, relatively speaking.

I agree. The lack of designs are going to be limiting for quite some time.

> The problem is that even automated manufacturing of raw diamond
> gives you military superiority in fighter jets, and isotopic separation
> and mass-produced nukes comes not too long after that.

You keep raising the fighter jet point. I'll simply make two in response -
a) How frequently do we use our fighter jets today (which are clearly
   superior to anything anyone else produces [*if* they even have them!])?
   (In other words "so what" if we have diamondoid fighter jets?)
b) Diamondoid fighter jets are easily countered by diamondoid surface-to-air
   missles (perhaps more so).

Your point only seems important if you are suggesting that some nasty
power (North Korea? Iraq?) gets diamondoid assembly and can produce
full scale macro-nano-designs in a time scale much shorter than it
would take us to develop complementary countermeasures? Is that
what you are suggesting might happen? Or are you afraid that the U.S.
will develop diamondoid nanoassembly and use it in U.S. fighter jets
to spread the U.S. version of democracy all over the world?

Now, with isotopic separation *where* do you get the raw uranium
(or plutonium)? Can this be hidden from current nonproliferation
checks or satellite observations? Even if you have mass-produced
nuclear weapons -- would you use them? Bioweapons development wasn't
enough to induce the use of nuclear weapons, why should nanoweapons
development be any different?

> If nanotech comes first, the two key questions are military
> destabilization and the effect of nanocomputers on AI.

You have made the key statement -- "destabilization". Currently
the country in the forefront of nanotech and biotech is the U.S.
Other countries are probably 2-10 years behind us. Can you make
a case that this situation is likely to change?

Then there is the problem of having a *designs* for nanoscale
nanocomputers. Nanosystems is 8 years old this year and there
still isn't an atomic scale design for a nanocomputer, though I
believe the software with which an atomic scale design could
be done does exist.

> If nanocomputers are classified as military material, so AI researchers
> (and specifically the Singularity Institute) can't get them, then the
> situation looks a lot worse.

I would be very surprised if Mitre doesn't produce a molecular computer
and win the Feynman Prize within the next 3-4 years. I'd be even more
surprised if HP/IBM/Lucent do not demonstrate prototype molecular computers
within the next 3-8 years. And that is without classical diamondoid
nanoassembly. I don't see the probability of being very high that these
will become classified. There may be fairly large lag between lab demonstrations
of these capabilities and end-user machines using them simply because we
don't currently have code (other than perpaps physical simulations) that
requires that kind of speed.


This archive was generated by hypermail 2b29 : Mon Oct 02 2000 - 17:34:26 MDT