From: Ramez Naam (mez@apexnano.com)
Date: Mon Nov 25 2002 - 16:14:27 MST
From: hal@finney.org [mailto:hal@finney.org]
> Drexler's proposal was to try to restrict the set of designs
> to those which could be modelled relatively cheaply. That's
> one reason they use so much diamondoid, it's stiff so the
> atoms don't move around much.
>
> I think the moving parts do need some careful simulation, I'm
> pretty suspicious of some of Drexler's exotic bearings and
> such. But for the most part these can be embedded within a
> diamondoid matrix so you can use the part in a modular
> design without having to simulate the whole system at once.
I agree with all of your points. You can indeed make the system
simpler to model if you overbuild such that various parts can simply
be considered "stiff" and if the subsystems are sufficiently separate
so as to allow separate modeling.
At this point I'd like to make clear that I'm not doubting the
feasibility of building an assembler, someday. I am, though,
surprised that many of the problems that make the design and control
of such a device an extremely challenging problem have not been widely
raised. Illustrating some of those issues is my goal in this thread.
As far as the techniques you mentioned to make modeling simpler go,
they are certainly valid approaches. At the there are three crucial
things that an assembler does that make modeling far more difficult
and which may undermine many "shortcut" methods.
1) Assemblers process information to follow their instructions,
monitor their environment, etc... This involves many and very
complicated moving parts and in effect brings all the complexities and
difficulties of prediction of software engineering into play.
2) Assemblers make and break chemical bonds. From a molecular
modeling standpoint, this automatically forces you to use more
expensive simulation methods, as the cheapest methods simply do not
handle the breaking or formation of chemical bonds. In addition, a
chemical interaction at one end of the assembler may end up having
surprising effects on previously "isolated" subsystems. How does a
subsystem react when the assembler a whole temporarily picks up or
loses an electrical charge, for example?
3) Perhaps most problematically, in at least some proposals assemblers
reproduce. This gets into an additional level of modeling and
simulation which I haven't even mentioned up to this point.
Essentially you start dealing with population biology, an area where
modeling is now being applied but where our ability to predict exact
outcomes is essentially zero.
> Plus there's always good old science. You don't necessarily
> have to simulate everything, you can build parts and measure
> them to see how they work. We've built machines for
> thousands of years without computer simulations.
I agree. At the same time I think this has risks (you're doing trial
and error experimentation with replicators) and limits our ability to
directly control and program our creations.
Finally, this approach undermines the goals of Foresight. If the goal
is to understand what it may be possible to create before we actually
gain the ability to create such things, then modeling and simulation
are crucial. As far as I can see, our modeling and simulation
capabilities will not be up to the task for quite a long time to come.
cheers,
mez
This archive was generated by hypermail 2.1.5 : Wed Jan 15 2003 - 17:58:23 MST