Re: NANO: Skeptical MIT Tech Review Interview

Eugene Leitl (eugene.leitl@lrz.uni-muenchen.de)
Sun, 22 Nov 1998 23:28:17 +0100

Peter C. McCluskey writes:
> it is silly to confidently reject an argument you haven't listened to,
> rather than admit to some doubt.

Agreed.

> It's running on a 180 mhz pentium with bandwidth that is probably above
> average. It may be moved to a better machine soon. The biggest problems are
> that Perl is slow for what it tries to do, and there are probably some
> unidentified inefficencies in how it handles network i/o. Nobody has
> volunteered to do much work in these areas.

You did at least part of that thing, right?

I think the bottleneck is not the Perl code, but that the database is in a single location. As Foresight is supposed to go transatlantic, it might make sense to clone that system somewhere in Europe, synching the database with a daily mirror, dispatching queries according to requester's domain. Since the stuff is in perl, much easier even would seem to ask academic servers all over the world to volunteer resources. If one could solve the database joining and distribution Crit would become much more usable. As a quick fix, you can use the current server as a master radiating changes to slaves elsewhere.

(I realize I should be volunteering to fix it instead of just whining, but my slow days are rapidly coming to an end...)

> To the extent I understand what you are saying, I can't see any similarity
> between Drexler's beliefs and the beliefs you are attacking. Possibly Merkle
> oversimplifies too much when dealing with people who ask the usual half-baked
> questions, but when responding to thoughfull questions I think he is more
> skeptical about those issues than you imply.

I am _far_ from attacking Drexler (who is a remarkably clear thinker and an impressive speaker) and Merkle (who is continuing to contribute a stream of amazing papers to the field), but rather the dark halo of 'nanoists'. As rapidly as the field moves lately (why, even that laggard Krautland is sprouting nanotechnology centers like mushrooms), the R&D impetus is likely to spill out into the mainstream, and thus into hands of ruthless and/or ignorant corporate and mad tinkerer types. A fair fraction of them don't see the technology's true potential (or, rather see it too clearly). Playing the constant Kassandra may enhance the awareness of the general public a bit.

> >malaise. Nanotechnology as practiced on the web is simply not a
> >science.
>
> I suspect you misunderstand Drexler precisely because most of his
> communication is via media you find less accesible than the web. (Most
> of what I know of his arguments comes mainly from Nanosystems and from
> hearing him in person).

As I said, Drexler et al. is impeccable. I've only had the pleasure of talking to him recently, but my copy of Nanosystems is _very_ tattered;) I can only hope that my next engagement will offer the opportunity to dive into academic science magazines, foremost 'Nanotechnology'. Actually, there are very decent NT information on the web, not only by the Foresight Institute and on nano.xerox.com and merkle.com, but web published stuff done by mainstream researchers as well.

> >Futurology is not science.
>
> A ridiculous claim. It isn't very succesfull as a science, but there are
> plenty of falsifiable hypotheses.

Yes, but not a priori. Accuracy of predictions goes down dramatically the farther they reach in the future, particularly during the time of rapid developments deemed to form the outskirts of the Singularity. Technology predictions over the course of the next 30 years might have been 100% accurate during early bronze, difficult during the times of James Watt, and very much snake oil today. Unexpected things happen all the time.

> > There exists no validated scheme to assign
> >probabilities to future histories. Hence, expecting any real-world
> >criticisms of pipe dreams is not exactly constructive.
>
> Engineering time forecasts are routinely made for incompletely specified
> blueprints, and are often of some help in resolving policy disputes.

True, but these are done in a narrow context, and are of very limited temporal depth. Estimating the duration and costs for the next Boeing model is vastly different from saying which new semiconductor litho processes will be available in a decade. Whatever happened to all that GaAs predicted? Why did parallel systems take so long to catch on? Why are we still doing comfortable UV instead of synchrotrone light sources?

> >Is such a device possible? No one can say, now. If there was a
> >detailed, mundane blueprint, we could simulate and validate that.
> >But there is none. How does one supposed to prove the existance and
> >met functionality of a mundane device which is not specified? The
>
> One of the many reasons that nobody has provided a detailed blueprint is
> that we don't have systems powerfull enough to adequately simulate them yet.

This is not not quite true. One of the chief advantages of machine-phase systems is that very little flexibility is present, and the trajectories are simple (they are machines, right?), and well-defined. It should be perfectly possible to run a sensible MD run of such a system up to a few million atoms large, using a parallel supercomputer or a commodity component cluster for the costs of a high-end workstation. The same is true of the mechanosynthetic processes occuring at the tip, which can be approximated by doing QM of small clusters. All this is not exactly rocket science, a single person with sufficient resources could validate such a design, or at least individual parts of it, in less than a year.

> >science/technology validation process does not work that way. It would
> >help a lot if there was a mundane, well-developed body of theory describing
> >mechanosynthetical processes. But there is none! (Don't say things
> >like Schroedinger or ab initio, that doesn't count). So we have to resort
>
> I'm not sure what kind of theory you are asking for. I am fairly confident
> that for any given feature that an assembler would require, a conservative
> design can be found such that most chemists would say (without feeling a need
> for simulation) that it would probably work if built. I think Drexler has

I happen to be a chemist (perhaps not a good one), but I would not believe any chemist stating such claims. Chemistry is more art than a science, and operates in the solvated domain. For instance, a chemist will say that cumulenes polymerize to product with the structure x at conditions y by the mechanism z. Finis. What I'd like instead, is that someone physisorbs, say, butatriene on HOPG and pokes it with a STM tip a bit, to see whether the graphite would open up and you could attach an end to it. At the same time, somebody else should see whether a cumulene strand extruded from a fluorine-terminated carbon nanotube kinked against a surface would do in a QM simulation, estimating the biradical character and its orientation vs. inner graphite surface would do. Estimations about whether one could use that mechanosynthetic process for deposition of a low-defect carbon nanotube should follow, best validated by an experiment. Then one would use the control precision estimate from an MD of the entire motion controller structure, and gauge the defect rate, and its impact on rigidity in the context of a clone sequence. If it all looks good, you color all parts of the preliminary assembler structure green, and either attempt to get rid of the uncoloured ones by a redesign, or attempt to think up another ms reaction which could deposit that. Reiterate that for that any reaction that follows. Integrate all the changes into the final design, and publish a thorough analysis of it. Whew.

If you do something like that, you can be certain that a lot of people start listening damn closely. Even if that particular design never flies, people will see the basic logic and also implications, and after one year ten thousands of new researchers will flock to the field.

> produced some clear reasons for suspecting his ideas, or something more
> powerfull, will eventually be built. I think the biggest uncertainty comes
> from the shortage of good tools for bootstrapping the construction of such
> features, and that many people are underestimating the progess that has
> been made recently on such tools.

A lot of progress is hampered by the absence of cheap computing power and adequate software. Drexler is still using Cerius2, which is not that bad, but it is no open source, and _very_ expensive, limits you to SGI boxes forcrunch, with low parallelism and a terrible price/performance ratio. Lab robotics and combinatorial chemistry is doing great, and SPM nanorobotics is doing fantastic, particularly since carbon nanotubes have proven to be so rich both structurally and functionally. Molecular biology flies on, but the protein folding problem is unsolved. Macromolecular chemistry is doing well, as do a large number of other fields. Taken together, they form an impressive resource.

But all this does not say anything about whether the whole autoreplicators-by-means-mechanosynthesis idea is sound. And if you're not looking at these fields from that angle, you don't see anything dramatic happening. You need a certain point of observation to realize that all the dots around you show a blue shift.

> I think a handfull of people (possibly Drexler, probably someone at Zyvex)
> have moderately clear guesses about what engineering steps will be taken when.
> Unfortunately, I haven't seen those guesses published anywhere.

Well, I hope the reality has enough respect, and will deliver.

> > Which does not mean that it is impossible -- simply
> >we. just. can't. tell. yet.
>
> That is a conclusion that Drexler and I would be glad for most people
> to adopt. The belief that it can't possibly happen in the next 20 years
> is potentially quite dangerous. Uncertainty poses much fewer risks.

If it's possible, it's certainly doable in the next 20-30 years.

> >Please, let's stop this nonconstructive discussion, and let's go back to
>
> My responses are often slow because I give them lower priority than working
> on molecular modelling software, but if discussions like these raise doubts

Your priority schedule is absolutely correct. In fact, I will be doing very much the same thing as you in a few months.

> among people who were previously confident that they could postpone worrying
> about the effects of nanotech, they can have very constructive effects on
> how prepared people are.

Amen to that.

> >the workstation/lab. Whoever can push the field of nanorobotics and
> >nanolithography into full-blown all-purpose mechanosynthesis is
> >scheduled to be the next Nobel laureate -- and Generic Saviour as well.
>
> I hope the phrase "Generic Saviour" is intended to be sarcastic, but I
> think it is dangerous to make claims that could be interpreted as
> encouraging complacency about the effects of nanotech.

No, 'Generic Saviour' is meant more about the lines of the 2nd Coming, with airborne trumpeting angels, vials of blood, Pale Riders, and sundry apocalyptic paraphernalia.

ciao,
'gene