I recently made a post – Making and doing – about the importance of moving the focus of radical nanotechnology away from the question of how artefacts are to be made, and towards a deeper consideration of how they will function. I concluded with the provocative slogan Matter is not digital. My provocation has been rewarded with detailed attempts to rebut my argument from both Chris Peterson, VP of the Foresight Institute, on Nanodot, and Chris Phoenix of the Center for Responsible Nanotechnology, on the CRNano blog. Here’s my response to some of the issues they raise.
First of all, on the basic importance of manufacturing:
Chris Peterson: Yes, but as has been repeatedly pointed out, we need better systems that make things in order to build better systems that do things. Manufacturing may be a boring word compared to energy, information, and medicine, but it is fundamental to all.
Manufacturing will always be important; things need to be made. My point is that by becoming so enamoured with one particular manufacturing technique, we run the risk of choosing materials to suit the manufacturing process rather than the function that we want our artefact to accomplish. To take a present-day example, injection moulding is a great manufacturing method. It’s fast, cheap, can make very complex parts with high dimensional fidelity. Of course it only works with thermoplastics; sometimes this is fine but everytime you eat with a plastic knife you expose yourself to the results of sub-optimal materials choice forced on you by the needs of a manufacturing process. Will MNT similarly limit the materials choices that you can make? I believe so.
Chris Peterson: But isn’t it the case that we already have ways to represent 3D molecular structures in code, including atom types and bonds?
Certainly we can represent structures virtually in code; the issue is whether we can output that code to form physical matter. For this we need some basic, low level machine code procedures from which complex algorthms can be built up. Such a procedure would look something like: depassivate point A on a surface. Pick up building block from resevoir B. Move it to point A. Carry out mechanosynthesis step to bond it to point A. Repassivate if necessary. Much of the debate between Chris Phoenix and Philip Moriarty concerned the constraints that surface physics put on the sorts of procedures you might use. In particular, note the importance of the idea of surface reconstructions. The absence of such reconstructions is one of the main reasons why hydrogen passivated diamond is by far the best candidate for a working material for mechanosynthesis. This begins to answer Chris Peterson’s next question…
Chris Peterson: How did we get into the position of needing to use only one material here?
…which is further answered by Chris Phoenix’s explanation of why matter can be treated with digital design principles, which focuses on the non-linear nature of covalent bonding:
Chris Phoenix: Forces between atoms as they bond are also nonlinear. As you push them together, they “snap” into position. That allows maintenance of mechanical precision: it’s not hard, in theory, for a molecular manufacturing system to make a product fully as precise as itself. So covalent bonds between atoms are analogous to transistors. Individual bonds correspond to the ones and zeros level.
So it looks like we’re having to restrict ourselves to covalently bonded solids. Goodbye to metals, ionic solids, molecular solids, macromolecular solids… it looks like we’re now stuck with choosing among the group 4 elements, the classical compound semiconductors and other compounds of elements in groups 3-6. Of these, diamond seems the best choice. But are we stuck with a single material? Chris Phoenix thinks not…
Chris Phoenix: By distinguishing between the nonlinear, precision-preserving level (transistors and bonding) and the level of programmable operations (assembly language and mechanosynthetic operations), it should be clear that the digital approach to mechanosynthesis is not a limitation, and in particular does not limit us to one material. But for convenience, an efficient system will probably produce only a few materials.
This analogy is flawed. In a microprocessor, all the transistors are the same. In a material, the bonds are not the same. This is obviously true if the material contains more than one atom, and even if the material only has one type of atom the bonds won’t be the same if the working surface has any non-trivial topography – hence the importance of steps and edges in surface chemistry. If the bonds don’t behave in the same way, a mechanosynthetic step which works with one bond won’t work with another, and your simple assembly language becomes a rapidly proliferating babel of different operations all of which need to be individually optimised.
Chris Phoenix: For nanoscale operations like binding arbitrary molecules, it remains to be seen how difficult it will be to achieve near-universal competence.
I completely agree with this. A classic target for advanced nanomedicine would be to have a surface which resisted non-specific binding of macromolecules, but recognised one specific molecular target and produced a response on binding. I find it difficult to see how you would do this with a covalently bonded solid.
Chris Phoenix: But most products that we use today do not involve engineered nanoscale operations.
This seems an extraordinary retreat. Nanotechnology isn’t going to make an impact by allowing us to reproduce the products we have today at lower cost; it’s going to need to allow us to make products with a functionality that is now unattainable. These products – and I’m thinking particularly of applications to nanomedicine and to information and communication technologies – will necessarily involve engineered nanoscale operations.
Chris Phoenix: For example, a parameterized nanoscale truss design could produce structures which on larger scales had a vast range of strength, elasticity, and energy dissipation. A nanoscale digital switch could be used to build any circuit, and when combined with an actuator and a power source, could emulate a wide range of deformable structures.
Yes, I agree with this in principle. But we’re coming back to mechanical properties – structural materials, not functional ones. The structural materials we generally use now – wood, steel, brick and concrete – have long since been surpassed by other materials with much superior properties, but we still go on using them. Why? They’re good enough, and the price is right. New structural materials aren’t going to change the world.
Chris Phoenix: A few designs for photon handling, sensing (much of which can be implemented with mechanics), and so on should be enough to build almost any reasonable macro-scale product we can design.
Well, I’m not sure I can share this breezy confidence. How is sensing going to be implemented by mechanics? We’ve already conceded that the molecular recognition events that the most sensitive nanoscale sensing operations depend on are going to be difficult or impossible to implement in covalently bonded systems. Designing band-structures – which we need to do to control light/matter interactions – isn’t an issue of ordinary mechanics, but of many-body quantum theory.
The idea of being able to manipulate atoms in the same way as we manipulate bits is seductive, but ultimately it’s going to prove very limiting. To get the most out of nanotechnology, we’ll need to embrace the complexities of real condensed matter, both hard and soft.
[…] terson, of the Foresight Institute, and Chris Phoenix. I responded to these criticisms in Bits and Atoms. Drexler and Smalley. The most high-profile scientific opponent of Drexle […]
Richard writes: “hydrogen passivated diamond is by far the best candidate for a working material for mechanosynthesis”
This is news to me! It’s never been claimed, that I know of, by any proponent of MNT. Diamond is the most-studied mechanosynthetic target–largely because of Nanosystems. And why was it featured in Nanosystems? Drexler wrote (section 8.6.1), “it is important to choose a few appropriately challenging models. …. a structure built entirely of rings of sp3 carbon atoms appears to maximize the basic challenges of bond formation, and diamond is such a structure.” In other words, he analyzed diamond because it was extra-difficult. As far as I know, no one has even begun to analyze cubic boron nitride, silicon carbide, or sapphire for mechanosynthetic suitability.
So the question “How did we get locked into one material here?” remains to be answered.
Are we restricted to covalent solids? It’s too early to say. But in any case, that doesn’t seem like too much of a restriction. Metals: heavy, rare. Macromolecular: No reason we can’t use those if we want. Ionic: Does anything I said rule out ionic bonds? I think you’re looking for extra limitations to trumpet, rather than actually trying to follow the argument.
As for example, “We‚Äôve already conceded that the molecular recognition events that the most sensitive nanoscale sensing operations depend on are going to be difficult or impossible to implement in covalently bonded systems.” I conceded no such thing. I said that we did not yet know the difficulty of achieving near-universal competence.
Richard: “This analogy is flawed. In a microprocessor, all the transistors are the same. In a material, the bonds are not the same.” You have misunderstood the analogy. The analogy is not between individual transistors and bonds. It is between the nonlinearity of transistors (in general) and the nonlinearity of bonds (in general). Specific bond formation operations are analogous to assembly language instructions. So… “If the bonds don‚Äôt behave in the same way, a mechanosynthetic step which works with one bond won‚Äôt work with another, and your simple assembly language becomes a rapidly proliferating babel of different operations all of which need to be individually optimised.” Assembly language instructions are parameterized. There are a few types–move, add, compare, jump, etc.–which are tweaked to do the exact operation desired. Similarly, I expect there will be be just a few tools which are brought in on different trajectories to do different bonding operations for similar atoms in slightly different conditions. This is not a Babel.
How to do specific binding in covalent solid: google for “molecular imprinting.” In particular, ””This is the first example of molecular imprinting in which a single molecular template is imprinted into a single macromolecule ‚Äì a highly branched polymer called a dendrimer,” said Zimmerman, a William H. and Janet Lycan Professor of Chemistry at Illinois. “Upon removal of the template, we have a synthetic molecular shell that can bind specifically shaped molecules and reject all others, just like a natural antibody.”” from Artificial antibodies created by new molecular imprinting process
I also found an intriguing reference: “However, with advances in technology for the inorganic synthesis of siliceous materials, molecular imprinting on silica has, since the late of 1980s, again become a focus of attention and is now applied to sensor, adsorbent and catalyst preparation.”
My mention of today’s products was not intended as a retreat. Molecular manufacturing will produce revolutionary capabilities in several areas. One is construction of macro-scale products. Another is construction of useful nanoscale structures. By the way, for an overview of nanoscale sensing with a structural toolbox, see Nanomedicine chapter 4.
“The idea of being able to manipulate atoms in the same way as we manipulate bits is seductive, but ultimately it‚Äôs going to prove very limiting.” If you can produce a desired function via a clunky construction that happens to be easy to engineer, is that a limitation or an advance? Digital electronics appears to be “limiting,” compared to analog. But digital continues to replace analog, even for processing of analog signals. Then, microprocessors are inefficient by orders of magnitude in comparison to special-purpose circuitry. But in modern computer designs, more and more functionality is moving into the CPU. At least 99% of the computation done by your computer is spent on saving time retroactively for the hardware and software engineers.
“To get the most out of nanotechnology, we‚Äôll need to embrace the complexities of real condensed matter, both hard and soft.” Here, I agree. A more flexible approach might be several orders of magnitude more efficient. In theory, very preferable. But practically speaking, the engineerability of covalent solids built by mechanosynthesis will probably be the deciding factor.
Chris
I wrote a long comment with lots of references. I forgot about the spam filter. Richard, could you dig that comment out and either email it to me or post it?
Briefly: For selective binding of molecules in covalent solids, google “molecular imprinting.” Including a mention of doing it in silica.
You misunderstood my analogy. Transistor is not bond. Non-linearity of transistor is non-linearity of bonds in general. Specific bonds are assembly language instructions–which are parameterized, as the use of tools will be to handle different bonding environments.
Your claim that “hydrogen passivated diamond is by far the best candidate for a working material for mechanosynthesis” has no support that I know of. No one has investigated cubic boron nitride, silicon carbide, or sapphire. Drexler investigated diamond because it is unusually difficult (Nanosystems 8.6.1).
Why are we restricted from using ionic solids and macromolecular systems? If covalent solids are available, then we have restoration of precision. That’s very useful. But that doesn’t mean they’re the only things we can handle.
I emphatically did not concede that “molecular recognition events that the most sensitive nanoscale sensing operations depend on are going to be difficult or impossible to implement in covalently bonded systems.” I said that “it remains to be seen how difficult it will be to achieve near-universal competence.”
You said, “The idea of being able to manipulate atoms in the same way as we manipulate bits is seductive, but ultimately it‚Äôs going to prove very limiting. To get the most out of nanotechnology, we‚Äôll need to embrace the complexities of real condensed matter, both hard and soft.”
Well, bits are also limiting. As I already pointed out, transistors are not digital. There are still people who say analog computing will make a comeback. I’ll believe it when I see it. The trend for the past six decades is toward easier engineering; people write whole applications in scripting languages, throwing away an order of magnitude of efficiency. That’s on top of the two orders of magnitude thrown away by using CPU’s instead of special-purpose logic. And there are various other inefficiencies that probably add up to another order of magnitude or two. That means that at least 99.99% of the computations your computer does are spent on retroactively saving time for the hardware and software engineers who designed it. So I agree that molecular manufacturing will not “get the most out of nanotechnology,” but I do not see this as a very important consideration compared with the engineerability of mechanosynthesis-built covalent solids.
By the way, make sure you’re distinguishing between mechanosynthesis and machine-phase chemistry.
Chris
“By the way, make sure you‚Äôre distinguishing between mechanosynthesis and machine-phase chemistry.”
Yes, Chris, distinctions are important. See this post for details.
I’m withdrawing from this debate now. Your post above convinces me that you will continue to put forward arguments (that are sometimes deeply flawed in terms of fundamental physics) on the basis of generalisations and an unwillingness to consider the detail of the underlying physics and chemistry. As has been pointed out by Richard , time and time again the tired Drexlerian argument is “if ‘x’ doesn’t work, well we’ll just try ‘y'”. This has now extended to shifting the focus from diamondoid – the subject of both Freitas’ model (which you championed only 3 months ago) and the vast majority of (perhaps all?) mechanosynthesis research to date – to other materials systems. But, as usual, there’s no attempt to consider the detailed chemistry.
For example, Chris, do you know how many different surface reconstructions SiC forms? Are you aware of the counter-intuitive effects of hydrogen passivation on SiC(100)? Just why do you think that a binary system such as SiC makes an easier mechanosynthetic target than an elemental system such as diamond? (It’d be helpful to cite something other than Nanosystems in response). I’m particularly ‘enamoured’ of the idea that Drexler, instead of picking the simplest prototypical platform that fulfilled the large number of criteria and constraints necessary to implement mechanosynthesis, went that one step further and chose an even more complicated system! Where in Nanosystems does Drexler convincingly put forward an alternative ‘simpler’ system to diamondoid for vacuum mechanosynthesis?
It’s clear to me what the game plan will be: the goal posts will continue to shift and widen and work which is not at all related to the Drexlerian vision will increasingly be misappropriated as a verification of Nanosystems . Almost ten years ago, the (in)famous Scientific American headline described the MNT community as “waiting for breakthroughs”. Let’s wait ten more years and see how far towards the Nanosystems goal of computer-controlled “building of products from the bottom up, molecule by molecule, with atomic precision” we get.
Philip
Richard,
I’ve just posted a comment and also forgot about the spam filter. The comment has been swallowed – I’d appreciate it if you could retrieve it. (It’ll be my last post on Soft Machines for some time…).
Thanks.
Philip
“Where in Nanosystems does Drexler…?” In my lost post was the quote from Nanosystems: “it is important to choose a few appropriately challenging models. …. In general, higher valence and participation in more rings makes an atom more difficult to bond correctly. …. Further, diamond has the highest atom and covalent-bond density of any well-characterized material at ordinary pressures, maximizing problems of steric congestion.”
Drexler does not spend much time in Nanosystems on other families of crosslinked molecules. But surely you are not arguing that *every* family of crosslinked molecule will be either unsuitable for implementing molecular machines or impossible to build with them? We only have to find one. And we are not restricted to vacuum mechanosynthesis (machine-phase chemistry). Silica can be built enzymatically underwater.
Why do I think binary systems might make an easier target? Simply because binary systems form at all, indicating that the binary state is energetically preferred to states with atoms misplaced. This implies that atoms being deposited may be less likely to indulge in unwanted side reactions. I don’t expect this to be universally true, because there are factors like surface/volume differences, and different configurations may be favored at different temperatures. All I said about SiC was that no one had investigated it. But I’m pretty interested in cubic boron nitride.
Linear polymers, crosslinked post-synthesis, may or may not count as mechanosynthesis, depending on the details of how they are synthesized. If someone developed a system of engineered molecular machines that would rapidly synthesize a polymer family which could be used to build those machines as well as a variety of high-performance products that were relatively easy to engineer, I would consider that a vindication of Drexler’s work regardless of whether the chemistry of it fit the precise definition of mechanosynthesis.
I am not trying to redefine mechanosynthesis or Drexler’s work. I’m pointing out how broad they already are. I’m sorry if you don’t like the fact that the field of molecular manufacturing is a lot broader than the few ideas that have been lampooned over the past decade. Insisting that I defend a narrow sliver of this field is a nice rhetorical trick, but it is not the way to advance the field.
Chris
Chris:
Why diamond? Because of the issues of surface reconstruction raised in your debate with Philip. Just because other surfaces haven’t been studied by the MNT community doesn’t mean there isn’t detailed knowledge available about their surface structures.
Why not ionic solids? Because the bonding is much less directional than covalent bonding, less non-linear, in your language, and because their surfaces, being charged, are pretty much impossible to get clean.
What about molecular imprinting? You’ll recall I’m a polymer surface scientist, so I know this literature a little. Almost all work on molecular imprinting is done, like the example with a dendrimer you quote, in polymer surfaces. This is essential because you need the squidginess of polymers to get a good fit to the templated molecules. Imprinting in silica is done by the sol-gel process – i.e. you first form Si-O-Si polymers, in which you do the imprinting, then you crosslink the polymers, and remove the organic moieties and solvent. The result is not crystalline silica, but a glass.
Chris writes: .Linear polymers, crosslinked post-synthesis, may or may not count as mechanosynthesis, depending on the details of how they are synthesized. If someone developed a system of engineered molecular machines that would rapidly synthesize a polymer family which could be used to build those machines as well as a variety of high-performance products that were relatively easy to engineer, I would consider that a vindication of Drexler’s work regardless of whether the chemistry of it fit the precise definition of mechanosynthesis.
Let me repeat the comment I made on the CRN blog in the context of Ned Seeman’s work, which comes closest to what you propose here:
if you want to know why Seeman’s work is quite different to mechanosynthesis, read my book again. If you now think that the approach to nanotechnology outlined in Soft Machines is the right one, that’s great, come out and say so. But the last time you mentioned the book, it was to say it was wrong.
I’d be happy to say that success in this project would vindicate Drexler’s broad vision, as enunciated in Engines of Creation. But it certainly wouldn’t vindicate the approach outlined in Nanosystems.
This seems an extraordinary retreat. Nanotechnology isn’t going to make an impact by allowing us to reproduce the products we have today at lower cost; it’s going to need to allow us to make products with a functionality that is now unattainable.
I think this statement is wrong. It’s like saying, in the 1940’s, that electronic computers would not make an impact because they only reproduce computations that could already be performed with mechanical calculators. If the cost and speed of producing an existing thing change beyond a certain threshold, the impact can indeed be revolutionary.
Hi Richard — Due to the similarity between my name and Chris Phoenix’s, I am moving to using “Christine” instead of Chris. I reply to the above discussion at: http://nanodot.org/article.pl?sid=05/02/22/050237.
–Christine
Oops, there’s a period interfering: try this URL instead:
http://nanodot.org/article.pl?sid=05/02/22/050237
William, your point is a very fair one. Another, current, example would be the impact on the world economy if photovoltaics became a factor of ten cheaper and manufacturable on a scale a factor of 100 larger.
Richard writes: “If you now think that the approach to nanotechnology outlined in Soft Machines is the right one, that‚Äôs great, come out and say so. But the last time you mentioned the book, it was to say it was wrong.”
What I said was that your book was wrong about the problems with stiff machines. That’s a small part of your book.
Mechanosynthesis can be done with biopolymers in water. Mechanical engineering (systems with well-defined shape and few degrees of freedom) can be done with at least some biopolymers. From the other direction, self-assembly may be part of a biopolymer-based molecular manufacturing system.
Chris