|
All machines are specializedOne of the central themes of nanotechnology, one of the ideas that distinguishes it from any other kind of technology, is the idea of the general-purpose machine. In nature, each tree bears its own characteristic fruit. In industry, each assembly line produces one product. There is no such thing as an assembly line that produces pillows, butter, sailboats, and rolls of film. Nor is there such a thing as a pillow factory that can be reprogrammed to produce sailboats. Factories are hardwired to produce one thing. If a car company needs to retool a factory that has been producing six cylinder engines so it can produce four cylinder engines, this retooling project takes years and costs hundreds of millions of dollars. Likewise for software. Programs are written one at a time. There is such a thing as a compiler generator -- give it the syntax of a new language, and it will generate a compiler for that language. There is such a thing as a report program generator -- give it the specs for the report you want, and it will generate a program to produce that report from your database. But even these program generators are specialized. There is no such thing as a general purpose program-generator. (A general purpose program-generator would be an IDE, which will "generate" whatever program you want, if you tell it how. The more it can do, the less automatic it will be. There is always a tradeoff between the generality of the machine and the complexity of the interface -- a point that has come up before.) But with nanotechnology, we are, supposedly, going to have general purpose machines that can produce anything. I maintain that this idea makes no more sense in the nano domain than anywhere else. Suppose we do have a nanomachine that can produce both butter and rolls of film. Scale it up to macroscopic size. What would it look like? In the International Herald Tribune, Nov 5-6, 1998, page 11, there is an article about a project at General Motors. By introducing a new production system, they hope to "cut the cost of creating a new vehicle from $1 billion to $360 million." Imagine General Motors scaled down to the nano level, so it fits into a desktop manufacturing system. There is a nanofactory in there which corresponds to a macroscopic car factory. Retooling this nanofactory would be a project of the same magnitude as retooling a macroscopic factory. It would still cost hundreds of millions of dollars. The cost of retooling is independent of the size of the machinery being retooled. Consider one of the little devices that Eric Drexler described in Nanosystems. For example on page 400, you have the stiff manipulator arm (minus the tip!), designed for placing atoms. There will be other molecular machines making these manipulators. What if you need to change the design for some reason? Maybe a manipulator that works for one purpose would not work for some other purpose, so you need to design a different one. That would require changing everything along the assembly line. It would be a task comparable to retooling a factory that has been making six cylinder engines so that the factory can make four cylinder engines. Retooling (or reprogramming) an assembly line that makes manipulator arms is just as complex, and it's going to be just as expensive, and take just as long, as retooling an assembly line that makes engines. Scaling an assembly line down to the atomic level is not going to make it any easier or any cheaper to retool it. The cost of retooling depends on the complexity of the machinery, not on its size. A Shape Shifter that could morph one machine into another would have to be programmed. There is no way around this. It's not a question of what can or can't be done with atoms, or what can or can't be done with computers. It's just that all systems have interfaces. Going back to an earlier section, "Autarky is unstable" -- one of my postulates is Finiteness Assumption #3: The cost of retooling or reprogramming a machine is nonzero, and the cost varies depending on what kind of retooling is done. A general purpose machine would be possible only in a universe in which reprogramming and retooling are instantaneous and free, because such work is done by Genies. In our universe, there are no Genies. Automated engineering systems already exist. Like anything else that's automated, they exist in a larger context which is not automated, and the work they do is not free. Planning is not free, design is not free, programming is not free, and retooling is not free, even if such work is done (or assisted) by automated systems. This will continue to be true of automated systems in the future. Atomic positioners will be specialized. In biochemistry there is a vast array of transport mechanisms. Hemoglobin moves oxygen from place to place, bacteriorhodopsin pumps protons across membranes, and so forth. There are as many transport mechanisms as there are things to be transported. This will also be true of atomic positioners. Twelve years after the publication of Engines of Creation, and six years after the publication of Nanosystems, we still don't have an assembler arm tip that will pick up a carbon atom and put it into place, but suppose somebody eventually does figure out how to do this -- the same tip that picks up carbon atoms will not be able to pick up hydrogen or fluorine or uranium atoms. An assembler arm tip that carries carbon atoms will only carry carbon atoms, for the same reason that hemoglobin only carries oxygen. Using lasers to position atoms would mitigate this problem to an extent, but only to an extent. Each element has unique properties and has to be handled in its own unique way. Besides, there is the other side of a positioner, which can't be taken for granted -- there has to be some way to hold the workpiece in place while the positioner adds atoms to it. Suppose we have a jig which holds a partially constructed diamond in place while carbon atoms are added to it. This same jig would not be able to hold a partially constructed lipid membrane in place while carbon, hydrogen, oxygen, nitrogen, and phosphorus atoms are added to it. There must also be some mechanism for feeding atoms to the positioner, and here, once again, there will be as many mechanisms as there are elements. Machines that use molecular manufacturing to make macroscopic objects, like the vat that makes rocket engines (Engines of Creation, pp 60-62), will be specialized. There will be one vat that is hardwired to make rocket engines, another that makes fuel cells, another that makes batteries, and so forth. They will be hardwired to make one thing, for the same reason any other factory is hardwired to make one thing.
Even programmable machines are specializedSome machines are programmable -- ribosomes, for example. To what extent does this affect my point in the previous section? First of all, ribosomes make proteins. That's all. They can make anything in that domain, not anything in general. Actually, we don't really know that a ribosome can make any protein. The fact that a ribosome can make any polypeptide chain doesn't necessarily imply that it can make any protein. Once you have a polypeptide chain, it has to be folded up before you can get a useful protein. To get it to fold up properly, other proteins, known as chaperones, have to be there to hold the first one in place while it folds up. Now since each ribosome only has a finite number of enzymes, it can only make a finite number of proteins. Even though it can make any polypeptide chain you want, it can only make so many useful ones. Most of them will not be able to fold up into something useful. I think when we find out more about ribosomes -- we don't know much about them in detail at this point -- I predict that it will turn out that they are specialized. There must be many different kinds. Each ribosome is specialized to make a certain set of proteins. That's my conjecture. Anyway, this is a digression. Even if my prediction turns out to be wrong, i.e. even if a ribosome can make any protein, that's still a long, long way from being able to make "anything." A ribosome can't be reprogrammed to produce microtubules or milk, not to mention steel girders. There are machines at the macro level that can be reprogrammed to some extent. A car factory can make a variety of cars on the same assembly line. They don't all have to have the same upholstery or the same stereo system. Many variations are possible. But the assembly line only produces cars, and only cars of a certain type. There are plastics factories where producing a different toy is just a matter of using a different mold and injecting a different mix of plastics. But these factories only produce small plastic objects. They can make anything in that domain, not anything in general. There are programmable machine tools, but once again, they can be programmed to make anything in a certain domain, not anything in general. This will also be true of nanotechnology. There will be machines that can produce anything in a certain domain. Some of them may take standard parts -- little diamondoid structural shapes -- and put them together into a variety of larger structures, just like a ribosome starts with amino acids and makes proteins. (But I haven't seen a design for such a device, or a proposal for a set of structural shapes to use as inputs.) They won't be able to make literally anything. A diamondoid machine won't be able to make lipid membranes any more than a ribosome can make diamonds. In fact there will be many kinds of machines at the nano level. Some will be specialized to make exactly one product (perhaps with slight variations, like a car factory). Others will be like ribosomes or programmable machine tools. They will be able to make anything in a certain domain. Some of them may make copies of themselves, but (a) this isn't really necessary -- as long as the system as a whole can make a copy of any part of itself, that's all we need, and (b) in any case, they will make something plus copies of themselves, not anything plus copies of themselves. If we ever get to the point where nanotechnology can produce as wide a variety of products as nature produces, then there will be as many different kinds of nanomachines as there are different kinds of cells, and nanomachines that do what cells do will be as complex as cells. Likewise, if we ever get to the point where nanotechnology can produce the same variety of products that the world economy produces, there will be as many different kinds of nanomachines as there are different kinds of machines in the present economy, and the nanomachines will be as complex as the machines they replace.
|