The assembly line technique of replication is, as far as I know, an untested human invention. Plants and animals (at least at the cellular level) replicate via mitosis. But here we are dealing with RNA coding which is beyond my ability to demonstrate as it occurs in nature let alone as it might occur in machine production. I think convergent assembly is a better descriptor of the assembly line technique we have in mind. Merkle states that, “The particular architecture proposed should be able to produce meter-sized products in a few minutes from nanometer-sized parts while going through about 30 stages.” This sounds pretty intriguing but all that he is proposing is an assembly technique comparable to the one Ford introduced in 1908 except now the work of people is done by chemistry, and the required chemistry is left unexplained.
There are more possibilities for self-reproduction out there (another) but they are all highly complex and unproven. An example,
(The machine would have four parts – (1) a constructor “A” that can build a machine “X” when fed explicit blueprints of that machine; (2) a blueprint copier “B”; (3) a controller “C” that controls the actions of the constructor and the copier, actuating them alternately; and finally (4) a set of blueprints f(A + B + C) explicitly describing how to build a constructor, a controller, and a copier. The entire replicator may therefore be described as (A + B + C) + f(A + B + C). Now, for the machine “X” that is to be manufactured, let us choose X = (A + B + C). In this special case, Controller C first actuates copier B which copies f(A + B + C) to produce a second copy f(A + B + C), then actuates constructor A to build a second constructor, copier, and controller – i.e., another (A + B + C) – and to tie them together with the second copy of the blueprints. Thus the automaton (A + B + C) + f(A + B + C) produces a second automaton (A + B + C) + f(A + B + C) and so self-replication has taken place. The fact that the description f(A + B + C) must be both copied (uninterpreted data) and translated (interpreted data) during replication is the key to avoiding the paradox of self-reference. Note that this replication takes place in an environment of stockroom parts) ; this isn’t necessarily that mind boggling until you delve into Neumann’s meaning of automaton.
The majority of nanoproduction that is currently going on is the production of lattice tubes, crystal structures, and molecular hybrids . The only nanomachine that has been created as of today is a paddle like object that spins around a tube. There is no information sent to the paddle, it rotates as a reaction to the proximity of two embedded cylinders. Our machines must do more than just react. They have to move along polar coordinates when we instruct them to and they must bond to each other and release at our instruction. How will they do that? We can assume nanochips will be available as they appear to already be in development. But in order for a quadrillion nanobots to be equipped with nanochips they will also have to be part of the self-replicating process, and this adds an immense degree of difficulty to an already complicated production dilemma. But lets assume that this is possible, is it possible to communicate with a quadrillion robots that are 100,000 times smaller than a hair on my head? Do we tell each robot what to do or just some of the robots? Maybe we don’t have to communicate with all of the bots, possibly the majority of bots are preprogrammed with some type of fitness criteria. And in the end, I am discussing communication with nanorobots when I don’t know how to communicate with a lego robot. Is there time to learn that and develop a believable production assembly? We don’t necessarily have to explicitly demonstrate everything but we have to careful not to look like alchemists creating something from nothing.
Additionally, is it appropriate to assume the existence of a smart material or should we create one?