Crystal growth does not qualify as a reproductive process, whilst the real process involving DNA molecules and a whole chemical environment remains far from being fully understood; nevertheless one is skeptical that it will be found to contain some complex computing mechanism. Certainly comprehension of biological reproduction has advanced considerably since those days when chromosomes were known but genes were merely conjectured on the basis of empirical observations, which was still the case when cellular automata came into being.
Questions of reproduction, be it sexual, asexual, or self reproduction, tend to be rather philosophical in nature, along with questions of consciousness, the borderline between the living and the nonliving, and similar issues. A robot is easy enough to imagine, not to mention the possibility of its having a certain dexterity and the ability to follow instructions of a sort. Never mind that functional examples may be in short supply, but it was just this gap between fanciful speculation and actual practice which led von Neumann to settle instead for self patterning wallpaper. The suggestion came from Stanislaw Ulam, who had been experimenting extensively with functional transformations and means of representing them.
By working with a medium which could support a variant of Alan Turing's hypothetical computer, the constructor which von Neumann sought could avoid the monotony of reproduction through crystal growth by calculating the nature of its intended construct, opening quite a few possibilities of variability, adaptation, and evolution. Presumably the use of a full Turing machine would leave the constructor with the utmost versatility possible in choosing its products, or the strategies for carrying out its constructions. Consequently it would appear that whatever such a model lacked in realism might be compensated by the possibility of establishing theorems similar to those which Turing proved for general computations, namely the existence of a universal machine and of undecidable calculations.
The resulting object was cumbersome, containing twenty nine state cells arranged on a two dimensional square lattice, and a finished universal constructor that would occupy an area thousands of squares on a side if it were ever to be ``built.'' Given that the construction originally contrived to prove a point is not necessarily the simplest or most elegant route to the same conclusion, it is not surprising that there are other alternatives, such as one with eight states found by Edgar Codd[15]. Nor is it surprising that von Neumann himself found that as he worked out the details of more and more of his proposed machine, he encountered ideas that would simplify the construction of the parts of the machine which he had already designed. Earlier this trap had thoroughly ensnared Babbage, who repeatedly laid aside plans for one machine in favor of another, still more elegant one, while gradually running out of financing.
The requirements which von Neumann laid on his machine may have been much more severe than necessary. If only the emulation of a Turing machine is required, Alvy Ray Smith III[65] has shown a fairly straightforward way to do so; if only reproduction is required, Christopher Langton[46],[47] has shown some simple variants of Codd's automaton which will grow, neither using a computer to guide the expansion, nor even their own description. Langton does argue, however, that a description of sorts is inherent in the form of the simplest reproducing unit of the organism.
The search for automata with elegant computational properties still goes on, and automata continue to be judged either by the power of the computation that they are able to carry on, or the power of the computer required to understand their behavior.