This paper is based on the keynote address delivered by the author at the Conference on Chemically-Based Computer Design in October 1983. The author proposes that the information processing capabilities of organic molecules could be used in computers in place of digital switching primitives. However, this is a purely hypothetical proposition, and no practical investigation of such systems is thought to be in progress.

If such computers could be produced they would, according to the author, operate in a very different fashion from current, so-called von Neumann computers. Von Neumann computers use large numbers of simple switching primitives arranged in highly programmable systems. Molecular computers would incorporate units capable of carrying out highly complex primitive operations, but would sacrifice programmability. The author postulates a trade-off principle: “A system cannot at the same time be effectively programmable, amenable to evolution by variation and selection, and computationally efficient.”

Much of the paper is taken up with justifying this principle and exploring its consequences. There are sections on programmability versus evolvability, programmability versus efficiency, and efficiency versus complexity. (Computer designers have long known that a special purpose machine can be made to run a particular job more effectively than a general purpose machine.) Biological organisms are presented as being capable of solving much larger problems than von Neumann computers and as being very much more efficient. As an example, the author cites the ability of enzymes to perform a tactile pattern recognition task. This task would require many steps using a digital computer; it would also involve, in order of magnitude, more energy expenditure. Digital computers are clearly not programmable in the sense that von Neumann computers are, so “How could a structurally nonprogrammable system be organized to perform useful functions?”

The author believes that there are many possible alternatives, but he focuses on one of them: “the principle of double dynamics.” His hypothetical example involves enzymes immobilized on a surface covered by a photochemically sensitive medium. This medium transduces spatial and temporal patterns of input photons into a chemical gradient (the first level of dynamics). At the second level of dynamics the chemcial concentrations are interpreted by the enzymes, each of which acts as a biosensor coupled to an output device (e.g., through a spectrometeric mechanism). No such device has yet been produced, but the author claims that his group at Wayne State University has simulated devices of this type and “has developed evolutionary learning algorithms for communicating desired performance to them.” The reactions used in these simulations are ones found in real neurons, and they are described in some detail in the paper. Finally the author considers the impetus from biotechnology and presents a sequence of 12 design stages relevant to molecular computer development. It is, the author admits, “patently futuristic.”

It’s all fascinating stuff, but there isn’t a remote chance of there being a system based on these ideas appearing in the foreseeable future. Thus, given that it is all highly speculative, it might have been more appropriately published elsewhere. One can only hope that the annals of computer science will not be filled up with derivative papers for years to come.