Previous Top

Appendix 5A:
First Attempt To Define A Self-Replicating System

By W. E. Bradley, June 1980

At a recent meeting a member of the NASA Advisory Council expressed excitement at the positive conclusions reached by the June 1979 Woods Hole Symposium concerning self-replicating mechanical systems. He said that he could not understand why a subject of such interest and importance to the exploration and utilization of space should be approached so timidly. Earnestly, he added: "After all, a lathe can produce a lathe, properly operated; nowadays numerically controlled lathes are available; so why not program one to reproduce itself?"

My reaction was the following:

• A lathe cannot produce another lathe without many added subsystems (e.g., driving motor, tool grinder, tool bit production, etc.).
• Some contemplation of the self-replicating system problems at the practical engineering level has been undertaken by a few individuals in the past few months. This work is incomplete as yet, but is aimed at practical, demonstrable systems with only a few critical parts supplied from outside the- system, including energy and raw materials for device fabrication. Energy and raw materials appear here in the role of "nutrient," the supply necessarily increasing as the system grows.
• The self-replicating system is indeed of great interest on fundamental grounds.
• The subject is appropriate to and important to NASA.

The work of the past few months (prior to the present study) relevant to self-replicating systems (SRS) is incomplete but has brought to light some principles and ideas of interest.

5A.1 Preliminary Investigation of the Self-Replicating Machine Shop

The town of Muncy is located in a somewhat remote part of central Pennsylvania. It is remarkable because of a nearly self-sufficient machine manufacturing capability in the Sprout-Waldron Company (now a division of another corporation, and therefore subject to change without notice). This company has manufactured agricultural and food-processing equipment as well as heavy machinery for the paper industry, especially pulp grinders. I became acquainted with them while searching for machines able to produce dense pellets for use as solid fuel from agricultural cellulosic wastes.

In the course of my visit, I was shown an excellent machine shop, a foundry, a woodworking shop, and a factory assembly space in which their machines were put together, painted and tested. They also had complete drafting and design engineering facilities. Of special interest was their toolmaking and repair shop, with which all of the milling machines, lathes, jig borers, punch presses, and so forth were kept in fine working order.

This complex, with the possible exception of the foundry, seemed to be a system which, with human assistance, could duplicate itself. In retrospect, it seems worthwhile to explore the possibility that the human operators might be replaced by general purpose automata, manufactured almost completely by the complex itself. The result would then be a major component of a self-replicating system. To complete the system would require manufacture of a prime power source which could be expanded as the complex grows, manufacture of a shelter system (sheds with roofs, walls, windows, and doors) similarly expandable, and possibly a casting and/or forging subsystem, and electronic and computer components of the automata. The foundry with its requirement for refractory furnace linings and high temperatures is a special problem and in some versions of the system may be bypassed.

Present machine shops. Each machine in a machine shop has a functional domain or "scope," assuming unlimited operator attention and guidance. Thus, a lathe (with no attachments) is able to produce objects with cylindrical symmetry having axial length and maximum diameter determined by the "bed length" and the "swing" of the machine. It can also make threads (helical structures), and, to a limited extent, can also make straight-line cuts or grooves which are more properly the work of a milling machine. Lathes can drill holes most readily on the axis of a workpiece of cylindrical symmetry and can achieve a high degree of accuracy of concentricity for this one type of drilling. Most drilling, however, is best accomplished on a jig borer.

The second major machine type in a shop is some form of drill press, or, better, a jig borer. The workpiece is held firmly in an accurately translatable and rotatable fixture, remaining stationary while holes are drilled by a drill or boring tool held in a chuck rotating about the principal axis of the machine. Such a device can produce clusters of accurately located holes with parallel axes.

The third important shop component is the milling machine. The workpiece is clamped firmly to an accurately controlled table. The workpiece moves continuously, slowly, during operations while the rotating milling cutter shaves or saws the surface being worked. The milling machine is usually used to make rectilinear cuts to form accurately related plane surfaces or grooves.

A well-equipped machine shop usually also includes a power hacksaw, a powerful press with forming dies for forming sheet metal and for punching holes with "punch and die" sets, a bending brake, tool grinders, and possibly a surface grinder to be used like a milling machine to produce flat surfaces.

Self-replicating shop and universal machines. Each machine or subsystem of such a shop can be separated into parts from which it can be reassembled. Each machine therefore has a "parts list," and each part either can or cannot be fabricated by the set of machines and subsystems comprising the shop. The criterion for replication thus may be stated as follows:

If all parts of all machines and subsystems can be fabricated within the shop, then if properly operated the entire shop can be replicated.

"Proper operation" in this context includes supplying raw materials, energy, and manipulatory instructions or actions necessary to carry out the large number of machine operations, parts storage, and parts assembly required. Human labor is now used for these functions.

It is not necessary that the shop be able to produce anything except a replica of itself which is in turn capable of producing another. Therefore, some simplifications appear possible, such as standardization and limitation of scope where feasible. For example, a universal machine can be imagined with a wider cross feed table than a conventional lathe and with a standardized vise and tool holder so that it can be used for milling. All three dimensions of translation and one axis of rotation could be provided on the table. The head stock could be arranged to hold work-pieces, milling cutters or drills. Hardened tools for the necessary cutting operations could be fabricated by the machine from carbon steel in the annealed condition, then tempered, drawn, and sharpened by a separate simpler machine including a small furnace and a tool grinding wheel equipped with tool-holder and feeds. By careful standardization of parts, tools, and fixtures, it is conceivable that such a "one-machine shop" could succeed in reproducing itself.

"Factons." After a shop had been tested with human operators and proven capable of self-replication, it would be possible to explore the replacement of the human operators by mobile computer-controlled manipulators, or "factons." Hopefully, all of the "numerical control" features could be contained in these general-purpose programmable devices which could handle the machines like a human operator. The factons would transfer work from operation to operation, adjust the machine, perform each operation. then transfer the work to a parts storage array. Finally, the parts would be assembled by the factons and the entire shop set up in a selected location and floor-plan. The facton itself has a parts list, most designed to be manufacturable by the shop. Here it is practically inevitable that the computer chips plus enormous memories will be needed which would fall outside the scope of the shop thus far envisioned. In other words most, but not all, of facton components could be fabricated by them in the shop. Still, given these extra components provided from outside, the factons could probably fully assemble themselves. The shop itself would require some exogenous elements, as noted above. Prime power, shaft power transmission such as belting or electric motors, abrasives, furnace heating arrangements for tool heat treatment, raw material such as basic feedstock including steel rods, strips, and plates are among the most obvious.

Using the same facton design, it should be possible to implement extensions of the shop, including an optical shop, a pneumatic and/or hydraulic equipment manufacturing shop, and ultimately even an integrated circuit shop. Note, however, that only the original shop with its factons and their programs would have to possess the capability for self-replication.

Computer components, probably provided from outside the system, might be furnished in an unprogrammed condition. Thus, factons would program the tapes, discs, or read-only memories by replication (and verification) of their existing programs. This procedure allows for the possibilities of "heritable" changes of program embodying "devolution" (simplification) or "evolution" (capability augmentation) by orderly program amendment.

5A.2 Program Extension Beyond Self-Replication

The "scope" of a self-replicating shop is much larger than is required for self-replication. Apparently the ability to replicate utilizes only a vanishingly small fraction of total capabilities (to produce various sizes and shapes of parts and to assemble them into machines and structures). The essential characteristic for self-replication is that the scope must be adequate to produce every part of every machine in the shop by means of a feasible program. This "closure condition" can be satisfied using only a small part of the shop's full capabilities.

A generic self-replicating shop can therefore, by means of a simple addition to its program, manufacture other machines and structures and, by means of them, interact with its environment. For example, it can construct and operate foraging systems to procure fuel or materials, waste disposal systems, or transporters to carry replica shops to other locations.

Obviously, self-replication of such an extended system requires replication of the program-memory. This memory can be partitioned into two parts: (1) The self-replication process memory, and (2) the external process (manufacturing) memory. The distinction between these two memories is that the first is required to reproduce the basic unit (shop machines plus factons) while the second memory contains the program to produce process equipment not essential to the self-replicating nucleus.

At this point it is clear that the effect of a self-replicating system on its environment may take many forms dependent on the external process program. Using such a program, the scope of the system can be extended by construction of machines and structures capable of producing complex subsystems including mineral processing plants, solar energy power supplies, etc.

All of these extended self-replicating systems would embody the same basic nucleus of machines, factons and self-replication programming. They would differ only by addition of the external process program segment peculiar to each type.

Reliability and redundancy. Reliability is a primary concern, especially in the case of self-replicating processes. Two ideas are most important here.

First, the self-replicating program accuracy can be verified by comparison with other replicas of the same program. If a discrepancy is found between two self-replicating programs, a third or fourth replica can be consulted and the error pinpointed and corrected. The test of correctness is the ability to self-replicate.

Second, machines tend to wear, and ultimately to fail, from excessive use. On the other hand, if the system can replicate itself it can make spare parts and install them itself. A special program segment, the "maintenance program," should be devised to check machine wear and perform repairs as needed. This segment would be part of the self-replication program, although another somewhat similar maintenance program should probably be used to care for machines and structures of the external process. This external maintenance program would be specialized for each extended system and is properly part of the second memory.

Speciation. Any self-replicating system is actually another species of SRS, the species being dependent upon the contents of the second memory.

A group of interacting extended self-replicating systems may form a still larger self-reproducing system with yet more complex capabilities. It is not immediately apparent what factors limit the possibilities of such systems. Separable subsystems manufactured by a self-replicating shop may be machines of considerable complexity, themselves incapable of self-replication. Their supply is therefore dependent on the self-replication shop and its program.

It is interesting to note that a facton equipped with an aberrant program may function like a virus, visiting a self-replicating shop and using its machines for reproduction of its own type without constructing any other machine. It could then replicate its program for installation in the new "virus facton" and reproduce this way, using materials and energy from a host self-replication shop. This possibility opens up a large field of problems related to the security of self-replication systems from facton defect or infection.

Next