The brilliant Hungarian-American mathematician John von Neumann (1903-57) was the first person to consider the problem of self-replicating systems seriously. His book on the subject, Theory of Self-reproducing Automata, was edited by a colleague, Arthur W. Burks (1915- ), and was published posthumously in 1966—almost a decade after von Neumann's untimely death due to cancer.
Von Neumann became interested in the study of automatic replication as part of his wide-ranging interests in complicated machines. His work during the World War II Manhattan Project (the top secret U.S. atomic bomb project) led him into automatic computing. Through this association, he became fascinated with the idea of large complex computing machines. In fact, he invented the scheme that is used today in the great majority of general-purpose digital computers—the von Neumann concept of serial processing stored-program—and that is also referred to as the von Neumann machine.
In 1945, von Neumann drafted a report in which he introduced the concept of the stored-program computer. He also recognized that base 2 represented a considerable gain in computer design simplicity over the base-10 approach, which had been used in the world's first working electronic calculator and digital computer, the Electronic Numerical Integrator and Computer (ENIAC). It was completed in 1946 and contained 18,000 vacuum tubes. While it was a major step forward in the evolution of "thinking machines," ENIAC stored and manipulated numbers in base 10. Von Neumann's suggestion of using the base 2 allowed circuits in the digital computer to assume only two states: on or off or 0 or 1 (in binary notation).
Following his pioneering work in computer science, of which he is one of the founding fathers, von Neumann decided to tackle the larger problem of developing a self-replicating machine. The theory of automata provided him a convenient synthesis of his early efforts in logic and proof theory and his more recent efforts (during and after World War II) on large-scale electronic computers. Von Neumann continued to work on the intriguing idea of a self-replicating machine and its implications until his death in 1957.
Von Neumann actually conceived of several types of self-replicating systems, which he called the kinematic machine, the cellular machine, the neuron-type machine, the continuous machine, and the probabilistic machine. Unfortunately, he was only able to develop a very informal description of the kinematic machine before his death in 1957.
The kinematic machine is the most frequently discussed von Neumanntype self-replicating system. For this type of SRS, von Neumann envisioned a machine residing in a "sea of spare parts." The kinematic machine would have a memory tape that instructed the device to go through certain mechanical procedures. Using manipulator arms and its ability to move around, this type of SRS would gather and assemble parts. The stored computer program would instruct the machine to reach out and pick up a certain part and then go through an identification and evaluation routine to determine whether the part selected was or was not called for by the master tape. (Note: in von Neumann's day, microprocessors, minicomputers, floppy disks, CD-ROMs, and multigigabyte-capacity laptop computers did not exist.) If the component picked up by the manipulator arm did not meet the selection criteria, it was tossed back into the parts bin (that is, back into the "sea of parts.") The process would continue until the required part was found, and then an assembly operation would be performed. In this way, von Neumann's kinematic SRS eventually would make a complete replica of itself—without, however, understanding what it was doing. When the duplicate was physically completed, the parent machine would make a copy of its own memory tape on the (initially) blank tape of its offspring. The last instruction on the parent's machine tape would be to activate the tape of its mechanical progeny. The offspring kinematic SRS could then start searching the "sea of parts" for components to build yet another generation of SRS units.
In dealing with his self-replicating system concepts, von Neumann concluded that these machines should include the following characteristics and capabilities: (1) logical universality; (2) construction capability; (3) constructional universality; and (4) self-replication. Logical universality is simply the device's ability to function as a general-purpose computer. To be able to make copies of itself, a machine must be capable of manipulating information, energy, and materials. This is what is meant by the term construction capability. The closely related term constructional universality is a characteristic that implies the machine's ability to manufacture any of the finite-sized machines that can be built from a finite number of different parts, which are available from an indefinitely large supply. The characteristic of self-reproduction means that the original machine, given a sufficient number of component parts (of which it is made) and sufficient instructions, can make additional replicas of itself.
One characteristic of SRS devices that von Neumann did not address but that has been addressed by subsequent investigators is the concept of evolution. In a long sequence of machines that make machines like themselves, can successive robot generations learn how to make themselves better machines? Robot engineers and artificial intelligence experts are exploring this intriguing issue as part of the larger question of thinking machines that are self-aware.
Can robots be made smart and alert enough to learn from the experiences that are encountered in daily operations and thus improve their performance? If so, will such improvements simply reflect a primitive level of machine learning, or will the smart machines somehow begin to develop an internal sense of "knowing" that they know? If and when this ever occurs, the smart robot will begin to mimic the consciousness of its human creators. Some AI researchers like to speculate boldly that an advanced "thinking" robot in the distant future could be capable
Replication Production Growth
Replication Production Growth
The five general classes of self-replicating system (SRS) behavior: production, replication, growth, evolution, and repair. (NASA)
of formulating the famous philosophical postulate of René Descartes (1596-1650): "Cogito, ergo sum" (I think, therefore I am). An SRS unit that exhibits the behavior of evolution might certainly be capable of achieving some form of machine self-awareness.
From von Neumann's work and the more recent work of other investigators, five broad classes of SRS behavior have been suggested:
1. Production. The generation of useful output from useful input. In the production process, the unit machine remains unchanged. Production is a simple behavior demonstrated by all working machines, including SRS devices.
2. Replication. The complete manufacture of a physical copy of the original machine unit by the machine unit itself.
3. Growth. An increase in the mass of the original machine unit by its own actions, while still retaining the integrity of its original design. For example, the machine might add an additional set of storage compartments in which to keep a larger supply of parts or constituent materials.
4. Evolution. An increase in the complexity of the unit machine's function or structure. This is accomplished by additions or deletions to existing subsystems or by changing the characteristics of these subsystems.
5. Repair. Any operation performed by a unit machine on itself that helps reconstruct, reconfigure, or replace existing subsystems—but does not change the SRS unit population, the original unit mass, or its functional complexity.
In theory, replicating systems can be designed to exhibit any or all of these machine behaviors. When such machines are actually built, however, a particular SRS unit will most likely emphasize just one or several kinds of machine behavior, even if it were capable of exhibiting all of them. For example, the fully autonomous, general-purpose self-replicating lunar factory, proposed in 1980 by Georg von Tiesenhausen and Wesley A. Darbo of the Marshall Space Flight Center (MSFC), is an SRS design concept that is intended for unit replication. There are four major subsystems that make up this proposed SRS unit. First, a materials processing subsystem gathers raw materials from its extraterrestrial environment (the lunar surface) and prepares industrial feedstock. Next, a parts production subsystem uses this feedstock to manufacture other parts or entire machines.
At this point, the conceptual SRS unit has two basic outputs. Parts may flow to the universal constructor (UC) subsystem, where they are used to make a new SRS unit (this is replication), or parts may flow to a production facility subsystem, where they are made into commercially useful products. This self-replicating lunar factory has other secondary subsystems, such as a materials depot, parts depot, power supply, and command and control center.
The universal constructor (UC) manufactures complete SRS units that are exact replicas of the original SRS unit. Each replica can then
make additional replicas of itself until a preselected SRS unit population is achieved. The universal constructor would retain overall command and control (C&C) responsibilities for its own SRS unit as well as for its mechanical progeny—until, at least, the C&C functions themselves have been duplicated and transferred to the new units. To avoid cases of uncontrollable exponential growth of such SRS units in some planetary resource environments, the human masters of these devices may reserve the final step of the C&C transfer function to themselves or so design the SRS units such that the final C&C transfer function from machine to machine can be overridden by external human commands.
Was this article helpful?