Life is characterized in part by the processes of birth, growth, and healing. Analogous processes in machines could make them cheap, plentiful, adaptable, reliable, and long-lasting.
“Self-replication” is a technical term with a more restricted meaning than “reproduction”. Reproduction is part of the evolutionary process, and implies the admixture of genetic information and the allowance for (random) variation in the offspring in preparation for selection by the environment. Self-replication, on the other hand, means that none of this happens; the self-replicating mechanism simply makes as true a copy of itself as possible.
I use the term “self-construction” to refer to a system which can “grow into” its useful configuration from a much smaller “seed” configuration. This is to be distinguished from self-assembly, which refers to a stochastic process where only the final state, and not the pathway taken to it, is specified.
“Self-maintenance” should be taken to include both continuous or scheduled operations mounted against normal wear and tear, and the detection as well as the repair of damage. This would encompass things like an “immune” system, sensory systems, internal consistency checkers, and so forth, as well as the ability to remove, replace, and possibly remanufacture broken parts.
These capabilities — which I will lump together under the heading of reflexive capabilities — are an essential part of a complete technological base for molecular manufacturing. The goal for molecular manufacturing can be simply stated: extend Moore’s Law from computers to general manufacturing.
Reflexive capabilities are, if anything, more important to this goal than the ability, by itself, to manipulate matter at the molecular scale. To illustrate this, let’s consider two scenarios for future technology with one and not the other:
First, assume there is the ability, say, using some scanning probe developments, to do mechanochemistry, and in general to build any desired object specified with atomic precision, one reaction at a time. If the manufacturing loop is not closed, we would expect to see things like advanced circuits and sensors, catalysts for chemical processes, and so forth, along the lines of today’s high-tech chips and materials, but still among the highest priced items made, on a pound-for-pound basis. My impression is that this is actually the model that many researchers and forecasters have of near-term nanotechnology.
Assume, on the other hand, that there is a factory that builds robots, which are general-purpose enough to build factories. Assume they use bricks and mortar, rolled steel and nuts and bolts and wires and the existing panoply of chips and computer hardware. Robots that can build and staff a robot factory can easily build most other kinds of factory. Now machines can move faster, apply greater forces, and tolerate greater environmental extremes than humans. So as the processing capabilities increase, the robot factories would tend to become more productive, not only in terms of their end output but also in terms of building more factories.
Suppose it costs $100,000 per year to hire a human worker (wages, benefits, employer tax contributions, equipment, etc). Suppose a robot can be built with $10,000 worth of parts and does the same things as the human, but twice as fast. Suppose it takes 100 humans a year to build a factory (you’ll have to pay them $10 million) which can build 100 robots a year if staffed by 100 humans.
You’d have to sell a robot for $120,000 (parts, labor, and interest on your factory) just to break even. However, in a year a factory staffed by robots can build 200 robots and another factory (built by the first 100 robots in the second half of the year). You can sell 100 robots for $32,000 each (parts and interest on the factory and the robots) and have two fully staffed factories. Next year you repeat the process, sell 200 robots for $21,000, and have 4 factories. Very quickly you can sell robots for essentially the cost of the parts. Apply some of your factories to building parts from raw materials… The pattern should be clear.
Combining the reflexive technology base with mechanochemistry and nanorobotics produces a cycle time measured in hours instead of years (not to mention more portable factories). However, thinking about the robots and factories model not only shows us how the system produces exponentially decreasing costs (asymptotic to raw materials) but also gives some insight into self-construction and -maintenance. A factory full of robots has some of the robots doing maintenance, repair, and maybe constructing extensions just as one with human workers would do. The trick is to think of the factory itself as the self-replicating, -constructing, or -maintaining machine.
For further reading, here is a good book and some Web pages. My Foresight conference paper goes into some detail about how I think these concepts could be translated to the nanoscale.
Robot: Mere Machine to Transcendent Mind
by Hans Moravec (Oxford, 1999)
Ralph Merkle’s self-replication page:
A theoretical overview of self-replication studies, by Moshe Sipper:
An article by George Friedman in the Assembler:
My paper from the recent Foresight Conference: