merkle@parc.xerox.com (Ralph Merkle) (04/09/91)
A few points might be in order: In general, the design objectives for a device capable of self replication will include low cost, ease of design, and ability to carry out the desired function. The objectives of low cost and simplicity of design both argue in favor of a single low cost source of fuel. For example, automobiles run on gasoline. Gasoline suitable for use by a car does not normally occur in nature, but must be produced artificially. In sharp contrast, biological systems can survive on a wide variety of foods. This is achieved, however, at the cost of a more complex design which will be less efficient in any specific environment than a design which was not so adaptable. Unlike cars, horses can eat hay, carrots, sugar lumps, grass, etc. Many non-fuel nutrients are extracted from this variety of foods, but from the point of view of energy, the only thing the horse gets is glucose in the blood. (Fat, used as an energy storage device, would not be used by the horse if a reliable supply of energy were available). Thus, if I were to re-design a horse as an efficient transportation device, I would largely eliminate the mouth, stomach, gut, and related parts. Instead, I would feed the horse sugar water with a broth of nutrients. This re-design would reduce the complexity of the horse, increase efficiency of fuel usage (it would only need to absorb glucose -- provided in the "broth,"), reduce overall weight, increase fuel economy, and reduce offensive waste products. The only cost imposed would be the production and sale of a standard "horse broth." In a competitive economy, this would be produced by the most efficient producer using the most efficient techniques. Indeed, I would have carefully designed the horse to use a "broth" that was cheap to produce, this being one of the design criteria. Someone designing a transportation device from the ground up would simply not include the various mechanisms found in biological systems that provide for flexible usage of a wide variety of fuel sources, and would either eliminate or drastically reduce fuel storage. A horse can survive for weeks and perhaps months before starving to death. A car can run for a few hours before running out of gas, and can be easily re-started when gas is again available. If we look at "nutrients," we find that biological systems are able to survive on a wide variety of sources. Again using the horse as an example, proteins and vitamins are extracted from a wide variety of food sources. Elaborate systems for storing nutrients and synthesizing missing nutrients from available nutrients are built in, so that the list of "vitamins" that the horse requires is relatively small. Again, our hypothetical re-design of the horse to make it a more efficient transportation system would largely eliminate these mechanisms. "Horse broth" would contain the right nutrients, in the right proportion, manufactured by the cheapest available technique. The number of "vitamins" in the horse broth would be sharply increased. Instead of having "essential" amino acids that can be used to synthesize the "less essential" amino acids, all 20 amino acids would be essential. Instead of using long metabolic pathways to synthesize small amounts of essential compounds, the compounds would be provided in the "broth." The re-designed horse would be quite unable to survive in the wild, for it would require compounds that simply were not natural and could not be found outside of "horse broth." Finally, if I did indeed want horses to replicate and make more horses (instead of being made in a factory) I would completely eliminate sexual reproduction. Asexual reproduction is lower cost, simpler, and more efficient. The only argument in favor of sexual reproduction is its ability to improve evolutionary adaptability. The last thing I want my re-designed horse to do is to evolve -- I've just spent a lot of time and effort getting the thing designed correctly, and I certainly don't want the design changing without careful planning. If design changes appear appropriate, then the Mark II horse would be introduced, upgraded and improved. In general, products that "evolve" are absolutely not wanted. The term "evolve," when applied to a product, typically means "incremental changes deliberately introduced by a design team to improve product function." "Evolve" in the biological sense involves random mutations, with the ones that improve survival (which is quite different from "product function") being retained while the countless other mutations result in degraded performance and customer complaints. In general, biological systems have a wonderful ability to adapt and evolve. Neither ability will be present in a system designed for mundane economic reasons, nor is it plausible that they would appear in such a product by chance. Anyone who does want to deploy a highly adaptable self replicating system should be carefully regulated (or just told "No!"). While it is true that some researchers will wish to use "evolution" as part of their research or product design efforts, they will use the ideas developed from such evolutionary experiments in products, but will not use the evolutionary experiments themselves as products. Such research is best carried out by computer simulation. Researchers who wish to build actual self replicating devices with the intent of allowing them to evolve should again be very carefully regulated. Regulations in this area must necessarily be international in scope. A uniform set of regulations that are adhered to by all nations is much, much better than stronger regulations that are not universally respected. It is reasonable to develop a social consensus that careful regulation of devices that can evolve and survive in the wild is a good idea. Systems that completely lack the ability to do either will require little or no regulation. The discussion so far has considered only systems designed by benign (though possibly incompetent) individuals who are basically interested in an economically viable product. They have no particularly malign intent. Modest regulations should suffice in most instances, with truly tight regulatory control being limited to a few special cases. Such tight regulatory control would have to be coupled with a broad international consensus that such controls are needed, or the regulations will fail. Military systems are another issue. A simple blanket condemnation of military research and development is not acceptable. Unless we are prepared to say with a straight face that we can guarantee that no one, anywhere, will deliberately design a weapon system to enhance their power, we must be prepared to counter military attack. For the U.S. to forego military R&D completely would guarantee that we would be overwhelmed by an attack based on weapons that we did not understand and could not counter. (I make no comment on the appropriate level of funding of military R&D, nor on any of the specific funding decisions in fact made by the U.S. government. I simply note that in general such funding is both reasonable and necessary, unless we can guarantee the benign intent of everyone, everywhere). Thus, it appears essential to support military R&D by the "most reasonable" countries, while discouraging it among the "less reasonable" ones. The alternative is worse. This suggests the following strategy: pick the country or countries you think are most stable politically and most likely to use their weapons reasonably, and encourage them to (a) stay one jump ahead in the technological arms race and (b) continue to use the weapons in a reasonable way. This leads to the somewhat contradictory stance of both strongly encouraging appropriate military R&D, while at the same time strongly discouraging use of the resulting systems. While the profit motive does not generate strong incentives to design and build a self-replicating system that can survive on a broad range of energy sources, the objectives of a military system might well include such capabilities. It does not appear, however, that even a military system would benefit from the ability to evolve in the uncontrolled Darwinian sense. Thus, designing military systems that are safe (in the sense that they only attack what the're supposed to attack) poses a greater challenge, but does not appear to be impossible. This, unfortunately, assumes that the designers of such a system are both responsible and competent. Furthermore, for the intended target a "safe" weapon system is an oxymoron. I doubt if the world will be destroyed by a commercial product gone awry. It might be destroyed by a weapons system that was badly designed or badly directed.