Engines of Creation: The Coming Era of Nanotechnology

NOTES AND REFERENCES

... Engines of Construction* ... The ideas in this chapter rest on technical arguments presented in my paper "Molecular Engineering: An Approach to the Development of General Capabilities for Molecular Manipulation" (Proceedings of the National Academy of Sciences (USA), Vol. 78, pp. 5275-78, 1981), which presents a case for the feasibility of designing protein molecules and developing general-purpose systems for directing molecular assembly.

... "Protein engineering* ... represents..." See "Protein Engineering," by Kevin Ulmer (Science, Vol. 219, pp. 666-71, Feb. 11, 1983). Dr. Ulmer is now the director of the Center for Advanced Research in Biotechnology.

... One dictionary* ... The American Heritage Dictionary of the English Language, edited by William Morris (Boston: Houghton Mifflin, 1978).

... modern gene synthesis machines* ... See "Gene Machines: The Second Wave," by Jonathan B. Tucker (High Technology, pp. 50-59, March 1984).

... other proteins* serve basic mechanical functions ... See Chapter 27 of Biochemistry, by Albert L. Lelininger (New York: Worth Publishers, 1975). This standard textbook is an excellent source of information on the molecular machinery of life. For a discussion of the bacterial flagellar motor, see "Ion Transport and the Rotation of Bacterial Flagella," by P. Lauger (Nature, Vol. 268, pp. 360-62, July 28,1977).

... self-assembling structures* ... For a description of molecular self-assembly, including that of the T₄ phage and the ribosome, see Chapter 36 of Lehninger's Biochemistry, (referenced above).

... Designing with Protein* ... Nature has demonstrated a wide range of protein machines, but this will not limit us to designing with protein. For examples of fairly complex non-protein structures, see "Supramolecular Chemistry: Receptors, Catalysts, and Carriers," by Jean-Marie Lehn (Science, Vol. 227, pp. 849 - 56, February 22, 1985), which also speaks of designing "components, circuitry, and systems for signal and information treatment at the molecular level."

... any protein they can design* ... Modern techniques can synthesize any desired DNA sequence, which can be used to direct ribosomes to make any desired amino acid sequence. Adding prosthetic group is another matter, however.

... These tasks may sound similar* ... For a comparison of the task of predicting natural protein structures with that of designing predictable structures, see "Molecular Engineering," referenced at the beginning of this section.

... in the journal Nature* ... See "Molecular Technology: Designing proteins and Peptides," by Carl Pabo (Nature, Vol. 301, p.200, Jan. 20, 1983).

... short chains of a few dozen pieces* ... See "Design, Synthesis, and Characterization of a 34-Residue Polypeptide That Interacts with Nucleic Acids," by B. Gutte et al. (Nature, Vol. 281, pp. 650-55, Oct. 25, 1979).

... They have designed from scratch* a protein ... For a reference to this and a general discussion of protein engineering, see Kevin Ulmer's paper (referenced near the beginning of this section).

... changing their behaviors* in predictable ways ... See "A Large Increase in Enzyme-Substrate Affinity by Protein Engineering," by Anthony J. Wilkinson et al. (Nature, Vol. 307, pp. 187-88, Jan. 12, 1984). Genetic engineering techniques have also been used to make an enzyme more stable, with no loss of activity. See "Disulphide Bond Engineered into T₄ Lysozyme: Stabilization of the Protein Toward Thermal Inactivation," by L. Jeanne Perry and Ronald Weutzel of Genentech, Inc. (Science, Vol. 226, pp. 555-57, November 2, 1984).

... according to biologist Garrett Hardin* ... in Nature and Man's Fate (New York: New American Library, 1959), p. 283

... in the journal Science* ... See "Biological Frontiers," by Frederick J. Blattner (Science, Vol. 222, pp. 719-20, Nov. 18, 1983).

... in Applied Biochemistry and Biotechnology* ... See Enzyme Engineering, by William H. Rastetter (Applied Biochemistry and Biotechnology, Vol. 8, pp. 423-36, 1983). This review article describes several successful efforts to change the substrate specificity of enzymes.

... two international workshops* on molecular electronic devices ... For the proceedings of the first, see Molecular Electronic Devices, edited by Forrest L. Carter (New York: Marcel Dekker, 1982). The proceedings of the second appear in Molecular Electronic Devices II, also edited by Forrest L. Carter (New York: Marcel Dekker, 1986). For a summary article, see "Molecular Level Fabrication Techniques and Molecular Electronic Devices," by Forrest L. Carter (Journal of Vacuum Science and Technology, B1(4), pp. 953-68, Oct.-Dec. 1983).

... recommended support for basic research* ... See The Institute (a publication of the IEEE), January 1984, p. 1.

... VLSI Research Inc* ... Reported in Microelectronic Manufacturing and Testing, Sept. 1984, p. 49

... a single chemical bond* ... The strength of a single bond between two carbon atoms is about six nano-newtons, enough to support the weight of about 30,000 trillion carbon atoms. See Strong Solids, by A. Kelly, p. 12 (Oxford: Clarendon Press, 1973).

... diamond fiber* ... Diamond is also over ten times stiffer than aluminum. See Strong Solids (referenced above), Appendix A, Table 2.

... chemists ... coax reacting molecules ... See "Sculpting Horizons in Organic Chemistry," by Barry M. Trost (Science, Vol. 227, pp. 908-16, February 22, 1985), which also mentions organic electrical conductors and the promise of molecular switches for molecular electronics.

... will do all that proteins can do, and more* ... Chemists are already developing catalysts that improve on enzymes; see "Catalysts That Break Nature's Monopoly," by Thomas H. Maugh II (Science, Vol. 221, pp. 351-54, July 22, 1983). For more on non-protein molecular tools, see "Artificial Enzymes," by Ronald Breslow (Science, Vol. 218, pp. 532-37, November 5, 1982).

... assemblers* ... See the first reference in this section. A device reported in 1982, called the scanning tunneling microscope, can position a sharp needle near a surface with an accuracy of a fraction of an atomic diameter. Besides demonstrating the feasibility of such positioning, it may be able to replace molecular machinery in positioning molecular tools. See "Scanning Tunneling Microscopy," by G. Binnig and H. Rohrer (Physica 127B, pp 37-45, 1985).

... almost any reasonable arrangement* ... Assemblers will be able to create otherwise improbable arrangements of reactant molecules (overcoming entropy-of-activation factors), and will be able to direct the action of highly reactive chemical species. This will allow the use in controlled synthesis of reactions that would otherwise proceed only at a negligible rate or with an excessive number and rate of side reactions. Further, assemblers will be able to apply mechanical forces of bond-breaking magnitude to provide activation energy for reactions, and they will be able to employ molecular-scale conductors linked to voltage sources to manipulate electric fields in a direct and novel fashion. While photochemical techniques will not be as useful (because typical photon wavelengths are large on a molecular scale), similar results may sometimes be achieved by transfer of electronic excitation from molecule to molecule in a controlled, localized way.

Though assemblers will be powerful (and could even be directed to expand their own toolkits by assembling new tools), they will not be able to build everything that could exist. For example, a delicate structure might be designed that, like a stone arch, would self-destruct unless all its pieces were already in place. If there were no room in the design for the placement and removal of a scaffolding, then the structure might be impossible to build. Few structures of practical interest seem likely to exhibit such a problem, however. (In fact, the reversibility of the laws governing molecular motion implies that all destructable objects are, in principle, constructable; but if the destruction mechanisms all involve an explosive collapse, then attempts at construction via the reverse mechanism may have a negligible chance of success, owing to considerations involving the uncertainty of the trajectories of the incoming parts and the low entropy of the target state.)

... the DNA-copying machinery* in some cells ... See "Comparative Rates of Spontaneous Mutation," by John W. Drake (Nature, Vol. 221, p. 1132, March 22, 1969). For a general discussion of this machinery, see Chapter 32 of Lehninger's Biochemistry (referenced above).

... repairing and replacing* radiation-damaged parts ... The bacterium Micrococcus radiodurans has vigorous molecular repair mechanisms that enable it to survive the equivalent of more than a million years' worth of normal terrestrial background radiation delivered in a single dose. (See "Inhibition of Repair DNA Synthesis in M. radiodurans after Irradiation with Gamma-rays," by Shigeru Kitayama and Akira Matsuyama, in Agriculture and Biological Chemistry. Vol. 43, pp. 229-305, 1979.) This is about one thousand times the lethal radiation dose for humans, and enough to make Teflon weak and brittle.

... life has never abandoned* ... Living organisms have built cell structures and simple molecular devices from lipids and sugars (and have built shells from silica and lime) but the lack of programmable assembly systems for these materials has kept life from exploiting them to form the main parts of complex molecular machines. RNA, like protein, has a structure directly determined by DNA, and it sometimes serves protein-like functions. See "First True RNA Catalyst Found" (Science, Vol. 223, p. 266, Jan. 20, 1984).

... R. B. Merrifield ... used chemical techniques* ... See Lehninger's Biochemistry. p. 119 (referenced above).

... during the mid-1800s, Charles Babbage* ... See Chapter 2 of Bit by Bit An Illustrated History of Computers, by Stan Augarten (New York: Ticknor & Fields, 1984).

... a billion bytes ... in a box a micron wide* ... If two different side groups on a polyethylene-like polymer are used to represent the ones and zeros of binary code, then the polymer can serve as a data storage tape. If one were to use, say, fluorine and hydrogen as the two side groups, and to allow considerable room for tape reading, writing, and handling mechanisms, then a half cubic micron would store about a billion bytes. Access times can be kept in the microsecond range because the tapes can be made very short. A mechanical random-access memory scheme allows storage of only about 10 million bytes in the same volume, though this can probably be bettered. For a more detailed discussion, see "Molecular Machinery and Molecular Electronic Devices," by K. Eric Drexler, in Molecular Electronic Devices II, edited by Forrest L. Carter (New York: Marcel Dekker, 1986).

... mechanical signals* ... These could be sent by pushing and pulling atom-wide rods of carbyne, a form of carbon in which the atoms are linked in a straight line by alternating single and triple bonds. See "Molecular Machinery and Molecular Electronic Devices," referenced in the above note.

... a scheme proposed* by ... Richard Feynman ... See his article "Quantum Mechanical Computers" (Optics News, Vol. 11, pp. 11-20, Feb. 1985). Feynman concludes that "the laws of physics present no barrier to reducing the size of computers until bits are the size of atoms, and quantum behavior holds dominant sway."

... a disassembler* ... There will be limits to disassemblers as well: For example, one could presumably design a sensitive structure that would fall apart (or explode) when tampered with, preventing controlled disassembly.

... "Think of the design process..." See The Sciences of the Artificial (Second Edition) by Herbert A. Simon (Cambridge, Mass: MIT Press, 1981). This book explores a range of issues related to engineering, problem-solving, economics, and artificial intelligence.

... Both strand and copy ... Because of the rules for nucleotide pairing, the copies actually resemble photographic negatives, and only a copy of a copy matches the original itself.

... Biochemist Sol Spiegelman ... A discussion of his work in this area appears in "The Origin of Genetic Information," by Manfred Eigen et al. (Scientific American, Vol. 244, pp. 88-117, April 1981).

... Oxford zoologist Richard Dawkins ... discusses replicators in The Selfish Gene (New York: Oxford University Press, 1976). This readable book offers an excellent introduction to modern concepts of evolution, focusing on germ-line replicators as the units that undergo variation and selection in evolution.

... As Richard Dawkins points out ... in The Selfish Gene (see above).

... Darwin's detested book ... The Origin of Species, by Charles R. Darwin (London: Charles Murray, 1859).

... the ... ideas of evolution were known before Darwin ... See p. 59 of The Constitution of Liberty, by Friedrich A. Hayek (Chicago: University of Chicago Press, 1960) for a discussion of the earlier work on linguistic, institutional, and even biological evolution, which apparently developed "the conceptual apparatus that Darwin employed." See also p. 23 of Law, Legislation and Liberty, Vol. 1, Rules and Order (Chicago: University of Chicago Press, 1973). Elsewhere, these books discuss the concept of liberty under law and the crucial distinction between a law and a command. These will be important to matters discussed in Chapters 11 and 12.

... As Richard Dawkins puts it ... in The Selfish Gene (see above).

... in The Next Whole Earth Catalog ... Edited by Stewart Brand (Sausalito, California: POINT: distributed by Random House, New York. 1980).

... Peters and Waterman ... See In Search of Excellence. Lessons from America's Best-Run Corporations, by Thomas J. Peters and Robert H. Waterman, Jr. (New York: Warner Books, 1982).

... as Alfred North Whitehead stated ... in Science and the Modern World (New York: Macmillan Company, 1925).

... only study, imagination, and thought ... Converting these into good computer graphics and video will help a lot, though.

... Richard Dawkins calls ... "Meme - is a meme that was launched in the last chapter of The Selfish Gene (see above).

... selfish motives can encourage cooperation ... In The Evolution of Cooperation (New York: Basic Books, 1984) political scientist Robert Axelrod uses a multisided computer game and historical examples to explore the conditions required for cooperation to evolve among selfish entities. Being nice, retaliatory, and forgiving is important to evolving stable cooperation. Chapter 7 of this valuable book discusses. "How to Promote Cooperation."

... In The Extended Phenotype ... by Richard Dawkins (San Francisco: W. H. Freeman, 1982).

... This meme package infected the Xhosa people ... See "The Self-Destruction of the Xosas," Elias Canetti, Crowds and Power (New York: Continuum, 1973), p. 193.

... "The critical attitude..." From Conjectures and Refutations: The Growth of Scientific Knowledge, by Sir Karl Popper (New York: Basic Books, 1962).

... Richard Feynman ... gave a talk ... "There's Plenty of Room at the Bottom," reprinted in Miniaturization, edited by H. D. Gilbert (New York: Reinhold, 1961).

... Bertrand Russell observed ... Quoted by Karl Popper in Objective Knowledge: An Evolutionary Approach (Oxford: Clarendon Press, 1972).

... to seem true or ... to be true ... Ideas that have evolved to seem true (at least to uncritical minds) can in fact be quite false. An excellent work that compares naive human judgment to judgment aided by scientific and statistical techniques is Human Inference, a book by Richard Nisbett and Lee Ross in the Century Psychology Series (Englewood Cliffs, New Jersey: Prentice-Hall, 1980). It shows that, just as we suffer from optical illusions and blind spots, so we suffer from cognitive illusions and blind spots. Other experiments show that untutored people share systematic misunderstandings of such elementary facts as the direction a ball will move when whirled in a circle and then released; learned medieval philosophers (who neglected to test their ideas against reality) evolved whole systems of "science" based on identical misunderstandings. See "Intuitive Physics," by Michael McClosky (Scientific American, Vol. 248, pp. 122-30, Apr. 1983).

... survivors ... huddle so close together ... Strictly speaking, this applies only to survivors that are themselves uniform, general theories. The theory that all rocks will fall straight up next Wednesday has not been disproved (and would have practical consequences), but the special reference to Wednesday makes it nonuniform.

... as Karl Popper points out ... See his Logic of Scientific Discovery, pp. 124 and 419 (New York: Harper & Row, 1965). See also Objective Knowledge, p. 15.

... As ... Ralph E. Gomory says ... in "Technology Development (Science, Vol. 220, pp. 576-80, May 6, 1983).

... their interplay of function and motion ... These were clear only at low speeds; while Leonardo surely had some intuitive sense of dynamics, a description of dynamics adequate to predict the behavior of fast-moving, high-acceleration machine parts did not arrive until Newton.

... designs even in the absence of the tools ... Familiarity with the steady progress in chip fabrication technology has led some companies to design microprocessors whose manufacture required techniques not available at the time of their design.

... computer-aided design of molecular systems ... To do this well will require the simulation of molecular systems. A discussion of one system for molecular simulation appears in Robert Bruccoleri's doctoral thesis, "Macromolecular Mechanics and Protein Folding" (Harvard University, May 1984). For the results of a simulation, see "Dynamics and Conformational Energetics of a Peptide Hormone: Vasopressin," by A. T. Hagler et al. (Science, Vol. 227, pp. 1309-15, Mar. 15, 1985). These references both describe classical simulations, which describe how molecules move in response to forces; such simulations will be adequate for most parts of a typical molecular machine. Other work requires more fundamental (and more costly) quantum mechanical simulations, which describe the distribution of electrons in molecules. These calculations will be required to describe the forming and breaking of bonds by assembler tools. For a discussion of molecular simulations that include quantum mechanical calculations of bond formation, see "Theoretical Chemistry Comes Alive: Full Partner with Experiment," by William H. Goddard III (Science, Vol. 227, pp. 912-23, Feb. 22, 1985). See also Lecture Notes in Chemistry, 19, Computational Aspects for Large Chemical Systems, by Enrico Clementi (New York: Springer-Verlag, 1980). Finally, for a discussion of present design tools, see "Designing Molecules by Computer," by Jonathan B. Tucker (High Technology, pp. 52-59, Jan. 1984). Parallel processing computers will greatly aid computational chemistry and computer-aided design.

... design ahead ... Early design-ahead efforts seem likely to aim at defining a workable assembler system; it need not be ideal, so long as it has fairly broad capabilities. Once the capabilities of this standard assembler design are fairly well specified - even before the design is complete - it will become possible (11) to begin developing a library of nanomachine designs suited to construction by this standard assembler (or by assemblers that the standard assembler could construct), and (2) to prepare a corresponding library of procedures for the assembly of these designs. Then when the first crude assembler is developed, it can be used (perhaps through an intermediate stage of tool building) to build a standard assembler. This in turn could be used to build anything in the design library.

Early assemblers will greatly extend our ability to make things. With even limited design ahead, the advent of assemblers will result almost immediately in substantial jumps in the quality of hardware. Since assemblers will be built by assemblers, some form of self-replicating system will be an immediate natural consequence of design ahead and the assembler breakthrough. Accordingly, the advent of assemblers may make possible not only a jump in hardware quality, but the almost immediate mass production of that hardware in unprecedented quantities (see Chapter 4). For better or for worse, this will make possible an unusually abrupt change in technology, in economics, and in world affairs.

... If every tool, when ordered ... From Scientific Quotations: The Harvest of a Quiet Eye, selected by A. L. Mackay, edited by M. Ebison (New York: Crane, Russak, 1977).

... a NASA scientist ... Former NASA administrator Robert Frosch said much the same thing at the IEEE Centennial Technical Convocation (see The Institute, p. 6, Dec. 1984).

... replicators, such as viruses, bacteria ... In an evolutionary sense, an animal's genes are replicators, but the animal itself is not; only changes to genes, not changes to an animal's body, are replicated in later generations. This distinction between genetic replicators and the systems they shape is essential to understanding evolution, but use of the term "replicator" to refer to the whole system is more convenient when discussing replicating systems as productive assets.

... Fujitsu Fanuc ... See "Production: A Dynamic Challenge". by M. E. Merchant (IEEE Spectrum, pp. 36-39, May 1983). This issue of the IEEE Spectrum contains an extensive discussion of computer-based automation.

... will instead resemble factories ... Cell-style organization nonetheless has advantages. For example, despite various active-transport mechanisms, cells typically transport molecular components by diffusion rather than by conveyors. This effectively connects every machine to every other (in the same membrane compartment) in a robust fashion; conveyors, in contrast, can break down, requiring repair or replacement. But it may be that properly implemented conveyor-based transportation has strong advantages and yet did not evolve. Conveyor-based systems would be harder to evolve because they require a new molecular machine to have a suitable location, orientation, and interface to the conveyor before it can function. If it failed to meet any of these requirements it would be useless, and selective pressures would generally eliminate it before a useful variant had a chance to appear. For a new molecular machine to function in a diffusion-based system, though, it need only be present. If it does something useful, selection will favor it immediately.

... A fast enzyme ... See Albert L. Lehninger's Biochemistry, p. 208 (in Chapter 1 references). Further, each molecule of the enzyme catalase can break down 40 million hydrogen peroxide molecules per second; see Enzyme Structure and Mechanism, by Alan Fersht, p. 132 (San Francisco: W. H. Freeman & Co., 1977). In typical enzymatic reactions, molecules must wander into position with respect to the enzyme's "tools," then wait for random thermal vibrations to cause a reaction, and then wander out of the way again. These steps take up almost all of the enzyme's time; the time required to form or break a bond is vastly smaller. Because the electrons of a bond are over a thousand times lighter and more mobile than the nuclei that define atomic positions, the slower motion of whole atoms sets the pace. The speed of typical atoms under thermal agitation at ordinary temperatures is over 100 meters per second, and the distance an atom must move to form or break a bond is typically about a ten billionth of a meter, so the time required is about one trillionth of a second. See Chapter 12 of Molecular Thermodynamics, by John H. Knox (New York: Wiley-Interscience, 1971).

... about fifty million times more rapidly ... This scaling relationship may be verified by observing (1) that mechanical disturbances travel at the speed of sound (arriving in half the time if they travel half as far) and (2) that, for a constant stress in the arm material, reducing the arm length (and hence the mass per unit cross section) by one half doubles the acceleration at the tip while halving the distance the tip travels, which allows the tip to move back and forth in half the time (since the time required for a motion is the square root of a quantity proportional to the distance traveled divided by the acceleration).

... a copy ... of the tape ... Depending on the cleverness (or lack of cleverness) of the coding scheme, the tape might have more mass than the rest of the system put together. But since tape duplication is a simple, specialized function, it need not be performed by the assembler itself.

... only a minute fraction misplaced ... due to rare fluctuations in thermal noise, and to radiation damage during the assembly process. High-reliability assemblers will include a quality-control system to identify unwanted variations in structure. This system could consist of a sensor arm used to probe the surface of the workpiece to identify the unplanned bumps or hollows that would mark a recent mistake. Omissions (typically shown by hollows) could be corrected by adding the omitted atoms. Misplaced groups (typically shown by bumps) could be corrected by fitting the assembler arm with tools to remove the misplaced atoms. Alternatively, a small workpiece could simply be completed and tested. Mistakes could then be discarded before they had a chance to be incorporated into a larger, more valuable system. These quality-control steps will slow the assembly process somewhat.

... working, like muscle ... See the notes for Chapter 6.

... The world stands ... Quoted from Business Week, March 8, 1982.

... As Daniel Dennett ... points out ... See "Why the Law of Effect Will Not Go Away," in Daniel C. Dennett, Brainstorms: Philosophical Essays on Mind and Psychology (Cambridge, Mass: MIT Press, 1981). This book explores a range of interesting issues, including evolution and artificial intelligence.

... Marvin Minsky ... views the mind ... See his book The Society of Mind (New York: Simon & Schuster, to be published in 1986). I have had an opportunity to review much of this work in manuscript form; it offers valuable insights about thought, language, memory, developmental psychology, and consciousness - and about how these relate to one another and to artificial intelligence.

... Any system or device ... American Heritage Dictionary, edited by William Morris (Boston: Houghton Mifflin Company, 1978).

... Babbage had built ... See Bit by Bit, in Chapter 1 references.

... the Handbook of Artificial Intelligence ... edited by Avron Barr and Edward A. Feigenbaum (Los Altos, Calif: W. Kaufmann, 1982).

... As Douglas Hofstadter urges ... See "The Turing Test: A Coffeehouse Conversation" in The Mind's I, composed and arranged by Douglas R. Hofstadter and Daniel C. Dennett (New York: Basic Books, 1981).

... We can surely make machines ... As software engineer Mark Miller puts it, "Why should people be able to make intelligence in the bedroom, but not in the laboratory?"

... "I believe that by the end of the century..." From "Computing Machinery and Intelligence," by Alan M. Turing (Mind, Vol. 59, No. 236, 1950); excerpted in The Mind's I (referenced above).

... a system could show both kinds ... Social and technical capabilities might stem from a common basis, or from linked subsystems; the boundaries can easily blur. Still, specific AI systems could be clearly deserving of one name or the other. Efforts to make technical AI systems as useful as possible will inevitably involve efforts to make them understand human speech and desires.

... social AI ... Advanced social AI systems present obvious dangers. A system able to pass the Turing test would have to be able to plan and set goals as a human would - that is, it would have to be able to plot and scheme, perhaps to persuade people to give it yet more information and ability. Intelligent people have done great harm through words alone, and a Turing-test passer would of necessity be designed to understand and deceive people (and it would not necessarily have to be imbued with rigid ethical standards, though it might be). Chapter 11 discusses the problem of how to live with advanced AI systems, and how to build AI systems worthy of trust.

... "May not machines carry out ..." See the earlier reference to Turing's paper, "Computing Machinery and Intelligence."

... Developed by Professor Douglas Lenat ... and described by him in a series of articles on "The Nature of Heuristics" (Artificial Intelligence, Vol. 19, pp. 189-249, 1982; Vol. 21, pp. 31-59 and 61-98, 1983; Vol. 23, pp. 269-93, 1984).

... Traveller TCS ... See preceding reference, Vol. 21, pp. 73-83.

... EURISKO has shortcomings ... Lenat considers the most serious to be EURISKO's limited ability to evolve new representations for new information.

... In October of 1981 ... In the fall of 1984 ... See "The 'Star Wars' Defense Won't Compute," by Jonathan Jacky (The Atlantic, Vol. 255, pp. 18-30, June 1985).

... in the IEEE Spectrum ... See "Designing the Next Generation," by Paul Wallich (IEEE Spectrum, pp. 73-77, November 1983).

... fresh insights into human psychology ... Hubert Dreyfus, in his well-known book What Computers Can't Do: The Limits of Artificial Intelligence (New York: Harper & Row, 1979), presents a loosely reasoned philosophical argument that digital computers can never be programmed to perform the full range of human intellectual activities. Even if one were to accept his arguments, this would not affect the main conclusions I draw regarding the future of AI: the automation of engineering design is not subject to his arguments because it does not require what he considers genuine intelligence; duplicating the human mind by means of neural simulation avoids (and undermines) his philosophical arguments by dealing with mental processes at a level where those arguments do not apply.

... virus-sized molecular machines ... See Chapter 7.

... build analogous devices ... These devices might be electromechanical, and will probably be controlled by microprocessors; they will not be as simple as transistors. Fast neural simulation of the sort I describe will be possible even if each simulated synapse must have its properties controlled by a device as complex as a microprocessor.

... experimental electronic switches ... which switch in slightly over 12 picoseconds are described in "The HEMT: A Superfast Transistor," by Hadis Morkoc and Paul M. Solomon (IEEE Spectrum, pp. 28-35, Feb. 1984).

... Professor Robert Jastrow ... in his book The Enchanted Loom: The Mind in the Universe (New York: Simon & Schuster, 1981).

... will fit in less than a cubic centimeter ... The brain consists chiefly of wirelike structures (the axons and dendrites) and switchlike structures (the synapses). This is an oversimplification, however, because at least some wirelike structures can have their resistance modulated on a short time scale (as discussed in "A Theoretical Analysis of Electrical Properties of Spines," by C. Koch and T. Poggio, MIT AI Lab Memo No. 713, April 1983). Further, synapses behave less like switches than like modifiable switching circuits; they can be modulated on a short time scale and entirely rebuilt on a longer time scale (see "Cell Biology of Synaptic Plasticity," by Carl W. Cotman and Manuel Nieto-Sampedro, Science, Vol. 225, pp. 1287-94, Sept. 21, 1984).

The brain can apparently be modeled by a system of nanoelectronic components modulated and rebuilt by nanomachinery directed by mechanical nanocomputers. Assume that one nanocomputer is allotted to regulate each of the quadrillion or so "synapses" in the model brain, and that each also regulates corresponding sections of "axon" and "dendrite." Since the volume of each nanocomputer (if equivalent to a modern microprocessor) will be about 0.0003 cubic micron (See "Molecular Machinery and Molecular Electronic Devices," referenced in Chapter 1), these devices will occupy a total of about 0.3 cubic centimeter. Dividing another 0.3 cubic centimeter equally between fast random-access memory and fairly fast tape memory would give each processor a total of about 3.7K bytes of RAM and 275K bytes of tape. (This sets no limit to program complexity, since several processors could share a larger program memory.) This amount of information seems far more than enough to provide an adequate model of the functional state of a synapse. Molecular machines (able to modulate nanoelectronic components) and assembler systems (able to rebuild them) would occupy comparatively little room. Interchange of information among the computers using carbyne rods could provide for the simulation of slower, chemical signaling in the brain.

Of the nanoelectronic components, wires will occupy the most volume. Typical dendrites are over a micron in diameter, and serve primarily as conductors. The diameter of thin wires could be less than a hundredth of a micron, determined by the thickness of the insulation required to limit electron tunneling (about three nanometers at most). Their conductivity can easily exceed that of a dendrite. Since the volume of the entire brain is about equal to that of a ten-centimeter box, wires a hundred times thinner (one ten-thousandth the cross section) will occupy at most 0.01 of a cubic centimeter (allowing for their being shorter as well). Electromechanical switches modulated by molecular machinery can apparently be scaled down by about the same factor, compared to synapses.

Thus, nanoelectronic circuits that simulate the electrochemical behavior of the brain can apparently fit in a bit more than 0.01 cubic centimeter. A generous allowance of volume for nanocomputers to simulate the slower functions of the brain totals 0.6 cubic centimeter, as calculated above. A cubic centimeter thus seems ample.

... dissipating a millionfold more heat ... This may be a pessimistic assumption, however. For example, consider axons and dendrites as electrical systems transmitting signals. All else being equal, millionfold faster operation requires millionfold greater currents to reach a given voltage threshold. Resistive heating varies as the current squared, divided by the conductivity. But copper has about forty million times the conductivity of neurons (see "A Theoretical Analysis of Electrical Properties of Spines," referenced above), reducing resistive heating to less than the level assumed (even in a device like that described in the text, which is somewhat more compact than a brain). For another example, consider the energy dissipated in the triggering of a synapse: devices requiring less energy per triggering would result in a power dissipation less than that assumed in the text. There seems no reason to believe that neurons are near the limits of energy efficiency in information processing; for a discussion of where those limits lie, see "Thermodynamics of Computation - A Review," by C. H. Bennett (International Journal of Theoretical Physics, Vol. 21, pp. 219-53, 1982). This reference states that neurons dissipate an energy of over one billion electron-volts per discharge. Calculations indicate that electrostatically activated mechanical relays can switch on or off in less than a nanosecond, while operating at less than 0.1 volt (like neurons) and consuming less than a hundred electron-volts per operation. (There is no reason to believe that mechanical relays will make the best switches, but their performance is easy to calculate.) Interconnect capacitances can also be far lower than those in the brain.

... pipe ... bolted to its top ... This is a somewhat silly image, since assemblers can make connectors that work better than bolts and cooling systems that work better than flowing water. But to attempt to discuss systems based entirely on advanced assembler-built hardware would at best drag in details of secondary importance, and would at worst sound like a bogus prediction of what will be built, rather than a sound projection of what could be built. Accordingly, I will often describe assembler-built systems in contexts that nanotechnology would in fact render obsolete.

... As John McCarthy ... points out ... See Machines Who Think, by Pamela McCorduck, p. 344 (San Francisco: W. H. Freeman & Company, 1979). This book is a readable and entertaining overview of artificial intelligence from the perspective of the people and history of the field.

... As Marvin Minsky has said ... in U.S. News & World Report, P. 65, November 2, 1981.

... engineers know of alternatives ... For orbit-to-orbit transportation, one attractive alternative is using rockets burning fuel produced in space from space resources.

... result will be the "lightsail" ... For further discussion, see "Sailing on Sunlight May Give Space Travel a Second Wind" (Smithsonian, pp. 52-61, Feb. 1982), "High Performance Solar Sails and Related Reflecting Devices," AIAA Paper 79-1418, in Space Manufacturing III, edited by Jerry Grey and Christine Krop (New York: American Institute of Aeronautics and Astronautics, 1979), and MIT Space Systems Laboratory Report 5-79, by K. Eric Drexler. The World Space Foundation (P.0. Box Y, South Pasadena, Calif. 91030) is a nonprofit, membership-oriented organization that is building an experimental solar sail and supporting the search for accessible asteroids.

... The asteroids ... are flying mountains of resources ... For a discussion of asteroidal resources, see "Asteroid Surface Materials: Mineralogical Characterizations from Reflectance Spectra," by Michael J. Gaffey and Thomas B. McCord (Space Science Reviews, No. 21, p. 555, 1978) and "Finding "Paydirt" on the Moon and Asteroids," by Robert L. Staehle (Astronautics and Aeronautics, pp. 44-49, November 1983).

... as permanent as a hydroelectric dam ... Erosion by micrometeoroids is a minor problem, and damage by large meteoroids is extremely rare.

... Gerard O'Neill ... See his book The High Frontier: Human Colonies in Space (New York: William Morrow, 1976). The Space Studies Institute (285 Rosedale Road, P.0. Box 82, Princeton, N.J. 08540) is a nonprofit, membership-oriented organization aimed at advancing the economic development and settlement of space, working chiefly through research projects. The L5 Society (1060 East Elm, Tucson, Ariz. 85719) is a nonprofit, membership-oriented organization aimed at advancing the economic development and settlement of space, working chiefly through public education and political action.

... Using this energy to power assemblers ... How much electric power can a given mass of solar collector supply? Since electric energy is readily convertible to chemical energy, this will indicate how rapidly a solar collector of a given mass can supply enough energy to construct an equal mass of something else. Experimental amorphous-silicon solar cells convert sunlight to electricity with about 10 percent efficiency in an active layer about a micron thick, yielding about 60 kilowatts of power per kilogram of active mass. Assembler-built solar cells will apparently be able to do much better, and need not have heavy substrates or heavy, low-voltage electrical connections. Sixty kilowatts of power supplies enough energy in a few minutes to break and rearrange all the chemical bonds in a kilogram of typical material. Thus a spacecraft with a small fraction of its mass invested in solar collectors will be able to entirely rework its own structure in an hour or so. More important, though, this calculation indicates that solar-powered replicators will be able to gather enough power to support several doublings per hour.

... The middle layer of the suit material ... To have the specified strength, only about one percent of the material's cross-sectional area must consist of diamond fibers (hollow telescoping rods, in one implementation) that run in a load-bearing direction. There exists a regular, three-dimensional woven pattern (with fibers running in seven different directions to support all possible types of load, including shear) in which packed cylindrical fibers fill about 45 percent of the total volume. In any given direction, only some of the fibers can bear a substantial load, and using hollow, telescoping fibers (and then extending them by a factor of two in length) makes the weave less dense. These factors consume most of the margin between 45 percent and one percent, leaving the material only as strong as a typical steel.

For the suit to change shape while holding a constant internal volume at a constant pressure, and do so efficiently, the mechanical energy absorbed by stretching material in one place must be recovered and used to do mechanical work in contracting material in another place - say, on the other side of a bending elbow joint. One way to accomplish this is by means of electrostatic motors, reversible as generators, linked to a common electric power system. Scaling laws favor electrostatic over electromagnetic motors at small sizes.

A design exercise (with applications not limited to hypothetical space suits) resulted in a device about 50 nanometers in diameter that works on the principle of a "pelletron"-style Van de Graaff generator, using electron tunneling across small gaps to charge pellets and using a rotor in place of a pellet chain. (The device also resembles a bucket-style water wheel.) DC operation would be at 10 volts, and the efficiency of power conversion (both to and from mechanical power) seems likely to prove excellent, limited chiefly by frictional losses. The power-conversion density (for a rotor rim speed of one meter per second and pellets charged by a single electron) is about three trillion watts per cubic meter. This seems more than adequate.

As for frictional losses in general, rotary bearings with strengths of over 6 nano-newtons can be made from carbon bonds - see Strong Solids, by A. Kelly (Oxford: Clarendon Press, 1973) - and bearings using a pair of triple-bonded carbon atoms should allow almost perfectly unhindered rotation. Roller bearings based on atomically perfect hollow cylinders with bumps rolling gear-fashion on atomically perfect races have at least two significant energy-dissipation modes, one resulting from phonon (sound) radiation through the slight bumpiness of the rolling motion, the other resulting from scattering of existing phonons by the moving contact point. Estimates of both forms of friction (for rollers at least a few nanometers in diameter moving at modest speeds) suggest that they will dissipate very little power, by conventional standards.

Electrostatic motors and roller bearings can be combined to make telescoping jackscrews on a submicron scale. These can in turn be used as fibers in a material able to behave in the manner described in the text.

... now transmits only a tenth of the force ... An exception to this is a force that causes overall acceleration: for example, equilibrium demands that the forces on the soles of the feet of a person standing in an accelerating rocket provide support, and the suit must transmit them without amplification or diminution. Handling this smoothly may be left as an exercise for future control-system designers and nanocomputer programmers.

... the suit will keep you comfortable ... Disassemblers, assemblers, power, and cooling - together, these suffice to recycle all the materials a person needs and to maintain a comfortable environment. Power and cooling are crucial.

As for power, a typical person consumes less than 100 watts, on the average; the solar power falling on a surface the size of a sheet of typing paper (at Earth's distance from the Sun) is almost as great. If the suit is covered with a film that acts as a high-efficiency solar cell, the sunlight striking it should provide enough power. Where this is inadequate, a solar-cell parasol could be used to gather more power.

As for cooling, all power absorbed must eventually be disposed of as waste heat - in a vacuum, by means of thermal radiation. At body temperature, a surface can radiate over 500 watts per square meter. With efficient solar cells and suitable design (and keeping in mind the possibility of cooling fins and refrigeration cycles), cooling should be no problem in a wide range of environments. The suit's material can, of course, contain channels for the flow of coolant to keep the wearer's skin at a preferred temperature.

... a range of devices greater than ... yet built ... A pinhead holds about a cubic millimeter of material (depending on the pin, of course). This is enough room to encode an amount of text greater than that in a trillion books (large libraries hold only millions). Even allowing for a picture's being worth a thousand words, this is presumably enough room to store plans for a wide enough range of devices.

... in a morning ... The engineering AI systems described in Chapter 5, being a million times faster than human engineers, could perform several centuries' worth of design work in a morning.

... replicating assemblers that work in space ... Assemblers in a vacuum can provide any desired environment at a chemical reaction site by positioning the proper set of molecular tools. With proper design and active repair-and-replacement mechanisms, exposure to the natural radiation of space will be no problem.

... move it off Earth entirely ... But what about polluting space? Debris in Earth orbit is a significant hazard and needs to be controlled, but many environmental problems on Earth cannot occur in space: space lacks air to pollute, groundwater to contaminate, or a biosphere to damage. Space is already flooded with natural radiation. As life moves into space, it will be protected from the raw space environment. Further, space is big the volume of the inner solar system alone is many trillions of times that of Earth's air and oceans. If technology on Earth has been like a bull in a china shop, then technology in space will be like a bull in an open field.

... As Konstantin Tsiolkovsky wrote ... Quoted in The High Frontier: Human Colonies in Space, by Gerard K. O'Neill (New York: William Morrow, 1976).

... drive a beam far beyond our solar system ... This concept was first presented by Robert L. Forward in 1962.

... Freeman Dyson ... suggests ... He discussed this in a talk at an informal session of the May 15, 1980, "Discussion Meeting on Gossamer Spacecraft," held at the jet Propulsion Laboratory in Pasadena, California.

... Robert Forward ... suggests ... See his article "Roundtrip Interstellar Travel Using Laser-Pushed Lightsails," (Journal of Spacecraft and Rockets, Vol. 21, pp. 187-95, Jan.-Feb. 1984). Forward notes the problem of making a beam-reversal sail light enough, yet of sufficient optical quality (diffraction limited) to do its job. An actively controlled structure based on thin metal films positioned by nanometer-scale actuators and computers seems a workable approach to solving this problem.

But nanotechnology will allow a different approach to accelerating lightsails and stopping their cargo. Replicating assemblers will make it easy to build large lasers, lenses, and sails. Sails can be made of a crystalline dielectric, such as aluminum oxide, having extremely high strength and low optical absorptivity. Such sails could endure intense laser light, scattering it and accelerating at many gees, approaching the speed of light in a fraction of a year. This will allow sails to reach their destinations in near-minimal time. (For a discussion of the multi-gee acceleration of dielectric objects, see "Applications of Laser Radiation Pressure," by A. Ashkin [Science, Vol. 210, pp. 1081-88, Dec. 5, 1980].)

In flight, computer-driven assembler systems aboard the sail (powered by yet more laser light from the point of departure) could rebuild the sail into a long, thin traveling-wave accelerator. This can then be used to electrically accelerate a hollow shell of strong material several microns in radius and containing about a cubic micron of cargo; such a shell can be given a high positive charge-to-mass ratio. Calculations indicate that an accelerator 1,000 kilometers long (there's room enough, in space) will be more than adequate to accelerate the shell and cargo to over 90 percent of the speed of light. A mass of one gram per meter for the accelerator (yielding a one-ton system) seems more than adequate. As the accelerator plunges through the target star system, it fires backward at a speed chosen to leave the cargo almost at rest. (For a discussion of the electrostatic acceleration of small particles, see "Impact Fusion and the Field Emission Projectile," by E. R. Harrison [Nature, Vol. 291, pp. 472-73, June 11, 1981].)

The residual velocity of the projectile can be directed to make it strike the atmosphere of a Mars- or Venus-like planet (selected beforehand by means of a large space-based telescope). A thin shell of the sort described will radiate atmospheric entry heat rapidly enough to remain cool. The cargo, consisting of an assembler and nanocomputer system, can then use the light of the local sun and local carbon, hydrogen, nitrogen, and oxygen (likely to be found in any planetary atmosphere) to replicate and to build larger structures.

An early project would be construction of a receiver for further instructions from home, including plans for complex devices. These can include rockets able to get off the planet (used as a target chiefly for its atmospheric cushion) to reach a better location for construction work. The resulting system of replicating assemblers could build virtually anything, including intelligent systems for exploration. To solve the lightsail stop ping problem for the massive passenger vehicles that might follow, the system could build an array of braking lasers as large as the launching lasers back home. Their construction could be completed in a matter of weeks following delivery of the cubic-micron "seed." This system illustrates one way to spread human civilization to the stars at only slightly less than the speed of light.

... space near Earth holds ... Two days' travel at one gee acceleration can carry a person from Earth to any point on a disk having over 20 million times the area of Earth - and this calculation allows for a hole in the middle of the disk with a radius a hundred times the Earth-Moon distance. Even so, the outer edge of the disk reaches only one twentieth of the way to the Sun.

... enough energy in ten minutes ... Assuming conversion of solar to kinetic energy with roughly 10 percent efficiency, which should be achievable in any of several ways.

... Dr. Seymour Cohen ... argues ... See his article "Comparative Biochemistry and Drug Design for Infectious Disease" (Science, Vol. 205, pp. 964-71, Sept. 7, 1979).

... Researchers at Upjohn Company ... See "A Conformationally Constrained Vasopressin Analog with Antidiuretic Antagonistic Activity," by Gerald Skala et al. (Science, Vol. 226, pp. 443-45, Oct. 26, 1984).

... a dictionary definition of holism ... The American Heritage Dictionary of the English Language, edited by William Morris (Boston: Houghton Mifflin Company, 1978).

... aided by sophisticated technical AI systems ... These will be used both to help design molecular instruments and to direct their use. Using devices able to go to specified locations, grab molecules, and analyze them, the study of cell structures will become fairly easy to automate.

... separated molecules can be put back together ... Repair machines could use devices resembling the robots now used in industrial assembly work. But reassembling cellular structures will not require machines so precise (that is, so precise for their size). Many structures in cells will self-assemble if their components are merely confined together with freedom to bump around; they need not be manipulated in a complex and precise fashion. Cells already contain all the tools needed to assemble cell structures, and none is as complex as an industrial robot.

... the T₄ phage ... self-assembles ... See pp. 1022-23 of Biochemistry, by Albert L. Lehninger (New York: Worth Publishers, 1975).

... lipofuscin ... fills over ten percent ... Lipofuscin contents vary with cell type, but some brain cells (in old animals) contain an average of about 17 percent; typical lipofuscin granules are one to three microns across. See "Lipofuscin Pigment Accumulation as a Function of Age and Distribution in Rodent Brain," by William Reichel et al. (Journal of Gerontology, Vol. 23, pp. 71-81, 1968). See also "Lipoprotein Pigments - Their Relationship to Aging in the Human Nervous System,"- by D. M. A. Mann and P. O. Yates (Brain, Vol. 97, pp. 481-88, 1974).

... about one in a million million million ... The implied relationship is not exact but shows the right trend: for example, the second number should be 2.33 uncorrected errors in a million million, and the third should be 4.44 in a million million million (according to some fairly complex calculations based on a slightly more complex correction algorithm).

... compare DNA molecules ... make corrected copies ... Immune cells that produce different antibodies have different genes, edited during development. Repairing these genes will require special rules (but the demonstrated feasibility of growing an immune system shows that the right patterns of information can be generated).

... will identify molecules in a similar way ... Note that any molecule damaged enough to have an abnormal effect on the molecular machinery of the cell will by the same token be damaged enough to have a distinctive effect on molecular sensors.

... a complex and capable repair system ... For a monograph that discusses this topic in more detail, including calculations of volumes, speeds, powers, and computational loads, see "Cell Repair Systems," by K. Eric Drexler (available through The Foresight Institute, Palo Alto, Calif.).

... will be in communication ... For example, by means of hollow fibers a nanometer or two in diameter, each carrying a carbyne signaling rod of the sort used inside mechanical nanocomputers. Signal repeaters can be used where needed.

... to map damaged cellular structures ... This need not require solving any very difficult pattern recognition problems, save in cases where the cell structure is grossly disrupted. Each cell structure contains standard types of molecules in a pattern that varies within stereotyped limits, and a simple algorithm can identify even substantially-damaged proteins. Identification of the standard molecules in a structure determines its type; mapping it then becomes a matter of filling in known sorts of details.

... in a single calendar year ... Molecular experiments can be done about a millionfold faster than macroscopic experiments, since an assembler arm can perform actions at a million times the rate of a human arm (see Chapter 4). Thus, molecular machines and fast AI systems are well matched in speed.

... extended ... lifespan ... by 25 to 45 percent ... using 2-MEA, BHT, and ethoxyquine; results depended on the strain of mouse, the diet, and the chemical employed. See "Free Radical Theory of Aging," by D. Harman (Triangle, Vol. 12, No. 4, pp. 153-58, 1973).

... Eastman Kodak ... according to the Press-Telegram, Long Beach, Calif., April 26, 1985.

... rely on new science ... Cell repair will also rely on new science, but in a different way. As discussed in Chapter 3, it makes sense to predict what we will learn about, but not what we will learn. To extend life by means of cell repair machines will require that we learn about cell structures before we repair them, but what we learn will not affect the feasibility of those repairs. To extend life by conventional means, in contrast, will depend on how well the molecular machinery of the body can repair itself when properly treated. We will learn more about this, but what we learn could prove discouraging.

... durability has costs ... See "Evolution of Aging," a review article by T. B. L. Kirkwood (Nature, Vol. 270, pp. 301-4, 1977).

... As Sir Peter Medawar points out ... in The Uniqueness of the Individual (London: Methuen, 1957). See also the discussion in The Selfish Gene, pp. 42-45 (in Chapter 2 references).

... Experiments by Dr. Leonard Hayflick ... See the reference above, which includes an alternative (but broadly similar) explanation for Hayflick's result.

... A mechanism of this sort ... For a reference to a statement of this theory (by D. Dykhuizen in 1974) together with a criticism and a rebuttal, see the letters by Robin Holliday, and by John Cairns and Jonathan Logan, under "Cancer and Cell Senescence" (Nature, Vol. 306, p. 742, December 29, 1983).

... could harm older animals by stopping ... division ... These animals could still have a high cancer rate because of a high incidence of broken clocks.

... cleaning machines to remove these poisons ... One system's meat really is another's poison; cars "eat" a toxic petroleum product. Even among organisms, some bacteria thrive on a combination of methanol (wood alcohol) and carbon monoxide (see "Single-Carbon Chemistry of Acetogenic and Methanogenic Bacteria," by J. G. Zeikus et al., Science, Vol. 227, pp. 1167-71, March 8, 1985), while others have been bred that can live on either trichlorophenol or the herbicide 2,4,5-T. They can even defluorinate pentafluorophenol. (See "Microbial Degradation of Halogenated Compounds," D. Chousal et al., Science, Vol. 228, pp. 135-42, April 12, 1985.)

... cheap enough to eliminate the need for fossil fuels ... through the use of fuels made by means of solar energy.

... able to extract carbon dioxide from the air ... In terms of sheer tonnage, carbon dioxide is perhaps our biggest pollution problem. Yet, surprisingly, a simple calculation shows that the sunlight striking Earth in a day contains enough energy to split all the carbon dioxide in the atmosphere into carbon and oxygen (efficiency considerations aside). Even allowing for various practical and aesthetic limitations, we will have ample energy to complete this greatest of cleanups in the span of a single decade.

... Alan Wilson ... and his co-workers ... See "Gene Samples from an Extinct Animal Cloned," by J. A. Miller (Science News, Vol. 125, p. 356, June 9, 1984).

... O my friend ... The Iliad, by Homer (about the eighth century B.C.), as quoted by Eric Hoffer in The True Believer (New York: Harper & Brothers, 1951). (Sarpedon is indeed killed in the battle.)

... Gilgamesh, King of Uruk ... From The Epic of Gilgamesh, translated by N. K. Sandars (Middlesex: Penguin Books, 1972).

... To Jacques Dubourg ... In Mr. Franklin, A Selection from His Personal Letters, by L. W. Labaree and W. J. Bell, Jr. (New Haven: Yale University Press, 1956), pp. 27-29.

... a new heart, fresh kidneys, or younger skin ... With organs and tissues grown from the recipient's own cells, there will be no problem of rejection.

... The changes ... are far from subtle ... Experiments show that variations in experience rapidly produce visible variations in the shape of dendritic spines (small synapse-bearing protrusions on dendrites). See "A Theoretical Analysis of Electrical Properties of Spines," by C. Koch and T. Poggio (MIT AI Lab Memo No. 713, April 1983).

In "Cell Biology of Synaptic Plasticity" (Science, Vol. 225, pp. 128794, Sept. 21, 1984), Carl W. Cotman and Manuel Nieto-Sampedro write that "The nervous system is specialized to mediate the adaptive response of the organism... To this end the nervous system is uniquely modifiable, or plastic. Neuronal plasticity is largely the capability of synapses to modify their function, to be replaced, and to increase or decrease in number when required." Further, "because the neocortex is believed to be one of the sites of learning and memory, most of the studies of the synaptic effect of natural stimuli have concentrated on this area." Increases in dendritic branching in the neocortex "are caused by age (experience) in both rodents and humans. Smaller but reproducible increases are observed after learning of particular tasks ..." These changes in cell structure can occur "within hours."

For a discussion of short- and long-term memory, and of how the first may be converted into the second, see "The Biochemistry of Memory: A New and Specific Hypothesis," by Gaty Lynch and Michel Baudty (Science, Vol. 224, pp. 1057-63, June 8, 1984).

At present, every viable theory of long-term memory involves changes in the structure and protein content of neurons. There is a persistent popular idea that memory might somehow be stored (exclusively?) "in RNA molecules," a rumor seemingly fostered by an analogy with DNA, the "memory" responsible for heredity. This idea stems from old experiments suggesting that learned behaviors could be transferred to uneducated flatworms by injecting them with RNA extracted from educated worms. Unfortunately for this theory, the same results were obtained using RNA from entirely uneducated yeast cells. See Biology Today, by David Kirk, p. 616 (New York: Random House, 1975).

Another persistent popular idea is that memory might be stored in the form of reverberating patterns of electrical activity, a rumor seemingly fostered by an analogy with the dynamic random-access memories of modern computers. This analogy, however, is inappropriate for several reasons: (1) Computer memories, unlike brains, are designed to be erased and reused repeatedly. (2) The patterns in a computer's "long-term memory" - its magnetic disk, for example - are in fact more durable than dynamic RAM. (3) Silicon chips are designed for structural stability, while the brain is designed for dynamic structural change. In light of the modern evidence for long-term memory storage in long-lasting brain structures, it is not surprising that "total cessation of the electrical activity of the brain does not generally delete memories, although it may selectively affect the most recently stored ones" (A. J. Dunn). The electrical reverberation theory was proposed by R. Lorente de No (Journal of Neurophysiology, Vol. 1, p. 207) in 1938. Modern evidence fails to support such theories of ephemeral memory.

... "striking morphological changes" ... For technical reasons, this study was performed in mollusks, but neurobiology has proved surprisingly uniform. See "Molecular Biology of Learning: Modulation of Transmitter Release," by Eric R. Kandel and James H. Schwartz (Science, Vol. 2 18, pp. 433-43, Oct. 29, 1982), which reports work by C. Baily and M. Chen.

... until after vital functions have ceased ... The time between expiration and dissolution defines the window for successful biostasis, but this time is uncertain. As medical experience shows, it is possible to destroy the brain (causing irreversible dissolution of mind and memory) even while a patient breathes. In contrast, patients have been successfully revived after a significant period of so-called "clinical death." With cell repair machines, the basic requirement is that brain cells remain structurally intact; so long as they are alive, they are presumably intact, so viability provides a conservative indicator.

There is a common myth that the brain "cannot survive" for more than a few minutes without oxygen. Even if this were true regarding survival of the (spontaneous) ability to resume function, the survival of characteristic cell structure would still be another matter. And indeed, cell structures in the brains of expired dogs, even when kept at room temperature, show only moderate changes after six hours, and many cell structures remain visible for a day or more; see "Studies on the Epithalamus," by Duane E. Haines and Thomas W. Jenkins (Journal of Comparative Neurology, Vol. 132, pp. 405-17, Mar. 1968).

But in fact, the potential for spontaneous brain function can survive for longer than this myth (and the medical definition of "brain death") would suggest. A variety of experiments employing drugs and surgery show this: Adult monkeys have completely recovered after a sixteen-minute cutoff of circulation to the brain (a condition, called "ischemia," which clearly blocks oxygen supply as well); see "Thiopental Therapy After 16 Minutes of Global Brain Ischemia in Monkeys, by A. L. Bleyaert et al. (Critical Care Medicine, Vol. 4, pp. 130-31, Mar./Apr. 1976). Monkey and cat brains have survived for an hour at body temperature without circulation, then recovered electrical function; see "Reversibility of Ischemic Brain Damage," by K.-A. Hossmann and Paul Kleihues (Archives of Neurology, Vol. 29, pp. 375-84, Dec. 1973). Dr. Hossmann concludes that any nerve cell in the brain can survive" for an hour without blood (after the heart stops pumping, for example). The problem is not that nerve cells die when circulation stops, but that secondary problems (such as a slight swelling of the brain within its tight-fitting bone case) can prevent circulation from resuming. When chilled to near freezing, dog brains have recovered electrical activity after four hours without circulation (and have recovered substantial metabolic activity even after fifteen days)" see "Prolonged Whole-Brain Refrigeration with Electrical and Metabolic Recovery," by Robert J. White et al. (Nature, Vol. 209, pp. 1320-22, Mar. 26, 1966).

Brain cells that retain the capability for spontaneous revival at the time when they undergo biostasis should prove easy to repair. Since success chiefly requires that characteristic cell structures remain intact, the time window for beginning biostasis procedures is probably at least several hours after expiration, and possibly longer. Cooperative hospitals can and have made the time much shorter.

... fixation procedures preserve cells ... For high-voltage electron micrographs showing molecular-scale detail in cells preserved by glutaraldehyde fixation, see "The Ground Substance of the Living Cell," by Keith R. Porter and Jonathan B. Tucker (Scientific American, Vol. 244, pp. 56-68, Mar. 1981). Fixation alone does not seem sufficient; long-term stabilization of structure seems to demand freezing or vitrification, either alone or in addition to fixation. Cooling in nitrogen - to minus 196 degrees C - can preserve tissue structures for many thousands of years.

... solidification without freezing ... See "Vitrification as an Approach to Cryopreservation," by G. M. Fahy et al. (Cryobiology, Vol. 21, pp. 407-26, 1984).

... Mouse embryos ... See "Ice-free Cryopreservation of Mouse Embryos at -196 degrees C by Vitrification," by W. F. Rall and G. M. Fahy (Nature, Vol. 313, pp. 573-75, Feb. 14, 1985).

... Robert Ettinger ... published a book ... The Prospect of Immortality (New York: Doubleday, 1964; a preliminary version was privately published in 1962).

... many human cells revive spontaneously ... It is well known that human sperm cells and early embryos survive freezing and storage; in both cases, successes have been reported in the mass media. Less spectacular successes with other cell types (frozen and thawed blood is used for transfusions) are numerous. It is also interesting to note that, after treatment with glycerol and freezing to minus 20 degrees C, cat brains can recover spontaneous electrical activity after over 200 days of storage; see "Viability of Long Term Frozen Cat Brain In Vitro," by I. Suda, K. Kito, and C. Adachi (Nature, Vol. 212, pp. 268-70, Oct. 15, 1966).

... researching ways to freeze and thaw viable organs ... A group at the Cryobiology Laboratory of The American Red Cross (9312 Old Georgetown Road, Bethesda, Md. 20814) is pursuing the preservation of whole human organs to allow the establishment of banks of organs for transplantation; see "Vitrification as an Approach to Cryopreservation," referenced above.

... cell repair ... has been a consistent theme ... As I found when the evident feasibility of cell repair finally led me to examine the cryonics literature. Robert Ettinger's original book, for example (referenced above), speaks of the eventual development of "huge surgeon machines" able to repair tissues' "cell by cell, or even molecule by molecule in critical areas." In 1969 Jerome B. White gave a paper on "Viral Induced Repair of Damaged Neurons with Preservation of Long Term Information Content," proposing that means might be found to direct repair using artificial viruses; see the abstract quoted in Man into Superman, by Robert C. W. Ettinger (New York: St. Martin's Press, 1972, p. 298). In "The Anabolocyte: A Biological Approach to Repairing Cryo-injury" (Life Extension Magazine, pp. 80-83, July/August 1977), Michael Darwin proposed that it might be possible to use genetic engineering to make highly modified white blood cells able to take apart and reconstruct damaged cells. In "How Will They Bring Us Back, 200 Years From Now?" (The Immortalist, Vol. 12, pp. 5-10, Mar. 1981), Thomas Donaldson proposed that systems of molecular machines (with devices as small as viruses and aggregates of devices as large as buildings, if need be) could perform any needed repairs on frozen tissues.

The idea of cell repair systems has thus been around for many years. The concepts of the assembler and the nanocomputer have now made it possible to see clearly how such devices can be built and controlled, and that they can in fact fit within cells.

... the animals fail to revive ... Hamsters, however, have been cooled to a temperature which froze over half the water content in their bodies (and brains), and have then revived with complete recovery; see Biological Effects of Freezing and Supercooling, by Audrey U. Smith (Baltimore: Williams & Wilkins, 1961).

... As Robert Prehoda stated ... in Designing the Future: The Role of Technological Forecasting (Philadelphia: Chilton Book Co., 1967).

... discouraged the use of a workable biostasis technique ... Other factors have also been discouraging - chiefly cost and ignorance. For a patient to pay for a biostasis procedure and to establish a fund that provides for indefinite storage in liquid nitrogen now costs $35,000 or more, depending on the biostasis procedure chosen. This cost is typically covered by purchasing a suitable life insurance policy. Facing this cost and having no clear picture of how freezing damage can be repaired, only a few patients out of millions have so far chosen this course. The small demand, in turn, has prevented economies of scale from lowering the cost of the service. But this may be about to change. Cryonics groups report a recent increase in biostasis contracts, apparently stemming from knowledge of advances in molecular biology and in the understanding of future cell repair capabilities.

Three U.S. groups presently offer biostasis services. In order of their apparent size and quality, they are:

The Alcor Life Extension Foundation, 4030 North Palm No. 304, Fullerton, Calif. 92635, (714) 738-5569. (Alcor also has a branch and facilities in southern Florida.)

Trans Time, Inc., 1507 63rd Street, Emeryville, Calif. 94707, (415) 655-9734.

The Cryonics Institute, 24041 Stratford, Oak Park, Mich. 48237, (313) 967-3115.

For practical reasons based on experience, they require that legal and financial arrangements be completed in advance.

... this preserves neural structures ... The growth of ice crystals can displace cell structures by a few millionths of a meter, but it does not obliterate them, nor does it seem likely to cause any significant confusion regarding where they were before being displaced. Once frozen, they move no further. Repairs can commence before thawing lets them move again.

... clearing ... the major blood vessels ... Current biostasis procedures involve washing out most of a patient's blood; the nanomachines recover any remaining blood cells as they clear the circulatory system.

... throughout the normally active tissues ... This excludes, for example, the cornea, but other means can be used to gain access to the interior of such tissues, or they can simply be replaced.

... that enter cells and remove the glassy protectant ... Molecules of protectant are bound to one another by bonds so weak that they break at room temperature from thermal vibrations. Even at low temperatures, protectant-removal machines will have no trouble pulling these molecules loose from surfaces.

... a temporary molecular scaffolding ... This could be built of nanometer-thick rods, designed to snap together. Molecules could be fastened to the scaffolding with devices resembling double-ended alligator clips.

... the machines label them ... Labels can be made from small segments of coded polymer tape. A segment a few nanometers long can specify a location anywhere within a cubic micron to one-nanometer precision.

... report ... to a larger computer within the cell ... In fact, a bundle of nanometer-diameter signal-transmission fibers the diameter of a finger (with slender branches throughout the patient's capillaries) can in less than a week transmit a complete molecular description of all a patient's cells to a set of external computers. Though apparently unnecessary, the use of external computers would remove most of the significant volume, speed, and power-dissipation constraints on the amount of computation available to plan repair procedures.

... identifies cell structures from molecular patterns ... Cells have stereotyped structures, each built from standard kinds of molecules connected in standard ways in accordance with standard genetic programs. This will greatly simplify the identification problem.

... Richard Feynman saw ... He pointed out the possibility of making devices with wires as little as ten or a hundred atoms wide; see "There's Plenty of Room at the Bottom," in Miniaturization, edited by H. D. Gilbert (New York: Reinhold, 1961), pp. 282-96.

... Robert T. Jones wrote ... in "The Idea of Progress" (Astronautics and Aeronautics, p. 60, May 1981).

... Dr. Lewis Thomas wrote ... in "Basic Medical Research: A Long-Term Investment" (Technology Review, pp. 46-47, May/June 1981).

... Joseph Lister published ... See Volume V, "Fine Chemicals" in A History of Technology, edited by C. J. Singer and others (Oxford: Clarendon Press, 1958).

... Sir Humphry Davy wrote ... See A History of Technology, referenced above.

... the limiting speed is nothing so crude or so breakable ... The principle of relativity of motion means that "moving" objects may be considered to be at rest - meaning that a spaceship pilot trying to approach the speed of light wouldn't even know in what direction to accelerate. Further, simple Minkowski diagrams show that the geometry of space-time makes traveling faster than light equivalent to traveling backward in time - and where do you point a rocket to move in that direction?

... Arthur C. Clarke wrote ... in Profiles of the Future: An Inquiry into the Limits of the Possible, first edition (New York: Harper & Row, 1962).

... its properties limit all that we can do ... For an account of some modern theories that attempt to unify all physics in terms of the behavior of the vacuum, see "The Hidden Dimensions of Spacetime," by Daniel Z. Freedman and Peter van Nieuwenhuizen (Scientific American, Vol. 252, pp. 74-81, Mar. 1985).

... peculiarities far more subtle ... For example, quantum measurements can affect the outcome of other quantum measurements instantaneously at an arbitrarily great distance - but the effects are only statistical and of a subtle sort that has been mathematically proved to be unable to transmit information. See the very readable discussion of Bell's theorem and the Einstein-Podolsky-Rosen paradox in Quantum Reality by Nick Herbert (Garden City, New York: Anchor Press/Doubleday, 1985). Despite rumors to the contrary (some passed on in the final pages of Quantum Reality), nothing seems to suggest that consciousness and the mind rely on quantum mechanics in any special way. For an excellent discussion of how consciousness works (and of how little consciousness we really have) see Marvin Minsky's The Society of Mind (New York: Simon & Schuster, 1986).

... Your victim might have said something vague ... But now physics can answer those questions with clear mathematics. Calculations based on the equations of quantum mechanics show that air is gaseous because nitrogen and oxygen atoms tend to form tightly bonded pairs, unbonded to anything else. Air is transparent because visible light lacks enough energy to excite its tightly bound electrons to higher quantum states, so photons pass through without absorption. A wooden desk is solid because it contains carbon atoms which (as shown by quantum mechanical calculations) are able to form the tightly bonded chains of cellulose and lignin. It is brown because its electrons are in a variety of states, some able to be excited by visible light; it preferentially absorbs bluer, higher-energy photons, making the reflected light yellowish or reddish.

... Stephen W. Hawking states ... In "The Edge of Spacetime" (American Scientist, Vol. 72, pp. 355-59, Jul.-Aug. 1984).

... Few other stable particles are known ... Electrons, protons, and neutrons have stable antiparticles with virtually identical properties save for opposite charges and the ability to annihilate when paired with their twins, releasing energy (or lighter particles). They thus have obvious applications in energy storage. Further, antimatter objects (made from the antiparticles of ordinary matter) may have utility as negative electrodes in high-field electrostatic systems: the field would have no tendency to remove positrons (as it would electrons), making mechanical disruption of the electrode surface the chief limit to field strength. Such electrodes would have to be made and positioned without contacting ordinary matter, of course.

Various physical theories predict a variety of other stable particles (and even massive, particle-like lines), but all would be either so weakly interacting as to be almost undetectable (like neutrinos, only more so) or very massive (like hypothesized magnetic monopoles). Such particles could still be very useful, if found.

... Trying to change a nucleus ... The molecular and field effects used in nuclear magnetic resonance spectroscopy change the orientation of a nucleus, but not its structure.

... the properties of well-separated nuclei ... It has been suggested that excited nuclei might even be made to serve as the lasing medium in a gamma-ray laser.

... would present substantial difficulties ... Before nuclei are pushed close enough together to interact, the associated atomic structures merge to form a solid, metal-like "degenerate matter," stable only under enormous pressure. When the nuclei finally do interact, the exchange of neutrons and other particles soon transmutes them all into similar kinds, obliterating many of the patterns one might seek to build and use.

... insulating against heat ... This is a simple goal to state, but the optimal structures (at least where some compressive strength is required) may be quite complex. Regular crystals transmit heat well, making irregularity desirable, and irregularity means complexity.

... the runners-up will often be nearly as good ... And in some instances, we may design the best possible system, yet never be sure that better systems do not exist.

... Richard Barnet writes ... in The Lean Years: Politics in the Age of Scarcity (New York: Simon & Schuster, 1980).

... Jeremy Rifkin (with Ted Howard) has written ... Entropy: A New World View (New York: Viking Press, 1980).

... "The ultimate moral imperative, then ..." Despite this statement, Rifkin has since struck off on a fresh moral crusade, this time against the idea of evolution and against human beings' modifying genes, even in ways that viruses and bacteria have done for millions of years. Again, he warns of cosmic consequences. But he apparently still believes in the tightly sealed, ever dying world he described in Entropy: "We live by the grace of sacrifice. Every amplification of our being owes its existence to some diminution somewhere else. "Having proved in Entropy that he misunderstands how the cosmos works, he now seeks to advise us about what it wants: "The interests of the cosmos are no different from ours... How then do we best represent the interests of the cosmos? By paying back to the extent to which we have received." But he seems to see all human achievements as fundamentally destructive, stating that "the only living legacy that we can ever leave is the endowment we never touched," and declaring that "life requires death." For more misanthropy and misconceptions, see Algeny, by Jeremy Rifkin (New York: Viking, 1983).

For a confident assertion that genetic engineering is impossible in the first place, made by Rifkin's "prophet and teacher," Nicholas Georgescu Roegen, see The Entropy Law and the Economic Process (Cambridge, Mass: Harvard University Press, 1971).

... exponential growth will overrun ... The demographic transition - the lowering of average birthrates with economic growth - is basically irrelevant to this. The exponential growth of even a tiny minority would swiftly make it a majority, and then make it consume all available resources.

... exploding outward at near the speed of light ... The reason for this rests on a very basic evolutionary argument. Assume that a diverse, competitive civilization begins expanding into space. What groups will be found at the frontier? Precisely those groups that expand fastest. The competition for access to the frontier provides an evolutionary pressure that favors maximum speed of travel and settlement, and that maximum speed is little short of the speed of light (see the notes to Chapter 6). In a hundred million years, such civilizations would spread not just across galaxies, but across intergalactic space. That a thousand or a million times as many civilizations might collapse before reaching space, or might survive without expanding, is simply irrelevant. A fundamental lesson of evolution is that, where replicators are concerned, a single success can outweigh an unlimited number of failures.

... need not contain every possible chemical ... Even the number of possible DNA molecules 50 nucleotides long (four to the fiftieth power) is greater than the number of molecules in a glass of water.

... The Limits to Growth ... by Donella H. Meadows et al. (New York: Universe Books, 1972).

... Mankind at the Turning Point ... by Mihajlo D. Mesarovic and Eduard Pestel (New York: Dutton, 1974).

... trouble enough controlling viruses and fruit flies ... We have trouble even though they are made of conventional molecular machinery. Bacteria are also hard to control, yet they are superficially almost helpless. Each bacterial cell resembles a small, rigid, mouthless box - to eat, a bacterium must be immersed in a film of water that can carry dissolved nutrients for it to absorb. In contrast, assembler-based "superbacteria" could work with or without water; they could feed their molecular machinery with raw materials collected by "mouths" able to attack solid structures.

... AI systems could serve as ... strategists, or fighters ... See "The Fifth Generation: Taking Stock," by M. Mitchell Waldrop (Science, Vol. 226, pp. 1061-63, Nov. 30, 1984), and "Military Robots," by Joseph K. Corrado (Design News, pp. 45-66, Oct. 10, 1983).

... none, if need be ... To be precise, an object can be assembled with a negligible chance of putting any atoms in the wrong place. During assembly, errors can be made arbitrarily unlikely by a process of repeated testing and correction (see the notes for Chapter 4). For example, assume that errors are fairly common. Assume further that a test sometimes fails, allowing one in every thousand errors to pass undetected. If so, then a series of twenty tests will make the chance of failing to detect and correct an error so low that the odds of misplacing a single atom would be slight, even in making an object the size of the Earth. But radiation damage (occurring at a rate proportional to the object's size and age) will eventually displace even correctly placed atoms, so this degree of care would be pointless.

... a cosmic ray can unexpectedly knock atoms loose from anything ... It might seem that shielding could eliminate this problem, but neutrinos able to penetrate the entire thickness of the Earth - or Jupiter, or the Sun - can stilt cause radiation damage, though at a very small rate. See "The Search for Proton Decay," by J. M. LoSecco et al. (Scientific American, Vol. 252, p. 59, June 1985).

... Stratus Computer Inc., for example ... See "Fault-Tolerant Systems in Commercial Applications," by Omri Serlin (Computer, Vol. 17, pp. 19-30, Aug. 1984).

...design diversity ... See "Fault Tolerance by Design Diversity: Concepts and Experiments," by Algirdas Avizienis and John P. J. Kelly (Computer, Vol. 17, pp. 67-80, Aug. 1984).

... redundancy ... multiple DNA strands ... The bacterium Micrococcus radiodurans apparently has quadruple-redundant DNA, enabling it to survive extreme radiation doses. See "Multiplicity of Genome Equivalents in the Radiation-Resistant Bacterium Micrococcus radiodurans," by Mogens T. Hansen, in Journal of Bacteriology, pp. 7 1-75, Apr. 1978.

... other effective error-correcting systems ... Error correction based on multiple copies is easier to explain, but digital audio disks (for example) use other methods that allow error correction with far less redundant information. For an explanation of a common error-correcting code, see "The Reliability of Computer Memory," by Robert McEliece (Scientific American, Vol. 248, pp. 88-92, Jan. 1985).

... intelligence will involve mental parts ... See The Society of Mind, by Marvin Minsky (New York: Simon & Schuster, 1986).

... "The Scientific Community Metaphor" ... by William A. Kornfeld and Carl Hewitt (MIT Al Lab Memo No. 641, Jan. 1981).

... AI systems can be made trustworthy ... Safety does not require that all AI systems be made trustworthy, so long as some are trustworthy and help us plan precautions for the rest.

... One proposal... This is a concept being developed by Mark Miller and myself; it is related to the ideas discussed in "Open Systems," by Carl Hewitt and Peter de Jong (MIT AI Lab Memo No. 692, Dec. 1982).

... more reliably than ... human engineers ... if only because fast AI systems (like those described in Chapter 5) will be able to find and correct errors a million times faster.

... other than specially designed AI programs ... See "The Role of Heuristics in Learning by Discovery," by Douglas B. Lenat, in Machine Learning, edited by Michalski et al. (Palo Alto, Calif: Tioga Publishing Company, 1983). For a discussion of the successful evolution of programs designed to evolve, see pp. 243-85. For a discussion of the unsuccessful evolution of programs intended to evolve but not properly designed to do so, see pp. 288-92.

... neglect to give replicators similar talents ... Lacking these, "gray goo" might be able to replace us and yet be unable to evolve into anything interesting.

... to correct its calculations ... Calculations will allow the system to picture molecular structures that it has not directly characterized. But calculations may lead to ambiguous results in borderline cases-actual results may even depend on random tunneling or thermal noise. In this case, the measurement of a few selected atomic positions (performed by direct mechanical probing of the workpiece's surface) should suffice to distinguish among the possibilities, thus correcting the calculations. This can also correct for error buildup in calculations of the geometry of large structures.

... Each sensor layer ... As described, these sensor layers must be penetrated by wires, which might seem to present a security problem: what if something were to get past the sensors by eating its way along a wire? In practice, anything that can transmit signals and power (including optical fibers, mechanical transmission systems, and so forth) could be used in place of wires. These channels can be made secure by basing them on materials with extreme properties: if a very fine wire is made of the most conductive material, or if a mechanical transmission system is made of the strongest material (and used near its breaking stress), then any attempt to replace a segment with something else (such as an escaping replicator) will show up as a greater electrical resistance or a fractured part. Thus the transmission systems themselves can act as sensors. For the sake of redundancy and design diversity, different sensor layers could be penetrated by different transmission systems, each transmitting signals and power to the next.

... If we destroy the records of the protein designs ... But how could people be made to forget? This is not really necessary, since their knowledge would be dispersed. In developing modern hardware systems, different teams work on different parts; they need not know what another team's part is (much less how it is made), because only how it interacts really matters to them. In this way people reached the Moon, though no one person or team ever fully knew how to do it; it could be likewise with assemblers.

Since the first assembler designs will be historic documents, it might be better to store them securely, rather than destroy them. Eventually they can become part of the open literature. But hiding design information will at best be a stopgap, since the methods used for the design of the first assembler system will be harder to keep secret. Further, sealed assembler labs might be used to develop and test machines that can make assemblers, even machines that can themselves be made without assemblers.

... no fixed wall will be proof against large-scale, organized malice ... Sealed assembler labs can work despite this. They do not protect their contents from the outside; in fact, they are designed to destroy their contents when tampered with. Instead, they protect the outside from their contents - and their sealed work spaces are too small to hold any large-scale system, malicious or not.

... giving the attacker no obvious advantage ... In the examples cited, organized entities were pitted against similar entities. These entities could, of course, be vaporized by hydrogen bombs, but faced with the prospect of retaliation in kind, no attacker has yet seen an advantage in launching a nuclear strike.

... occupy hostile powers ... In principle, this could be a minimal form of occupation, controlling only research laboratories, but even this would require a degree of coercion roughly equivalent to conquest.

... as open as possible ... It may be possible to devise forms of inspection that give a group great confidence in what a system under development will (and will not) be able to do, without letting that group learn how those systems are made. Compartmentalized development of a system's components could, in principle, allow several groups to cooperate without any single group's being able to build and use a similar system independently.

... we naturally picture human hands aiming it ... For a discussion of autonomous spacecraft, see "Expanding Role for Autonomy in Military Space," by David D. Evans and Maj. Ralph R. Gajewski (Aerospace America, pp. 74-77, Feb. 1985). See also "Can Space Weapons Serve Peace?" by K. Eric Drexler (L5 News, Vol. 9, pp. 1-2, Jul. 1983).

... while providing each with some protection ... Saying that a symmetrical, 50 percent effective shield would be worthless is like saying that a bilateral 50 percent reduction in nuclear missiles - a real breakthrough in arms control - would be worthless. The practicality of such a shield is another matter. Until really good active shields become possible, the question is not one of making a nuclear attack harmless, but at best of making it less likely.

... limiting technology transfer ... In fact, President Reagan has spoken of giving away U.S. space defense technology to the Soviet Union. See the New York Times, p. A15, March 30, 1983. See also - Sharing "Star Wars, technology with Soviets a distant possibility, says head of Pentagon study group," by John Horgan (The Institute, p. 10, Mar. 1984). Richard Ullman, professor of international affairs at Princeton University, has proposed a joint defense program with extensive sharing of technology; see "U.N.-doing Missiles" (New York Times, p. A23, Apr. 28, 1983).

In principle, a joint project could proceed with little technology transfer. There is a great difference between (1) knowing what a device cannot do, (2) knowing what it can do, (3) knowing what it is, and (4) knowing how to make it. These define four levels of knowledge, each (more or less) independent of the levels beyond it. For example, if I were to hand you a plastic box, a superficial examination might convince you that it cannot fly or shoot bullets, but not tell you what it can do. A demonstration might then convince you that it can serve as a cordless telephone. By inspecting it more closely, you could trace its circuits and gain an excellent idea of what it is and of what its operating limits are. But you still wouldn't necessarily know how to make one.

The essence of an active shield lies in what it cannot do - that is, that it cannot be used as a weapon. To conduct a joint active-shield project relying on high-technology components, one would need to share knowledge chiefly on levels (1) and (2). This requires at least limited sharing on level (3), but need not require any on level (4).

... basic issues common to all active shields ... Such as those of their control, purpose, and reliability, and the fundamental issue of political understanding and acceptance.

... a U.S. National Science Foundation survey ... as quoted by NSF Director John B. Slaughter (Time, p. 55, June 15, 1981).

... Advice and Dissent ... subtitled "Scientists in the Political Arena," by Joel Primack and Frank von Hippel (New York: Basic Books, 1974).

... Hazel Henderson argues ... in Creating Alternative Futures: The End of Economics (New York: Berkley Publishing, 1978).

... Harrison Brown likewise argues ... in The Human Future Revisited: The World Predicament and Possible Solutions (New York: Norton, 1978).

... Debates ... over the safety of nuclear power ... For a discussion of the failures at the Three Mile Island nuclear power plant, and a discussion of (1) the remarkable degree of agreement on the problems reached by an expert panel and (2) how the media mangled the story, and (3) how the federal government had failed to respond to reality, see "Saving American Democracy," by John G. Kemeny, president of Dartmouth College and chairman of the presidential commission on Three Mile Island (Technology Review, pp. 65-75, June/July 1980). He concludes that "the present system does not work."

... Disputes over facts ... Worse yet, two people can agree on the facts and on basic values (say, that wealth is good and pollution is bad) and yet disagree about building a factory - one person may be more concerned about wealth, and the other about pollution. In emotional debates, this can lead each side to accuse the other of perverted values, such as favoring poverty or caring nothing for the environment. Nanotechnology will ease such conflicts by changing the trade-offs. Because we can have much more wealth and much less pollution, old opponents may more often find themselves in agreement.

... AI researchers ... See "The Scientific Community Metaphor," by William A. Kornfeld and Carl Hewitt (MIT AI Lab Memo No. 641, Jan. 1981); see also the discussion of "due-process reasoning" in "The Challenge of Open Systems," by Carl Hewitt (Byte, Vol. 10, pp. 223-41, April 1985).

... procedures somewhat like those of courts ... These might use written communications, as in journals, rather than face-to-face meetings: judging the truth of a statement by the manner in which it is said is useful in courts, but plays a lesser role in science.

... Kantrowitz ... originated the concept ... He did so in the mid-1960s. See his discussion in "Controlling Technology Democratically" (American Scientist, Vol. 63, pp. 505-9, Sept.-Oct. 1975).

... used (or proposed) as a government institution ... This is the original usage of the term "science court," and many criticisms of the due-process idea have stemmed from this aspect of the proposal. The fact forum approach is genuinely different; Dr. Kantrowitz is presently pursuing it under the name "Scientific Adversary Procedure."

... backed by the findings of an expert committee ... which in 1960 had drawn up a proposed space program for the Air Force. It emphasized that learning to assemble systems in Earth orbit (such as space stations and Moon ships) was at least as important as building bigger boosters. During the subsequent debate on how to reach the Moon, Kantrowitz argued that Earth-orbital assembly would be perhaps ten times less expensive than the giant-rocket, lunar-orbit-rendezvous approach that was finally chosen. But political factors intervened, and the matter never received a proper public hearing. See "Arthur Kantrowitz Proposes a Science Court," an interview by K. Eric Drexler (L5 News, Vol. 2, p. 16, May 1977).

For an account of another abuse of technical decision-making during Apollo, see The Heavens and the Earth: A Political History of the Space Age, by William McDougall, pp. 315-16 (New York: Basic Books, 1985).

... a proposed procedure ... This is described in "The Science Court Experiment: An Interim Report," by the Task Force of the Presidential Advisory Group on Anticipated Advances in Science and Technology (Science, Vol. 193, pp. 653-56, Aug. 20, 1976).

... a colloquium on the science court ... see the Proceedings of the Colloquium on the Science Court, Leesburg, Virginia, Sept. 20-21, 1976 (National Technical Information Center, document number PB261 305). For a summary and discussion of the criticisms voiced at the colloquium, see "The Science Court Experiment: Criticisms and Responses," by Arthur Kantrowitz (Bulletin of the Atomic Scientists, Vol. 33, pp. 44-49, Apr. 1977).

... could move toward due process ... The formation of the Health Effects Institute of Cambridge, Massachusetts, created in 1980 to bring together adversaries in the field of air pollution, has been a step in this direction. See "Health Effects Institute Links Adversaries," by Eliot Marshall (Science, Vol. 227, pp. 729-30, Feb. 15, 1985).

... knowledge is ... guarded ... An open question is the extent to which non-public procedures embodying some due process principles can improve the judging of classified information.

... an experimental procedure ... Reported in "Science court, would tackle knotty technological issues," by Leon Lindsay (Christian Science Monitor, p. 7, Mar. 23, 1983).

[Note: More recent information is available on the Web at "Twenty-Five Year Retrospective on the Science Court"]

... Roger Fisher and William Ury ... See Getting to Yes (Boston: Houghton Mifflin Company, 1981).

... Both sides ... The procedures described here treat issues as two-sided, but this may seem too limited, because "issues," as commonly understood, often have many sides. In the energy debate, for example, gas, coal, nuclear, and solar power all have their advocates. Yet multisided issues contain many two-sided questions: Is the probability of a reactor meltdown low or high? Are the effects of coal burning on acid rain small or large? Will a solar collector cost little or much? Are gas reserves small or large? Multisided issues thus often resolve at their factual roots into numerical micro-questions.

Judicious scientists and engineers will seldom argue for high or low numbers as such; they will argue for the particular numbers they think most likely, or simply state evidence. But since holding a forum presupposes a dispute, advocates will be involved, and they will often wish to push far in one direction - nuclear advocates would like to prove that reactors are very cheap and safe; their opponents would like to prove that they are very expensive and deadly. Because numbers measuring cost and risk can only be larger or smaller, these micro-questions will tend to be two-sided.

... Tohru Moto-oka ... He is a professor at Tokyo University and the titular head of Japan's Fifth Generation Computer Project.

... one system's structure ... Their approach to hypertext, now in the demonstration stage, is called the Xanadu system. I have examined the proprietary data structures on which their system is based, and it is clear that powerful hypertext systems are indeed possible. For a less ambitious yet still quite powerful system, see "A Network-Based Approach to Text-Handling for the Online Scientific Community," a thesis by Randall H. Trigg (University of Maryland, Department of Computer Science, TR-1346, Nov. 1983).

... Theodor Nelson's books ... See Computer Lib/Dream Machines (self-published, distributed by The Distributors, South Bend, Ind., 1974), and Literary Machines (Swarthmore, Pa: Ted Nelson, 1981). Computer Lib is an entertaining and idiosyncratic view of computers and their potential, including hypertext; a new edition is in preparation. Literary Machines focuses on hypertext.

... Time magazine reports ... on p. 76, June 13, 1983.

... increasing the quantity of information available ... A hypertext system might store the most commonly used information in the home, or in a local branch library. Compact disks of the sort used for audio recordings cost about three dollars to manufacture and can store as much text as about 500 books. See "Audio Analysis II: Read-only Optical Disks," by Christopher Fry (Computer Music Journal, Vol. 9, Summer 1985).

... bring abundance and long life to all who wish them ... But the limits to exponential growth ensure that universal, unconditional abundance cannot last indefinitely. This raises questions regarding the distribution and ownership of space resources. Three basic approaches might be considered:

One is a first-come, first-served approach, like the claiming of homesteads or mining sites through use. This has roots in the Lockean principle that ownership may be established by mixing one's labor with a previously unowned resource. But this might allow a person with a suitable replicator to turn it loose in space to rework - and thus claim - every unclaimed object in the universe, as fast as it could be reached. This winner-take-all approach has little moral justification, and would have unpleasant consequences.

A second extreme would be to distribute ownership of space resources equally among all people, and to keep redistributing them to maintain equality. This, too, would have unpleasant consequences. In the absence of universal, stringent, compulsory limitations on childbearing, some groups would continue to grow exponentially; evolutionary principles virtually guarantee this. In a surprisingly short time, the result of endless redistribution would be to drag the standard of living of every human being down to the minimum level that allows any group to reproduce. This would mean hunger and poverty more extreme and universal than that of any Third World country. If 99 percent of the human race voluntarily limited its birth rate, this would merely allow the remaining one percent to expand until it absorbed almost all the resources.

A third basic approach (which has many variations) takes a middle path: it involves distributing ownership of the resources of space (genuine, permanent, transferable ownership) equally among all people - but doing so only once, then letting people provide for their progeny (or others') from their own vast share of the wealth of space. This will allow different groups to pursue different futures, and it will reward the frugal rather than the profligate. It can provide the foundation for a future of unlimited diversity for the indefinite future, if active shields are used to protect people from aggression and theft. No one has yet voiced a plausible alternative.

From a socialist perspective, this approach means equal riches for all. From a libertarian perspective, it violates no one's property rights and provides a basis for a future of liberty. In Thomas Schelling's terms, equal division is a focal point solution in a coordination game (see The Strategy of Conflict, by Thomas Schelling, Cambridge, Mass: Harvard University Press, 1960). What "equal division" actually means is a messy question best left to lawyers.

For this approach to work, agreement will be needed not just on a principle of division, but on a date. Space has been declared by treaty to be "the common heritage of all mankind," and we need to choose an Inheritance Day. Schelling's analysis suggests the importance, in a coordination game, of finding a specific, plausible proposal and of making it visible as soon as possible. Does a date suggest itself? A round-numbered space-related anniversary would seem appropriate, if it were not tied exclusively to the U.S. or U.S.S.R., or too soon, or too near a millennial date on the calendar. These constraints can be met; the most plausible candidate is perhaps April 12, 2011: the thirtieth anniversary of the flight of the world's first reusable spacecraft, the space shuttle, and the fiftieth anniversary of the flight of the first human into space, Yuri Gagarin.

If, before this date, someone finds and employs a means to raise human reproduction rates by a factor of ten or more, then Inheritance Day should immediately be made retroactive to April 12 of the preceding year, and the paperwork sorted out later.

... to secure a stable, durable peace ... Active shields can accomplish this reliably only through the use of redundancy and ample safety margins.

... but nature seems uncooperative ... For a discussion of the apparent impossibility of time machines in general relativity, see "Singularities and Causality Violation," by Frank J. Tipler in Annals of Physics, Vol. 108, pp. 1-36, 1977. Tipler is open-minded; in 1974 he had argued the other side of the case.

... patterns that resemble ... "fractals" ... Fractal patterns have similar parts on different scales - as, a twig may resemble a branch which in turn resembles a tree, or as gullies, streams, and rivers may all echo each other's forms. See The Fractal Geometry of Nature, by Benoit B. Mandelbrot (San Francisco: W. H. Freeman, 1982).