ON MARCH 27, 1981, CBS radio news quoted a NASA scientist as saying that engineers will be able to build self-replicating robots within twenty years, for use in space or on Earth. These machines would build copies of themselves, and the copies would be directed to make useful products. He had no doubt of their possibility, only of when they will be built. He was quite right.
Since 1951, when John von Neumann outlined the principles of self-replicating machines, scientists have generally acknowledged their possibility. In 1953 Watson and Crick described the structure of DNA, which showed how living things pass on the instructions that guide their construction. Biologists have since learned in increasing detail how the self-replicating molecular machinery of the cell works. They find that it follows the principles von Neumann had outlined. As birds prove the possibility of flight, so life in general proves the possibility of self-replication, at least by systems of molecular machines. The NASA scientist, however, had something else in mind.
The ancient myth of a magical life-force (coupled with the misconception that the increase of entropy means that everything in the universe must constantly run down) has spawned a meme saying that replicators must violate some natural law. This simply isn't so. Biochemists understand how cells replicate and they find no magic in them. Instead, they find machines supplied with all the materials, energy, and instructions needed to do the job. Cells do replicate; robots could replicate.
Advances in automation will lead naturally toward mechanical replicators, whether or not anyone makes them a specific goal. As competitive pressures force increased automation, the need for human labor in factories will shrink. Fujitsu Fanuc already runs the machining section in a manufacturing plant twenty-four hours a day with only nineteen workers on the floor during the day shift, and none on the floor during the night shift. This factory produces 250 machines a month, of which 100 are robots.
Eventually, robots could do all the robot-assembly work, assemble other equipment, make the needed parts, run the mines and generators that supply the various factories with materials and power, and so forth. Though such a network of factories spread across the landscape wouldn't resemble a pregnant robot, it would form a self-expanding, self-replicating system. The assembler breakthrough will surely arrive before the complete automation of industry, yet modern moves in this direction are moves toward a sort of gigantic, clanking replicator.
But how can such a system be maintained and repaired without human labor?
Imagine an automatic factory able to both test parts and assemble equipment. Bad parts fail the tests and are thrown out or recycled. If the factory can also take machines apart, repairs are easy: simply disassemble the faulty machines, test all their parts, replace any worn or broken parts, and reassemble them. A more efficient system would diagnose problems without testing every part, but this isn't strictly necessary.
A sprawling system of factories staffed by robots would be workable but cumbersome. Using clever design and a minimum of different parts and materials, engineers could fit a replicating system into a single box - but the box might still be huge, because it must contain equipment able to make and assemble many different parts. How many different parts? As many as it itself contains. How many different parts and materials would be needed to build a machine able to make and assemble so many different materials and parts? This is hard to estimate, but systems based on today's technology would use electronic chips. Making these alone would require too much equipment to stuff into the belly of a small replicator.
Rabbits replicate, but they require prefabricated parts such as vitamin molecules. Getting these from food lets them survive with less molecular machinery than they would need to make everything from scratch. Similarly, a mechanical replicator using prefabricated chips could be made somewhat simpler than one that made everything it needed. Its peculiar "dietary" requirements would also tie it to a wider "ecology" of machines, helping to keep it on a firm leash. Engineers in NASA-sponsored studies have proposed using such semireplicators in space, allowing space industry to expand with only a small input of sophisticated parts from Earth.
Still, since bulk-technology replicators must make and assemble their parts, they must contain both part-making and part-assembling machines. This highlights an advantage of molecular replicators: their parts are atoms, and atoms come ready-made.
But engineers seem more likely to develop other approaches to replication. Evolution had no easy way to alter the fundamental pattern of the cell, and this pattern has shortcomings. In synapses, for example, the cells of the brain signal their neighbors by emptying bladders of chemical molecules. The molecules then jostle around until they bind to sensor molecules on the neighboring cell, sometimes triggering a neural impulse. A chemical synapse makes a slow switch, and neural impulses move slower than sound. With assemblers, molecular engineers will build entire computers smaller than a synapse and a millionfold faster.
Mutation and selection could no more make a synapse into a mechanical nanocomputer than a breeder could make a horse into a car. Nonetheless, engineers have built cars, and will also learn to build computers faster than brains, and replicators more capable than existing cells.
Some of these replicators will not resemble cells at all, but will instead resemble factories shrunk to cellular size. They will contain nanomachines mounted on a molecular framework and conveyor belts to move parts from machine to machine. Outside, they will have a set of assembler arms for building replicas of themselves, an atom or a section at a time.
How fast these replicators can replicate will depend on their assembly speed and their size. Imagine an advanced assembler that contains a million atoms: it can have as many as ten thousand moving parts, each containing an average of one hundred atoms - enough parts to make up a rather complex machine. In fact, the assembler itself looks like a box supporting a stubby robot arm a hundred atoms long. The box and arm contain devices that move the arm from position to position, and others that change the molecular tools at its tip.
Behind the box sits a device that reads a tape and provides mechanical signals that trigger arm motions and tool changes. In front of the arm sits an unfinished structure. Conveyors bring molecules to the assembler system. Some supply energy to motors that drive the tape reader and arm, and others supply groups of atoms for assembly. Atom by atom (or group by group), the arm moves pieces into place as directed by the tape; chemical reactions bond them to the structure on contact.
These assemblers will work fast. A fast enzyme, such as carbonic anhydrase or ketosteroid isomerase, can process almost a million molecules per second, even without conveyors and power-driven mechanisms to slap a new molecule into place as soon as an old one is released. It might seem too much to expect an assembler to grab a molecule, move it, and jam it into place in a mere millionth of a second. But small appendages can move to and fro very swiftly. A human arm can flap up and down several times per second, fingers can tap more rapidly, a fly can wave its wings fast enough to buzz, and a mosquito makes a maddening whine. Insects can wave their wings at about a thousand times the frequency of a human arm because an insect's wing is about a thousand times shorter.
An assembler arm will be about fifty million times shorter than a human arm, and so (as it turns out) it will be able to move back and forth about fifty million times more rapidly. For an assembler arm to move a mere million times per second would be like a human arm moving about once per minute: sluggish. So it seems a very reasonable goal.
The speed of replication will depend also on the total size of the system to be built. Assemblers will not replicate by themselves; they will need materials and energy, and instructions on how to use them. Ordinary chemicals can supply materials and energy, but nanomachinery must be available to process them. Bumpy polymer molecules can code information like a punched paper tape, but a reader must be available to translate the patterns of bumps into patterns of arm motion. Together, these parts form the essentials of a replicator: the tape supplies instructions for assembling a copy of the assembler, of the reader, of the other nanomachines, and of the tape itself.
A reasonable design for this sort of replicator will likely include several assembler arms and several more arms to hold and move workpieces. Each of these arms will add another million atoms or so. The other parts - tape readers, chemical processors, and so forth-may also be as complicated as assemblers. Finally, a flexible replicator system will probably include a simple computer; following the mechanical approach that I mentioned in Chapter 1, this will add roughly 100 million atoms. Altogether, these parts will total less than 150 million atoms. Assume instead a total of one billion, to leave a wide margin for error. Ignore the added capability of the additional assembler arms, leaving a still wider margin. Working at one million atoms per second, the system will still copy itself in one thousand seconds, or a bit over fifteen minutes - about the time a bacterium takes to replicate under good conditions.
Imagine such a replicator floating in a bottle of chemicals, making copies of itself. It builds one copy in one thousand seconds, thirty-six in ten hours. In a week, it stacks up enough copies to fill the volume of a human cell. In a century, it stacks up enough to make a respectable speck. If this were all that replicators could do, we could perhaps ignore them in safety.
Each copy, though, will build yet more copies. Thus the first replicator assembles a copy in one thousand seconds, the two replicators then build two more in the next thousand seconds, the four build another four, and the eight build another eight. At the end of ten hours, there are not thirty-six new replicators, but over 68 billion. In less than a day, they would weigh a ton; in less than two days, they would outweigh the Earth; in another four hours, they would exceed the mass of the Sun and all the planets combined - if the bottle of chemicals hadn't run dry long before.
Regular doubling means exponential growth. Replicators multiply exponentially unless restrained, as by lack of room or resources. Bacteria do it, and at about the same rate as the replicators just described. People replicate far more slowly, yet given time enough they, too, could overshoot any finite resource supply. Concern about population growth will never lose its importance. Concern about controlling rapid new replicators will soon become important indeed.
To make large objects rapidly, a vast number of assemblers must cooperate, but replicators will produce assemblers by the ton. Indeed, with correct design, the difference between an assembler system and a replicator will lie entirely in the assembler's programming.
If a replicating assembler can copy itself in one thousand seconds, then it can be programmed to build something else its own size just as fast. Similarly, a ton of replicators can swiftly build a ton of something else - and the product will have all its billions of billions of billions of atoms in the right place, with only a minute fraction misplaced.
To see the abilities and limits of one method for assembling large objects, imagine a flat sheet covered with small assembly arms-perhaps an army of replicators reprogrammed for construction work and arrayed in orderly ranks. Conveyors and communication channels behind them supply reactive molecules, energy, and assembly instructions. If each arm occupies an area 100 atomic diameters wide, then behind each assembler will be room for conveyors and channels totaling about 10,000 atoms in cross sectional area.
This seems room enough. A space ten or twenty atoms wide can hold a conveyor (perhaps based on molecular belts and pulleys). A channel a few atoms wide can hold a molecular rod which, like those in the mechanical computer mentioned in Chapter 1, will be pushed and pulled to transmit signals. All the arms will work together to build a broad, solid structure layer by layer. Each arm will be responsible for its own area, handling about 10,000 atoms per layer. A sheet of assemblers handling 1,000,000 atoms per second per arm will complete about one hundred atomic layers per second. This may sound fast, but at this rate piling up a paper-sheet thickness will take about an hour, and making a meter-thick slab will take over a year.
Faster arms might raise the assembly speed to over a meter per day, but they would produce more waste heat. If they could build a meter-thick layer in a day, the heat from one square meter could cook hundreds of steaks simultaneously, and might fry the machinery. At some size and speed, cooling problems will become a limiting factor, but there are other ways of assembling objects faster without overheating the machinery.
Imagine trying to build a house by gluing together individual grains of sand. Adding a layer of grains might take grain-gluing machines so long that raising the walls would take decades. Now imagine that machines in a factory first glue the grains together to make bricks. The factory can work on many bricks at once. With enough grain-gluing machines, bricks would pour out fast; wall assemblers could then build walls swiftly by stacking the preassembled bricks. Similarly, molecular assemblers will team up with larger assemblers to build big things quickly - machines can be any size from molecular to gigantic. With this approach, most of the assembly heat will be dissipated far from the work site, in making the parts.
Skyscraper construction and the architecture of life suggest a related way to construct large objects. Large plants and animals have vascular systems, intricate channels that carry materials to molecular machinery working throughout their tissues. Similarly, after riggers and riveters finish the frame of a skyscraper, the building's "vascular system" - its elevators and corridors, aided by cranes - carry construction materials to workers throughout the interior. Assembly systems could also employ this strategy, first putting up a scaffold and then working throughout its volume, incorporating materials brought through channels from the outside.
Imagine this approach being used to "grow" a large rocket engine, working inside a vat in an industrial plant. The vat - made of shiny steel, with a glass window for the benefit of visitors - stands taller than a person, since it must hold the completed engine. Pipes and pumps link it to other equipment and to water-cooled heat exchangers. This arrangement lets the operator circulate various fluids through the vat.
To begin the process, the operator swings back the top of the vat and lowers into it a base plate on which the engine will be built. The top is then resealed. At the touch of a button, pumps flood the chamber with a thick, milky fluid which submerges the plate and then obscures the window. This fluid flows from another vat in which replicating assemblers have been raised and then reprogrammed by making them copy and spread a new instruction tape (a bit like infecting bacteria with a virus). These new assembler systems, smaller than bacteria, scatter light and make the fluid look milky. Their sheer abundance makes it viscous.
At the center of the base plate, deep in the swirling, assembler-laden fluid, sits a "seed." It contains a nanocomputer with stored engine plans, and its surface sports patches to which assemblers stick. When an assembler sticks to it, they plug themselves together and the seed computer transfers instructions to the assembler computer. This new programming tells it where it is in relation to the seed, and directs it to extend its manipulator arms to snag more assemblers. These then plug in and are similarly programmed. Obeying these instructions from the seed (which spread through the expanding network of communicating assemblers) a sort of assembler-crystal grows from the chaos of the liquid. Since each assembler knows its location in the plan, it snags more assemblers only where more are needed. This forms a pattern less regular and more complex than that of any natural crystal. In the course of a few hours, the assembler scaffolding grows to match the final shape of the planned rocket engine.
Then the vat's pumps return to life, replacing the milky fluid of unattached assemblers with a clear mixture of organic solvents and dissolved substances - including aluminum compounds, oxygen-rich compounds, and compounds to serve as assembler fuel. As the fluid clears, the shape of the rocket engine grows visible through the window, looking like a full-scale model sculpted in translucent white plastic. Next, a message spreading from the seed directs designated assemblers to release their neighbors and fold their arms. They wash out of the structure in sudden streamers of white, leaving a spongy lattice of attached assemblers, now with room enough to work. The engine shape in the vat grows almost transparent, with a hint of iridescence.
Each remaining assembler, though still linked to its neighbors, is now surrounded by tiny fluid-filled channels. Special arms on the assemblers work like flagella, whipping the fluid along to circulate it through the channels. These motions, like all the others performed by the assemblers, are powered by molecular engines fueled by molecules in the fluid. As dissolved sugar powers yeast, so these dissolved chemicals power assemblers. The flowing fluid brings fresh fuel and dissolved raw materials for construction; as it flows out it carries off waste heat. The communications network spreads instructions to each assembler.
The assemblers are now ready to start construction. They are to build a rocket engine, consisting mostly of pipes and pumps. This means building strong, light structures in intricate shapes, some able to stand intense heat, some full of tubes to carry cooling fluid. Where great strength is needed, the assemblers set to work constructing rods of interlocked fibers of carbon, in its diamond form. From these, they build a lattice tailored to stand up to the expected pattern of stress. Where resistance to heat and corrosion is essential (as on many surfaces), they build similar structures of aluminum oxide, in its sapphire form. In places where stress will be low, the assemblers save mass by leaving wider spaces in the lattice. In places where stress will be high, the assemblers reinforce the structure until the remaining passages are barely wide enough for the assemblers to move. Elsewhere the assemblers lay down other materials to make sensors, computers, motors, solenoids, and whatever else is needed.
To finish their jobs, they build walls to divide the remaining channel spaces into almost sealed cells, then withdraw to the last openings and pump out the fluid inside. Sealing the empty cells, they withdraw completely and float away in the circulating fluid. Finally, the vat drains, a spray rinses the engine, the lid lifts, and the finished engine is hoisted out to dry. Its creation has required less than a day and almost no human attention.
What is the engine like? Rather than being a massive piece of welded and bolted metal, it is a seamless thing, gemlike. Its empty internal cells, patterned in arrays about a wavelength of light apart, have a side effect: like the pits on a laser disk they diffract light, producing a varied iridescence like that of a fire opal. These empty spaces lighten a structure already made from some of the lightest, strongest materials known. Compared to a modern metal engine, this advanced engine has over 90 percent less mass.
Tap it, and it rings like a bell of surprisingly high pitch for its size. Mounted in a spacecraft of similar construction, it flies from a runway to space and back again with ease. It stands long, hard use because its strong materials have let designers include large safety margins. Because assemblers have let designers pattern its structure to yield before breaking (blunting cracks and halting their spread), the engine is not only strong but tough.
For all its excellence, this engine is fundamentally quite conventional. It has merely replaced dense metal with carefully tailored structures of light, tightly bonded atoms. The final product contains no nanomachinery.
More advanced designs will exploit nanotechnology more deeply. They could leave a vascular system in place to supply assembler and disassembler systems; these can be programmed to mend worn parts. So long as users supply such an engine with energy and raw materials, it will renew its own structure. More advanced engines can also be literally more flexible. Rocket engines work best if they can take different shapes under different operating conditions, but engineers cannot make bulk metal strong, light, and limber. With nanotechnology, though, a structure stronger than steel and lighter than wood could change shape like muscle (working, like muscle, on the sliding fiber principle), An engine could then expand, contract, and bend at the base to provide the desired thrust in the desired direction under varying conditions. With properly programmed assemblers and disassemblers, it could even remodel its fundamental structure long after leaving the vat.
In short, replicating assemblers will copy themselves by the ton, then make other products such as computers, rocket engines, chairs, and so forth. They will make disassemblers able to break down rock to supply raw material. They will make solar collectors to supply energy. Though tiny, they will build big. Teams of nanomachines in nature build whales, and seeds replicate machinery and organize atoms into vast structures of cellulose, building redwood trees. There is nothing too startling about growing a rocket engine in a specially prepared vat. Indeed, foresters given suitable assembler "seeds" could grow spaceships from soil, air, and sunlight.
Assemblers will be able to make virtually anything from common materials without labor, replacing smoking factories with systems as clean as forests. They will transform technology and the economy at their roots, opening a new world of possibilities. They will indeed be engines of abundance.