18

  Very Small Legos

The principles of physics, as far as I can see, do not speak against the possibility of maneuvering things atom by atom. It is not an attempt to violate any laws; it is something, in principle, that can be done; but in practice, it has not been done because we are too big.

Richard Feynman, "There's Plenty of Room at the Bottom," a talk delivered in 1959¹

We all know that atoms are small. Avogadro’s number describes just how small they are. Written out in full it is about 602,400,000,000,000,000,000,000. That is the ratio between grams, the units we use to measure the mass of small objects – a dime weighs slightly over two grams – and the units in which we measure the mass of atoms. An atom of hydrogen has an atomic weight of about one, so Avogadro’s number is the number of atoms in a gram of hydrogen.

Looking at all those zeros, you can see that even very small objects have a lot of atoms in them. A human hair, for example, contains more than a million billion. The microscopic transistors in a computer chip are small compared to us but large compared to an atom. Everything humans construct, with the exception of some very recent experiments, is built out of enormous conglomerations of atoms.

We ourselves, on the other hand, like all living things, are engineered at the atomic scale. The cellular machinery that makes us run depends on single molecules – enzymes, proteins, DNA, RNA, and the like – each a complicated structure of atoms, every one in the right place. When an atom in a strand of DNA is in the wrong place, the result is a mutation.² As we become better and better at manipulating very small objects it begins to become possible for us to build as we are built, to construct machines at the atomic level, assembling individual atoms into molecules that do things. That is the central idea of nanotechnology.³

One attraction of the idea is that it lets you build things that cannot be built with present technologies. Since the bonds between atoms are very strong, it should be possible to build very strong fibers from long strand molecules. It should be possible to use diamond – merely a particular arrangement of carbon atoms – as a structural material. We may even be able to build mechanical computers, inspired by Babbage’s failed nineteenth-century design. Mechanical parts move very slowly compared to the movement of electrons in electronic computers. But if the parts are on an atomic scale, they do not have to move very far.⁴

In some cases, small is the objective. A human cell is big enough to have room for the multitude of molecular machines that make us function. With sufficiently advanced nanotechnology, it ought to be possible to add one more – a cell repair machine. Think of it as a robot submarine that goes into a cell, fixes whatever is wrong, then exits that cell and moves on to the next. If we can build mechanical nanocomputers, it could be a very smart robot submarine.⁵

The human body contains about 100 trillion cells, so fixing all of them with one cell repair machine would take a while. But there is no reason to limit ourselves to one. Or ten. Or a million. Which brings us to another advantage of nanotechnology.

Carbon atoms are all the same (more precisely, Carbon 12 atoms are all the same, but I am going to ignore the complications introduced by isotopes in this discussion). So are nitrogen atoms, hydrogen atoms, iron atoms. Imagine yourself, shrunk impossibly small, building nanomachines. From your point of view, the world is made up of identical interchangeable parts, like tiny Legos. Pick up four identical hydrogens, attach them to one carbon atom, and you have a molecule of methane. Repeat and you have another, perfectly identical.

We cannot shrink you that small, of course, since you yourself are made of atoms and we can’t shrink them. So our first project, once we have the basics of the technology worked out, is to build an assembler. An assembler is a nanoscale machine for building other nanoscale machines. Think of it as a tiny robot – where tiny might mean built out of fewer than a billion atoms. It is small enough so that it can manipulate individual atoms, assembling them into a desired shape. This is far from trivial, since atoms are not really Legos and cannot be manipulated and snapped together in the same way. But we know that assembling atoms into molecules is possible; we, and other living creatures, do it routinely, and some of the molecules we build inside ourselves are very complicated ones.⁶ Organic chemists, with much less detailed control over material than an assembler would have, succeed in deliberately assembling moderately complicated molecules.

Once you have one assembler, you write it a program for building another. Now you have two. Each of them builds another. Four. After 10 doublings you have more than 1,000 assemblers, after 20 more than a million. Now you write a program for building a cell repair machine and set your assemblers to work. Once you have a billion or so cell repair machines you inject them into your body, sit back, and relax. When they are finished you feel like a new person – and are.⁷

It sounds like magic. But consider that your 100 trillion cells started out as one cell and reached their present numbers by an analogous – but much more complicated – process.

ext A friend of mine (Albert R. Hibbs) suggests a very interesting possibility for relatively small machines. He says that, although it is a very wild idea, it would be interesting in surgery if you could swallow the surgeon.

ES Richard Feynman

A cell repair machine would be a very complicated piece of nanotechnology indeed; although we may eventually get such things, it is unlikely to happen very soon. Super strong materials, or medical drugs designed on a computer, one atom at a time, are likely to be earlier applications of the technology. To keep us going while we wait for the cell repair machine, Ralph Merkle proposed and Robert Freitas further developed an ingenious proposal for an improved version of a red blood cell: a nanoscale compressed air tank.⁸ Its advantage becomes clear the day you have a heart attack and your heart stops beating. Instead of dropping dead you pick up the phone, arrange an emergency appointment with your doctor, get in the car and drive there – functioning for several hours on the supply of oxygen already in your bloodstream.

Nanotechnology could be used to construct large objects as well as small ones. It takes a lot of assemblers to do it. But if we start with one assembler, instructions in the form of programs it can read and implement, plenty of atoms of all the necessary sorts and a little time, we can produce a lot of assemblers. With enough assemblers and the software to control them, we can build almost anything. If the idea of a very large object built by molecular machinery strikes you as improbable, consider a whale.

ext It doesn’t cost anything for materials, you see. So I want to build a billion tiny factories, models of each other, which are manufacturing simultaneously, drilling holes, stamping parts, and so on.

ES Richard Feynman

Like most new and unproven technology, nanotech is still controversial, with some authors arguing that the technology is and always will be impossible for a variety of reasons.⁹ The obvious counterexample is life, a functioning nanotechnology based on molecular machines constructed largely of carbon.

One might suppose that, even if nanotechnology does develop, anything really good that it can produce will already have been produced by evolution. Compressed air blood cells would have been useful to us and other living things quite a long time ago, so if the design works why don’t we already have them?

The answer is that although evolution is a powerful design system, it has some important limitations. If a random mutation changes an organism in a way that increases its reproductive success, that mutation will spread through the population; after a while everyone has it, and the next mutation can start from there. So evolution can produce large improvements that occur through a long series of small changes, each itself a small improvement. Evolutionary biologists have actually traced out how complicated organs, such as the eye, are produced through such a long series of small changes.¹⁰

But if a large improvement cannot be produced that way, if you need the right twenty mutations all happening at once in the same organism, evolution is unlikely to do it. The result is that evolution has explored only a small part of the design space, the set of possible ways of assembling atoms to do useful things.¹¹

Human beings also design things by a series of small steps. The F111 did not leap full-grown from the brains of the Wright brothers, and the plane they did produce was powered by an internal combustion engine whose basic design had been invented and improved by others. But what seems a small step to a human thinking out ways of arranging atoms to do something is not necessarily small from the standpoint of a process of random mutation. Hence we would expect that human beings, provided with the tools to build molecular machines, would be able to explore different parts of the design space, to build at least some useful machines that evolution failed to build. Very small compressed air tanks, for example.

Readers interested in arguments for and against the workability of nanotechnology can find and explore them online. For the purposes of this chapter I am going to assume that the fundamental idea of constructing things at the atomic scale using atomic scale assemblers is workable and will, at some point in the next 100 years, happen. That leaves us to consider the world that technology would give us.

SOFTWARE WRIT LARGE

To build a nanotech car I need assemblers – produced in unlimited numbers by other assemblers – raw material, and a program, a full description of what atoms go where. The raw material should be no problem. Dirt is largely aluminum, along with large amounts of silicon, oxygen, possibly carbon, and nitrogen; iron is the fourth most abundant element in the earth’s crust. If I need additional elements that the dirt does not contain, I can always dump in a shovel full of this and that. Add programmed assemblers, stir, and wait for them to find the necessary atoms and arrange them. When they are done I have a ton or two less dirt, a ton or two more car. It sounds like magic – or the process that produces an oak tree.

I have left out one input: energy. An acorn contains design specifications and machinery for building an oak tree, but it needs sunlight to power the process. Similarly, assemblers will need some source of energy. One obvious possibility is chemical energy, disassembling high-energy molecules to get both power and atoms. Perhaps we will have to dump a bucket of alcohol or gasoline on our pile of dirt before we start stirring.

Once we have the basic technology, the hard part is the design; there are a lot of atoms in a car. Fortunately we don’t have to separately calculate the location of each one; once we have the first wheel designed, the others can be copied, and similarly with many other parts. Once we have worked out the atomic structure for a cubic micron or so of our diamond windshield, we can duplicate it over and over for the rest, with a little tweaking of the design when we get to an edge. But even allowing for all plausible redundancy, designing a car, as good a car as the technology permits you to build, is going to be a big project.

I have just described a technology in which most of the cost of producing a product is the cost of the initial design. We already have a technology with those characteristics: software. Producing the first copy of Microsoft Office took an enormous investment of time and effort by a large number of programmers. The second copy used a CD burner, consumed one CDR disk, and cost under a dollar. One implication of nanotechnology is an economy for producing cars very much like the economy that presently produces word-processing programs.

A familiar problem in the software economy is piracy. Not only can Microsoft produce additional copies of MS Office for a dollar apiece, I can do it too. That raises problems for Microsoft or anyone else who expects to be rewarded for producing software with money paid to buy it.

Nanotechnology raises the same problem, although in a somewhat less severe fashion; I cannot simply put my friend’s nanotech car or nanotech computer into a disk drive and burn a copy. I can, however, disassemble it. To do that, I use nanomachines that work like assemblers, but backwards. Instead of starting with a description of where atoms are to go and putting them there, they start with an object – an automobile, say – and remove the atoms, one by one, keeping track of where they all were.

Disassembling an automobile with one disassembler would be a tedious project, but I am not limited to one. Using my army of assemblers I build an army of disassemblers, each provided with some way of getting the information it generates back to me – perhaps a miniature radio transmitter, perhaps some less obvious device. I set them all to work. When they are done the car has been reduced to its constituent elements and a complete design description. If there were computers big enough to design the car, there are computers big enough to store the design. Now I program my assemblers and go into the car business.

One approach to dealing with the problem of copying is an old legal technology, copyright, applied to a new subject matter by suitable amendments to the statute. Having created my design for a car, I copyright it. If you go into business selling duplicates, I sue you for copyright violation. This should work at least a little better for cars than it now does for computer programs, both because the first stage of copying – disassembling, equivalent to reading a computer program from a disk – is a lot harder for cars, and because cars are bigger and harder to hide than programs.

The solution may break down if instead of selling the car the pirate sells the design to individual consumers, each with his own army of assemblers ready to go to work. We are now back in the world of software. Very hard software. The copyright owner has to enforce his rights, copy by copy, against the ultimate consumer, which is a lot harder than enforcing them against someone pirating his property in bulk and selling it.

Suppose that, for these reasons or others, copyright turns out not to do the job. How else might people who design complicated structures at the molecular level get paid for doing so? One possibility is tie-ins with other goods or services that cannot be produced so cheaply – land, say, or backrubs. You download from a (very broad bandwidth) future internet the full specs for building a new sports car, complete with diamond windshield, an engine that burns almost anything and gets 100 miles a gallon, and a combined radar/optical pattern recognition system that warns you of any obstacle within a mile and, if the emergency autopilot is engaged, avoids it. You convert the information into programmed tapes – very small programmed tapes – for your assemblers, find a convenient spot in the backyard, and set them to work. By next morning the car is sitting there in all its splendor.

You get in, turn the key, appreciate the purr of the engine, but are less happy with another feature – the melodious voice telling you everything you didn’t want to know about the lovely housing development completed last week, designed for people just like you. On further investigation, you discover that turning off the advertising is not an option. Neither is disabling it; the audio system is a molecular network spread through the fabric of the car. If you want the car without the advertising you will have to design it yourself. You cast your mind back to the early years of the internet, thirty or forty years ago, and the solution found by web sites to the problem of paying their bills.¹²

Another possibility is a customized car. What you download, this time after paying for it, is a very special car indeed, one of a kind. Before starting, it checks your fingerprints (read from the steering wheel), retinal patterns (scanner above the windshield), and DNA (you’ll never miss a few dead skin cells). If they all match, it runs. The car is safe from thieves, since they cannot start it. You do not even have to carry a key; you are the key. But if you disassemble it and make lots of copies, they will not be very useful to anyone but you. If your neighbor wants a car, he will have to buy his own car, customized to him.¹³

This again is an old solution, although not much used for consumer software. While we do not have adequate biometric identification just yet, the equivalent for computers is fairly easy; all it requires is a CPU with its own serial number. Given that or some equivalent, some identifier specific to a particular computer, it is possible to produce a program that will only run on one machine. One version of this approach uses a hardware dongle – a device not easily copied that attaches to the computer and is recognized by the program.

A third possibility for producing nanotech designs is open source: a network of individuals cooperating to produce and improve designs, motivated by some combination of status, desire for the final product, and whatever else motivated the creators of Linux, Sendmail, and Apache.

As these examples suggest, a mature nanotechnology raises issues similar to those raised by software. Those issues can be dealt with in similar ways – imperfectly, but perhaps well enough.

It also raises other issues of a different, and more disturbing, sort.

THE GRAY GOO SCENARIO

ext “Plants” with “leaves” no more efficient than today’s solar cells could out-compete real plants, crowding the biosphere with an inedible foliage. Tough, omnivorous “bacteria” could out-compete real bacteria: they could spread like blowing pollen, replicate swiftly, and reduce the biosphere to dust in a matter of days. Dangerous replicators could easily be too tough, small, and rapidly spreading to stop – at least if we made no preparation. We have trouble enough controlling viruses and fruit flies.

ES Drexler, Engines of Creation

Life is, on the whole, a good thing – but we are willing to make an exception for certain forms of life, such as smallpox. Molecular machines are, on the whole, a good thing. But there too there might be exceptions.

An assembler is a molecular machine capable of building a wide variety of molecular machines, including copies of itself. It should be much easier to build a machine that copies only itself: a replicator. For proof of concept, consider a virus, a bacterium, or a human being – although the last doesn’t produce an exact copy.

Now consider a replicator designed to build copies of itself, which build copies, which …. Assume it uses only materials readily available in the natural environment, with sunlight as its power supply. Simple calculations suggest that, in a startlingly short time, it could convert everything from the dirt up into copies of itself, leaving only whatever elements happen to be in excess supply. That is what has come to be referred to, in nanotech circles, as the gray goo scenario.

If you happen to be the first one to develop a workable nanotechnology, precautions might be in order. One is to avoid, so far as possible, building replicators. Of course, you will want assemblers, and one of the things an assembler can assemble is another assembler. But at least you can make sure nothing else is designed to replicate – and an assembler, being a large and very complicated molecular machine, should pose less of a threat of going wild than simpler machines whose only design goal is reproduction.

One precaution you could apply to assemblers as well as other replicators is to design them to require some input, whether matter or energy, not available in the natural environment. That way they can replicate usefully under your control but pose no hazard if they get out. Another is to give them a limited lifetime, a counter that keeps track of each generation of copying and turns the machine off when it reaches its preset limit. With precautions like these to supplement the obvious one of keeping your replicators in sealed environments, it might be possible to make sure that no replicator you have designed to be safe poses any serious threat of turning the world into gray goo.

Nanotech replicators, like natural biological replicators, can mutate. A cosmic ray might knock an atom off the instruction tape that controls copying, producing defective copies – and one defect might turn off the limit on number of generations. It might even, although much less probably, somehow eliminate the need for the one element not available in a natural environment. Freed of such constraints, wild nanotech replicators could gradually evolve, just as biological replicators do. Like biological replicators, their evolution would be toward increased reproductive success – getting better and better at converting everything else in existence into copies of themselves. And it is at least possible that, by exploiting design possibilities visible to the humans who designed their ancestors but inaccessible to the continuous processes of evolution, they would do a better job of it than natural replicators.

It should be possible to design replicators, if one is sufficiently clever, with safeguards. One way is through redundancy. You might, for example, give the replicator three copies of its instruction tape and design it to execute an instruction only if all three agree; the odds that three cosmic rays will each remove the same atom from each tape are low. But low is not zero; our cells contain triply redundant safeguards against uncontrolled growth, yet cancer still occurs. So one might also want to make sure that elements not available in the natural environment play a sufficiently central role in the working of the replicator so that there is no plausible way of mutating around the constraint. After designing your replicator and before building it you might want to run it in simulation, using a computer to run through many generations with a very large number of possible changes to see if any of them could break the replicator free of your designed-in controls. Alternatively, you might decide that building replicators is not such a good idea after all.

  Almost Worse Than the Disease

I have described a collection of precautions that could work in a world in which only one organization has access to the tools of nanotechnology and that organization acts in a prudent and benevolent fashion. Is that likely? On the face of it such a monopoly seems extraordinarily unlikely in anything much like our world. But perhaps not. Suppose the idea of nanotechnology is well understood and accepted by a number of organizations with substantial resources, probably governments, at a point well before anyone has succeeded in building an assembler. Each of those organizations engages in extensive computerized design work, figuring out exactly how to build a variety of useful molecular machines once it has the assemblers to build them with. Those machines include designer plagues, engineered obedience drugs, a variety of superweapons, and much else.

One organization makes the breakthrough; it now has an assembler. Very shortly, after about forty doublings, it has a trillion assemblers. It sets them to work building what it has already designed. A week later it rules the world. One of its first acts is to forbid anyone else from doing research in nanotechnology.

It seems a wildly implausible scenario, but I am not sure it is impossible; I do not entirely trust my intuition of what can or cannot happen, given a technology with such extraordinary possibilities. The result would be a world government with very nearly unlimited power. I can see no reason to expect it to behave better than past governments with such power. It would, I suppose, be an improvement on gray goo, but not much of an improvement.

  If you want a picture of the future, imagine a boot stamping on a human face – for ever.

  George Orwell¹⁴

  Between a Rock and a Hard Place

Suppose we avoid world dictatorship and end up instead with multiple independent governments, some of them reasonably free and democratic, and fairly widespread knowledge of nanotechnology. What are the consequences?

One possibility is that everyone treats nanotech as a government monopoly, with the products but not the technology made available to the general public. Eric Drexler has described in some detail a version of this in which everybody is free to experiment with the technology but only in a (nanotechnologically constructed) sealed and inaccessible environment, with the actual implementation of the resulting designs under strict controls.¹⁵ Once the basic information on how to do nanotech is out, the enforcement of such regulations may depend on the government’s lead in the nanotech arms race providing it with devices for surveillance and control that will make the video mosquitoes of an earlier chapter seem primitive. Again not a very attractive picture, but an improvement on all of us turning into gray goo.

The problem with this solution is that it looks very much like a case of setting the fox to guard the hen house. Private individuals may occasionally do research on how to kill people and destroy stuff, but the overwhelming bulk of such work is done by governments for military purposes. The very organizations that, in this version, have control over the development and use of nanotech are the ones most likely to spend substantial resources finding ways of using the technology to do what most of the rest of us regard as bad things.

One possible result is gray goo deliberately designed as a doomsday machine by a government that wants the ability to threaten everyone else with universal suicide. In a less extreme case, we could expect to see a lot of research on designing molecular machines to kill large numbers of (selected) people or destroy large amounts of (other nations’) property. Governments doing military research, while they prefer to avoid killing their own citizens in the process, are willing to take risks – as suggested by incidents such as the accident in a Soviet germ warfare facility that killed several hundred people in a nearby city. And they work in an atmosphere of secrecy that may make it hard for other people to notice and point out risks in their work that have not yet occurred to them. There is a very real possibility that deliberately destructive molecular machines will turn out to be even more destructive than their designers intended or will get released before their designers want them to be.

Consider two possible worlds. In the first, nanotechnology is a difficult and expensive business; it takes billions of dollars of equipment and skilled labor to create and implement workable designs for molecular machines that do useful things. In that world, gray goo is unlikely to be produced deliberately by anybody but a government, and any organization big enough to produce it by accident is probably well enough organized to take precautions. In that world defenses against gray goo – more generally, molecular machines designed to protect human beings and their property from a wide variety of risks, including destructive molecular machines, tailored plagues, and more mundane hazards – will be big sellers, with very large resources devoted to designing them commercially. In that world, making nanotech a government monopoly will do little to reduce the downside risk, since governments will be the main source of that risk, but might substantially reduce the chance of protecting ourselves against it.

In the second world, perhaps the first world a few decades later, nanotech is cheap. Not only can the U.S. Department of Defense design gray goo if it wants to, you can design it too – on your desktop. In this world, nothing much short of a small number of dictatorships maintained in power over rivals and subjects by a lead in the nanotech arms race is going to keep the technology out of the hands of anyone who wants it. And it is far from clear that even that would suffice.

In this second world, the nanotech equivalent of designer plagues will exist for much the same reasons that computer viruses now exist. Some will come into existence the way the original internet worm did, the work of someone very clever, with no bad intent, who makes one mistake too many. Some will be designed to do mischief and turn out to do more mischief than intended. And a few will be deliberately created as instruments of apocalypse by people who for one reason or another like the idea.

Before you conclude that the end of the world is upon you, consider the other side of the technology. With enough cell repair machines on duty, designer plagues may not be a problem. Human beings want to live and will pay for the privilege. The resources that will go into designing protections against threats, nanotechnological or otherwise, will be enormously greater than the (private) resources that go into creating such threats – as they are at present, with the much more limited tools available to us. Unless it turns out that, with this technology, the offense has an overwhelming advantage over the defense, nanotech defenses should almost entirely neutralize the threat from the basement terrorist or careless experimenter. The only serious threat will be from organizations willing and able to spend billions of dollars creating really first-rate molecular killers – almost all of them governments.

The previous paragraph contained a crucial caveat, that offense not be a great deal easier than defense. The gray goo story suggests that it might be, that simple molecular machines designed to turn everything in the environment into copies of themselves might have an overwhelming advantage over their more elaborate opponents.¹⁶

The experiment has been done; the results so far suggest that that is not the case. We live in a world populated by molecular machines. All of them, from viruses up to blue whales, have been designed with the purpose¹⁷ of turning as much of their environment as they can manage into copies of themselves. We call it reproductive success. So far, at least, the simple ones have not turned out to have any overwhelming advantage over the complicated ones: Blue whales, and human beings, are still around.

That does not guarantee safety in a nanotech future. As I pointed out earlier, nanotechnology greatly expands the region of the design space that is accessible; human beings will be able to create things that evolution could not. It is conceivable that, in that expanded space of possible designs, gray goo will turn out to be the winner. All we can say is that so far, in the more restricted space of carbon-based life capable of being produced by evolution, it has not turned out that way.

In dealing with nanotechnology, we are faced with a choice between centralized solutions – in the limit, a world government with a nanotech monopoly – and decentralized solutions. As a general rule I much prefer the latter. But a technology that raises the possibility of a talented teenager producing the end of the world in his basement makes the case for centralized regulation look a lot better than it does in most other contexts, good enough to have convinced some thinkers, among them Eric Drexler, to make at least a partial exception to their usual preference for decentralization, private markets, laissez-faire.

While the case for centralization is in some ways strongest for so powerful a technology, so is the case against. There has been only one occasion in my life when I thought there was a significant chance that many of those near and dear to me might die. It occurred a little after the 9/11 terrorist attack, when I started looking into the subject of smallpox.

Smallpox had been officially eliminated; so far as was publicly known, the only remaining strains of the virus were held by U.S. and Russian government laboratories. Because it had been eliminated, and because public health is a field dominated by governments, smallpox vaccination had been eliminated too. It had apparently not occurred to anybody in a position to do anything about it that it was worth maintaining sufficient backup capacity to reverse that decision quickly. The United States had supplies of vaccine, but they were adequate to vaccinate only a small fraction of the population. So far as I could tell, nobody else had substantial supplies either.

Smallpox, on an unvaccinated population, produces mortality rates as high as 30%. Most of the world’s population is now unvaccinated; those of us who were vaccinated forty or fifty years ago may or may not still be protected. If a terrorist had gotten a sample of the virus, either stolen from a government lab or cultured from the bodies of smallpox victims buried somewhere in the arctic at some time in the past – nobody seems to know for sure whether or not that is possible – he could have used it to kill hundred of millions, perhaps more than a billion, people. That risk existed because the technologies to protect against replicators – that particular class of replicators – had been under centralized control. The center had decided that the problem was solved.

Fortunately, it didn’t happen.