Engines Of CreationThe Coming Era of NanotechnologyFOREWORDby Marvin MinskyK. Eric Drexler's Engines of Creation is an enormouslyoriginal book about the consequences of new technologies.It is ambitious and imaginative and, best of all, thethinking is technically sound.But how can anyone predict where science and technologywill take us? Although many scientists and technologistshave tried to do this, isn't it curious that the mostsuccessful attempts were those of science fiction writerslike Jules Verne and H. G. Wells, Frederik Pohl, RobertHeinlein, Isaac Asimov, and Arthur C. Clarke? Granted,some of those writers knew a great deal about the scienceof their times. But perhaps the strongest source of theirsuccess was that they were equally concerned with thepressures and choices they imagined emerging from theirsocieties. For, as Clarke himself has emphasized, it isvirtually impossible to predict the details of futuretechnologies for more than perhaps half a century ahead.For one thing, it is virtually impossible to predict indetail which alternatives will become technicallyfeasible over any longer interval of time. Why? Simplybecause if one could see ahead that clearly, one couldprobably accomplish those things in much less time given the will to do so. A second problem is that it isequally hard to guess the character of the social changeslikely to intervene. Given such uncertainty, looking

ahead is like building a very tall and slender tower ofreasoning. And we all know that such constructions areuntrustworthy.How could one build a sounder case? First, thefoundations must be very firm - and Drexler has built onthe soundest areas of present-day technical knowledge.Next, one must support each important conclusion step inseveral different ways, before one starts the next. Thisis because no single reason can be robust enough to standbefore so many unknowns. Accordingly, Drexler gives usmultiple supports for each important argument. Finally,it is never entirely safe to trust one's own judgments insuch matters, since all of us have wishes and fears whichbias how we think - without our knowing it. But, unlikemost iconoclasts, Drexler has for many years courageouslyand openly exposed these ideas to both the mostconservative skeptics and the most wishful-thinkingdreamers among serious scientific communities like theone around MIT. He has always listened carefully to whatthe others said, and sometimes changed his viewsaccordingly.Engines of Creation begins with the insight that what wecan do depends on what we can build. This leads to acareful analysis of possible ways to stack atoms. ThenDrexler asks, "What could we build with those atomstacking mechanisms?" For one thing, we could manufactureassembly machines much smaller even than living cells,and make materials stronger and lighter than anyavailable today. Hence, better spacecraft. Hence, tinydevices that can travel along capillaries to enter andrepair living cells. Hence, the ability to heal disease,reverse the ravages of age, or make our bodies speedieror stronger than before. And we could make machines downto the size of viruses, machines that would work atspeeds which none of us can yet appreciate. And then,once we learned how to do it, we would have the option ofassembling these myriads of tiny parts into intelligentmachines, perhaps based on the use of trillions of

nanoscopic parallel-processing devices which makedescriptions, compare them to recorded patterns, and thenexploit the memories of all their previous experiments.Thus those new technologies could change not merely thematerials and means we use to shape our physicalenvironment, but also the activities we would then beable to pursue inside whichever kind of world we make.Now, if we return to Arthur C. Clarke's problem ofpredicting more than fifty years ahead, we see that thetopics Drexler treats make this seem almost moot. Foronce that atom-stacking process starts, then "only fiftyyears" could bring more change than all that had comeabout since near-medieval times. For, it seems to me, inspite of all we hear about modern technologicalrevolutions, they really haven't made such largedifferences in our lives over the past half century. Didtelevision really change our world? Surely less thanradio did, and even less than the telephone did. Whatabout airplanes? They merely reduced travel times fromdays to hours - whereas the railroad and automobile hadalready made a larger change by shortening those traveltimes from weeks to days! But Engines of Creation sets uson the threshold of genuinely significant changes;nanotechnology could have more effect on our materialexistence than those last two great inventions in thatdomain - the replacement of sticks and stones by metalsand cements and the harnessing of electricity. Similarly,we can compare the possible effects of artificialintelligence on how we think - and on how we might cometo think about ourselves - with only two earlierinventions: those of language and of writing.We'll soon have to face some of these prospects andoptions. How should we proceed to deal with them? Enginesof Creation explains how these new alternatives could bedirected toward many of our most vital human concerns:toward wealth or poverty, health or sickness, peace orwar. And Drexler offers no mere neutral catalog ofpossibilities, but a multitude of ideas and proposals for

how one might start to evaluate them. Engines of Creationis the best attempt so far to prepare us to think of whatwe might become, should we persist in making newtechnologies.MARVIN MINSKYDonner Professor of ScienceMassachusetts Institute of Technology-----------------------------Engines Of Construction(Chapter 1)------------------------------Protein engineering . represents the first major steptoward a more general capability for molecularengineering which would allow us to structure matter atomby atom. KEVIN ULMERDirector of Exploratory ResearchGenex CorporationCOAL AND DIAMONDS, sand and computer chips, cancer andhealthy tissue: throughout history, variations in thearrangement of atoms have distinguished the cheap fromthe cherished, the diseased from the healthy. Arrangedone way, atoms make up soil, air, and water; arrangedanother, they make up ripe strawberries. Arranged oneway, they make up homes and fresh air; arranged another,they make up ash and smoke.Our ability to arrange atoms lies at the foundation oftechnology. We have come far in our atom arranging, from

chipping flint for arrowheads to machining aluminum forspaceships. We take pride in our technology, with ourlifesaving drugs and desktop computers. Yet ourspacecraft are still crude, our computers are stillstupid, and the molecules in our tissues still slide intodisorder, first destroying health, then life itself. Forall our advances in arranging atoms, we still useprimitive methods. With our present technology, we arestill forced to handle atoms in unruly herds.But the laws of nature leave plenty of room for progress,and the pressures of world competition are even nowpushing us forward. For better or for worse, the greatesttechnological breakthrough in history is still to come.TWO STYLES OF TECHNOLOGYOur modern technology builds on an ancient tradition.Thirty thousand years ago, chipping flint was the hightechnology of the day. Our ancestors grasped stonescontaining trillions of trillions of atoms and removedchips containing billions of trillions of atoms to maketheir axheads; they made fine work with skills difficultto imitate today. They also made patterns on cave wallsin France with sprayed paint, using their hands asstencils. Later they made pots by baking clay, thenbronze by cooking rocks. They shaped bronze by poundingit. They made iron, then steel, and shaped it by heating,pounding, and removing chips.We now cook up pure ceramics and stronger steels, but westill shape them by pounding, chipping, and so forth. Wecook up pure silicon, saw it into slices, and makepatterns on its surface using tiny stencils and sprays oflight. We call the products "chips" and we consider themexquisitely small, at least in comparison to axheads.

Our microelectronic technology has managed to stuffmachines as powerful as the room-sized computers of theearly 1950s onto a few silicon chips in a pocket-sizedcomputer. Engineers are now making ever smaller devices,slinging herds of atoms at a crystal surface to build upwires and components one tenth the width of a fine hair.These microcircuits may be small by the standards offlint chippers, but each transistor still holds trillionsof atoms, and so-called "microcomputers" are stillvisible to the naked eye. By the standards of a newer,more powerful technology they will seem gargantuan.The ancient style of technology that led from flint chipsto silicon chips handles atoms and molecules in bulk;call it bulk technology. The new technology will handleindividual atoms and molecules with control andprecision; call it molecular technology. It will changeour world in more ways than we can imagine.Microcircuits have parts measured in micrometers - thatis, in millionths of a meter - but molecules are measuredin nanometers (a thousand times smaller). We can use theterms "nanotechnology" and "molecular technology"interchangeably to describe the new style of technology.The engineers of the new technology will build bothnanocircuits and nanomachines.Molecular Technology TodayOne dictionary definition of a machine is "any system,usually of rigid bodies, formed and connected to alter,transmit, and direct applied forces in a predeterminedmanner to accomplish a specific objective, such as theperformance of useful work." Molecular machines fit thisdefinition quite well.

To imagine these machines, one must first picturemolecules. We can picture atoms as beads and molecules asclumps of beads, like a child's beads linked by snaps. Infact, chemists do sometimes visualize molecules bybuilding models from plastic beads (some of which link inseveral directions, like the hubs in a Tinkertoy set).Atoms are rounded like beads, and although molecularbonds are not snaps, our picture at least captures theessential notion that bonds can be broken and reformed.If an atom were the size of a small marble, a fairlycomplex molecule would be the size of your fist. Thismakes a useful mental image, but atoms are really about1/10,000 the size of bacteria, and bacteria are about1/10,000 the size of mosquitoes. (An atomic nucleus,however, is about 1/100,000 the size of the atom itself;the difference between an atom and its nucleus is thedifference between a fire and a nuclear reaction.)The things around us act as they do because of the waytheir molecules behave. Air holds neither its shape norits volume because its molecules move freely, bumping andricocheting through open space. Water molecules sticktogether as they move about, so water holds a constantvolume as it changes shape. Copper holds its shapebecause its atoms stick together in regular patterns; wecan bend it and hammer it because its atoms can slip overone another while remaining bound together. Glassshatters when we hammer it because its atoms separatebefore they slip. Rubber consists of networks of kinkedmolecules, like a tangle of springs. When stretched andreleased, its molecules straighten and then coil again.These simple molecular patterns make up passivesubstances. More complex patterns make up the activenanomachines of living cells.Biochemists already work with these machines, which arechiefly made of protein, the main engineering material ofliving cells. These molecular machines have relatively

few atoms, and so they have lumpy surfaces, like objectsmade by gluing together a handful of small marbles. Also,many pairs of atoms are linked by bonds that can bend orrotate, and so protein machines are unusually flexible.But like all machines, they have parts of differentshapes and sizes that do useful work. All machines useclumps of atoms as parts. Protein machines simply usevery small clumps.Biochemists dream of designing and building such devices,but there are difficulties to be overcome. Engineers usebeams of light to project patterns onto silicon chips,but chemists must build much more indirectly than that.When they combine molecules in various sequences, theyhave only limited control over how the molecules join.When biochemists need complex molecular machines, theystill have to borrow them from cells. Nevertheless,advanced molecular machines will eventually let thembuild nanocircuits and nanomachines as easily anddirectly as engineers now build microcircuits or washingmachines. Then progress will become swift and dramatic.Genetic engineers are already showing the way.Ordinarily, when chemists make molecular chains - called"polymers" - they dump molecules into a vessel where theybump and snap together haphazardly in a liquid. Theresulting chains have varying lengths, and the moleculesare strung together in no particular order.But in modern gene synthesis machines, genetic engineersbuild more orderly polymers - specific DNA molecules - bycombining molecules in a particular order. Thesemolecules are the nucleotides of DNA (the letters of thegenetic alphabet) and genetic engineers don't dump themall in together. Instead, they direct the machine to adddifferent nucleotides in a particular sequence to spellout a particular message. They first bond one kind ofnucleotide to the chain ends, then wash away the leftovermaterial and add chemicals to prepare the chain ends tobond the next nucleotide. They grow chains as they bond

on nucleotides, one at a time, in a programmed sequence.They anchor the very first nucleotide in each chain to asolid surface to keep the chain from washing away withits chemical bathwater. In this way, they have a bigclumsy machine in a cabinet assemble specific molecularstructures from parts a hundred million times smallerthan itself.But this blind assembly process accidentally omitsnucleotides from some chains. The likelihood of mistakesgrows as chains grow longer. Like workers discarding badparts before assembling a car, genetic engineers reduceerrors by discarding bad chains. Then, to join theseshort chains into working genes (typically thousands ofnucleotides long), they turn to molecular machines foundin bacteria.These protein machines, called restriction enzymes,"read" certain DNA sequences as "cut here." They readthese genetic patterns by touch, by sticking to them, andthey cut the chain by rearranging a few atoms. Otherenzymes splice pieces together, reading matching parts as"glue here" - likewise "reading" chains by selectivestickiness and splicing chains by rearranging a fewatoms. By using gene machines to write, and restrictionenzymes to cut and paste, genetic engineers can write andedit whatever DNA messages they choose.But by itself, DNA is a fairly worthless molecule. It isneither strong like Kevlar, nor colorful like a dye, noractive like an enzyme, yet it has something that industryis prepared to spend millions of dollars to use: theability to direct molecular machines called ribosomes. Incells, molecular machines first transcribe DNA, copyingits information to make RNA "tapes." Then, much as oldnumerically controlled machines shape metal based oninstructions stored on tape, ribosomes build proteinsbased on instructions stored on RNA strands. And proteinsare useful.

Proteins, like DNA, resemble strings of lumpy beads. Butunlike DNA, protein molecules fold up to form smallobjects able to do things. Some are enzymes, machinesthat build up and tear down molecules (and copy DNA,transcribe it, and build other proteins in the cycle oflife). Other proteins are hormones, binding to yet otherproteins to signal cells to change their behavior.Genetic engineers can produce these objects cheaply bydirecting the cheap and efficient molecular machineryinside living organisms to do the work. Whereas engineersrunning a chemical plant must work with vats of reactingchemicals (which often misarrange atoms and make noxiousbyproducts), engineers working with bacteria can makethem absorb chemicals, carefully rearrange the atoms, andstore a product or release it into the fluid around them.Genetic engineers have now programmed bacteria to makeproteins ranging from human growth hormone to rennin, anenzyme used in making cheese. The pharmaceutical companyEli Lilly (Indianapolis) is now marketing Humulin, humaninsulin molecules made by bacteria.Existing Protein MachinesThese protein hormones and enzymes selectively stick toother molecules. An enzyme changes its target'sstructure, then moves on; a hormone affects its target'sbehavior only so long as both remain stuck together.Enzymes and hormones can be described in mechanicalterms, but their behavior is more often described inchemical terms.But other proteins serve basic mechanical functions. Somepush and pull, some act as cords or struts, and parts ofsome molecules make excellent bearings. The machinery ofmuscle, for instance, has gangs of proteins that reach,

grab a "rope" (also made of protein), pull it, then reachout again for a fresh grip; whenever you move, you usethese machines. Amoebas and human cells move and changeshape by using fibers and rods that act as molecularmuscles and bones. A reversible, variable-speed motordrives bacteria through water by turning a corkscrewshaped propeller. If a hobbyist could build tiny carsaround such motors, several billions of billions wouldfit in a pocket, and 150-lane freeways could be builtthrough your finest capillaries.Simple molecular devices combine to form systemsresembling industrial machines. In the 1950s engineersdeveloped machine tools that cut metal under the controlof a punched paper tape. A century and a half earlier,Joseph-Marie Jacquard had built a loom that wove complexpatterns under the control of a chain of punched cards.Yet over three billion years before Jacquard, cells haddeveloped the machinery of the ribosome. Ribosomes areproof that nanomachines built of protein and RNA can beprogrammed to build complex molecules.Then consider viruses. One kind, the T4 phage, acts likea spring-loaded syringe and looks like something out ofan industrial parts catalog. It can stick to a bacterium,punch a hole, and inject viral DNA (yes, even bacteriasuffer infections). Like a conqueror seizing factories tobuild more tanks, this DNA then directs the cell'smachines to build more viral DNA and syringes. Like allorganisms, these viruses exist because they are fairlystable and are good at getting copies of themselves made.Whether in cells or not, nanomachines obey the universallaws of nature. Ordinary chemical bonds hold their atomstogether, and ordinary chemical reactions (guided byother nanomachines) assemble them. Protein molecules caneven join to form machines without special help, drivenonly by thermal agitation and chemical forces. By mixingviral proteins (and the DNA they serve) in a test tube,

molecular biologists have assembled working T4 viruses.This ability is surprising: imagine putting automotiveparts in a large box, shaking it, and finding anassembled car when you look inside! Yet the T4 virus isbut one of many self-assembling structures. Molecularbiologists have taken the machinery of the ribosome apartinto over fifty separate protein and RNA molecules, andthen combined them in test tubes to form workingribosomes again.To see how this happens, imagine different T4 proteinchains floating around in water. Each kind folds up toform a lump with distinctive bumps and hollows, coveredby distinctive patterns of oiliness, wetness, andelectric charge. Picture them wandering and tumbling,jostled by the thermal vibrations of the surroundingwater molecules. From time to time two bounce together,then bounce apart. Sometimes, though, two bounce togetherand fit, bumps in hollows, with sticky patches matching;they then pull together and stick. In this way proteinadds to protein to make sections of the virus, andsections assemble to form the whole.Protein engineers will not need nanoarms and nanohands toassemble complex nanomachines. Still, tiny manipulatorswill be useful and they will be built. Just as today'sengineers build machinery as complex as player pianos androbot arms from ordinary motors, bearings, and movingparts, so tomorrow's biochemists will be able to useprotein molecules as motors, bearings, and moving partsto build robot arms which will themselves be able tohandle individual molecules.Designing With ProteinHow far off is such an ability? Steps have been taken,but much work remains to be done. Biochemists have

already mapped the structures of many proteins. With genemachines to help write DNA tapes, they can direct cellsto build any protein they can design. But they stilldon't know how to design chains that will fold up to makeproteins of the right shape and function. The forces thatfold proteins are weak, and the number of plausible waysa protein might fold is astronomical, so designing alarge protein from scratch isn't easy.The forces that stick proteins together to form complexmachines are the same ones that fold the protein chainsin the first place. The differing shapes and kinds ofstickiness of amino acids - the lumpy molecular "beads"forming protein chains - make each protein chain fold upin a specific way to form an object of a particularshape. Biochemists have learned rules that suggest how anamino acid chain might fold, but the rules aren't veryfirm. Trying to predict how a chain will fold is liketrying to work a jigsaw puzzle, but a puzzle with nopattern printed on its pieces to show when the fit iscorrect, and with pieces that seem to fit together aboutas well (or as badly) in many different ways, all but oneof them wrong. False starts could consume many lifetimes,and a correct answer might not even be recognized.Biochemists using the best computer programs nowavailable still cannot predict how a long, naturalprotein chain will actually fold, and some of them havedespaired of designing protein molecules soon.Yet most biochemists work as scientists, not asengineers. They work at predicting how natural proteinswill fold, not at designing proteins that will foldpredictably. These tasks may sound similar, but theydiffer greatly: the first is a scientific challenge, thesecond is an engineering challenge. Why should naturalproteins fold in a way that scientists will find easy topredict? All that nature requires is that they in factfold correctly, not that they fold in a way obvious topeople.

Proteins could be designed from the start with the goalof making their folding more predictable. Carl Pabo,writing in the journal Nature, has suggested a designstrategy based on this insight, and some biochemicalengineers have designed and built short chains of a fewdozen pieces that fold and nestle onto the surfaces ofother molecules as planned. They have designed fromscratch a protein with properties like those of melittin,a toxin in bee venom. They have modified existingenzymes, changing their behaviors in predictable ways.Our understanding of proteins is growing daily.In 1959, according to biologist Garrett Hardin, somegeneticists called genetic engineering impossible; today,it is an industry. Biochemistry and computer-aided designare now exploding fields, and as Frederick Blattner wrotein the journal Science, "computer chess programs havealready reached the level below the grand master. Perhapsthe solution to the protein-folding problem is nearerthan we think." William Rastetter of Genentech, writingin Applied Biochemistry and Biotechnology asks, "How faroff is de novo enzyme design and synthesis? Ten, fifteenyears?" He answers, "Perhaps not that long."Forrest Carter of the U.S. Naval Research Laboratory, AriAviram and Philip Seiden of IBM, Kevin Ulmer of GenexCorporation, and other researchers in university andindustrial laboratories around the globe have alreadybegun theoretical work and experiments aimed atdeveloping molecular switches, memory devices, and otherstructures that could be incorporated into a proteinbased computer. The U.S. Naval Research Laboratory hasheld two international workshops on molecular electronicdevices, and a meeting sponsored by the U.S. NationalScience Foundation has recommended support for basicresearch aimed at developing molecular computers. Japanhas reportedly begun a multimillion-dollar program aimedat developing self-assembling molecular motors andcomputers, and VLSI Research Inc., of San Jose, reportsthat "It looks like the race to bio-chips [another term

for molecular electronic systems] has already started.NEC, Hitachi, Toshiba, Matsushita, Fujitsu, Sanyo-Denkiand Sharp have commenced full-scale research efforts onbio-chips for bio-computers."Biochemists have other reasons to want to learn the artof protein design. New enzymes promise to perform dirty,expensive chemical processes more cheaply and cleanly,and novel proteins will offer a whole new spectrum oftools to biotechnologists. We are already on the road toprotein engineering, and as Kevin Ulmer notes in thequote from Science that heads this chapter, this roadleads "toward a more general capability for molecularengineering which would allow us to structure matter atomby atom."Second-Generation NanotechnologyDespite its versatility, protein has shortcomings as anengineering material. Protein machines quit when dried,freeze when chilled, and cook when heated. We do notbuild machines of flesh, hair, and gelatin; over thecenturies, we have learned to use our hands of flesh andbone to build machines of wood, ceramic, steel, andplastic. We will do likewise in the future. We will useprotein machines to build nanomachines of tougher stuffthan protein.As nanotechnology moves beyond reliance on proteins, itwill grow more ordinary from an engineer's point of view.Molecules will be assembled like the components of anerector set, and well-bonded parts will stay put. Just asordinary tools can build ordinary machines from parts, somolecular tools will bond molecules together to make tinygears, motors, levers, and casings, and assemble them tomake complex machines.

Parts containing only a few atoms will be lumpy, butengineers can work with lumpy parts if they have smoothbearings to support them. Conveniently enough, some bondsbetween atoms make fine bearings; a part can be mountedby means of a single chemical bond that will let it turnfreely and smoothly. Since a bearing can be made usingonly two atoms (and since moving parts need have only afew atoms), nanomachines can indeed have mechanicalcomponents of molecular size.How will these better machines be built? Over the years,engineers have used technology to improve technology.They have used metal tools to shape metal into bettertools, and computers to design and program bettercomputers. They will likewise use protein nanomachines tobuild better nanomachines. Enzymes show the way: theyassemble large molecules by "grabbing" small moleculesfrom the water around them, then holding them together sothat a bond forms. Enzymes assemble DNA, RNA, proteins,fats, hormones, and chlorophyll in this way - indeed,virtually the whole range of molecules found in livingthings.Biochemical engineers, then, will construct new enzymesto assemble new patterns of atoms. For example, theymight make an enzyme-like machine which will add carbonatoms to a small spot, layer on layer. If bondedcorrectly, the atoms will build up to form a fine,flexible diamond fiber having over fifty times as muchstrength as the same weight of aluminum. Aerospacecompanies will line up to buy such fibers by the ton tomake advanced composites. (This shows one small reasonwhy military competition will drive molecular technologyforward, as it has driven so many fields in the past.)But the great advance will come when protein machines areable to make structures more complex than mere fibers.These programmable protein machines will resembleribosomes programmed by RNA, or the older generation ofautomated machine tools programmed by punched tapes. They

will open a new world of possibilities, letting engineersescape the limitations of proteins to build rugged,compact machines with straightforward designs.Engineered proteins will split and join molecules asenzymes do. Existing proteins bind a variety of smallermolecules, using them as chemical tools; newly engineeredproteins will use all these tools and more.Further, organic chemists have shown that chemicalreactions can produce remarkable results even withoutnanomachines to guide the molecules. Chemists have nodirect control over the tumbling motions of molecules ina liquid, and so the molecules are free to react in anyway they can, depending on how they bump together. Yetchemists nonetheless coax reacting molecules to formregular structures such as cubic and dodecahedralmolecules, and to form unlikely-seeming structures suchas molecular rings with highly strained bonds. Molecularmachines will have still greater versatility inbondmaking, because they can use similar molecularmotions to make bonds, but can guide these motions inways that chemists cannot.Indeed, because chemists cannot yet direct molecularmotions, they can seldom assemble complex moleculesaccording to specific plans. The largest molecules theycan make with specific, complex patterns are all linearchains. Chemists form these patterns (as in genemachines) by adding molecules in sequence, one at a time,to a growing chain. With only one possible bonding siteper chain, they can be sure to add the next piece in theright place.But if a rounded, lumpy molecule has (say) a hundredhydrogen atoms on its surface, how can chemists split offjust one particular atom (the one five up and threeacross from the bump on the front) to add something inits place? Stirring simple chemicals together will seldomdo the job, because small molecules can seldom se

like Jules Verne and H. G. Wells, Frederik Pohl, Robert Heinlein, Isaac Asimov, and Arthur C. Clarke? Granted, some of those writers knew a great deal about the science of their times. But perhaps the strongest source of their success was that they were equally concerned with the pressures and choices they imagined emerging from their societies.