The Ribosome Already Solved Digital Manufacturing

Asked what he has learned at the boundary of bits and atoms, Gershenfeld answers with three claims that sound, he admits, presumptuous: why von Neumann and Turing made fundamental mistakes, the secret of life, and how to solve many of the world's most important problems. He could ask people who had been in the room. Andrew Gleason had been Turing's American counterpart; Marvin Minsky had been a student of von Neumann. Neither computer-science hero, Gershenfeld argues, intended the architecture that now bears their names.

The Turing machine has a head that moves, reads, and writes a tape. The head is distinct from the tape, which means persistence of information is separate from interaction with it. Von Neumann then wrote what Gershenfeld calls a horrible memo, the First Draft of a Report on the EDVAC, that roughly built Turing's split into a machine. The computer someone is using to watch the interview spends much of its effort moving information from storage transistors to processing transistors, even though they have the same computational complexity.

Computer science, he says, teaches a ridiculous taxonomy of about a hundred models of computation. They are all fictions. In physics, a patch of space occupies space, stores state, takes time to transit, and can be interacted with. That is the only physical model. He felt this most clearly after a keynote for the supercomputer industry, walking the exhibit floor: like *Metropolis*, people frolic in the garden upstairs while someone moves levers in the basement. Software pretends it is not physical. Every scaling problem in computing, and every opportunity, lives on that boundary.

Turing and von Neumann both knew. The canon credits them for an architecture neither meant to leave as the last word. Both ended their lives studying how software becomes hardware: von Neumann on self-reproducing automata, Turing on morphogenesis—how genes give rise to form. Working that boundary, Gershenfeld's lab took part in early faster-than-classical quantum computations and in creating a minimal synthetic organism. In both domains, hardware and software will not come apart.

In high school Gershenfeld wanted vocational school: welding, cars, houses. He was told no, he was smart, he had to sit in a room. At Bell Labs before deregulation he walked into the workshop to make something and drew union grievances: no, he was smart, he had to tell somebody what to do. He did not understand the prohibition until MIT, and traces it to the Renaissance. Liberal arts meant the path to liberation—the trivium and quadrivium, language and natural science. Beside them appeared the illiberal arts: anything for commercial gain, "just making stuff," unfit for serious study. Micromachining and embedded coding, he says, are as expressive as a painting or a sonnet. The means of expression have changed. The status hierarchy has not.

The path into the Center for Bits and Atoms ran through music. As a junior fellow at Harvard he visited MIT through Minsky because he cared about the physics of instruments. At Cornell he crossed the street from physics to play bassoon, trimming reeds that were beautiful until they got soggy. In the music-department basement was David Borden, a pioneer of electronic music; Bob Moog, another Cornell physics student who crossed the same street, had been kicked out and invented the synthesizer. That pairing posed a question: what is the computational capacity of a musical instrument?

Minsky introduced him to Tod Machover at the Media Lab, who was starting a project with Yo-Yo Ma. They instrumented a cello to pull Ma's playing into computational environments. Modeling the physics at relevant time and length scales is no longer heroic—a good GPU can do what used to take a national supercomputer. Ma himself was unsentimental about the Stradivarius. The magic was not ineffable wiggles in the wood. It was the cello as a controller: resolution and bandwidth between the body and sound. Instrument everything he does, connect it to almost anything, and it still sounds like Yo-Yo. Asked what it would take to put the Strad away, Ma said logistics. The rig was a rack of electronics, cables, and graduate students. Once the technology became as invisible as the instrument, he would take it.

An accident in that sensing became a hundred-million-dollar-a-year airbag business. To watch the bow without touching it, Gershenfeld set up local electromagnetic fields. Ma's hand kept registering instead of the bow. His first graduate student, Josh Smith, wrote a thesis on seeing in three dimensions with electric fields. Through Machover and Joe Paradiso that work became a Penn and Teller trick in Las Vegas, a séance for Houdini. Then Phil Rittmueller ran into the lab: airbags were killing infants in rear-facing seats. A car needed to tell a front-facing adult from groceries from a baby. They took the cello-and-magic sensing to an auto show; the car companies asked when they could buy it. ELESYS became a leading occupancy sensor, with almost no publicity, because the pitch is that the car does not kill you.

Gershenfeld had expected to take over a lab at IBM Research, among people he calls gods of the foundations of computing. At the last minute he pivoted to MIT's Media Lab, which is famous for Nicholas Negroponte. Less famous is Jerome Wiesner: MIT president, Kennedy science advisor, frustrated at the end of his life by how knowledge was segregated. Wiesner wanted a department of none of the above. As president he could not announce that without a revolt, so—while attention was on Negroponte—he hid Media Arts and Sciences inside the Media Lab. Students who helped create quantum computing or synthetic life still get that degree. That explanation, later confirmed by Negroponte, is why Gershenfeld came.

With Machover, Paradiso, and Mike Hawley he started Things That Think, around the birth of the Internet of Things and RFID. The work then stretched into quantum computing, cryptography, materials, logic, and microfluidics, which needed heavier infrastructure and longer arcs. About twenty people wrote an NSF proposal that was, roughly, one of every tool to make anything of any size. Nanometers, micrometers, millimeters, and meters are usually segregated, as are inputs and outputs. They wanted to look, literally, at how digital becomes physical and physical becomes digital. NSF funded it on a good day. With Joe Jacobson, Isaac Chuang, and Scott Manalis, they launched the Center for Bits and Atoms.

A well-equipped research lab has electron microscopes and focused-ion-beam probes for nanostructures, laser micromachining and X-ray microtomography for microstructures, multi-axis machining and 3D printing for macrostructures. The difference is that those tools usually live in different buildings, run by technicians, billed to accounts, used when you already know what you are doing. CBA is for inchoate work that has to cross length scales. Once a project is running, there is no charge for time and no formal proposal to schedule; users run the tools. Today's range, he says, goes from zeptojoule electronics for lowest-power computing, to diamond micromachined for ten-million-RPM bearings, to robots meant to build hundred-meter structures in space.

To explain the biggest things the lab has tried to build, Gershenfeld brings in Claude Shannon. Digital does not mean zero and one. The deeper answer is in what he calls the best master's thesis ever: Shannon, annoyed by Vannevar Bush's differential analyzer—a room of gears and pulleys whose answer got worse the longer it ran—invented modern digital logic. At Bell Labs he asked not how to make a phone line a little better but where the limit was. The threshold theorem for channel capacity says that above a noise threshold you are doomed; below it, a linear increase in energy drops the error rate exponentially. That is why unreliable things can work reliably, and why a podcast arrives as packets that can be corrected rather than as a sound wave that only gets worse with distance.

Von Neumann applied the same restoring of state to computation, showing an unreliable computer can run reliably. The result was forgotten for years and had to be rediscovered in quantum computing, when devices were unreliable again. Fabrication still has not learned it. MIT's 1952 numerically controlled mill—Gershenfeld notes the messy history, MIT effectively taking the idea from an inventor who wanted an Air Force partnership—put information in the control computer, not in the materials. 3D printing, including Formlabs, a billion-dollar spinout from his lab by Max Lobovsky, is the same analog story. Make a mistake and it accumulates, like Bush's gears.

The real birth of computerized digital manufacturing, he says, is the evolutionary age of the ribosome. A genetic code goes to a molecular factory that assembles about twenty amino acids into you. Mix chemicals and the error rate is about one in a hundred; elongate a protein in the ribosome and it is about one in 10^4; replicate DNA with an extra layer of correction and it is one in 10^8. You do not need a ruler. Geometry comes from the parts. A child snapping LEGO builds a tower more accurate than the child's motor control, can join dissimilar materials, and produces no LEGO trash, because the parts contain enough information to come apart. Those are the properties of a digital code, embodied not for communication or computation but for construction.

A digital material, in the lab's precise sense, is a discrete set of parts, reversibly joined, with global geometry determined by local constraints. Aerospace brought the first large test. Spirit AeroSystems, Boeing's factories, asked how to join composite wings and fuselages, because joints were where things failed. Instead of a few huge parts, the group made little loops of carbon fiber and linked them in a geometry that balances under-constraint and over-constraint. They set a world record for the highest-modulus ultralight material: carbon-fiber LEGO. A few part types could tune the properties. Then came robots that walk on the cellular structure, error-correct their position on it, and navigate on it, so you do not need a tool the size of a jumbo jet to make a jumbo jet.

With NASA they made morphing airplanes; former students Kenny Cheung and Ben Jenett built one the size of NASA Langley's biggest wind tunnel. With Toyota they made ultra-efficient race cars. They are talking with NASA about space telescopes and habitats. The analogy is the ribosome making an elephant one molecule at a time. Ribosomes are slow, about one molecule a second, but ribosomes make ribosomes, so you get trillions. Likewise the little assembly robots can make giant structures because the robot can make the robot. Two students, Amira and Miana, published in *Nature Communications* a robot made from the parts it is making.

What that robot looks like depends on length scale. Biology's hierarchy is the template: primary structure is the messenger-RNA code; secondary is motifs such as helices and sheets; tertiary is functional elements; quaternary is molecular motors. Electronics vendors stock hundreds of thousands of resistor, capacitor, and transistor types. All of biology is twenty amino acids. The lab is trying to create the engineering equivalent: all of technology from about twenty parts. Early on Gershenfeld tried to climb from the bottom to the top in a straight line and decided that was a bad idea. They now work the layers in parallel and let them grow together. 3D-printing a house with a printer the size of the house is hype he would rather replace with swarms of table-scale robots walking the structure to place parts.

Complaining about "the von Neumann architecture," Gershenfeld adds, is unfair to von Neumann, who never claimed it. The late work is the one he wants remembered, and he insists on the wording: how a computation communicates its own construction. A computation can store a description of how to build itself. The hard problem is transferring that description and waking up a thing that can contain it—giving birth to something that knows how to make itself. With Stanislaw Ulam, von Neumann invented cellular automata as a simulation. The lab's stack, from Will Langford's microrobots up through Miana's larger ones, is an attempt to realize the idea in hardware, at the intersection of communication, computation, and fabrication.

Intellectually, self-reproducing automata matter because they are the foundation of life: how to create life in non-living material. Technologically they matter because that is how you scale capacity. Assemblers make assemblers. He then puts numbers on the current AI boom. A 100-petaflop supercomputer does about 10^17 operations per second; so does a brain, with roughly 10^15 synapses running at about 100 hertz. A year or two before the interview, big computers matched brains in that crude sense, with about 10^15 transistors in processors and as many in memory. If they could not do roughly the same things, he says, that would be derelict.

The comparison that actually interests him is construction speed. The chip fab making the supercomputer places about 10^10 transistors a second. Digestion places about 10^18 parts a second. Eight orders of magnitude. Computational capacity has caught up. The remaining hole is the rate at which biology can build versus the rate of state-of-the-art manufacturing. Closing it is not a better analog printer. It is digital fabrication in the deep sense: a description does not describe a thing; the description becomes the thing. Communication had its Moore's Law, computation had its analogue, and he expects fabrication to be the third.

Casual digital fabrication is a computer controlling a tool, MIT's 1952 mill. Deep digital fabrication is what the ribosome does, the path to a *Star Trek* replicator, and it does not exist yet. The roadmap, he says, runs through fab labs. With colleagues he accidentally started a network that was then about 2,500 digital-fabrication community labs in 125 countries, doubling every year and a half—Lass's Law, after Sherry Lassiter. CBA had so many machines that learning them would take a lifetime of classes, so they offered a modestly named course: How to Make Almost Anything, aimed at a few research students. They were unprepared. Hundreds try to take it every year. Students asked whether MIT was allowed to teach something that seemed too useful.

The projects were not industrial demos. Kelly Dobson, a sculptor with no engineering background, built a device that swallows a scream when you are angry and plays it back later. Mejin made a dress with sensors and spines that defends personal space when someone creepy comes close. Someone built a web browser for parrots. Someone built an alarm clock you have to wrestle to prove you are awake. Gershenfeld connects that to computing history at MIT: Whirlwind, the first real-time computer, transistorized as TX, spun out as the PDP minicomputers that helped create the internet. DEC, Prime, Wang, Data General sat outside MIT and then failed when computing became personal. Ken Olsen of Digital famously said you do not need a computer at home.

He had been asking how to do digital fabrication, not why. The students answered. Kelly did not make the scream-saver for a thesis, a paper, or a business model. She wanted one. That is personal expression in new materials, the thing vocational school had been for. The course is now taught across multiple MIT labs, with as many people teaching as taking it, and even a Harvard lab serving the MIT class. The deep research on self-replicating robots and Kelly's box sit on the same line: one end tests the physics of manufacturing, the other asks why it is so hard for a person who wants a thing to have a thing.

Another MIT mentor, Seymour Papert, had studied with Piaget and come to campus to let children play with early computers. Piaget had shown that kids learn like scientists until it is throttled out of them. Papert's line, with Mitch Resnick, led to Logo, LEGO Mindstorms, and the rest. As fab labs spread into education, Papert made a gesture Gershenfeld still quotes: the Logo turtle connected to a mainframe had been a thorn, because the goal was never for kids to program the robot. It was for kids to create the robot. Fab labs, which for Gershenfeld had been an accident, were the fulfillment of that arc.

Someone once added up MIT spinouts and ranked them as the world's tenth economy, between India and Russia. Gershenfeld treats that as a bad number: it is only a few thousand people, not uniquely the brightest on earth, just a productive environment. In rural Indian villages, African shantytowns, and Arctic hamlets he finds the same profile. Hours above Tromsø, so far north the satellite dishes look at the ground, a student named Hans Christian was considered unteachable; after a few project examples he was designing robot vehicles. In an apartheid-era South African township, the technical institute taught kids to make bricks and fold sheets as punishment; in the fab lab, Chapiso was doing the work of the MIT class. Historically the answer to that kid was: you are smart, go away. The lab network can bring the world to them instead.

He maps the scaling onto computing. One Earth, a thousand cities, a million towns, a billion people, a trillion things. One Whirlwind; thousands of PDPs; millions of hobbyist machines; billions of PCs; trillions of IoT devices. 1952 was the NC mill; now there are thousands of fab labs, at about the cost and complexity of a minicomputer—a room, not a pocket, but already the scale of a work group rather than a corporation. Email, video games, and word processing were invented in that era. Fab labs are inventing how the world works if anybody can make anything. They are now moving from buying machines to machines making machines, the hobbyist-computer stage. Assemblers and self-assemblers are the smartphone and IoT stage. The warning from computing is that the core change happened in the minicomputer transition; later exponential growth did not rewrite the idea. Close your eyes, treat the whole fab lab as one machine, and you can already make almost anything—if you accept a lot of inputs. Bit by bit the inputs and the room will shrink.

How hard is a self-replicating assembler? Students including Nadya Peek and Jonathan Ward started using the lab's tools to make the lab's tools. Around the world—Danielle in Germany, Jens in Norway—people now walk into a fab lab and make a machine. Super fab labs in places such as Bhutan and Kerala add the precision tools that ordinary labs cannot, so machines get cheaper. That is self-reproduction with a caveat: you still feed the system bearings and microcontrollers. He numbers the stack as Fab 1 through 4: computers controlling machines, machines making machines, assemblers, self-assemblers. They are moving from one to two; three and four are the research. Sustainability of feedstocks is already part of Fab 2: a colleague in Chile, Alicia, leads work on turning forest products, coffee grounds, and seashells into the high-tech materials the lab needs.

In the lab they already have robots that build structures and assemble more such robots, finer robots that build micromechanical walkers, and a project called DICE, discrete assembly of integrated electronics. The smallest commercial discrete transistor is now the size of an early integrated-circuit transistor, so you can imagine placing parts instead of running a billion-dollar chip fab. He has a nearly unclaimed student prize: a thesis that can walk out of the printer, carrying the description of how it did so. The vision is billions of robots at many scales. The "but" is gray goo. He does not worry about it. To take over the world, mutant assemblers would have to compete with biology for water and sunlight, and biology already knows everything he has described. What is new is doing it in non-living materials. If the machine can make the machine, it is also not a great centralized business. The economic win of the digital era was platforms; this one, he argues, is distributed.

Weapons are a well-met market. Hobbyists have made guns for ages; a fab-lab gun today is not a very good gun, and guns are already easy to get. Labs in war zones, he says, are used as an alternative to making weapons. Trash, by contrast, is an analog concept. There is no trash in a forest, and none in a LEGO room, because the parts know how to come apart. The real technological change is not additive versus subtractive printing. It is assembling and disassembling, collapsing inventories and supply chains, first interesting for satellites at the end of long logistics, later for eliminating technical trash. He does not expect to print a whole cell phone on one printer. He does expect to assemble one.

The fear he takes seriously is biological. A fab lab can roughly make a bio lab; How to Grow Almost Anything teaches skills that can make serious biological threats, and large language models plus protein-folding tools tighten the design loop. Unlike gray goo, "somebody could learn to make a virus" is something he really worries about. Command-and-control—copier serial numbers, even Asilomar-style research management—fails when anybody can do the work anywhere. You cannot contain it or forbid it. What they have found useful is incentives for transparency: do it in the light, with help and community, rather than in a basement. He offers that not as dewy-eyed theory but as years of watching communities. The intelligence-community anecdote he prefers is a briefing that ended with a clip of Valentina in Ghana making surface-mount electronics. One official was livid at the waste of time; another yelled that nothing was more important, because a girl who can master that work changes assumptions about conflict, migration, and the economy.

Assume the machines-making-machines transition succeeds. You still have to design unimaginable complexity. Nowhere in the genome is the number five. HOX genes and other morphogenes store a developmental program—grow up a gradient, break symmetry—and at the end you have five fingers: a growth plan, not a body plan. One reason is compression. The deeper one is that random genome edits are fatal or boring, while edits to developmental programs take you from swimming webs to fingers, or from walking to wings. That is a search space in which search is interesting. AI's recent success, he says, is less a new way to search than good representations for search. Developmental programs are molecular intelligence, as profound as the cognition in a brain.

Once self-reproducing assemblers place 10^18 parts, you cannot "design the phone" as a representation of the phone. You have to learn to evolve, in the morphogenetic sense, not by trivial random mutation. He does not start from copying biology; he starts from problems and rediscovers it. Valentina in Ghana still needs a distant chip fab for the processor on her board. For her to make that processor locally you need the deep stack. David Gingery's books take you from a charcoal furnace to a machine shop, a personal industrial revolution; NASA's in-situ resource utilization has essentially assumed that path, recreating a hundred thousand resistor types. What they are finding is that a technological civilization's minimum kit is about twenty properties—conducting, insulating, semiconducting, magnetic, dielectric, flexural—enough for logic and actuation, hence microrobots, hence bigger robots, hence structures. Amino acids are not extreme; they are good enough to compose. His favorite sensor is a step response on two electrodes: from user interfaces to chemistry, depending on frequency.

Where do the ideas come from? Vannevar Bush's *Endless Frontier* taught a pipeline from basic research to impact. Gershenfeld calls that a mistake, because almost every example in the conversation ran backwards: basic research from applications, and from failures. Ready, aim, fire hits what you aim at. He prefers ready, fire, aim: do the homework, close your eyes, then look very carefully at where you actually pointed. Early quantum computing is the exhibit. Short-term work on shoplifting tags, before modern RFID, needed nonlinearity to distinguish many objects. Nuclear spins were a bad idea for supermarket checkout and a good idea for computation. With Isaac Chuang and others, talking to David DiVincenzo, Charles Bennett, and the IBM physics-and-computing circle, they programmed spins to run Grover's search and later Shor's factoring. Those NMR systems do not scale past a few qubits; the control techniques lived on. The whole arc began as a failed tag.

A second exhibit: with a student, Manu, he tried to build a ribosome for fluids to place tiny parts. Bubbles kept wrecking the system. After half a year of fighting them, Manu said the bubbles were better than the original plan. In viscous microchannels a stuck bubble is a switch and a memory; two bubbles meeting at a crossing are a logic gate. That is a universal computer, plus a fifty-cent bubble microscope, plus techniques later used to transplant genomes for synthetic life—all from a crashed fluidic ribosome. Fail fast, he says, usually skips ready and aim. You have to do homework before the failure and aim after it. Milestones and deliverables work on a straight line. They do not work for this kind of discovery. Part of what the fab-lab network does is give more people the chance to fail that way.

Geometry dominates molecular biology: shape is function, which is why protein folding matters, and the stack from amino acids to molecular motors looks, in simulation, like science-fiction walkers on a discrete lattice. Fridman asks how hydrogen and helium become a brain. Gershenfeld goes to thermodynamics: batteries run down, cars fail, life seems to resist entropy. Maxwell posed a demon that opens a door for fast molecules and closes it for slow ones, sorting a gas into hot and cold without touching anything, then running an engine—an apparent free lunch so infuriating it attached itself to a grim run of suicides among people who worked on it. Leó Szilárd reduced the puzzle to one molecule and one bit; Shannon credited that analysis for the invention of the bit. Rolf Landauer finally located the dissipation: as long as the demon remembers what it did, you can run the process backwards; forgetting is where reversibility dies. Charles Bennett then shocked Landauer by showing you can compute with arbitrarily little energy if the machine can run backwards. Data centers burn tens of megawatts; we are orders of magnitude from that limit. A student, Cameron, is working with Lincoln Laboratory on superconducting machines near it.

Life exists, in this telling, because it can locally violate thermodynamics, and it can do that because of intelligence—molecular intelligence. Gershenfeld goes further: this looks like the fifth AI boom-bust. Automation, expert systems, early neural nets, each promised to change everything and then had the goalposts moved. Chess was going to remake the world; now computers play chess and people still play chess. Large models will take another chunk of tasks, with roughly brain-scale capacity, and we will keep thinking. The eight orders of magnitude left are not in transistors or in the brain. They are in embodied intelligence, the intelligence of the body. Molecular systems already represent, store, search, and adapt. Their output is not a thought. It is us.

Consciousness, he does not put on quantum mechanics. A famous conversation in a Mussolini-era villa ran, in his parody: consciousness is weird, quantum mechanics is weird, therefore one explains the other. He was once a DARPA program manager on quantum biology. Biology trivially uses quantum mechanics because it is made of atoms; coherence is rarer and expensive: photosynthesis, probably olfaction, probably magnetoreception in birds. There is no evidence of it in cognition. He finds the question of whether a large language model is conscious silly. Biology is full of hacks that work; numerically, big computations now have brain-like degrees of freedom and a lot of the phenomenology. He sees no reason to invoke anything else. A preview of what digital fabrication might grow is topology optimization: you specify loads on a bicycle or a table and get organic forms that look like a forest because they solve loads the way biology does. That is a peek at teaching machines how to design, not merely generating a shape from a wish.

Cellular automata look like a mystery of complexity from simple rules. Gershenfeld says understanding them is easier than it sounds. Turing never meant his machine as an architecture; he used it to prove results about computability, and that work is exquisite. Almost everything is computationally universal. Norman Margolus and Tommaso Toffoli showed that a cellular automaton modeling billiard balls is a universal computer: you need persistent state, a nonlinear interaction, and connectivity. Cristopher Moore showed it is easy to make physical systems, bouncing balls on surfaces, that are uncomputable in Turing's deep sense. There is not yet a comparable notion of fabricational universality. Connect the two, and you get growth, adaptation, and evolution. That is why the heart of the conversation, for him, is morphogenesis: how a small genome becomes a body that can exist, reproduce, grow, and evolve. You cannot separate communication, computation, and fabrication there.

Is the universe a computer? He thinks quantum computing is overhyped except for modeling other quantum systems and a few cryptanalytic cases. The inventors were often asking the reverse question: if quantum mechanics can compute, that is a description of nature. Partial differential equations are an information technology from the pencil-and-paper era. Schrödinger's and Maxwell's equations are not fundamental; a point in that theory contains infinite information, which cannot be right, because information is a resource tied to energy. Ask a cosmologist whether information is conserved since the Big Bang, he says, and they stutter. Start instead with information and computation, already quantized, and build up to baseballs rolling down slopes. In that sense he does think of the universe as a computer—trivially because almost anything is universal, deeply as a way physics should be written.

Advice to the young is unfashionably plain. Junior faculty who try to reverse-engineer MIT tenure are miserable and do not get it; the ones who cannot imagine doing anything else do. Ninety-nine students in a hundred say the work is fascinating and they would like to work for him. One in a hundred says he is wrong, names the mistake, and starts. He does not know whether that resource is everyone or one percent, only that it is orders of magnitude larger than we employ. Once a year the fab network meets—Bhutan that year—and every body shape and language is the same person: bright, inventive joy. To join: cba.mit.edu for the research, fabfoundation.org, fablabs.io, fabacademy.org, fab.city. From one lab to a thousand they counted; from a thousand to a million counting stops being interesting. In that phase you go to a lab not to use a machine but to make one, and a "fab lab in a box," with SolidWorks, tries to ship the knowledge with the tools. Near-term AI in the lab becomes a collaborator that helps you start.

The research, he says, is in a sense just turning a crank. What he had not anticipated is the social question: if anybody can make almost anything anywhere, how do you live, learn, work, and play? Work-for-money-for-shopping is only a few decades old. A Sámi village in north Norway is deeply sustainable; the project is to reconnect that to global society rather than revert. Making crosses sectarian lines as an observation, not a philosophy. Deep in the brain is shaping our environment; the most important thing made in the labs is making itself. He has no idea what the meaning of life is, and maybe that is it. He leans toward stacked sigmoids rather than a Kurzweil singularity, and toward the selfish-gene view that life is the propagation of life, each level of organization the next violation of thermodynamics. Fab labs began as outreach and now send more knowledge back than they take. For him, tapping the brainpower of the planet is the meaning of his life. Morphogenesis or recursion all the way down: either is a fair summary. Life is short, he adds, but it is pretty fun.