Neil Gershenfeld: Self-Replicating Robots and the Future of Fabrication | Lex Fridman Podcast #380

1. 0:00 – 1:29

   Introduction

   1. NGNeil Gershenfeld0:00

      The ribosome, who I mentioned a little while back, can make an elephant one molecule at a time.

   2. LFLex Fridman0:04

      Mm-hmm.

   3. NGNeil Gershenfeld0:05

      Ribosomes are slow. They run at about one molecule a second, but ribosomes make ribosomes, so you have trillions (laughs) of them and that makes an elephant. In the same way, these little assembly robots I'm describing can make giant structures, at heart because a r- the robot can make the robot. (laughs) So more recently, two of my students, Amira and Miana, had a Nature Communication paper showing how this robot can be made out of the parts it's making, so the robots can make the robots, so you build up the capacity of robotic assembly.

   4. LFLex Fridman0:40

      The following is a conversation with Neil Gershenfeld, the director of MIT's Center for Bits and Atoms, an amazing laboratory that is breaking down boundaries between the digital and physical worlds, fabricating objects and machines at all scales of reality, including robots and automata that can build copies of themselves and self-assemble into complex structures. His work inspires millions across the world as part of the maker movement to build cool stuff, to create, the very act that makes life so beautiful and fun. This is the Lex Fridman Podcast. To support it, please check out our sponsors in the description, and now, dear friends, here's Neil Gershenfeld.

2. 1:29 – 6:53

   What Turing got wrong

   1. LFLex Fridman1:29

      You have spent your life working at the boundary between bits and atoms, uh, so the digital and the physical. What have you learned about engineering and about nature of reality from, uh, working at this divide, trying to bridge this divide?

   2. NGNeil Gershenfeld1:45

      I learned why von Neumann and Turing made fundamental mistakes.

   3. LFLex Fridman1:50

      (laughs) That's good stuff.

   4. NGNeil Gershenfeld1:52

      Um, I learned the secret of life.

   5. LFLex Fridman1:55

      Yeah.

   6. NGNeil Gershenfeld1:56

      Um, I- I- I learned how to ss-solve many of the world's most important problems, which all sound presumptuous, but all of those are things I learned at that boundary.

   7. LFLex Fridman2:05

      Okay, so, uh, Turing and von Neumann, let's start there.

   8. NGNeil Gershenfeld2:07

      Okay.

   9. LFLex Fridman2:07

      Some of the most impactful, important humans who have ever lived in computing. Why were they wrong?

   10. NGNeil Gershenfeld2:12

       So I worked with, uh, Andy Gleason, who was Turing's counterpart. So ju- just for background if anybody doesn't know, Turing is credited with the modern architecture of computing, among, uh, many other things. Uh, Andy Gleason was his US counterpart, and you might not have heard of Andy Gleason, but you might have heard of the Hilbert problems and Andy Gleason solved the fifth one, so he was a really notable mathematician. Uh, during the war, he was Turing's counterpart. Then von Neumann is credited with the modern architecture of computing and one of his students was Marvin Minsky, so I could ask Marvin what Johnny was thinking and I could ask Andy what Alan was thinking. And w- what came out from that, what I came to appreciate... As background, I never understood the difference between computer science and physical science.

   11. LFLex Fridman3:01

       Mm-hmm.

   12. NGNeil Gershenfeld3:01

       But Turing's machine that's the foundation of modern computing has a simple physics mistake, which is the head is distinct from the tape. So in the Turing machine, there's a head that programmatically moves and reads and writes the tape. The head is distinct from the tape, which means persistence of information is separate from interaction with information.

   13. LFLex Fridman3:25

       Yeah.

   14. NGNeil Gershenfeld3:26

       Then von Neumann wrote deeply and beautifully about many things, but not computing. He wrote a horrible men- memo called the First Draft of a Report on the EDVAC, which is how you program a very early computer. In it, he essentially s- roughly took Turing's architecture and built it into a machine. So the legacy of that is the computer somebody's using to watch this is spending much of its effort moving information from storage transis- transistors to processing transistors even though they have the same computational complexity. So in computer science, when you learn about computing, there's a ridiculous taxonomy of about a hundred different models of computation, but they're all fictions. In physics, a patch of space occupies space, it stores state, it takes time to transit, and you can interact. Th- that is the only model of computation that's physical. Everything else is a fiction. So I- I really came to appreciate that a few years back when I did a keynote for the annual meeting of the supercomputer industry, and then went into the halls and spent time with the supercomputer builders and came to appreciate... Oh, th- the, the, the... Let's see, if you're familiar with the movie The Metropolis-

   15. LFLex Fridman4:49

       Mm-hmm.

   16. NGNeil Gershenfeld4:49

       Uh, people would frolic upstairs in the gardens and down in the basement people would move levers.

   17. LFLex Fridman4:54

       Yes.

   18. NGNeil Gershenfeld4:54

       And that's how computing exists today, that we pretend software is not physical, it's separate from hardware, and the whole canon of computer science is based on this fiction that bits aren't constrained by atoms, but all sorts of scaling issues in computing come from that boundary, but all sorts of opportunities come from that boundary. And so you can trace it all the way back to Turing's machine making this mistake between the head and the tape. Von Neumann in- in... Um, crea- he never called it von Neumann's architecture. He wrote about it in this dreadful memo and then he wrote beautifully about other things we'll talk about.

   19. LFLex Fridman5:32

       Mm-hmm.

   20. NGNeil Gershenfeld5:32

       Now to end a long answer, Turing and von Neumann both knew this, so all of the canon of computer scientists credits them for what was never meant to be a computer architecture. Both Turing and von Neumann ended their life studying exactly how software becomes hardware. So von Neumann studied self-reproducing automata, how a machine communicates its own construction. Uh, Turing studied morphogenesis, how genes give rise to form. They ended their life studying the embodiment of computation, something that's been forgotten by the canon of computing, but developed sort of off to the sides by a really interesting lineage.

   21. LFLex Fridman6:12

       So, there's no distinction between the head and the tape, between the computer and the computation. It is all computation.

   22. NGNeil Gershenfeld6:19

       R-r-right. So I never understood the difference between computer science and physical science. And w-working at that boundary helped lead to things, like my lab was part of doing, with a number of interesting collaborators, the first faster-than-classical quantum, uh, computations. Uh, we were part of a collaboration creating the minimal synthetic organism where you design life in a computer.

   23. LFLex Fridman6:42

       Mm-hmm.

   24. NGNeil Gershenfeld6:42

       Those both involve domains where you just can't separate hardware from software. The embodiment of com- you know, computation is embodied in these really profound ways.

3. 6:53 – 20:00

   MIT Center for Bits and Atoms

   1. NGNeil Gershenfeld6:54

   2. LFLex Fridman6:54

      So the first quantum computations, synthetic life, so in the space of biology, the space of physics, at the lowest level, and the space of biology at the lowest level. So, uh, let's talk about, uh, CBA, Center of Bits and Atoms. What's the origin story of this MIT, legendary MIT center that you were, uh, a part of creating?

   3. NGNeil Gershenfeld7:16

      In high school, I really wanted to go to vocational school where you learn to weld and fix cars and build houses. And I was told, "No, you're smart. You have to sit in a room."

   4. LFLex Fridman7:26

      Mm-hmm.

   5. NGNeil Gershenfeld7:27

      And nobody could explain to me why I couldn't go to vocational school. Uh, I then worked at Bell Labs, this wonderful place, uh, before deregulation, legendary place.

   6. LFLex Fridman7:39

      Mm-hmm.

   7. NGNeil Gershenfeld7:39

      And I would get union grievances because I would go into the workshop and try to make something, and they would say, "No, you're smart. You have to tell somebody what to do." And it wasn't until MIT, and I'll explain how CBA started, but I could create CBA, that I came to understand this is a mistake that dates back to the Renaissance. So in the Renaissance, the liberal arts emerged, and liberal doesn't mean politically liberal. This was the path to liberation, birth of humanism. And so the liberal arts were the trivium, quadrivium, roughly language, natural science, and th- at that moment, what emerged was this d- dreadful concept of the illiberal arts.

   8. LFLex Fridman8:20

      Mm-hmm.

   9. NGNeil Gershenfeld8:20

      So anything that wasn't the liberal arts was for commercial gain and was just making stuff and wasn't valid for serious study. And so that's why we're left with learning to weld wasn't a subject for serious study. Um, but the means of expression have changed since the Renaissance, so micromachining or- or embedded coding is every bit ex- expressive as painting a painting or writing a sonnet. So, uh, never understanding this difference between computer science and physical science, uh, th- the path that led me to create CBA with colleagues was I was what's called a junior fellow at Harvard. I was visiting MIT through Marvin because I was interested in the physics of musical instruments.

   10. LFLex Fridman9:06

       Mm-hmm.

   11. NGNeil Gershenfeld9:07

       I, uh, this will be another slight digression. I, uh, in Cornell, I would study physics, and- and then I would cross the street and go to the music department where I played the bassoon and I would trim reeds and play the reeds.

   12. LFLex Fridman9:19

       Right.

   13. NGNeil Gershenfeld9:19

       And they'd be beautiful, but then they'd get soggy. And then I discovered in the basement of the music department at Cornell was David Borden, uh, who you might not have heard of, but is legendary in electronic music because he was really the first electronic musician. So Bob Moog, who invented, um, Moog synthesizers, was a physics student at Cornell, like me, crossing the street, and eventually he was kicked out and invented electronic music. David Borden was the first musician who created electronic music, so he's legendary for people like Phil Glass and Steve Reich. And so that got me thinking about I would behave as a scientist in the music department, but not in, uh, in the physics department, but not in the music department. It got me thinking about what's the computational capacity of a musical instrument. And through Marvin, he introduced me to Tod Machover at the Media Lab who was just about to start a project with Yo-Yo Ma to, um, that led to a collaboration, uh, to instrument a cello, to- to extract Yo-Yo's data and bring it out into computational environments.

   14. LFLex Fridman10:20

       What is the computational capacity of a musical instrument? Uh, uh, as we continue on this tangent and when we shall return to CBA.

   15. NGNeil Gershenfeld10:26

       Yeah. So o- one part of that is to understand the computing. And if you look at, like, the finest time scale and length scale you need to model the physics, it- it's not heroic. You know, a- a- a good GPU can do teraflops today. That- that used to be a national class supercomputer. Now it's just a GPU. And that's about, if you take the time scales and length scales relevant for the physics, that's about the scale of the physics computing. For Yo-Yo, what was really driving it was he's completely unsentimental about the Strad. It's not that it makes some magical wiggles in the sound wave. It's, it's performance as a controller, h- how he can manipulate it as an interface device.

   16. LFLex Fridman11:09

       Interface between what and what exactly?

   17. NGNeil Gershenfeld11:11

       H- h- him and sound.

   18. LFLex Fridman11:12

       Okay, him and sound.

   19. NGNeil Gershenfeld11:13

       And so- so what it led to was l- I had started by thinking about ops per second, but Yo-Yo's question was really, um, resolution and bandwidth. It's, um, how fast can you measure what he does and, um, uh, the- the- the bandwidth and the resolution of detecting his controls and then mapping them into sounds.

   20. LFLex Fridman11:37

       Mm-hmm.

   21. NGNeil Gershenfeld11:38

       And wha- what we found, what he found was if you instrument everything he does and connect it to almost anything, it sounds like Yo-Yo, that- that- that the magic is in the control, not in ineffable details in how the wood wiggles. And so with- with Yo-Yo and Tod, that led to a piece, and towards the end, I asked Yo-Yo what- what it would take for him to get rid of his Strad and use our stuff, and his answer was just logistics. It was, at that time, our stuff was like a rack of electronics and lots of cables and some grad students to- to make it work. Once the technology becomes as invisible as the Strad...... that, then sure, absolutely, h- he would take it. And by the way, as a footnote on the footnote, an accident in the sensing of Yo-Yo's cello led to $100 million a year auto safety business to control airbags in cars.

   22. LFLex Fridman12:29

       How did that work?

   23. NGNeil Gershenfeld12:31

       I had to instrument the bow without interfering with it. So I, um, set up, um, local electromagnetic fields-

   24. LFLex Fridman12:37

       Mm-hmm.

   25. NGNeil Gershenfeld12:38

       ... where I would, um, detect, um, how those fields interact with the bow he's playing. But we had a problem that his hand, whe- whenever his hand got near these sensing fields, I would start sensing his hand rather than the materials on the bow.

   26. LFLex Fridman12:53

       Nice.

   27. NGNeil Gershenfeld12:53

       And I didn't quite understand what was going on with those, that, that interference. So my very first grad student ever, uh, Josh Smith, uh, did a thesis on tomography with electric fields, how to see in 3D with electric fields. Then through Todd and a, at that point a research scientist in my lab, Joe Paradiso, it led to a collaboration with, uh, Penn & Teller who, um, where we did a magic trick in Las Vegas to contact Houdini. And sort of these fields are sort of like, you know, contacting spirits.

   28. LFLex Fridman13:27

       Mm-hmm.

   29. NGNeil Gershenfeld13:28

       So we did a magic trick in Las Vegas and then the, the crazy thing that happened after that was, uh, Phil Rittmueller, um, came running into my lab. He worked with, um, this became with Honda and NEC, airbags were killing infants in rear-facing child seats. Um, cars need to distinguish, uh, a front-facing adult where you'd save the life versus a bag of groceries where you don't need to fire the airbag versus a rear-facing infant where you would kill it. And so the, the, the seat needed to, in effect, see in 3D to understand the occupants. And so we took the Penn & Teller magic trick derived from Josh's thesis from Yo-Yo's cello to an auto show and all the car companies said, "Great. Where, when can we buy it?" And so that became Elesis, a hu- and it was $100 million a year business making sensors. There wasn't a lot of publicity because it was in the car so the car didn't kill you. So they didn't sort of advertise, "We have nice sensors so the car doesn't kill you."

   30. LFLex Fridman14:26

       Yeah.
4. 20:00 – 26:36

   Digital logic

   1. NGNeil Gershenfeld20:00

      Yeah. So what is digital? Uh, the casual obvious answer is digital in one and zero-

   2. LFLex Fridman20:08

      Mm-hmm.

   3. NGNeil Gershenfeld20:08

      ... but that's wrong. There's a much deeper answer which is Claude Shannon at MIT wrote the best master's thesis ever.

   4. LFLex Fridman20:18

      Mm-hmm.

   5. NGNeil Gershenfeld20:18

      In his master's thesis, he invented our modern notion of digital logic. Where it came from was, uh, Vannevar Bush, uh, was a grand old man at MIT. Uh, he created the post-war research establishment that led to the National Science Foundation and he made an important mistake which we can talk about. But he also made the la- the differential analyzer which was the last great analog computer. So it was a room full of gears and pulleys and the longer it ran, the worse the answer was.

   6. LFLex Fridman20:49

      (laughs)

   7. NGNeil Gershenfeld20:49

      And Shannon-

   8. LFLex Fridman20:50

      Yeah.

   9. NGNeil Gershenfeld20:51

      ... worked on it as a student and he got so annoyed, in his master's thesis he invented digital logic.

   10. LFLex Fridman20:56

       Mm-hmm.

   11. NGNeil Gershenfeld20:57

       Um, but he then went on to Bell Labs and what he did there was communication was beginning to expand, there was more demand for phone lines and so there's a question about h- how much, how many phone lines you could, phone messages you could send down a wire and you could try to just make it better and better. He asked a question nobody had asked which is rather than make it better and better, what's the limit to how good it can be?

   12. LFLex Fridman21:23

       Mm-hmm.

   13. NGNeil Gershenfeld21:23

       And, uh, he proved a couple of things but one of the main things he proved was a threshold theorem for channel capacity. And so what he showed was my voice to you right now is coming as a wave through sound-

   14. LFLex Fridman21:38

       Mm-hmm.

   15. NGNeil Gershenfeld21:38

       ... and the further you get, the worse it sounds. But people watching this are getting it as, uh, as inf- packets of data in a network. Um, when they get... When the computer they're watching this gets the packet of information, um, it- it can detect and correct an error. And what Shannon showed is if the noise in- in the cable to the people watching this is above a threshold, they're doomed. But if the noise is below a threshold for a linear increase in the energy representing our conversation, the error rate goes down exponentially. Exponentials are fast, there's very few of them in engineering, and the exponential reduction of error below a threshold if you restore state is called a threshold theorem.

   16. LFLex Fridman22:27

       Mm-hmm.

   17. NGNeil Gershenfeld22:27

       That's what led to digital. That- that means unreliable things can work reliably. So Shannon did that for communication. Then von Neumann was inspired by that and applied it to computation and he showed how an unreliable computer can operate reliably by using the same threshold property of restoring state. It was then forgotten many years. We had to rediscover it in effect in the quantum computing era when things are very unreliable again.

   18. LFLex Fridman22:54

       Mm-hmm.

   19. NGNeil Gershenfeld22:55

       Um, but now to go back to how does this relate to the biggest things I've made? So in fabrication, MIT invented computer-controlled manufacturing in 1952.

   20. LFLex Fridman23:09

       Mm-hmm.

   21. NGNeil Gershenfeld23:09

       Uh, jet aircraft were just emerging. There is a limit to turning cranks on a machine, on a milling machine, to make parts for jet aircraft. Now this is a messy story. MIT actually stole computer-controlled machining from an inventor who brought it to MIT-

   22. LFLex Fridman23:25

       Uh-oh.

   23. NGNeil Gershenfeld23:25

       ... wanted to do a joint project with the Air Force and MIT effectively stole it from him. So it's kind of a messy history. But that sounds like the birth of computer-controlled machining, 1952. Um, there are a number of inventors of 3D printing. One of the companies spun off from my lab by Max Lobovsky is Formlabs, which is now a-

   24. LFLex Fridman23:46

       Mm-hmm.

   25. NGNeil Gershenfeld23:47

       ... billion-dollar 3D printing company. That's the modern version. But all of that's analog, meaning the information is in the control computer, there's no information in the materials. And so it goes back to Vannevar Bush's analog computer. If you mista- make a mistake in printing or machining, just the mistake accumulates. The real birth of computerized digital manufacturing is four billion years ago. That's the evolutionary age of the ribosome. So the way you're manufactured-

   26. LFLex Fridman24:20

       Mm-hmm.

   27. NGNeil Gershenfeld24:21

       ... is there's a code that describes you, the genetic code. It- it goes to a- a micromachine, the ribosome, which is this molecular factory that builds the, um, molecules that- that are you.The key thing to know about that is it, there are about 20 amino acids that get assembled.

   28. LFLex Fridman24:43

       Mm-hmm.

   29. NGNeil Gershenfeld24:43

       And in that machinery, it does everything Shannon and von Neumann taught us. You detect and correct errors. So if you mix chemicals, the error rate is about a part in 100. When you make, elongate a protein in the ribosome, it's about a part in 10 to the 4. When you replicate DNA, there's an extra level of error correction. It's a part in 10 to the 8. And so in the molecules that make you, you can detect and correct errors, and you don't need a ruler to make you. The geometry comes from your parts. So now, compare a child playing with Lego and a state-of-the-art 3D printer or computerized milling machine. Uh, the tower made by a child is more accurate than their motor control because the act of snapping the bricks together...

   30. LFLex Fridman25:31

       Mm-hmm.
5. 26:36 – 37:04

   Self-assembling robots

   1. NGNeil Gershenfeld26:36

      My lab was working with the aerospace industry. So Spirit Aero was Boeing's factories. Uh, they asked us for how to join composites. When you make a composite airplane, you make these giant wing and fuselage parts, and they asked us for a better way to stick them together.

   2. LFLex Fridman26:51

      Mm-hmm.

   3. NGNeil Gershenfeld26:52

      'Cause the joints were a place of failure. And what we discovered was instead of making a few big parts, if you make little loops of carbon fiber, and you reversibly link them in joints, and you do it in a special geometry that balances being under-constrained and over-constrained with just the right degrees of freedom, um, we set the world record for the highest modulus ultralight material just by ef- in effect, making carbon fiber Lego.

   4. LFLex Fridman27:24

      (laughs) .

   5. NGNeil Gershenfeld27:24

      So the, uh, so lightweight materials are crucial for energy efficiency. This let us make, make the, the lightest weight high modulus material. We then showed that with just, just a few part types, we can tune the material properties, and then you can create really wild robots that instead of having a tool the size of a jumbo jet to make a jumbo jet, you can make little robots that walk on these cellular structures to build the structures, where they error correct their position on the structure and they navigate on the structure. And so using all of that, uh, with, um, NASA, we made morphing airplanes. Um, a, a former student, Kenny Cheung and Ben Jeannette made a morphing airplane the size of NASA Langley's biggest wind tunnel. Um, with Toyota, we've made super-efficiency race cars. We're right now looking at projects with NASA to build these for things like space telescopes and space habitats, where the ribosome, who I mentioned a little while back, can make an elephant one molecule at a time.

   6. LFLex Fridman28:27

      Mm-hmm.

   7. NGNeil Gershenfeld28:28

      Ribosomes are slow. They run at about one molecule a second, but ribosomes make ribosomes. So you have thousands of them, uh, trillions of them, and that makes an elephant. In the same way, these little assembly robots I'm describing can make giant structures, uh, at heart because a, the robot can make the robot. So more recently, two of my students, Samira and Miana, had a Nature Communication paper showing how this robot can be made out of the parts it's making. So the robots can make the robots. So you build up the capacity of robotic assembly.

   8. LFLex Fridman29:02

      A self- it can self-replicate. Can you linger on what that robot looks like? What is a robot that can walk along and do error correction, and what is a robot that can self-replicate, uh, from the materials it is given? What does that look like? What are we talking about?

   9. NGNeil Gershenfeld29:16

      So, um...

   10. LFLex Fridman29:17

       This is fascinating.

   11. NGNeil Gershenfeld29:17

       Yeah. Th- the answer is di- different at different length scales. S- so, to explain that, uh, in biology, primary structure is the code in the messenger RNA that says what the ribosome should build.

   12. LFLex Fridman29:30

       Yeah.

   13. NGNeil Gershenfeld29:30

       Um, secondary structure are geometrical motifs. They're things like helices or sheets.

   14. LFLex Fridman29:35

       Mm-hmm.

   15. NGNeil Gershenfeld29:35

       Tertiary structures are functional elements, like electron donors or acceptors. Quaternary structure is things like, um, molecular motors that are moving my mouth or making the synapses work in my brain. So there's that hierarchy of primary, secondary, tertiary, quaternary. Now what's interesting is if you want to buy electronics today from a vendor, there are hundreds of thousands of types of resistors or capacitors or transistors, huge inventory. All of biology is just made from this inventory of 20 parts, the amino acids.

   16. LFLex Fridman30:11

       Mm-hmm.

   17. NGNeil Gershenfeld30:11

       And by composing them, you can create all of life. And so, uh, as part of this digitization of materials, we're in effect trying to create something like amino acids for engineering, creating all of technology from 20 parts. I, um, I, l- see, uh, as another digression, I helped start an office for science in Hollywood, and, um, there was a fun thing for the movie The Martian where-... I did a program with Bill Nye and a few others on how to actually build a civilization on Mars that they described in a way that I like, as I was talking about how to go to Mars without luggage.

   18. LFLex Fridman30:49

       (laughs) Yeah.

   19. NGNeil Gershenfeld30:49

       And the, at heart, it's sort of how to create life in non-living materials. So if- if you think about this primary, secondary, tertiary, quaternary structure, um, in my lab, we're doing that, but on different length scales for different purposes. So we're making micro-bots out of, like, nano bricks. And to make the robots to build large scale structures in space, the elements of the robots now are centimeters rather than, um, micrometers. And so the assembly robots for the bigger structures are, uh, they are the cells that make up the structure, but then we have functional cells. And so cells that can process and actuate. Each- each cell can, like, move one degree of freedom, or attach or dis- detach or, um, process. Now, those elements I just described, we can make out of the still smaller parts. So eventually there's a hierarchy of the little parts make little robots that make bigger parts of bigger robots that, up- up through that hierarchy.

   20. LFLex Fridman31:52

       And- and that way you can move up the length scale.

   21. NGNeil Gershenfeld31:54

       Right. Early on, I tried to go in a straight line from the bottom to the top, and that ended up being a bad idea. Instead, we're kind of doing all of these in parallel, and then they're growing together. And so to make the larger scale structures, we, um, like there's a lot of hype right now about 3D printing houses, where you have a printer the size of the house.

   22. LFLex Fridman32:13

       (laughs)

   23. NGNeil Gershenfeld32:13

       We're right now working on using swarms of these, you know, uh, table scale robots that walk on the structures to place the parts, um, much more efficiently.

   24. LFLex Fridman32:22

       That's amazing. But you're saying you can't, for now, go from the very small to the very large.

   25. NGNeil Gershenfeld32:27

       Th- that'll come. Um, that- that'll come in stages.

   26. LFLex Fridman32:30

       Can we just linger on this idea? Starting from von Neumann's, uh, self-replicating automata that you mentioned, it's just a beautiful idea.

   27. NGNeil Gershenfeld32:39

       So that's at the heart of all of this, in the stack I described. So one student, Will Langford, made these micro-bots out of little parts that then we're using for Miana's bigger robots up through this hierarchy. And it's really realizing this idea of the self-reproducing automata. So von Neumann, when I complained about the von Neumann architecture-

   28. LFLex Fridman33:00

       Mm-hmm.

   29. NGNeil Gershenfeld33:01

       ... it's not fair to von Neumann because he never claimed it as his architecture. He really wrote about it in this one fairly dreadful memo that led to all sorts of lawsuits and fights and, about the early days of computing. He did beautiful work on reliable computation and unreliable devices. And towards the end of his life, what he studied was how, and- and I have to say this precisely, how a computation communicates its own construction.

   30. LFLex Fridman33:27

       Yeah.
6. 37:04 – 47:59

   Digital fabrication

   1. NGNeil Gershenfeld37:04

      the thing.

   2. LFLex Fridman37:04

      So you're saying, I mean, this is one of the cases you're making and, uh, that this is this third revolution. We've seen the Moore's Law in communication, we've seen the Moore's Law-like type of growth-

   3. NGNeil Gershenfeld37:15

      Right.

   4. LFLex Fridman37:15

      ... in, uh, computation. And you're anticipating we're going to see that in digital fabrication. Can you actually, first of all, describe what you mean by this te- term, digital fabrication?

   5. NGNeil Gershenfeld37:26

      So the casual meaning is a computer controls a tour- tool to make something.

   6. LFLex Fridman37:31

      Uh-huh.

   7. NGNeil Gershenfeld37:31

      And that was invented when MIT stole it in 1952. (laughs)

   8. LFLex Fridman37:35

      Yeah.

   9. NGNeil Gershenfeld37:36

      Um, there's the deep meaning of what the ribosome does, of a computa- uh, uh, uh, of- of a des- a, a digital description doesn't describe a thing, a digital description becomes the thing.

   10. LFLex Fridman37:47

       Yeah.

   11. NGNeil Gershenfeld37:48

       That's where the- that's- that's the path to the Star Trek replicator, and that's the thing that doesn't exist yet. Now, I think the- the best way to understand what this roadmap looks like is to now s- bring in Fab Labs and how they relate to all of this.

   12. LFLex Fridman38:04

       What are Fab Labs?

   13. NGNeil Gershenfeld38:06

       So here- here's the sequence. Uh, um, with colleagues, I accidentally started a network of what's now 2,500 digital fabrication community labs called Fab Labs right now in 125 countries and they double every year and a half. That's called Lass' Law after Sherry Lassiter, who I'll explain. So here's the sequence. Uh, we started Center for Bits and Atoms-

   14. LFLex Fridman38:30

       Mm-hmm.

   15. NGNeil Gershenfeld38:31

       ... to do the kind of research we're talking about. We had all of these machines and then had a problem, it would take a lifetime of classes to learn to use all the machines. So with, you know, colleagues who helped start CBA, we began a class modestly called How to Make Almost Anything.

   16. LFLex Fridman38:48

       Yeah.

   17. NGNeil Gershenfeld38:48

       And there's no big agenda. It was just, it was aimed at a few research students to use the machines and, uh, we were completely unprepared for the first time we taught it. We were swamped by, e- every year since, hundreds of students try to take the class. It's one of the most oversubscribed classes at MIT. Um, students would say things like, you know, "Can you teach this at MIT? It seems too useful." It's just how to work these machines.

   18. LFLex Fridman39:12

       Mm-hmm.

   19. NGNeil Gershenfeld39:13

       And the students in the class, I w- you know, I would teach them all the skills to use all these tools and then they would do projects integrating them and they were amazing. So Kelly was a sculptor, no engineering background. Uh, her project was she made a device that saves up screams when you're mad and plays them back later and-

   20. LFLex Fridman39:31

       Uh, saves up screams when you're mad and plays them back later.

   21. NGNeil Gershenfeld39:34

       Right. Y- you scream into this device and it- it- it deadens the sound, records it, and then when it's convenient, releases your scream.

   22. LFLex Fridman39:40

       Can we just- just, like, pause on the brilliance of that invention, creation, the art, I don't know, the brilliance? Who is this that created this?

   23. NGNeil Gershenfeld39:50

       Kelly Dobson.

   24. LFLex Fridman39:51

       Kelly Dobson.

   25. NGNeil Gershenfeld39:52

       Gone on to do a number of interesting things. Uh, Mijin, who's gone on to do a number of interesting things, uh, made a dress instrumented with sensors and spines and when somebody creepy comes close, it would defend your personal space.

   26. LFLex Fridman40:02

       Which are also very useful.

   27. NGNeil Gershenfeld40:03

       Um, uh, uh, another project early on was a web browser for parrots which have the cognitive ability of a young child and lets parrots surf the internet, you know? Another w- another was an arm clock- an alarm clock you wrestle with and prove you're awake.

   28. LFLex Fridman40:16

       Mm-hmm.

   29. NGNeil Gershenfeld40:17

       And what connects all of these is... So MIT made the first real-time computer, the Whirlwind.

   30. LFLex Fridman40:23

       Mm-hmm.
7. 47:59 – 55:45

   Self-reproducing machine

   1. NGNeil Gershenfeld47:59

   2. LFLex Fridman47:59

      So how difficult is it to create a self-replicating assembler, self-replicating machine that builds copies of itself or builds a more complicated version of itself, which is kind of the dream towards which you're pushing in a generic arbitrary sense?

   3. NGNeil Gershenfeld48:16

      I had a student, Nadia Peak with Jonathan Ward, who- who for me started this idea of how do we use the tools in my lab to make the tools in the lab.

   4. LFLex Fridman48:25

      Yes.

   5. NGNeil Gershenfeld48:26

      Uh, in a very clear sense, they are making self-reproducing machines. So one of the really cool things that's happened is there's a whole network of machine builders around the world. So there's Danielle in- now in Germany and Jens in Norway and, um, um, each of these people is- has learned the skills to go into a Fab Lab and make a machine. And so we've started creating a network of super fab... So the Fab Lab can make a machine, but it can't make a number of the precision parts of the machine. So in places like Bhutan or Kerala in the south of India, we started creating super Fab Labs that have more advanced tools to make the parts of the machines so that the machines themselves become even cheaper.... um, that, that is self-reproducing machines, but you need to feed it things like bearings or microcontrollers.

   6. LFLex Fridman49:20

      Sure.

   7. NGNeil Gershenfeld49:20

      They can't make those parts. But other than that, they're making their own things. And I should note as a footnote, the stack I described of computers controlling machines, to machine making machines, to assemblers, to self-assemblers, view that as Fab 1, 2, 3, 4.

   8. LFLex Fridman49:35

      Hmm.

   9. NGNeil Gershenfeld49:35

      So we're transitioning from Fab 1 to Fab 2, and the research in the lab is 3 and 4. At this Fab 2 stage, a big component of this is, uh, sustainability in the material feedstocks. So Alicia, a colleague in Chile, is leading a great effort looking at how you take, um, forest products and coffee grounds and seashells and a range of locally available materials and produce the high-tech materials that go into the lab. So all of that is machine building today. Then back in the lab, what we can do today is we have robots that can build structures and can assemble more robots that build structures. We have finer resolution robots that can build micromechanical systems, so robots that can build robots that can walk and manipulate.

   10. LFLex Fridman50:26

       Mm-hmm.

   11. NGNeil Gershenfeld50:27

       And we're just now, we have a project at the layer below that, where there's endless attention today to billion-dollar chip fab investments.

   12. LFLex Fridman50:36

       Mm-hmm.

   13. NGNeil Gershenfeld50:37

       Uh, but a- (laughs) a really interesting thing we passed through is, today, the smallest transistors you can buy is a single transistor, just commercially for electronics, is actually the size of an early transistor in an integrated circuit. So we're, we're using these machines-making-machines-making-assemblers to place those parts to not use a billion-dollar chip fab to make integrated circuits, but actually assemble little electronic components.

   14. LFLex Fridman51:03

       So have, uh, fine enough, precise enough actuators and manipulators-

   15. NGNeil Gershenfeld51:08

       So-

   16. LFLex Fridman51:08

       ... that allow you to place these transistors.

   17. NGNeil Gershenfeld51:10

       Right. We're, that's a research project in my lab on D- called DICE, on Discrete Assembly of Integrated Electronics. And we're just at the point to really start to take seriously this notion of not having a chip fab make integrated electronics, but having, not a 3D printer, but a thing that's a cross between, a pick-and-place makes circuit boards in 2D.

   18. LFLex Fridman51:31

       Yeah.

   19. NGNeil Gershenfeld51:32

       Um, the 3D printer extrudes in 3D.

   20. LFLex Fridman51:34

       Mm-hmm.

   21. NGNeil Gershenfeld51:34

       We're making sort of a micro-manipulator that acts like a printer, but it's placing to build electronics in 3D. A-

   22. LFLex Fridman51:41

       But this micro-manipulator is distributed? So there's a bunch of them or is this one centralized thing?

   23. NGNeil Gershenfeld51:45

       Oh, so, that's why that's a great question. So, um, I have a prize that's almost but not been claimed for the students whose thesis can walk out of the printer.

   24. LFLex Fridman51:54

       Oh, nice.

   25. NGNeil Gershenfeld51:55

       So you have to print the thesis with the means to, to exit the printer, and it has to contain its description of the thesis that says how to do that.

   26. LFLex Fridman52:08

       It's a really good, uh, I mean, it's a, it's a, (laughs) it's a fun example of exactly the thing we're talking about.

   27. NGNeil Gershenfeld52:13

       A- and I've had a few students almost, uh, get to that. Um, and so, um, in, in what I'm describing, there's this stack where we're getting closer, but it's still quite a few years to really go from a s- So, there's a layer below the transistors where we assemble the base materials that become the transistor. We're now just at the edge of assembling the transistors to make the circuits. W- w- we can assemble the microparts to make the microrobots. We can assemble the bigger robots. And in the coming years, we'll be patching together all of those, uh, scales.

   28. LFLex Fridman52:47

       Do you, so do you see a vision of just endless billions of robots s- at the different scales self-assembling, um, s- self-replicating and building m- complicated structures?

   29. NGNeil Gershenfeld52:59

       Yes. Uh, uh, a- and the but to the yes but is, let me clarify two things. One is, that immediately raises King Charles' fear of gray goo, of runaway mutant self-reproducing-

   30. LFLex Fridman53:14

       Sure.
8. 55:45 – 1:00:41

   Trash and fabrication

   1. NGNeil Gershenfeld55:45

      it does.

   2. LFLex Fridman55:45

      Yeah, but what are the different ways, uh, the evolution of the exponential scaling of digital fabrication can evolve? So you said, um, yeah, self-replicating nanobots, right? This is the- the gray goo fear. It's a caricature of a fear, but nevertheless there's interesting, just like you said, spam and all these kinds of things that came with the scaling of c- communication and computation. What are the different ways that malevolent actors will use this technology?

   3. NGNeil Gershenfeld56:15

      Yeah, well first let me start with a benevolent story.

   4. LFLex Fridman56:17

      Yes.

   5. NGNeil Gershenfeld56:17

      Which is, uh, trash is an analog concept.

   6. LFLex Fridman56:23

      Yeah.

   7. NGNeil Gershenfeld56:24

      There's no trash in a forest. All the parts get disassembled and reused. Trash means something doesn't have enough information to tell you how to reuse it.

   8. LFLex Fridman56:33

      Yeah.

   9. NGNeil Gershenfeld56:34

      It's as simple as th- there's no trash in a Lego room.

   10. LFLex Fridman56:38

       Mm-hmm.

   11. NGNeil Gershenfeld56:38

       Wh- when you assemble Lego, the Lego bricks have enough information to dis- disassemble them. So one of the... So as you go through this fab one, two, three, four story, one of the implications of this transition to, from printing to assembling... So the real breakthrough technologically isn't additive versus subtractive, which is a subject of a lot of attention and hype. Um, yep, 3D printers are useful, um, you know, we spun off companies like Formlabs, led by Max, for 3D printing, but in a fab lab it's one of maybe 10 machines. It's- it's used but it's only part of the machines. The real technological change is when we go from printing and cutting to assembling, uh, and disassembling.

   12. LFLex Fridman57:22

       Mm-hmm.

   13. NGNeil Gershenfeld57:22

       But that reduces inventories of hundreds of thousands of parts to just having a few parts to make almost anything.

   14. LFLex Fridman57:29

       Mm-hmm.

   15. NGNeil Gershenfeld57:29

       It reduces global supply chains to locally sourcing these building blocks. But one of the key implications is it gets rid of technological trash, because you can disassemble and reuse the parts, not throw them away. And so initially, that's of interest for things at the end of long supply chains, like satellites on orbit, but one of the things coming is eliminating technical trash through reuse of the building blocks.

   16. LFLex Fridman57:54

       So, like, when you think about 3D printers, you're thinking about-

   17. NGNeil Gershenfeld57:57

       So-

   18. LFLex Fridman57:57

       ... addition and subtraction. When you think about the other options available to you in that parameter space, as you call it-

   19. NGNeil Gershenfeld58:05

       Yeah.

   20. LFLex Fridman58:05

       ... that's going to be assembly, disassembly, cutting, you said?

   21. NGNeil Gershenfeld58:09

       So the 1952 NC mill was, uh, subtractive. You remove material.

   22. LFLex Fridman58:15

       Mm-hmm.

   23. NGNeil Gershenfeld58:15

       And, um, 3D printing, a- additive, and there's a couple claims to the invention of 3D printing. That's closer to what's called net shape, which is you don't have to cut away the material you don't need. You just put material where you do need it.

   24. LFLex Fridman58:27

       Mm-hmm.

   25. NGNeil Gershenfeld58:27

       And so that's the 3D printing revolution. But there are all sorts of limitations on 3D printing, to, um, the kinds of materials you can print, um, the kind of functionality you can print. We're- we're- we're just not gonna get to making a, um, everything in a cellphone on a single printer. But I do expect to make everything in a cellphone with an assembler. And so instead of printing and cutting technologically, it's this transition to assembling and disassembling, y- going back to Shannon and von Neumann going back to the ribosome four billion years ago. Now, you- you come to malevolent. Um, let me tell you a story about... I was doing a briefing for, uh, th- the National Academy of Sciences group that advises the intelligence communities. And I talked about the kind of research we do, and at the very end, I showed a little video clip of Valentina in Ghana, um, making, uh, a local girl making surface mount electronics in the fab lab. And I showed that to this room full of people. Uh, one of the in- members of the intelligence community got up, livid, and said, "How dare you waste our time showing us a young girl in an African village making surface mount electronics? We're looking e- we need to know about disruptive threats to the future of the United States."

   26. LFLex Fridman59:52

       (laughs) Yeah.

   27. NGNeil Gershenfeld59:53

       And somebody else got up in the room and yelled at him. "You idiot, I can't think of anything more important than this."

   28. LFLex Fridman1:00:00

       Yeah.

   29. NGNeil Gershenfeld1:00:00

       But for two reasons. One reason was, um, because if we rely on, like, informational superiority in the battlefield, it means other people could get access to it. But this intelligence person's point, bless him, wasn't that, it was getting at the root causes of conflict, is if this young girl in an African village could actually master surface mount electronics, it changes some of the most fundamental things about recruitment for terrorism, um, uh, impact of economic migration, basic assumptions about an economy. It's ju- just existential for the future of the planet.

   30. LFLex Fridman1:00:40

       But,

9. 1:00:41 – 1:04:56

   Lab-made bioweapons

   1. LFLex Fridman1:00:41

      you know, we've just lived through a pandemic. I would love to linger on this, 'cause the possibilities that are positive are endless.

   2. NGNeil Gershenfeld1:00:50

      Yeah.

   3. LFLex Fridman1:00:50

      But the possibilities that are negative are still nevertheless extremely important. What's both positive and negative? What do you do with a large number of general assemblers?

   4. NGNeil Gershenfeld1:01:01

      Yeah. With the fab lab, you could roughly make a bio lab then learn biotechnology. Now, that's terrifying because making self-reproducing gray goo that out-competes biology I consider doom because biology knows everything I'm describing and is really good at what it does.

   5. LFLex Fridman1:01:19

      Mm-hmm.

   6. NGNeil Gershenfeld1:01:20

      Um, in...... how to grow almost anything. You learn skills in biotechnology that would let, that let you make serious biological threats.

   7. LFLex Fridman1:01:32

      And when you combine, uh, some of the innovations you see with large language models, some of the innovations you see with AlphaFold, so applications of AI for designing biological systems, for, uh, writing programs which you can with l- large language models increasingly. So there seems to be an interesting dance here of automating the design stage of complex systems using AI, and then that's the, th- that's the bits.

   8. NGNeil Gershenfeld1:02:02

      Mm-hmm.

   9. LFLex Fridman1:02:02

      And you can leap now the innovations you're talking about. You can leap from the complex systems in the digital space to th- the printing to the creation to the assembly, uh, at scale i- i- of, uh, complex systems in the physical space.

   10. NGNeil Gershenfeld1:02:18

       Yeah. So something to be scared about is a fab lab can make a bio lab, a bio lab can make biotechnology, somebody could learn to make a virus.

   11. LFLex Fridman1:02:28

       True.

   12. NGNeil Gershenfeld1:02:28

       Uh, that's scary. That, that's, uh, unlike some of the things I said I don't worry about, that's something I really worry about that is scary.

   13. LFLex Fridman1:02:36

       Okay.

   14. NGNeil Gershenfeld1:02:37

       Now how do you deal with that? Uh, prior threats we dealt with command and control.

   15. LFLex Fridman1:02:45

       Mm-hmm.

   16. NGNeil Gershenfeld1:02:46

       So like, uh, e- early color copiers had unique codes and you could tell which copier made them. Eventually you couldn't keep up with that. Uh, there, there was a famous meeting at Asilomar in the early days of recombinant DNA-

   17. LFLex Fridman1:03:02

       Mm-hmm.

   18. NGNeil Gershenfeld1:03:02

       ... where that community recognized the dangers of what it was doing and put in place a regime to help manage it, and so that led to the kind of research management. So, you know, MIT has an office t- that supervises research and it works with the national office. That works if you can identify who's doing it and where.

   19. LFLex Fridman1:03:22

       Mm-hmm.

   20. NGNeil Gershenfeld1:03:23

       It doesn't work in this world we're describing.

   21. LFLex Fridman1:03:25

       Yeah.

   22. NGNeil Gershenfeld1:03:26

       So anybody could do this anywhere. And so what we've found is you can't contain this. It's already out. You can't forbid because there isn't command and control. The most useful thing you can do is provide incentives for transparency.

   23. LFLex Fridman1:03:44

       Yes.

   24. NGNeil Gershenfeld1:03:44

       So but really, the heart of what we do is you could do this by yourself in a basement for nefarious reasons or you could come into a place in the light where you get help and you get community and you get resources, and there's an incentive to do it in the open, not in the dark. And that might sound naïve, but in the sort of places we're working, oh, you know, um, again, bad people do bad things in these places already, but, uh, providing openness and providing transparency is a key part of managing these. And so it, it, it transitions from regulating risks as regulation to, to soft power to manage them.

   25. LFLex Fridman1:04:30

       So there's so much potential for good, so much capacity for good, that, uh, fab labs and the, uh, the, the ability, um, and the tools of creation really unlock that potential.

   26. NGNeil Gershenfeld1:04:44

       Yeah. And I, I don't say that as sort of dewy-eyed naive. I, I say that empirically from just-

   27. LFLex Fridman1:04:50

       Yeah.

   28. NGNeil Gershenfeld1:04:50

       ... y- years of seeing how this plays out in communities.

   29. LFLex Fridman1:04:52

       I wonder if it's the early days of personal computers, though, before we get spam,

10. 1:04:56 – 1:16:48

    Genome

    1. LFLex Fridman1:04:56

       right?

    2. NGNeil Gershenfeld1:04:57

       In the end, most fundamentally, literally the mother of all problems is who designed us. So w- so assume success in that we're gonna transition to the machines making machines and all of these new sort of social systems we're describing will help manage them and curate them and democratize them. I- i- if we close the gap I just led off with of 10 to the 10 to 10 to the 18 between chip fab and you, um, we're ultimately, in marrying communication, computation, and fabrication, gonna be able to create unimaginable complexity.

    3. LFLex Fridman1:05:42

       Mm-hmm.

    4. NGNeil Gershenfeld1:05:43

       Um, and how do you design that? And so I'd say the deepest of all questions that I've been working on is, goes back to the oldest part of our genome. So, uh, i- in our genome, what are called Hox genes, and these are morphogenes, and nowhere in your genome is the number five. It doesn't store the fact that you have five fingers.

    5. LFLex Fridman1:06:18

       Mm-hmm.

    6. NGNeil Gershenfeld1:06:19

       What it stores is what's called a developmental program. It's a series of steps and the steps have the character of like grow up a gradient or break symmetry.

    7. LFLex Fridman1:06:29

       Mm-hmm.

    8. NGNeil Gershenfeld1:06:30

       And at the end of that developmental program, you have five fingers.

    9. LFLex Fridman1:06:33

       Mm-hmm.

    10. NGNeil Gershenfeld1:06:34

        So you are stored not as a body plan but as a growth plan.

    11. LFLex Fridman1:06:43

        Mm-hmm.

    12. NGNeil Gershenfeld1:06:44

        And there's two reasons for that. One reason is just compression. Billions of genes can place trillions of cells. Um, but the much deeper one is evolution doesn't randomly perturb. Almost anything you did randomly in the genome would be fatal or inconsequential but not interesting.

    13. LFLex Fridman1:07:06

        Mm-hmm.

    14. NGNeil Gershenfeld1:07:07

        But when you modify things in these developmental programs, you go from like webs for swimming to fingers, or you go from walking to wings for flying. It's a space in which search is interesting. So this is the heart of the success of AI. In part, it was the scaling we talked about a while ago, and in part, it was the representations......for which search is effective. A- a- AI has found good representations. It's- does- hasn't found new ways to search, but it's go- found good representations of search.

    15. LFLex Fridman1:07:45

        And that's- you're saying that's what biology, that's what evolution has done is created representations, structures, biological structures through which search...

    16. NGNeil Gershenfeld1:07:55

        Correct.

    17. LFLex Fridman1:07:55

        ...is effective.

    18. NGNeil Gershenfeld1:07:56

        And so the- the developmental programs in the genome beautifully encapsulate the lessons of AI. And this is- it's embodied- it's- it's molecular intelligence. It's AI embodied in our genome. It- it- it- it's every bit as profound as the cognition in our brain, but now this is sort of thinking in molecular thinking in how you design. And so, um, I'd say the most fundamental problem we're working on is it's kind of tautological that when you design a phone, you design the phone, you represent the design of the phone. But that actually fails when you get to the sort of complexity that we're talking about. And so there's this profound transition to come, you know, once I can have self-reproducing assemblers placing 10 to the 18 parts, um, y- you need to not sort of metaphorically but create life in that you need to learn how to evolve. But evolutionary design has a really misleading trivial meaning. It's not as simple as you randomly mutate things. It's this much more deep embodiment of- of AI and morphogenesis.

    19. LFLex Fridman1:09:13

        Uh, is there a way for us to continue the kind of evolutionary design that led us to this place from the early days of bacteria, single cell organism, to ribosomes, and the 20 amino acids?

    20. NGNeil Gershenfeld1:09:24

        Y- you mean for human augmentation or...

    21. LFLex Fridman1:09:26

        For life augment- I mean, what would you call assemblers that are self-replicating and placing parts, what is that? The co- the- the dynamic complex things built with digital fabrication, what is that? That's life.

    22. NGNeil Gershenfeld1:09:41

        So yeah, so ultimately, absolutely, w- if you add up everything I'm talking about, it's building up to creating life in non-living materials.

    23. LFLex Fridman1:09:51

        Yes.

    24. NGNeil Gershenfeld1:09:51

        And I- I don't view this as copying life. I view it as deriving life. I- I didn't start from how does biology work and then I'm gonna copy it.

    25. LFLex Fridman1:10:02

        Mm-hmm.

    26. NGNeil Gershenfeld1:10:02

        I- I start from how to solve problems and then it- it- it leads me to, in a sense, rediscover biology. So if we go back to Valentina in Ghana making her circuit board-

    27. LFLex Fridman1:10:14

        Mm-hmm.

    28. NGNeil Gershenfeld1:10:14

        ...um, she still needs a chip fab very far away to make the processor in her circuit board. For her to make the processor locally, for all the reasons we described, you actually need the deep things we were just t- talking about. And so it- it really does lead you... So let's see, there- there's a wonderful series of books by Gingery.

    29. LFLex Fridman1:10:38

        Mm-hmm.

    30. NGNeil Gershenfeld1:10:38

        Book one is How to Make a Charcoal Furnace, and at the end of book seven, you have a machine shop.
11. 1:16:48 – 1:21:19

    Quantum computing

    1. NGNeil Gershenfeld1:16:48

       I mentioned the early quantum computing. So quantum computing is this power of using quantum mechanics to make computers that for some problems are dramatically more powerful than classical computers. Before it started, there was a really interesting group of people who knew a lot about, um, physics and computing that were inventing what became quantum computing before it was clear anything... there was an opportunity there. It was just studying how those relate. Here's how it fits to the ready, fire, aim, in that I was doing really short-term work in my lab on shoplifting tags.

    2. LFLex Fridman1:17:25

       Mm-hmm.

    3. NGNeil Gershenfeld1:17:26

       On... Y- this was b- really before there was modern RFID, and so how you put tags in objects to sense them.

    4. LFLex Fridman1:17:35

       Mm-hmm.

    5. NGNeil Gershenfeld1:17:35

       Something we just take for granted commercially. And there was a problem of how you can sense multiple objects at the same time. And so I was studying how you can remotely sense materials to make low-cost tags that could let you distinguish multiple objects simultaneously.

    6. LFLex Fridman1:17:53

       Mm-hmm.

    7. NGNeil Gershenfeld1:17:53

       To do that, you need non-linearity so that the signal is modulated, and so I was looking for material sources of non-linearity, and that led me to s- look at how, um, nuclear spins interact.

    8. LFLex Fridman1:18:08

       Mm-hmm.

    9. NGNeil Gershenfeld1:18:09

       Just- just, uh, for- for, um, spin resonance. This is the sort of things you use when you, like, go in an MRI machine. And so I was studying how to use that, and it turns out that it was a bad idea. You couldn't remotely use it for, um, shoplifting tags. But I realized you could compute. And so, um, with a group a- of colleagues thinking about early quantum computing, like David DiVincenzo and Charlie Bennett, was articulating what are the properties you need to compute, and then looking at how to make the tags.

    10. LFLex Fridman1:18:45

        Mm-hmm.

    11. NGNeil Gershenfeld1:18:45

        It turns out the tags were a terrible idea for, um, sensing objects in a supermarket checkout, but I realized they were computing. So with Ike Chuang and a few other people, we realized we could program nuclear spins to compute. And so that's what we used to do Grover's search algorithm and then it was used for a Shor's factoring algorithm, and it worked out. Um, the systems we did it in, nuclear magnetic resonance, don't scale beyond a few qubits.

    12. LFLex Fridman1:19:15

        Mm-hmm.

    13. NGNeil Gershenfeld1:19:15

        But the techniques have lived on, and so, uh, you know, all the current quantum computing techniques grew out of the ways we would talk to these spins. But I- I'm telling this whole story because it- it came from a bad way to make a shoplifting tag.

    14. LFLex Fridman1:19:32

        Starting with an application, mistakes led to breakthrough-

    15. NGNeil Gershenfeld1:19:35

        The fundamental science.

    16. LFLex Fridman1:19:36

        ... fundamental science.

    17. NGNeil Gershenfeld1:19:37

        Yeah.

    18. LFLex Fridman1:19:37

        I mean, can you, can you just linger on that? I mean, ju- just t- just the using nuclear spins to do com- computation that, like...What gave you the guts to try to think through this? The, from, from a fabrica- from a digital fabrication perspective actually. How to leap from one to the other?

    19. NGNeil Gershenfeld1:19:54

        I wouldn't call it guts. I would call it collaboration. So, I-

    20. LFLex Fridman1:19:57

        Ah.

    21. NGNeil Gershenfeld1:19:57

        So at IBM, there was this a- a- amazing group of, like I mentioned, Charlie Bennett and David DiVincenzo, and Rolf Landauer and Nabil Amir, and these were all gods of thinking about physics and computing. So-

    22. LFLex Fridman1:20:13

        Ah.

    23. NGNeil Gershenfeld1:20:14

        ... I- I- I- I yelled at the whole computer industry being based on, uh, a fiction, Metropolis, you know, programmers frolicking in the garden while somebody moves levers in the basement.

    24. LFLex Fridman1:20:25

        Mm-hmm.

    25. NGNeil Gershenfeld1:20:26

        There's a complete parallel history of, um, uh, Maxwell to Boltzmann to Szilard to, um, Landauer to Bennett, m- most people won't know most of these names, but this whole parallel history, thinking deeply about how computation and physics relate. So, um, I was collaborating with that whole group of people. And then, y- at, at MIT I was in this high traffic environment. I wasn't deeply inspired to think about better ways to detect shoplifting tags, but, you know, stumbled across companies that needed help with that and was thinking about it, and then I realized those two worlds intersected and we could use the failed approach for the shoplifting tags to make, um, early, um, quantum computing algorithms.

    26. LFLex Fridman1:21:14

        And this kind of stumbling is fundamental to the Fab Lab idea, right?

    27. NGNeil Gershenfeld1:21:18

        Right. Here's

12. 1:21:19 – 1:26:41

    Microfluidic bubble computation

    1. NGNeil Gershenfeld1:21:19

       one more example. With a student, Manu, we talked about ribosomes, and I was trying to build a ribosome, um, that worked on fluids so that I could place the little parts we're talking about, and we, it kept failing because bubbles would come into our system and the bubbles would make the whole thing stop working.

    2. LFLex Fridman1:21:38

       Mm-hmm.

    3. NGNeil Gershenfeld1:21:39

       And we spent about half a year trying to get rid of the bubbles, um, then Manu said, "Wait a minute. The bubbles are actually better than what we're doing. We should just use the bubbles." And so w- we invented how to do universal object with little bu- logic with little bubbles and fluid.

    4. LFLex Fridman1:21:55

       Okay. You have to, you have to explain this microfluidic bubble logic, please.

    5. NGNeil Gershenfeld1:22:00

       Please.

    6. LFLex Fridman1:22:00

       How does this work?

    7. NGNeil Gershenfeld1:22:01

       So-

    8. LFLex Fridman1:22:01

       This is really- (laughs)

    9. NGNeil Gershenfeld1:22:02

       Yeah.

    10. LFLex Fridman1:22:02

        That's super interesting.

    11. NGNeil Gershenfeld1:22:03

        Yeah. And so wh- so I'll come back and explain it, but what it led to was, um, we showed fluids could do, um... it had been known fluid could do logic, like your old automobile transition- transmissions do logic, but that's macroscopic. It didn't work at little scales. We showed with these bubbles we could do it at little scales. Th- then I'm gonna come back and explain it, but what came out of that is Manu then showed you could make a 50-cent microscope using little bubbles.

    12. LFLex Fridman1:22:30

        Mm-hmm.

    13. NGNeil Gershenfeld1:22:31

        And then, um, the techniques we developed are what we used to transplant genomes to make synthetic life, all came out of the failure of trying to make, uh, the genome, the- the- the ribosome. Now, so the way the bubble logic works is, um, in a little cha- channel, uh, fluid at small scales is fairly viscous. It's sort of like pushing jello, think of it as. Um, i- i- if a bubble gets stuck-

    14. LFLex Fridman1:23:04

        Mm-hmm.

    15. NGNeil Gershenfeld1:23:04

        ... the fluid has to detour around it.

    16. LFLex Fridman1:23:06

        Mm-hmm.

    17. NGNeil Gershenfeld1:23:08

        So now imagine a, a channel that has two wells and one bubble. If the bubble is in one well, the fluid has to go in the other channel. If the fluid is in the other well, it has to go in the first channel.

    18. LFLex Fridman1:23:22

        Mm-hmm.

    19. NGNeil Gershenfeld1:23:23

        So the, the bu- the position of the bubble can switch. It's a switch, it can switch the fluid between two channels. So now we have one element of switch, and it's also a memory because you can detect whether or not a bu- bubble is stored there. Then if two bubbles meet, um, i- i- if you have two channels crossing, a bubble can go through one way or a bubble can go through the other way, but if two bubbles come together, then they push on each other and one goes one way and one goes the other way.

    20. LFLex Fridman1:23:55

        Mm-hmm.

    21. NGNeil Gershenfeld1:23:56

        That's a logic operation.

    22. LFLex Fridman1:23:57

        Mm-hmm.

    23. NGNeil Gershenfeld1:23:57

        That's a logic gate. So we now have a switch, we have a memory, and we have a logic gate, and that's everything you need to make a universal computer.

    24. LFLex Fridman1:24:05

        I mean, the fact that you did that with bubbles and microfluidics just kind of brilliant, you know?

    25. NGNeil Gershenfeld1:24:12

        Well, so to, I mean, to stay with that example, uh, it- it- what we proposed to do was to make a fluidic ribosome and the project crashed and burned. It was a disaster. Um, this is what came out of it. And so it was precisely ready, fire, aim, in that we had to do a lot of homework to be able to make these microfluidic systems. The- the fire part was we didn't think too hard about making the ribosome. We just tried to do it.

    26. LFLex Fridman1:24:41

        Mm-hmm.

    27. NGNeil Gershenfeld1:24:41

        The aim part was we realized the ribosome failed but something better had happened. And if you look all across research funding, research management, it doesn't anticipate this. So fail fast is familiar-

    28. LFLex Fridman1:24:56

        Mm-hmm.

    29. NGNeil Gershenfeld1:24:56

        ... but fail fast tends to miss ready and aim. You, you can't just fail. You have to do your homework before the fail part and you have to do the aim part after the fail part. And so the whole language of research is about, like, milestones and deliverables. That works when you're going down a straight line, but it doesn't work for this kind of discovery. And to leap to something you said that's really important is I view part of what the Fab Lab Network is doing is giving more people the opportunity to fail.

    30. LFLex Fridman1:25:29

        You've, uh, said that geometry is really important in biology. Um, what does fabrication biology look like? Why is geometry important?
13. 1:26:41 – 1:35:27

    Maxwell's demon

    1. LFLex Fridman1:26:41

       Just stepping back, does it amaze you that from small building blocks where, um, amino acids you mentioned, molecules, let's go to the very beginning of hydrogen and helium at the start of this universe, that we're able to build up such, um, complex and beautiful things like our human brain?

    2. NGNeil Gershenfeld1:27:00

       So studying thermodynamics, which is exactly the question of, you know... Bat- batteries run out and need recharging.

    3. LFLex Fridman1:27:13

       Mm-hmm.

    4. NGNeil Gershenfeld1:27:15

       You know, e- equipment, you know, cars get old and fail-

    5. LFLex Fridman1:27:19

       Mm-hmm.

    6. NGNeil Gershenfeld1:27:20

       ... yet life doesn't and that's why there's a- a sense in which life seems to violate thermodynamics, although of course it doesn't.

    7. LFLex Fridman1:27:30

       It seems to resist the march towards entropy somehow.

    8. NGNeil Gershenfeld1:27:33

       Right. And so Maxwell, who helped give rise to the science of thermodynamics, uh, posited a- a- a- a problem that was so infuriating, it led to a series of suicides. There was a series of a- a- a- um, advisors and advisees, um, three in a row that all ended up committing suicide, that happened to work on this problem, and Maxwell's demon is- is this simple but infamous problem where, right now in this room, we're surrounded by molecules and they run at different velocities. Um, imagine a container that has a wall and it's got gas on both sides and a little door, and if the door is a molecular sized creature and it could watch the molecules coming, and when a fast molecule is coming, it opens the door, when a slow molecule is coming, it closes the door.

    9. LFLex Fridman1:28:33

       Mm-hmm.

    10. NGNeil Gershenfeld1:28:34

        After it does that for a while, one side is hot, one is cold. Once something is hot and is cold, you can make an engine, and so you close that and you make an engine and you make energy.

    11. LFLex Fridman1:28:45

        Mm-hmm.

    12. NGNeil Gershenfeld1:28:47

        So the demon is violating thermodynamics because it's- it's not, it's never touching the molecule, yet by just opening and closing the door, it can make arbitrary amounts of energy and power a machine-

    13. LFLex Fridman1:29:03

        Mm-hmm.

    14. NGNeil Gershenfeld1:29:03

        ... and in thermodynamics, you can't do that. So that's Maxwell's demon. Uh, that problem is connected to everything we just spoke about for the last few hours.

    15. LFLex Fridman1:29:14

        Mm-hmm.

    16. NGNeil Gershenfeld1:29:15

        So, uh, Leo Szilard, uh, around, um, early 1900s, was a deep physicist who then had a lot to do with, um, also, uh, po- sort of post-war anti-nuclear things. But he reduced Maxwell's demon to a single molecule. So the molecule, one, there's only one molecule, and the question is, which side of the partition is it on?

    17. LFLex Fridman1:29:44

        Mm-hmm.

    18. NGNeil Gershenfeld1:29:45

        That led to the idea of one bit of information. So Shannon credited Szilard's analysis of Maxwell's demon for the invention of the bit.

    19. LFLex Fridman1:29:55

        Mm-hmm.

    20. NGNeil Gershenfeld1:29:56

        Um, for many years, people tried to explain Maxwell's demon by, like, the energy in the demon looking at the molecule or the energy to open and close the door and nothing ever made sense. Um, finally, Rolf Landauer, one of the colleagues I mentioned at IBM, finally solved the problem. He showed that you can explain Maxwell's demon by, you need the mind of the demon.

    21. LFLex Fridman1:30:28

        Mm-hmm.

    22. NGNeil Gershenfeld1:30:29

        Um, when the demon open and closes the door, as long as it remembers what it did, you can run the whole thing backwards-

    23. LFLex Fridman1:30:39

        Mm-hmm.

    24. NGNeil Gershenfeld1:30:39

        ... but when the demon forgets, then you can't run it backwards, and that's where you get dissipation and that's where you get the vi- uh, violation of thermodynamics.

    25. LFLex Fridman1:30:51

        Mm-hmm.

    26. NGNeil Gershenfeld1:30:51

        And so the explanation of Maxwell's demon is that it's- it's in the demon's brain.

    27. LFLex Fridman1:30:57

        Mm-hmm.

    28. NGNeil Gershenfeld1:30:57

        So then Rolf's collea- colleague Charlie at IBM, uh, then shocked Rolf by showing you can compute with arbitrarily low energy. So one of the things that's not well covered is that the big computers used for big machine learning, the data centers use tens of megawatts of power. They use as much power as a city. Um, Charlie showed you can actually compute with arbitrarily low amounts of energy by making computers that can go backwards as well as forwards. And what limits the speed of the computer is how fast you want an answer and how certain you want the answer to be. But we're orders of magnitude away from that. So I have a student, Cameron, working with Lincoln Labs on making superconducting computers that operate near this Landauer limi- limit that are orders of magnitude more efficient.Um, so stepping back to all of that, that whole tour was driven by your question about life. And, uh, you know, right at the heart of it is Maxwell's demon.L- life exists because it can locally violate thermodynamics. It can locally violate thermodynamics because of intelligence and its, its molecular intelligence that, you know... I would even go out on a limb to say we can already see we're beginning to come to the end of this, eh, current AI phase. So depending on how you count, this is, I'd say the fifth AI boom bust cycle.

    29. LFLex Fridman1:32:36

        Mm-hmm.

    30. NGNeil Gershenfeld1:32:37

        And you can already... You know, it, it's exploding but you can already see where it's heading, you know, how it's going to saturate, wh- what happens on the far side. Um, th- the big thing that's not yet on horizons is, is embodied I- AI, molecular intelligence. So to step back to this AI story, um, th- there, there was automation and that was gonna change everything. Then there were expert systems. Um, uh, there was then the, you know, the first phase of the neural network systems. There've been about five of these.

Episode duration: 2:07:04