Full Transcript for Diffraction Limit, Microscopy, and Cell Biology | Eric Betzig on Super-Resolution Microscopy

Almost everything you learn in biology textbooks is a hallucination. You guys have probably seen on the web, you know, there's those beautiful things of here's a cargo on a kinesin walking along a microtubule like this, and it's all like in this vast empty space. I don't know any cell that's a bunch of vast empty space. I'm sorry. It's crowded as f***.

There's 10 billion protein molecules. There's 10 billion carbohydrates. It's by far the most complex matter in the known universe. We understand the interiors of neutron stars far better than we understand the interior of cells. There's a reason why only 9% of the drugs that enter phase 1 come out phase 3 because we don't know what the f*** we're doing.

We don't know the real mechanisms that are going on. And when you start to f***ing look at the dynamics, not just the structure, you realize that you had it all wrong. And you realize that so many of the things that they thought they knew, you can't be sure that they know. We have to reinvestigate all of it. This week, Mike and Misha sit down with Eric Betzig, Nobel laureate and UC Berkeley professor who pioneered super -resolution microscopy.

Eric traces his path from Cornell to Bell Labs, including two stretches of unemployment that produce some of his best ideas. He reflects on his work in the auto industry and the living room where he built a microscope to beat the diffraction limit. Now he's looking ahead to his new cell observatory, using AI to turn petabytes of live cell imaging into a queryable model of biology.

Please welcome Eric Betzig. So, Eric, you've been working, like, battling with diffraction limits throughout your career. What is the diffraction limit? How would you explain it? Well, it was first figured out around the end of the 19th century. And it kind of makes sense that if light is waves and you're trying to get an image through light, then you can think of the light colloquially as like little fingers trying to feel the sample, right? And if the fingers are too fat, you're not going to feel the structure below the level of those fingers, right? And so basically about half the wavelength of light is what Ernst Abbe mathematically determined is the limit.

You know, he did it. But, you know, as soon as people understood the wave nature of light going back to Huygens or certainly Maxwell, all of this could have been easily deduced. But he was the guy who really realized from a real imaging perspective that that was a fundamental limit. You know, in many fields, of course, you know, when I first started getting in in the 80s in graduate school, there was obviously already semiconductors and other structures that were pushing those same limits in terms of their manufacture and so forth.

So it was already known as an issue. And, of course, this is why the electron microscope was invented in the 30s with Ruska and others, right, is to get around that limit.

In biology, it was perhaps less appreciated that you'd necessarily want to go further. But, you know, the diffraction limit is such that it's roughly 100 times smaller than a cell. So you can still learn a lot. But it's 100 times larger than the molecules that make up the cell. So, obviously, there's an incentive to try to do better. Maybe if you could give us, like, some taste of, like, maybe some numbers in terms of, let's say, there's, like, available on the shelf microscope someone can get, like, with objectives.

Right. And how small those numbers are. So, yeah. So with visible light, you're going to be down to around 200 nanometers. Okay. So, again, a protein molecule might be 4 nanometers. So you're 50 times too coarse to get to that level. But a cell might be 20 microns, which is 20 ,000 nanometers. So then 200 nanometers means you're roughly, you know, can see 100 spots across the width of a cell.

But you have to have a good microscope. I mean, the kind that you'll have in a high school is probably not going to cut it. Right. You need good objectives. It's an art and a science for those guys. And that was really the thing is, in fact, there's an interesting story is that when Abbe came up with the understanding of this limit, not only did he do that, but he also understood exactly how to create objectives, lenses that would actually reach that limit.

He was a professor at University of Jena, but he was working with Carl Zeiss, who had his microscope company in Jena. And so he had all these prescriptions about how many lenses, how to grind them, you know, how to space them exactly to get to the diffraction limit. Zeiss goes and follows those exact rules, and it was crap. And they're like, why is this the case? And they realized it was because the quality of the glasses were not good, okay? That, you know, this is often the way as an experimentalist, you have a mental picture in your head, but reality bites you in the ass.

And so they went to Otto Schott, who was making glasses, particularly in that time there was a lot of street lamps that were gas lamps. And so he was making the borosilk head glass that covers the lamps, and they went to him and say, hey, can you make us better glasses? And so basically Schott figured out exactly how to make exactly the types of refractive indices they needed.

Once they did that, bang, they were right to the diffraction limit. And so, you know, Abbe, he did okay. Zeiss, he made a fair amount of money. But the real winner was Schott because Schott is today still one of the largest makers of glasses for every application in the world. Schott and Corning are the two big names in this stuff, right, to this day.

And so he became uber rich because of his ability to make these glasses. And was it like impurity issues? Impurities was part of it, but also because you have to introduce impurities as well specifically to get certain refractive indices, right? You need to go anywhere from, you know, glasses is normally 1 .5 index, and they would go down to below 1 .4 all the way up to 1 .9, right? And so you have a lot more tools at your disposal as an optical engineer if you have all these different refractive indices to make your lenses from.

Why did it matter that much? Because they could also use, like, you know, the curvature. It's not enough. You don't have enough knobs to tweak to get to where you need to be with that alone. You actually need to be able to have different indices of glass to make it work. Where are we now with objectives? Oh, I mean, so there's many, many fringe benefits of winning the Nobel Prize.

So one of them is I've had a long relationship with Zeiss because they licensed our patents on Palm. They licensed our patents on Lattice Lightsheet. But normally I go to Jena, but one trip I got to go to Oberkachen. And Oberkachen is where they make the big EUV lenses that then go into the ASML machines that then go into making NVIDIA chips and so forth.

And that is the closest I've ever seen to alien technology in my life. I mean, these lenses are, you know, taller than me. They're this fat around. They have– they're all reflective lenses because you can't use glasses there, right? But the coatings to make it reflective at EUV wavelengths, they have like 30 layers. And because of the curvature, even though it's like sub– a monatomic level of coating thicknesses varying across the diameter.

And then they have to put little impurities in so the atoms don't diffuse across the boundaries between the different layers. And this whole thing is literally gigantic and it's precise down to the atomic level in what they do. It's just absolutely– you wouldn't believe that this would be physically possible. But they do it and it's incredible.

Have you thought about a biological imaging microscope based on these giant mares? Could you imagine one? Would it help? Well, if a living cell could live under EUV wavelengths, life would be great, wouldn't it? But sadly, that's not the case. I guess you could image like cryo samples, right? Yeah, yeah, yeah, yeah. Well, one of my main points that I make in all my talks today, right? And one of the reasons I pivoted from structural super resolution is it's my firm belief that you cannot understand life without looking at it live.

And I feel like a big problem in biology today and in pharma today is that there's been– because of the diffraction limit, because we reached this wall at the end of the 20th century and it was 100 times too small to see the molecules, science pivoted from this holistic look at cells with– that they could look at live cells to doing reductionism.

And EM is one of those reductionist tools. You're doing structural imaging. But you will never understand life without looking at it live, okay? So having high– everybody was thinking, what's the point of a microscope? High resolution. No. It's one of the things you're interested in. But you're also interested in speed because the cell is moving.

You're interested in non -invasiveness. So what's the point of looking at a cell if you're killing it while you're looking at it? So all of these other metrics matter too. And so UV is not the path to glory to study living things. In fact, while we've learned a lot from the reductionist tools like EM and particularly like structural biology like cryo -EM and alpha -fold and all of that, there's a whole slew of companies five miles across the bay here trying to do drug discovery based on protein structures determined by alpha -fold.

So those proteins and that stuff is such an infinitesimal part of the whole dynamic complex system that creates life. They're burning money for nothing. I guarantee that unless they pivot to pull in other types of information, biological information, those companies will not be here in five years. So what kind of information do you see needs to be there to develop therapeutics? To be able to see the dynamics and to see the interactions that actually happen in living matter.

That's the only way. Imagine you try to reverse engineer an internal combustion engine. And if all you have is the reductionist tools that rule biology today and ruled in the 20th century, those are biochemistry, how different proteins and other molecules interact with one another, structural biology such as alpha -fold or EM or stuff like that, and molecular biology, the central dogma, knowing DNA makes RNA makes proteins and all of that.

But those are incredibly reductionist tools. It would be like trying to reverse engineer life from that would be many, many orders of magnitude harder than having a random pile of internal combustion engine parts and trying to figure out how internal combustion engines work. That's where we are today. That's where most of the field is. And it drives me insane because I keep making this point over and over again that we need to go back to a holistic understanding of this complexity that exists in the cell instead of being focused on just little parts.

So you mentioned Icon is a company we have. So the head of Icon is Roger Perlmutter, who used to be head of research at Merck. He's one of the most prolific and successful drug discoverers of the 20th century. And he has a line which is, you know, it's a miracle if we ever find a drug that works because we have no idea what we're doing. And that is accurate because of this focus on such a crazy level of reductionism instead of understanding the system holistically.

You know, we were discussing this yesterday. Like, I mean, obviously, we've kind of skipped a lot of the what is super resolution microscopy or anything. But is it do you have you ever thought about or like have you ever kind of like come across an idea of somehow making a cell where it's genetically engineered to be friendly to microscopy? So meaning like you genetically engineer as many proteins, as many structures as possible in the cell to be easier to image so that you can image like all of them.

I mean, in the in the limit, it would be like they're all fluorescent in different wavelengths. Yeah, right, right. Well, that's that's exactly the way we do engineer cells today. Right. Is to put fluorescent tags on specific proteins so they'll light up. And that's one of the most limiting parts of optical microscopy still, because in in visible wavelengths, there's a limited number of colors that you can do.

And there's 20 ,000 different types of proteins in the cell. It would be great if we could see them all at once. But that technology does not exist. And that is a Nobel waiting to happen if somebody can directly interrogate proteins in some way and deduce them without having to put those tags on. But more generally, if you start modifying a cell in order to make it tractable to image it, you're not looking at real cells anymore.

You're looking at the thing you created. Right. And and so, yeah, that's I don't think that's a path. Is there have people try this where they try to like make the proteins more absorbing or more some more maybe like I mean, the thing is, like, if you want to add, let's say, a green fluorescent protein, that's a big protein. So it's going to start changing.

Of course. No kidding. It's a bowling ball. But maybe there's a two kilo Dalton bowling ball added to added to your protein. That's right. Yeah. Is there is there any has that actually has that type of approach of actually genetically engineering the cells yielded any kind of like interesting results? Yeah. Well, I mean, again, we genetic engineer all the time to put those labels on.

You put the you you you you do also all sorts of as many controls as you can to try to show that the cell is still behaving physiologically with that particular tag on. And a lot of times you'll find it isn't OK. And a lot of times it's a black art about the linker, for example, the little the little peptide linker between the protein and the bowling ball.

And and all sorts of crap that you have to just it's it's an art, not a science in terms of figuring out what works and recapitulates as much as we can tell the native physiology. But there's it's it's it's just it's just like, you know, the uncertainty principle. Anytime you observe anything, you're going to perturb the thing. And so it's really, really important to make sure that you do as many controls as you can to to believe your physiology.

So what kind of like cells or objects, organisms you are targeting? So many, many, many things. So we start, of course, with with cultured cells because that's the easiest. Right. And certainly with palm and so forth, that's where we had had to start. Super resolution started with with single culture. In fact, at first, it starts with dead cells, fixed cells, chemically fixed so that so they're not wiggling around because the super resolution techniques, at least initially and still in many cases today are slow.

And so things would blur out if you try to look at super resolution of something that's wiggling and moving real fast. So you chemically lock it into place and then you image that. Right. So that's where you start. But those fixation protocols, which started with electron microscopy, are incredibly perturbative to the ultra structure at the nanoscale you want to see.

It's fine if you're looking at at regular optical resolution. But when you when you dig down the super resolution, you're looking at artificiality. And that was one of the reasons I pivoted from super resolution, because most of the time people electron microscopy has known this for years. And so their gold standard for them is not to chemically fix cells.

It's to high pressure freeze cells so that in in milliseconds you're locking in the structure in vitreous ice, not crystalline ice, and then look at it that way. But that's a lot of work to do. Harold and I have done it at Janelia and done correlative EM and super resolution. But it is not it's not a protocol for the faint of heart. It takes a lot of technology and a lot of skill to make that kind of thing work.

You try to combine both microscopes. Yeah. Yeah. Correlative system. Yeah, yeah, yeah. So, well, you go from one to the other. Right. So so you vitreously freeze the cells. Then you put them in your super resolution cryostat that's now at four degrees Kelvin. And so you do cryos cryopalm, we call it. Right. And you and you bleed out all the molecules, find out where they are.

Then you take that and then you take it out. While still frozen vitreously, you do freeze substitution and you basically end up putting it in resin in a way that still preserves that ultra structure. Then that gets sliced and put in the Herald's three dimensional focused ion beam milling electron microscope. So then you can get the three dimensional profile that way.

And then you take the two data sets, which, of course, have all sorts of like weird little distortions and stuff. And then you do kind of a warping and register them all together. And then you got your core loops. That's this one right up there. You see that? That's correlative super resolution. And that orange thing with that. I see like overlapping them there.

Yeah. They're exactly mapped on one another right there. Yes. So you have to slice basically the. In the EM. So there's a focused ion beam, like a gallium beam that comes and takes a nanometer or two off. How thin is the layer? Like a couple nanometers, really thin. And then you image that, you image that, you image that. And Herald is the god of fibs.

So so connectomics is a big thing in neuroscience right now. So one of the first connectomes was made possible by Herald's fibs at Janelia, where I used to work. So that has to be done a couple nanometers at a time. You couple of nanometers, image that surface, couple nanometers, image that surface. And it's less crazy than those lenses that I talked about at Zeiss, but it's pretty crazy.

Not very scalable. It's not terribly scalable. It's not going to be a consumer product anytime soon. No, no. But none of what we're talking about is. Yes. Maybe we can go back to like. Sure. So you're jumping all over. You're a physicist, right? So how did you get into all this like wet, you know, stuff? So so in curiosity. Right. So, yeah, I started as a physicist and, you know, as a kid, I wanted to be an astronaut because I grew up with Apollo and Star Trek.

Right. And but by the time I was in college and by the time I graduated from college, the shuttle was coming online. And I knew at the time the shuttle was the biggest mistake you could possibly make. It was a horribly engineered system, a complete waste of money. Nothing but nothing but a political gift to the legacy aerospace contract. Did you like the Soviet rockets better? Well, the Soviet the Soyuz has been flying for, what, 50 freaking years or something.

Right. That's pretty good. Right. But obviously, SpaceX has completely changed the nature of the game. Right. So this is very exciting to me because I think that, you know, I honestly believe there is a finite chance that I could still reach my dream of making it to space before I die. And that is the number one thing on my bucket list is to make it to orbit.

Yeah. So you have to take a break from that dream. Yeah. But but anyway, anyway, so getting back to your point, how did I get into the wet and squishy stuff? Right. So I went to graduate school and I've always wanted to be like I wanted to do physics, not because I wanted to be a Feynman or anything. I wanted to do engineering physics. OK, I wanted you know, my dad was an engineer.

I wanted to be an engineer. I like making things. And so at that time, there were only two applied physics departments in the country. One was at Stanford and I hated California. I wanted to get the hell out. And then the other was Cornell and Cornell was a lot more like Michigan where I grew up. So I went to Cornell and there was two professors.

One was a guy who was an electron microscopist and the other guy was a Raman spectroscopist. And they had this crazy idea that if you could use your electron microscope to drill a hole smaller than a wavelength in a black film, then you could press it against the cell and you would have a little nano flashlight that would only illuminate one spot.

You drive it around. You get a super resolution image. And, you know, the idea is what if we could make a microscope that could look at living cells with the resolution of an electron microscope? All you have to do is say that sentence and you go, oh, my God, that would be incredibly revolutionary. Right. And you say, OK, that's the kind of thing I'd like to engineer.

So that was my entrance into working super resolution for my thesis. And so I did near field microscopy, as it was called, because you're in very close to the aperture. You're not in sort of the sort of far field optics limit. You have to get into the details of the of electric fields that decay exponentially away from the thing. So that's called near field optics.

Did you use a fiber? I and not at that at that time we used. So one of the things that was new at that time was something something in electrophysiology called patch clamping, where you could basically look at single ion channels in a membrane by pulling glass pipettes down to the nanometer scale and then pop that thing as an electrode on top of the ion channel, actually measure sort of single ion currents going through pores in cells.

OK. And so we realized that, well, if we coat that thing with metal, then the hole in the end of that glass thing is going to be our little aperture. And that's a lot easier than having to go to the electron microscope and drill these holes in this in this thin film. And the thin film, if you looked at it the wrong way, would shatter. And so the pipettes were a lot more robust and simpler.

And so that's what I did. And then I took that technology in my own lab at Bell Labs and then realized once I was there that we could get much better delivery of light by instead taking optical fiber. And pulling them like a piece of taffy. And then it would taper down and break with cleave at the flat end and then illuminate the sides. And then it would be optically guided in the wave guide down to that aperture.

So we'd have a much brighter light source. And then that was where I developed my reputation as a super resolution guy. And the experiments I did there, although I didn't know it at the time, kind of set the stage for what would become Palm and other super resolution. How small can you get in terms of the 20 nanometers? I see. And that was the resolution limit.

Yeah. Yeah. Tough to get. The trouble, the key thing that killed near field for biology is that, again, because it is near, the light that comes out of that hole spreads incredibly rapidly. So if you're even 20 nanometers away from the surface, you know, you've lost much of your resolution. And I didn't know a lot of biology at the time, but I knew cells were a lot rougher than 20 nanometers.

Right. And so there was there really wasn't any way I was going to be able to follow, even get the surface and follow the contours of the surface with this this big ass probe and and get the resolution that I needed. And there was no way I was going to get to the interior of the cells. So I did some experiments on cells then, but I picked cells that were known to be crazy flat.

Right. And not the whole cell, but just out at the periphery where it's just kind of, you know, and demonstrated that I could do it. But it was really for biology, pretty much a dead end. By the way, related maybe to diffraction limits, seems like it's a good demonstration that if you have a tiny hole, like 10, 20 nanometers, like 10 times smaller than the wavelength, that the light wants to expand.

Is there really, really fast? I mean, I can derive it like, let's say, from the formulas. But as an effinescent field, it's exponential, right? Right. Is there intuitive explanation for it? I don't know. Not really. Unless you want to think of it as like a fire hose, right? Or something like that, right? If that's your level of intuition.

But generally speaking, you know, it was something that was a concern of mine from day one. In fact, the first thing I did when I got in there as a graduate student is I couldn't do a round hole, but I developed a theory for figuring out how fast the diffraction is from a slit. And I did that on an IBM PC, the 8088. And so I was able to do that calculation.

I said, yeah, it's pretty fast. But I didn't necessarily have enough confidence to think that, well, let's just do the experiment and find out how bad it is. How were you getting the probe so close to the cell without, like, how do you, like, five nanometer type controls, right? So what you need is an independent feedback mechanism to regulate the height of the aperture above the surface.

So there's two answers to that question. The first is if you– at Bell Labs, I had lots of hits. I had a number of, you know, science papers, you know, at one– you know, I became well known as a scientist at Bell Labs. But if you look at all of the applications I published, they have one thing in common. The samples were really, really flat.

Okay. So we did, like– we had the world record for high -density data storage at one time with near -field optical data storage because you're looking at very flat, you know, ferroelectric film that you can switch the bits in, right? And– but still, it was limited to very flat things. So your samples were the magnetic samples? Yeah, yeah, yeah, yeah, yeah, yeah, exactly.

It seems like it was a big theme to develop data storage systems. Oh, yeah, yeah. You know, back I, you know, flew out here and had talks with Seagate about commercializing and shit like that. None of it ever went anywhere. But at least at the time, it was something that was worthy of thinking about, right? But to get back to your question, in order to get that regulation, that I developed at Bell Labs.

And it turned out– so this was– near -field was a form of what is called today's scan probe microscopy. The most famous example of that is scanning tunneling microscopy, right? And that's the one that– that won the Nobel in 86. That was while I was– where was I? I was still in grad school in 86, right? And so at first, we were trying to– you know, because our tips were– our near -field tips were coated with metal in order to make them opaque.

Well, now I can try to use that to do tunneling microscopy against the sample if I have a conductive sample. That didn't work so well because those tunneling distances are not nanometers. They're angstroms, which gets even harder, right? But another technology that developed quickly out of scanning tunneling microscopy was something called atomic force microscopy, where instead with a tip that's attached to a spring, you can kind of feel the forces at maybe not the atomic level but sort of at the nanometer level, you could kind of do that.

And so the trouble was that was done with a tip on a cantilever, so the cantilever is very floppy, right? And my probe has to go this way and it's stiff this way, right? And so what I did instead is I dithered it in the floppy direction and then as I came close to the surface, there would be enough forces between the tip and this that it would slightly change the resonant frequency of that thing.

It's like damping it. Damping it, exactly. And so you could see that slight change in the damping frequency and use that as a regulatory. That worked really well. So that worked on cells, worked on everything. So sheer force feedback is a thing today and that was developed specifically for Nearfield first, yeah.

When did you realize that like Nearfield wasn't going to do it? Because eventually I got to the point after six years in which I pretty much exhausted every flat thing worth looking at.

Okay. But at the same time, and this gets now into sort of philosophy of science and why I'm not really a scientist, right, is I dropped Nearfield in part because I dropped science altogether. I got really fed up with science because when we first started in graduate school to do Nearfield, everybody told us we were nuts. There's no way. Fraction limit.

Blah, blah, blah. This will never work. And it was enough to convince the people of Bell with the results I had to do it. And it just took off. I mean, again, it was a terrific– best time of my life was at Bell. And things were going great. But I knew the limits, right? And they were real. But at the same time, science is very faddish, okay? And every new thing, just people jump on, right? Just tons.

And I don't mind that. That's fine. But they have to be careful, and they're not, okay? And most people who jump– there's– first off, if you're willing to drop what you're doing as a scientist to chase something else, what was the value of the thing you were working on in the first place, right? What is motivating you in order to do this? It's fine if there's an opportunity– particularly if that opportunity reads upon what you were doing.

But if you're just like, whoosh, let's do this all of a sudden, right? So the field blew up overnight. It was part of the blowup of scan probe microscopy in general with STM, with AFM, with near field, all of this. But the trouble is is that most of the people are not careful. And one of the things– one of my stock phrases from that era was, it's very easy to get an image and very difficult to get a meaningful image.

So it's very easy to see artifacts in your images all the time and things. And this was just endemic. And I kind of felt like every good paper that we published was the justification for 100 pieces of crap that followed in its wake. And everything that I was doing was a net negative to society and a waste of the taxpayer's money. So that's the way I still feel about most of science.

Do you have here an image taken with that? With near field? No, I don't think I have a near field one anywhere. Nope, sorry. Yeah. It's gone. I mean, I have them somewhere. I have some– oh, actually, you see those two big -ass black books? Right above where he is there. That's my thesis, one of the biggest theses in the history of applied physics at Cornell University.

Later on, if we want, and I can give you– there's plenty of images in there from my near field microscopy for now. So– And it was written without LLMs. It was– yeah. It was– you know what it was written on? A Wang word processor. There wasn't even– you know, didn't even have a PC to do it on then. So, yeah. Yeah. I think this complaint about the, like, people publishing essentially artifacts, right? Yeah.

Right. Like, I've heard it before, actually, from an AFM guy that I'm friends with that he worked on– he had a job for a while doing, like, hero experiments at Bruker. Mm -hmm. So, they want to show, like, the power of the microscope. Exactly. So, he would put a lot of effort into, like, let's say, imaging the helix of DNA or whatever. Right, right, right.

But then he was saying that, like, yeah, if you look at papers, like, you're often just looking at artifacts. Absolutely. Absolutely. Yeah, what's– ScanPro microscopy is loaded with artifacts. Yeah. Is there, like– do you have, like, a hack that would, you know, help reduce the amount of artifacts in literature? Have you ever thought of, like, if you could, like, wave a wand? So, one of my favorite lines– and I only learned it a few years ago, but I think it explains everything in the world– is from Charlie Munger.

Show me the incentives and I'll show you the outcome. Okay. So the incentives in science are not incentives to increase the store of knowledge. They're to get grants, to get awards, to do things of this sort, right? And if one can do that… Receive the Nobel Prize? Yeah, whatever. Yeah, yeah. I can rant forever about how toxic I think awards are to science.

Every award that's ever given out, there's 50 people who think they should have gotten and are bitter about it, right? And it's all subjective. It's a beauty contest, right? Right? Who's to… Science is a collective work of many, many people. Who's to say one guy should get all the f***ing credit for what an entire field has done? Or worse, a guy who runs a lab of 50 people, right? These giant supergroups, right? Where it's all like, you know, field marshal at the top and lieutenants and this and that and that all the way down.

Some kid down at the bench has some ID and the guy up here gains all the credit, right? I mean, what the f*** is that, right? The whole system is frankly corrupt. Were you already… Is this what you were thinking when you were leaving IBM? Like when you… No, IBM Bell Labs. Sorry, Bell Labs, yes. I was thinking this when I was in graduate school.

Are you kidding me? The scales fell off my eyes pretty damn early, right? But I felt that… This was a lot of the reason I didn't just leave Bell Labs. I didn't just leave Nearfield. I left science, right? I mean, I was done with it, right? So, yeah. What did you do? Where did you go? I worked for my dad. So, he… Timing was good because he worked for a machine tool company in Michigan, which… So, a machine tool is… So, Michigan, of course, is home of the big three in the auto industry and they were producing millions of cars a year.

So, that means you need to make millions of brake calipers a year and you may need to make a million intake manifolds a year. That's a lot, okay? You don't do that by monkeys typing out Shakespeare by having a thousand guys each on its own milling machine and personally doing that. You have to have automation. So, CNC machines existed then but they were not robust, okay? You could not produce the million parts a year with the technology with CNC, computer numerically controlled machine tools at that time.

So, you… What… Where my dad worked is you actually design the machine specifically for that part. So, two years before the car ever is in the showroom, the big three says, okay, we've designed this. This is, for example, the brake caliper, okay? They go to these different companies and they say, quote for me a machine customized specifically to make a million of these brake calipers a year.

And so, they do that and that machine will be, you know, ten times the size of this room, okay? And because it has… It's working in parallel doing… Here it's doing a milling operation. Here it's doing a drill. Here it's tapping at different stations as it goes around. And every 30 seconds, pop, part comes up, pop, part comes up. So, these machines even back then cost hundreds of thousands to a million dollars.

And you have a year in order to get that machine running and it's got to run 24 -7 on the floor or you are out of business because they will never buy from you. If you stop the assembly line because they don't have that brake caliper, okay. So, you have to make really, really robust machines. So, my dad did that and then he decided to make his own company doing that.

And so, that happened while I was at Bell. Right when I started Bell, he had left the other company and started a competing company. So, that was 1988. You guys are probably too young to know but 1990s was the era of us boomers becoming real consumers. The minivan was invented. You know, the SUV was invented. It was boom time for the auto industry.

So, my dad's timing could not have been better. And so, his company– so, by the time of 97, he had 300 people and was doing 70 million in sales from zero, right? And like, you know, that's great. And so, he could have– when I left– so, therefore, in 95, when I left Bell, they were already on their upswing and he could afford a leech like me to come work there.

And try to see if there was some way I could use, you know, my physics experience and like that to make machine tools better. And so, you know, when I was completely fed up with science, I always kind of knew I would have this as a potential backup because he was always saying, I want you to come work for me.

And eventually, it was like, okay, dad, yeah, I'll come work for you. But can I just kind of try to be the guy who's– your R &D department, okay? I'm going to try to think if there are other ways of doing things or other markets we could do. And so, he was willing to humor me with that, right? And so, I did that and I did that for six years and I developed two projects in that time.

The first one was when the parts come off the machine, they had damn well better bolt together than the rest of the parts, right? And so, they couldn't inspect every part but what they do is they take every one in a zillion parts and they go to a clean room where there's something called a coordinate measuring machine. This thing that's on a gantry and it has this little sapphire ball at the end and it touches the part here, touches the part there.

And it kind of feels its way around and figures out to sort of 10 microns or so precision where everything is, right? But you're only looking at one out of a zillion parts and then you hope statistically you look at enough parts and you hope the variability of the process is good enough. It wasn't. It clearly wasn't, okay? And so, what I came up with was because I was still filled with optics and still filled with super resolution, right?

You know, I knew, in fact, one of the two experiments I did at Bell which led to the idea for the Nobel Prize was I was the first person to see single molecules at room temperature. And furthermore, I was able to localize their positions to about a 50th of the wavelength of light because I knew what the energy distribution inside of the near field aperture was.

So, I could fit a profile to that and therefore figure out where the molecule was to much better than the width of the aperture itself by doing that fit, right? Just like you can find the center of a Gaussian to much better than the width of a Gaussian if you have enough signal noise in the nature of that Gaussian. Did you do it with near field microscopy? I did that with near field microscopy.

How do you also prove that it's a single molecule? Oh, that's easy because there were two things that were really conclusive on that. I love that. That was probably one of the best papers of my life. The first one is it bleaches. It doesn't go down slowly where the molecules bleach. They're there and then the next instant they're gone, okay? So, you fried it, okay? The other is that fluorescent molecules are dipoles.

That's how they work. And so, it had a dipole orientation. So, actually what happened is because of that dipole orientation, it was actually mapping out the electric fields in that subwavelength aperture. So, I could actually map the fields in the aperture to almost one nanometer precision as I drove it along that molecule. And I compared it to a theory that Hans Bethe developed in the 40 for what diffraction would be through a subwavelength hole and it matched up exactly.

It was an awesome paper. So, you could see the dipole near field distribution. And I could turn that around and since I knew from Bethe's theory what the electric field profile was like, I could not only determine the position of the molecules, I could determine the orientation of every f***ing dipole.

So, it was really cool. Is it the single dipole? Single dipole. No, no, no. Fluorophores are generally just a single dipole. And how do you collect the fluorescence in that situation? Is it through the same fiber? You can either come back up or you can have a detector on the opposite side if it's on a transparent substrate. Got it. Yeah. The other key thing then was having these avalanche photodiodes.

So, before that was all photomultiplier tubes which sucked both in terms of quantum efficiency and noise. But the avalanche photodiode absolutely changed imaging when that came on the scene. And so, again, it was being in the right place at the right time. And one of the other things I would say is one of the reasons I've been successful as a scientist is I've always been one of the first adopters of new technology.

I've always got my sniffing around for whatever is the latest widget that I can– and how combinatorically it adds with all of the other widgets and all the other experience I have to do something new. How would you explain it? How do you do it? Like to be in the right place at the right time? Again, it's– I hate going to conferences. But I go to conferences to go to the trade shows, right? Because that's where you kind of learn what's the new when you talk to the people.

You even learn things that aren't quite– they're not quite ready to release yet and so forth. And so, yeah. I started actually to do the same. Yeah. Okay. There you go. For wallets. Yeah. Absolutely. Yeah. I learn way more from those guys than I ever learn in any talks at something like the optics conferences and Moscone or whatever, right? Yeah.

Yeah. But how did this– you were going to talk about how that microscopy work translated to the CNC inspection? Oh, yeah, yeah, yeah. We got off that thread. Exactly. So, instead of just measuring one part every so often, I developed this thing that had about 30 little cheap CCD cameras that were all around the thing. And then I could measure to sub -pixel precision where, say, the edge of a part was or where a hole was located or all of that.

And I could do it in a fraction of a second. So it could be done for every part. So, therefore, we can look at every part that comes off the machine and we can reject those parts that don't fit the tolerances of the blueprint, right? And so I built one of those. We put it on a machine that we had shipped to New York.

And what happened? It starts rejecting parts, right? Because what happened is the tolerances that the companies would spec– this was the era of GE and the Six Sigma bullshit with Jack Welch and all of that, right? Six Sigma is crazy. You know what Six Sigma means in terms of precision? It's like 10 to the minus 6 or something. It's crazy shit, right? I mean nobody can achieve that type of stuff, right? So basically it's rejecting parts because of this, right? And tacitly everybody knew that, right? And so with the old method of just using the coordinate measuring machine to statistically get some parts, as long as the f***ing things bolted on to the axle, life was good,

okay? And so they spec them so tight that even if you were a factor of five off in terms of tolerance, it was still plenty good that it would end up bolting onto the thing right. That was the way the world worked. And so what they did is because it rejected parts, right? They turned it off and then they let the parts go through and then they measure them the old way.

And so that was my first lesson is that, okay, well, the thing that is king is productivity. I developed something which hampered productivity, right? So instead for my second thing I did with my dad is can I do something that increases productivity, okay? I've learned my lesson, okay? So there was an opportunity there as I learned is that, you see, in order to have the robustness that these machines needed to have, they were incredibly big, okay? So you're talking about each– and it would do stuff in parallel.

So if you have like on one surface eight holes you're being drilled at once, you make a spindle that has eight spindles together linked together and in the exact positions of where those holes will and it all goes in in parallel together and it drills all of those holes at once, right? And so that's the only way you get the speed. And so you're moving masses for these– in order to make it robust, in order to make aggressive milling cuts across cast iron or something, you need to make things that are really heavy.

So each station around these machines might be a ton apiece, right? Maybe more because, again, you're making three axis moves, right? So you have three different– one axis on top of the other, on top of the other to make these moves. And so it's like a rocket, right, where you have three stages, right, in order to make this type of move.

And so the bottom stage is moving all the upper stages, right, in order to do all that. So it has to be even bigger and more robust in order to do that, right? So that's how it worked.

And in that day, you had two options for trying to move this stuff. You could use a big electric motor, which was tied to a ball screw, and the ball screw would then move the stage forward. Or you could use hydraulics and use a hydraulic cylinder that would just push it forward. So you could use either a hydraulic cylinder or a ball screw.

The trouble with the hydraulics was it was cheap, but it wasn't very precise, okay? So generally what you do is you just move it forward until you come up against stuff. But you couldn't really make milling moves with this, just kind of drilling moves. And the trouble with the ball screw stuff is it was really limited in speed because, you know, you're putting all of the force of this tons through a screw, right? And so the screw could get ripped up if you don't do it slowly enough, right? So it meant that most of the time in which you're making these parts, you're not making chip.

You're just moving the masses to the point where it's going to make the chip. And there's an old saying in the business that if you're not making chips, you're not making money, right? And so the duty cycle of actually making chips was small. And so– but I had learned enough control theory while I was in grad school in Bell Labs that I realized that there was no need to do all this open -loop hydraulics.

And there were some new– again, getting to the things that are just new– there were some very impressive servo valves developed by Bosch in Germany. And so I was able to do some nonlinear control theory coupled to those servo valves, coupled to hydraulic cylinders, coupled to the final thing, which is energy storage. So normally when you're doing an electric move, right, is– if I want to, for example, move that whole big column of metal forward in a certain given period of time, if I wanted to do it in half the time, that means I'm going at twice the speed, okay? But twice the speed means I've got four times the kinetic energy, mv squared, right, that I have to put

in there. But I have to put that energy in in half the time. So the peak power goes up as the cube. So that's ruinous with an electric motor, right? Because now I have to have electric motors that are eight times bigger to do half the time, and their mass goes up eightfold to have that much extra horsepower. And so now it's like having a rocket in which your rocket fuel is really shitty and your mass fraction is really bad, okay? So that was not the– but hydraulics, you can fit 100 -horsepower hydraulic motor in the palm of your hand because basically it's just delivering what's the original source.

And with hydraulics, they have accumulators, basically short -term batteries for hydraulics. And it's nothing more than a big -ass cylinder which stays remote, right? And it's got a little nitrogen bladder in the top, and you use your 20 -horsepower electric motor to drive a pump to basically compress that gas up there. And now it's a spring, right? And then when you want to make your move, boom, you can blast that out at hundreds of horsepower through the hydraulic lines and move the load.

So I made this thing. It's somewhere here, somewhere on one of the walls. Do you do like acceleration? There it is right here. That's it. Okay. That's fast, flexible, adaptive, servo -hydraulic technology. So you see that thing there? See this big -ass mass here a human sits like right here? See this big thing here? That's a couple tons. I could move that at eight Gs of acceleration and position it anywhere within a meter cubed to five micron precision.

That's sick. Yeah, it was sick. It was sick. It's been taken over since by linear electric motor technology. This is why Starship started with hydraulics, but they quickly pivoted for the flaps and so forth and moving the Raptors around to electrics. But it was only long since I left that electrics could get to that point. But man, it worked great.

It worked great. I was so proud of that machine. And so I spent three years developing it and two years trying to sell it. And in the end, I sold two. Two. Because it was– well, a couple reasons. A, there's a lot of pushback from the UAW because we had to use 3 ,000 PSI hydraulics instead of the 300 PSI they were used to. Now, aerospace at that time was already using 10 ,000 PSI hydraulics.

And there's no problem with 3 ,000 PSI. But that doesn't mean that they're not going to fight it anyway. The UAW was definitely not on board with that. Is it because it would produce more parts per person? So there's like fewer jobs? That's possible. But I think it was just– it wasn't just the UAW. It was also– so this gets into the human aspect of how orders are placed and so forth, right? Is that typically what happens is, again, the big three say, hey, we're going to make this particular new car.

Here are the parts. Here's the part print. And what happens is the guy who is responsible for getting that part, a million parts per year to the company, he's up and coming. He's probably about 40 years old. He's a middle manager. This is going to help define his career, right? So should I go with what's always worked or should I put my neck out and do this crazy thing that this crazy kid who– well, he's not a kid anymore.

But he's wild -eyed and nutty and talking about– and I look at his machine and we would bring people by and they would look at him and say, okay, here, we're going to make this part for you. And they would literally jump out of their skin when it first moved because it was like a hummingbird. And it had to be tied not just to the concrete floor but into the rebar in the thing for the back reaction of the inertia, right? And so it just scared the shit out of people, okay? Particularly the people who were going to sign the check in order to buy it, right?

And so, yeah. I'm a good scientist but, man, am I a horrible businessman and an even worse salesman. And it's just not– I cannot make people feel comfortable, right? And that's still a problem today in trying to go to philanthropy and raise money is because I come across as too fucking crazy, right? And even philanthropists who claim that they're risk -tolerant are– fundamentally, they don't want to look stupid, right? And so this is a problem but, yeah.

And so what happened next? So did you go back to science or– Yeah, so that's where we then get into the palm story, right? So after six years of that, I apologized to my dad for wasting a couple million dollars of his money and so forth. But the company was still doing fine. And, you know, of course, he wanted me to stay in what he always wanted.

He wanted his son to take over his business someday. But I did– he was just– he was so good. He was the inverse of me. He was so people -friendly and he could make everybody feel comfortable and like him. And he just was really terrific. There's no way I was ever going to fill his shoes, nor did I want to, right? I wanted to do something– I still had in my head I'm a scientist.

I'm a technologist. I want to make the warp drive, okay? It's not my dream to run a machine tool company. I want to build a warp drive, okay? So what happened to the business? Oh, that's a sad story. So it went fine again for several years after I left. But then, you know, my dad got old. The other technologies– remember I said eventually, particularly CNC's started to take over more and more of the business.

They were still doing fine, but my dad was getting old, okay? And the business was changing. And of course, through the 90s, not only were the minivans doing well, but there was encroachment of the Japanese and so forth. It was putting pressure on the big three. And so the company was– they got up to 70 and then they stayed around 70 for most of the time until my dad retired, 70 million.

And then it was time for him to retire and so he ended up selling the company and then– this was 2005. And then 2008– what happened in 2008? GM goes bankrupt. The great financial crash. You got it, guys. Okay. It was a bloodbath in Michigan, a bloodbath. Unemployment hit over 30%, okay? My dad's company, the people who went into receivership, 300 guys.

300 guys, many of them who I knew out of a job. I was by then a Janelian and living high on the hog, okay? But a lot of those guys– and it was a scramble for those guys. A lot of them just odd jobs here and there, whatever.

We never got the reckoning we deserved for what happened in 2008. There were not enough heads who rolled for that whole f***ing thing. Basically, moral hazard went out the window and allowing people to actually f***ing take the penalty for what they f***ing did never happened. It's one of the things that– I kind of feel like the United States lost the thread of what it was starting in 2008 and has not recovered.

That there is no reckoning for bad behavior, financial or otherwise. And I think it really started right then. And again, eventually those guys found things. But it was brutal, really brutal. So anyway, let's switch topics from that. But yeah, yeah. Coming back to the poem. Yeah, I just want to say one more story about that, right? It's another thing that– in some ways they– well, certainly I believe the financial types have primary responsibility for this.

The big three in the unions did not help themselves. And a story for that was on one trip when I put in one of my vision systems.

This is a transmission plant in Kokomo, Indiana. Kokomo, Indiana is literally in the middle of nowhere. It's in cornfields. It's right next to Seymour, Indiana. Do you guys know John Mellencamp, the singer at all? But I did my piece at Purdue. So we had a lot of– There you go. Urbana Champ– yeah, right. So Kokomo is– trust me, Urbana Champagne is New York City compared to Kokomo.

So anyway, so I made the grievous error because I went with some of the service guys who were doing other things on that machine. I made the grievous error of taking one of our– we had a fleet of cars that the service guys would take out, right, when they went on jobs. I took a Ford to a Chrysler plant. Okay. So I'm working all day trying to get my machine up and running in this Chrysler plant.

I have never seen a closer approximation to hell in my life. It was August. It's 120 degrees inside of this plant. The plant is probably like 250 ,000 square feet, chocked full of machinery and people. Probably 120 decibels in there. Every frequency from subsonic grinding to supersonic milling. Oof. Stuff for the brain. There's coolant mist in the air everywhere, chips.

There were a thousand people working in that plant. And I worked there from 6 in the morning until 11 o 'clock at night. And I go out to my car bone tired, ready to go to the hotel. And there was– it was 90 degrees outside that day. And so there was– somebody had poured chili all over my windshield that had dried because I took a Ford to a Chrysler plant.

Can you imagine if I took a Toyota? But the thing I learned from there, and this is something which I'm sorry still sticks– I am never going to be an academic. I am an academic. I have– the first class of the day, I have to teach this. There's no way I would ever be at this place if it were not for my wife who wanted to be here. You're trying so hard not to be an academic.

Yes. It pulls you back. But the thing is, is those people work their ass off. And they work under just miserable conditions, right? Eventually, that's gone. That transmission plant is also gone. All of those people were thrown out of work in the end, right? As everything changes and more and more stuff gets done overseas, right? I get it.

I believe in competition, okay? But those people worked really, really hard. And then all of a sudden, skipping past the palm, I'm at Janelia working for Hughes, a $1 billion building that's just luxuriously appointed. The first thing that happened when– I was the first group leader on staff. And when the first six of us were together, they took us in a room, showed us six different executive desk chairs and said, pick the one that you like the best.

When I was in my dad's company, I worked on a Naugahyde stool that had the jute poking out of it into my ass for six years because it was good enough. It was good enough. Money mattered, right? And all of a sudden, I'm in La La Land. And then I start going to conferences again, right? And I go to these conferences. They can't be– put the conference in Kokomo, Indiana.

It has to be in f***ing Crete or someplace, right? So everybody flies to Crete. They're eating their lobster. And they talk about the dumb rubes and the flyover states who vote for Trump. Well, why the f*** do you think they're voting for Trump? This is Marie Antoinette all over again. They have no– most academics have no appreciation, no gratitude for the grant money that they get and not an understanding that that money comes from those people in Kokomo, Indiana and 100 million other people just like them all around the country who scrabble every day.

And they think that their tax money is being well used to support research. So anyway, sorry.

Is that the reason why you don't apply for the grants? I have never written a grant in my life. I will go to my grave never writing a grant. And that's only part of it. The other part of it, of course, is that peer review is toxic, right? It enforces conformity. They never want to take a risk, right? Because if you stick out, your peers are never going to support that.

They all want to support the stuff that everybody else already does, right? That's how it works. Again, show me the incentives and I show you the outcome. They are not incentivized to support crazy ideas, right? So you don't get crazy ideas. You get– but the lack of gratitude of academics towards the money they have or a willingness to realize that every penny they spend is off of the labor of people doing all sorts of shit.

I have a vacation house near Yosemite and I cross the Central Valley every couple of weeks. The people working so hard in the fields there, particularly in the summer, and the smell from the pesticides and the fertilizers and so forth. And Jesus, it just drives me crazy that the people don't realize the basis of what everything– of the comforts that we have, where the hell it comes from.

And– or an appreciation of all of the technology behind– I wrote a tweet just the other day about SiriusXM, right? And how much I love being able to not have to be just limited to FM and CDs and tapes like when I was younger but have this infinite variety of musical sources and so forth. And the technology in order to do that, right? In order to have 10 ,000 satellites going on in orbit and the switching from satellite to satellite and a phased ray antenna I have in my backyard for just a couple hundred bucks.

And all of this stuff and people just don't realize that there's magic all around us, right? And they're completely oblivious to that magic, completely ungrateful or unappreciative of that magic. They expect it. And it's so easy to lose it if we don't have gratitude for it or an understanding of it. And so it's just nuts. Anyway, I'm going off topic.

We haven't even gotten to Palmer. We're way off the thread of microscopy and have been for quite some time.

This is good. I actually– I worked in Michigan a bit. Cool. Where at? So I was– I had like built a factory in China and I was then directed by the CEO of this company that had acquired my company. And I built the factory for them in China. And then he was like a– he was from Michigan and he wanted to bring manufacturing back there. Oh, cool.

So we tried to move the factory to Detroit. And this was like 2015 -ish, 2016. Okay, yeah, yeah. And it was probably 2016. And that was like a– when I tried to do it in Detroit, I was like, yeah, I think some of these things had to go out of business. Yeah, right. So we ended up– we failed to do it in Detroit and we ended up doing it in West Michigan, which was much better.

And the explanation– Kalamazoo or– It was Grand Rapids. Grand Rapids. The explanation that I got, which I'm Curious, like what you think of this is that the flavor of Christianity in West Michigan, this is just what I was told, so I never verified this, but it doesn't allow you to join a union because you can only be in one organization, which is like your, I think it's like some type of Baptist church or whatever.

Yeah, I don't know. And so because of that, they never had like the same penetration of unions. Right. And so people were just much more reasonable. Like when I worked with them, they were flexible, you know, just the typical, like reasonable. Whereas like in Detroit, everybody was kind of, there was this like goo you would move through, you know, when you tried to do anything.

You would just be in a room with people. I would call it solid cement. Something like that. And I was like, yeah, I kind of get why this didn't work out. Like, you're kind of like, you know, the whole thing felt very hard to get through. Yeah, yeah. And I think the Ohio plants that started and the ones in Tennessee, it was a new story then.

But there was a lot of legacy stuff, obviously, in southeastern Michigan, right, that made it extremely difficult. Yeah, I think it's just hard to keep a culture. Like, obviously, Detroit was incredible for whatever, 50 years or however long. But it's hard to keep that culture, right? Because people get. When I was born, Detroit was the fifth largest city in the United States.

You know, it had three million people. Okay. Now it has 600 ,000. Yeah, yeah, yeah. Yeah. Wow, that's crazy. Yeah. The 632 nanometer podcast doesn't have any sponsorships. It's an art project and we're going to keep it that way. But we wanted to do an unpaid advertisement for Tesla Autopilot. Both of us use it. We've both done probably over a thousand miles on autopilot now.

And it's just an incredible technology. It's gotten amazing over the last two years. And I'm personally very passionate about it because I got to, you know, was hit by somebody who ran a red light. And my pregnant wife was in the car. And it's just something that really shouldn't happen anymore. Like, we have the technology to stop this. You know, if that person was driving a Tesla, it wouldn't have happened.

So luckily, it all, you know, it all turned out fine. Driving in Boston is a hell. And I think that definitely helps with finding the right way. It's mind -bending and life -saving technology. I highly recommend. You have your hands free. You're mind free. You can think. Yes, you have to, like, pay attention to the road to an extent. But it really drives itself.

But they really improve it all the time. I don't think I've had to take over probably in the last six months or something like that. And that's literally all of my driving is on autopilot. Anyway, go out and try this. I think every car should have it, not just Tesla. Like, this software should be in every car on the road. If you are a fan of the 632 nanometer podcast, you most likely love quantum computers.

And now I'm in the lab at Quera Computing, where we build the most advanced neutral atom quantum computers. We have plenty of job openings. And I'm hiring for the position of the quantum machine builder. If you have a relevant skill set and want to contribute to the race for building the first fault -oriented quantum computer, join us and enjoy the rest of the episode.

How did you end up building a microscope in a living room? In a living room. Yeah, all right. So now we'll pivot to that. All right. So picking up that thread. So I failed from my dad, right? And I left. And so I was unemployed. You know, I was unemployed for about a year after I left Bell. And I was unemployed for two years after I left my dad's company.

And the first, it's like, well, what am I going to do? I've blown up my scientific career. And I've blown up my backup plan of working for my dad. Is it good to be unemployed to get some fresh ideas? That's a major understatement. By far, the best periods of my life were my two periods of unemployment. So, for example, getting to the Palm, the first key idea that led to Palm.

And in fact, the Nobel Committee cited two papers for why I was going to share that prize. Both of them, one each, was from each of those two periods of unemployment. And they were done by me and me alone. Well, not, no, the second one was with Harold. But, you know, the key was that I was unemployed in both cases. So, the first one is after I left Bell and I was trying to think of, well, am I going to work for my dad? What am I going to do? My wife, my first wife, was at Bell and still working.

So, at least we weren't going to starve. And we had had a baby by that time. And so, the baby is like six months old. And so, I'm a house husband. And so, I was pushing the baby in a stroller. And there were two– remember, I said I just had this streak of hits with near field. I mentioned the single molecule one already. But another was one I did with my friend Harold where we looked at– again, because Harold was a low temperature physicist.

We were looking at semiconductor lasers at cryogenic temperatures to understand exactly where in these quantum well layers the light emission was occurring. And people have studied that optically for ages and understood the spectrum of the emission and like that. But when we did it with a near field probe, we found out that that spectrum, which normally looks like a normal spectrum, you know, it's got little humps on and like that.

But when you look at it with near field, it completely broke up into sharp spectral lines. Boom, boom, boom, boom, boom, boom. And every time you moved the probe by even 50 nanometers, the lines would change dramatically. And so, what we found was that the light wasn't formed anywhere. It was formed in discrete spots, which were usually just different points of roughness, single monolayer changes in the roughness like potholes in the quantum wells.

That would change the quantum confinement slightly and hence change the spectrum of the emission. And even though– even with our near field probe that was 50 nanometers, you know, those potholes were too close together to resolve. But because the emission from them was so narrow, you could still resolve them in a multidimensional space based on wavelength as the additional dimension.

So, you could individually study them because you had enough spectral resolution in the one dimension coupled with enough spatial resolution thanks to the near field and the other dimensions to resolve all of them. Interesting. So, I was pushing– and then the second one was the fact that I was the first guy to be able to see single fluorescent molecules and then localize them to subwaves and dimensions.

So, while pushing the dollar in the stroller, it just– I wasn't really even thinking about it. But it literally just popped from nowhere that you could combine those two ideas. And so, if I had some way– say the molecules– I know– I knew fluorescent molecules at room temperature at broad spectrum, but say they didn't. Say that I was at a cryogenic temperature and they were equally sharp, right? Well, then I could then at low temperature with a spectrograph see these single molecules.

And with my near field tip, I could see just a few of them at a time underneath the thing and they'd be isolated because of their different wavelengths. So, I isolate them, but now I can find the center of the emission even though it's 25 nanometers big with the near field. Hell, I could have done it with regular light, you know, diffraction -limited light.

But I can still find the center of emission to much better precision than the width of the emission, okay? And so, then I could find the position of every molecule to nanometer precision. So, this would be a path to super resolution, okay, is that you could do. So, I published that paper in Optics Letters and I described how you might do this at cryogenic temperatures and so forth.

But it would have been a hero experiment. I could have gone back to Harold who was still at Bell. I was still in the neighborhood because my wife was working there. I could have tried to do that experiment. But it would have, A, been a hero experiment to do it. And the second thing is you're f***ing at four degrees Kelvin. There's not a lot of biology, live cell biology to study at four Kelvin, right? And so, I just published the paper and left it at that.

And then that's when– then I went to work for my dad, okay? So, fast forward six years. We talked to my dad's story. I'm back unemployed, okay? So, it's like no one really take it like seriously? Actually, there were a couple papers. So, there was this guy Brackenhoff in the Netherlands used a confocal microscope around 2000 or so to cryogenically look at like six fluorophores inside a volume and see them there, right? So, that was really like the first– and there was a couple other things where people used like blinking of molecules to see two molecules in one spot.

But not at the– you know, if– in order to have super high resolution, you need to have lots and lots of molecules to decorate your sample or else it's just a bunch of dots, right? So, this is something called the Nyquist limit, right? Is that you have to sample at at least half of the finest resolution you want to have a complete picture of what's going on.

So, that means you need a– if I want to have 50 nanometer resolution, I need a molecule every 25 nanometers localized to a few nanometers to have resolution at that level. So, you need to have lots. And there was no good way to get lots in the same diffraction limited region at the time. How would you like sample them to like uniformly cover the space? Again, you would– you decorate them at high density but at the original concept, it's– I generalized the concept even in that paper to have any sort of discriminating third dimension.

If there's sufficient resolution in that discriminating third dimension, whatever that means is, it could have been the lifetime of the molecules. It could have been the polarization of the molecules. It could be– then you can discriminate in that space. But you have to have– the more– the higher the density of molecules, the more resolution you need in that third dimension to discriminate them.

Was there like a possibility to make a single layer or they were just like stuck on top of each other? No, no. Even in a single layer, there's– you know, the diffraction limit is big, you know, 100 times bigger than the molecule, right? And so you need to be able to discriminate, have resolution of 100 times better than whatever you had in the third dimension in order to do that discrimination to say that this molecule is different from this one within that same spot, right? So there wasn't a great way to do it.

Maybe cryogenically you could have done it. But then fast forward and during my second round of unemployment– well, first, it's why go back to science? I hated science. I hated everything about academia, right? I think I've made that clear by this point.

But the key for me was I realized that I really miss science. It's a love and hate. Definitely it's love, hate. It's definitely– I really miss doing science, okay? And I wanted to see if there was some way somehow I could get back to doing science. And so I reconnected with Harold and after I left in 95, this was the time period in which– you know, in 84 is when they broke up the monopoly that provided the financial underpinning for Bell Labs.

And so they continued to support it through the mid -90s. But by then they were starting to think they need to be financially relevant and contributive to AT &T's bottom line. And so then they– among other things, they purchased the National Cash Register, right? And then– so the environment changed so quickly at Bell that when I started in 88, in 90, I had another scientist, another friend come by and said, you know, what you are doing is technology.

And that doesn't really matter here at Bell. The only thing that matters is how many FISREV letter papers you publish, okay? So fast forward two years after that, after the NCR purchase and the head of Bell Labs research, Arno Penzias, who did the background theory for the Big Bang and won the Nobel, every year gives a state of the union address or something.

And he used my work as the sort of applied work that we should all aspire to because I had had my first few hits then, right? A couple of years after that with the National Cash Register purchase, right after I left, Harold was asked to spend some fraction of his time not doing low temperature STM but instead see if he could figure out how to use spectroscopy to figure out which fruit was in the shopping cart so you wouldn't have to put those little four -digit stickers on it and would automatically– that's how quickly the culture changed, okay? So anyway, so there was this diaspora out of Bell and by 2000, the Henrik Schoen scandal was the final nail in the coffin and

so everybody was gone after that, right? And so Harold had gone into working for a company that makes test equipment for the disc drive industry in San Diego but– so I reconnected with Harold because he's been my best friend forever.

And, you know, he was feeling dissatisfied to a degree, right, with that and we was thinking, well, he'd like to get back into science too. So we started going to different national parks where we would meet up and, you know, hike around and think about ideas. And I developed– I decided, of course, I couldn't get away from microscopy. I knew I didn't want to do microscopy but then I started to think about it and the thing that changed my mind was starting to read the scientific literature for the first time in the decade and I ran across the paper on green fluorescent protein.

Green fluorescent protein came out the exact same time I quit Bell Labs and I did not look at the scientific literature at all. The idea that you could snip a piece of DNA from a glowing jellyfish and have that attached to any protein of interest in a live cell, my jaw was down on the ground for a week after learning that in 2003, okay? I was like, oh my god.

Oh my god. This is going to revolutionize imaging, okay? I said, shit, I got to do imaging. So I tried to come up with an idea and I came up with an idea about how to interfere multiple light beams from different directions to create different types of optical lattices. This is used now a lot in the AMO field, right, to do stuff. But at that time it wasn't.

And so I came up with these theories of optical lattices to make a massively multifocal excitation field that I could try to do parallelized three -dimensional imaging of living cells. And so I tried to get Harold to come in with me on that idea. And he said, it's a nice idea and I'd like to help you but I'd be chewing your cud because it's your idea.

But he tried to help me. And so he– there was– so part of that diaspora out of Bell Labs was a lot of good people were everywhere. People I knew. I had contacts everywhere because I had made a name for myself at Bell. And these people had all gone to academia or government labs or whatever. And so, man, did I use that contact network.

So one of them was– Harold was friends with this guy, Greg Bobinger, who had been at Bell and had since become head of the National High Magnetic Field Lab in Tallahassee. And so Greg had been trying to recruit Harold to become a scientist at the magnet lab. And in an earlier visit, he met this weird dude, Mike Davidson. And Mike's job was to use microscopes to look at the grain boundaries in the wires that would go in the magnets because if they weren't the right type of structure, the magnets would just blow apart when you're running at 30 Tesla, right? And so– but Mike's real love was live cell imaging.

And so Mike had made himself independently wealthy by using his microscopes to look at cocktail mixes under polarized light microscope and print those on neckties and then sell them online. And so he made millions off of this, okay? And so then Mike was– used that money to follow his dream of doing live cell imaging and live cell imaging meant GFP in that era, right? Green fluorescent protein.

And so Mike became a cloner. And in fact, he– because Mike was kind of self -taught, barely made it through college, self -supported all the time. He took all these kids who were flunking out of Florida State and hired them as techs to be cloners to start knocking in fluorescent proteins on every f***ing protein under the sun. So he had the world's biggest library of fluorescent protein fusions, about 3 ,000 different knock -ins by that time.

And so we went to visit– and so Harold thought, well, Mike's into live imaging. Maybe he'll give you some space in the lab to do this. So we went there. We hit it off great. And he then told us– I said how much I love fluorescent proteins. And he said, yeah, well, there's this kind of weird one that just came on the scene. What happens is when you shine the normal 488 laser on it, it doesn't glow green.

Nothing happens. But if you first shine 405 light on it, it activates it, and then it glows green. And so, okay, well, that's interesting. So we finished the trip, and Harold and I are in the airport in Tallahassee. And it hits us both at the same instant. Oh, my f***ing God. If you turn down that violet light really low, only a few photons at a time are going to hit the sample.

You're going to only photo convert a few of the molecules. Statistically, they'll be separated. Their fuzzy balls will be separated by more than the fraction limit, even in a regular optical microscope. And then you can find the center of those fuzzy balls. Then you turn those molecules off, turn on another subset, do that again and again and again, and you get a super resolution image with a standard microscope.

No funny, fancy tricks at all other than having this photo switchable fluorescent protein. And so we said, well, f*** that optical lattice shit. Let's do this. And now Harold's fully invested because it's our shared idea, right? And so we were terrified. We were like, my God, that paper is like four, five, six months old. Why hasn't anybody done this already? I pitched that idea back in 94 how this would be possible if you had something like– nobody had connected the dots.

It's like we were– we're nothing. We have no lab, nothing, right? And anybody who had a reasonable lab, this would be trivial to do. It's like, holy f***. We've got to do it now. And so we figured we'll do it at Harold's place because when Harold left Bell, he took all of his equipment with him because Bell didn't have any use for it. So a lot of that was a lot of the spectroscopy and optic shit we did when we did our quantum well experiment back in the day.

So we had some lasers and detectors and other shit. And then we had to put about 50K each of our own money into it for other shit that didn't happen and machining the microscope. And that's that guy right there. That's Harold's living room right there. That's where we did it. And you said the EMCCD was the– The EMCCD was the big ouch because that was 30 grand.

And I struck a deal with Andor to at least return it to them at half price if it didn't work out. But that was the whole thing. We put that together and– we went from the idea to shipping it. So that was the other part of it is we didn't know shit about biology. We didn't know how to clone a cell to get fluorescent proteins in or anything.

We needed a f***ing biologist to work with us. Mike was willing to help and that was great. But I had another in which is that– so who invented this photoactivated fluorescent protein? Two biologists who were at the National Institutes of Health. Well, guess what? I had a friend from Bell Labs who had gone to NIH. Bell Labs Network. Bell Labs Network.

So I called up him, Rob Tico. He was an STM guy who had pivoted to other things. And I said, you know, I'm trying to find a job. Can I come give a– would you host me to give a talk?

And in fact, I had set that up before– I misspoke because actually I spoke to him before we had the idea for this because it was one of the many ways I was trying to sell my optical lattice ID and get into a lab. Because I asked him could I give a talk to try to pitch my lattice ID. Okay. And so when the day came to give that talk, I said, Rob, would you please ask these two people, George Patterson and Jennifer Lippincott Schwartz, to come to my talk? Because I'd really, really, really like to meet them.

And so they came to the talk and I went to them afterwards. And I said, can I take you guys to launch? And I said, okay. And I took them to launch and said, my buddy and I have this crazy idea and we need you guys. We need you guys badly for your photo -activated flare suppress. And Jennifer said, fantastic. Sure, bring it by. And so we were already building the scope then.

And so from conception to shipping that thing out to NIH was three months. Within– Oh, so you ship– We ship that whole thing to NIH. Right. We built it there. Why not just to get the– Because there's a lot of biology and shit you got to– It's easier to take the microscope to the biology than take the biology to the microscope. Okay.

So we did that and within less than a month, we had a sample in– Jennifer's people would– rather than taking a whole cell, we did cryo -sectioning to take thin sections, plate them out because you're worried about the third dimension, right? There would be a lot of out -of -focus stuff like that. So we wanted to make sure we were really two -dimensional.

So we had these slices that were then on the sample. Could you use confocal approach? Not really. Again, just– You want to see like the whole field of view? Yeah, yeah. Exactly. Exactly. And so we did that and so we had some slices through some lysosomes. That photo is somewhere in here, I think. Yes, right there. There it is. Okay. That was one of the first.

But in the first– so we turned on the violet light and then turned on– poof, these molecules came on. We said, f***, we got it. It works. It's going to work. And then we just raced like hell. We had only a 10 -milliwatt laser. So we spent– so we worked around the clock just babying the microscope because the focus would draft because it's just this cheap little thing that we put together, right? And so we would– it was like November and it was cold and it was a concrete– it was an old dark room that was converted.

And we kept it– her lab was big but we kind of kept what we were doing secret except for George. And so they would see us going back and forth and they didn't know who these two old dudes were because everybody is just a postdoc or a grad student, right? We were these old guys talking to George, right? They called us the gruesome twosome.

And so we were doing this and once we had that, everything went like– so yeah, we went– within three months, we had all the data that was in that science paper. So you realized that before you did the post -processing of the– Oh, yeah, yeah. Oh, yeah, yeah. The post -processing is easy, yeah. So yeah, it was obvious. We had it. What was the most difficult charging part in building the setup? Nothing.

It was like a gift from God, right? It was just trivial. Everything was easy. It was just the ripest plow you could imagine that was just waiting to be plucked. Why did it take so long to actually collect the data? What was the– Oh, well, because we were unemployed and had to shell out of our own pocket. For a 10 -milliwatt 561 laser to excite the fluorescent protein.

Okay. So we didn't have much money. So we dribbled those molecules out. Nowadays, you know, you got even– the SCMOS cameras are much faster than the CCDs back then. The lasers are far more powerful. You can do it just– you know, ICON takes a fucking petabyte of day to day. Okay. So it's a different world than it was back in 2005. How did you control the lasers? Because you need to, like, switch on and off? Yeah, yeah.

There's just TTL pulses that would go to control the laser from a PC. And so we had– you know, there's a PC somewhere in that– you know, you can at least see the screen of the PC, whatever. So they had, like, the kind of triggering? Yeah, yeah, yeah. Yeah, exactly. You could trigger the lasers, the 405 and the 561 and so forth. Yeah. It was really simple.

How did you build the software part to control? Just ourselves, you know. I mean, you know, you just have national instruments, cards, and, you know, shit, and you control it through either LabVIEW or Matlab. Was it LabVIEW already? LabVIEW was around. I just did it with Matlab, but, yeah, so, yeah. And also the localization code was done in Matlab, and, yeah, so, yeah.

And for the objective, did you need a special objective? Well, yeah, you need a good– we used a high NA turf objective, and that was another one of the ouches, okay, in terms of– that was probably at least 10 grand, something like that for that objective, yeah. So that– those images proved that the method could work. Yes. What was kind of the first use of the method that generated, like, biologically interesting results? That depends on who you talk to, right? I would argue that most single -molecule super -resolution hasn't revealed much of anything.

In fact, I would argue that most super -resolution in general has not revealed much of anything, okay? We can go deep down that rabbit hole if you want, but my feeling is that even back in 2014, I felt like it was incredibly premature to give a Nobel Prize for that. I think the only reason they did is because– well, they broke this fundamental limit.

But I think anybody who knew anything from an optics point of view would know that it's not a fundamental limit in the same way that the uncertainty principle is fundamental, right? It was a practical limit that could be circumvented by a trick, right? And the trick is you can localize to better precision than the wavelength, right? And that's the trick, right? So anyway, where were we going with this? What was the biological– Oh, oh, the first– Oh, oh, the first– yeah.

So again, so overall, the first– in my opinion, the first real hit and the one that led to ICON was when we started not to look at dead and fixed samples. The reason you look at dead and fixed samples is you're only looking at a few molecules at a time, and it takes a f***ing long time to bleed out every molecule to get that Nyquist criterion of having every molecule on the thing decorated to put together that whole image.

And you're throwing a lot of light at the sample, and if it were alive, it would be cooked with all of that light on it for that time.

So yeah, I know we're never going to make it through everything in the period I'm talking. But– Well, it's a great story. It's like where else can you find it? Exactly. This is like what we're– Not written in any paper. Yeah. Well, most of the stuff is– you can find somewhere if you dig hard enough. But the first real hit was– and this really kind of set the stage for the whole rest of my career up to the present, which is that at– when I was at Janelia, the president of– Janelia is funded by the Howard Hughes Medical Institute.

And at that time, the president of the Howard Hughes Medical Institute was Bob Tegen, who is famous as a biochemist who was one of the key guys to unravel the methods of transcription, how different proteins come together to recruit the polymerase, which is necessary to unzip DNA and then produce RNA. Okay.

And Teej had– was a biochemist, but he understood the potential of trying to see at a single molecule level how this transcription is actually occurring. And so we started to use palm -like techniques in order to do that.

And so in their models, in their biochemical models, they had come up with this idea that many different transcription factors, they're proteins that come together at the start of the gene first before it recruits the polymerase. And so there's a sequence of events that has to occur with different proteins. And they believed it had formed this large– call it micromolecular complex, but multiple proteins, dissimilar proteins that had to start at the beginning of the gene before the polymerase would come in.

And it was believed that this would take minutes to hours to happen. When we started to look at the end of– by palm -like stuff in live cells at these transcription factor molecules, none of them were binding to the DNA for more than a second or two. And so it was like, holy f**k. Our whole model of how this– how transcription works is completely wrong.

And in fact, this is the take -home story of this whole talk, of this whole morning, all right, is that almost everything you learn in biology textbooks is a hallucination. Because they– it is. It is. It's because– or at least cell biology– because what they're doing is they're taking three reductionist tools, biochemistry, molecular biology, and structural biology.

And then hypothesizing what– how those little pieces, tiny, tiny little bits, come together both structurally, stoichiometrically, and dynamically to create the cell. They have no direct knowledge of the stoichiometry or the arrangements, spatial arrangements, or the dynamics. All of that is hallucination, okay? They have little bits. I'm exaggerating.

But largely speaking, you guys have probably seen on the web, you know, there's those beautiful things of, you know, oh, all these molecules coming together. Here's a cargo on a kinesin walking along a microtubule like this. You know, if you've seen these things. And it's all like in this vast empty space. I don't know any cell that's a bunch of acidity space.

I'm sorry. It's crowded as fuck. There's 100 trillion water molecules in every cell. There's 10 billion protein molecules. There's 10 billion carbohydrates. There's 10 billion lipids. There's metabolites. It's by far the most complex matter in the known universe. We understand the interiors of neutron stars far better than we understand the interior of cells.

It's crazy complex. And yet we've got a whole industry of pharma. That's back to Roger's point. There's a reason why only 9% of the drugs that enter phase one come out phase three because we don't know what the fuck we're doing. We don't know the real mechanisms that are going on. And this is the real hit from single molecule stuff was the realization that that's the case.

That when you start to fucking F***ing look at the dynamics, not just the structure. You realize that you had it all wrong. And you realize that so many of the things that they thought they knew, you can't be sure that they know. We have to reinvestigate all of it, right? And so that's where I pivoted to live imaging, right? What was the improvement in the method that actually allowed you to look at transcription live? Well, there was no improvement.

It was basically applying palm -type techniques to photoactivate a subset of these transcription factors and look at them in a live cell. Instead of having to get to that Nyquist criterion of trying to get every molecule, right, in order to look at a structure, I didn't care about that. I just want to understand their kinetics, okay? And so if I want to understand their kinetics, I can do a subsample, right? And if that principle was right, that they would form these stable complexes, I should be able to see that.

But they weren't. They weren't stabilized anywhere. They were staying for a second or two. And so the model that we've built up is that basically, mitochondria is not the powerhouse of the cell. Brownian motion is the powerhouse of the cell. That's how everything happens is statistically.

Molecules diffuse at– on the order of 20 micron squares a second, that means you'll go across a whole cell in like three seconds, okay? But in that time, you're not going in a straight line. You're bouncing off of literally one trillion other molecules in those two seconds to get to that other side. Your instantaneous velocity based on Maxwell -Boltzmann is like 100 meters a second, okay? But you're going– that's what makes it all work, okay? Is that it's all stochastic but you're throwing the dice so many times that eventually you collide with something where it's energetically good for you to stick to that other guy.

That's how order comes out of disorder, okay? And that's how you build up successively larger structures. And so it's that bouncing around, getting to that start codon on the gene for that first transcription factor. Then the second one comes along and he sticks to him. And the third one comes along and the first one is gone by then but the second– and so you get a cascade.

And then eventually polymerase comes in and then it scoots along with the DNA, right? So that's the picture that emerges. So then how do the pharma companies, like you said, they succeed in like 10 percent? Yeah. What are those cases when it works? It's like blindfold.

No, there's more to it than that. There's all sorts of– I'm not saying that there's no utility to the methods they use but it's incredibly inefficient. It is. And all they really think about is the mechanism of action at the molecular level and thinking of it in terms of the binding, right? Because that's really what's key. Well, again, single molecule tracking is great for studying binding because you can get on times, off times, diffusion rates, all of that stuff.

That's what ICON is all about, right, is having an assay to really look at that, right? But it's still a crapshoot because among other things, there's multiple– living matter involves emergence from many different levels starting from the stochastic motion of single molecules to macular and molecular assemblies to membrane -bound organelles to cells to tissues.

To populations to the whole f***ing biosphere. Everything about life is emergence from single molecules to that level, okay? And if you just focus on that molecular level alone, you're going to be really limited in your ability to discover drugs, right? Is that you have to be thinking about multiscale mechanisms of action. Okay, this protein has to be here in order for me to conquer this disease.

What if the protein can't get there, right? I mean, what good is it that you have– you know that you have a drug that will interact with this protein when it's there if the protein never gets there, right? What happens in terms of if that drug affects some other of the 20 ,000 other types of proteins that creates an off -target effect over here, right? So maybe it's doing what you want, but it's doing a thousand other things you don't want, right? They don't look for that directly.

That doesn't happen until you get to clinical trials and then either you have ineffective stuff or you have dead people on your hands, right? So, gee, wouldn't it be nice if we actually looked at all link scales if possible while we're doing initial drug discovery in order to find these off -target effects? But how do you do that? By using all the other microscopes that we've developed since, right? So we have microscopes that look at all link scales from the molecular up to the whole organism, right? And they all have a different niche that they serve and they all have a purpose.

But the trouble is now we can look anywhere from milliseconds to days, anywhere from nanometers to millimeters to centimeters. What does that mean? You do that in three spatial dimensions, one dimension of time. You're covering seven orders of magnitude of time and about 14 orders of magnitude of volume. And we have the ability to do all that with our microscopes.

What does that mean? Voxels, five -dimensional voxels, a f***ing lot of five -dimensional voxels, petabytes and petabytes of five -dimensional voxels. Now you're making the full cycle to the storage. No, but actually what we're coming to is I know we're running out of time. I'm trying to get to the present day. Okay. So the point is that we have enormously powerful tools now.

But you can see all of these pictures on the wall are from many of these tools right behind your head right there. That's what a neutrophil actually looks like when it's moving inside an organism. This is a zebrafish. What you're seeing at the top is the skin cells. What you're seeing in that cavity is a mesenchymal space. What you're seeing at the bottom, that funky thing is the neutrophil.

They're in crazy dynamic. That's part of the immune system. Yes, right. So if you watch the movie of that thing, it will blow your mind in terms of how it works and how it moves. Do you have one to show? I have a f*** load of movies I can send you, right? I mean more movies than you could ever want. Okay. So– but the thing is that we evolved to see in 2D plus time.

Okay. When you're looking at that, you're looking at 2D. It looks 3D, but again, you're not seeing the interior of that thing. The data is there. It's blocked by all the other cells in front of it in that particular view, right?

Life happens in five dimensions. X, Y, Z, T, and molecular species. The 20 ,000 proteins, all the lipids, the carbohydrates, all the rest, right? We can't– even with the microscopes that take our petabytes of data at all of those link scales, it's f***ing bits on a drive and it does nothing. It's so f***ing frustrating to have petabytes of data on drives that are completely worthless because there is no scalable way to look and understand that data.

What we need is to build a five -dimensional mind, right, that can look and see in five dimensions. A five -dimensional vision transformer. That's what we need to be able to crack this nut. I was the last human on Earth who ever wanted to have anything to do with AI because I hate doing what everybody else is doing. But I'm forced in this direction because I think it is the only possible scalable way to really extract meaning at scale from the data that we can take.

And you're in the right location, close to all these companies. God, you don't know how– over the last 18 months of pitching this, you don't know how many times I've pitched, how many close calls I think. You know how– it is the right place and it's the wrong place. The reason it's the wrong place is at f***ing Cal, how much do you think I can hire an AI engineer for compared to what he can get five miles away from here, right? That's problem number one.

You have to find the crazies, the crazies like me and Harold who don't give a f*** about the money. They give a f*** about the problem, right? They're hard to find, okay? When you talk to AI people about this, do they think it's like a tractable problem in terms of the type of data it is, the size? It's a great question. Even I– and I'm not an AI guy– I feel like this is at the very f***ing bleeding edge of what's tractable.

It is a big scale– it would make alpha fold look like a picnic, okay? So it is– building a vision language model on this level is really bleeding edge, okay? So we would need one of the big hyperscalers to bite in the end. But we could– but to triage that risk, there's a lot of initial ablation studies and so forth we could do to try to better answer that question.

But we can't get our paws even on enough GPUs to do that, right? What's the– maybe just to like go into the detail a little bit. So what is the data set versus the– like what are you training towards? So are you trying to predict the next change in the pixels? No prediction. Well, no, no. I mean you'll do things like next token– well, not token but prediction for– It's like which pixel is going to move where or something like that? Not that.

But the first task that you have to do is robust 4D segmentation, all right? The fundamental unit of life is the cell, right? You would like to be able to see the cells individually. Beyond that, you would like to see the organelles inside of the cell. So we need robust 4D segmentation. In this modern age, with everything that we have, even 2D segmentation is imperfect, okay? Biological 2D segmentation– you know, Meta had SAM, you know, the segment anything, right? And they have SAM2 and other things.

So what do you mean by segmentation? Just very specifically what does it mean? Find the boundaries of objects. Okay. Tell me where– this is this organelle. And where does this cell begin and the next cell end, right? This kind of thing, right? And the machine have some understanding of that. Because if it doesn't have that understanding of where cells begin and end, it's not going to get very far in being this Sherpa that I want to lead us.

I want to be able to ask through an LLM interface. What happens when a T cell is through immuno -oncology engaging with the tumor? What particular proteins are expressed at the surface? What is the course of its motility in order to get to the tumor? And if I have all of that data on drives, okay? But I can't access it because I can't find out exactly where it is.

I need something that can recognize that, right? And we have labels to label the T cell so we can train the model to understand what a T cell looks like and all of that. But we need to build the model to be able to do that. And segmentation is that first step is to know what is a cell to understand what a different cell type from that cell type, from a mitochondrion, from an endoplasmic reticulum or whatever, right? I mean, all of that is doable, okay? But again, it requires compute on a vast scale to get to that point.

But the first step is segmentation. And I believe if we had enough GPUs and enough people working on the problem, we could get robust segmentation because nobody's really tried. So much of the imaging that has been used has been two -dimensional imaging, right? So that's why Meta has like segment anything. It's a two -dimensional tool. And people try to apply that by doing plane by plane the 2D tool.

That's not the way to do it because you're missing a prior. The prior is there's reasonable continuity between successive planes. And then likewise, there's reasonable continuity in time. The cell doesn't go like to this right away. It moves continuously, right? So you need to build an inherently native 4D model that takes use of those priors.

So when you collect the images, are you collecting them in slices? Yeah, but we do it so fast. Oh, I see. So it's like–

Yeah. I'm interested in this among many, many others. And it's just everybody's got their own thing. What do you think is the first– if you grew this tree, what is the first fruit it bears? Like what's the first interesting thing that comes out of this? Let's say if you put, I don't know, billion dollars of compute into it and labeling and people and so on.

We estimate the whole– we call this thing cell observatory, right? And we're doing all we can to develop the biological reagents, knocking in all these fluorescent proteins and dozens, hundreds of different tags, cell types and like that. We've got the microscopes. I'll show you guys later and doing– to take the data and all of that.

But where was I going with this one? Well, my question is let's say if you took a billion dollars or something and put it into– We estimate it would take us $50 million. But what's like the first kind of– The first thing we're– I mean it's always hard to know. But like what would be like the first kind of result that you might imagine could come out of it? So the first thing would be just robust segmentation, right, which would be valuable in many contexts.

But the other is, again, if you had the ability to identify cell types, right, if the model could do that. And that would be trivial once you have a robust segmentation. Now you can ask all sorts of f***ing interesting biological questions, right? That gets back us into the pharma and the off -target effects, right? Like I mean I can look through– we use zebrafish as a model organism because it's transparent and it's 70 percent genetically homologous with humans and it's a vertebrate.

But to the extent that those things are recapitulated, if we use that as an organism, we could create an entire company that's nothing but a contract research organization for pharma everywhere where if they got something they're about to put in phase one, they bring it to us. We put it into the fish and we see if bad s*** happens, okay? Because we can holistically look practically down to the molecular level.

We have a baseline of what normal activity is like across all organs, all cell types and whatever and we see what the f*** this drug does, right? How many– if you have a zebrafish, how many– how much of it can you look at at a given moment? Like what's like the– Yeah. Again, obviously the bigger the field you could do, the less speed you have, right? Because you have to cover it.

But again, you would look over small fields of view of about 50 microns at maybe 100 millisecond intervals, which is pretty f***ing fast, right? Yeah, yeah. You can look over larger organs or half the organism maybe every 10 minutes if you cover a young adult, right? So there's obviously a huge dynamic range of application depending on what– but a lot of this stuff– yeah.

I mean not only the basic biology but the farm implications I think are immense and it just– it drives me crazy that I haven't– as hard as I've tried, I have not been able to interest any philanthropist. Not even– I'm still HHMI. I still have a lab at Genelia. I haven't been able to interest them. They have their own things they're doing in AI, which I think are stupid.

But apparently they think what– but they– apparently they think what I want to do is stupid too. So– This can happen sometimes.

All right. That's like a very common experience. I'm too old to have any sensor circuit left at all, okay, guys? But yeah. That's interesting. It's– but I wonder– so I mean for these problems where you have like– that people have like cracked with AI, a lot of times there's a lot of like good labeled data. So for example, in the car, they have the person intervening and driving the car.

And so Tesla has collected a ton of that data. Of course. So I'm just thinking like what's the labeled– what is the labeled data for in this case? The labels from us comes out of the fluorescence, right? No, but what I mean by labeled is like what is– let's say you want to predict like– It has to be annotated. Like this cell is attacking that cell, right? So that has to be human annotated initially or like how would you see that? Not necessarily.

So again, once– for example, if I had a different fluorescent label for my target cell and my T cell, right, the model itself just needs to know that– Like it merged or something like that. Exactly. Yeah, yeah. That's interesting. And this is why you have to do something through a transformer because it has to be self -supervised with minimal annotation thereafter, right? That's the only way it's going to work because you just can't use human annotation at scale.

Are there any existing efforts? I feel like you hear sometimes about efforts to like model the entire cell. Are there any existing– That's another pet peeve of mine, the virtual cell. We're back to these fucking reductionists, right? So they do spatial transcriptomics or alpha fold or things like that and think that they're going to predict response to perturbation on the basis of just that.

That's crazy talk. It's one part in 10 to the– what? One part in 10 to the 10th of what's going on in the cell and they're going to recapitulate all of cellular behavior on the basis of that. So it's naive to the point of craziness in my opinion. What's the advantage of transcriptomics over microscopy? Is it just cheaper, easier to collect the data? Transcriptomics is– spatial transcriptomics is a form of microscopy.

Ah, because it's spatial also. You're doing the sequencing in -situ. So what do they get out of it? They get, oh, well, okay. Well, these transcripts are here in this cell. So they determine where cells of different cell types are. It's valuable to an extent whether it actually– it certainly won't tell you because they're basically looking at what cell types are where.

They're not telling you anything. This is all fixed cells. Oh, yeah. These are fixed slices. This is all totally fixed tissue. Another huge pet peeve I have about modern biology is there's so much f***ing stamp collecting going on. And spatial transcriptomics is part of it. Connectomics is part of it. Everybody's creating huge atlases. And so we're getting all sorts of data and no understanding, no fundamental understanding.

Just collect, collect, collect, collect, collect. That's where we are right now. But at least I understand the limitations of it and I want to get understanding out of it. So what do you think needs to happen so that those AI people would feel former to jump? I ask you that question. I've tried everything I can for 18– you don't believe how many doors I've not.

And I've never had problems in the past getting money and philanthropy to support me. But on this one, it's been a bridge too far. I'm still trying. As long as I'm alive and kicking, I'll still swing at the pitch. Is the data publicly available? In pieces, right? But again, all they can make sense of is the movies. I can show you a crap load of movies, right? But those movies, again, are two -dimensional movies, right? Because they're projections of– No, but let's say the thing you were saying that there are these petabytes of data that are generated by the microscopes.

Is there a single place– like let's say I'm a researcher in France, okay? And I happen to have access to $50 million in computers. All of our data is open source. Is the data warehouse somewhere? Oh, yeah, absolutely. How would I access it? Like is there a practical way to access it? Yeah, you can download the data and we have metadata that goes with it that says this particular file has got this cell taken under these conditions and da -da -da.

You have to have all of that if you're going to actually build any understanding on it. All of that's available. But again, most people don't even have the– first, they don't have the tools to download the data. How do you share this? Then they don't have the tool. Yeah, exactly. Then they don't have the tools. We have two 100 -gig pipes going out of this building right now, right? But it took us forever to get you to give us those 200 -gig pipes going out of this place.

So there's a real limitation. There's like a physical limitation even to handling the quantities. Oh, yeah. Most people are nowhere near capable of handling. How many places are generating data in this amount? Like are there other microscopy centers? Maybe astrophysicists? Well, yeah. I mean, yeah, exactly. No, I mean in the cell world. Nobody.

Really? Nobody like us. We collect– I wouldn't be– well, no, that's not true. The people doing the connectomics now are taking data at big scale, okay? Yeah, yeah. That would be the only other application I can think of where they're really going balls to walls with the– Do you think the Transformers even has the teeth to grind through this data? It's our best option now until something better comes along, okay? If there's– this is maybe off target but since– I love space, right? And I think maybe I might get there.

And if I get there, well, because of data centers in space, which I'm still not sold about is making much sense. But if that happens, one of the things I'm hoping for– remember Munger, show me the incentives and I'll show you the outcome. We're on this crazy build out of AI right now and data centers, right? Crazy.

But I can think of two things that would change that. One is people come up with new model architectures that are orders of magnitude more efficient than the model architectures. Obviously, there's a huge incentive to do that. Some nuts somewhere will be thinking about do I need this many nodes and this many layers in order to get the same result, right? If they could reduce that by an order of magnitude, what would that mean in terms of all of this CapEx? There might be some spare compute.

There might be more than enough spare compute. There might be a bloodletting, right? And then the other one is what if there's a new architecture? What if there's a hardware architecture that would be much more efficient? What if somebody could– that we don't need to wait 10 years from an A100 to a B300 but somebody comes up with something that's poof, a different way of doing things.

That's an order of magnitude faster all of a sudden. Do we have in the world somewhere like resources for this kind of scientific compute? So let's say like have a data center which would be dedicated for like– I don't know, astrophysics. Or in your case. Right now, not really, right? I mean there's a big one up on the hill, half a mile from here, an LBL.

There's the ProMotor supercomputer. It has like 30 ,000 A100s in it, right? And now they're building the– what's the next one? The Doudna one. That's going to have like 30 ,000 B200s in it, right? But do you know how many people want a piece of that? Yeah, yeah, yeah. You never get 30 ,000. You don't get 100. You get tiny, tiny little bits for tiny bits of time because it's all diffused to a billion different projects, right? So I would kill to have 128 B200s, OK? We just– we have limited money.

We haven't been good with philanthropy recently, but we just shelled out for 32 B200s. A, it took us nine months to get them. They just came in now. B, we had to fight like hell to get the power and the cooling to put them in. 32 f***ing B200s, right? But it was a huge hit for us, but we need something we can use to do more ablation studies to get back to your question of what is actually physically possible at the state of the art, right? What's an ablation study? Oh, just try different tests about, again, how many nodes you need, how much of this you need, whatever.

What is the size in order to– and again, be able to do prediction, right? Make sure that you're making accurate predictions as you take away stuff or you add stuff or you do different types of tools, right? So– but it's– just the demand for the GPUs now is such that it's– Yeah, it's very interesting. Just as the technology is becoming broadly applicable, it's hard to get your hands on it.

Yeah, it's very interesting. It makes sense for the same reason, right? It's supply and demand, right? Yeah, yeah. Are there opportunities in other countries or– People need cat videos. It's sad, but it's true. Well, it's fine. It's fine. It's fine. I mean, come on. I mean, it's already transformed everything. You know, all the optics stuff I do in microscopes now, I mean, I used to do in ZMAX.

Now, you know, now I just grok it, you know? I mean, I ask, you know, I need to have, you know, a telecentric lens that takes me from here to here with this focal length and please build it and tell me what the spacing of lenses are. Tell me the sources and components and out it comes in five minutes, right? So, yeah. How do you now digest the data to build the same videos? What's that? How do you– How do you digest the stuff that's about some data? Oh, and poorly.

So, there's various software packages that will do renderings, right, of the data, four -dimensional data as three -dimensional images, glacially slow. Just loading it up, loading the data into even a good quality workstation and then rendering those images in the view. It's just– it can– you know, a 10 -second movie can take days. There's at least a hundred to one difference between the time it takes to take the data and the time it would take to visualize the data.

Maybe sometimes ten thousand to one difference. So, to start with, like at least like accelerating the rendering with already how– Oh, yeah. Absolutely. Absolutely. The scientists can look into the movies. So, this is another area that, of course, we investigate is can you do some kind of dimensional reduction to save it, right? So, basically– so, we're working with Gaussian splatting, for example, as a way of trying to have an alternative reduced space or reduced data representation that still recapitulates most of the structure and the dynamics that's in the data, right? If– that's a heavy area of interest right now because of our poverty, right, in terms of

compute in order to try to make something that– not just for the AI but also for the visualization is we may make a different type of renderer based on a Gaussian splatting model that would be much more efficient than what exists today. But I would love it if somebody out in the world did it instead. Last thing I want to do is reinvent the wheel if there's people better at this type of shit than us neophytes, right? That's super interesting.

Yeah, it's– the idea of modeling the entire cell is definitely very– it's just– it's very– I don't even want to talk about it. I haven't gotten through to you guys how f***ing complex it is, okay? We're not ready to model. We have to observe. We have to be Tycho, Brahe, and guys like that, right? We have to be looking at the thing and seeing these epicycles.

We don't f***ing know the inverse square law yet, but we have to see the epicycles first before we ever get to the inverse square law, right? So it's– science starts with observation. We haven't done enough observation of these types of systems to sensibly even think about virtual cells or prediction or shit like that. What I'm asking for is to help us do the observation.

Putting it as bits on drive is not observation. Having understanding in quotes either through a machine or through humans or best humans and machines working together is the only path forward. What do you think besides– so obviously you have a strong opinion about applying AI to this. Is there something– what's the future of the actual microscopes, the hardware? Well, they get better all the time.

Like I say, I have my ears to the ground on anything new that comes on. So I'll show you some microscopes. We have one that's coming online now. They'll be more performative than anything we've had before because we need to– we think our best estimates is we'll need on the order of 20 to 50 petabytes by the time we're done. And we'd like that to happen in two years instead of 15 years, okay? So that– so we want to get an order of magnitude more out of our scopes, right? And so we're building scopes.

Take multi -well plates and we'll be automated in order to do this, do perturbation experiments and all the rest. So we're developing that now. Is there only automation or is there like fundamental biology or optics that's still needed? I mean the optics, you have to be clever. But most of what I've been doing for I'd say the last eight years is more about just mixing together.

I take this off the shelf of what I know in optics. I take this off the shelf and I mix them together in different ways and put it together. So it's more about how you put it together and get the pieces to work.

If someone is like a, let's say, synthetic biologist or cell biologist, is there something that– like something missing that if somebody built in biology would really help with these experiments? Oh, many things. Many, many things. What are like some things people could work on? Well, you know, we talked about the bowling ball before, right? Yeah, yeah, yeah.

If there was some smaller type of– the other big thing that helped particularly single molecule is a colleague of mine at Janelia, Luke Lavis, made a brighter fluorophore called Janelia fluoridyes, right? They allow us to track much longer. They allow us to look at any organelles. These are small molecules? Small molecules. So they need to be then attached to the thing and that can be done with nowadays reasonable specificity by using bowling balls called like halo tag that are attached to that.

Then there's a ligand on the dye that then attaches to– to get the dions. So it's– fluorescent proteins are more flexible but in terms of the photophysics, the JF dyes are much better. Now, if, again, somebody can make a better fluorescent protein or smaller bowling balls or, again, if somebody could particularly make– this would be a dream is if you could make fluorophores.

There's– ions have very, very narrow emission widths, right? If you could somehow have a caged ion that maintained its narrow emission width in a buckyball or– I don't know, right? And get that to attach and you can get that to attach with a thousand different ways of attaching. Then we could look at a thousand proteins at once because all of their spectra would be– The spectra would be narrow.

So someday that– after I'm dead and buried, maybe that kind of shit will happen but that would be another frontier, right? So it's a narrow line with fluorophores. Narrow line with– I always have my ear to the ground trying to look at has anybody come up with anything clever in that regard. And how is it now? Like what's So, basically, you have about 50 nanometers of emission spectra.

You can let them overlap and unmix, right? Again, you turn your lasers on and off for each one individually, you put them on multiple cameras that look at each spectral band and all of those, they overlap.

It's doable. Are you excited? Are you just like white light? No, no, no, no, lasers, all lasers, right? So, we're narrow in that regard, right? At least we can say that. But, yeah, the excitation spectra are just as broad by vibronic bands as the emission spectra. And you can temporarily also separate the signals, maybe? Yeah, yeah, to a degree.

Yes, yeah, yeah. That's what we do. Anyway. Has there been anything useful out of, like, plasmonics or resonators or anything like that? Not yet, not yet. That was another area I was looking at early on, right, is little silver resonators. And so forth as to whether I could do something like that. But so far, no.

Again, delivery is a pain, right? Yeah, yeah, yeah. You know, you're now in an environment which isn't a vacuum anymore. And so the electric fields around where you are can also perturb your lines, right, and so forth. So, you know, just stark shifts or whatever else from what else is around. All sorts of shit can happen, right? It's messy.

But, yeah. So, you know, incredible to see all those microscopes and kind of the full stack of everybody working on even just trying to process the data coming out. If you weren't doing this, what would you be like something? What would you do like something completely different? Like what is the thing that you wish, you know, if you had two lives, like what would you be working on right now? Yeah.

I actually did during the pandemic when it was difficult to do much. I went down two rabbit holes. One was nuclear space propulsion and tried to see if that was feasible and worthwhile as, you know, since Musk has already talked about Mars and so forth is maybe use starships for people. But maybe there's cargo or something that might be better transported by other means.

I went down that rabbit hole and I kind of decided that based on orbital mechanics and so forth, probably Starship is the best solution to that. The other rabbit hole was energy.

I really want to understand the energy economy. I read this book from an expert on it called Baklav Shmil called How the World Really Works, which is about the energy economy. And, you know, what is the right mix of fossil fuels, nuclear, solar, wind, et cetera. I went pretty deep down that rabbit hole and came up with the conclusion. It's obvious that the right long -term answer is nuclear.

Just no doubt about it. It will be scalable.

It's done right by far the cheapest method. And it's basic physics, right, because the costs and everything come down to energy density, right? And because energy density is materials and materials is cost. And the energy density of nuclear is orders of magnitude better than hydrocarbons, which is orders of magnitude better than wind and solar.

Okay. So anybody who thinks that we're going to replace hydrocarbons with wind and solar are completely delusional, absolutely delusional. So the reason that nuclear has stagnated so much is because of the Nuclear Regulatory Commission. Is that, you know, so it was formed in 72 out of the AEC. And in that time, 50 years, there's been hundreds of proposals for new reactors in the United States.

Only two just last year came online, which were the Vogtel reactors in Georgia. Two out of hundreds of proposals over 50 years. The regulations are insane. They have certain things like– in fact, there was a review of this just under the new administration to try to see if they should modify this called ALARA, as low as reasonably achievable, right? What does that mean? What that means is no matter how much you spend, if you have another dime, you should spend more to get the background radiation from the plant even lower than the native background radiation.

What kind of sense does that make? Or the Vogtel reactors where they had already poured the concrete and rebar for the thing and then they changed the specifications and made them rip it up and do it all over again. How are you– they're intentionally sabotaging the ability of nuclear to be cost competitive. And so you'll see study after study by solar and wind proponents that nuclear is unaffordable.

They've made it unaffordable. The energy density is such that done right, it will be by far the cheapest. By far– and if you look at our world in energy or things like that, in terms of deaths per kilowatt hour produced, nuclear is orders of magnitude safer than any other technology that exists. Any reasonable, rational person, economic or physicist would realize that this is the right answer.

But the hysteria and the politics have just absolutely killed it. Did you– when you did your deep dive– it's kind of funny because I actually did one of these also. I had like a week off. We shut down our startup every year during Christmas. So a lot of people are from Europe and they go home and everything. And I read like a nuclear textbook and talked to a lot of nuclear friends.

Right, right. And I kind of came away from it thinking– finding kind of similar feeling like it's not really about the technology that much. It's really like you could just build the PWR reactor. There's so many good reactors. You could do PWR but it's not the best. Yeah, there are better ones but even the PWRs are pretty good. You should have a high temperature reactor, okay? Like again, you know, Triso with helium cooling like X Energy is doing or Radiant.

In fact, the good news was Radiant, which is also using Triso, just got an award from the army to do some initial micro -reactor projects, right? So if they get off the ground, that's– there's one by a guy here at Berkeley, Per Peterson. I can't remember the company but in Alameda. They're also using Triso pellets in sort of a molten salt bath.

When you were looking into this, did you have– besides the regulatory aspect, was there kind of like a– oh, if this piece of technology turns over, suddenly a lot of things become possible. Was there like kind of a technology insight or a physics insight? I really feel like the technologies really exist already. It's all there, okay? And I particularly like the high temperature designs because a huge amount of our energy economy is process heat, right? Right.

And if you have a light water reactor, you don't have the temperatures necessary for a lot of the process heat applications you'd like to do. So if we can do high temperature reactors at scale, that's the thing that would make fossil fuels basically just a materials thing as opposed to an energy thing, right? But yeah. But the wind and solar has just been proven wrong.

You look at Germany. They're over 40 percent solar now. It's the stupidest place on earth except maybe the North Pole to go all– and so what happens is they have their– in the winter, there's no wind and there's no sun, right? Because it's just a big cloud cover over it. So they buy electricity from either the hydro parts in Norway or from the nuclear reactors from the Mesmer project in France.

Yeah, yeah. Now, France is still 70 percent nuclear, right? But those reactors are crazy old and they're nearing end of life and they haven't thought anything about how they're going to replace them. And that's going to come up– we keep extending the life of the reactors we have in the US and they're good but they won't last forever, okay? And so it's the most insane– there are many things that– like I said before, I kind of feel like there was sort of– whether it was 2008 or before, it's kind of an inflection point where it felt to me like things kind of went nuts.

And not being rational about energy policy is really, really high on my list. I mean as Nobel laureate, you get invited every year to this Lindau thing on Lake Constance to talk to 600 -odd young scientists about whatever you– so I gave an energy talk there. But again, I just– it was great because I got attacked left and right but because a lot of the Germans are very pro -solar, right? And so I took them out all for beers afterwards so we could debate it, right? I don't think anybody changed anybody's mind but I tried, okay? But it just– it's not science that's holding nuclear back.

It's sociology and politics, yeah. So earlier in your conversation, you said that the best time of your life was in Bell Labs. Right, yeah. Can you tell us what's the magic about this place? Do we want to bring it back and what do we need to do? It was magic. It would be wonderful if we could bring it back and I doubt we ever will. So the magic was– it understood how science is done.

Creativity is not done by committee. Creativity comes out of the individual human mind. It's useful to have other people to bounce ideas off of but it's the guy waking up in the middle of the night where just the different pieces fall into place. And in order to be creative like that, you have to be fully immersed in a problem. You can't be like a university where in a couple hours I'll be teaching a class.

You don't have– you can't be writing grants. You can't be doing all of this other extraneous stuff and still focus your mind completely. This is how Shannon worked. This is how Hamming worked. This is how towns worked. This is how 12 f***ing Nobel Prizes out of one institution. The reason is they let scientists do science all the time. And furthermore, in the physical research division where I was, there were 100 PIs.

We were called members of technical staff. It didn't matter whether you were a new hire or whether you had a Nobel Prize. You could not have in your group more than one postdoc and one technician. That was it. You could not build empires. So as a result, how do you get stuff done? Well, you collaborate. And furthermore, there was no focus.

There was no departmentalization. There was no– what's the idea of having a chemistry department over here, a physics department over here, and a mathematics department over there? That's insanity. Siloization of academia is nuts. You need to put people together because the advances happen at the interface between disciplines. Bell understood that.

So we were all in one corridor of a massive building. There was one straight line with labs on either side going all the way down. In order to get lunch, you had to pass 50 other people. And those stochastic interactions you would have with other people either going to lunch or in the lunchroom was what made the whole damn thing work.

Okay? I mean it was– you felt like do I belong here when you first go into the door there? There's a lot of history there, right? And you're like, my god. Can I measure up? And so you bust hump in order to try to prove yourself, right? And when Harold and I were there, we would come in at four in the morning and we would always fight for the first spot in the parking lot.

And if he beat me, I would put my hand on the hood of his car to guess by temperature how many minutes he beat me by, right? And then we would work until the sun rose. Then we would play tennis and we'd come back to the lab. Then we'd work until six. Then we'd go to the same Chinese restaurant two blocks away for dinner. Then we'd come back to the lab and work to ten.

And we did that seven days a week. Boom, boom, boom, boom, boom, boom, boom, boom. You get a lot done if you do that. A lot done. Okay. And again, when you're playing the tennis or when you're having the Chinese dinner and your mind is off of it, or when you're sleeping, the subconscious mind is turning over all of the shit that has been put into your mind by your conscious effort during the day and the conversations you had with other people.

And the answers come out just f***ing out of nowhere. The answers just pop up out of nowhere.

And the fact that there was no focus, you know, in my time at Nearfield, what did I do? I did high density data storage. I did looking at fibroblast cells in their cytoskeleton. I looked at single molecule imaging. I looked at profiles of emission from fibers. I worked on fiber lasers. I worked on tissue sections. I worked on a zillion different things with so many different people.

And I gained not just in the application but from the technology. I started going with my pulled pipette tips. And then once I learned about the fibers and how I could exploit those, I pivoted to the fibers. And, oh, well, the fibers, in order to pull the tips, well, that's silica glass. That's way too hot to pull with a heated filament. Well, guess what? The guy who invented the carbon dioxide laser is down the hall.

So I just go and borrow a carbon dioxide laser from him, which then absorbs enough in the glass that I could then melt the glass and do that. I don't know how many times I was stuck on something. And it turns out that the guy I needed was– or I'd start reading the literature about something. And it's the f***ing guy down the hall that I've nodded to a hundred times and never really had a conversation with.

It was magical. To be– what is it? Not politically correct, though. It had other aspects to it. And that was– it was not an environment for a happy family life.

It was not about work -life balance at all. So it was in the era where most of the guys had stay -at -home wives. I think that was– I think those people deserve the Nobel Prizes as much as the guys who got the Nobel Prizes because they wouldn't have happened without those people. It would not have happened. We live in a world now where it doesn't happen, where it's not expected that you're going to have a stay -at -home wife and you're going to be working 16 hours a day, seven days a week.

That's a unicorn now. It doesn't happen. So that's the trade we've made. But don't think that you can just put a building up like Bell Labs and have people work even 50 -hour weeks. And most people don't want to even work 50 hours these days, at least twice 100 hours like we did, right? But there is no substitute for just full immersion in something, and you can't do that when you have a family.

So that's the price. Yeah, that's interesting. So kind of the implication is basically that it's not just about the environment or the freedom or anything. It's also the fact that people were working nonstop. Basically, it's difficult. You think that's important enough that it would be difficult to replicate without that aspect? Absolutely.

You can't. Impossible. Literally impossible. To me, it's an immutable law of the way the human brain works. It's that you have to be fully immersed in a problem, fully immersed. Any distraction, your mind will latch onto instead of doing the job it's supposed to do, right? So also, you know, with all the distractions of that damn box in the corner that you type into, right? And you spend your life in front of a screen.

Spending your life in front of a screen, unless you're a theorist, is not the way to get science done, okay? So, you know, when I work, I intentionally make sure I turn my Wi -Fi off, okay? So if I had any interest to, like, click on something, I'm just going to come up with a blank web page, and it's a reminder of myself, f***ing focus, dude.

Don't mess around, okay? Yeah, yeah, yeah. So, yeah, it's– there's a cost to everything, okay? And, you know, call it the sort of– again, you're fighting back sort of an– you're trying to create order out of nothingness when you're creating some new scientific instrument or principle or whatever. Creating that ordered structure requires enormous energetic input, okay? If you don't put the energetic input in by working insane hours, it ain't going to be built, okay? So.

Yeah, I mean, I think– I wonder if, you know, there probably are some people working that way today, but some of it is also, like, maybe you get more of them in places where it's more commercial. That's one. Startups is the logic. You know, if you have your back against the wall in a startup, some of those people work insanely hard. Yeah, and the other one is that even people working very hard in academia might be spending half their time on paperwork.

So it's like you might be working that hard, but, like, half of your work is paperwork, right? Yeah, and as soon as– there are times, particularly when I'm focused on something and I know the path, that you have to tune out the rest of the scientific community. You don't listen to the scientific community. You focus, right? I mean, you don't need those distractions.

So I talk about blinders on, blinders off. There are times in your life when you have to have the blinders on, when you know what your goal is, when you know what you have to do when you do it. And any time you take the blinders off, it's just temporary to find the missing piece to kind of go on. But you're focused– and then there are times when, you know, my unemployment are like that.

You have to learn to take the blinders off and I could do anything. I could go and become a chef in a diner or whatever, a cook, you know. But– so you need to take the blinders off to survey widely. But yeah,

you know, academia is just almost the worst possible system one could create if the goal is to get science done. And the other problem is so many things in academia. Not only is it about having to get the money. Not only is it the peer review causes everybody to have to think the same way in order to get the money. So it punishes creativity.

It's also– and the teaching and the interruptions that you get from that.

It's– and that– again, show me the incentives and I show you the outcome. What are the incentives for academics? The incentives are to publish papers. It doesn't necessarily mean they have to be good papers. They have to be papers. And to train the next generation. And so– and the more grant money you get in, the more postdocs and graduate students you have to have in order to administer that stuff and to do that work.

And so– and the more that you're going to move up the tenure ranks and so forth, if there's more overhead going to the university because you have more grants. So it's all incentivized to grow large groups. Okay. Until I came here and started Cell Observatory, I never had in my entire life more than three people working for me at one time.

Okay. Because I viewed every project as a 50 -50 collaboration between me and that other person. But as soon as you start to do that and you grow these groups, all the– throughout my entire career, since I was in grad school, people were talking about science is in crisis. We don't have enough money. Okay. It's now louder more than ever under the Trump administration.

But they haven't hardly done anything. Okay. The NIH budget is just about the same as it was last year. Okay. Yes, they're slow pedaling some of the stuff. But overall, science has grown at an enormous rate. And why has it grown? Because you hire all these grad students and postdocs who want to become PIs. And so more and more, this whole structure gets bigger and bigger and bigger.

But there aren't enough universities for all of these people, right? There's a lot of washouts and flameouts and unhappy people who spent the better part of their 20s, very productive years, being just slaves at crappy salaries to work for some PI. The PI gets all of the glory. And they end up either starting again as an assistant professor at some cow college in the middle of nowhere or else they flame out and they do something else, right? The system is seriously broken.

There's nothing wrong with academia that getting rid of 95% of it wouldn't cure.

But that's how professors like you or like some other groups find a way around and succeed, right? I say that I use my HHMI money and the other sources of money we have to put a force field around the third floor of Barker Hall to keep the rest of Cal out. That's the way I view it, right? I'm trying to create my own little Janelia Bell Labs within this microcosm that I have here, right? But yeah.

Yeah, that's interesting. What's like the– if you were 20 years old today, where would you go to try to get that type of environment? I know exactly where I would go if I were 20 years old today. I would beg for a job at SpaceX. SpaceX is the new Bell Labs, okay? They have a somewhat narrower focus. But do you know how many technological problems they are solving? They've created new steels for the Starship and the engines.

They probably understand more about computational fluid dynamics than anybody ever in history. If they're going to be building a moon colony or a– do you know all the human factors, the environmental factors, all the other stuff? They already know a lot of this. Hell, they've got the dragon capsule, right? There's so much– to be able to do space travel is such an interdisciplinary problem.

It's at least as interdisciplinary as anything. Oh, the control algorithms to do a hover slam for the Falcon 9 rocket. That was non -trivial too, okay? The compute that they do, the real -time compute that's necessary. There's so much immense science and technology happening within that organization now and they're just f***ing getting started.

I would tell any young person who's listening to this in any field of endeavor that if you want to have a really f***ing exciting career and you do like the idea of working your ass off because those guys work their asses off. If you ever watch those launches though, have you ever seen anybody happier than those guys watching their rocket do what it's supposed to do like on Flight 13 recently? It brings tears to my eyes to see how happy those young people are doing that.

That's the way you should live. You should be making a better world and these people are making a better world. I just can't believe how much shit that company gets and how much shit Musk gets for basically pulling us up, kicking and screaming to a better world than we've had before. How much of an impact has Starlink had in third world countries already? I mean, come on.

It's insane. Absolutely insane that he gets the shit he does for the impact that he's had. And his whole team, you know, they paint the same brush for all of those young people doing that work. And they're producing miracles daily. Yeah, I think, is there a SpaceX of biology? Like, I guess if you're a biologist, where are you going? Well, as a biologist, I don't know.

Every, you know, the thing about biology is it's a wonderful, you know, this is the last stage of my career and I'm never going to be a biologist. But I study biology and the wonderful thing about it is it's the last frontier. It's mystery everywhere. We understand so little. We're, again, in the pre -Kepler era. We're in the phlogiston era, okay, of understanding biology.

There are so many things we think we know that are just bullshit, you know, that it's going to– and it really is the final frontier because it is the most complex matter in the known universe. And it's going to take generations, if not eons, if ever, to have enough of a mechanistic understanding to really do a virtual cell like they say they want to do now.

People just don't understand exactly what big of a mystery it is. Yeah, I mean maybe in that sense it's more– it's a little bit like a better– it's an easier field to do some interesting work in even if you're not at a SpaceX. Absolutely. Because there's just mysteries. And there's, I think, a lot of opportunities to do it on a small scale.

You don't need to build an empire in order to do that stuff, right? Mm -hmm. It's not like trying to do fusion or quantum computing or something where you're going to need, you know, bigger resources or whatever. There's still opportunities to do it on a small scale. Nice. Okay. I think we should wrap it up. Okay. This was awesome. Thanks for listening.

That's amazing, guys. Yeah, yeah. Thanks for your time. Yeah, well, thank you, guys. I appreciate your coming out. Awesome. Thank you. Thank you. Thank you.

Thank you.