No Priors Ep. 34 | With Ginkgo Bioworks Co-Founder and CEO Jason Kelly

1. 0:00 – 6:47

   Synthetic Biology and AI Revolution

   1. EGElad Gil0:00

      (instrumental music plays) Biology is undergoing a digital revolution as we build developer tools and production infrastructure for synthetic biology. How will it change industries? How does it intersect with AI, and how do we rethink biosecurity? This week, Sara and I are joined by Jason Kelly, co-founder and CEO of Ginkgo Bioworks, to discuss their goal of making cells as easy to work with as computers, their data strategy, and the tech keeping the next pandemic at bay, and in general, what cell programming will do for the future of food, medicine, and agriculture. Jason, thanks so much for joining us today.

   2. JKJason Kelly0:34

      Yeah. Thanks for having me on.

   3. EGElad Gil0:36

      So I think there's a lot of talk about synthetic biology and how biology and DNA and proteins are effectively just code and you can manipulate them in different ways now and things like that. I'd love to just get your view of both what Ginkgo does as well as what does synthetic biology actually mean?

   4. JKJason Kelly0:50

      So the... I think the founding idea of synthetic biology is that DNA is code, right? And inside of cells are A, Ts, Cs, and Gs, essentially on, like, a tape, and it, and it is very, like, surprisingly analogous to zeros and ones, uh, you know, inside memory in a computer. Th- that's roughly where the similarities end. Okay? Like, once you get to the next step of what the cell does with that code, we are in a totally different world. It- it- it is not virtual, is the first thing, right? It is a physical thing. The code itself is literally physical, right? It is a polymer, uh, and it is going to use that to make proteins, which are basically little pieces of nanotechnology, and they're all gonna be bumping into each other, and it's all crazy. It's not physically isolated like you would imagine with a semiconductor chip. It's not built by humans. So you have this really interesting thing where the hook is there for people in tech to engage with biology, but then once they get in, they're like, "What the fuck?" Uh, and, and-

   5. EGElad Gil1:45

      (laughs)

   6. JKJason Kelly1:45

      ... and so, and so, like, I'm happy to talk about those pieces, but I, I think you're right. The core idea of synbio is that it runs on code, and then what can we bring over from programming into this world that actually sticks? And so I think what synthetic biology has been, you know, really since it got going, I met, um, you know, the founders of Ginkgo back when we met at MIT in 2002. That was, like, early days of synbio. It's about 20 years now. It's basically engineers asking the question of what can they bring over into biology that's actually gonna work? And some stuff has been left by the wayside and some things do work, and the, and the latest technology that's being tried now is AI.

   7. EGElad Gil2:18

      C- can you walk us through what you actually think does transfer over and then where there are one or two unique challenges and then how does AI help to solve for s- some of those things?

   8. JKJason Kelly2:26

      I'll tell you, like, a funny story, right? So, so, uh, one of the fellows I started the company with is this guy, Tom Knight, right? And Tom Knight started on the faculty at MIT in 1972, okay? Right? Like mainframe computers, punch card computers. He was a computer architect for a very famous mini computer, which was, like, the size of a refrigerator called the Lisp Machine. Okay? Like, Symbolics, that company as well, the founders of. Like, old school classic Stephen Levy Hackers in the book kind of guy, right? Mid '90s, he realizes this thing about DNA as code and basically it's like, "Forget computers. I'm moving into programming DNA." He's still Tom, right? He's been, like, teaching the semiconductor course for 20 years at MIT at this point. Opens a wet lab in the MIT computer science building. Starts growing bacteria. Freaking everybody out, right?

   9. EGElad Gil3:08

      (laughs)

   10. JKJason Kelly3:08

       And he puts up this flag and he's like, "Hey, computer scientists. Like, DNA is code. If you're interested in, in this thing, like, come over and try it out." Right? Some of us came over and we're like, "All right, cool." We got there, got our hands wet, and we're okay with it. A lot of computer scientists, they'd get there. They... Tom's like, "Okay, here's the lab bench. Remember, this code is physical. So if you wanna compile it, I'm gonna have to teach you how to do molecular cloning. And here is a pipette. And you're gonna sit at this bench and you're gonna do these steps." Okay? And, and the person would do them and they'd get a result the next day. They're like, "Wow, that's really interesting." And then, and then they'd do the same thing again the next day and they would get a different result. And they'd be like, "Tom, I just did the exact same thing and I got two different results. Like, what's going on?" And he was like, "You're never gonna know."

   11. EGElad Gil3:51

       (laughs)

   12. JKJason Kelly3:51

       And, and, and, and it would just break their brains, right? Like, they'd be like, "Fuck this. I'm out of here." You're leaving a world in computer science of, like, pure logic, right? At the end of the day, if there's a bug, you can always run it to ground. And, and that's because, A, these are systems that run like clocks. B, we design them. We design them, right? Uh, and, and so, like, you at the end of the day can go in and figure it out. And in biology, it's like, sometimes you can, right? And sometimes it's just a part of the biology that we just frankly don't understand well enough that's broken and, like, tough luck. And, and you, and you gotta, like, stomach that. And, and one of the things, my favorite things about AI is like, how does that neural net work? Nobody knows.

   13. EGElad Gil4:32

       Yeah, yeah, yeah.

   14. JKJason Kelly4:32

       You're about to experience, like, like, like the analysis of these neural nets is gonna look like systems biology, right? It, it's gonna be like go in and like try to back figure out a thing that you didn't design, my friends. And, and so like-

   15. EGElad Gil4:45

       (laughs)

   16. JKJason Kelly4:45

       ... that, that'll be your first taste of really feeling like a biological engineer, right? But, like, why bother? Like, why work with these neural nets that, my God, we, you actually can't easily debug and understand why it's hallucinating and all this stuff? And the answer is 'cause they're powerful. It's worth it. And that's the same reason you wanna do biological engineering. Like, even though it's unpredictable, even though you're gonna, it's gonna be so frustrating, it's not gonna do what you want and blah, blah, blah. People are gonna, like... It's because the substrate is incredible, right? It self-replicates, it self-assembles. We have nothing else like it in the physical world. So you wanna work with it even though it's hard. That, you know, that, that's, that's what, that's what ultimately gets people passionate about this stuff.

   17. EGElad Gil5:23

       Yeah, totally makes sense and I think one can argue that neural networks are actually heading even more in that direction because as people build systems that can code themselves, we're gonna end up with evolutionary systems that are completely n- non-designed.

   18. JKJason Kelly5:33

       Yeah.

   19. EGElad Gil5:34

       And I think then we're truly in the world of biology where you have evolution kicking in and, you know, to your point, evolution is really messy, right? It's always optimizing for the utility of something versus the form of it. It reuses parts aggressively. It creates enormous redundancies in weird ways that you don't know that there's a p- a perturbation here and it propagates across, like, in weird ways. And so, I think people are really underestimating what happens once we have self- ev- self-evolving neural nets, which I, I think is coming quite soon.

   20. JKJason Kelly6:01

       ... be still my heart. It's, it's gonna be great. It's so cool. I mean, this, it's so cool. It's so worth it. Yeah. I mean, it's neat because there's a magic to it, right? Like, again, uh, people, people like different stuff. And, and, like, I get that. I think this would be, like, one of the big cultural divides. Th- there's a certain kind of mind that really likes things to be predictable, and there's some people who just like the magic, right? Like, some of what's cool about biology is the, is how hard it is to understand.

   21. EGElad Gil6:22

       So how does, how does Ginkgo go about harvesting all this, you know, this shift in biology, in terms of the ability to manipulate these systems using molecular biology and molecular cloning techniques, and then software tooling and other things? Could, could you tell us a bit more about the company and where you focus and what you've done with it all to date?

   22. JKJason Kelly6:37

       So one of the ideas that we tried to bring over from computer science was abstraction, right? And what is abstraction? Well, in Tom's era of computing,

2. 6:47 – 14:54

   Abstraction Layers and AI in Bioengineering

   1. JKJason Kelly6:47

      in order to be a computer scientist, you had to be an electrical engineer. Because how do you program a computer if you don't know how a computer works? Now, obviously today, like, an eight-year-old or something is, is able to, to program a thing on their iPad by drawing boxes around, so, like, what happened, right? Like, well, assembly language, operating system, programming language, graphical programming language, right? W- we build all these abstraction layers to, to split the disciplines of electrical engineering and, and computer science into their own paths, both of which had very long roads, right? And so one of the big things we did at Ginkgo when we started the company, you know, 15 years ago, like, unimaginably a long time here, was to, to do that split from the get-go. So we ha- we have part of our infrastructure, we call it a foundry, taking a page from semis, that is basically a group whose whole job is automate and scale the lab work, okay? And move away from a system where that lab work is being done by hand by a scientist. And then the DNA programmers, who are really typically scientifically trained and, and PhD biology types, they order from that system to get their work done. That is actually very difficult to pull off. It's culturally difficult because, like, a scientist does not want somebody else to do their experiments and, like, they're a good scientist. You know, like, there's all... No, no, no. There's a, a whole long laundry list of why it's hard, not to mention that when you first try to build the infrastructure, it sucks, okay? And so The Foundry team has been able to drive enormous scale economics in doing the lab work, which gives us the data of lots of different genetic designs that we've tested, which is exactly gonna be useful for the AI stuff, but is also just generally useful, right? Because y- you're, you gotta try a lot of designs to get the cell to do the thing you want it to do. And so that's, that's been probably, like, one of the biggest activities the last decade at Ginkgo has been, like, driving that scale.

   2. SGSarah Guo8:24

      Jason, what's the right way to think about the abstraction between, like, your customers and then your DNA programmers? Like, what's the spec that gets passed over, or how should we understand the science they do versus you?

   3. JKJason Kelly8:35

      So today, the way that it works is basically, a customer of Ginkgo's would be, like, you know, um, like, a recent customer is Merck. Okay? Like, Merck, Novo Nordisk, Biogen on the pharma side, Bayer, Syngenta, Corteva, the biggest ag companies in the world, all customers, then a lot of startups, right? And, and the, and the way they interface with us is that we basically are agreeing on a spec. We're like, "Okay, here's what I would like the cell to do." They'll, uh, they, they tell us, and we agree on, like, a timeline to develop it, and we're kind of like a prop software development shop. Like, we're, we're gonna, like, make it for the customer and then license it to them, and they'll take, and they'll own it to go develop their product. In exchange for that, we'll get a royalty, and we'll also get some payment along the way. All right? That's the business model. Today, their, uh, their scientists don't use our infrastructure. My scientists, you know, or our scientists here at Ginkgo, they do the, they use the infrastructure, and they have this interface with the customer about hitting goals. That's mostly a technical limitation. I think it would be very cool, ultimately, to have scientists at all these companies accessing our infrastructure directly. It's just too early. That's the problem.

   4. EGElad Gil9:36

      And then how does the AI come into the picture? Or when, when, and how did you start using it? And has this current wave of AI impacted you, or how much do division models, LLMs, et cetera, matter relative to what you're doing?

   5. JKJason Kelly9:46

      So the short answer is, like, we have u- we do a lot of protein engineering, so, you know, you want to program a bacteria, right? Okay, so bacteria has a three million letter genome, and, like, a customer has asked you to, uh, express a protein. And remember, a protein are basically like the little pieces of nanotechnology inside the cell that bump into each other and, like, do all the things. Like, you're st- sitting there, you're like a big giant bag of proteins, right? Uh, a- and so, and so, like, they want to make a lot of this protein because it's gonna go into cold water laundry detergent. Okay? So people don't realize this, but, like, the reason cold water laundry detergent doesn't need hot water is 'cause there's enzymes in there, okay? Proteins. And so they want to make a lot of this protein. And, by the way, if they could make it more active, like break up dirt faster, like whatever reaction it is catalyzing... So if you remember in chemistry class, like, a catalyst makes a certain chemical reaction happen faster than if you don't have the catalyst. Okay, so this enzyme is a catalyst, and I want to make it also just better. Like, so I want to improve the quality of the enzyme, and I want to make a ton of it. All right? That's the spec. And so, how do we do that today? Well, we would have, for example, a host strain that's really good at producing a lot of protein to begin with. Okay? So, so think of that more like an existing software library. So that's one form of leverage from existing data assets is literally, like, a hard physical asset, an actual microbe with a genome that I engineered in a project previously that is useful for your project. So now, now we're starting from a good place. We're already starting to make a lot of protein, but you wanted to make it more active. You want to make more of the cat- catalysis. Okay. How do you do that? Well, remember that protein is encoded in DNA, and the sequence of DNA determines the, uh, effectively everything about that protein. But in this case, what you care about is how good of a catalyst is it. A- and so you go in there, and you have c- certain tools to try to model the protein, this, that, and the other thing, and you make some choices and along with the software tools, and you say, "I want to try these 1,000 designs of the, of the DNA in the lab and see how they do." All right. The, and you then get that data back on how they perform. You use that to update these design tools you have, and you do it again. And that's what we've been doing at Ginkgo for a long time, including with neural nets and all this stuff, the latest and greatest, you know, all that. But just on that, like, data asset.I think the new idea is, is on the back, and this is... We just announced a deal with Google, um, a couple weeks ago. The new idea is, can I make a foundation model that will be additive to what I'd previously been doing just with the data I was getting on enzymes? And so now I have a foundation model that really is, is not specific to catalysis or anything else. It, like, speaks protein, you know? Right? Like, just like GPT-4 speaks English, right? That's what we're gonna try out with Google. That's what we think is, like, a really, uh, new i- new idea. And people... You know, there's people obviously who work on it, and obviously Google themselves with AlphaFold was, like, one of the first generations of this. But we th- we see a lot of ways to make it better and make it bigger and all the things. And so we'll see. A- and, uh, we'll see how it goes. Let me just also give you, like, one other thing why I think bio is particularly interesting for folks that are interested in AI in general. So this whole idea of, like, a foundation model plus fine-tuning with specialized data, right? Like, we all... Like, all the people who pay attention to AI, like, understand this idea, right? So let's, let's just take it in, in one of the categories of English, like legal, right? Like LexisNexis, they have all this data. We're gonna fine-tune GPT 4 also. Okay, the... That thing has to compete with, like, a lawyer at Ropes & Gray, right? And a lawyer at Ropes & Gray has trained for 15 years, uh, you know, being taught by other humans how to do law. They are writing contracts that were designed to be understood by human brains. They work the way we think. Uh, they're writing that contract in English, a language that co-evolved with our brains. You know, just language in general co-evolved with our brains, uh, to also, uh, give us leverage from how our brains work. And we're asking this computer brain, the neural net, to compete with us on our turf. But it's a pretty high bar that it's got to compete with. Now, let's go over into biology. I remind you, it runs on code, sequential letters. Feels a lot like language. It ain't our language, right? We did not invent it. We do not understand it. We do not speak it. We do not read it, write it. And so I feel like these computer brains are gonna kick our ass a lot faster in this domain than they do in English, right? Like, you... You know, like, I think the early applications in AI are all gonna be, like, replace the intern, uh, not, not the partner at Ropes & Gray a- at best for a while. Whereas over in bio, it should be... It sh- it could quickly become the best. I- I- if you're looking to understand, like, where AI is really gonna flip the script and not be, like, kind of a low-level Clay Christensen-style disruption, which I think is sort of what's happening in English, but rather be a, like, split in the atom, it's bio.

   6. EGElad Gil14:42

      When people talk about protein folding-related models, you know, and to your point, there's things like AlphaFold, and there's a few new companies that have been set up to basically focus on protein folding models because of the breakthroughs in AI, they kind of divide it into

3. 14:54 – 19:48

   AI Applications in Biology and Pharma

   1. EGElad Gil14:54

      a few markets, right? There's sort of the pharma market, which is better designed for pharmaceuticals and biologics that are used as drugs. And then there's more the industrial, the catalysts, the ag, those sorts of things. I'm sort of curious, like, how do you think about those relative markets just in terms of sheer market size? Because if you look at the cost of developing a drug, it's like 1.5 billion per drug, but very little of it actually goes to the underlying molecule that's being used, relative. Most of it's clinical trials. But for ag or for catalysts or for other things, a lot of it could actually go into the molecules. So I'm just sort of curious how you view those as relative markets for this kind of stuff.

   2. JKJason Kelly15:28

      So Giggle has, like, an AWS-style business model, right? Like, you need compute, I don't really care if it's... You know, if you're AWS, whether it's medicine or they're a startup or you're video streaming, like rock and roll, right? So, so, so we have an attitude that, like, we're supporting cell engineering wherever it is. It, it is definitely different by market, like pretty substantially, right? Both the assets you need, the enterprise sales, everything is different. Um, the biotech, you know, it's not, it's not a small industry. But it, it's... It depends, like, which side of the house you're looking at. If it's fees, it's probably a little more the, the more valuable markets that you could get more research fees. Um, but royalty is a different game, right? Like, certain things go to market a lot faster. That's how I see it. But the real problem is, like, there aren't any platform services (laughs) . So, like, the other thing that's just wild to people in the tech industry is like, "Where's the SaaS platforms? Like, where is all the horizontal stuff? Where's the operating systems?" And it's like, nowhere. Like, like, vertical integration, vertical integration, vertical integration. Merck, Pfizer, Bayer, Syngenta. Like, every one of these companies is like its own tech stack (laughs) top to bottom, you know? And the, and the closest thing they do to anything vertical is buy equipment from the same people (laughs) . Uh, but like i- it's really fascinating, like totally different industry structure, and so that surprises people, right?

   3. SGSarah Guo16:37

      Jason, what's the most rational way to understand that? 'Cause y- you, you look at that as a non-bio person, and you're just... Yeah, it doesn't make any sense, right?

   4. JKJason Kelly16:46

      So the rational way is the work being done cross-product is too dissimilar to support common platforms. I don't agree with this. Obviously, my entire business model is predicated on that statement being false. But that, that is the reason. You're like, "Oh, well, because this... The, the platform you build for customer A will never w- read on customer B." And so now, now you're just bad, right? Like you're, you're having to build a new platform for every customer. You're getting no leverage. You're getting... None of the reasons that, like, operating systems and, and data ce- Like, why does data centers work? 'Cause c- compute is really generic, right? And like, you can use software to make it different. But the hardware underneath is all common, and now we're seeing a little edge case difference here, right? Like, actually, the CPUs aren't that useful for the AI. Now everyone's freaking out about the GPUs. So we're having, like, an instance of, like, hardware variability. But the argument would be that, that type of hardware variability you're seeing is sort of like per company. Okay? Right? Like, you know, or at least per modality in pharma, which modality is like a fancy word for type of drug, right? Like gene therapy is gonna have very different stuff than, you know, whatever. And there's truth to that, right? Like, like it's not... That's not a false statement. It's just a question of degree, and we happen to believe that when it comes to the engineering of organisms, that that is at parts common. But plenty of people think we're wrong.

   5. SGSarah Guo18:01

      Yeah, it seems like if you just go back to this analogy of like...... human-designed computation where you're building systems from the ground up that you can understand, but you're all discovering the same set of systems with common building blocks and the need for data analysis. It would, it would shock me if that would not eventually be true, and it's a, like a temporal cultural figment of these companies. Okay, maybe one more general question about how to think about AI in, in pharma writ large. Why do you think we haven't seen AI-discovered drugs yet? Because people have been talking about it for a long time. Will we, and will we see it soon?

   6. JKJason Kelly18:36

      Well, a- and I... So first off, I would say, like, the, people have been talking about it for a long time is sort of like saying three years ago, like, "Why haven't we seen good natural language processing in AI?" People have been talking about it forever. Right? So, so I think there is an element of, like, you need the breakthrough, right? Have the neural nets been big enough? The big limitation i- in bio is the availability of d- of data to train these things, right? And so you have this tough situation where, like, everyone who's doing these models are training on the same data, right? And, and so one of our advantages at Ginkgo is we just have a ton of data. That's a real gap that I think is... I think partially it's like, have people gone big enough for it to have happened yet? And I think, and I think it... Now people are trying, like, working on trying Recursion is another great example. And, and, like, you know, it might still not be big enough or more likely it's not enough data, right? And, like, there's no- nothing stopping you from making a giant n- neural net at this point. You know, like, uh, the tech industry is gonna commoditize th- that infrastructure. But, like, y- you might not have enough data to give it to solve the problem you're asking, Sarah. Right?

   7. SGSarah Guo19:33

      Where does Ginkgo's data come from? Is it like your own experimental data?

   8. JKJason Kelly19:37

      Yeah, we have a 300,000 square foot robotic lab that they, uh, built in the last 10 years. And so we generate that and we do it in service of our customer projects. We can do our own data generation, but yeah, that, that's where it comes from.

   9. SGSarah Guo19:47

      I

4. 19:48 – 31:42

   Rational Pandemic Response Program Building

   1. SGSarah Guo19:48

      want to talk a little bit about one area that Ginkgo has been an expert in, which is, um, infectious disease, right? Can you talk a little bit about the work you guys do here? And I think a, a question everybody, like, cares about is like, are we prepared for another global pandemic? What has changed since COVID-19?

   2. JKJason Kelly20:05

      Yeah. I, I think, like, the reality is infectious disease is, is really scary and bad, right? But like the, the big, the big lesson of COVID is modern healthcare systems and our current infrastructure does not render us immune even in the developed world from pandemic scale infectious disease. Period. We don't just allow ourselves to not have defenses against things that are like society killers that are known to exist, you know? Right? Like, like this is not like, like, you know, a fanciful idea. Like it fricking happened two years ago, right? Like, you know, and, and so the... So what should we build? Right? A- and, and the answer is like a lot of different things, right? Like we should build rapid vaccine response, which is really good through Operation Warp Speed and kind of what we figured out with mRNA vaccines and just like, I got a target, I got vaccines for the entire country in three months. Not every version of this thing is vaccinatable. The other one we've been big believers in is like monitoring, like radar. I grew up in Florida, right? Like, we have radar systems that warn us for hurricanes. Okay? My co-founder Tom Knight, obviously a lot older than me, was explaining to me that when he was a kid, they would get three hours warning for a hurricane. Currently we find out about Hurricane COVID after it has landed in New York City a week ago. Okay? Like, unacceptable, right? So, so, so one of the things we're doing is like with the CDC, we run programs where we collect wastewater from inbound airplanes and we sequence the DNA and we look for pathogens. We look, we mi- we monitor variants and all this, both for flu and COVID. And I can add other things to that list. We have a similar program actually in Doha Airport in Qatar. We've got a program in Ukraine and think of these like bio-radar stations. And that's gets you baseline 'cause you also wanna... You look for anomalies, right? So, so like that whole thing has been missing. And so we think that's like y- where you start and then you want rapid response to be able to like basically patch. Think like cybersecurity. Like it should feel like that's your answer, right? Like, like cybersecurity I think is a bit the mental model... For like what the future of infectious disease response looks like. Persistent monitoring, rapid response, kill it. And, and the beautiful thing is, and it's scary and beautiful, remember these things replicate. So if you can snuff it out at the beginning, you win. Like speed will matter, right? I'm chair of a national security commission down in DC on emerging biotech and like the DOD just put out their biosecurity... It's like bas- basically biodefense posture review. And like the DOD maintains like millisecond preparedness in this country, right? Like I know that seems crazy but like that's kind of like how you ought to treat these things.

   3. EGElad Gil22:35

      Yeah. You actually saw that with SARS, right? 'Cause SARS, they both snuffed out with the original form, but then it leaked four times in the first two years after it was cultured in a lab. It kept leaking from what eventually became the Wuhan Institute of Biology when they moved it from Beijing to Wuhan. And it was really rapid response in terms of shutting down SARS outbreaks that really helped prevent it from spreading. And so I think to your point, there's good precedent in terms of trying to prevent spread and having it be effective assuming that the coefficients on the disease spread are reasonable.

   4. JKJason Kelly23:02

      Yeah, exact- i- certain ones are gonna be harder than others, right? But like, it's not really about precedent. You just need logic. If you could get it early, you win, right? Now the, here's the question of how hard is that? And like COVID was a lot harder than MERS, right? 'Cause for a variety of reasons. But there's probably a level of tooling that could even have stopped, that could even like snuff a COVID since we had nothing when that happened.

   5. EGElad Gil23:24

      Sure. I guess one of the things that people in the AI safety community bring up quite a bit is that one of the big risks that are associated with the use of AI and LLMs and these foundation models for biology is that there's some risk of a lone actor somewhere deciding to, to build a virus that is infectious and deadly and can sort of run through the population rapidly. How much of a risk do you think that really is?

   6. JKJason Kelly23:46

      The idea that like we know how to like exactly like design for that sort of thing is low. You could try something, but it's not like, "Oh, I know for sure someone's waiting to do it." Remember, you need data. It's hard to accumulate that for... Yeah, it's easy for me to accumulate data on, on enzymatic catalysis. It is a little bit hard to accumulate data on case fatality rate.

   7. EGElad Gil24:04

      A- and it's very hard to do because the argument I always hear from the safety community is, "Oh, the lone actor will of course be somebody who isn't that well-versed in biology anyhow 'cause the people who are well-versed in biology are unlikely to do these types of attacks." So it's a, it's kind of this really weird needle that's threaded in the community to try and make arguments that-... to your point, seem to not really hold up relative to the reality of what's needed to actually pull something like that off. At least, uh, today.

   8. JKJason Kelly24:28

      I would basically agree with that today. The, the only thing I would say though is like, w- we are unacceptably exposed to these things. Like, we would not tolerate, like, on our computers, our human defenses against viruses. Like, we would not allow our computers to be as exposed to viruses as we allow ourselves to be. Okay? We're talking about, like, technological solutions, things in the background, right? And there's an entire edifice handling that, right? Including, like, detection, letting other nodes know all around the world, all this stuff. Like, like, but all happening in the background. That's what this should feel like if it's done properly. So whether you're worried about a lone actor or not, you don't even have to worry about that. Nature's gonna toss it out on us again. We should, we should have it ready for that.

   9. EGElad Gil25:09

      Yeah, I think you laid out, like, a really rational program around pandemic response. And I think most of the, the people lay out things that are, I feel, in some cases, actually subtractive. So I think your points on global monitoring makes a ton of sense. Your points on having rapid response vaccine generation makes a ton of sense. So I think those are, like, really smart, grounded approaches. It's kind of interesting because one of my big lessons from COVID when I looked at the biosafety levels that were actually enacted at some of these labs, and you're collecting large masses of bat viruses, right? And I remember when I used to work in a lab at MIT, I'd be working with different viruses and different agents and things like that for gene therapy purposes. And you look at the biosafety level and there'd be somebody in a hood and they'd kind of rub their shirt and they'd walk out the door.

   10. NANarrator25:48

       (laughs) Yeah.

   11. EGElad Gil25:48

       And, you know, one of the things that I almost get comfort in is I'm like, wow, there's been so few actual lab leaks over time relative to the poor behavior in labs themselves that it's really hard to actually have something jump into humans.

   12. JKJason Kelly26:01

       Well, I mean, look, look, Eli. I mean, to some degree, w- we are the evolutionary product of being able to defend against that stuff, right?

   13. NANarrator26:08

       (laughs)

   14. JKJason Kelly26:09

       Like, like, you know, like, if it was easy, we, like, we wouldn't have made it, right? Like, you know? Like, there's other species that, like, didn't get our imm- didn't get the immune system we got. And what you have in every organism on the planet is the integration of four billion years of incident solar radiation on the entire planet.

   15. EGElad Gil26:24

       They, they've been trained over evolutionary time, yeah.

   16. JKJason Kelly26:26

       A, a time that we have a really hard time comprehending how much energy that is because of the timescale.

   17. EGElad Gil26:31

       Yeah. One, one of the things that I think is really unique about Ginkgo is some of the decisions you've made as you've built a startup. So for example, I believe you have super voting shares for all the employees as, like, a public company. Could you tell us a little bit more about ideas like that that you've enacted and how you thought about them? It's, it's really cool stuff.

   18. JKJason Kelly26:46

       Yeah, we were kind of a weird bug, right? Like, we started out of grad school in 2008, right? So, like, straight out of school. Totally common in tech, not at all common in biotech. Okay, right? So we were, like, unfundable, couldn't raise money. It was 2008, and we weren't developing a drug. And so, like, biotech people don't back that. We were developing, like, a platform for programming cells, and the tech people wouldn't back it 'cause, like, what? Are you kidding me? A wet lab? And, and so we were, like, not fundable. We, we gr- five years of government grants. DARPA, ARPA-E, NSF, SBIR. So we were the first biotech to do YC 'cause Sam had just taken over from Paul, and he wanted to do nuclear, biotech, all these things, right? I do think, like, the entrepreneur energy is r- is common. So even if you're a hard tech company, I don't think there's, like, a different. You don't need a different entrepreneurial training than what, like, YC has perfected, you know? Right? Like, that, that's a good thing. So I think there, you're okay. But I think we were, like, finding out too late that things like social networks and large-scale tech platforms also have enormous real world consequences. But they were sort of getting a pass because they were in the world of bits. And people were like, "Yeah, bits." You know? Right? Like, whatever. It doesn't feel that dangerous, right? But once you're talking about a drug, you're, you know, you're putting a, a thing in a kid's body, you know? Right? Like, like, it's like a medicine. You know, there's just stuff that's like, you know, you can't fuck around with it, right? And, and so we're building a powerful platform at Ginkgo. So one of the questions was, who should control it? Like, how do you make these decisions about, like, who can use it? Like, platform ethic stuff. All the stuff that now is being talked about in AI 'cause AI is finally making people be, like, a little bit like, "Oh, bits." You know? Right? Like, maybe it is scary. Uh, and, and so, like, it, it's just a little more at the front for those of us that have already been hanging out in Adam's world, right? And this is why people, like, I think there's, like, a hard time between pharma and tech in terms of cultural, like, non-overlap because, like, the tech people feel like the, the therapeutics people are, like, losers and slow 'cause they're lame and not ambitious. But they're all, but these guys have clinical trials where people die, right? And, and so, like, it's being at the coalface of actually building things that really inflect on, on people and the world in a way that's not second order, like information technology, creates a different kind of culture, okay? And so who should control then a platform that in the, if we're right and is successful will powerfully read on people's lives, right? One answer is, like, the founders, right? That's Facebook, right? Like, Mark's got super voting and Finn, his kids do or some insanity. The other option is capital markets, right? You know? Like BlackRock, you know? Right? Fidelity. Like, like, that's just every normal corporation that isn't founder-controlled in Silicon Valley where, like, the, the voting shares are, are majority held by Arm's Length Capital and they... if the CEO's not doing the thing they want, they bring in a different board and fire the person. But, like, it, it comes down to the control is, is who has the share votes when this is in a public company ultimately hires and fires the CEO ultimately then sets the control of the platform, okay? And, and so at Ginkgo, we took this Silicon Valley idea of, of founder super voting shares and extended it to the entire employee base. So it's not just us. Anyone who's at the company, and it goes away if you leave, uh, gets 10X voting for their B shares versus the A shares, which have 1X voting. And so the way the math works is that the employees own more than 9.1% collectively. You multiply that by 10 and it outvotes the remainder. And so the theory was who should control the platform? Humans. Okay? Not, not divorced capital because that's not their priority. They're kind of like, "My job is to get a high return." At, at the end of the day, it's your job, company leadership, to, like, decide how to do it." But in reality what that means is company leadership primacy is the return. Okay? And, and so that seems eh. And then what we've decided is a persistent thing is the employees. Okay? The workers. Because they are humans and they actually work there and they go home to their families on Thanksgiving and have to explain, like, why they work at this company and are proud of it. And that, that may be long term-This is a theory. Uh, i- is a, is a good group to give governance to.

   19. EGElad Gil30:35

       Yeah. It's a r- it's a really cool approach, uh, and I think a lot of that early sort of super voting share stuff was pioneered in the media world, right? So the New York Times, the same family still controls it, you know, 100 years later because of super voting shares in the family.

   20. JKJason Kelly30:48

       And that's why it's the New York Times.

   21. EGElad Gil30:50

       Sure.

   22. JKJason Kelly30:50

       The only reason that place is the New York Times is because, like, uh, like, humans have control, not capital.

   23. EGElad Gil30:58

       So I guess one potential question about the model, 'cause I think it's a really smart and unique model, is sometimes CEOs have to do really unpopular things, and if you don't have the founder authority and you come in and you do something really unpopular, y- maybe you do a big riff and the people who are left are really upset about it, but you really have to do it for the business to survive. Or, you make tough choices that may be at odds with the employee base. How does that impact governance later, where, to some extent, you could argue the motivation if... You're not really answering to your board, you're answering to your employees purely. Uh, you may fall into more dynamics around popularity contests or trying to appease people around tough decisions to make. And f- as a founder, you have the moral authority to do those. As a hired gun CEO, I think it's much harder.

   24. JKJason Kelly31:41

       I agree

5. 31:42 – 37:00

   Discussion on AI, Evolution, and Architecture

   1. JKJason Kelly31:42

      completely. Yeah. I- I- I- I don't dispute any of that. So... And the answer is share voting, okay? So it's not, like, one person, one vote.

   2. EGElad Gil31:50

      Sure.

   3. JKJason Kelly31:51

      So how do you accumulate more shares in a company? Work there longer, okay? To build the value. Like, work there when it was cheaper and, and grow the value. So there will be a waiting. Again, it's all theory (laughs) . R- really, I d- I don't know. But, like, like, like, what I'm imagining-

   4. EGElad Gil32:05

      Yeah, yeah.

   5. JKJason Kelly32:05

      ... is you'll have, you know, employees wh- who own a lot and are like, "Yeah, that, that's a hard decision, but it's right for the organization." Right? A- and, like, I- I don't think that that's out o- out of school. I think you can see that happen, but we'll see. I don't know. We're trying it.

   6. EGElad Gil32:20

      Yeah. No, it's a very exciting experiment. That's cool.

   7. JKJason Kelly32:22

      But it originates from the platform gov- governance. That, that, that's, like, why we're actually doing it. I think some of these other reasons, I think, are interesting and I like them 'cause I'm, like, generally, I lean worker ownership. But I think the real point is, like, at the end of the day, someone has to have platform governance.

   8. SGSarah Guo32:34

      Jason, you've expressed, like, this awe around the result of humans. You know, in four, four billion years of evolution, we're incredibly energy efficient versus neural networks. Everybody knows that. You now have the very largest labs talking about spending literally a trillion dollars in compute over the next decade. If you think of that as maybe half energy, and then you have to make assumptions about energy prices, but now you're talking trillions of kilowatt hours, and maybe you're off by a magnitude, but, like, "Where's that money going to come from?" becomes a big question. These things have to get more efficient. If you compare that to, like, humans, like, maybe we spend 5,000 kilowatt hours before we learn to read, right? Do you think we get more biological inspiration in AI, or do you have any point of view on the intersection of this from an architecture perspective?

   9. JKJason Kelly33:19

      I- I- I- I don't have a great intuition, um, other than I think, you know, the other option is we just do giant brain in a vat and we throw that up against GPT-4. You know? Like, why are, why are we just limiting ourselves to a brain that fits in our head? Why don't we grow a room-sized brain and just go straight biological? Yeah. Have you thought about that, Eli?

   10. EGElad Gil33:39

       (laughs)

   11. JKJason Kelly33:39

       I mean, I think the... what's cool about, um, neural nets is that, like, brains basically allowed, uh, computer scientists to, like, escape their world of, like, logic. Like, back to the beginning of our conversation, it was an excuse for them to basically build a piece of software that was gonna, like... Well, they weren't gonna understand how it worked. There's probably a lot more things like that, because the community that builds software wants to understand it because that's the kind of people that historically have been good... been good at building software, right? So, like, open your minds, right? Like, like, I'm sure the neural net is not the best architecture, right? But, like, you know, and I know people are working on it, but, like, really go in different directions, right? Like, do something crazier.

   12. EGElad Gil34:15

       You're raising a really key point, which I think was back to part of the conversation around evolution is the driver for all sorts of optimizations that you don't expect, and if you are a rational person and you look at a biological system, right? There's the gene that, uh, can be coded, that you can, uh, produce RNA in either direction and it produces two different proteins, and why would you ever do that? And one of them is actually duplication of this thing that got repositioned for catalysis for this other thing, and so it's really messy, weird systems that evolved. And I think the second you have self-replicating systems where you have code writing its own code and you start going down that evolutionary path, you should have hyper-optimization for energetics and for all sorts of other things, because it's just gonna be part of the utility function that gets selected for by that system. And I think that's when you get out of the realm of, you know, handpicked design and logic and laying it down and just, like, an explosion of stuff, right? It's kind of... The Cambrian explosion happened for a reason, and that reason was evolution and resources, right?

   13. JKJason Kelly35:11

       Yeah. There's a very good book, um, if you're, if you like to nerd out on this particular line of stuff, uh, called The Plausibility of Life. Um, it's awesome. A- and I personally think instead of learning from brains, I- I agree with you, Eli, you wanna learn from evolution. Like, because evolution itself evolved things to become more evolvable, right? Like, the example they give in the book is really cool, I'll just say. It's, like, a skeletal system. So your skeletal system... Like, you can have a person you've seen l- that can have, like, a sixth finger. Like, that happens sometimes, right? Have you noticed it's not just, like, a bone jutting out of their hand? It's, like, wrapped in skin and nerves. Well, that's because your skin and nerves are adaptive to bones, and that, as you can imagine for exploring body plans, is a much more efficient way to explore the space of body plans. So if you jut out a bone, maybe it's gonna be better, but it's definitely not gonna be better if it's not wrapped in skin. Okay? A- and so, like, the... for wha- however it happened, there was, like, this layering in evolution where, like, we created... The, the system created skin and nerves to be adaptive, and the exploratory part of it is the bones. And again, I'll emphasize, four billion years is a long time, lot of energy that has been spent. I know it was a random walk, but, like, evolution figured a lot of cool stuff out, and I- I think that is, like, totally untapped.

   14. SGSarah Guo36:23

       For any of the scientists exploring alternative architectures out there, if you're, like, gonna do any sort of crazy mixture of exports routing, like, "Wrap the skin around the bones," is the advice we have from Jason.

   15. JKJason Kelly36:34

       (laughs)

   16. SGSarah Guo36:35

       Thank you so much for doing this. This was really fun.

   17. JKJason Kelly36:37

       Yeah, super fun. Thanks, Sarah. Thanks, Eli.

   18. SGSarah Guo36:38

       (instrumental music) Find us on Twitter @nopriorspod. Subscribe to our YouTube channel if you wanna see our faces. Follow the show on Apple Podcasts, Spotify, or wherever you listen. That way, you get a new episode every week. And sign up for emails or find transcripts for every episode at no-priors.com.

Episode duration: 37:00