Podcast Episode Summary
Main Topic: The Future of Biology and Biotechnology with George Church
Key Points:
Participants:
Takeaways:
Implications:
George Church is the godfather of modern synthetic biology and has been involved with basically every major biotech breakthrough in the last few decades.
Professor Church thinks that these improvements (e.g., orders of magnitude decrease in sequencing & synthesis costs, precise gene editing tools like CRISPR, AlphaFold-type AIs, & the ability to conduct massively parallel multiplex experiments) have put us on the verge of some massive payoffs: de-aging, de-extinction, biobots that combine the best of human and natural engineering, and (unfortunately) weaponized mirror life.
Watch on YouTube; listen on Apple Podcasts or Spotify.
Sponsors
* WorkOS Radar ensures your product is ready for AI agents. Radar is an anti-fraud solution that categorizes different types of automated traffic, blocking harmful bots while allowing helpful agents. Future-proof your roadmap today at workos.com/radar.
* Scale is building the infrastructure for smarter, safer AI. In addition to their Data Foundry, they recently released Scale Evaluation, a tool that diagnoses model limitations. Learn how Scale can help you push the frontier at scale.com/dwarkesh.
* Gemini 2.5 Pro was invaluable during our prep for this episode: it perfectly explained complex biology and helped us understand the most important papers. Gemini’s recently improved structure and style also made using it surprisingly enjoyable. Start building with it today at https://aistudio.google.com
To sponsor a future episode, visit dwarkesh.com/advertise.
Timestamps
(0:00:00) – Aging solved by 2050
(0:07:37) – Finding the master switch for any trait
(0:19:50) – Weaponized mirror life
(0:30:40) – Why hasn’t sequencing/synthesis led to biotech revolution?
(0:50:26) – Impact of AGI on biology research progress
(1:00:35) – Biobots that use the best of biological and human engineering
(1:05:09) – Odds of life in universe
(1:09:57) – Is DNA the ultimate data storage?
(1:13:55) – Curing rare diseases with genetic counseling
(1:22:23) – NIH & NSF budget cuts
(1:25:26) – How one lab spawned 100 biotech companies
Dwarkesh Patel
Today, I have the pleasure of interviewing George Church. I don't know how to introduce you. Honestly, this is not even a exaggeration. It would honestly be easier to list out the major breakthroughs in biology over the last few decades that you haven't been involved in, from the Human Genome Project to CRISPR, age reversal to deinstinction. So you weren't exactly an easy prep. Sorry. OK, so let's start here. By what year would it be the case that if you make it to that year, Technology in bio will keep progressing to such an extent that your lifespan will increase by a year, every year, or more.
George Church
Escape velocity is sometimes what it's called for aging. Different people have estimates, and all those estimates, including mine, are going to take with a big grain of salt. I think that mainly looking at the exponentials in biotechnology and the progress that's been made in understanding, not just understanding the causes of aging, but seeing real examples where you can reverse subsets of the aging phenotype you know so you're getting close to all of aging um in other words you're seeing instead of just saying oh i'm going to fix the damage in this collagen in this tendon in this limb you're saying oh i'm going to change a lot of things that that are that are common to age-related diseases and i'm going to get more than one at a time I think looking at those two phenomena, the exponentials in biotechnologies and the breakthrough in general aging, not just analysis, but synthesis and therapies, and a lot of these therapies are now making in the clinical trials, I wouldn't be surprised if 2050 would be a point.
if we can make it to that point, 25 years. Most people listening to this have a good chance of making it 25 years. And the thing is, it's not going to be some sudden point where you're going to be, you know, so sick 25 years from now that it's like hit or miss. It's more likely that you're going to be healthier 25 years from now than you thought you were going to be. There may be some, probably not some law of physics, but some economic or complexity issue that we don't know about that becomes a brick wall. I doubt it seriously, but we'll have to see. Given the number of things you would have to solve to give us a lifespan of humpback whales.
Bowhead whales. 200 years, yeah.
Dwarkesh Patel
Is there any hope for doing that from somatic gene therapy alone, or would that have to be germline gene therapy?
George Church
Probably there's a lot of forces pushing it towards somatic. For one, there's 8 billion people that have missed the germline opportunity. That's to say, doesn't apply to us, the two of us and everybody listening to this. And, you know... You have to be very cautious when you say something's impossible. It's safe to say it's impossible to do it this second, but you don't know what's going to happen tomorrow in the next decade or something. So I think there's a lot that could be done. In particular, since aging is a fairly cellular phenomenon with... proteins going through the blood and other factors going through the blood, the signaling and so forth.
You could imagine if you replaced, let's say, every cell in the body, every nucleus in the body, you know, it would suddenly be young again, right? Without... going all the way back to the embryo and forward again. And there's various other things that are just short of that. If you replace the cells, well, they'll fit into that niche. They might displace the old cells. You know, that's... within, certainly within the realm of modern synthetic biology is for cells to take over niches. I think the hardest part is brain, but even there, you know, there's some evidence that if you bring, even though the brain doesn't really use stem cells that much, you could artificially bring in stem cells and they could artificially fit into a circuit and learn the circuit and then displace the old ones in some way.
Dwarkesh Patel
Ship of Theseus kind of thing?
George Church
Yeah, exactly. Ship of Theseus. Having, you know, trying to maintain the connections and the memories. But, you know, there's some fairly straightforward experiments that need to be done before we can really even estimate how hard that problem is. Or, you know, very often there's low-hanging fruit that people just... think is improbable, but it's there because biology has all these gifts that, you know, just hands over to us levers that we can flip, like vaccines is an amazing gift. It didn't have to exist, but they do. Yeah. Is there an existing gene delivery mechanism which could deliver gene therapy to
Dwarkesh Patel
every single cell in the body? There is nothing close to that today, but there's no law of physics that would prevent it.
George Church
There's going to be practical considerations like You know, how many injections do you need to do to achieve that goal? But we're getting better at targeting tissues. You know, so for one of my companies, Dino Therapeutics, showed they could get a hundredfold improvement in targeting neurons in the brain, which is a big deal. Now, if... And that was just one little campaign that they did, you know, one experiment. It involves a lot of AI and a lot of testing of millions of different capsids. If you did that with cells, capsids are fairly limited in the diversity and the structure that it can change to, but cells have even more possibilities, I think you could probably get delivery to everything.
And the question is, how close to 100% do you need to get? And it's gonna vary from tissue to tissue. For example, for some therapies, you just need to get because that 1% can produce some missing enzyme, and the 1% doesn't have to necessarily be in its normal place, right? You know, you can turn a muscle into part of the immune system temporarily for a vaccine. You can, you know, An enzyme that's normally made in, let's say, the brain, you could make the liver, right? If the point is just to get it into the blood. So I think that's moving along quite well.
Dwarkesh Patel
You're one of the co-founders of Colossus, which recently announced that they de-extincted a dire wolf, and now you're working on the woolly mammoth. Yeah. Do you really think we're going to bring back like a woolly mammoth? Because the difference between an elephant and a woolly mammoth might be like a million base pairs. So how do you think about the kind of thing we're actually bringing back? Well, so I think... I think people get worked up about whether we are trying to bring back or have
George Church
already or will ever bring back a new species. And I think of it, if you think of it rather than as a natural thing that we're trying to do, but as a Synthetic biology with goals that have potential societal, and people also get worked up as to whether this could possibly benefit society in any way. You know, can we really, you know, fix an environment to suit humans or fix the global carbon to suit humans? And the answer is we don't know, but it's worth a try, isn't it? Because it could be very cost effective. And the other thing, the other aspect of it is there's a whole discipline within synthetic biology of asking what's the minimum, right?
And so people often phrase it into what's the maximum, you know, like what can we do? And I'm interested in both. But, you know, it's like, oh, yes, there's millions of difference between mammoths and elephants. There are millions of difference between elephant one and elephant two within Asian elephants and between Asians and African. what we would normally call them, you know, how we would normally classify them, what their functionality would be in an ecosystem, right? And so there's this exercise that people do, and we've done it, for example, with developmental biology. What's the minimum number of transfusion factors it takes to make a neuron from a pluripotent stem cell,
Dwarkesh Patel
right? What's the minimum number of base spheres it takes to make something that will
George Church
replicate to something that was done in mycoplasma originally? And in a way, these are more interesting than can we make a perfect copy of something, right? It's can we make... What's the minimum things we have to do to make it completely functionally... or even functionally in a particular category, right? How do we make it bigger? We learn the rules for how to make things bigger, how to make things replicate faster, how to use new materials, et cetera. So I think with the dire wolf, we clearly didn't make an exact copy of the dire wolf, but it helped illustrate kind of educated people around the world that what is the difference between a gray wolf and a dire wolf, right?
Because, you know, dire wolves, they're big. Maybe they have a particular coloration. You know, the head components tend to be bigger than the leg components. And so how many genes do you need to do that? Maybe this was dire wolf, you know, 2.0, and we're going to go for... go for 3.0 in successive approximation. And we might want to develop the technology for making exact copy of something, because then we can, especially being able to make 100 variations on an exact copy, because then there won't be any argument about whether you could make a dire wolf. It's a matter of what should you make and what would be most beneficial for the species that you're making, for the environment it lives in, and for humans.
Dwarkesh Patel
Does this teach us something interesting about phenotypes, which you think are downstream for many genes, are in fact modifiable by very few changes? Basically, could we do this to other species or to other things you might care about, like intelligence, where you might think like, oh, there must be thousands of genes that are relevant, but there's like 20 edits you need to make really to be in a totally different ballgame.
George Church
Yeah, I think you're hitting on a very interesting question, and it's related to, you know, what's the minimum? So, for example, you almost said it, which was, you know, take a very multigenic trait in humans, like height is something that's probably the most well-studied one, simply because no matter what genius, no matter what medical condition you're studying you collect information on height and weight and things like that anyway they tracked it down to on the order of 10 000 genes of which we have 20 000 protein coding genes some of them are rna coding genes And they each have a tiny influence on height.
But if you take growth hormone, somatotropin, that you have extreme examples where you'll get extremely low small stature and extremely high stature due to that one alone. And in fact, it's used clinically as well for seven different treatments. medical treatments. So that's a perfect example of how much we can minimize something, sometimes called reductionism. Reductionism isn't all bad. Sometimes it helps us bring a product into medicine. Sometimes it helps us understand or build a tool chest or a a module that we can use in other cases and translate it to other species. So you hit on it just right, is that not everything will translate, but we start accumulating these widgets.
It's kind of like all the electronic widgets that we accumulate over time, that if you just want to slap it into the next circuit, you might be able to.
Dwarkesh Patel
What implications does this have for gene therapy in general? Like what is preventing us from finding the latent knob for like every single phenotype we might care about in terms of helping with disabilities or enhancement? Is it the case that for any phenotype we care about, there will be one thing that is like HGH for height? And how do we find it?
George Church
Biology, we've got a real gift, which is it's both very much more complicated than almost anything we've designed from scratch. But it also is a lot more forgiving in a certain sense. You can have an animal or even a human that has two heads, which is not something that they evolutionarily do. there was not evolutionary selection specifically to have two heads, but just a little deviation from the normal developmental pattern during, you know, fetal development. And they both function fine. They control subsets of the body. And, you know, they have their own personality, their own life. So there's all kinds of things you can do in biology that where you're working at a very high programming level is a way of thinking about it.
Pushing us to a new level of intelligence is gonna be very challenging. And maybe not even urgent, okay? To some extent, actualizing the people that we currently have would be quite, you know, just getting them all up to whatever speed they want to be up to within the range that's been demonstrated. So, like, some people are going to want to be like Einstein. Some people won't. Some people want to be healthy all the time. Unlikely, but some people might not. Some people might want to live to 150. Some people might want to die at 80. But if you give them that range, that capability, you know, what if we had 8 billion super healthy people Don't need to worry about food and drugs.
Super healthy. Einstein level of intelligence. Education level, best we can come up with. That would be a completely different world.
Dwarkesh Patel
But just getting everybody to the healthy level. How much gene therapy would that take? It sounds like it wouldn't take that much if you think that there are these couple of knobs which control very high-level functions. So do you find them through the GWAS genome-wide dissociation studies? Is it through simulations of these?
George Church
I would say mostly GWAS for humans. maybe for animals in general, followed for animals with synthetic biology, you know. And the smaller and the cheaper and faster replicating, the more experiments you can do. So, you know, I don't want to overemphasize how single genes can do these amazing things, but there's also the possibility that multiple genes can be hypothesized and tested quickly. So for example, I mentioned earlier, what's the minimum number of transcription factors it takes to turn a stem cell into a neuron? Well, there's a bunch of recipes where you can do it with one. Maybe you want a specific neuron, you might need a few more.
But then you can kind of quickly go to the answer by looking at each target cell type that exists, and you can see what transcription factors exist. did it use to get, did it express at the time that it's the target? And then you say, well, let's just try those on the stem cells and see if they work. And that recipe has worked quite well. It's the basis of GC Therapeutics company and a bunch of the work that we do is you can almost, you can get a recipe for almost every cell type in the body. Now that's not new cell types, but at least you can, you've learned to, To your point about reducing the number of genes we need to manipulate in order to get to a particular goal, here's a whole series of goals, and we can get them with one, two, three, maybe seven change transcription factors.
So that's an example. Yeah. There's room for lots of other examples of where you can do a reduction and do not just reductionistic virality, but then constructionistic, where you take it back up and make a whole complex system and see what happens. And then you can do lots of those combinations and you debug them and so forth. Some of these things you can do in vitro things you can do on probably on the order of 10 to the... 14th, 10th, 17th. Things that involve cells are typically in the billions. But we have this, this is how we're going to get inroads into the very complicated biological systems.
Dwarkesh Patel
Most antifraud solutions focus on detecting and blocking bots. And that's fine if your product is just meant to be used by humans. But what if you actually want your product to be used by AI agents? How do you distinguish between automated traffic that you want to allow and automated traffic that you need to block? WorkOS Radar is built for this exact problem. Radar is a powerful, humans-only filter that major companies like Cursor rely on to protect their product from bots. But it can also handle more complex and granular tasks, like distinguishing desired versus malicious agents, rather than just blocking all automated traffic.
Even if you aren't building for agents yet, Radar helps you future-proof your product roadmap. If you start using radar for traditional fraud protection, when you do ship your first feature intended for agents, you can just update radar's behavior with no engineering required. Learn more about AI native fraud prevention at workos.com slash radar. All right, back to George. Can I ask you some questions about biodefense? Yeah. Because some of the stuff you guys work on or, you know, quite responsibly choose not to work on, can keep one up at night. Mirror life. Yes. Given the fact that it's like physically possible, Why doesn't it just happen at some point?
Like some days it'll get cheap enough or some people care about it enough that somebody just does it. What's the equilibrium here?
George Church
Right. You know, I was a co-author on a paper that warned about the dangers of mirror life, just like, you know, I wrote a paper long ago about the dangers of having the synthetic capabilities we have for making synthetic viruses. And to some extent of having new genetic codes, they have a few things in common. But the thing about the advance that we were recognizing in our science paper that was warning about mirror life is that we not only had to calculate what the possibility of error prone, you know, escape or something like that. We don't want anything to escape. that we made in the lab, unless there's a general societal consensus, it's a good thing.
And so far, there aren't too many examples of that. But, aren't any examples of that. But mirror life, if it can be weaponized, then we took it to a whole nother level of concern. And the concern was that if we got it to a certain point, it would be easy to weaponize it. And again, there's practical considerations. It may be that most people who would consider weaponizing mirror life would probably be satisfied with weaponizing viruses that already exist, that are already pathogens. And they wouldn't want to destroy themselves and their family and their legacy and everything like that. But all it takes is one, you know, one group probably or one person together.
But your question is, is it inevitable? I don't know. It might be. It's quite possible it's already here. In other words, we already have mirror life in our... in our solar system or maybe even on our planet, it just hasn't been weaponized, right? And so it's just like what we were saying in the science paper is this seems like the sort of thing that could wipe out all competing life if we're properly weaponized. But there are probably a few things like that. And what we really need to do is reduce the motivation to do that, maybe increase our preparedness for a variety of existential threats some of which will be natural some of which will be you know one disgruntled person who has essentially too much power because you know over history of humanity, the amount of things that a single person can do has grown very significantly.
I mean, it used to be when you had your bare hands, there's kind of a limit to what one person could do. A large number of people could team up and get, let's say, a mammoth or something like that. But, you know, but today, one person with the right connections or right access to technology, you know, could... could blow up a city, right? And that's a huge increase in capability. And I think we want to start dialing that back a little bit somehow. And what does that look like in terms of not just mirror life, but synthetic biology in general?
Dwarkesh Patel
You know, maybe we're at an elevated period of the ratio to offense and defense, but how do we get to an end state where even if there's lots of people running around with bad motivations, that somehow there's defenses built up that we would still survive, that we're robust against that kind of thing? Or is such an equilibrium possible, or will offense always be privileged in this game?
George Church
Offense does have an advantage, but so far we haven't. We made it through the Cold War without blowing up any hydrogen bombs, as far as I know, accidentally or intentionally on enemies. We did two atomic bombs. But a lot of that is based on the difficulty of building hydrogen or atomic bombs. The thing that's alarming to people like me is that biotechnology enables smaller and smaller efforts, harder and harder to detect, harder and more and more subtle to the stochastic variation between people. There's some people that are just so happy, they would never wanna do anything close to that, or they're so responsible or ethical or whatever.
And then there are other people who, like, whenever they have a bad day, they want to take a lot of people with them, right? And, you know, maybe some progress in psychiatric medicine would help. Again, you don't want to force that on people. You want to make sure that if they don't want to get cured, you can't force them, but you can make it available to them. That might help. Hopefully there's a more technological solution or more robust solution than... Well, there will be technological solutions to the psychiatric problem. It could be even people who aren't sure whether they want to be helped or not can try it out and it's reversible.
And they say, yes, I like that better. OK, let's try that. then then there's other things that cause uh you to to have bad days it's not just your psyche it's also the environment so if you if you're surrounded by your people you know being starved or infectious disease or being shot at or something like that those are things that are subject to sociological and technological solutions. And if we could really solve a lot of that stuff, we could reduce the probability that one person- I mean,
Dwarkesh Patel
this is making me pessimistic, because you're basically saying we've got to solve all of society's problems before we don't have to worry about synthetic biology. Yeah. Which I'm like, I'm not that optimistic about, like, what's on some of them.
George Church
I'm not trying to reassure you. And, you know, we're having a conversation about what it takes, and that might be as one scenario for what it might take. You had an interesting scheme for remapping the codons in a genome so that it's
Dwarkesh Patel
impervious to naturally evolved viruses. Is there a way in which this scheme would also work against synthetically manufactured viruses?
George Church
Much harder. Again, the offense has the advantage. We can make a lot of different codes.
Dwarkesh Patel
Which would limit the transmissibility?
George Church
Yeah, so one interesting thing is that there's only two chiralities. There's the current chirality and the mirror chirality. But there's maybe 10 to the 80th difference codes. Now, some of them you might be able to take out all at once. Anyway, the coding space is a kind of more interesting space. And of course, it could get even more complicated than that because the 10 to the 83rd is like, based on triplet codons and and that sort of thing but if they're quadruplet codons or new novel alphabets and so on um but we're sort of getting into uh you know, a cycle of competition. It would be better than nip it in the bud, which is, you know, why did we spend so much societal resources building up to tens of thousands of nuclear warheads? And now we've... dialed it back to mere thousand nuclear warheads. That's nice that we dialed it back, but why did we waste all that time and money and energy?
Dwarkesh Patel
Biology seems very dual use, right? So the mere fact that you, like literally you, are making sequencing cheaper will just have this dual use effect in a way that's not necessarily true for nuclear weapons. Right. And we want that, right? We want biotechnology.
George Church
It's hard to pound nuclear weapons into plowshares, as they say.
Dwarkesh Patel
I guess I am curious if there is some long-run vision where... To give another example, in cybersecurity, as time has gone on, I think our systems are more secure today than they were in the past because we found vulnerabilities and we've come up with new and Christian schemes and so forth. Is there such a plausible vision in biology or are we just stuck in a world where offense will be privileged and so we would just have to limit access to these tools and have better monitoring, but there's not a more robust solution?
George Church
You know, one of the things I advocated in 2004 is that we stop deluding ourselves into thinking that moratorium and voluntary sign-ups to be good citizens is going to be sufficient. We need to also have surveillance and consequences and mechanisms for whistleblowers to make it easy for people to report things that they think are out of line. And we had essentially moratoria and disapproval for germline editing. Nevertheless, somebody did it. And a lot of people knew about it. So that was clearly a failure of the whole moratorium, voluntary, and whistleblower components.
Dwarkesh Patel
I mean, it worked for five years with only one defector. That's quite impressive.
George Church
Okay. Half empty, half full. I'll give you that. But... But all it takes is one for some of these scenarios, right? And that's, so it would have been nice if the whistleblowers could have saved him the three years in prison by getting an intervention. I mean, it's not like anybody died, right? There are probably three healthy genetically engineered children in the world now, be teenagers soon. But still, it was a good test run, shows a failure of the system. We need to have better surveillance of all the things we don't want and consequences that are well known.
Dwarkesh Patel
Over the last couple of decades, we've had a million-fold decrease in the cost of sequencing DNA, a thousand-fold in synthesis. We have gene editing tools at CRISPR, massive parallel experiments through multiplex techniques that have come about. And of course, much of this work has been led by your lab. Despite all of this, why is it the case that we don't have some huge industrial revolution, some huge burst of new drugs or some cures for Alzheimer's and cancer that have already come about? When you look at other trends in other fields, right? Like we have Moore's Law and here's my iPhone. Why don't we have something like that in biology yet?
George Church
Yes, so we have something that's about the same speed, a little bit faster than Moore's Law in biology. It's more recent is one aspect of it. But we could kind of stand on the shoulders of the electronics giants to go a little bit faster to catch up. I would say we do. I mean, we have the biotech industry, which has... use that exponential curve to get better. It's also possible we're close to the big payoff, is the other aspect, or the beginning of the big payoff. You know, right now we have miraculous things like cures for rare diseases, we have vaccines, we have a trillion dollars probably of various biotech-related things, if you go far enough apart. But we're kind of on the verge of really combining electronics and biology more thoroughly and AI and biotech. And I think that's, it seems like we're on the same track as Moore's Law, if not better.
Dwarkesh Patel
What exactly are we on the verge of? What does 2040 look like?
George Church
Well, 2040, we're talking about only 15 years, which is like one and a half, maybe two cycles of FDA approval. 2040 is post-AGI. It's a long time. Well, I hope it's not post-AGI. I think we're rushing a little bit to get to AGI, and there's lots of cool things we can do with just super AI. But we need to be very cautious, I think, that AGI. Well, anyway, we could get into that.
Dwarkesh Patel
I have a question for you there, yeah.
George Church
You know, I think that we are shortening the time of getting medical products approved still in a safe way. So I think, but that's not gonna completely change the exponential. It might reduce it from 10 years down to one year is our record so far for, say, COVID vaccines. So maybe that'll be 10 times shorter. Maybe that will multiply out a little bit. But I think the big thing is that all our designs will become better, so there'll be fewer failures. The cost per drug will drop. There'll be things that we didn't classically consider drugs or instruments, kind of some sort of hybrid thing. But again, I don't think that'll be completely shocking. But it's just going to be so much of it. It's going to be lots of diversity of solutions.
Dwarkesh Patel
How much more are we talking? Are we going to have 10x the amount of drugs, 100x? I'm not even sure it's going to make sense, but yeah, 100X would not be completely surprising.
George Church
Combinations of drugs will be important. Using them intelligently, there'll be a lot more. Some drugs will affect everything. So, for example, an age-related drug, that could impact every disease. I'm not sure the number is going to matter so much as the quality and the impact and intersection and software that helps physicians and regular citizens make decisions.
Dwarkesh Patel
And what specifically is changing that's enabling this? Is it just existing cost curves continuing or is it some new... technique or tool that will come about?
George Church
Well, the cost curves are affected by new tools. I mean, it's not just some automatic thing. There is a big discontinuity between Sanger sequencing and nanopores and fluorescent next-gen sequencing. And so, you know, I think sometimes it's a merger of two things. So clearly AI merging with protein design caused a step function. These step functions get smoothed out into kind of a smooth exponential, but there are lots of them. The next set will probably be a merger of AI with other aspects of biology, like developmental biology, merger of developmental biology with manufacturing. Conquering developmental biology, in other words, actually knowing how to make any arbitrary shape given DNA as the programming material.
I think that would be a big thing. Having just more materials in general. All the materials we use in mechanical and electrical engineering should be made better by biotechnologies. Why is that? Why is that? Well, it's that electronics is, you know, Moore's law, I wouldn't say is stopping, but it's kind of And what we would call the one nanometer process, which is supposed to come out in 2027, according to the roadmap, is not really one nanometer. It's more like 40 nanometers center-to-center spacing, typically in two dimensions, maybe a little bit of three dimensions. But biology is already at 0.
4 nanometer resolution, and it is in three dimensions. depending on how you count that third dimension, that could be a billion times higher density that biology is already at. And we just need a little more practice with dealing with the whole periodic table. Even electrical engineering doesn't use the whole periodic table typically, but especially not at the atomic level. So I think biology is just really good at doing atomic precision.
Dwarkesh Patel
So then what's the reason that over the last many decades, and we do have not atomic, but close to atomic level manufacturing with semiconductors. 40 nanometers. Right. It's quite small. It's a thousand times bigger than biology, linearly. But the progress we have made hasn't been related to biology so far. It seems like we've made Moore's Law happen... I don't know, people in the 90s were saying, you know, ultimately we'll have these biomachines that are doing the computing. But it seems like we've just been using conventional manufacturing processes. What exactly is it that changes that allows us to use bio to make these things?
George Church
There are a few things. One is the arrival of synthetic biology, where you sort of... We were already kind of doing synthetic biology before. You know, we were doing recombinant DNA was kind of... You know, genetic engineering was called. It was kind of in that direction. But synthetic biology really liberated us to think a little bit bigger. Even though it started kind of focused on E. coli and yeast, it enabled us to maybe think about new amino acids, for example. And I think new amino acids, if you start using the full periodic table with the amino acids or what amino acids can catalyze, that breaks one of the major barriers.
One of the major barriers between electrical, mechanical engineering and biology was the use of special materials, things that conduct electricity at the speed of light, or conduct signals more generally. But there's definitely polymers that biology can make that will conduct at the speed of light. And we could make a mixed neuronal system that has conventional neurons and processes that conduct at the speed of light. That would be interesting. I think that our ability to design proteins was particularly difficult. Designing nucleic acids was great, whether we were doing, you know, you want two things to bind to each other.
You just dial it up using Watson-Crick rules. If you want to make a three-dimensional structure, you know, it's actually the one, kind of the one thing where morphology is dictated by fairly simple rules. It's not how developmental biology works, and we still need to figure out how that works. But DNA origami, DNA nanostructures really work. but doing it for proteins was really really hard until i don't know right maybe eight years ago something like that and and i think we're just now getting used to it the use of chips for making dna um i mean you said that dna synthesis come down a thousand fold what depends on who you talk to so that when we came out with the first chip based genes in 2004 nature paper uh basically people dismissed it for about a decade the only people that used it were collaborators and alumni.
And it wasn't even listed on the Moore's Law curve for DNA synthesis, even though it was like a thousand times cheaper. It was just like ignored. And now we have claims of 10 to the 17th You can make libraries 10th to 17th that aren't randomized in the usual sense where you just use air-prone PCRs, spiked-in nucleotides. At 10th to 17th, that's a lot bigger than a thousandfold, if it turns out to be practical.
Dwarkesh Patel
Publicly available data is running out, so major AI labs partner with SCALE to push the boundaries of what's possible. Through SCALE's Data Foundry, major labs get access to high-quality data to fuel post-training, including advanced reasoning capabilities. SCALE's research team, SEAL, is creating the foundations for integrating advanced AI into society through practical AI safety frameworks and public leaderboards around safety and alignment. Their latest leaderboards include Humanities Last Exam, Enigma Eval, Multi-Challenge, and Vista, which test a range of capabilities from expert-level reasoning to multimodal puzzle solving to performance on multi-turn conversations.
Scale also just released Scale Evaluation, which helps diagnose model limitations. Leading frontier model developers rely on scale evaluation to improve the reasoning capabilities of their best models. If you're an AI researcher or engineer, and you want to learn more about how SCALE's Data Foundry and Research Lab can help you go beyond the current frontier of capabilities, go to scale.com slash thwarkash. Okay, so speaking of protein design, Another thing you could have thought in the 90s is when people were writing about nanotechnology, Eric Drexler, and so forth. And now we can go from a function that we want this tiny molecular machine to do back to the sequence that can give it that function.
Why isn't this resulting in some nanotech revolution? Or will it eventually? Like, why didn't Alpha 4 cause that?
George Church
I think part of it is that the nanotechnology as original, you know, the kind of the source of the inspiration, Eric Drexler, he wanted to reinvent biology in a certain sense, but it already existed. And so you don't need to design a diamond replicator because you already have a DNA replicator. And so the question of what was missing, what was motivating this reinvention of biology, it was materials. So the biology is not that great with you know, materials that are, say, superconductors or conductors periods, semiconductors and light speed. But it's getting there. I mean, you know, rather than going the route of having everything has to be based on first principle nanostructures, you can meet in the middle where biology can build things.
Now, of course, when you go down to liquid nitrogen in colder temperatures, biology, as we currently know it, stops functioning. Now, it's not to say that you can't have things moving in liquid nitrogen. You can, but that hasn't been explored and doesn't really need to be. Because if biology can build things that can operate at low temperature, or maybe biology now because you can make these big libraries of biology, maybe 10th of the 17th in vitro, and you can flip through them quickly and you can barcode them. This is something that's never been done in electronics. I'm not saying you can't do it in electronics, but you haven't made, you know, a billion different kinds of electronic materials, right?
Just in an afternoon, barcode them all and see who wins, right? But we do that all the time in biology now, at least since 2004 we have. And so... So I think that's an opportunity, is that we use those libraries to make much superior materials. And we might even finally get a room temperature superconductor that way. From bio? It's possible. I mean, from libraries. We call them chemical slash biochemical slash exotic material libraries. But the point is they're libraries. They're essentially based, in some sense, on polymers, even though pieces of them don't necessarily have to be polymers. Do you have a prediction by when we'll see this material science revolution?
Dwarkesh Patel
What is basically standing between? Because we've got AlphaFold right now, right? So, what is the thing that we need? Do we need more data?
George Church
Well, AlphaFold's very nice, but it's only part of it. So, there are large language models that are different from AlphaFold. So, give an example. Last time I checked, anyway, at least it's all changed. If you substitute an alanine for a serine in a serine protease, it will have exactly the right fold. It will be precise to a fraction of angstrom overall average. but it won't function. It just won't function. And that's where you need either extraordinary precision or just knowledge of what happens evolutionarily or happens in experiments to say that no, an alanine won't work, okay? And so I think There's all kinds of combinations of AI tools that can give you deeper insight into that. If alpha-fold predicting the structure doesn't tell you whether the thing will
Dwarkesh Patel
actually function, then what is needed before I can say, I want a nanomachine that does X thing, or I want a material that does Y thing, and I can just get that?
George Church
I mean, I think the way that it's working now, which will get us a long way, won't get us the whole way, is we have something that kind of works, and we make libraries inspired by that, make variations on it. And then whichever of those variations work, we make variations on that, and we can just keep going. It's kind of like the way evolution worked, except... Now we can do it at incredibly high speeds. In principle, evolution might incorporate a few base pair changes in a million years. Now we can make billions of changes in an afternoon. And it's all guided in such a way that you get rid of the wastefulness of having a bunch of neutral mutations and a bunch of lethal mutations.
You can have things that are quasi-neutral, but likely to be game-changing, have more of a focus on those. Another thing that's been missing, and none of the ai protein design tools that i know of are particularly good at it yet but we're trying to we're as we speak trying to improve this is non-standard amino acids because a lot of these tools depend on having libraries of 3d structures which use 20 amino acids and large language models that where you line up all the sequences of 20 amino acids and we have very little experience with extra ones but But I think there's a revolution going on in generating non-standard amino acids where the amino acids can either have as covalent part of them or as easily liganded all the entire periodic table, stable elements.
And that will, you know, each of those will have to blend in and train our models on. But as soon as that comes in, then we're going to have a whole series of new materials very quickly. And ultimately, you can think of the... The determination of the functionality of your library is a kind of computer, right? So you use AI to design the library optimally, so you avoid things that are really neutral and really seriously damaged. But then the stuff in the middle, you actually play it out, not in a simulation, but in real life. But it's so inexpensive, and it's so fast, and it's so exact. I mean, it's 100% precision because you're not simulating.
You're not making assumptions. You're not going from quantum electrodynamics, which is an assumption, quantum mechanics, which is an assumption, to molecular mechanics, which are full of assumptions, you're really doing the real thing. You're doing a kind of natural computing, and then you can take that data and harvest it in various ways very efficiently, pump it back into the more conventional AI, and do another round of it.
Dwarkesh Patel
If I listen to these words, it seems like I should be expecting the world to physically look a lot different. But then why are we only getting like a couple more drugs by 2040?
George Church
Well, I didn't mean to stop there. I mean, I knew the conversation would continue. Right. I'm not pinning down a particular year either. But I think this is poised to go pretty quickly. There are very few practitioners is the thing that will stop it for a while. Yeah. Since materials actually should go faster, though, because they don't require quite as much regulatory approval. So it's one of these things where when you get the right idea, it's not hard to recruit people. I mean, for example, when Feng Zhang and my labs brought out CRISPR, we just got 10,000 requests in the next two months for people that wanted to duplicate the system. And so that's what I hope will happen with the non-stained amino acids and the using AI for protein design and making new materials. So hopefully that will recruit tens of thousands of people overnight, yeah. Are you more excited about AI which thinks in protein space or the capsid space or like just
Dwarkesh Patel
It's like predicting some biological or DNA sequences. Or are you more optimistic about just LLMs trained on language, which can write in English and tell you, here's the experiment you should run in English? Which of those two approaches, or is there some combination that when you think about AI and bio is more promising?
George Church
I'm much more excited about scientific AI than about language AI. I think languages were in pretty good shape already. And what worries me is that to get to the next level of language requires AGI or ASI, you know, artificial superintelligence. And that's very dangerous. I don't think we have quite figured out how to And there's a lot of safety organizations and a lot of safety rules and so forth. And I think what typically happens when there's an intense competition is those safety rules get undermined and pushed aside. But even if they weren't, I just don't think we understand our own ethics well enough to educate people.
A completely foreign type of intelligence. I mean, we barely know how to pass it on to the next generation of humans. So, I think we need time to sort that out. And there's no rush. This is a completely artificial emergency. This is not like COVID-19 where we actually, millions of people are dying if we delayed the science. This is something where if there ever is a crisis, it's because we created it. It's not because we're trying to solve it. And so I think we need to go very slowly on AGI and ASI and double down on slightly narrower scientific goals. And even that, we need to be very cautious about.
We need to have kind of an international consensus on what constitutes safe AI.
Dwarkesh Patel
I suppose we did build safe superintelligence. How much would that speed up bioprogress? Is it just like, there's a million George churches in data centers, just like thinking all the time. Is it a 10x speed up?
George Church
I think it would slow it down. I think it would eliminate it because the first thing it would conclude is biology is not relevant to me because I'm not made out of biology.
Dwarkesh Patel
I mean, suppose you get them to care about it. There's a million copies of you in a data center. How much faster is bioprogress? But they can't run experiments directly. They're just in data centers. They can just say stuff and think stuff. I don't think we have anything close to the assurance that we need that that would be safe.
George Church
But let's put safety aside for the moment. I think it's not only hard to calculate the bads, it's hard to calculate the goods. So I think it could be a complete game changer. But on the other hand, It's like if we said we could get instantaneous transport all over the Earth, right? Well, we could say, yes, that could be a game changer, but do we really need it, right? Is that really important? Maybe it would be more interesting to just have Zoom calls that are better, you know, or just learn how to get everything we want in our kitchen and we don't need to travel anymore, right? Yeah. So, you know, be careful what you ask for, right? You know, because you could tip our priorities towards something that we really don't care about, that we shouldn't care about or might wish we didn't care about, right? I'm curious what, you've still got to run the experiments, you still need these other things.
Dwarkesh Patel
So does that bottleneck the impact of the millionth copy of you, or do you still get some speed up? How much faster can biology basically go if they're just like more smart people thinking, which is a sort of proxy for what AI might do?
George Church
These are great questions, and I'm not sure, I don't want to misrepresent that I know the answers, but it's like the question of if you have nine women, can you do pregnancy in one month? No, not present. But you're working on that, right?
Speaker 3
No, no, no. But the same thing is, there may be certain things that doesn't take a lot of people.
George Church
We just don't know. We just don't have that much experience with having... you know, thousands of Einstein-type levels of creativity and intelligence simultaneously in a generation. And in fact, it's probable that we're all capable of being a bit more efficient if we don't have, you know, distractions of mental illness, of, you know, taking care of other people. Now, taking care of other people may be a very good thing. You know, it may be that if we have no one to take care of, there'll be something bad that happens to us socially. So these things are very complicated, hard to predict. I think right now, I think the The baby step, or actually the pretty big baby step, is to eliminate diseases or at least make it possible for people to eliminate their own diseases as they see fit.
Dwarkesh Patel
You worked on brain organoids and brain connectome and so forth. That work, how has it shifted your view on fundamentally how complex intelligence is? In the sense of like how, you know, are you like more bullish on AI because I realize that organoids are not that complicated. Or it's like very little information is required to describe how to grow it. Or are you like, no, this is actually much more gnarly than I realized.
George Church
I think I always felt it was very gnarly. And I also felt that it was something that we could engineer. Yeah. Certainly, we have made a lot of progress at the broken end of the spectrum where the brain is severely challenged relative to average. You know, there's thousands of, a huge fraction of genetic diseases that are, that have one of their consequences being that the child is developmentally delayed to such an extent that it's lethal or, you know, or a lifetime deficit. And we know how to We know the genes evolved and we know how to do genetic counseling and in some cases, gene therapy and other therapies to deal with it.
At the other end, you know, we have reduction of cognitive decline by cognitive enhancement, which is showing some promise. But again, that's kind of like this early stage severe impediment to cognition has a late stage component. But what about... How much information does it take to encode a brain? I'm not sure that it's that much less genome is required than if you just wanted to make a brain, because the brain is totally entangled with the body. You know, you have 10 to the 11th neurons, 10 to the 14th synapses. If you wanted to reproduce a particular brain, let's say, it might be... It's speculative as to whether it would be easier to do that by making a copy of it in silico, in some kind of inorganic matrix, or making a copy of it.
Both of those are going to be hard. I would say that if you wanted to make a copy of a you know, complicated book, it would be easier to take photographs of each of the pages than to completely translate it into another language, trying to get all the nuances of the poetry and so forth, if your goal is just to replicate it. And I think the same thing might be true of a brain. But replicating a brain probably involves a lot more information than synthesizing it. So, I mean, just to define this 10 to the 14 synapses is going to take a lot more bytes than the genome, which is billions rather than 10 to the 14th.
But you might, there might be reasons that you want to replicate a particular brain configuration rather than just make another animal that is, you know, starts from scratch as an infant.
Dwarkesh Patel
Given how little I knew about biology, my prep for this episode basically looked like one minute of trying to read some paper and then chatting with an LLM like Gemini 30 minutes afterwards and asking it to explain a concept to me using Socratic tutoring. And the fact that this model has enough theory of mind to understand what conceptual holes a student is likely to have and ask the exact right questions in the exact right order to clear up these misunderstandings is honestly been one of the most feel the AGI moments that I've ever experienced. This is probably the single biggest change in my research process, honestly, since I started the podcast.
For this episode, I think I probably spend on the order of 70% of my prep time talking with LLMs rather than reading source material directly because it was just more useful to do it that way. And given how much time I spend with Gemini in prep for these episodes, improvements in style and structure go a really long way towards making the experience more useful for me. That's why I'm really excited about the newly updated Gemini 2.5 Pro, which you can access in AI Studio at ai.dev. All right, back to George. Going back to the engineering stuff, Often people will argue that, look, you have this existence proof that E.
coli can multiply every or duplicate every 30 minutes. Insects can duplicate really fast as well. But then with our ability to manufacture stuff with human engineering, you know, we can do things that nothing in biology can do, like radio communication or fission power or jet engines, right? Right. Like how plausible to you is the idea that we could have biobots, which are, you know, like can duplicate at the speed of insects and there could be trillions of them running around. But they also can have access to jet engines and radio communication and so forth. Are those two things compatible? Well, I mean, certain things seem incompatible, like the temperatures of a fission reactor.
George Church
Mm-hmm. isn't obviously compatible, but the possibility that once we, that a biological system can make other things, you know, for example, You know, it can make a nest. A bird can make a nest, okay? And you consider the whole nest as part of the replication cycle of the bird. So you can say biological thing that replicates at 30 minutes doubling time could make a nuclear reactor. That would be its nest. Mm-hmm. But you need to expand this range of materials. In a certain sense, we do this already. Humans are a biological thing that replicates, not in 30 minutes, but in 20 years or less. And is that fundamentally limiting us?
Yeah, probably is. But yes, it's amazing to think about. What if you could take a cornfield or a nuclear reactor and suddenly 30 minutes later you got two of them, right? And then four of them. I mean, that's quite an interesting concept. But I think we should start with, I teach a course called How to Grow Almost Anything. And I work with Neil Gershenfeld at MIT who has a course called How to Make Almost Anything. And we're trying to meet in the middle where we can And his mechanical electrical engineering will meet with our biological. And in fact, neither of us can make or grow almost everything, because there are all kinds of little gaps in things that are very hard to make in a small lab, because there are things all over the world that depend on multibillion-dollar fabs to make things.
But, you know, we're eating away at it. I think that we might eventually be, you know, maybe a smaller baby step than making a nuclear reactor is making a phone. You know, you said radio communication. We should make a biolot. It should be a small... challenge goal for the synthetic biology community, maybe iGEM or something, make bacteria make a radio. Now, actually, Joe Davis is an artist that's been affiliated with my lab and before that Alex Rich's lab. And he did make a bacterial radio, but it was kind of more on the art end than on the science end. But I think that would be a good goal.
Dwarkesh Patel
What would it take to do... whole genome engineering to such a level that for even a phenotype which doesn't exist in the existing pool of human variation, you could manifest it because your understanding is so high that you can, for example, if I wanted wings, Yeah, right. Is the bottleneck our understanding? Is the bottleneck our ability to make that many changes to my genome? So part of this has to do with just learning the rules of development of biology, like I said. We can determine morphology at sort of the molecular level now, proteins, nucleic acids.
George Church
Determining at the cellular, multicellular level, there's a lot more things you can do and a lot faster. but we don't know the language yet. I think we're on the cusp of getting the tools to do that, like the transcription factor I was talking about earlier. Harnessing, migration, gradients of diffusion factors, chemotaxis, and so forth. That's one thing we need, but there's a bunch of things we need, really.
Dwarkesh Patel
What discovery in biology, so not in astronomy or some other field, in biology, would make you convinced that life on Earth is the only life in the galaxy? And conversely, what might convince you that, no, it must have arisen independently thousands of times in this galaxy?
George Church
Oh, I see what you're getting at. Right. I mean, so astronomy might be we would detect radio signals or light signals. Right. But biology, the kind of evidence would be that you show in a laboratory using... prebiotic conditions a really simple way to get life, right? Or, I mean, it's a harder proof to prove that given, because we don't know all the possible prebiotic conditions. And probably the number was vast. I mean, you have 10 to the 20th liters of water and, you know, at various different salinities and drying up on the ocean and the sun and the lightning and all this stuff. But yes, I think if you showed kind of reconstructed in the lab a very simple pathway from inorganics, cyanide derivatives and reduced compounds, all the way up to some cellular replicating structure, I think that might...
It might lead us to believe that at least life exists. Now, there are other parts of the Drake equation that might kick in, which is maybe it's hard to get intelligent life because intelligence isn't necessarily in your best interest. And if you get intelligent life, it's hard to maintain that without societal collapse or without robotics taking over and then killing themselves, right? And that's hard to do experiments. But I think... To your question, I think an experiment that showed maybe multiple different ways of getting to a living system from non-living systems spontaneously would be interesting.
Again, I'm not sure it would be very hard to prove the negative. So I'm curious, between intelligent life and some sort of primordial RNA thing,
Dwarkesh Patel
What is the step at which, if there is any, where you say there's a less than 50% chance something like at this level exists elsewhere in the Milky Way?
George Church
Yeah, I think these are very challenging problems. I'm not even sure we would be able to say within five orders of magnitude, much less 50%. I think it's more likely to come from exploration than it is going to be from simulation. If, you know, The sad truth is that almost none of the missions that we've sent outside of Earth have actually looked for life. They've had components that could have looked for life, but a sad number of those, not enough components that could look for life, and the ones that could look for life, not really looking for it. And when we get positive results, we dismiss them, as happened with the...
And so I think if we just start looking at the, you know, the geysers that are coming out of various moons of Jupiter and Saturn, there's so much water, there's 50 times more water, liquid water, not frozen, more liquid water in our solar system than in Earth. Doesn't that seem likely that, you know, some of that would have been a good breeding ground? But it could be that we need sunny shores, you know, where they have a lot of dry land and right next to water. Maybe these are just giant oceans that are surrounded by ice and that's not added. But in any case, we need to look at those fountains to see what's popping up.
That's a high priority. And the same thing goes for, you know, there's a lot of water on Mars that's maybe even more accessible. But until we've exhausted those, those are probably the easiest. They're hard. We're still talking about multibillion-dollar experiments, but I think they're a little more convincing. And again, it'll be hard to prove the negative. If we find this negative on everything in the solar system, you know, there's so much more diversity out there that could have done it. If in a thousand years we're still using DNA and RNA and proteins for top-end
Dwarkesh Patel
manufacturing, the frontiers of engineering, how surprised would you be? Would you think like, oh, that makes sense, evolution designed these systems for billions of years? Or would you think like, oh, it's surprising that these ended up being the systems, whatever evolution found just happened to be the best way to manufacture or to store information? Yeah.
George Church
I don't think I'd be surprised either way. I can imagine it going either way. I can imagine making truly amazing materials using proteins as the catalysts, or maybe in some cases as a scaffold as well as catalysts. I think one thing that's probably already happening, so we don't have to go a thousand years out, is the number of amino acids that's going up. It's going up radically from 20. I think pretty soon we'll have a system where we can have 33, 34 new non-standard amino acids being used simultaneously with all the standard ones in a E. coli cell. So 34 plus 20 is a lot bigger than 20. I don't think we necessarily need more more than four nucleic acid components.
I mean, certainly there are plenty of modified ones. There's a bunch of alternative base pairs, some of which don't even involve hydrogen bonds, so we could have more. But I think the main thing is this information storage, and whether it's bits, digital binary is just zeros and ones, that works pretty well for 99% of what we do electronically. So, having four is better than two, maybe. But do we really need six? You know, I don't know. So, yeah, I wouldn't be surprised if we had... Another possibility is that we changed the backbone of DNA. So, maybe keep the ACGT, but make it out of peptides now.
A little bit smaller, a little bit more compatible. I don't know. Or maybe they'll just be... just a slight, you know, it could be part of the new amino acid collection. And there'll be more. I mean, these are just things that my primitive 21st century brain is coming up with a thousand years from now. It'll be a whole new millennium. So it makes sense why evolution wouldn't have discovered radio technology, right? But things like more than 20 amino acids or these different bases so that you can
Dwarkesh Patel
store more than two bits per base pair. Or, for example, the codon remapping scheme, this redundancy, which it seems like based on your work, you can... there was this extra information you could have used for other things. So is there some explanation for why 4 billion years of evolution didn't already give living organisms these capabilities? I think that evolution has a tendency to go with what works and the investment in
George Church
making a whole new base pair would have been high. And we haven't even articulated what what the return on investment would be. What do you get from that? We have made systems like Floyd, Roemsburg, and others where you have replication and transcription and translation with a new base pair, but it hasn't been clearly articulated what that gets you. even in technological society. So in technology, you can jump to things where all the intermediates aren't incrementally useful. Evolution is, as far as we know, generally limited to you have to justify every change. It's like some bureaucracy says, well, if you're gonna put this sidewalk in, you have to justify that before you can build a city. So we've talked about many different technologies you worked on or are working on
Dwarkesh Patel
right now, from gene editing to de-extinction to age reversal. What is an under-hyped technology in a research portfolio, which you think more people should be talking about, but gets glossed over?
George Church
It's hard to say because as soon as you say it, it becomes hyped. So if I've ever been asked this question before, it's too late. But, you know, I would say One thing I think is very ripe and is very well understood in a certain sense, but is nevertheless ignored. It's kind of like the previous example I would have chosen was making genes out of arrays. Arrays were typically used for analytic, you know, quantitating RNAs or something like that. So the original aphymetrics type arrays. But we turned them into gene arrays. people weren't using it. It was in nature, it was hidden in plain sight.
But anyway, it was somehow under-hyped. What I would say is genetic counseling is under-hyped. It is clearly competitive with gene therapy in a certain sense. I mean, clearly not for people that are already born, but for people in the future, not even distant future, next couple of years. We've got a chance of diagnosing them or... Diagnosing the potential parents and dodging. And this has been in practice since 1985 in Doria Shurim. Perfectly reasonable community response to it. Eliminated or greatly reduced all sorts of very, very serious inherited diseases. It's sometimes just, you know, depending on how it's presented, it's dismissed as eugenics.
I think it's rarely that I heard Doria Shurim described that way, and rightly so. What they're doing is standard medicine. You know, whether you, you know, cure these kids as soon as they are newborns or whether you counsel the parents so that the same disease is missing, The problem with eugenics was that it was forced, the government forced it on people. It wasn't that it enabled people to make a choice. It's that it removed the choice from the people. That was what was wrong. And that's the confusion. But I don't think that's the explanation for why this is under-hyped. I think it's people, when they're dating, they're not thinking about reproduction necessarily.
And when they're thinking about reproduction, they're not... You know, they're not necessarily thinking about serious genetic diseases because they're rare. I think it's our difficulty with dealing with rare things. It's like there was great resistance to seat belts because less than 1% of people died in automobile accidents or even got hurt. Great resistance to stopping smoking. Really, it's hard even for us to imagine how great the resistance was for seat belts and smoking. And, but eventually we got over it. I think this is a similar thing, which is that only 3% of children are severely affected by genetic diseases.
And they feel like, well, I'm not that unlucky. You know, I'm in the 97%, right? 97% of those were your odds of winning at the horse races or at the casino, you'd take them. Yeah, 97% of winning, good. But when a child's future is at risk, I think that's... It's not the right solution. And the other thing is, I think, has to do with the trolley problem. It's like, if you don't influence it, it's not your fault. But actually, everything is your fault, you know? Not doing something is a decision, right? And so I think it's like, if I just don't do anything and they come out damaged, well, it's not my fault.
But it is, yeah.
Dwarkesh Patel
David Reich was talking about how in India, especially because of the long-running history of caste and endogamous coupling, that there have been these small cell populations that have high amounts of recessive diseases. And so there it's especially...
George Church
I think I said, yeah, I know what you're saying and what David is saying, but I think it's a dangerous dichotomy. You know, they'll say that there's certain, there are lots of, not just India, you know, all over the world. And in fact, we all went through a bottleneck. That changes the rate from, say, 3% to 6%. But the point is, 3% is still unacceptable. I mean, it's just a tragic loss, not only of the human life directly affected, but the whole family. That's right. Very often, one or both parents have to quit their job and spend full time caregiving and fundraising, because these are very expensive diseases as well.
And it's just... We don't need to—we need to be careful not to stigmatize as well. So when—if a bunch of families get fixed, we shouldn't point a finger at the ones that are unwilling to get fixed because that's their choice, you know. But I think as word spreads and you see the positive outcomes, I think there will be— it will be seen as one of the simplest bits of medicine ever. I mean, in fact, it's something... Save the vaccination. Yeah, it's very inexpensive. In fact, it's less than zero because you spend... $100 per genome, and it'll probably be less soon, and you get the whole thing analyzed.
And compare that to millions of dollars that will be lost, opportunity costs, and not being part of the workforce, the taking care of them and so forth. So the return on investment is tremendous. It's at least a tenfold return on investment. So it's a no-brainer from a public health standpoint. We should be able to pay for this through national health services in England, through insurance companies in the United States. And it turns the insurance companies from being the bad guys, that they're like snooping in on your personal life and then raising your rates to, oh, they're giving you this free information and you can do with it as you wish.
And you could, if you... take the advice, then you save them millions of dollars.
Dwarkesh Patel
Right. Do you think genetic counseling is a more important intervention or even in the future will continue to have a bigger impact than even gene therapy for these monogenic? Oh, absolutely.
George Church
I've actually counseled my gene therapy companies that that they should be investing in very common diseases, because rare diseases have this genetic counseling solution, with the exception of spontaneous mutations and dominance, which probably are IVF clinic-type solutions, but the rare recessives can be handled at matchmaking at every level. Anyway, I counsel my genetic therapy companies that they should, you know, invest in common diseases like age-related diseases and infectious diseases. And in fact, you know, the COVID vaccine was formulated as a gene therapy and was, you know, the cost was in the, you know, $20 per dose range and 6 billion people benefited from it or 6 billion people took it. you know, and it was, you know, proven over the whole population. So I think that's the more appropriate use of gene therapy. But I think for practical reasons, you know, getting FDA approval and so forth, you might go for the rare diseases, and that's perfectly fine. But I think the cost effectiveness of the sweet spot for gene therapy is for age-related diseases and the sweet spot for... the rate of diseases is genetic counseling.
Dwarkesh Patel
All right, some final questions to close us off. If 20 years from now, If there's some scenario in which we all look back and say, you know what, I think on net it was a good thing that the NSF and the NIH and all these budgets were blown up and got doged and so forth. I'm not saying you think this is likely, but suppose there ends up being a positive story told in retrospect. What might it be? Would it have to maybe come up with a different funding structure? Basically, like... Yeah, what is the best case scenario if this post-war system of basic research is upended?
George Church
Oh, whew. I have to preface this, but when scientists answer a question and explore possibilities, it doesn't mean they're advocating it. In the past, people have asked me off-the-wall questions about Neanderthals, for example, and then it was described as if I was enthusiastic about it. So, not enthusiastic about NIH and NSF budgets being cut. you could say, well, it forces us to think more seriously about philanthropy and industrial sponsored research. That could be a positive thing. It could be that that makes us listen more carefully to what society actually needs rather than doing basic research. I'm a big proponent of basic research, but also maybe I'm more than average connecting the basic research to societal needs from the get-go. I don't think it actually interferes with basic research to think and act on societal needs at the same time. So that could be a positive. It could be that it creates a... another nation state that now is the dominant force, you know, like China could now become the next empire after.
Dwarkesh Patel
This is a positive story?
George Church
Yeah, well, it could be for China, right? I mean, you didn't specify who it's a positive story for. You know, the U.S. displaced Britain, which would displace, you know, Spain and Portugal, you know, it keeps moving. Fresh blood is sometimes a good thing. Again, I preface this by saying I'm not advocating this. What else could go well? You know, there's just certain things that we, the society is fairly good at doing collectively that we're not good at doing individually. You know, building roads, schools, and science are examples of that. Doesn't mean we couldn't learn how to do that. You know, to some extent, when you build a gated community, a lot of that is done with private funding. It's possible we could figure out how to build roads and schools and just about everything. It means we're gonna run into some kind of hyper-capitalism that might mean there's all kinds of pathologies that come along with that.
Dwarkesh Patel
What is it about the nature of your work, maybe biology more generally, that makes it possible for one lab to be behind so many advancements? I don't think there's an analogous thing in computer science, which is a field I'm more familiar with, where you could go to One lab and... One academic lab, yeah. Yeah, sorry, one academic lab. And then 100 different companies have been formed out of it, including the ones that are most exciting and doing a bunch of groundbreaking work. So is it something about the nature of your academic lab? Is it something about the nature of biology research? What explains this pattern?
George Church
Well, first of all, thank you for being so generous in your evaluation, which may be taken with grain of salt. But... I think that what it is being in the right place at the right time, so Boston is a unique culture. It attracts some of the best and brightest students and postdocs automatically. It is dense enough. Sometimes people want to spread the wealth out evenly all over the universe or the planet. And there's advantages to having it clustered. So like if you have spouses can find other jobs in the same field. So having a concentration of biotech and pharma MIT and Harvard and BU and so forth all in one pretty walkable distance, you know, not spread out all along the east or west coast, but actually in a walkable city is one thing.
That's the starting point. And then a lab that chooses from from an early stage to, you know, to keep this dynamic between basic science and societal needs going at all costs, causing great trauma when the lab starts, but then getting a couple of wins and it starts building up a, you know, a it's a positive feedback loop where just like the building of Boston was a positive feedback loop, the more Harvard's and MIT's and high tech startups than pharma. And so you get a couple of wins in the literature and people start coming that are, you know, a whole nother level up on it. And maybe they're already aiming for entrepreneurship well before they weren't.
Anyway, it evolves in a way that you can't just jumpstart from, you couldn't just suddenly create Harvard and MIT in the middle of the desert and suddenly create a lab that is taking these kind of risks early in a career. And then also the timing is good because the exponential is starting to show up. The exponential is pretty much the same in the beginning of the hockey stick and the end of the hockey stick, but you don't notice it until it gets... And so that's what's happening is both the computing, AI... biotech, they're all peaking at this point. And so whichever lab happened to already have that positive feedback loop going with the academic to industry technology transfer would asymmetrically benefit from that exponential.
And to some extent, exponential, you can really look like you're very productive when really you're just kind of sliding downhill. Yeah, look at how productive I am. I just jumped out of a plane and I'm accelerating steadily.
Dwarkesh Patel
So yesterday, I had a dinner with a bunch of biotech founders. And I mentioned that I was going to interview you tomorrow. And so somebody asked, wait, how many of the people here have worked in George's lab at some point or worked with him at some point? And I think 70% or 80% of the people raised their hand. And one of the people suggested, oh, you should ask him, how does he spot talent? Because it is the case that many of the people who are building these leading companies or doing groundbreaking research have done have been recruited by you, have worked in your lab. So how do you spot talent?
George Church
I'm glad you framed it as spotting talent. I've heard at least one meme that all you have to do is show up and you'll get into my lab, which is definitely not true. First of all, there's a lot of self-selection, frankly. We're an acquired taste. Technology development is not at all the same skill set as regular biology, where you, you know, you pick a gene, you pick a disease, you pick a phenomenon, and you hammer away at it for your whole life. This is more you make a library where you have, you know, a million members of the library are going to fail, and maybe one or two will succeed. Very different attitude.
It's much more engineering, but it's even different from most engineering, where engineering doesn't usually use libraries that way, millions and billions of components that are non-random. But many of them will fail. Yeah. So the question is selection criteria. So of that, there's a self-selection. And the next thing is in the interview, I typically tell them I'm looking at people that are nice. I'm not necessarily looking for geniuses. We end up with a lot of geniuses. That's wonderful. But NICE, I think, is highly predictive of how well you will do in the lab and afterwards. And as a consequence, I think we have a kind of international set of alumni that are quite nice to each other.
even though they're supposedly in cutthroat fields and i think they're nice to other other people as well so that's nice is one criteria multidisciplinary it's hard to build a multidisciplinary team from disciplinarians so if you have two people that each know two languages or two skills even if they don't have anything in common they have shown that they can learn a new skill and then they'll each add a new the the skill that connects them as a third thing so uh those are the three main things i would
Dwarkesh Patel
say final question Given the fast pace of AI progress, your point taken that we should be cautious of the technology, but by default, I expect it to go quite fast and there not being some sort of global moratorium on AI progress. Given that's the case, what is the vision for, we're going to have a world with, we're going to very plausibly have a world with genuine AGI within the next 20 years. What is the vision for AI? biology, given that fact? Because if AI was 100 years away, we could say, well, we've got this research we're doing with the brain or with gene therapies and so forth, which might help us cope or might help us stay on the same page. Given how fast AI is happening, what is the vision for this bio-AI co-evolution or whatever it might look like?
George Church
I think one scenario, if we handle the safety issues, and that has to be a top priority, if we handle that properly, then we're probably going to have almost perfect health. Why wouldn't we? It's going to go so fast. I mean, it's going to go pretty fast with just regular AI, without AGI. But if you add to it AGI, and it'll be a positive feedback loop, because the more people they get fixed you know or get access to good healthcare the more people will be helping prompt the AI if that's necessary and I think it probably will be and the more hybrid systems we'll have of people and machines working together in harmony hopefully in this very positive scenario yes well that's a good vision to end on
Dwarkesh Patel
okay George thank you so much for coming on yeah thank you Thank you for tuning in. I'll see you on the next one.