Today I have the pleasure of  interviewing George Church. 

I don't know how to introduce you. This is not even an 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 de-extinction. So you weren't exactly an easy prep. Okay,  

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? Escape velocity is sometimes  

what it's called for aging. Different people have estimates  

and all those estimates, including mine, are  going to be taken with a big grain of salt. 

Mainly looking at the exponentials in  biotechnology and the progress that's been  

made in understanding—not just understanding  causes of aging, but seeing real examples  

where you can reverse subsets of the aging  phenotype—you're getting close to all of aging. 

In other words, 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 are common to age-related diseases and  I'm going to get more than one at a time." 

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 now  

making it 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. The thing is, it's not going to be some  

sudden point where you're going to be 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. Sorry, yeah. Is there any hope  

for doing that from somatic gene therapy alone,  or would that have to be germ line gene therapy? 

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, it doesn't apply to us, the  two of us and everybody listening to this. 

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. 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,  signaling and so forth—you could imagine that if  

you replaced every nucleus in the body, it would  suddenly be young again without going all the way  

back to the embryo and forward again. There's various other things that  

are just short of that. If you replace the cells,  

will they fit into that niche? They might displace the old cells. 

That's certainly within the realm of modern  synthetic biology, for cells to take over niches. 

I think the hardest part is the brain. Even there, 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. 

A Ship of Theseus kind of thing in the brain. Yeah exactly, Ship of Theseus, trying to maintain  

the connections and the memories. There's some fairly straightforward  

experiments that need to be done before we can  really even estimate how hard that problem is. 

Very often there's low-hanging fruit  that people just think is improbable. 

But it's there because biology  has all these gifts where it  

just hands over to us levers that we can flip. Like vaccines are this amazing gift that didn't  

have to exist, but they do. Is there an existing gene  

delivery mechanism which could deliver gene  therapy to every single cell in the body? 

There is nothing close to that today. But there's nothing, no law of physics,  

that would prevent it. Again, there's going to be practical  

considerations, like how many injections  do you need to do to achieve that goal? 

But we're getting better at targeting tissues. One of my companies, Dyno Therapeutics,  

showed they could get a hundredfold  improvement in targeting  

neurons in the brain, which is a big deal. That was just one little campaign that they did. 

That one experiment involved a lot of AI and a  lot of testing of millions of different capsids. 

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… The question is how  close to 100% do you need to get? 

It's going to vary from tissue to tissue. For example, for some therapies you just  

need to get 1% because that 1%  can produce some missing enzyme. 

And that 1% doesn't have to  necessarily be in its normal place. 

You can turn a muscle into part of the  immune system temporarily for a vaccine. 

An enzyme that's normally made in, let's  say the brain, you could make in the liver,  

if the point is just to get it into the blood. So I think that's moving along quite well. 

You're one of the co-founders of  Colossal which recently announced  

that they de-extincted a dire wolf. Now you're working on the woolly mammoth. 

Do you really think we're going  to bring back a woolly mammoth? 

The difference between an elephant and a woolly  mammoth might be like a million base pairs. 

How do we think about the kind of  thing we're actually bringing back? 

People get worked up about whether we are  trying to bring back, or have already,  

or will ever bring back a new species. If you think of it, rather than as a  

natural thing that we're trying to do, but as  synthetic biology with goals that has potential  

societal… People also get worked up as to whether  this could possibly benefit society in any way. 

Can we really fix an environment to suit  humans or fix the global carbon to suit humans? 

The answer is we don't know. But it's worth a try, isn't it? 

Because it could be very cost-effective. The other aspect of it is there's a whole  

discipline within synthetic biology  of asking, "What's the minimum?" 

People often phrase it into, "What's the maximum?  What can we do?" I'm interested in both. Yes,  

there's millions of differences  between mammoths and elephants. 

There are millions of differences between  elephant one and elephant two, within Asian  

elephants and between Asians and African. But not all of those are definitive in terms  

of what we would normally call them and how we  would normally classify them, and what their  

functionality would be in an ecosystem. There's this exercise that people do. 

We've done it, for example,  with developmental biology. 

What's the minimum number of  transcription factors it takes to  

make a neuron from a pluripotent stem cell? What's the minimum number of base pairs it  

takes to make something that will replicate to  something that was done in mycoplasma originally? 

In a way, these are more interesting than,  "Can we make a perfect copy of something?" 

What's the minimum things we have to do to make  it completely functionally, or even functionally  

in a particular category? 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, etc. With the dire wolf, we clearly  

didn't make an exact copy of a dire wolf. But it helped illustrate and educate  

people around the world on, "What is the  difference between a gray wolf and a dire  

wolf?" Because direwolves, they're big.  Maybe they have a particular coloration. 

The head components tend to be  bigger than the leg components. 

How many genes do you need to do that? Maybe this was Direwolf 2.0, and we're  

going to go for 3.0 and successive approximations. We might want to develop the technology for making  

exact copies of something, 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'll be a matter of what you should make  and what would be most beneficial for  

the species that you're making, for the  environment it lives in and for humans. 

Does this teach us something interesting  about phenotypes which you think are  

downstream from many genes, and 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, 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. You're hitting on a very interesting question. 

It's related to, "What's the minimum?" For example, you almost said it. 

Take a very multigenic trait in humans. Height is probably the most well-studied one,  

simply because no matter what gene, 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 something  on the order of 10,000 genes, of which we  

have 20,000 protein coding genes. Some of them are RNA coding genes. 

They each have a tiny influence on height. But if you take growth hormone, somatotropin,  

you have extreme examples where you'll get  extremely low small stature and extremely  

high stature due to that one alone. In fact, it's used clinically as well  

for seven different medical treatments. 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 module that we could use in  other cases and translate it to other species. 

You hit on it just right. Not everything will translate,  

but we start accumulating these widgets. It's kind of like all the electronic  

widgets that we accumulate over time. If you just want to slap it into the next  

circuit, you might be able to. What implications does this  

have for gene therapy in general? What is preventing us from finding the latent knob  

for 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? How do we find it? With biology we've got a real gift which is  

that it's both 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, evolutionarily, there was  no selection specifically to have two heads. 

But just a little deviation from the  normal developmental pattern during  

fetal development and they both function fine. They control subsets of the body and  

they have their own personality, their own life. There's all kinds of things you can do in biology  

where you're working at a  very high programming level. 

That’s a way of thinking about it. But pushing us to a new level of  

intelligence is going to be very  challenging and maybe not even urgent. 

To some extent, actualizing the people that  we currently have would be quite impactful,  

just getting them all up to whatever  speed they want to be up to within the  

range that's been demonstrated. Some people are going to want  

to be like Einstein, some people won't. Some people will want to be healthy all the time. 

Unlikely, but some people might not. Some people might want to live to be 150,  

some people might want to die at 80. But if you give them that range,  

that capability… What if we had 8 billion, super  healthy, don't need to worry about food and drugs,  

super healthy Einstein-level intelligence,  education level the best we can come up with…  

That would be a completely different world. 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 a couple of knobs  which control very high level functions? 

So do you find them through the GWAS,  genome wide association studies? 

Is it through simulations of these… I would say mostly GWAS for humans,  

maybe for animals in general. For animals with synthetic biology,  

the smaller and the cheaper and faster  replicating, the more experiments you can do. 

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. 

For example I mentioned earlier, what's the  minimum number of transcription factors it  

takes to turn a stem cell into a neuron? 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. You can see what transcription factors does it  

express at the time that it's the target? Then you say, "Let's just try those on  

the stem cell and see if they work." That recipe has worked quite well. 

It's the basis of GC Therapeutics  and a bunch of the work that we do. 

You can get a recipe for almost  every cell type in the body. 

Now, that's not new cell types, but at  least you've learned, 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 1, 2, 3, maybe 7 changed  

transcription factors. That's an example. There's  room for lots of other examples of where you can  

do reduction and do not just reductionistic  biology, but then constructionistic. 

You take it back up and make a whole  complex system and see what happens. 

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 probably on the order of 10^14, 10^17. 

Things that involve cells are  typically in the billions. 

But this is how we're going to get inroads  into the very complicated biological systems.

Can I ask you some questions about biodefense? Because some of the stuff you guys work on,  

or quite responsibly choose not to work on,  can keep one up at night. Mirror life. Given  

the fact that it's physically possible,  why doesn't it just happen at some point? 

Some day it'll get cheap enough. Somebody will care about it enough, that somebody  

just does it. What's the equilibrium here? I was a co-author on a paper that  

warned about the dangers of mirror life. Just like 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. 

The advance that we were recognizing,  in our Science paper that was warning  

about mirror life, is not only that  we had to calculate the possibility of  

error-prone escape or something like that was. We don't want anything to escape that we  

made in the lab unless there's a general  societal consensus that it's a good thing. 

So far there aren't any examples of that. Mirror life, if it can be weaponized, that  

would take it to a whole other level of concern. The concern was that if we got it to a certain  

point, then it would be easy to weaponize it. Again, there's practical considerations that  

maybe that most people who 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, one group  

probably, or one person. 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 solar system or maybe even on our planet.  

It just hasn't been weaponized. What we were  saying in the Science paper is that this seems  

like the sort of thing that could wipe out all  competing life if were properly weaponized. 

But there are probably a few things like that. 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 one disgruntled person who has  

essentially too much power. Over the history of humanity,  

the amount of things that a single person  can do has grown very significantly. 

It used to be, when you had your bare hands, there  was kind of a limit to what one person could do. 

A large number of people could team up  and get a mammoth or something like that. 

Today, one person with the right connections or  right access to technology could blow up a city. 

That's a huge increase in capability. I think we want to start dialing  

that back a little bit somehow. What does that look like in terms of not just  

mirror life, but synthetic biology in general? Maybe we're at an elevated period of  

the ratio to offense and defense. How do we get to an end state  

where—even if there's lots of people running  around with bad motivations—somehow there are  

defenses built up that we would still survive,  where we're robust against that kind of thing?  

Is such an equilibrium possible? Or will  offense always be privileged in this game? 

Offense awfully 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 do 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 that are harder and  

harder to detect, more and more subtle to  the stochastic variation between people. 

There's some people that are just so happy they  would never want to do anything close to that. 

Or they're so responsible or ethical or whatever. Then there are other people who, whenever they  

have a bad day, they want to  take a lot of people with them. 

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 that. 

Well, there will be technological  solutions to the psychiatric problem. 

It could be that even people who aren't  sure whether they want to be helped or not  

can test, try it out, and it's reversible. They say, "Yes, I like that better." Okay,  

let's try that then. Then there's other things  that cause you to have bad days. It's not just  

your psyche. It's also the environment. So if  you're surrounded by your people being starved,  

infectious disease, or you’re being  shot at or something like that,  

those are things that are subject to  sociological and technological solutions. 

If we could really solve a lot of that stuff,  we could reduce the probability that one person… 

This is maybe pessimistic because you're basically  saying we have to solve all of society's problems  

before we don't have to worry about synthetic  biology, which I'm not that optimistic  

about. We'll solve some of them. Right? You shouldn’t be. I'm  

not trying to reassure you. We're having a conversation  

about what it takes and that might be  one scenario for what it might take. 

You had an interesting scheme for  remapping the codons in a genome so that  

it's impervious to naturally evolved viruses. Is there a way in which this scheme would also  

work against synthetically manufactured viruses? It’s much harder. Again,  

the offense has the advantage. We can make a lot of different codes. 

Which will limit the transmissibility? 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 10^80 different codes. 

Some of them you might be  able to take out all at once. 

Anyway, coding space is a kind  of more interesting space. 

Of course it could get even more complicated  than that because the 10^83 is based on triplet  

codons and that sort of thing. But if they're quadruplet  

codons or novel alphabets and so on… We're sort of getting into a cycle of competition. 

It'd be better to nip it in the bud. Why did we spend so much societal  

resources building up to tens of  thousands of nuclear warheads? 

Now we've dialed it back to  mere thousand nuclear warheads. 

It's nice that we dialed it back, but  why did we waste all that time and money? 

Biology seems very dual-use, right?  The mere fact that you, literally you,  

are making sequencing cheaper will just have this  dual-use effect in a way that's not necessarily  

true for nuclear weapons. And we want that,  right? We want biotechnology to advance. 

It's hard to pound nuclear weapons  into ploughshares, as they say. 

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 encryption 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 so we'll just have to limit access  

to these tools and have better monitoring  but there's not a more robust solution? 

One of the things I advocated in 2004 is that  we stop deluding ourselves into thinking that  

moratorium and voluntary signups 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. 

We had essentially moratoria and  disapproval for germline editing. 

Nevertheless, somebody did it and  a lot of people knew about it. 

That was clearly a failure of the whole moratorium  and voluntary and whistleblower components. 

It worked for five years with only  one defector. That's quite impressive. 

Okay, half empty, half full, I'll give you that. 

But all it takes is one for  some of these scenarios. 

It would have been nice if the whistleblowers  could have saved him the three years in prison  

by getting an intervention. It's not like anybody  died. Right. There are probably three healthy  

genetically-engineered children in the world now.  They’ll be teenagers soon. But it was a good test  

run that 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. 

Over the last couple of decades we've  had a million-fold decrease in the cost  

of sequencing DNA, a thousandfold in synthesis. We have gene editing tools like CRISPR, massive  

parallel experiments through multiplex  techniques that have come about. 

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 a huge Industrial Revolution, a huge  

burst of new drugs, or cures for Alzheimer's  and cancer that have already come about? 

When you look at other trends in other fields,  we have Moore's law and here's my iPhone. 

Why don't we have something  like that in biology yet? 

We have something that's about the same speed,  a little bit faster than Moore's Law in biology. 

It's more recent, that’s one aspect of it. We could stand on the shoulders of the  

electronics giants to go a little bit  faster to catch up. I would say we do.  

We have the biotech industry, which has  used 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. 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. 

We're on the verge of really combining electronics  and biology more thoroughly, and AI and biotech. 

It seems like we're on the same  track as Moore's law, if not better. 

What exactly are we on the verge  of? What does 2040 look like? 

Well with 2040 we're talking about only 15 years. Which is 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. There's lots of cool things we can do with just  

super AI, but we need to  be very cautious with AGI. 

Anyway we can get into that. I have questions for you there. 

We are shortening the time of getting medical  products approved still in a safe way. 

But that's not going to  completely change the exponential. 

It might reduce it from 10 years down to… One  year is our record so far for say COVID vaccines. 

Maybe that'll be 10 times shorter. Maybe that will multiply out a little bit. 

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,  

some sort of hybrid thing. But again that won't be completely shocking. 

It's just going to be so much of it. There's going to be lots  

of diversity of solutions. How much more are we talking about? 

Are we going to have 10x  the amount of drugs? 100x? 

I'm not even sure it's going to make sense,  but 100x would not be completely surprising. 

Combinations of drugs will be important, using  them intelligently. There'll be a lot more.  

Some drugs will affect everything, 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. 

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? The cost curves are affected by new tools. 

It's not just some automatic thing. There was a big discontinuity between  

Sanger sequencing and nanopores and  fluorescent next-gen sequencing. 

Sometimes it's a merger of two things. Clearly AI merging with protein  

design caused a step function. These step functions get smoothed out into 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. 

After that, the merger of developmental  biology with manufacturing,  

conquering developmental biology. In other words, it’d be actually  

knowing how to make any arbitrary shape  given DNA as the programming material. 

That would be a big thing. Just having more materials in general. 

All the materials that we use in mechanical and  electrical engineering should be made better by  

biotechnologies. Why is that? 

Well in electronics, I wouldn't say Moore’s law  is stopping, but think about what we would call  

the 1 nanometer process, which is supposed  to come out in 2027 according to the roadmap. 

It's not really 1 nanometer, it's more like  40 nanometers, center-to-center spacing,  

typically in two dimensions, maybe  a little bit of three dimensions. 

Biology is already at 0.4 nanometer  resolution and it is in three dimensions. 

Depending on how you count that third  dimension, it could be a billion times  

higher density that biology is already at. We just need a little more practice with  

dealing with the whole periodic table. Even electrical and mechanical engineering  

don't typically use the whole periodic table  typically, especially not at the atomic level. 

Biology is just really good  at doing atomic precision. 

So then what's the reason that over the last  many decades 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. 

People in the 90s were saying ultimately we'd 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? 

A few things. One is the  arrival of synthetic biology. 

We were already kind of doing synthetic  biology before, we were doing recombinant DNA,  

a kind of genetic engineering. 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. 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  and 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 are definitely polymers that biology  

can make that will conduct at the speed of light. We could make a mixed neuronal system that has  

conventional neurons and processes that  conduct at the speed of light. That would  

be interesting. So I think that our ability  to design proteins was particularly difficult.  

Designing nucleic acids was great. 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,  

it's actually the one kind of thing where  morphology is dictated by fairly simple rules. 

It's not how developmental biology works. 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 maybe eight years ago, something like that. I think we're just now getting used to it. 

The use of chips for making DNA. You said that DNA synthesis has come down  

a thousandfold, it depends on who you talk to. When we came out with the first chip-based  

genes in a 2004 Nature paper, basically  people dismissed it for about a decade. 

The only people that used it  were collaborators and alumni. 

It wasn't even listed on the Moore's law curve for  DNA synthesis, even though it was thousand times  

cheaper. It was just ignored. Now we have claims  of 10^17 genes that you can make libraries of,  

that aren't randomized in the usual sense,  where you just do error-prone PCR or spiked  

nucleotides. 10^17, that's a lot bigger than a  thousandfold if it turns out to be practical.

Speaking of protein design, another thing you  could have thought of in the 90s—People were  

writing about nanotechnology,  Eric Drexler and so forth. 

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?  Why didn't AlphaFold cause that? 

Part of it is that nanotechnology as  originally… The source of the inspiration,  

Eric Drexler, he wanted to reinvent biology  in a certain sense but it already existed. 

So you don't need to design a diamond replicator  because you already have a DNA replicator. 

The question was, what was missing? What was motivating this reinvention of  

biology? It was materials. Biology is not  that great with materials that are, say,  

superconductors, conductors, semiconductors, and  light speed. But it's getting there. Rather than  

going the route of everything having to be based  on first principle nanostructures, you can meet  

in the middle where biology can build things. Of course, when you go down to liquid nitrogen  

and colder temperatures, biology as we  currently know it stops functioning. 

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 temperatures… Or maybe  biology now, because you can make these big  

libraries of biology, maybe 10^17 in vitro,  you can flip through them quickly and you can  

barcode them and you can… 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 a billion different kinds  of electronic materials just in an afternoon,  

barcode them all and see who wins. But we do it all the time in biology  

now, at least since 2004 we have. So I think that's an opportunity. 

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, from libraries. We call it chemical  / biochemical / exotic material libraries. 

The point is that 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 materials science revolution? 

What is standing between now and that? Because we've got AlphaFold right now. 

So what is the thing that we  need? Do we need more data? 

AlphaFold is very nice, but it's only part of it. There are large language models  

that are different from AlphaFold. To give an example, with AlphaFold—last  

time I checked anyway—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. That's where you need either  

extraordinary precision or just knowledge of what  happens evolutionarily, or happens in experiments,  

to say that, "No, alanine won't work. Okay?" So  I think there's all kinds of combinations of AI  

tools that can give you deeper insight into that. If AlphaFold predicting the structure doesn't tell  

you whether the thing will 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 that Y thing,  

and I can just get that"? 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. We make variations on it and then whichever  

of those variations work, we make variations  on that. 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. 

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 and have  more of a focus on those. Another thing  

that's been missing. None of the AI protein  design tools that I know of are particularly  

good at it yet but as we speak, we're trying  to improve it, is nonstandard amino acids. 

Because a lot of these tools depend on adding  libraries of 3D structures, which use 20 amino  

acids, and large language models where you  line up all the sequences of 20 amino acids. 

We have very little experience with extra ones. But there's a revolution going on in generating  

nonstandard amino acids, where the amino acids  can either have as part covalent part of them,  

or as easily liganded, all the stable  elements in the entire periodic table. 

Each of those we’ll 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. 

Ultimately, the determination of the functionality  of your library is a kind of computer. 

You use AI to design the library optimally. You avoid things that are really  

neutral and really seriously damaged. But 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. It's a hundred percent precision, because you're  

not simulating. You're not making assumptions.  You're not going from quantum electrodynamics,  

which is an assumption, to quantum mechanics,  which is an assumption to molecular mechanics,  

which are full of assumptions. You're really doing the real thing. 

So you're doing a kind of natural computing. 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. 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  a couple more drugs by 2040? 

Well, I didn't mean to stop there. I knew the conversation would continue. 

I'm not pinning down a particular year either,  but this is poised to go pretty quickly. 

There are very few practitioners, which is  the thing that will stop it for a while. 

Materials actually should go  faster though, because they  

don't require quite as much regulatory approval. It's one of these things where when you get the  

right idea, it's not hard to recruit people. For example, when Feng Zhang and my labs  

brought out CRISPR, we each got 10,000  requests in the next two months for people  

that wanted to duplicate the system. That's what I hope will happen with the  

nonstandard amino acids and using AI for  protein design and making new materials. 

Hopefully that will recruit tens  of thousands of people overnight. 

Are you more excited about AI which  thinks in protein space, or capsid space,  

just predicting some biological or DNA sequences? Or are you more optimistic about LLMs just  

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  

it some combination, when you think  about AI and bio is more promising? 

I'm much more excited about scientific  AI than I am about language AI. 

With languages, we're in  pretty good shape already. 

What worries me is that to get to the  next level of language requires AGI or  

ASI. That's very dangerous. I don't think we  have quite figured out how to handle that. 

There's a lot of safety organizations  and a lot of safety rules and so forth. 

What typically happens when there's  an intense competition is those safety  

rules get undermined and pushed aside. Even if they weren't, I don't think we  

understand our own ethics well enough to educate  a completely foreign type of intelligence. 

We barely know how to pass it onto  the next generation of humans. 

So we need time to sort that out. There's no  rush. This is a completely artificial emergency. 

This is not like COVID-19, where millions of  people were 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. So I think we need to go very slowly  

on AGI and ASI, and double down on  slightly narrower scientific goals. 

With even that, we need to be very cautious. We need to have kind of an international  

consensus on what constitutes safe AI. Suppose we did build safe superintelligence. 

How much would that speed up bio progress? There's a million George Churches in data centers  

just thinking all the time, is it a 10x speed-up? 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. Suppose you could get them to care about it. 

There's a million copies of you in a data center. How much faster is bio progress? 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. 

But let's put safety aside for a moment. It's not only hard to calculate the bads,  

it's hard to calculate the goods. 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. Well, we could say,  

"Yes, that could be a game changer." But do we really need it? Is that really  

important? Maybe it'd be more interesting to just  have Zoom calls that are better, or we can just  

learn how to get everything we want in our  kitchen and we don't need to travel anymore. 

So be careful what you ask for. You could tip our priorities towards something  

that we really don't care about, that we shouldn’t  care about, or we might wish we didn’t care about. 

But I'm curious, you've still got to run the  experiments, you still need these other things. 

So does that bottleneck the  impact of the millionth copy  

of you or do you still get some speed up? Basically, how much faster can biology go  

if there are just more smart people thinking,  which is a sort of proxy for what AI might do? 

These are great questions and 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 at present. But you're working on that, right? 

No, but the same thing is that  there may be certain things that  

don't take a lot of people. We just don't know.  We don't have that much experience with having  

thousands of Einstein-type levels of creativity  and intelligence simultaneously in a generation. 

In fact, it's probable that we're all  capable of being a bit more efficient  

if we don't have distractions of mental  illness, or taking care of other people. 

Now, taking care of other  people may be a very good thing. 

Maybe 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 and hard to predict. 

I think right now, 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. You've worked on brain organoids  

and brain connectome and so forth. How has that work shifted your view  

on fundamentally how complex intelligence is? Are you more bullish on AI because you realized  

the organoids are not that complicated,  or rather very little information is  

required to describe how to grow them. Or are you like, "No, this is actually  

much more gnarly than I realized." I always felt it was very gnarly. 

I also felt that it was  something that we could engineer. 

Certainly we have made a lot of progress  at the broken end of the spectrum where the  

brain is severely challenged relative to average. There are a huge fraction of genetic diseases that  

have as one of their consequences the child  being developmentally delayed to such an extent  

that it's lethal or causes a lifetime deficit. We know the genes involved and we know how to  

do genetic counseling in some cases, gene  therapy and other therapies to deal with it. 

At the other end, 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  having a late stage component. 

But what about, how much information  does it take to encode a brain? 

I'm not sure that much less genome is required  than if you just wanted to just make a brain,  

because the brain is totally  entangled with the body. 

You have 10^11 neurons, 10^14 synapses. If you wanted to reproduce a particular brain,  

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 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. 

The same thing might be true of the brain. But replicating a brain probably involves  

a lot more information than synthesizing it. Just to define this, 10^14 synapses is going  

to take a lot more bytes than the genome,  which is billions rather than 10^14. 

But there might be reasons that you want to  replicate a particular brain configuration  

rather than just make another animal  that starts from scratch as an infant.

Going back to the engineering stuff, often people  will argue, "Look, you have this existence proof  

that E. coli can duplicate every 30 minutes. Insects can duplicate really fast as well. 

But then with our ability to manufacture stuff  with human engineering, we can do things that  

nothing in biology can do, like radio  communication or fission power or jet engines." 

How plausible to you is the idea that we could  have biobots which can duplicate at the speed  

of insects—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? 

Certain things seem incompatible. Like  the temperature of a fission reactor  

isn’t obviously compatible. But it is a possibility that  

a biological system can make other things. For example, it can make a nest. 

A bird can make a nest. You consider the whole nest as  

part of the replication cycle of the bird. So you can say the biological thing that  

replicates at a 30 minute doubling  time could make a nuclear reactor. 

That would be its nest but you need  to expand its 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.  Is that fundamentally limiting us? Yeah,  

probably it is. But it's amazing to think about. What if you could take a cornfield or a nuclear  

reactor, and suddenly 30 minutes later you've got  two of them, then four of them, and eight of them.  

That's quite an interesting concept. I teach  a course called How to Grow (Almost) Anything. 

I work with Neil Gershenfeld at MIT who has  a course called How to Make Almost Anything. 

We're trying to meet in the middle where  his mechanical electrical engineering  

will meet with our biological. In fact, neither of us can make  

or grow almost anything because there's  all kinds of little gaps and things  

that are very hard to make in a small lab. There are things all over the world that depend  

on multi-billion dollar fabs to make things. But we're eating away at it. 

Maybe a smaller baby step than making a nuclear  reactor is making a phone. You said radio  

communication. It should be a small challenge goal  for the synthetic biology community, maybe iGEM or  

something: make bacteria make a radio. Actually Joe Davis is an artist—he’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. 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. 

For example, if I wanted wings. Is the bottleneck  our understanding? Is the bottleneck our ability  

to make that many changes to my genome? Part of this has to do with just learning  

the rules of developmental biology, like I said. We can determine morphology at the molecular level  

now: proteins, nucleic acids. 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. 

What discovery in biology—so not in astronomy or  some other field—would make you convinced that  

life on Earth is the only life in the galaxy? Conversely, what might convince you that no,  

it must have arisen independently  thousands of times in this galaxy? 

Oh, I see what you're getting at. Astronomy might be that we would  

detect radio signals or light signals. With biology, the kind of evidence would be  

that you show in a laboratory using prebiotic  conditions a really simple way to get life. 

It's harder to prove the negative because we  don't know all the possible prebiotic conditions.  

Probably the number was vast. You have 10^20  liters of water at various different salinities  

and drying up on the ocean and the sun  and the lightning and all this stuff. 

I think if you showed, reconstructed in the lab,  a very simple pathway from inorganics, cyanide  

derivatives and reduced compounds, all the way up  to some cellular replicating structure, that might  

lead us to believe that at least life exists. Now that there are other parts of the  

Drake equation that might kick in. Maybe it's hard to get intelligent  

life because intelligence isn't  necessarily in your best interest. 

And if you get intelligence life, it's hard to  maintain that without societal collapse or without  

robotics taking over and then killing themselves. That's hard to do experiments on. 

But to your question, 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'd be very hard to prove the negative. Between intelligent life and some sort of  

primordial RNA thing, what is the step at  which, if there is any, you say there's a  

less than 50% chance something like at this  level exists elsewhere in the Milky Way? 

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. 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. 

A sad number of those had not enough  components that could look for life. 

The ones that could look for life  were not really looking for it. 

When we get positive results, we  dismiss them as happened with Pioneer. 

I think if we just start looking at 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 but liquid water, in  our solar system than on Earth. 

Doesn't that seem likely that some of that  would have been a good breeding ground? 

It could be that we need sunny shores where  you have a lot of dry land right next to water. 

Maybe these are just giant oceans that are  surrounded by ice and maybe that's not an ideal. 

In any case, we need to look at those fountains  to see what's popping up. That's a high priority.  

The same thing goes for 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. 

Again, it'll be hard to prove the negative. If we find this negative on everything  

in the solar system, 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 manufacturing,  

the frontiers of engineering,  how surprised would you be? 

Would you think, "Oh, that makes sense. Evolution designed these systems  

for billions of years." Or would you think, "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"? 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. 

One thing that's probably already happening,  we don't have to go a thousand years out,  

is that the number of amino acids is going up. It's going up radically from 20. 

I think pretty soon we'll have a system where we  can have 34 new non standard amino acids being  

used simultaneously with the standard ones in a  E. coli cell. 34 plus 20 is a lot bigger than 20. 

I don't think we necessarily need more  than four nucleic acid components. 

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—whether it's  bits, digital binary—it’s just zeros and ones. 

That works pretty well for 99%  of what we do electronically. 

Having four is better than two maybe, but do  we really need six? I don't know. I wouldn't  

be surprised. Another possibility is  if we changed the backbone of DNA. 

Maybe we keep the ACGT, but make it out  of peptides now, a little bit smaller,  

a little bit more compatible. I don't know. It  could be part of the new amino acid collection.  

There'll be more. 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. 

It makes sense why evolution wouldn't  have discovered radio technology. 

But things like more than 20 amino acids  or these different bases so that you can  

store more than 2 bits per base pair or  for example, the codon remapping scheme,  

this redundancy, which it seems like based  on your work there was this extra information  

you could have used for other things. 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. 

The investment in making a whole  new base pair would have been high. 

We haven't even articulated what  the return on investment would be. 

What do you get from that? We have made systems,  

like Floyd Romesberg 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. In technology you can jump to things where all  

the intermediates aren't incrementally useful. But evolution, as far as we know, is generally  

limited to… You have to justify  every change, like some bureaucracy,  

"If you're going to put this sidewalk in, you  have to justify that before you build a city." 

We've talked about many different technologies you  worked on or are working on right now, from gene  

editing to de-extinction to age reversal. What is an underhyped technology in your  

research portfolio which you think more people  should be talking about but gets glossed over? 

It's hard to say because as soon  as you say it, it becomes hyped. 

If I've ever been asked this  question before, it's too late. 

One thing I think is very ripe and is very  well-understood in a certain sense but is  

nevertheless ignored… The previous example I  would have chosen was making genes out of arrays. 

Arrays were typically used for analytic,  quantitating RNA, or something like that,  

the original Affymetrix-type arrays. But we turned them into gene arrays  

and people just weren't using it. It was  in Nature. It was hidden in plain sight. 

But anyway it was somehow underhyped. What I would say is genetic  

counseling is underhyped. It is clearly competitive  

with gene therapy in a certain sense, clearly  not for people that are already born but for  

people in the future, not even distant  future but in the next couple of years. 

We've got a chance of diagnosing them or  diagnosing the potential parents and dodging… This  

has been in practice since 1985 in Dor Yeshorim,  a perfectly reasonable community response to it. 

It eliminated or greatly reduced all  sorts of very serious inherited diseases. 

Sometimes, depending on how it's  presented, it’s dismissed as eugenics. 

Rarely have I heard Dor Yeshorim  described that way and rightly so. 

What they're doing is standard medicine  whether you cure these kids as soon as  

they are newborns or whether you counsel  the parents so 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 underhyped. I think it's because when people are they're  

dating, they're not thinking  about reproduction necessarily. 

And when they're thinking about reproduction,  they're not necessarily thinking about serious  

genetic diseases because they're rare. I think it's our difficulty  

with dealing with rare things. There was great resistance to seat belts  

because less than 1% of people died in  automobile accidents or even got hurt. 

There was great resistance to stopping smoking. It's hard even for us to imagine how great the  

resistance was for seat belts and smoking. But eventually we got over it. 

I think this is a similar thing. Only 3% of children are severely  

affected by genetic diseases and they feel like,  "I'm not that unlucky. I'm in the 97%." If those  

were your odds of winning at the horse races or at  the casino, you'd take them. 97% of winning, good. 

But when a child's future is at risk,  I think that's not the right solution. 

The other thing is that I think it  has to do with the trolley problem. 

If you don't influence it, it's not your fault. But actually everything is your fault. 

Not doing something is a decision. So I think it's like, "If I just  

don't do anything and they come out  damaged, it's not my fault", but it is. 

David Reich was talking about how in  India—especially because of the long running  

history of caste and endogamous coupling—there  have been these small subpopulations that have  

high amounts of recessive diseases. So there, it's an especially  

valuable intervention. I know what you're saying, and what David  

is saying, but I think it's a dangerous dichotomy. There are lots of, not just in India but  

all over the world… In fact we  all went through a bottleneck. 

But that changes the rate from say 3% to 6%. But the point is 3% is still unacceptable. 

It's just a tragic loss, not only of the human  life directly affected, but the whole family. 

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. 

We need to be careful not to stigmatize as well. So 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. 

But I think as word spreads and you see  the positive outcomes, I think it will be  

seen as one of the simplest bits of medicine  ever. It's very inexpensive. In fact,  

it's less than zero because you spend a hundred  dollars per genome. It'll probably be less  

soon. You get the whole thing analyzed. Compare that to millions of dollars that  

will be lost in opportunity costs and  them not being part of the workforce,  

taking care of them and so forth. So the return on investment is tremendous. 

It's at least a tenfold return on investment. It's a no brainer from a public health standpoint. 

We should be able to pay for this through  the National Health Service in England,  

through insurance companies in the United States. It turns the insurance companies from being the  

bad guys snooping in on your personal  life and then raising your rates to  

them giving you this free information  and you can do with it as you wish. 

If you take the advice then you  save them millions of dollars. 

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 disorders? 

Absolutely, I've actually counseled my  gene therapy companies 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 rather than… But the rare recessives  can be handled at matchmaking and at every level. 

Anyway, I counseled my genetic therapy companies  that they should invest in common diseases like  

age-related diseases and infectious diseases. In fact the COVID vaccine was formulated as  

a gene therapy and the cost was in the $20 per  dose range. 6 billion people benefited from it,  

or 6 billion people took it and it  was proven over the whole population. 

So I think that's the more  appropriate usage of gene therapy. 

For practical reasons, getting FDA approval  and so forth, you might go for the rare  

diseases and that's perfectly fine. But I think the cost effectiveness…  

The sweet spot for gene therapy is for  age-related diseases and the sweet spot for  

rare diseases is genetic counseling. Alright, some final questions to close us off.  

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 DOGE’d 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, what is the best case scenario if this  

post-war system of basic research is upended? I have to preface this. 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. 

I’m not enthusiastic about  NIH and NSF budgets being cut. 

You could say that 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 at 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. That could be a positive. It  could be that it creates another nation-state  

that now is the dominant force. Like China could now become the  

next empire after the US. Is this a positive story? 

It could be for China. You didn't  specify who it's a positive story for. 

The US displaced Britain, which displaced Spain  and Portugal. It keeps moving. Fresh blood is  

sometimes a good thing. Again, I preface this  

by saying I'm not advocating this. Let's see, what else could go well? 

There are just certain things that society is  fairly good at doing collectively that we're  

not good at doing individually. Building roads, schools,  

and science are examples of that. It doesn't mean we couldn't learn how to do that. 

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 going to run into  some kind of hypercapitalism. 

That might mean that there's all kinds  of pathologies that come along with that. 

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 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. 

Is it something about the  nature of your academic lab? 

Is it something about the nature of biology  research? What explains this pattern? 

First of all, thank you for being  so generous in your evaluation. 

Maybe take it with a grain of salt. But I think that what it is is being  

in the right place at the right time.  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. 

There are advantages to having it clustered. Spouses can find other jobs in the same field. 

Having a concentration of biotech and pharma  and MIT and Harvard and BU and so forth,  

all in one pretty walkable distance, 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 an early stage 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. 

It starts building up a positive  feedback loop, just like the  

building of Boston was a positive feedback loop. The more Harvards and MITs and high tech startups,  

then pharma… So you get a couple of wins  in the literature and people start coming  

that are a whole other level up. Maybe they're already aiming for  

entrepreneurship while before they weren't. Anyway, it evolves in a way that you can't  

just jump start from scratch. You couldn't just suddenly create  

Harvard and MIT in the middle of the desert  and suddenly create a lab that is taking these  

kinds of risks early in a career. 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 at the end of the hockey stick,  

but you don't notice it until it gets going. That’s what's happening in computing,  

AI, and biotech. They're all peaking at this point. 

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. 

To some extent with the exponential, you can  really look like you're very productive when  

really you're just kind of sliding downhill. It's like, "Yeah, look at how productive I am. 

I just jumped out of a plane  and am accelerating steadily." 

Yesterday I had dinner with  a bunch of biotech founders. 

I mentioned that I was going  to interview you tomorrow. 

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?" 

I think 70% or 80% of the  people raised their hand. 

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  been recruited by you, have worked in your lab. 

So how do you spot talent? Well, 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 pick a gene,  you pick a disease, you pick a phenomenon,  

and you hammer away at it for your whole life. This is more like you make a library where you  

have a million members of the library  that are going to fail and maybe one or  

two will succeed. It’s a very different  attitude. It's much more engineering,  

but it's even different from most engineering. Engineering doesn't usually use libraries that  

way, millions and billions of components that  are non-random but many of them will fail. 

So the question was selection criteria.  There's self-selection. The next thing is,  

in the interview, I typically tell them  that I'm looking for 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. 

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  people as well. So nice is one criteria.  

Multidisciplinarity. It's hard to build a  multidisciplinary team from disciplinarians. 

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 the  

skill that connects them. That's the third thing.  Those are the three main things I would say. 

Final question. Given the fast pace of AI  progress—your point taken, that we should  

be cautious of this 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? 

We're going to very plausibly have a world  with genuine AGI within the next 20 years. 

What is the vision for biology given that fact? If AI were 100 years away we could say 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? 

If we handle the safety issues,  and that has to be a top priority,  

then we're probably going to have almost perfect  health. Why wouldn't we? It's going to go so fast. 

It's going to go pretty fast  with just regular AI without AGI. 

But if you add to it AGI… It'll also be a  positive feedback loop, because the more  

people that get fixed or get access to good  healthcare, the more people will be helping  

prompt the AI, if that's necessary. I think it probably will be. 

The more hybrid systems we'll have of  people and machines working together in  

harmony in this very positive scenario. Well, that's a good vision to end on. 

George, thank you so much for coming on. Thank you.