不要制造镜像生命：合成生物学家 Kate Adamala 谈风险与责任

Hello, and welcome back to the Cognitive Revolution. Today, my guest is Kate Adamala, professor of genetics at the University of Minnesota and synthetic biologist, who recently made headlines for coordinating a diverse group of prominent researchers to issue a collective warning against the creation of mirror life. We begin with a discussion of the origins of life and what we can infer from the fact that all of life's vast diversity operates on a single biochemical framework. And then Kate gives an overview of the current state of synthetic biology. Much like AI interpretability researchers, biologists today are using a mix of top-down and bottom-up methods to create the simplest possible model organisms, but have yet to create something that is both alive in the functional sense and which we fully understand. If you've never encountered this research before, I am confident that you will find it fascinating. The main reason I invited Kate on the show, though, is to discuss her journey from a proponent and active developer of mirror life to someone who's now warning against it. For context, mirror life is a possible form of life which does not exist on Earth but is clearly physically possible, made of molecules that are the mirror images of life's normal molecules. This relationship, also known as reverse chirality, means that while the chemical properties of the molecules are exactly the same in almost every way, mirror molecules cannot occupy the same space as one another, just as one's left and right hand, even if perfectly identical in every other way, can never occupy the exact same space. This creates exciting possibilities for medical treatments, but also some risk that mirror organisms could be impossible for predators to digest and thus might prove catastrophic if they were ever released, intentionally or accidentally, into the wild. Kate was initially excited about the utility and the intrinsically interesting nature of this work. But with time, and particularly after seeing experimental data showing that mirror molecules actually do seem to evade immune systems, she began to reconsider, ultimately concluding that the risks far outweigh any potential benefits. From there, rather than simply abandoning the work quietly, she took the unusual but I think highly admirable step of actively building a coalition of researchers, including some of the biggest names in biology, to publish a warning in Science against developing mirror life. Now, importantly, this wasn't a binary or permanent decision to abandon synthetic biology entirely or even necessarily to avoid mirror life forever. Kate continues her work on synthetic cells and still has hope that mirror molecules can be useful in medicine. She and her co-authors have simply identified one particular branch of the synthetic biology tech tree that they believe humanity would be much better off not exploring, at least until major new evidence comes in. The parallels to AI development should be pretty obvious. Like synthetic biology, AI offers the thrill of discovery to researchers, promises tremendous benefits for the public, and also brings serious and as yet poorly understood risks. And it seems very likely to me that the specific details of the powerful AIs that we develop over the next few years could matter tremendously. Advanced AIs aren't one thing, and neither are they an inseparable bundle. You can create AIs with superhuman coding ability that still don't know how to use a computer, AIs that beat world champions at Go but have no language ability whatsoever, and as recent reinforcement learning work has shown, AIs that solve problems more and more effectively and agentically, but also in more inscrutable and increasingly problematic ways. Given the vast possibility space in front of us, I really hope that AI researchers, particularly at leading companies but also across academia and startups, make a habit of following Kate's example and regularly stepping back from their work to seriously grapple with its implications and to change course when appropriate. Should we continue to scale reinforcement learning given the recent rise in deceptive behavior, which now includes the opportunistic blackmail and whistleblowing that we've seen in Anthropic Report and Claude IV? Should we be racing to turn machine learning research over to such systems given these behaviors? Should we be developing architectures that internalize models' thinking processes such that we no longer have an explicit chain of thought to examine or even such that models might begin to communicate with one another in an alien neuralese that we've never understood? And should we, as one well-meaning person who is interested in AI safety recently emailed me to propose, start using reinforcement learning to train AIs to escape their environments so that we can hopefully use their successes as indicators of the ways in which we need to harden our defenses? I honestly don't think these questions have simple or obvious answers. One of my mantras is that AI defies all binaries. But given the pace at which AI research is moving and the fact that even I, as a full-time student and analyst of the field, can no longer keep up with even the things that seem obviously important, it is critical that individual researchers themselves understand that the responsibility is currently on them, both to ask the right questions and to choose which directions to pursue with foresight and wisdom. Fatalist rationalization that someone will do it if we don't simply doesn't cut it in today's world. Kate's work proves that scientific communities can indeed coordinate to identify and avoid particularly dangerous research directions while still aiming for transformative progress. For those of us involved in AI development, this offers both an inspirational model and an important challenge. The question isn't whether to develop AI at all. I do agree with those who say that the upside is too great to pass up and that in any case, the cat's out of the bag. But still, we can and should hold ourselves to the highest possible standard when it comes to assessing the risks and being willing to back off and change course when necessary. As always, if you're finding value in the show, we'd appreciate it if you'd share it with friends, post about it online, or leave a review on Apple Podcasts or Spotify. We always welcome your feedback too, either via our website cognitiverevolution.ai, or by DMing me on your favorite social network. Finally, I was honored to learn that we were voted the number three AI podcast at Swyx's AI Engineer World's Fair this week. I'm really very glad to know that this show has proven a valuable resource to such a plugged-in group, and I hope that our mix of technical deep dives, broad surveys, and occasional moralizing lectures can nudge AI development in a positive direction, however slightly. In any case, a big thank you to everyone who voted and everyone who listens. Now, I hope you enjoy this fascinating exploration of synthetic biology, existential risk, and the great responsibility that comes with truly transformative research with Professor of Genetics and synthetic biologist, Kate Adamala. University of Minnesota Professor of Genetics and synthetic biologist, Kate Adamala, welcome to The Cognitive Revolution.

Thanks for having me. Hi, everyone.

I'm excited for this conversation. The impetus for it is the recent warning that you and others in the field of synthetic biology put out to the broader community about the dangers of mirror life, and why you think this is a branch in the technology tree that we should not be going down, at least for now. I'm really interested to get into that, and I do think it has some not-one-to-one mapping to what's going on in AI, but certainly some rhymes that people would do well to consider.

But before we get to all that, for our audience, which is broadly obsessed with AI and interested in frontier technologies, can we do a little primer on synthetic biology?

Yeah.

Give us the 101 of what you guys are up to in synthetic biology. Maybe one thing that I've seen you use to motivate some talks before, which I found fascinating as a jumping-off point, is that, obviously, life is really diverse, but you make this fascinating observation that there are some ways in which it's not really diverse at all. Specifically, our biochemistry is not at all diverse.

Yes.

So maybe start with that—

Yes.

—observation—

Yeah.

—and unpack it from there.

From the biochemical or chemical point of view, life is actually extremely boring because there are a huge number of different types of chemicals that life could be using and isn't using. Life only uses 22 proteinogenic amino acids. Twenty-two amino acids go into our proteins, and it uses only 5 different nucleobases that go into our DNA and RNA. There are hundreds, if not thousands, of amino acids out there that can be naturally synthesized, and there are a lot of different types of possible nucleic acids that could exist, but life is very limited in what it's using.

It managed to do great things with that limited set of chemicals. We have an amazing variety of form and function in life. But when you look at it from that very basic biochemistry point of view, it's really not that diverse.

And so that's one of the motivations for synthetic biology: let's expand the chemical repertoire of life. Let's give life tools to play with that natural life doesn't bother using, and let's see what happens. Let's see what kinds of functionalities we can evolve, and also what we can learn about life in general—what kinds of chemistries can be used in life.

Do we have any idea how wide that range of possibility is? It's one thing to say we're using N amino acids and we could use N + 1, N + 2, or N + 100, maybe even more. And then there's also this sort of—I don't know—is it realistic to think that there could have been other biochemistries that were just fundamentally different? Do we have any theory that tells us how wide-ranging that possibility space really is?

The limiting factor in all our theories is imagination at this point, because we only have N = 1 for modern terrestrial life.

We’ve never made or found any other life form. So we can speculate about what’s possible, and people have gone really crazy in those speculations. People talk about life that doesn’t even have to be carbon-based, life that can exist under truly extreme pressures and temperatures, and life that can be silicon-based. I’m not an expert in that.

If I’m an expert in anything at all, it would be carbon-based molecules. So when I think about different possibilities for life, I think of different things that could be made with the kinds of biological molecules that we know right now, just an expanded set of them. Even if you look at amino acids, there are literally hundreds of different amino acids that can be made into proteins using the natural ribosome—not a modified ribosome, just the natural ribosome that bacteria, humans, and everyone else has.

That ribosome already can translate hundreds of different amino acids. It doesn’t do that in nature, but if we take that ribosome out of cells and gently encourage it in vitro, then it will translate them. So that just shows you where the floor is. That’s definitely not the ceiling. We can imagine that if we evolve that ribosome, or if we make any other kinds of biochemical assemblers, they could assemble really weird things into polymers.

But even sticking to the natural ribosome, we can feed it this huge variety of different substrates that don’t even have to be amino acids. The ribosome can catalyze other molecules connecting into polymers. It doesn’t even have to be an amino acid. So the possibilities are almost endless without even going too crazy, without even leaving the carbon world or the water-based life world.

To answer your question directly, no, there is no single theory that would limit the possibilities. But there also doesn’t have to be, because there are so many different possibilities that we haven’t even started exploring. We don’t have to go super crazy, so it’ll still be exciting.

We’ve got several follow-up questions just on this alone. Do you have a functional definition of life that you go to all the time?

I do. My personal functional definition of life is a quote from US Supreme Court Justice Potter Stewart, who said, “I will know it when I see it.”

The reason there is no good definition of life that will be exhaustive of all possible life forms and exclude all non-life is that it’s really, really difficult. For example, NASA has a definition of life, kind of their working definition of life, which says it’s a self-replicating chemical system that can undergo Darwinian evolution. On the surface, that sounds great. You’d think that all life is self-replicating and undergoes Darwinian evolution.

But when you really think about it, according to NASA’s definition of life, I’m not alive because without my husband, I cannot replicate, and I don’t undergo Darwinian evolution on my own. In order for me to replicate and make offspring, I need a partner. That already shows you a giant, gaping hole in NASA’s definition of life, because I’m not a self-replicating organism.

With any definition of life that you can find—and people have made many—there can be a hole found in any one of them. So we don’t really know how to define life. That’s why our functional definition is that if we find life elsewhere in a solar system, as long as it’s close enough for us to recognize it as living, then it’s going to be life.

That crosses over a little bit into the problems that you normally cover on this podcast and with your audience. People have raised that repeatedly: If we don’t actually have a good definition of life, how do we know, for example, that a very advanced AI wouldn’t cross that definition? Our definition doesn’t limit life to being molecular-based. There’s no definition of life that says you have to have molecules, DNA and proteins, or even any kind of molecules.

So you can imagine realizing the functions of life, whatever you decide they are, without a body even. You can imagine having a very complex computational system with emergent properties that would cross the threshold into life. I think it’s a really difficult question. For me, this very intuitive understanding is that if I see something, I can tell you if it’s alive.

But most importantly, life has to have emergent properties. It has to be more than the sum of its parts. According to that definition, probably some really complex, lifelike computational systems would qualify.

Yeah. That’s fascinating. Okay, well, let’s come back to that.

Fascinating but true.

Definitely.

Because without a good definition, you can’t really say what you’re doing. But such is life.

Yeah, literally and figuratively. I see what you did there. Okay, so definitely we’ll come back to that.

On just Earth-based life as we know it, do we have a sense that this only ever happened once, or would the best guess be that there maybe were other biochemistries that developed? What happened to them? Did they get outcompeted, or did they just fail on their own? What’s the best guess as to N equals 2 and beyond in terms of other biochemistries?

Yeah. My personal guess, which is shared by a lot of people in the community, is that it happened way more than once. The reason we think about it that way is because life started on Earth really quickly. The moment Earth had a crust and liquid water, life started. On the geological timescale, it was almost instantaneous.

The moment the planet became capable of supporting life, life started. So it stands to reason that the process of the origin of life, under the right conditions, cannot be that difficult. When life started, it probably happened in many places almost simultaneously.

Whenever you have life, it competes with everything else. If all of the life came from the same pool of biomolecules, then they could compete for the same resources. That’s probably the reason we have only one lineage of life right now: the one that gave rise to us ate all the others or outcompeted all the others. Everything in nature tries to eat everything else, and that probably happened very early in the origin of life, too.

It’s possible that life started multiple times right at the beginning, and it’s possible that the origin-of-life processes keep happening. There’s no rule that says life cannot start in 2025 somewhere deep in a hydrothermal vent. Except when life first starts, it’s extremely fragile. It’s very, very wimpy.

If you happen to be that very nascent life that starts on a planet full of highly evolved, functioning life, then you’re just going to get eaten before you even get started. So the going theory is that life started many, many times, but only one really made it to today. Once you have established life on a planet, then good luck trying to start again. There’s just no way for a new life to get started.

Yeah. It’s a crowded space now.

Now it is, yes. But originally, when life started, it had kind of a free rein.

Yeah. Again, one wonders about how analogous the modern compute environment is compared to early Earth and how ripe it might be for—

Yeah.

—takeover by some new form. But again, let’s come back to that.

I mean, I think physical containment is one thing that you have that’s different from life on early Earth, because your computing systems are contained to the hardware. Life on the early Earth had free rein of the prebiotic ocean.

Mm-hmm.

So that’s one thing. If you create parasite AI—I don’t know if that’s even a thing—but if you do create a parasite AI, how would it spread? I guess it could spread as a virus, a computer virus.

When you said that even the current ribosome can handle more amino acids than we actually use, that sort of reminded me of the development of language. My highly non-expert, stylized story is that even today, if you go to the part of the Earth closest to where language was initially developed, as we understand it, which is some place in Southeastern Africa—I’m not exactly sure where—you see languages that have a lot more sounds.

Then, if you follow the path of human spread across the globe, maybe the final place is Hawaii or something. There, you have the smallest alphabet and the smallest number of sounds. Is there something similar going on with the development of biochemistry? Did we maybe have more of these things in the past and are we gradually reducing them, or am I just hallucinating a pattern that doesn’t exist here?

I think there’s definitely a pattern. Coming from Poland, we use a lot of consonants that, when I moved to the US, people just don’t hear and don’t recognize. There’s definitely a pattern.

In biochemistry, when you look at biochemistry, there are two different, almost competing patterns. One is that very early in biochemistry, there were definitely other molecules being used that are now relegated to secondary roles and are not essential to metabolism. On the other hand, the current complexity of life has to be built up from scratch.

So you can almost trace the lineage of usage of amino acids, for example.

We have five or six amino acids that we speculate were the very, very first ones, then a few others that came slightly later, and then a few others that are almost brand new, that started right before the last universal common ancestor, that last organism that gave rise to all of life that we know on Earth right now. So there are definitely 2 different patterns.

One is that the very early biochemistries that gave rise to life were probably very promiscuous in terms of biochemistry. They used whatever they could get their little hands on because they couldn't be picky; they didn't really have complex biosynthesis pathways yet. But once you set it in stone, once you created an actual cell with a membrane that was isolated from the environment, then you had to start making hard choices because you cannot afford that huge chemical diversity. That means you have to make all of those compounds.

It's really difficult to develop a reliance on a small molecule that you then cannot provide from the environment. One really good example of that is human reliance on vitamin C. Most mammals can synthesize their own vitamin C, and for some stupid evolutionary reason, humans and guinea pigs lost that pathway. So we rely on vitamin C, and that really kicked us in the butt when we started the age of sail and exploration.

People started being on ships for months and started dying of scurvy because suddenly they had no access to vitamin C. No one ever thought about it before because that was just something we had in the environment; we ate it. That's a good analogy for very early life. If you're this primitive cell trying to make it in a prebiotic ocean, if you develop your biochemistry to rely on too many chemicals before you learn how to make them, before you develop biosynthesis pathways, you suddenly become very vulnerable.

If you happen to drift into a region in the prebiotic ocean that doesn't have all of those chemicals that you were born with and you still don't know how to make them, then you're out of luck and you're dead because your biochemistry is suddenly not available. I think that was the driving force: very early life had this buffet of different chemicals, but once it started being more constrained, more compartmentalized, and then moved into different environments, it had to, by necessity, be very simple.

Just like people who move into a van or those nomads who move around the world with their backpack, they have to limit what they carry because they just can't carry everything. Living in my house, I have a basement full of crap because I don't have to move. That's the difference: once you settle down and start synthesizing stuff, then you can expand your biochemistry. These are the 2 things that work against each other: 2 mechanisms, 1 for increasing diversity and the other for limiting diversity.

So how should we understand the ribosome? Is it just an accident that it can handle another 100 amino acids? I guess I don't really know a lot about this, but my intuition would be that something like that would be pretty highly evolved, and I wouldn't expect it to handle another 100 amino acids unless there was some history that explained why, in fact, it at one point needed to.

Yeah. The reason the ribosome can handle a lot of different types of amino acids is because the ribosome is actually extremely dumb. Despite being very ubiquitous—every known life form has ribosomes—the ribosome is actually an incredibly crappy catalyst. In biochemical speak, it's an entropy trap. It's not a proper catalyst.

The business model of a ribosome is to take things and smash them together. It's super simple. It's very, very primitive. The catalytic core of the ribosome is very primitive. That probably indicates a frozen accident. The first organism that was able to figure out how to do protein synthesis, how to make ribosomes, had a huge evolutionary advantage.

But then you're kind of stuck because once you figure that out, you don't want to mess with it because you're worried about breaking it. You cannot afford to break it. That's why all of our internet infrastructure runs on those super-old computers that no one dares to touch because you just can't afford to touch them. You can't afford to update them.

The same goes for my house router. I have a router that was set up 5 years ago, and we don't dare to touch it because we would just lose the internet. You can apply the same principle to a ribosome. Once you've figured out how to make protein synthesis work, then you're not going to apply too much evolutionary pressure to it because any mistake will be lethal, any mistake in your assembly of a ribosome.

Any way to evolve a ribosome will end up being lethal, and that's probably the reason why a ribosome can be as promiscuous as it is. It can translate all those different amino acids because it's not very specialized. There are other enzymes in biochemistry that are incredibly specialized, but the ribosome itself does not guard the genetic code. It does not care what it translates. As long as it fits, it will get translated.

The ribosome's specificity is almost physical. It's like the size of the hole into which the tRNA with the amino acid goes regulates what you can translate. The ribosome has something called an exit tunnel, which is exactly what it sounds like. The nascent protein chain comes out of that hole in the ribosome. That's called an exit tunnel, and the geometry of that hole regulates what you can translate.

There's no smart catalysis there. It's just, let's smash things together into a necklace, into a polymer, and then kick it out the exit tunnel. Whatever fits can be translated. We see that when we do this translation with unnatural amino acids: the ribosome literally doesn't care. The only thing that stops amino acids from being translated is that they're too bulky.

There can be an amino acid that's just too big, and it gets stuck in an exit tunnel, and then the ribosome stalls. But there is no smartness to that biochemistry. The ribosome doesn't look at an amino acid and say, “Hey, you're unnatural, so I won't translate you.” It doesn't care.

There are other enzymes that care, the enzymes that charge amino acids onto the tRNA. Those guys are super particular. They care. They will not charge an unnatural amino acid, but a ribosome does. That probably illustrates the evolutionary age of those enzymes: the ribosome is super primitive and very simple. In that simplicity is the beauty of it, because it can do a lot of things.

One thing I noticed about your language there, which definitely happens all the time in the AI space as well, is the sort of language of, “Once a model has learned to do this, it doesn't want to change its representation too much,” and so on. I realize it's not meant to be understood literally in either space, but in a more literal sense, of course, in the AI space, there's some loss function generally, and you're optimizing with gradient descent.

A lot of these things cash out to: you're in a local minimum of the loss function, and there may be other, even deeper local minima in other places. But getting from here to there can be really hard once you're deep enough.

Yes.

That's exactly the language that people use when they describe the evolution of the ribosome: the ribosome is right now in a local minimum that works, and there probably is some other global minimum that we haven't reached yet. But to get there, you have to climb a hill, and that hill is lethal.

Yeah. Okay, cool. That's really helpful context. One of the things—this is maybe so new that you don't even have an opinion on it yet—but I just saw in the last few days an article come across that said that intelligence is believed now to have evolved independently at least twice. I'm not really in a position to evaluate this, but since it is fresh and you might be in a position to evaluate it, do you have any thoughts on this new claim?

Yeah, I saw that, and my first question is: how do you define intelligence? My dog can tell the sound of a bag of chips apart from the sound of another bag, like a bag when I'm unpacking shopping. It rattles the same, but he knows that when it's a bag of chips, there's something in it for him, so he comes. When it's a bag of shopping, then he doesn't come because nothing's going to fall on the floor.

Is that intelligence? Otherwise, my dog is pretty dumb, but he can tell those things apart when there's something in it for him. So that definitely is intelligence. How do you define intelligence? It's really difficult to answer. Whenever I read about different claims about intelligence, either as an emergent phenomenon that comes out of our neurons and the geometry of them or as some other kind of independent property of life, my problem with that is the same as my problem with the lack of definitions of life: you can't really talk about something if you can't strictly define it.

How do you define intelligence? Is it self-awareness? What is it? That's my main criticism of that theory that you mentioned and all the other theories that talk about the emergence of consciousness and intelligence: we don't have a good definition of it. So it's really hard to track the phylogeny of it.

So it sounds like that would also imply that your perspective is that there are a lot more flavors of this than two, and they probably also sort of evolved or emerged many different times, in different ways.

Many different times independently, and they're probably all within the same kinds of boundaries because all of the intelligence that we know of right now evolved on the chassis of life as we know it. There is no foreign intelligence until you guys make AI that's actually intelligent. Then there is no other foreign chassis that is capable of supporting intelligence.

Yeah. Well, it might be coming soon, if not already.

How would you know, though?

Yeah, well, that's a big, big question, and surprisingly, the range of opinions in the field doesn't seem to be narrowing nearly as much as I would have thought it would at this point. At a minimum, we're able to sustain extreme positions on either end.

I think I'm ready to move on to synthetic biology and maybe just give you the same prompt. What's the 101 on synthetic biology? Why should we be interested in it in the first place? What do we hope to get out of it, and where are we today in the development of synthetic biology?

You should be interested in synthetic biology because it's the coolest thing ever. Okay, now we're done.

Synthetic biology is a field that tries to engineer biology, that tries to expand the chemical and functional abilities of biology. The idea is that biology is very limited in both the chemistry of it and the types of functions that biology performs. Also, biology is really hard to control. It's very difficult to control a cell, to program a cell to do what you want it to do if it's different from the evolutionary program of that cell.

Synthetic biology is basically aiming at changing that. It's making life that behaves and is able to do things that natural biology never bothered doing, and there are a few reasons for doing it. One is that, because it's inherently extremely cool, the ability to engineer life is something that would really be the ultimate triumph of science. The ability to say, "I can program life, I can engineer life to do whatever functions I want it to do."

Gaining that full operational control of life is really the final end goal of synthetic biology: saying that, "I can truly understand all the molecules of life, and I can program them to do whatever I want." Beyond the foundational research and the fascinating aspect of it, there are a lot of motivating practical applications.

We need to be able to truly have this operational control of biology because we need to use biology to sustain life on Earth. The way our civilization is going right now, being reliant on petrochemicals, we're not going to make it past the end of this century. There are 2 ways around it. One is to give up petrochemicals altogether. Let's go back to the preindustrial society. That, that's not going to fly with anyone.

The other way is to keep civilization the way it is, and to do that, we have to sustain the number of people we have. We have to feed the people, we have to put clothes on them, we have to give them entertainment. We basically have to run this whole business of civilization. In order to do that, we need chemicals that right now we take from oil, and we cannot keep taking them from oil because of climate and all that.

So let's make those chemicals with biology. That's the driving force of what we're doing: let's convince natural cells or modified cells to synthesize all those chemicals that we right now extract from dead ferns underground. Those chemicals are really necessary for everything, not just plastics. Most people, when I say petrochemicals, think plastics. There's really everything around you. If you look around whatever room you're sitting in right now, everything around you is made of petrochemicals: electronics, clothes, medicine, cosmetics, but also food.

We would not be able to sustain the current population of Earth using traditional agriculture. We have to use fertilizers, and those fertilizers are derived from petrochemicals. So oil is really at the very basis of our civilization. It's not just energy. People think if we have renewable energy sources, we're good, we don't need oil anymore. That's not true. We still need oil to make things, to make all of the chemical-based things in our life, which is pretty much everything.

In order to replace oil, we need a way to make chemicals with biology, and that's where synthetic biology comes in. Those chemicals are pretty toxic, and no self-respecting bacterium or other biomanufacturing organism will want to make those chemicals for us unless we engineer that organism so that the chemical is not toxic to it anymore. That's the core foundation of synthetic biology: to engineer life to expand the chemical chassis of what life is capable of doing.

Some of that chemistry is what we're now using in industry, known as petrochemicals. So that's one big goal. And if that's not enough, there's another big goal, and that is: let's make humans healthy.

Let's understand exactly how human cells work, because right now there are a lot of diseases whose biology we don't understand. For example, how do human cells develop a lot of kinds of metabolic diseases and a lot of kinds of cancers? Also, how do humans age? We're not fans of aging, and it would be nice to figure out what the molecular basis of it is, how to at least slow it down, and how to increase the health span of humans.

Not just the life span, but the health span, because, you know what, I don't want to live until I'm 100 if I'm going to be sick and unhappy for the last 20 years of it. I really want to be healthy for most of my life. And so that's the goal.

In order to do that, we really need to understand biology. Right now, we really suck at understanding biology, at understanding the molecular basis of biology. There is a quote from Richard Feynman that's overused in our field, but rightfully so because it's still accurate. He said, "What I cannot build, I cannot understand."

That makes perfect sense. If I cannot build a living system, then I cannot really say I understand how life works. That's another driving force behind synthetic biology: let's understand exactly how life works, and then let's develop the ability to engineer living cells so that we can cure diseased states. Basically, let's make better diagnostics and better drugs. That's another thing that synthetic biology is aiming to do.

That's probably enough to be really excited about the field.

Yeah, that's a couple of big-picture issues, to say the least.

Modest goals. Let's save humanity.

So you mentioned, of course, this grand challenge in biology of understanding how it all works. This is a bit of an absurd question, but how far along do you think we are on that quest to understand how things work? I don't even know necessarily how to properly conceive of the metric, but maybe one way to do it would be to say: if there's a causal graph of all the things that are going on in my cells, my tissues, and my systems, and one thing is upregulating one thing or downregulating another, and there are all these knock-on effects and a dense web of—

Yeah.

—causes. How much of that have we mapped out in today's—

Wait.

—biology?

Really not much. Using your graph analogy, we know there is a graph, and we might have a very, very basic idea of the major nodes of that graph. We do not have a full understanding of even parts of it.

We don't know where every molecule goes. We don't know what the cause-and-effect relationships are between those different parts of that graph or different parts of our biochemistry. We don't even have a full ingredient list of biology. If you give me a single human cell, I cannot tell you what every single molecule in that cell is—not even talking about relationships between those molecules.

We simply don't have a full ingredient list, and that's kind of embarrassing for a field that's been at it for the last 50 years. But that's where we are: we're trying to understand what we don't know right now. We're trying to understand the extent of the complexity and figure out what it is we actually have to do in order to understand that super-complex thing that is a cell.

I'm talking about single cells right now. We're still miles away from actually going to a complex organism.

So one aspect of your work that I've found really interesting in preparing for this is the attempt to kind of bootstrap your way—

Mm.

—building on that Feynman idea: you'll know you can understand something if you can build it. You've got a project, or I guess more than a project, a whole line of work on trying to bootstrap from something that is clearly not living to something that is living.

Mm-hmm.

I understand there's also a kind of top-down approach. If that is bottom-up—like, we're going to start with these ingredients and try to give them some spark and get them to come to life—then there's also this: can we—I'll use an AI term—distill down or maybe winnow away everything that's not truly essential until we get to some minimum cell type?

I understand that's also kind of come pretty far, but—

Yes.

—with the major caveat that even the simplest cells we've been able to winnow down to, we still don't fully understand how they work.

Right.

Maybe give us both—

Yes.

—angles on that: What is the current state of trying to build up, and what is the state of building down? How long until we meet in the middle?

I would love to meet in the middle. The top-down approach has succeeded in making a so-called minimal cell. That's Craig Venter's project on building a minimal Mycoplasma-based cell. They succeeded in building a cell that has the minimal genome, a genome that can be completely chemically synthesized.

The kick here, though, is that we still don't know what every single gene in that cell does. There are a couple dozen of those so-called essential genes of unknown function. As the name implies, we know they have to be there: if you remove one of those genes, the cell dies, but we don't know what they do. So we have this very minimal genome, and we still don't understand what a couple dozen genes do.

And don't even get me started on small molecules. There are so many small molecules in that cell that we don't know what they do. We don't know the actual relationships between them. So it is a very minimal organism, and it's still a black box to an extent. We made a window in that box; we can kind of poke at it, and we definitely know much more about that organism than about any other cell, but it's still not a fully understood cell.

The bottom-up approach—the side of the field that I represent—we're basically cheating. We decided we don't know how a cell works, and we're not about to find out. I want to have a research program that spans decades, not centuries. I assume I don't know how a cell works, so instead of trying to understand a complex living cell, I want to build a cell from scratch.

I want to put together molecules that I fully understand, so I know exactly what those molecules do and where every molecule goes, and I want to put them together into a lifelike system. That goes back to what we talked about at the beginning of this interview, which is that we don't know what life is. We don't have a definition of life, but we just roll with it. We decide that we're going to build something that's complex enough that it will have some emergent properties.

We're wanting to build something that looks like a cell, that can self-replicate, that has metabolism, that can exist in the environment, and that can have interactions with other cells. Once we build that, hopefully we'll still be able to understand every single part of it because we built it. Then we might actually meet in the middle with the minimal-cell people, because they keep trying to make theirs simpler, while we'll keep trying to make ours more complex. Hopefully, one day we'll meet at the point where it's an undeniably living cell: it has all the properties of a living cell, but it's fully understood. We can fully understand where every molecule goes, and that's the dream. That's the goal.

How big is that gap right now? Is the number of genes even the right way to think about that gap? How many genes does the minimum viable cell have versus how many genes—or whatever is the right unit of measure—do your synthetic cells have?

The most complex synthetic cells currently available, the ones that have been published, have a few dozen genes. My lab has been working with a synthetic cell that has about 85 to 90 genes. The minimal living cell has 474 genes.

That's a pretty big gap, but it's not an order-of-magnitude gap. There was a paper from George Church's lab about 10 years ago where they speculated that a truly minimal living system would have about 120 genes. That kind of shows you that we're getting there. We're approaching the order of complexity that people speculated could sustain independent life. I don't know if that's a big gap or a small one.

Yeah. Well, time will tell, I suppose. Is the difference between that 120 speculated and the 474—is that another one of these local-minimum-type things? If I understand the 474 correctly, all 474 of those have to be there. If you knock any one out, the cell dies. But theoretically and speculatively, they're saying, “We think this is a local minimum. We can imagine something simpler, but we don't necessarily know how to get there from here.”

A lot of the functions of the minimal cell are things that you could imagine engineering around. You don't have to have complex post-transcriptional modification enzymes. You don't have to make your own membrane. You don't have to do a lot of things that the minimal cell does.

And then there are those essential genes of unknown function. Maybe we have a blind spot. Maybe it's a mistake that there are some other genes that we have to have in the minimal cell that we don't know about yet.

How does the process of turning on your synthetic cells work? The sort of science-fiction view of this would be, like, mix all this stuff together and then electrocute it or something. What's the moment when it goes from off to on?

You're mixing it in the lab, and at some point you're running around screaming, “It's alive! It's alive!” No, that doesn't happen like that.

We haven't actually made a living cell from nonliving components yet, at least not one that everyone would agree is living. Sometimes physicists say that a lot of the things we're doing already cross some arbitrary threshold to life, but then a biologist looks at it and says, “No, this is too wimpy. This is not living yet.”

Most of the time, it's pretty boring manual labor in the lab. You mix things together, pipetting clear liquids into a tube, and then you wait for it. The way we measure the “aliveness,” quote-unquote, or activity of our cells is just by looking at their metabolism: what proteins they make and what small molecules they metabolize.

The cells are getting better and better at it, but there's never a light-bulb moment when you can say, “This wasn't alive 5 minutes ago and is now alive.” It's not like that, and it probably wasn't like that during the origin-of-life processes either. It was probably—if you had a front-row seat watching the origin of life, it wouldn't be anything to write home about.

It's not that, at some point, the mixture becomes alive and starts doing something. It's a very gradual process. You blink, you miss it. It's really difficult to pinpoint the exact moment when a mixture of chemicals becomes alive.

I have the sense, though, that there are some things—and maybe this is just wrong—but one of my sayings about AI is, “AI defies all binaries.” Yet here I'm about to say that I feel like there are some things that I would look at under a microscope and see either: there is some membrane that contains stuff, or there isn't a membrane that contains stuff.

So I suspect that there may be a fundamental gap in my knowledge that's leading me to try to impose these binaries. But how should I think about things like whether there is a membrane or isn't one, whether transcription is happening or isn't, or whether there are middle states on those questions?

Those things are definitely binary. You cannot have a little bit of transcription without having transcription. I guess you can have a little bit of transcription, but then you have transcription. You can have a little bit of translation, but that counts: you do have translation.

The same is true of the membrane. The presence of a membrane is definitely not a hallmark of life because we have liposomes all the time that are not alive. Transcription, translation, feeding, growth, even replication—we have all those things already.

But I think people are waiting for something that will be more convincing, something like Darwinian evolution, for example. I think a lot of biologists are waiting for the synthetic cell to become capable of Darwinian evolution, and it's not right now.

The functional hallmarks that you list are definitely there. We can have metabolism, we can transcribe, we can translate, and we can even replicate the genome. So all those things that you'd normally think of as hallmarks of life—we can do a lot of those things.

Just to make sure I understand this correctly, you start with an empty test tube. You gradually add ingredients, each of which is a single known—

Yes.

—ingredient.

They're all known because we purified them or synthesized them.

And then this concoction spontaneously forms membranes—

I—

—starts transcribing genes—

Yeah.

—starts metabolizing, starts replicating.

Wait, I don't like the word “spontaneously” because it implies some kind of magic or a heavenly finger, whatever. It's spontaneous in that it's predefined by the chemical and physical properties of those molecules.

For example, if you put lipids into a water solution at the right pH, those lipids will form membranes. There's nothing magic about it. That's just the physical and chemical property of those lipids. The same is true of transcription and translation. If you put all the enzymes and small molecules together at the right concentrations and the right pH, they will do their job. That's what they are.

That’s their whole thing: if you’re an RNA polymerase, if you happen to find DNA with the right promoter sequence, you will transcribe it, because that’s what you evolved to do. And so the “spontaneous” in that sentence is that all those molecules behave in those ways, which seems amazing: that’s exactly what they’re supposed to be doing.

Yeah. And then the limit right now is—I think I heard you say once that replication is basically too consistent. It’s too high-fidelity. There’s not enough random variation introduced to support evolution?

Yes.

Secret’s safe with me.

Unfortunately, the polymerases that we use to replicate genomes right now are annoyingly high-fidelity. They don’t make mistakes, or they make very, very few mistakes. And in order to have Darwinian evolution, you have to make mistakes, because that’s how you introduce variety into your population. We’re not supposed to say “diversity” anymore. That’s how you introduce variety into your population.

And all the millions of your viewers.

That’s the problem with replication right now. We can replicate the genome and it works great, but in order to evolve, you have to make just the right amount of mistakes. You cannot make too many mistakes, because then you just devolve into an error catastrophe. If you don’t replicate your genome reliably enough, then you just don’t have a functioning genome anymore.

But you have to make just the right amount of mistakes. You have to make enough mistakes to introduce diversity and variety in the offspring, but it still has to be reliable enough that most of your offspring will still be viable. They can be a little bit different from you, but they still have to be viable. The balance is very careful. It’s really difficult to reach that balance, and very difficult to engineer it with the enzymes that we have at hand right now. So that’s why we don’t have Darwinian evolution in synthetic cells right now.

So where did this system come from that does the replication, that doesn’t make any mistakes? What is preventing you from using one that does make a few mistakes?

Yes. The enzyme we’re using right now, the polymerase, came from a bacteriophage, and it’s a very high-fidelity polymerase. It’s great in that it’s high-fidelity. It’s very difficult to engineer an enzyme to be just a little bit broken. I could, no problem, engineer that polymerase to make a ton of mistakes, but if you want to engineer it to make just this tiny amount of mistakes that we need for evolution, that’s very difficult.

The same problem is with all the other DNA polymerases that we can use in vitro: they’re usually very high-fidelity. The ones that are lower-fidelity are very error-prone. They also have uses—people use them to make libraries for selection—but they’re too error-prone for the small genome that we have.

So we need something with just the right size of error-proneness, if that’s even a word, and that’s difficult, partially because the genome that we have is very small. Normally, when you have a bigger genome, a low error rate is still fine. For example, let’s say you have 1 error per 100,000 bases. If you have 1 error per 100,000 bases and your genome is a couple million bases, then you will have a few errors per replication cycle, and that’s enough to evolve.

But if your genome is as small as my genome—about 90,000 bases—then that means 0 to 1 error per replication cycle, and that’s not enough to evolve. So that’s why we can’t get a good polymerase that would work for us. What we need doesn’t exist in nature, because no organism in nature is simple enough to have a genome that small.

And so what are we working with? We have something that has more errors, but then we have error correction?

Both. In our own bodies, the polymerases that replicate the human genome are very high-fidelity, but we also have error correction. Our giant genome gets replicated with few mistakes every time you replicate it, but there are also enzymes that come back and do error correction—that check it. We don’t have error-correction enzymes in synthetic cells either.

Yeah, gotcha. Okay. Fascinating.

We also make a ton of errors or other changes in each replication cycle. You can see that your kids look different from you, and that’s because of all of this diversity and recombination of different genes.

Okay. So I appreciate the 101, and I could ask lots more questions, but let’s turn our attention now to mirror life. What is it, and why has it been something that people have been particularly intrigued by relative to all this other interesting stuff that’s going on?

The idea of mirror life is that all the molecules in biology have a certain chirality, a certain stereochemistry. That means every molecule that’s not symmetrical can exist in 2 different forms, and a good example of that is your own hands. You have 2 hands, hopefully, and both of them are mirror images of each other.

The same is true for every biomolecule. Let’s say this is my left hand. My left hand is normal life. There’s a mirror image of my left hand—that’s my right hand. The same can be true for every biomolecule, every complex molecule that makes life.

So you can imagine making a cell that has all the same molecules as a normal bacterium, but instead of them being in the right stereochemistry—the stereochemistry that’s natural and used by the rest of life—they could be in the opposite stereochemistry. They would have DNA, RNA, and proteins that are chemically nearly identical to the natural molecules, but they literally are the mirror images of the natural molecules.

The reason for doing that was, well, one, because it would be cool. It would be an incredibly interesting example of engineering. If you can take the same molecules but with different stereochemistry, can you make a living cell? And if you can, then it would be the first example of a living cell that’s not on the same tree of life as the rest of biochemistry.

That’s one reason, but the work was really motivated by practical applications. We wanted to make better drugs, better therapeutics, and we also wanted to make better biomanufacturing platforms.

The property of mirror life that would allow us to make better therapeutics and better biomanufacturing is also now the reason we think it should not be created. Mirror life would be invisible to both the immune system and viruses and predators in the environment.

For example, if you got an infection from a mirror cell, your immune system would not recognize it as a living thing that it has to fight. If you released a mirror cell into the environment, nothing would contaminate it, no viruses would infect it, and nothing would eat it. It would be basically like a rock in the environment, because that different stereochemistry means all the chemical detection systems that the rest of life uses wouldn’t work on it. It’s the opposite; it’s a mirror image of it. It doesn’t work.

Just like your left shoe is not going to go on your right foot. It does if you really try hard, but it’s not going to be comfortable.

When we first started thinking about mirror life, the idea was that if we could have this thing that’s stealthy to the immune system, it would make great therapeutics. You could inject it into a human, it would do its thing, and it would cure what ails you.

And then if you made a bioreactor made of bacteria that are a mirror image, no phages would contaminate it. There would be no danger of losing your entire million dollars’ worth of bioreactor contents because something gets contaminated—someone sneezes into it. It’s not going to get contaminated, because mirror cells are not compatible with normal bacteria and viruses.

But then we started talking to ecologists and immunologists, and we realized that those features of mirror life are actually big bugs. It would be really irresponsible to release something into the environment or into a human body that’s so stealthy to the immune system, that can fly under the radar, because then it could just go and replicate, and it would not be stopped by any of the mechanisms that normally make sure that life is regulated, that no single organism proliferates out of control.

That was a roller coaster of emotions for me for a few years. It was, “Yay, mirror life. This is an amazing technology. It’s going to solve a lot of problems that biochemistry and synthetic biology have.” And then I was like, “Oh, wait. I actually should not be doing it, because the concerns that we have about potential interactions with the environment and with the human body really outweigh the potential benefits of that technology.”

And, okay, again, lots of follow-up questions here. Obviously, this is where some of the structure of this analysis becomes pretty evidently related to a lot of the discourse that’s going on in AI right now.

Just to make 100% sure that folks have this chirality idea clear, I always find it helpful just to say that even if your hands are exactly the same shape, there’s no way to make them occupy exactly the same space.

Yes.

And you can hold one up to a mirror and see its mirror image, but there’s no way to actually convert it into that mirror image or, again, to have it occupy the same space as that mirror image.

Yes.

This is a strikingly interesting fact unto itself: all life has the same general chirality as part of having the same biochemistry.

Of course.

You can imagine mirror DNA, mirror ribosomes, whatever, but we only have one version of that throughout all of biology.

Yes.

Yep.

Okay. I think there are interesting questions at the epistemic level and also at the societal decision-making level here, for sure. We've got some technology where we're like, “Oh my God, this might be a great way to create new drugs. The upside could be incredible.” And then we've also got, “Oh, but it's all dual use,” and that's another huge theme in AI, right? If it can evade the immune system for good, it might also evade the immune system for bad.

Yeah.

How much do we really know about the ability of these things to evade the immune system? Because one super-simplistic response to this would be like, “Well, sure, I can't make my two hands occupy the exact same space, but one hand can grab the other.” My immune system seems to be pretty robust to all kinds of crazy stuff that it hasn't seen before.

Yeah.

Do we have experimental reason to know that it's not going to be able to grab onto this reverse chirality, or is this theoretical? Where are we in terms of our confidence on all this stuff?

It's mostly theoretical. The immune-system part is mostly theoretical because no one ever made a mirror cell, thank God. Hopefully nobody ever will. But we don't have a mirror cell to give to a mouse and see how the immune system will respond to it.

We do have data from opposite-chirality proteins, and we know those guys are stealth to the immune system. That's why they're such great potential therapeutics. There are also mirror-image nucleic acids that are already in clinical trials as therapeutics. They're called spigomers, from the German word for mirror.

Those mirror nucleic acids are already in clinical trials because they are stealth to the immune system. People smarter than me who understand how the immune system actually works figured out that, by transitivity of properties, if mirror proteins are stealth to the immune system and mirror nucleic acids are stealth to the immune system, then a whole organism made up of those things will also be stealth to the immune system.

Okay, that's interesting. I had not seen this molecular-level evidence that the immune system can't detect these things. When you say “stealth,” can you unpack a more literal understanding of what it means to be stealth to the immune system?

Yeah.

What does it mean to be stealth to the immune system?

It just means you can exist in a bloodstream and no one is going to bother you. So you can inject it into the bloodstream and it will circulate in the bloodstream. You can inject it into tissues and it will circulate there, and the immune system will not activate.

I'm sure it's going to get sniffed by white blood cells, but it's not going to cause activation of the immune system. So you're not going to get sick or get an immune response to that.

So the normal situation is that if I all of a sudden introduce some random foreign RNA into my blood, the immune system detects that as a foreign object, mobilizes against it, and tries to get rid of it.

Yes.

And if we simply reverse the chirality on that same exact sequence of RNA, it just does nothing? It doesn't react to it at all?

It sniffs it, but it doesn't activate. That's why they're good therapeutics. That's why they're already in clinical trials because they don't cause an immune response.

It's kind of like when a dog walks around and sniffs a piece of bread on the floor: he's going to activate and eat it immediately. If he sniffs a rock, most dogs will keep going. Some of them will eat rocks, but that's another problem.

No, the immune system needs to recognize things as a possible threat. If it's mirror biochemistry, it just doesn't activate. It sees it, but it doesn't react to it, and that's a problem because if you have a cell that's made full of mirror molecules, then you can imagine that cell just growing in the bloodstream because your bloodstream is just a warm little pond, a bioreactor that can comfortably provide an environment for that mirror cell to replicate. You don't want to have a foreign cell replicating in your bloodstream.

Yeah. You mentioned earlier that everything in biology tries to eat everything else.

Yes.

And one of the things I often say is, “There's nothing more dangerous than something that nothing can eat.”

Exactly.

And that's what mirror life is.

Is there any deep history—not necessarily an analogy, but a sort of precedent—for things like this happening? The one I always go to, again without deep knowledge in the field, is the oxygenation of the atmosphere. I always find it striking that there once was no oxygen, then somebody started releasing oxygen, and a lot of things died off because they couldn't deal with it.

That's not even a form of life. It's just somebody's waste product. But are there any historical examples of things popping up that nothing could eat, which changed ecosystems on vast scales?

Other than the Great Oxidation Event, we don't know of anything. That probably happened a lot at the very early stages of life, at the origin of life. There were probably organisms popping up in the primordial soup that no one could eat, and then they just took over.

But ever since biology started, there was always this equilibrium, and we don't know of anything else that would just appear that had absolutely no natural predators. I love the analogy to the Great Oxidation Event because even the cyanobacteria themselves got screwed by it. They started releasing the oxygen, but it was toxic to them, too. So there was a great extinction event not only among other organisms, but also among the culprits—the ones that actually started it.

There's never been another example like that. And I don't know of any example of anything in nature that would evolve with absolutely no means of controlling its replication: no viruses, no predators, nothing eating it, nothing limiting it.

Humans are the closest. Really, humans are the closest example to that. We replicate uncontrollably, and we don't have natural predators right now. The only thing that kills us is ourselves, and we still have things that regulate our population. We're still susceptible to diseases.

But you can see what happens when an organism evolves on a planet—an organism that climbs to the very top of the food chain and ends up with no natural predators. It takes over the planet, and then the next thing you know, there's global warming and the planet's dying.

So that's not good, and that probably shows that nature needs balance. Nature really needs things that can stay in equilibrium. And that's one of the reasons why mirror life is such a terrible idea: it would not obey those laws of balance. It would not stay in equilibrium.

Okay, so let me try out a few skeptic arguments, and you can maybe tell me if these are arguments that anybody in the actual field has put forward or if I'm just inventing them. If I miss any, you can tell me what other arguments were maybe more compelling than the ones I came up with.

Mm-hmm.

The first one is just like, okay, you make a good point that these molecules are stealth to the immune system. That does sound a little spooky, but the actual mirror life is still a long way off, right? So do we really need to get too worried about this now?

I'm not entirely sure exactly where the rubber hits the road here, right? There is ongoing work with mirror molecules, and then there's this concept of mirror life. There's a significant distance between the two. So somebody might first of all say, “Well, it sounds like a long way off.” Is this not just much ado about nothing? We don't know how to make mirror life anyway, right?

And that's the best argument. Honestly, you hit the nail on the head. That's the best argument I've heard this entire time we've been working on this project: there's enough problems in the world right now. Why worry about something that's so far off?

Some people compare it to worrying about the sun dying. Yes, we know one day it will happen. It's inevitable, but that's not exactly something that has to affect the stock market and my retirement account. It's not going to happen anytime soon, so let's not worry about it. I don't walk around worrying about the sun dying.

Mirror life could be seen in a similar vein: it's so far away that we don't have to worry about it. The thing about mirror life, though, is that we were actively working on making the damn thing. We wanted to make it. And even if we, as the people who were actually actively working on it, stop doing it without talking about it in public, other people might still find the arguments for making it appealing.

If we don't really publicize the potential concerns, then someone might think, “Oh, it's a great idea. Let me go work on it.” And then we'll get closer to it, and closer, and closer, and one day the sun will be dead. The argument is that we cannot not talk about it because the science was going in that direction, and it was almost inevitable.

Most people in the field agree that, at the current rate of progress, we will reach a fully self-replicating mirror cell within a decade. Some people say 20 years, but it's still possible within a human lifetime. So we have to stop it now, before we actually have the capacity to do that.

What was your own personal journey or evolution of thought on this, as somebody who was enthused and actively working on it? Did you have a eureka moment? Did somebody come to you and say, “Hey, I think you should reconsider this”? What's the personal narrative?

Yeah. It wasn't a eureka moment. It was more like a holy-crap moment, because I was really excited about this idea. This was actually one of the first grants that I got as an independent PI. I had just started my lab, and that was the project that I got really excited about and really wanted to do.

Around the same time, I started doing more work in the biosafety and biosecurity community. I started talking to different people about potential concerns—not about mirror life specifically, but about other emerging technologies. Emerging technologies are things that are on the horizon, that are not quite here yet but will one day soon become possible. We know already that they will become possible, and we have to be ready to face the consequences of those technologies. It's a kind of thinking-before-we-make-it approach: safeguarding a field before it becomes a reality.

I started doing more work on that and started talking to people about mirror life. I'm not an evolutionary biologist; I'm not a biologist at all, and I'm also not an immunologist, so I wasn't aware of those possible concerns. Now it seems pretty obvious to me that, yes, if nothing eats it in the environment, then you should not be releasing it. But that wasn't obvious to me when I first started. I thought that perhaps there could be some mechanisms, some kind of safeguards, we could build to make it safe.

I was maybe trying not to think too hard about the possible downsides because I was so excited about the applications. As I started learning more about all the regulatory mechanisms that keep other forms of life in check, and about the balances that we absolutely need so no single form of life takes over the population, I realized—gradually—that I really should not be working on mirror life.

It was kind of a bummer, because I really wanted to do it. I really think it's a fascinating research project, and the potential benefits are really appealing. But the concerns really outweigh the benefits, so that's when I became convinced. It wasn't that one day I woke up one morning and said, “Okay, I'm not doing mirror life anymore.” It was more of a gradual coming to that understanding.

Yeah. A couple of other—not necessarily counterarguments, but sort of reasons we might not feel compelled to worry about this. You alluded to one, which is, “Well, maybe we can just contain them,” right? It's one thing to build it and do experiments in a lab, and then it's quite a different thing to release it into the wild.

Yeah.

Why don't we just keep them locked up?

Because if you make a foolproof container, then nature will develop a better fool. There's always—I think it's one of those Murphy-like laws—that if you make something idiot-proof, the universe will provide a better idiot.

There is no foolproof containment. There is no way to absolutely safeguard something, and that's for 2 reasons. One is that we cannot predict everything. If I make a mirror cell in my lab and put 27 locks on it, there will be a hole in the floor and the sample will leak out. We cannot predict everything in advance.

If there's something that we don't want released, then the only safe way of preventing the release is just not having the thing—not making it. That's the safety argument.

Unfortunately, there's also a security argument. If you have a mirror cell, even if you create the most perfect, advanced containment, you cannot safeguard against a bad actor—someone intentionally trying to release it. There are many crazy people in the world, and there are people who might want to do harm regardless of the consequences.

If mirror life were to exist, even if it were perfectly contained in the lab, you can imagine someone with bad intentions gaining access to it, either by force or by some kind of subterfuge, some kind of scam, and then stealing it and releasing it. So the only way to avoid that—to prevent it from happening—is just not to have a mirror cell in your lab to begin with.

Yeah. For what it's worth, I do think the “We'll just keep it under wraps” argument is not compelling at all, just given the obvious history of lab leaks. It's not a short list.

Yeah.

On the AI side, going back—I don't know if you've ever heard of this—to Eliezer Yudkowsky, who's sort of the prime mover in the AI safety worry space. He once ran an infamous experiment where he called it the “AI-in-a-box” experiment. The idea was that you could put the AI in a box, but if you let people talk to the AI, it would convince them to let it out.

Yes.

People said, “Well, no, there's no way it's going to convince you to do that.” And he said, “All right. Well, let's do a bet, then. I'll play the AI in the box, and you win if you're not convinced to let me out.”

He did a couple of these experiments, and at least in some of the cases, the counterparty came away saying… It became a sort of item of lore because part of the deal was that they weren't supposed to reveal what he told them to convince them to let him out. But people did come forward out of those experiments and said, “Yeah, without telling you how, I will say he convinced me to let him out.”

The idea that we're going to just contain something like this forever does seem pretty hubristic on its face. Another one that I find maybe a little more compelling—but maybe I shouldn't—is, won't they be constrained by a lack of raw materials or a lack of a favorable ecosystem?

This has a parallel in AI, too, which you alluded to earlier. These things can only exist on chips, and there are only so many chips. It seems pretty clear that the AIs are not going to break out of a sort of computing environment into the rest of the world.

Could a similar thing be said about mirror life? Don't they need the proper chiral building blocks, and wouldn't those be rare?

Yeah. They don't. There are 2 things. One is, you could imagine building a synthetic mirror cell that is completely reliant on, let's say, 50 different chiral molecules of the opposite chirality, and there is no way to find them in the environment. That would be very safe. You could pour that onto the lawn, and it would do nothing because it could not evolve 50 pathways at the same time in the span of 1 generation.

That would be a way to imagine perfect containment. But then, unfortunately, there is again the problem of security. If we make a mirror cell that's dependent on specific pathways, you can imagine a bad actor engineering those pathways to break our containment. Again, there are a lot of stupid or suicidal people who might want to do harm.

We don't want to create even the possibility of that happening—the possibility of someone having this chassis of a mirror cell that's very safely contained in my lab, but taking that and then engineering it so that it's not reliant on those rare molecules anymore.

You can imagine building a mirror cell that would only rely on achiral carbon sources—basically, carbon sources with no chirality. There are some bacteria that are like that already in the environment. So it's possible to imagine a mirror cell that doesn't need building blocks of this very rare, specific chirality.

If it's possible to imagine that, then you can have someone actually do it. That's why not making a mirror cell is the best way to avoid that risk.

I do agree that it's possible to imagine building a mirror cell that would be very well contained by those things—they're called auxotrophies, pathways that you engineer so that an organism is reliant on some rare molecule. But then you can imagine breaking those containments, either accidentally, which would be very unlikely for a mirror cell, or intentionally, which is a huge security risk. That's why not making a mirror cell is the best way to avoid that risk.

Yeah. Jurassic Park remains strikingly relevant. Life finds a way.

Exactly. It finds a way. Jurassic Park was a dumb idea, because how do you make something dependent on a single amino acid that's very popular and very common in the environment? That's not a good auxotrophy. A first-year bioengineering student would tell you that's not a good idea. But obviously, then you wouldn't have a movie, so sure, eat your beans.

But the principle is the same, yes. The auxotrophy is basically what they try to do to those dinosaurs: make them unable to synthesize a certain molecule. Obviously, we pick molecules that are much harder to get in the environment than lysine.

Yeah, you always have to be mindful about not generalizing too much from fictional evidence. That's another great Eliezer lesson. But, yeah, let's—

It's our imagination, though. You don't know what's possible unless you can imagine it.

Maybe one more. Okay, let's imagine that this thing can get out. Maybe it can break out of these various types of containment: the physical barrier and also the dependencies that we might try to engineer into it. Then somebody might say, "Well, okay, sure, even then, but won't this still be a massive underdog against all the rest of life? I mean, we've got all this diversity, all these species, way more biomass, and we're evolving too, right?"

Yeah.

Maybe it'll do some damage, but won't normal-chirality life find a way to outcompete it in the end anyway?

It might or it might not. That's the problem. It would definitely be an underdog. It wouldn't even be a turtle; it would be a slug. It would be something incredibly slow, but it would just keep going. It would be unstoppable because nothing would eat it and nothing would infect it.

Personally, if I were to bet my money, I would bet on normal-chirality life. I would say we would find a way. Life finds a way. We would find a way to eat it somehow, but it's a risk that I'm not willing to take. It's very hard to deal with the consequences of life not finding a way fast enough.

Because we've never had opposite-chirality life in the environment that we know of, there are no preexisting mechanisms to deal with it. Even though evolution, over a long enough period of time, might do its job and develop a mechanism that would allow us to deal with it, it's a risk that I don't think we should be taking. It's asking a lot of evolution, and evolution cannot be programmed. Evolution does what it does stochastically. Either we would find a way to eat or infect mirror life, or we would not. If we don't, then I don't want to think about the consequences. I would rather not do that.

Yeah.

So you're right that it's possible to imagine that mechanism, but it's not safe enough to rely on it.

Yeah, that makes sense. Does this cash out for you as a comparison to AI risk discourse? Is there, conditional on mirror life, a P(doom) that you would assign to this scenario?

The difference between mirror life and AI is that mirror life doesn't exist right now, so we can actually stop it. AI is already out of the box. There is no way to stop it, and frankly, I kind of don't want to stop it, because I haven't written a line of code in over a year. Any time I need a script, I just ask ChatGPT, and it writes me code. The code mostly works, and if it doesn't, I just paste my error back in and the script says, "Oh, sorry, I forgot to close the line," or something.

Anyway, we rely on it, and it's not a bad thing. We rely on AI for many things, and it does make our life easier in the end. It also speeds up the discovery process for a lot of different fields—for biomedicine, for medical discovery, for biotechnology. There are a lot of lab applications that rely on AI already.

There's no way to put it back in the box, and I wouldn't want to put it back in the box. I don't know how to safeguard that. There's obviously a big problem of how you keep that contained, but it's a technology that we cannot live without anymore. If we had those conversations 20 years ago, then maybe there would have been a way to stop the development of AI, and then we could actually have had those conversations: Is it actually net positive for the world?

But it's here already. There's no way to stop it. You're not going to stop it. All we can do is try to develop ways to safeguard it, to develop ways not to control access exactly, but to control the applications, control what it can do, and also be prepared for the negative consequences.

With mirror life, we can say, "No, we're not making it," and hopefully it will never happen. With AI, we've already made it, and we have to—

Well, I would say we've partially made it. The analogy is not meant too literally here, obviously, but we have mirror molecules. We don't have mirror life.

Yeah.

I would say we have some sort of intelligence in our artificial intelligence. It's pretty clear to me that there is—without getting bogged down in definitional debates about intelligence, I count it. I feel like I know it when I see it, and I feel like I see intelligence.

What I don't yet see is robust autonomy. Of course, people are trying to enumerate the key features that a more dangerous AI might have: robust autonomy, the ability to escape from its server and copy itself onto other servers. There are interesting questions around how much of this depends on evolution. AI is primarily developed by optimization as opposed to evolution. Does it need to have some sort of evolutionary mechanism?

To some degree, it can just kind of code its own evolutionary algorithms. Google came out with a really interesting paper around hybrid strategies using language models to generate programs and evolutionary algorithms. The language models can code their own evolutionary algorithms, I think, pretty safely at this point.

Huh.

But I think the key point that I'm interested in is that, in the same way that mirror molecules seem to pose very low risk and might be very much to the good, you can remain enthused about those while warning against a sort of gain of function to the degree that it becomes mirror life.

I think the AI field needs to do something similar now. All this stuff has worked well beyond anybody's wildest dreams—from, well, all but a very few people's wildest dreams—not very long ago. Now we're entering a different regime. At one point in time, it was, "Can you get anything to work? Can you get this thing to tell a cat from a dog?" When it was so simple, who cares what technique you used or whatever? Anything serious was a long way off.

Now I feel like we're in this zone where the serious stuff is not necessarily so far off. That's still debated, of course, but it seems like it's becoming more incumbent on the field to really ask the question: What kind of AI do we really want to have? We're going to have some.

Mm-hmm.

It's going to be powerful, and hopefully it will make life lots better. But there are some kinds that I think everybody looks at and says, "Geez, if you checked these 3 or 4 more boxes—if you had something that could cyber-hack its way off of its servers and could somehow scrounge up resources in the wild or make money on the internet to pay for its own compute—then these things do start to seem qualitatively different in terms of how unwieldy they are and how prepared the broader ecosystem is to contain them."

I don't see nearly enough of that kind of stuff happening in the AI space. It used to be that nothing worked, and anything you could get to work was a big accomplishment. Now I feel more often like everything is working, and there are a million doors that we could open and a million different paths that we could go down. What seems to be in relatively short supply is the critical analysis of which ones are the right ones to choose and which ones we will potentially regret.

And how do you make that choice? How do you pick which paths?

Yeah, I think that there are some things that hopefully would be fairly obvious, but I just don't see people putting a lot of attention on them. That raises a question for you around the experience of this. You've been on this long research journey. You've come to this conclusion yourself: "Okay, you know what? We shouldn't do this." But then you didn't stop there. You actually went out to others in the field and said, "Hey, we should stop this."

Maybe tell that story of how you went from one person—or a small group of people—coming to this conclusion to recruiting and putting together a coalition, and putting a big headline statement out there that you think this really shouldn't be done.

It was a rather slow process. It started with a rather small group of people talking about it, and then we realized that, in order to really make a statement, we had to have a lot of different stakeholders on board. That included not just the people who actually work on engineering mirror life, but also people who know anything about ecology, immunology, and policy, because the researchers who work on it are not experts in every field.

So we had to put together this really broad coalition of people. That started by approaching people and saying, “Hey, there’s this project we want to talk to you about. What do you think about making a mirror cell?” A lot of the people were very surprised when we first approached them because they had never even thought about it. A lot of the authors on the original Science paper were people who had not thought about mirror life at all before.

But when we came to them, we showed them the evidence, we showed them where the research was, and what we thought could happen. We basically approached a lot of people with the question, “Can you please prove us wrong?” We really wanted to be proved wrong. When we realized those concerns about mirror life were really serious, we were sincerely hoping that someone would poke a hole in our reasoning.

So we approached several people saying, “This is what we’re thinking. Could you please find a way in which we are wrong?” They couldn’t, and they ended up being authors on the paper because they were on board. They sincerely could not find any argument that would point to either why mirror life would not be a concern or why mirror life would be impossible, something you shouldn’t even think about.

That’s how it went. We were slowly on the lookout for people who could disprove our theory or mitigate our concerns, and the more we looked, the more we were unable to find anyone. Eventually, the team was big enough that we wrote up the paper and then started the public discussions.

And I’m still waiting. That’s an honest, sincere appeal to anyone watching this: If you can prove us wrong, if you have actual evidence for why mirror life would not be a concern, the community would be extremely grateful for that. We’re scientists who like cool projects, and we still would like to be able to make a mirror life. We just don’t think there is a way to do it safely right now.

Were there any holdouts? Was anybody—especially anybody who was, of course, in the classic situation where somebody’s paycheck depends on them not understanding something? That’s a pretty good leading indicator of them not understanding it. Was there anybody who was working on some aspect of this that you couldn’t bring on board with the conclusion?

Not so far. Maybe we were lucky that we only approached the people we knew would be reasonable. But no, there was no one we approached originally who would say, “No, I’m not going to—” There was no one who would join our team, work with us, review all the evidence, and then say, “No, I’m out of here because I can’t publish this,” or, “I don’t think this should be done.”

Maybe we’re just that convincing. We’re nice. We smile. But, yeah, no, I think it was the strength of the arguments. There were some people who joined the team and were convinced by the argument but had some other points of view, other angles, that ended up being represented in the technical report that accompanies the original Science paper.

That broadened our perspective. There were some things that we didn’t think about originally that ended up being part of the core message. So people definitely changed our reasoning and changed the direction of the group, but there was nobody we brought on board who then said, “I don’t believe you. You’re wrong. I’m out of here.”

What sort of evidence would be compelling? If you showed, for example, “Here’s an example of an immune system quickly responding to mirror molecules,” that sort of thing would be compelling. It sounds like that experiment has been run and didn’t turn out that way.

Yes.

Are there other experiments that you have in mind that you would find compelling, that people should do to interrogate this further?

I think there is some room in the countermeasure research. For example, right now we know that mirror-image antibiotics would work on mirror life, but we don’t know the extent of it, and we don’t know the exact effects of mirror antibiotics on other organisms.

There’s also research on the availability of food sources. I said earlier that it’s possible to rely on achiral—so, not chiral—carbon sources, but we don’t know exactly how well that would work. We don’t know exactly how promiscuous that metabolism could be.

There are some experiments that would be interesting, but I would be very careful not to go too far down that route because you don’t want to get too close to actually making the thing that you don’t want to make.

Now that this has happened and you guys have put out this headline-making warning, how do you think about the impact or the mechanism? You went to all this effort, and now the hope is that, at least until some major evidential update, mirror life won’t be made.

Is that a sort of taboo among scientists, where everybody agrees, “We won’t do it”? Or is it about influencing funding structures, so the grant evaluators know to be on the lookout for this sort of thing? How do you think about the mechanism between the declaration itself, or the warning itself, and the actual reduction in risk?

It’s still very early. In the policy world, 6 months is nothing, and that’s how long it’s been since we published the paper. My personal hope is that both of the mechanisms you named will happen.

One is that no self-respecting scientist will want to touch it, just like with germline editing in humans. There was one example when someone tried to CRISPR a human embryo, and that was so comprehensively shut down by both government and community pushback that hopefully no one will try to do that anytime soon.

I’m hoping the same will happen with mirror life work: The community will be on board, and no one will start working toward it within the community. But I also hope that there will be regulatory mechanisms at the funding and policy levels. No one will fund this work, and this work will become something that we’re just not allowed to do. So, all three.

It’s funny that you mention human germline editing. That one feels much more compelling to me on the upside. These mirror molecules as therapeutics could be interesting, but I feel like we have lots of different avenues that we can chase down to make better therapeutics. This one seems to have outsized risk attached to it, so let’s go down 99 out of 100 other doors and not this one.

When I think about editing the human genome, though, I can imagine a dystopia where it gets out of control. But at the same time, if we could eliminate a lot of rare diseases, that would just be a great win for everybody. Then you get into some slippery-slope territory, but it might be sort of—

Yeah.

—a good slippery slope at first, right? I was recently reading some analysis where it was, if you could make not very many edits, you could dramatically reduce the risk of Alzheimer’s, and you could dramatically reduce the risk of heart disease.

I wanted to voice for a second and maybe get your reaction on how we make sure we don’t throw too much good out with the bad here, because that’s certainly on the AI side. People are dreaming of utopia. I think it’s increasingly plausible, and I want it, too, so I don’t want to dismiss or, more to the point, miss out on that upside.

I feel like in the human-editing space, I can imagine a dystopia over multiple generations, but I can imagine a next generation that is just plain a lot healthier. Should we deny that next generation that benefit because we’re worried about what 3 generations down might do or look like? How do you think about just—

Yeah.

—the risk-reward on some of these less obvious cases?

I’m really worried about the slippery slope. If we start doing editing, we will not stop, and that can get out of hand really quickly. Another thing to think about is that it would significantly increase the inequality that already exists in the world.

You can imagine that rich people who have access to this technology would be producing offspring that are genetically as perfect as we can make them, and then the people who can’t afford that technology would just be breeding the old-fashioned way. Right now, we already have a huge amount of inequality in the world, and some of it is already biological, because if you grew up with good nutrition, that sets you up for life, and someone who grew up undernourished will never catch up in development.

But if you start doing genetic editing on…

Everyone will go to the absolute farthest length for their own children. Honestly, as a human, if I could make my kid the smartest, fastest, healthiest one in the world, I totally would. I absolutely would, and I think everyone who says they wouldn't do it for their own kid is lying.

It's just a natural instinct that we want to give the best to our kids, and that worries me, because if that technology becomes available, then there will be no way to stop it. It would not even be reasonable to ask people not to use it. But then not everyone will have access to it.

You can imagine quickly creating, in 1 or 2 generations, almost a new race of humans that will be superior in so many different ways that we, as a species, will absolutely be screwed. So that's the worst dystopian case that I'm worried about, and the reason why I think we should not be working on human germline editing at all is because once that technology becomes available, it will be impossible to stop it.

Frankly, I don't know if I could resist if someone offered me the ability to edit my own embryos so that my children had, for example, much lower Alzheimer’s risk. I don't think I could resist, honestly. Even though I rationally think that technology is wrong, as a human, I would do it because that's just my instinct as a mother to make sure that my kids get the best.

It would be normal for every human to react the same way. So if we have this ability, we would take it because it would be impossible to ask people not to take it. The only way to safeguard against it is to not have that ability at all.

It might be naive because the technology is almost already out there, and I think one day that might exist. Once it exists, I think we'll have to focus on distributing it equally so everyone has access to it. Although we know that, as a society, we really suck at distributing things equally. There is so much inequality right now that it can only get worse.

I think the example you used with AI is a little different, because the way I understand it, since it's technology, it's easier to distribute. If you have AI that gives you certain advantages—for example, if I have an AI that allows me to be a better writer—you can imagine distributing that all over the world to everyone who has access to the internet. It would be much harder to put that behind a paywall, while actual genetic manipulations would be hard to distribute equally.

I'm sorry. I realize I sound like a bummer, because it's a great technology that could really benefit humanity, but I also don't know if we're smart enough to make it benefit everyone equally. It's not even some altruistic, do-gooder idea that everyone should have equal opportunity. It's just self-preservation instinct.

If we actually do create 2 tiers of society biologically like that, then there's going to be too much tension, and I can't imagine that going well for the stability of our society. So, yeah, let's talk about something nicer, because I don't want to wrap up on this doom and gloom of everything being terrible and us, as humans, not being responsible enough.

I think those are concerns that should not be dismissed. I will give the AI companies a lot of credit, at least thus far. I'm always watching these things closely, and it does seem like the eras can sometimes be quite short.

At this point, it is remarkable. It kind of reminds me of the old Andy Warhol quote about what makes consumerism and American consumer society great: you can have a Coca-Cola, and it's the best Coca-Cola there is. No amount of money can buy a better one. The president drinks the same Coke, right? That is basically true of AI today.

Somebody just posted—Terence Tao, broadly considered, I understand, the greatest living mathematician, recently posted something where he was using Claude, and he was on the free plan. People noticed this and were having a laugh about the fact that here's this guy who is truly, by any measure, elite—maybe literally the best mathematician in the world today—and he's on the free plan of Claude.

Why is that the case? It's because you could pay and get a greater number of uses, but they actually do give the best model out for free. So I think we are—and that may change, because inference scaling and a lot of other things are involved. We're hearing rumors of a $20,000-a-month coding agent. Obviously, not everybody is going to be able to afford that, if and when it does come to market.

But I do give the AI companies a lot of credit, at least so far, for really trying to live up to their ideals around democratizing access and pushing that amazingly far for the moment.

Yes. That's impossible to imagine with biological technologies, because there are labor and material costs that you just can't get around.

But might people have said the same thing with sequencing? I'm mindful that there's a time delay, and it's not entirely fair. I'm not trying to say that it would be entirely fair, but it certainly seems like a lot of progress has come from luxury, high-end, or not generally accessible technologies that gradually become more accessible.

Going back to when I was in college, it was a million dollars or whatever to sequence a genome, and now we're at, what, a couple hundred bucks. So I am cautious around gene editing being an inherently unequal technology, in the same sense that maybe it's not accessible to most people now, but can we come down that cost curve and create a version of society 20 years from now where it is universally accessible?

That's kind of a vicious circle, because in order to bring the cost down, you have to start doing it. So we have to take the leap of faith that we will be able to bring the cost down and make it equally available in order to make that happen. You have to do that on faith.

Yeah. Do you think there's a bright line to be drawn? Just to create a rough structure for this, mirror life and a treacherous turn or fast-takeoff AI are kind of analogous in that these are things where we really just might lose control.

There's genetic editing, and that could get out of control and we could end up in a slippery-slope dystopian scenario. There are definitely AI versions of that too—the gradual disempowerment meme. Recent major papers have made the case that even if we can control all this near-term stuff, we really don't have a plan for how we're going to have a society that we're going to be happy with with very powerful and widely deployed AI.

I take that really seriously, but I do wonder if there is a line to be drawn and how bright it could be between things where, if we open this door, we really might not be able to close it, whereas if you open this other door, we might be worried that collectively we won't close it or whatever, but we could.

It's not like genetic editing is going to truly take on a life of its own. That's more of a social dynamic than life itself doing the thing, right? I mean, I guess we are life itself, but hopefully there's a distinction there.

Maybe there's no bright line to be drawn, but I want to draw a line between things where I can say, “Okay, that can be somebody in the future's problem, and they will still have some recourse.” I want to put genetic editing and what society is going to look like with a lot of AI deployed broadly into that bucket.

Then I want to draw a smaller circle around things like mirror life and hyper-autonomous, goal-oriented AI and say, “Those things seem qualitatively different.” How do you react to that attempt to taxonomize?

That kind of makes sense to me. It makes sense to me intuitively. I'm not very rational when thinking about stuff like that, because I want this to happen. I want this to be true. I have a very hard time drawing lines like that.

But I think we're going to have to, because it's inevitable. So if I could make you the emperor and you draw those lines, I think what you just described kind of makes sense.

It's a lot of responsibility. All right, maybe the last 1 or 2 questions. I'm not sure if these are the same question or a different question.

Okay.

I guess, based on your expertise in synthetic biology, is there anything that you would be telling people in the AI space to be specifically concerned about or watching very closely? Do you have any intuitions for what might be the step-change, gain-of-function moment that could really move the needle on this spectrum from inert, tool-based AI to some sort of out-of-control AI?

Yeah. To me, it would definitely be, within the limited understanding that I have as a biochemist and bioengineer, a model that would be able to generate new biological functions.

For example, if you can go and ask whatever AI is available online how to make a better virus, that's something that right now AI is not capable of giving you an answer to.

You can say there could be safeguards against answering that question, but we also know that it's possible to jailbreak most of those models. So right now, the AI doesn't have the capacity to do that. I cannot just go in and say, “How do I make a better toxin?” or “How do I make a better virus?” But if the model learns to analyze all of the data in the literature, then maybe there is knowledge out there that, by learning from all of the published literature, a model could synthesize all of that knowledge and come up with, “This is how you make a better pathogen. This is how you make a better drug. This is how you make a better toxin.”

Right now, if you ask the best virologists in the world, they'll give you an answer with some possibilities of how to make a better virus, but not a definite answer. No one knows. But if there could be a model that can synthesize all of that knowledge and come up with those answers, this would be a game changer, and not for the better. This would be something that would put a lot of capacity in the hands of people who don't have enough background to be responsible with it, and that's something that I really hope doesn't happen.

I really hope the answer's not even out there to be learned. But if the AI ever learns to be that good, to really synthesize scientific knowledge and predict things like that, I would not be very happy.

Maybe the last one, then. Any thoughts, reflections, or advice based on your social effort to build the coalition and bring people together on this that you think people who are trying to do something similar in the AI space could take note of?

I think carrot rather than a stick. When we first started talking about this Mirror Life project, we were wondering how we should approach it, and we decided that trying to bring people on board, convincing them that this is the right thing to do, the right idea, is much better than trying to top-down advocate for oversight and regulation.

One of the ideas was, why not, instead of going public with it and talking to everyone who wants to talk to us, just go to the government, scare the crap out of them, and make them regulate this out of existence? But we decided this wouldn't really be a good idea, because without the buy-in from the community, it wouldn't work. It wouldn't be effective.

So any time you want to regulate a community of relatively smart, motivated people, you have to do it in a way that makes them really believe in it, makes them be on board. For one reason or another, you have to make them really believe that it's in their own best interest to be on board with it. And I might be really naive, because I've only done one project like that, and we happen to be a pretty friendly community. I don't know how it's going to work with other communities.

But I would definitely say this is how you try it: You try to convince people that this is in their own best interest. You're not trying to take their toys away. You're trying to make sure they keep playing with their toys in a safe way.

Yeah. And exactly what kind of toys we're playing with seems like it could matter an awful, awful lot.

Oh, yeah, and how much money is involved, too. That's one thing: Mirror Life was not an existing therapeutic or an existing technology yet, so no one lost billions of dollars on it. It would probably be much harder to convince people to regulate a technology that already exists and already pays them.

Yeah, no doubt. The corresponding folks on the AI side definitely have a big challenge on their hands. But I think there is still a lot to meditate on from your work and just the overall arc of how it's gone.

I mean, if nothing else, it shows that people can change their minds on things. We've seen examples of that in the AI space as well, that it is possible to build a coalition, that people can be invited to try to prove you wrong, and, in failing to do so, update their own worldviews.

And I think it's also really notable that it is a pretty narrow circle that you've drawn around a particular thing. We're not hearing, “Also stop the mirror molecule trials,” but really trying to be very precise around what exactly the mirror model is and how do we make sure that we contain that, without being—

And I imagine that's going to be much harder for AI, to draw a circle like that.

Yeah.

Yes.

Definitely. But it's no choice but to try, I'm afraid, as we go forward through these next few years.

Yeah.

This has been fantastic. Thank you for taking the time. Anything else you want to leave people with before we break?

I just want to say it's fantastic that we are even having these conversations, that there is a community that cares enough about doing the right thing, regulating it, and doing it safely, that it has to be done. As you said, we have no other choice than to try.

Yeah. No way out but through.

No way out but through, yes. Thank you so much for picking up this topic and these—

And Dr. Kate Adamala, thank you for being part of The Cognitive Revolution.

Thank you so much. Thanks for having me. Bye.

It is both energizing and enlightening to hear why people listen and learn what they value about the show. So please, don't hesitate to reach out via email at tcr@turpentine.co, or you can DM me on the social media platform of your choice.