00:38

Levin

Yeah. Hey, Curt. Great to see you again. I think that one of the biggest myths in biology is that the best explanations come at the level of molecules. So this biochemistry—what they call molecular mechanism—is usually currently taken to be the gold standard of what we’re looking for in terms of explanations. And I think in some cases that has the most value, but in most cases I think it’s not the right level. And Denis Noble has a really interesting set of papers and talks on biological relativity. So this idea that different levels of explanation, especially in biology, provide the most bang for the buck. And I think in terms of looking forward—as to: what do these explanations let you do as a next step?—I think we really need to go beyond the molecular level for many of these things.

01:36

Levin

I think that emergence in general is basically just the measure of surprise for the observer. So I think a phenomenon is considered emergent by us if whatever it’s doing was something that we didn’t anticipate. And so it’s relative. I don’t think emergence is an absolute thing that either is there or isn’t, or is weak or strong. It’s just a measure of the amount of surprise. How much extra is the system doing that you—knowing the rules and the various properties of its parts—didn’t see it coming? That’s emergence, I think.

02:14

Levin

I mean, people have certainly made that kind of ladder. I’m not sure what we can do with that. I think that it’s more like these different levels have their own types of concepts that are useful for understanding what’s going on, and they have their own autonomy. That’s really important: that I think there’s a lot of utility in considering how some of these higher levels are autonomous, and they do things that the lower levels don’t do. And that gives us power; that gives us experimental power.

03:00

Levin

Yeah, it really depends on what you mean by “derived.” But, for example, there are certainly aspects of, let’s say, cognitive functions that we normally associate with fairly complex creatures, brains, and so on, that we are now seeing in very simple basal media. So even something as simple as a gene regulatory network can have six different kinds of learning and memory—for example, Pavlovian conditioning. And so there what you would see is, on the one hand, you say, “Okay, well, look, here’s a simple bit of chemistry that’s actually doing these complex things that we normally associate with behavior science and with investigating cognitive systems.” You can always tell a chemistry story about any effect. In fact, you can you could also tell a physics story as well. If you look under the hood, all you ever find is chemistry and physics, right?

But the more interesting thing about that story is that you actually can use paradigms from behavioral science—such as learning and conditioning and communication and active inference, and things like that—and that lets you do way more than if you were to restrict yourself to do an understanding at the level of mechanism.

04:48

Levin

I think that’s somewhat true, although it’s not as true as we think. So oftentimes you actually can tell an interesting story cashed out in the terms of behavioral science—for example, active inference; for example, various learning modalities—of objects that are really thought to be the domain of physics and chemistry. So I think there’s more of that than we tend to understand as of now. But it is also the case that the fact that you can tell a physics story doesn’t mean that there’s that much value in it necessarily.

So, for example, imagine that there’s a chess game going on. So you could absolutely tell the story of that chess game in terms of particle movements, or maybe quantum foam, or I don’t know—whatever the bottom level is. So you can do that. How much does that help you in playing the next game of chess? Almost not at all, right? If your goal is to understand what happened at the physical level—yes, you can tell a story about all the atoms that we’re pushing and pulling and all of that. But if your goal is to understand the large-scale rules of the game of chess, the principles, and more importantly, to play the next game of chess better than what you just witnessed, that story is almost completely useless to you. And this abounds in biology, where you can use chemistry and physics to tell a story looking backwards about what just happened and why, but in terms of understanding what it means and developing that into a new research direction, new biomedicine, new discoveries, it often means that you actually have to tell a much larger scale story about memories, about goals, about preferences, and these other kinds of concepts.

06:33

Levin

Well, the thing that ties all of our work together—I mean, we do a lot of different things. We do cancer biology, we do developmental biology, we do regeneration, aging, not to mention AI and some other things. What ties it all together is really an effort to understand embodied mind. So the center focus of all my work is really to understand cognition, broadly, in very unconventional, diverse embodiments. And in biology, what we try to understand is how life and mind intersect, and what are the features that allow biological systems to have preferences, to have goals, to have intelligence, to have clever problem solving in different spaces.

07:39

Levin

Yeah, what ties it all together is the effort to understand intelligence. All of those things are really different ways that intelligence is embodied in the physical world. And so, for example, when we made xenobots and anthrobots, our goal with all of these kinds of synthetic construct is to understand where the goals of novel systems come from. So typically, when you have a natural plant or animal and it’s doing various things, not only the structure and the behavior of that organism, but also whatever goals that they have, we typically think of are driven by evolutionary past. Right? So these are set by the evolutionary history, by adaptation to the environment. So for eons, everything that wasn’t quite good at pursuing those goals died out, and then this is what you have. That’s the standard story.

So the reason that we make these synthetic constructs is to ask, “Okay, well, for creatures that never existed before, that do not have an evolutionary history of selection, where do their goals come from?” So that’s just part of that research program to understand: where do the properties of novel cognitive systems come from that do not have a long history of selection as that system? Everything else that we do is an extension of our search for intelligence—so basically cancer, right? So the way we think about cancer is as the size of the cognitive light cone of living or of cognitive systems. So what I mean by that is: every agent can be defined by the size of the largest goal it can pursue. So that’s the cognitive light cone in space and time: what are the biggest goals that it can pursue?

So if you think about individual cells, they have really sort of humble, single-cell scale goals. They have metabolic goals, they have proliferative goals, and things like that. They handle the situation on a very small, single-cell kind of scale. But during embryonic development, and during evolution in general, policies for interaction between these cells were developed that allow them to scale the cognitive light cone. So groups of cells—for example, the groups of cells that are involved in making a salamander limb—they have this incredibly grandiose goal in a different space. Instead of metabolic space, they are building something in anatomical space. So they have a particular path that they want to take in the space of possible anatomical shapes. So when a salamander loses their limb, the cells can tell that they’ve been pulled away from the correct region of anatomical space. They work really hard. They rebuild. They take that journey again, they rebuild that limb, and then they stop. And they stop because they know they’ve gotten to where they need to go. That whole thing is a navigational task, right? And there’s some degree of intelligent problem solving that they can use in taking that task.

So that amazing scale-up of the cognitive light cone that, you know, from—the light cone measures: what do you care about? So, you know, if you’re a bacterium, maybe you care about the local sugar concentration and a couple of other things, but that’s basically it. If you’re a set of salamander cells, what you really care about is your position in anatomical space: do we have the right number of fingers, is everything the right size? And so on. And so understanding that scaling of cognition, the scaling of goals, how does a collective intelligence work that takes the very competent tiny components that have little tiny light cones, and how do they come together to form a very large cognitive light cone that actually projects into a different anatomical space now instead of metabolic space? That’s fundamentally a question of intelligence.

And the other thing about it is that this kind of thinking—so this is really important—that this kind of thinking is not just, you know, sort of philosophical embellishment, because it leads directly to biomedical research programs. If you have this kind of idea, what you can ask is: well, let’s see, then. If cancer is that type of phenomenon—it’s a breakdown of that scaling and that basically individual cells just shrink their light cone back down to the level of a primitive microbial cell as they were once, and that boundary between self and world now shrinks; whereas before the boundary was that whole limb, now the boundary is just: every cell is the self—so from that perspective, they’re not more selfish, they just have smaller selves. So that’s really important. Because a lot of work in cancer models, cancer cell behavior from a perspective of game theory, they’re less cooperative and they’re more selfish. Actually, I’m not sure that’s true at all. I think they just have smaller selves. And so—

12:02

Levin

Yeah, I think one of the things agents do is operate—it’s not the only thing they do, but one of the things they do—is they operate in their self-interest, but the size of that self can grow and shrink. So individual cells, when they’re tied into these large networks, they’ll be using electrical cues, chemical cues, biomechanical cues. They’re tied into these larger networks that partially erase their individuality—and we can talk about how I think that happens. But what you end up with is a collective intelligence that has a much bigger local cognitive light cone; the goals are much bigger. And then, of course, you can scale that up. I mean, humans, you know, have enormous, enormous cognitive light cones, and whatever’s beyond that. But that’s the thing: it’s the size of your cognitive light cone that determines what kinds of goals you can pursue and what space you pursue them in.

13:56

Levin

Yeah, let’s see. And just a quick thing to finish real quick the previous thought, which is that the reason I think that these are not questions of philosophy. These are very practical questions of science, because they lead to specific actionable technology. So the idea that what’s going on in cancer is a shrinking of the cognitive light cone leads directly to a research program where you say, “Well, instead of killing these cells with chemotoxic chemotherapies, because I believe that they’re genetically irrevocably damaged, maybe what we can do is force them into better electrical networks with their neighbors.” And that’s exactly what we’ve done. And so we’ve had lots of success expressing strong human oncogenes in frog embryos cells, and then leaving the leaving oncoproteins be, but instead making sure that they’re connected into tight electrical networks with their neighbors, and they normalize. They make nice skin, nice muscle. They do their normal thing instead of being metastatic.

And so that kind of thing, you know, it’s very important to me—all of these ideas, and so we’ll talk about in a minute about platonic space and things like that—it’s important to me that all of these ideas don’t just remain as kind of philosophical musings, but they have to make contact with the real world, and specifically not just explaining stuff that was done before, but facilitating new advances. They have to enable new research programs that weren’t possible before. That’s what I think is the value of all of these kinds of deep philosophical discussions.

So let’s see. With respect to the definition of intelligence—so, for practical purposes I like William James’ definition, which is: some degree of the ability to reach the same goal by different means. And it’s an interesting definition, because it presupposes that there is some problem space that the agent is working on, but it does not say you have to be a brain. It doesn’t say you have to be natural versus artificial. It doesn’t say any of that. It’s a very cybernetic definition. What it says is that you have some amount of skill in navigating that problem space to get to where you want to go, despite interventions, despite surprise, despite various barriers in your way. How much competency do you have to achieve that? And that was his definition of intelligence. Now, I will certainly agree that doesn’t capture everything that we care about in cognition. So a focus on problem solving doesn’t handle play, and it doesn’t handle emotions, and things like that. But for the purposes of our work, I focus on the problem solving aspects of intelligence.

So within that, your point about the Platonic space. So now, to be clear, everything that I was saying before, I think we have very strong empirical evidence for. And so now I’m going to, in answering this question, I’m just going to talk about stuff that I’m not sure of yet, and that these are ideas I don’t feel strongly that we’ve shown that any of this is true. But this is my opinion and this is where our research is going now. I actually think that the Platonic view is more correct. And I know this is not how most biologists think about things. I think that, in the exact same way that mathematicians that are sympathetic to the Platonic worldview—this idea that there are, in fact, in some way, there is a separate world in which various rules of number theory and various facts of mathematics and various properties of computation and things like that, that there’s a separate space where these things live. And importantly, the idea is that they think that we discover those things. We don’t invent them or create them, we discover them, and that they would be true still if all the facts of the physical universe were different, they would still be true.

I extend that idea in a couple of ways. One is: I think that what exists in that Platonic space is not just rules of mathematics and things like that, which are, in a certain sense, low-agency kind of objects, because they just sort of sit there. They don’t do much. I actually think that there’s a spectrum of that. And some of the components of that Platonic space have much higher agency. I think it’s a space of minds as well, and of different ways to be intelligent. And I think that just like when you make certain devices, you suddenly harness the rules of computation, rules of mathematics, that are basically free lunches. There are so many things you can build that will suddenly have properties that you didn’t have to bake in. You get that for free, basically. I think intelligence is like that. I think some of the components of that Platonic space are actually minds. And sometimes what happens when you build a particular kind of body—whether it’s one that’s familiar to us with a brain and so on, or really some very unfamiliar architectures, whether they be alien or synthetic or whatever—I think what you’re doing is you’re harnessing a preexisting intelligence that is there in the same way that you harness various laws of mathematics and computation when you build specific devices.

19:40

Levin

That’s actually even more specific than what I was saying. I mean, that’s interesting. And I think what I was saying that, in general, kinds of minds. So there are different—I don’t know quite what to think about individual instances—but the kinds of minds, you know.

20:03

Levin

Well, I think—yes, I think the space is extremely large, possibly infinite. In fact, probably infinite. But what I was talking about was mainly types of different minds, types of cognitive systems. Now, for individuals, you raise a really interesting point, which is: I call them “selflets,” which are kind of the thin slices of experience that—

20:25

Levin

Selflets. Because you have a self, and then the different slices of it are the selflets, and you can look at it as, you know, kind of like that special relativity bread loaf—you know, slice them to pieces; that kind of thing. So I think what’s interesting about asking that question about the self and what is the through line is: what’s really important there is to pick a vantage point of an observer. Again, kind of akin to what happens in relativity. You have to ask: from whose perspective?

So one of the things about being a continuous self is that other observers can count on your behaviors and your properties staying more or less constant. So the reason that we identify “Oh yes, you know, you’re the same person as you were before” is basically: nobody cares that you have the same atoms or you don’t, or the cells have been replaced. What you really care about is that I can have the same kind of relationship with them that I had before. In other words, they’re consistent. I can expect the same behaviors, the things that I think they know, they still know, and so on. And so, of course, in our human lives that often breaks down, because humans grow from being children to being adults, all their preferences change, the things that they remember and the things that they value change. In our own lives we sometimes change, right? And that’s a much more important change. You know, are you the same person, even if all your material components remain the same, but you changed all your beliefs, all your preferences, would you still be the same person? Right? So I think what we mean when we say “the same” is not about the matter at all, it’s about what kind of relationship we can still have, and what do I expect from you behavior-wise, and so on. And there’s some really interesting—and so that’s from the perspective of an external observer.

Now, the latest work that I just published a couple of days ago looks at: what does that mean from the perspective of the agent themselves? And this idea that you don’t have access to your past. What you have access to are memory engrams—you know, traces of past experience that were deposited in your brain and possibly in your body, that future you is going to have to interpret. And so that leads to a kind of scenario where you treat your own memories as messages from your past self. The idea, then, is that those memories have to be interpreted. You don’t necessarily know what they mean right away, because you’re different. You’re not the same as you were, right—especially over long periods of time.

And this comes out very starkly in organisms that change radically, like caterpillars to butterfly. So memories persist from caterpillar to butterfly. But the actual detailed memories of the caterpillar are of absolutely no use to the butterfly. It doesn’t eat the same stuff. It doesn’t move in the same way. It has a completely different body, completely different brain. You can’t just take the same memories. And so I think what happens in biology is that it is very comfortable—in fact, it depends on the idea—that the substrate will change. You will mutate yourselves, some cells will die, new cells will be born, material goes in and out. Unlike in our computational devices, you’re not committed to the fidelity of information the way that we are in our computation, you are committed to the salience of that information. So you will need to take those memory traces and reinterpret them for whatever your future situation is. In the case of the butterfly: completely different. In the case of the adult human: meh, somewhat different than your brain when you were a child. But even during adulthood, your mental context, your environment, everything changes, and I think you don’t really have an allegiance to what these memories meant in the past. You reinterpret them dynamically.

So this gives a kind of view of the self, kind of a process view of the self, that what we really are is a continuous dynamic attempt at storytelling, where what you’re constantly doing is interpreting your own memories in a way that makes sense about a coherent story about what you are, what you think, what you believe about the outside world. And it’s a constant process of self-construction. So that, I think, is what’s going on with cells.

24:47

Levin

Yeah, yeah, I think that’s exactly right. I think that a lot of what we’re doing now at any given moment is behavior that is going to enable or constrain your future self. You know? You’re setting the conditions in which the environment in which your future self is going to be living, including changing yourself. You know, anything you undertake as a self-improvement program—or conversely, you know, when people entertain intrusive or depressive thoughts—that changes your brain. That literally changes the way that your future self is going to be able to process information in the future. Everything you do radiates out, not only as messages and a kind of niche construction—you know, where you’re changing the environment in which you’re going to live, in which everybody else is going to live. We are constantly doing that to ourselves and to others.

And so that really forces a kind of thinking about your future self as kind of akin to other people’s future selves. And I think that also has important ethical implications. Because once that symmetry becomes apparent—that your future self is not quite you, and also others’ future selves are also not you—that suggests that the same reason you do things so that your future self will have a better life, you might want to apply that to others’ future selves. Like, breaking down this idea that—and I’m certainly not the first person to say this—but breaking down the idea that you are a persistent object separate from everybody else, and sort of persisting through time, breaking that down into a set of selves, and that what you are doing now is basically for the good of a future self, I think makes you think differently about others’ future selves at the same time.

26:37

Levin

Well, what does “better” mean? I mean, I think that, certainly, there are extremely successful organisms that do not have a large cognitive light cone. Having said all that, you know, the size of an organism’s cognitive light cone is not obvious. We are not good at detecting them. It’s an important research program to find out what any given agent cares about, because it’s not easily inferrable from measurements directly. You have to do a lot of experiments. So assuming we even know what anything’s cognitive light cone is, I think lots of organisms do perfectly well. But then, what’s the definition of success?

So in terms of the way we think about—well, the way many people think about evolution is in terms of, you know, how many copy numbers; like, how many of you are there? That’s your success. So just persistence and expansion into the world. From that perspective, I don’t think you need a particularly large cognitive light cone. You know, bacteria do great. But from other perspectives, if we sort of ask ourselves what the point is, and why do we exert all this effort to exist in the physical world, and try to persist, and exert effort in all the things we do in our lives, one could make an argument that a larger cognitive light cone is probably better in the sense that it allows you to generate more meaning, and allows you to bring meaning to all the effort and the suffering and the joy and the hard work and everything else. From that perspective, one would want to enlarge one’s cognitive light cone.

And, you know, I collaborate with a number of people in this group called CSAS, the Center for the Study of Apparent Selves. And we talk about this notion of something—for example, in Buddhism, they have this notion of a bodhisattva vow, and it’s basically a commitment to enlarge one’s cognitive light cone so that, over time, one becomes able to have a wider area of concern or compassion, right? The idea is you want to work on changing yourself in a way that enlarges your ability to really care about a wider set of beings. And so from that perspective, maybe you want the larger cognitive light cone.

29:14

Levin

Yeah. The relevance is, I think, really important. One thing I often think about is: what’s the fundamental unit that exists in the world? Is it genes? Is it information? What is it that’s really spreading through the universe and then differentially reproducing, and all that? I tend to think it’s perspectives. I think that what’s really out there as a way to describe really diverse agents. I think one thing they all have is a commitment to a particular perspective. So perspective in my definition is a bundle of commitments to: what am I going to measure about the outside world? What am I going to pay attention to? And how am I going to weave that information into some sort of model about what’s going on and, more importantly, what I should do next?

So there are many, many different perspectives and, as you said, it’s critical that every perspective has to shut out way more stuff than it lets in. And interestingly, this gets back to your original point about the different levels of description—you know, physics and chemistry and all that—because if you wanted to be a Laplacian demon—if you wanted to track microstates, all the particles, I’m just going to track reality as it really is, I’m just going to watch all the particles, that’s all I’m going to do—no living system can survive this way, right, because you would be eaten before anything else happens; you would be dead.

So I think that any agent that evolves under resource constraint, which is all of life—and it remains an interesting question: what does that mean for AIs and so on—but any agent that evolves under constraints of metabolic and time resources is going to have to coarse-grain: they’re going to have to not be a reductionist, they’re going to have to tell stories about agents that do things as a means of compression, as a way of compressing the outside world and picking a perspective. You cannot afford to try to track everything—it’s impossible.

So that compression also comes back to the memory engrams issue that we were talking about. Because as your past self compresses lots of diverse experiences into a compact representation of a memory trace of what happened, that kind of learning is fundamentally compression: when you’re compressing lots of data into a short, pithy generative rule, or a memory that you’re going to remember that you can use later, right? So not the exact experiences that you had, but some compressed representation. Well, the thing about compression is that the most efficient—data that’s compressed really efficiently starts to look random, right? Because you’ve gotten rid of all the correlations and everything else, the more you compress it, the more random it looks. And you’ve really gotten rid of a lot of metadata. You have to, that’s the point of compression. You’ve lost a lot of information.

32:25

Levin

That’s true because you’ve already compressed the hell out of it. That’s why, right? That’s why it’s incompressible: because you’ve already compressed it as much as it can. This is the issue that, for example, the SETI people—the Search for Extraterrestrial Intelligence—that they come up against. Because the messages that are going to be sent by truly advanced civilizations are going to look like noise to us, because really effective compression schemes, unless you know what the decomp—so you’ve compressed it—unless you know what the algorithm is to reinflate it, the message itself looks like noise. It doesn’t look like anything. Because you’ve pulled out all of the correlations. All of the order that would have made sense to a more naïve observer is now gone, and if you don’t have the key to interpret it with, it just looks like noise.

And so that means that, if you think about—there’s a bowtie kind of architecture that’s important here, where you take all these experiences, you compress them down into an engram (that’s your memory), and future you have to reinflate it again and figure out, “Okay, so what does this mean for me now?” It’s a simple rule that I inferred. Let’s say you learned an associative learning task or you’ve learned something general, you’ve learned to count a number or something like this. So rats, I think it takes about, if I recall correctly, like 3,000 trials before they understand the number three, that’s distinct from any instance. So it’s not like three peanuts or three sticks, it’s like the number three of anything, right? So after some thousands of trials they get it. And so now they have this compressed rule—they don’t remember the sticks or the peanuts or whatever, but they remember the actual rule, right? And so that’s the engram at the center of your bowtie. Well, future you has to reinflate it. It has to uncompress it and expand. Okay. But now I’m looking at three flowers, is that the same or not? How do I apply my rule to this?

And I think that what’s important is that you can’t do that deductively, because you’re missing a lot of that basic information. You have to be creative. When you interpret your own engrams, there’s a lot of creative input that has to come in to understand what it was that you were thinking at the time. And I think that kind of a thing—I mean, we know that recall of memories actually changes the memory, right? So in neuroscience they know this: that there’s no pure non-destructive reads of memories; that when you access your memory, you change it. And I think that’s why. Because the act of interpretation is not just a passive reading of what’s there, it’s actually a construction process of trying to recreate: So what does that mean for me? What does it mean now? And that’s part of that process of the dynamic self, you know: is trying to figure out—and obviously all this is subconscious—but trying to figure out what your own memories mean.

35:44

Levin

Yeah.

36:22

Levin

Well, so a few things. So the fact that there are an incredible number of processes under the hood, it has been known for a really long time. So not only all the physiological stuff that’s going on—I mean, you also don’t have the experience of running your liver and your kidneys, which are very necessary for your brain function. You are also not aware of all the subconscious motivations and patterns and traits and everything else. So let’s assume right now that whatever our conscious experience is, there is tons of stuff under the hood—not just the thing that I just said, but everything else that neuroscience has been studying for a hundred years or more—there’s lots going on under the hood. And that doesn’t define you, it enables you to do certain things. It constrains you from doing certain other things that you might want to do. The hardware does not define you.

I think the most important thing—and look, I think this is a really important question, because I get lots of emails from people who say, “I’ve read your papers. Now I understand I’m a collective intelligence of groups of cells. What do I do now? I don’t know what to do anymore,” right? And my answer is: do whatever amazing thing you were going to do before you read that paper. All of this information about what’s under the hood is interesting, and it has all kinds of implications, but the one thing it does not do is diminish your responsibility for living the best, most meaningful life you can. It doesn’t affect any of that.

And one way I think about this—there’s lots of sci-fi about this—but one thing that you might remember is, have you seen the film Ex Machina?

38:04

Levin

And so there’s one scene there where the protagonist, he’s standing in front of a mirror, and he’s completely freaked out because the AI is so lifelike. He’s now wondering: maybe he’s a robotic organism too. And so he’s cutting his hand, and he’s looking in his eye, and he’s trying to figure out what he is. And so let’s just dissect that for a minute. The reason he’s doing this—and what happens to most people, which I think is quite interesting—is that, if they were to open their arm, and they find a bunch of cogs and gears inside, I think most people would be super depressed. Because I think where most people go with this is: I just learned something about myself. Meaning: I know what cogs and gears can do, they’re a “machine,” and I just learned that I’m full of cogs and gears, therefore I’m not what I thought I was. And I think this is a really unfortunate way to think, because what you’re saying is: your experience of your whole life (and all of the joys and the suffering and the personal responsibility and everything else that you’ve experienced), you’re now willing to give all that up because you think you know something about what cogs and gears can do. I would go in the exact opposite direction and I would say: amazing! I’ve just discovered that cogs and gears can do this incredible thing! Like, wow! And why not? Because why do you think that ions and proteins and the various things in your body—those are great for true cognition. Like, I always knew I was full of protein and lipids and ions and all those things, and that was cool. I was okay with being that kind of machine. But cogs and gears? No way.

And so I think one thing that we get from our education focused on certain kinds of materialism is that we get this unwarranted confidence in what we think different materials are capable of. And we believe it to such an extent—I mean, I find this amazing as a kind of an educational or sociological thing—that we imbibe that story so strongly that we’re willing to then give up everything that we know about ourselves in order to stick to a story about materials. One thing that I think Descartes had really, really right is that the one thing you actually know is that you are an agentic being with responsibility and a mind and all this potential, and whatever you learn is on the background of that. So if you find out that you are made of cogs and gears, all you should conclude is: well, great, now I know this stuff can do it as well as proteins can. And so what I really hope that people get from this is simply the idea that pretty much no discovery about the hardware, no discovery about the biology or the physics of it, should pull you away from the fundamental reality of your being that, whatever it is that you are—groups of cells, an emergent mind pulled down from platonic space, or whatever—whichever of these things are correct, the bottom line is: you are still the amazing integrated being with potential and a responsibility to do things.

42:16

Levin

Yeah, for sure. And Iain McGilchrist makes a point of this too. He says that we’ve underestimated matter. You know, when we talk about materialism, we have been sold (and are selling still) this notion of matter as lacking intelligence. And I think that we need to give up this unwarranted confidence in what we think matter can do. We are finding novel, proto-cognitive capacities in extremely minimal systems—extremely minimal systems—and they’re surprising to us. They’re shocking, when we find these things. And I think we are really bad at recognizing what kinds of systems can give rise to minds, and therefore being depressed because we think that we are a particular kind of system, and there’s no way that system can be this majestic, agentic being. It’s way too early for that. We just, we absolutely—I mean, this is one of, I think, the major things: is that we have this idea that we know what different kinds of matter can do. And obviously, I’m not just talking about, you know, homogenous blocks of material, I’m talking about specific configurations. But really minimal kinds of things can have some very surprising cognitive qualities. And so, yeah, it’s way too early to think that we know what’s going on here.

44:00

Levin

Yeah. So bioelectricity is not magic. It is not unique in the sense that there are probably, certainly out in the universe, there are probably other ways of doing what it does. But what it does here on Earth in living forms is something very interesting. It functions as a kind of cognitive glue. So when you have collective intelligence, you need to have policies and you need to have mechanisms for enabling competent individuals to merge together into a larger emergent individual that’s going to do several things. First of all, the larger level is going to distort the option space for the lower levels. So the parts are doing things that the parts do, but they’re now doing it in a way that is coherent with a much higher-level story of what’s going on, higher level goals, because their action landscape is bent: it’s distorted by the larger level, their perception, their energy landscape is distorted.

So that collective new self is going to have memories, goals, preferences, a cognitive light cone. That requires some very specific features, and there are a bunch of them. And one of the modalities that lets that happen—that lets the cognitive light cone scale, that lets the intelligence scale—is bioelectricity. So by taking cells and enabling them to be part of an electrical network, there are some really interesting larger-level dynamics—which, you know, this is what we exploit in artificial neural networks, of course, right? Biology noticed this since the time of bacterial biofilms: that electricity is just a really good way for higher-level cells to emerge and higher level computation to emerge. There are probably other ways of doing it, but here on Earth, bioelectricity tends to be the way to go.

46:27

Levin

Yep. Yep.

46:34

Levin

Bad vocabulary. You’re a hundred percent right. The only thing that matters—because it doesn’t necessarily mean the next level up is not necessarily smarter than the lower level. Usually, but that doesn’t guarantee that at all. Not necessarily bigger or smaller in physical scale. So the only thing I mean by—and we really don’t have a great vocabulary yet for all this stuff—but the only thing I mean by “higher” is something akin to set membership: just the fact that a tissue is made up of cells, and cells are made up of molecular networks. That’s it. That’s all I’m talking about. I’m not saying that it’s bigger or more intelligent or more valuable. All I mean is that in this heterarchy, certain things are made up of other things. That’s it. That’s all I mean.

47:44

Levin

Yeah, the means in James is—at least the way I read him—James, when he says means, he means the path. That is the path: a means to an end, right? It’s a path that takes you to that end. So, you know, this is the kind of stuff we see in biology. Just to give you an example: one thing that people often think when they hear me talk about the intelligence of development and so on, people often think I mean the complexity, right? Just the fact that you start from an egg, and you end up with, I don’t know, a salamander or something: that there’s an increase in complexity. And then, rightly, people think: well, there’s lots of examples where simple rules give rise to complex outcomes. That’s just emergent complexity, that’s not intelligence. And, right, that is not what I mean. That’s not intelligence.

What I mean by intelligence is the problem solving of the following kind. And so, let’s say you have an egg belonging to a salamander. One of the tricks you can do is prevent it from dividing while the genetic material is copying. And so you end up with polyploid newts. So instead of two-end, you can have four-end, five-end, six-end, eight-end, that kind of thing. Well, when you do that, what happens is the cells get bigger in order to accommodate the extra genetic material, but the actual salamander stays the same size. So let’s say we take a cross-section of a little kidney tubule that runs to the kidneys, normally there’s like eight to ten little cells that go in the circle to make that tubule, and then there’s a lumen in the middle. So if you make the cells gigantic, the first thing you notice is that, well, first of all, having multiple copies of the genetic material, you still get a normal newt. That’s pretty amazing already. Second amazing thing is: the cells scale to the amount of genetic material, so the cells get larger. That’s amazing. Then you find that, well, actually, since the cells are really big, only a few of them are now working together to make the exact same size tubule. So they scale the number of cells to make up for the aberrant size of the cell. Right? That makes sense? And then the most amazing thing of all happens when you make truly gigantic cells, there’s not even room for more than one. One cell will bend around itself, leaving a lumen in the middle. The reason that’s amazing is that that requires a different molecular mechanism. That’s cytoskeletal bending, whereas before you had cell-to-cell communication.

And so that kind of thing, right? So just think about this. You’re a newt coming into the world. You have no idea. You can’t count on how much genetic material you’re going to have, how many cells you’re going to have, what size cells you’re going to have. What you do have is a bunch of cool tools at your disposal. You have cytoskeletal dynamics, you have gene regulatory networks, you have bioelectricity, you have all this stuff. And what you’re able to do under totally novel circumstances is pick from your bag of tools to solve the problem. I go from an egg to—in morphospace I take this journey—from an egg to a proper newt. Not only can I not count on the environment being the same, I can’t even count on my own parts being the same, right? That kind of—you know, another way to call this attitude is beginner’s mind. It’s like you don’t over-train on your evolutionary priors. You have a bag of tools and you’re not just a fixed solution. This is why I think evolution doesn’t just produce solutions to specific environmental problems, it produces problem-solving agents that are able to use the tools they have. I mean, what’s a better example of intelligence than something that can use the tools it has in novel ways to solve a problem it’s never seen before? Right? Like, that is a version of intelligence. And that’s what is all over the place in biology: the ability to navigate these pathways not only to avoid various barriers and so on, but to use the tools available to them in creative ways to solve the problem.

And we see some of this in extremely minimal systems. It does not require a brain. It doesn’t even require cells. Very minimal systems have surprising problem-solving capacities. And this is why we should be extremely humble when we try to make claims about what something is or isn’t, or what competencies it has. We are not yet good at recognizing those things. We do not have a mature science yet of knowing what the properties of any of this stuff is.

52:50

Levin

Yeah. The thing to keep in mind here is that estimates of intelligence, and I think all cognitive terms—so everything about, you know, all the words that people use: “sentience,” “cognition,” “goal-directed,” and so; as all of them—I think we have to remember that those are not objective properties of a given system. IQ is not the property of whatever system you’re trying to gauge the IQ of. It is your best guess about what kind of problem-solving you can expect out of that system. So it’s as much about you as it is about the system. And we’ve shown this in our experiments a lot of times. That, when people talk about certain kinds of developmental constraints, or they talk about the competency of tissues to do one thing or the other, it’s much more about our own knowledge of the right stimuli and the right ways to communicate with that system, and not so much about the system itself. When you make an estimate of the intelligence of something, you are taking an IQ test yourself. All you’re saying is: this is what I know, and this is what I can see, in terms of what kind of problem-solving I can see. And this applies to animals, this applies to AIs, this applies to humans in various economic environments. The simple version of this is: you show somebody a human brain and they say, “That’s a pretty awesome paperweight!” And I can see that it can do least action and hold down my papers against gravity. And that’s all I think it can do. And somebody else says, “You’ve missed the whole thing!” Right? This thing does all this other stuff. So I think that type of mistake, where (A) we think that it’s an objective property of the system, and (B) that we think that we’re good at determining what that is, is what bites us a lot when we’re dealing with especially unconventional systems.

So, to use your example: if someone looks at a kid with that environment and says, “Well, I don’t think this kid has as much intelligence,” the problem isn’t on the side of the kid. The problem is that somebody else might come along and says, “Oh, you don’t get it. In a different environment, this kid would exhibit all these amazing behaviors.” The good news about all of this—and it’s certainly not in my wheelhouse to comment on any of the kind of economic stuff or the sociology of it—but for the biology and for the computer science and so on, the good news is that all of these things are empirically testable.

So when we come across a certain system, each of us is going to guess: what is the problem space it’s operating in? What are its goals? And what capabilities do we think it has to reach those goals? And then we do the experiment, and then we see who’s right. That’s the thing: that this is not a philosophical debate. This is absolutely experimental. So if you say, “I don’t think these cells have any intelligence. I think they’re just the feed-forward emergent dynamics.” And I think, “Oh no, I think they’re actually minimizing or maximizing some particular thing, and they’re clever about doing it.” We do the experiment. We put a barrier in between them and their goals, and we actually see: do they or do they not have what I claimed to be their competency? And then we find out how much each of our views lets us discover the next best thing. So these are all empirically testable kinds of ideas.

56:23

Levin

Yeah, that sounds like—I think that might have been a review with Pamela Lyon. Yeah.

56:42

Levin

To be completely honest, I don’t remember this part. I mean, yes, I certainly don’t remember the pages or the—

56:49

Levin

Yeah, I mean—so, okay. So the idea for basal cognition is basically that, whatever cognitive capacities we have, they have to have an origin and we have to ask where that they come from. Because this idea that we are completely unique, and they suddenly sort of snap into place, it doesn’t work evolutionarily and it doesn’t work developmentally. Both of those are very slow processes. So the stories we have to tell are stories of scaling. To really understand these processes, we have to understand how simple information-processing capacities scale up to become larger cognitive light cones, more intelligent systems, project into new problem spaces, and so on. So basal cognition is the question of: okay, so where did our cognitive capacities come from? So that means looking at the functional intelligence of cells, tissues, slime molds, microbes, bacteria, and minimal matter, active materials, that kind of stuff. That’s basal cognition.

Well, what do the really primitive versions of cognition look like? And it’s a really important skill to practice that kind of imagination, because often what trips people up is, they imagine, for example, panpsychist views, right? So somebody says, “Oh, you’re trying to tell me that this rock is sitting there having hopes and dreams.” Well, no, that’s not the claim. The claim isn’t that these full-blown large-scale cognitive properties that you have are exactly there everywhere else. The claim is that it’s a spectrum or a continuum, and that there are primitive tiny versions of them that also should be recognized, because we need to understand how they scale. So that’s basal cognition.

59:07

Levin

Well, I personally am on board with that. I think potential energy and least action principles are the tiniest hopes and dreams that there are. So I agree with that completely. I think that is the most basal version. And in our universe—so this goes way beyond my pay grade—but, for example, I’ve talked to Chris Fields, who is really an expert in this stuff, and I asked him: is it possible to have a universe without least action laws? Right? And he said the only way you can have that is if nothing ever happens. So if that’s the case, that tells me that, in our world, there is no zero on the cognitive scale, and everything is on it.

But again, we have to ask, then—so I agree with you. I think if you’re on the spectrum, then you’re on and that’s it, and I think in this universe everything is on. But we have to ask ourselves: what do we want this terminology to do for us? So that’s why some people critique these kinds of perspectives by saying, “Well then, if everything is on it, then the word means nothing. Then why do we even have to work?” If everything is cognitive—and I didn’t say “consciousness” yet—but let’s say everything is cognitive, then why do we need the word? Everything is. And I really think that we need to focus on what we expect the terminology to do for us.

So let’s imagine. Let’s just dissect this for a minute. The old paradox of the heap, right? So you got a pile of sand, and you know that if you take off one little piece of sand, you’ve still got a pile. But eventually you have nothing. So how do you define the pile? So I think for all of these—so my answer to this (and I think the solution to all of these kinds of terminological issues) is that it’s not about the object itself, it’s about what tools you are going to use to interact with it. So if you call me and you say, “I have a pile of sand,” all I want to know about the definition of “pile” is: am I bringing tweezers, a spoon, a shovel, a bulldozer, dynamite? What are the tools that I have to do what we need to get done? So that is the only value in this terminology.

So by saying that everything is cognitive, does that by itself help us with anything? No. I think what does help is if you tell me what kind of cognition, and how much. And that’s an empirical question, and then we can argue about it. And the answer to that question is: what are the tools that help us the most? So you show me a bunch of cells, and you say, “I think the right way to do this is physics, chemistry, and feed-forward emergence, and complexity. That’s how I think we’re going to interact with it.” And I look at it and say, “I think the way to interact with this is through some interesting concepts from cognitive neuroscience, including active inference, learning, training, and so on.” Then we get to find out who’s right. If I can show that, using my concepts, I got to new discoveries that you didn’t get to, well, there you go. On the other hand, if I waste all my time writing poems to a rock and nothing ever comes of it, well, then you’re right. And so I think that the point of all this terminology—yes, we can say it’s all on the spectrum, but now comes the fun and interesting work of saying: okay, so what does the spectrum look like, and where on the spectrum do the various things that we see land?

1:03:14

Levin

You should have him on. This is getting way beyond anything I could argue with you about, but you should have him on and you guys could tell—I would watch that for sure.

1:03:27

Levin

Yeah, yeah.

1:03:48

Levin

Well, let’s see. I can say a few things. What I definitely don’t have yet—I’m working on it, but I don’t have anything that I would talk about now—a new theory of consciousness. So I do not have anything brilliant to add to this that somebody else hasn’t already said. So I’m just going to kind of tell you what I have to say now.

1:04:08

Levin

I think that one thing that’s really hard about consciousnes,s and what makes it the hard problem, is that unlike everything else that we work with, we don’t have any idea what a correct theory would output. So what format would the predictions—

1:04:32

Levin

No. No, no. No, we don’t have any idea of what the output of a correct theory would look like. What would it give you? Right? So, for everything else, a good theory gives you numbers, predictions about specific things that are going to happen. What does a good theory of consciousness give you? So, what we would like is something that—we say: okay, here’s a cat, or here’s a cat with three extra brain hemispheres grafted on that also has wings. What is it like to be these creatures? Right? What is the output of a correct theory of consciousness? Because if it outputs patterns of behavior or physiological states, then what you’ve explained is physiology and behavior. There are going to be people that say, “Well, you haven’t explained the consciousness at all.” In fact, almost all theories of consciousness look kind of eliminativist. Even the ones that aren’t trying to be, even the ones that say, “No, no, we’re not trying to explain away consciousness. It’s real and I’m going to explain it.” Then you look at the explanation and you always feel like: yeah, but you haven’t explained the actual consciousness. You’ve explained some kind of behavioral propensity as physiological states or whatever. So that’s the problem that we have.

Consciousness is one of those things that cannot exclusively be studied in the third person. Everything else you can study as an external observer, and you don’t change much as the observer by studying them. Consciousness, you can really only, in a full way, you can only study consciousness by being part of the experiment: by experiencing it from the first-person perspective. So the weak version of this is, you might say: well, a good theory of consciousness is art. What it outputs is art, poetry, whatever—that, when we experience it, it makes us experience that conscious state, and then we say, “Oh, so that’s what it’s like. I see.” Right? So that’s kind of a weak form.

You can do a stronger form and you say, “Well, the real way to do it is to have a rich brain interface.” So if I want to know what some other system’s consciousness is like, we need to fuse together. Now, caveat to that is: you don’t find out what its, if you do that—so let’s say a rich kind of brain interface, so we really connect our brains together or something—you don’t get to find out what it’s like to be that system. Both of you find out what it’s like to be a new system composed of the two of you. So it’s still not. So from that perspective it’s really hard. I mean, I suppose people who do meditation or take psychedelics, I suppose they’re doing experiments in actual consciousness. But third-person experiments in consciousness are really hard. You can do things like turn it off, you know? So there’s general anesthesia, and you can say, “Oh look, the consciousness is gone.” And even then, some people will say, “Yeah, but I experienced floating above my body while you did the surgery, and I saw you drop the scalpel and do this and that.” So it’s still, even with that amazing reagent of being able to supposedly shut off consciousness, you still got some issues.

So the study of consciousness is hard for those kinds of reasons. And I think that about the only useful thing I could say here is that, for the same reasons that we associate consciousness with brains, for exactly those same reasons we should take very seriously the possibility of other forms of consciousness in the rest of our bodies, and also lots of other things. So Nick Rouleau and I are working on a paper on this, where you can sort of look at all the different popular theories of consciousness on the table, and you can just ask: which of them are specific for brains, and why? Like, what aspects of each of those theories really tells you that it’s got to be brains? My guess is—we haven’t finished the paper yet—but my guess right now is that there’s not a single one that can distinguish brains from other types of structures in your body. And so I think we should take very seriously the possibility that other subsystems of the body have some sort of consciousness. It’s not verbal consciousness—

1:08:41

Levin

Yeah, we ask: what is it about that theory that says it’s in the brain rather than somewhere else? Let’s say you’re a liver.

1:08:56

Levin

Right, right. And I agree with that. Now, as far as I understand, IIT also has an exclusion postulate that says there should only be one central consciousness in the system. I think that’s true—at least it used to be true. I don’t know. Giulio may disagree with that. But I think we are actually a collection of interacting perspectives and interacting consciousnesses for that reason. And then sometimes people say, “Well, I don’t feel my liver being conscious.” Right, you don’t, but you don’t feel me being conscious either. Of course you don’t. And the fact that your left hemisphere has the language ability for us to sit here and talk about it and the liver doesn’t, doesn’t actually mean that it’s not conscious. It just means that we don’t have direct access to it, and we don’t have direct access to each other. So that doesn’t bother me. So that’s my suspicion about consciousness—is that, for the same reason that people think it’s in the brain, we should take very seriously that it’s in other places in the body. And then, more generally, other types of constructs that are not human bodies at all, or not even animal bodies.

1:10:32

Levin

Yeah. Boy, I don’t—you know, it’s hard for me to know what most people agree or disagree with him about. I really don’t know. We agree on a lot of things. We agree on, I think, the primacy of consciousness. I think that, you know, his idealist position has a lot to recommend it. One thing I think we disagree on is the issue of compositionality. So if I recall correctly from a talk that we had together a little while ago, he felt that it is important, in order to be a true self, to have a conscious experience and an inner perspective. You have to be—you know, he focuses on the view of embryonic development as a single system that, you know, whatever, subdivides and develops, but it starts out as a single system. And I was arguing that that that really is just a contingent feature of biology. I mean, we certainly can take two early embryos and mush them together. You get a perfectly normal embryo out of it. And in general there are lots of biological systems—like our xenobots, like anthrobots—that you can create by composition; by pulling other things together. So I don’t put as much emphasis on a system being demarcated from the outside world because it started out that way and it sort of remained disconnected. I think that’s kind of a superficial aspect of the biology, and you can do things a different way. I don’t think that’s what’s responsible for it.

But, you know, I think he thinks it’s important that individual selves are not compositions. They’re not made as compositions. They’re somehow individualized from the word go. Which, again, even the egg, right? So we humans like eggs because we can see it as a distinct little thing with a membrane, and you say, “Ah, there’s an individual.” But even an egg is composed by the maternal organism from molecular components. Like, I see no point at which any of this is truly distinct from anything else. So I put less emphasis on it, but I think he thinks it’s important.

1:12:59

Levin

What I’m saying is—and I, you know, I may not be doing justice to his view, and I think you should ask him more about this—but I think he thinks that it’s important, in order to be a unified—so I think we were discussing what makes for a unified inner perspective, right? So we don’t feel like billions of individual brain cells. I mean, I have no idea what—we kind of do, because that’s what it feels like to be billions of individuals of neurons. That’s really what it feels like. But we do feel—at least most of us, most of the time—feel like some kind of unified, centralized inner perspective. And so we were talking about how that comes about. And I think he felt that having that in the physical universe is importantly related to arising from a single origin. So he sees the egg as a single point of origin, and arising from that, that’s how you are a separate individual from others. And I see it as much more fluid, and I see the boundary between self and world as something that can change all the time. I think it changes in embryogenesis—and that’s the story of the scaling of the cognitive light cone that we talked about. I think it can shrink during cancer. I think it can change during metamorphosis, during maturation. I think it’s much more fluid than that.

1:14:24

Levin

I don’t know. I mean, certainly I think most—well, religions that have a God, anyway, as far as my understanding is—they would think that, yes, that God has extreme agency; in fact, higher than ours. I don’t know what that really buys us in any helpful way.

1:14:50

Levin

Okay, so that’s an interesting question. So let’s start with, first of all: how do we know when anything has agency? And that’s an experimental research program. So you basically hypothesize what problem space it’s working in, what you think its goals are, and then you do experiments to figure out what competency it has. And then you find out: did I guess well? Poorly? Do I need to make a better guess? And so on. So, for example, people have said to me, “Well, your kind of panpsychist (almost) view says that the weather should be cognitive.” And I don’t say that it is or isn’t, because we haven’t done the experiments. Do I know that weather systems—let’s say, let’s say hurricanes or so on—do I know that they don’t show habituation, sensitization, that they couldn’t be trained if you had the right scale of machinery? I have no idea. But what I do know is that this is not a philosophical thing that we can decide arguing in an armchair. Yes, it is—no, it isn’t. But you have to do experiments, and then you find out.

So now the question is: okay, so what about the galaxy? What about the universe? Right? Are these the Gaia ecosystems? Again, I think these are all empirical questions. Now, some of them are intractable. We don’t have the capability to do experiments on a planetary scale. But, for example, one thing that I did try to do once was design a gravitational synapse. So: design a solar-system-size arrangement where masses would fly in, and based on the history of masses flying in, it would respond to new masses in a different way. So you can do historicity, and you can have habituation and sensitization and things like that. So could you have something like that, that would be very, sort of, ponderously slowly on an enormous scale, computing something and having sort of simple thoughts? I bet you could. I bet you could. Is the real universe doing that? I have no idea. We have to do experiments.

So here you bump up against another question, which is: how do you know if and when you are part of a larger cognitive system? How do we know if we are in fact part of a bigger mind? So, I don’t know. My suspicion is that there is some sort of Gödel-like theorem that will tell you that you can never know for sure, and you can never be certain. But I bet that you could gather evidence for it—for or against. And I often think about a kind of a mental image. Imagine two neurons in the brain, and one is a kind of a strict materialist and one’s a little more mystical. And the one neuron says, “We just run on chemistry, and the outside world is a cold mechanical universe, and it doesn’t care what we do. There’s no mind outside of us.” And the other one says, “I can’t prove it, but I kind of feel like there’s an order to things. And I kind of feel like our environment is not stupid. I kind of feel like our environment wants things from us. And I kind of feel these waves backpropagating through us that are like almost rewards and punishments. I feel like the universe is trying to tell us something.” And the first one says, “Nah, you’re just seeing faces and clouds. It doesn’t exist.” And of course, in my example the second one is correct, because they are in fact part of a larger system. They’re part of a brain that is learning things. And it’s very hard for any one node in that system to recognize that, or even a subnetwork. But I wonder if we could—having a degree of intelligence ourselves—if we could gain evidence that we were part of a larger system that was actually processing information. And I don’t know exactly what that would look like, but my hunch is that it would look like what we call synchronicity. I think that what it would look like are events that don’t have a causal connection at our lower level—like mechanistically, like by physics, there’s no reason why that should be—but at a larger scale in terms of meaning and the greater meanings of things, that they do have some kind of interpretation. And I think that’s what it would look like to be part of a larger system. I think it would look and feel like synchronicity. So does it exist? I don’t know. But that’s what I think it would feel like.

1:19:11

Levin

Yeah. Okay, let’s see. Well, it started—I mean, it started in my head when I was pretty young. Like, it was a dream that I had to do this kind of stuff. I mean, I consider myself to be the luckiest person in the world. I get to do the funnest stuff with the best people. So yeah, I think it’s super, super fortunate. But I kind of had this idea when I was very young. I had no idea what it was like. I was pretty sure that it was actually impossible. I never really thought it would be practically feasible, but I figured I would push it as far as I could before I would have to go back to coding and, you know, because, yeah—

1:19:55

Levin

Right, right. Yeah, yeah. Yeah, I learned to program pretty young. And at that time that was a pretty good way to make money. And I just figured I would do the biology as long as I could, and then eventually I would get kicked out, and then I would just go back to coding. So yeah, my lab actually began in September of 2000. That’s when I got a faculty position at the Forsyth Institute at the Harvard Medical School. And we opened our doors in 2000. It was just me at first, and then me and one other technician. There’s like 42 of us now. But at the time it was just me and a tech named Adam. And that was the first time, you know, starting then was the first time that I could really start to be practical about some of the ideas I had about bioelectricity and cognition, all these things prior to that. I was building up a tool chest. So I was building up skills, techniques, you know, information, and so on. But being a grad student and then a postdoc, I wasn’t able to talk about any of these things. But then, when I was on my own, and then that was the time to get going.

So just a couple of interesting milestones. Already by the time I got—you know, we opened the lab. I was involved in a collaboration with Ken Robinson and his postdoc, Thorly Thorlin [?], and together with my postdoc mentor, Mark Mercola, we really showed the first molecular tools for bioelectricity. So we had a paper on left-right asymmetry, and we showed the first bioelectric tracking of non-neural bioelectric states in the chicken embryo. We showed that it was important for setting up which side is left, which side is right, and then manipulating that information using injected ion channel constructs. So that was the first time any of that, you know, reading and writing the mind of the body—which is how I certainly wouldn’t have said it back then, but this is how I see it now—that was the first time that was done in a molecular way. So that cell paper came out in, I think, 2002, I think it finally came out. But that was a really early project.

The other really early project was: as a postdoc, I started gathering tools for this whole effort. And a lot of those tools were DNA plasmids encoding different ion channels. And so what I would do is: I would send emails or letters to people working in electrophysiology, gut physiology, inner ear, and they would have some potassium channel that I had cloned, and I would say, “Could I get one of these plasmids?” And I was telling them what I was going to do. I would say, “And what I’m going to do is use it to express it in embryos in various locations, and use it to, in a very targeted way, change the bioelectrical properties of these things.” And most people were very nice, and they sent me these constructs. One person sent a letter to my postdoc mentor to say that clearly I had had a mental breakdown and that he should be careful, because this is so insane that I’m obviously off my rocker, right? And so I remember—

1:19:55

Levin

Well, okay, so I didn’t see the letter, but my boss came to me. So my boss came to me, and he was laughing, and he said, “Look at this. This guy says you’re nuts. You asked him for a plasmid. He told me to watch out. He says you’re having a psychiatric break.” So what I’m relaying is what he said to me.

1:23:37

Levin

But nevertheless, most people sent constructs. And when we got to lab, when my lab opened, I started doing that. I started mis-expressing these things in embryos just to see the space of possible changes, right? What does bioelectricity really do? I mean, nobody knew at the time. It was thought—it was really crazy—it was thought that membrane voltage was a housekeeping parameter, that it was an epiphenomenon of other things that cells were doing, and that if you mess with it, all you’re going to get is uninterpretable death. Everybody thought this was a stupid idea.

And so we started doing this. And I had this graduate student—she was in the dental program; her name was Ivy Chen, and she was in the dental program—and she had amazing hands. And so I taught her to micro-inject RNA into cells in the embryos, because she had really, really good hands.

1:24:34

Levin

Yeah, well, she was a dental student, and so I talked to her. She wanted to do research, and I said, “Tell me what you do,” and she said, “Oh, I do these, you know, I sew up people’s, you know, gums and whatever,” and I said, “Okay, you probably could do this.”

1:24:47

Levin

No, no, no—well, she may have been also, but I don’t know anything. I don’t know that.

1:24:51

Levin

What I know is that she was doing surgeries in people’s mouths, and I thought that she may be able to, in tight, confined places, you know, with the glasses and everything, so I thought she would be able to do this through a microscope—and she was. And so we did this together, and we injected these constructs. And I still remember to this day, she calls me in one day, and she says, “So I looked at the embryos, and they’re covered with some kind of black dots,” And I said, “Black dots? Let me see, let’s go look.” So I come out, and we look through the microscope—the black dots were eyes. What she had done was: the potassium channel that she injected was the one that we later—it took years to publish the paper—but what she had discovered is a particular bioelectrical state that tells cells to build an eye. It’s remarkable, because right there and then, you knew that (A) bioelectricity was instructive, it was not an epiphenomenon because it controls which organs you get, (B) that the whole system is modular and hierarchical, because we did not tell the cells how to build an eye. So we didn’t say where the stem cells go, or what cells go next to what other cells, or what genes should be expressed. We did none of that. We triggered a high-level subroutine that says: build an eye here. So right away that one experiment told us all these amazing things. Then eventually—and this was the work of a grad student, Sherry [???] in my group, who took on the project, and she did a whole PhD on this, showing that.

The other amazing thing about it is that, if you only target a few cells, what they do is: they get their neighbors to help out, because they can tell there’s not enough of them to build an eye. Kind of like ants: ants recruit their buddies to take on a bigger task. So that tells you that the material you’re working with has these amazing properties that you don’t have to micromanage, right? It’s a different kind of engineering. It’s engineering—as I put it recently in a recent paper—it’s engineering with agential materials, because it’s a material with competencies with an agenda. You don’t have to control it the way you do wood and metal and things like that.

So okay, so anyway, so that kind of thing. Then we had a bunch more work on left-right asymmetry and showing how the cells in the body decide which side they’re on based on these electrical cues. Then we discovered that the way they interpret these electrical cues had to do with the movement of serotonin. So long before the nervous system or the brain shows up, the body is using serotonin to interpret electrical signals. So this was really underscoring the idea that a lot of the tools, concepts, reagents, pathways, mechanisms from neuroscience really have their origin much earlier. And so this was a completely new role for serotonin, right? So serotonin is a neurotransmitter, does many interesting things, but long before your brain appears, it also controls which direction of your body your heart and your various other organs go to. So trying to understand, in the short term, how does electrical activity control cell behavior, but bigger picture, while these neural-like processes are going on in cells that are absolutely not neurons much long before that. So that was kind of cool.

Then I hired a postdoc named Danny Adams, who later became a faculty member and a colleague. And one of the things that she did was to pioneer the early use of voltage-sensitive dyes to read these electrical potentials. And so she discovered, in the work that she did in my group, she discovered this thing we call the electric face.

1:28:26

Levin

Oh, this is the electric face. This is probably 2008? Something like that. 2007, 2008. And what she had discovered was that, if you look at the nascent ectoderm that later will regionalize to become eyes and mouth and all of that, that early on, before all the genes turn on that determine where all those things will go, the bioelectric pattern within that ectoderm looks like the face. It shows you where all the stuff is going to go. And then ultimately we were able to show that—and, by the way, there was that eye spot, which is why the eye thing worked. And we were able to show that all kinds of birth defects that mess up the formation of face do it by inhibiting the normal bioelectric pattern, and that you could fix it. You could exert repair effects that way. So that was interesting.

Then we started looking at regeneration. And again, the early work was done also by Danny, and then later on by Kelly Chang, who’s now a faculty member at University of Las Vegas, where what we did was we showed that tadpole tail regeneration was also bioelectrically driven. And that was our first gain of function effect in regeneration, where we were able to show that we could actually induce new regeneration. So the tail is a very complex organ, it has a spinal cord, muscle, bone—well, not bone—vasculature, peripheral innervation, skin. And so we took tadpoles that normally do not regenerate, and there’s a stage at which they can’t regenerate their tails, and we developed a bioelectric cocktail that induces it to grow. My postdoc at the time, Kelly Chang, said, “I soaked them.” I said, “How long did you soak them for?” And she said, “An hour.” And I thought, “But that’s gotta be too short. There’s no way an hour soak is going to do anything.” And sure enough, that hour soak led to eight days of regeneration where we don’t touch it at all. And the most recent version of that work is in the frog leg, where we show that 24-hour stimulation with our cocktail induces a year-and-a-half of leg growth, during which time we don’t touch it at all.

So the amazing thing there is, again, this is not micromanagement, this is not 3D printing, this is not us telling every cell where to go during this incredible year-and-a-half long process. This is at the very earliest moment you communicate to the cells: “Go down the leg building path, not the scarring path,” and that’s it. And then you take your hands off. It’s calling a subroutine, it’s modularity, it’s relying on the competency of the material, where you’re not going to micromanage it.

So that was the first time that kind of became obvious that that was possible, is when she showed that just an hour stimulation of the correct bioelectrical state got the whole tail to commit to regenerate itself. So that was the beginning of our regeneration program, after which we went into limbs. And now, of course, we’re trying to push into mammalian limbs. Along the way Celia Herrera-Rincon, and Nirosha Murugan were other post-docs that showed leg regeneration in frog and so on.

Around that time, we had another thing. I really wanted to work on cancer. And I really wanted to work on this idea that there’s a bioelectric component to it. And the way you can think about it is simply: not why is there cancer, but why is there anything but cancer? So why do cells ever cooperate instead of being amoebas? Why do they ever cooperate? And so we know that the bioelectric signaling is the kind of cognitive glue that binds them together towards these large-scale construction projects—maintaining organs and things like that. And so, yeah, so we wanted to study that bioelectrically. And so I had two students, Brook Chernet and Maria Lobikin, who undertook that. And we were able to show that, using this bioelectrical imaging, you can tell which cells were going to convert ahead of time. You could also convince perfectly normal cells to become metastatic melanoma just by giving them inappropriate bioelectric cues about their environment. So you can—no genetic damage, no carcinogens, no oncogenes, but just the wrong bioelectric information, and they become like metastatic melanoma. And, best of all, they were able to show that you can actually reverse carcinogenic stimuli—for example, human oncogenes—by appropriate bioelectric connections to their neighbors. So we had a whole set of papers showing how to control cells bioelectrically. And Juanita Matthews in my lab now is trying to take all those strategies into human cancer.

1:33:17

Levin

This was—yeah, the first experiments were done around 2010, 2011, something like that.

1:33:35

Levin

Well, to be clear, the very first person who talked about this was Clarence Cohn in 1971. So Clarence Cohn in 1971 wrote that—he had a couple of papers in Science where he showed that resting potential of cells was an important driver of cell proliferation, and he conjectured that it might have something to do with cancer. So this idea had been floated. Nobody had ever done anything with it. And the tools to study this at a molecular level didn’t exist until we made them. So that idea—just that bioelectricity is important in cancer—had been around before. What I think we brought to it that was completely new is the notion that this is also related to cognition. The idea that it’s not just that it drives proliferation and cancer, that this is really involved in limiting the size of the cognitive light cone at which point cells acquire ancient metastatic-like behavior the way that amoebas do. That aspect of it, I think, is completely new with us. That idea that this really is about the boundaries of the self. And I’ve never seen anybody else talk about that.

So around that same time something else interesting that happened in 2009, which is that we were studying plenaria, we were studying flatworms, and we had shown that when you cut plenaria into pieces, the way that these pieces decide how many heads they were going to have is actually related to the ability of cells to communicate with each other using gap junctions, these electrical synapses. And so we had made some two-headed worms and so on. And around 2009, I had this visiting student from BU, her name was Larissa, and I asked her to re-cut the two-headed worms just in plain water, no more manipulation of any kind. We—

1:35:44

Levin

Again, yeah. So you have a normal one-headed worm, you cut off the head and the tail, you have the middle fragment, you soak that middle fragment in the drug that blocks the cells from electrically communicating with each other, and they develop heads at both ends.

1:36:03

Levin

It wasn’t even a cocktail. It was one single chemical. It was really simple. It was just one single chemical, and all it does is block gap junctions. And so what that did was change the electric circuit properties that the cells have, and both wounds decided they had to be heads. So now you get these two-headed worms. So yeah, so Larissa re-cuts these two-headed worms in plain water, no more manipulation, and she gets more two-headed worms. It’s permanent once you’ve convinced them. Now, the genetics are untouched, right? No genomic editing, no trans genes. It’s genetically identical. But the two-headed worms are a permanent line now.

So a couple of interesting things there. One is: it shows the interesting memory properties of the medium, meaning once you’ve brought it to a new state, it holds. It remembers the two—so it’s a kind of memory—it remembers the two-headed state. Another interesting thing is that two-headed worms were first seen—well, they were first described in the early 1900s. So people had seen—they had to be made by other means—people had seen two-headed worms. Apparently, to my knowledge, and I don’t think anybody’s ever written about this, nobody thought to recut them until we did it in 2009. And I think the reason is because it was considered totally obvious what would happen. I mean, their genome is normal, you cut off that second ectopic head, and of course it’ll just go back to normal, that’s what people assume. So this is another example of why thinking in these different conceptual ways, it matters, it leads to new experiments. Because if you don’t think about this as memory, if you’re focused on the genes as driving phenotypes, then it doesn’t make any sense to recut them. But if you start thinking, “Well, I wonder if there’s a physiological memory here,” then that leads you to this experiment, right? So thinking in this way leads to new experiments.

And then the other thing it points out is something really interesting. So, for pretty much any animal model, you can call a stock center and you can get lines and genetic mutants. So you can get flies with curly wings, and mice with crooked tails and weird coat patterns, and chickens with funky toes. So you can get any kind of mutant lines. In planaria, there are no mutant lines. Nobody’s ever succeeded in making anything other than a normal planarian, except for our two-headed form, and that one’s not genetic. And so there’s a deep reason—which I didn’t understand back then. In fact, I think I only really figured out what I think it means in the last few months. But it was striking that the only unusual permanent planarian form out there was the one that we had made, and that’s the one that’s not genetic. It’s not done by the way that you would do this with any other animal. Yeah.

1:38:52

Levin

Well, it’s not so much a discovery. It’s more of a new way of thinking about it. So, one of the weird things about planaria is that, because the way they reproduce—at least the ones that we study—the way they reproduce is they care themselves in half, and then each half grows the rest of the body. And typically, what happens for most of us that reproduce by sexual reproduction is that, when you get mutations in your body during your lifetime, those mutations are not passed on to your offspring. Right? They disappear with your body and then the eggs go on and so on. Well, in planaria it’s not like that. In planaria, any mutation that doesn’t kill the cell gets expanded into the next generation, because each half grows the remainder of the body. And so their genome is very messy. They have, I mean, in fact, cells can be mixed-employed. They can have different numbers of chromosomes. That’s very weird. And I always thought: isn’t it strange—and nobody ever talked about this in any biology class that I’ve ever had—isn’t it strange that the animal that is the most regenerative, apparently immortal (they don’t age), cancer resistant, and by the way, resistant to transgene (so still, nobody’s been able to make transgenic worms), is also the one with the most chaotic genome? And that’s bizarre. You would think, from everything that we were told about genomes and how they determine phenotypes, that the animal with all those amazing properties should have really pristine hardware. You would think that you should have a really clean, really stable genome if you’re going to be regenerative and cancer resistant and not age and whatever. It’s the exact opposite. I always thought that was incredibly weird.

And so finally, I think—and we’ve done some computational work now to show why this is—I think we now understand what’s happening. And what I think is happening is this. Let’s go back to this issue of developmental problem-solving. So if you had a passive material such that you’ve got some genes, the genes determine what the material does, and so therefore you have an outcome, and that outcome gets selected—either it does well or not—and then there’s differential reproduction. So the standard story of evolution. Then everything works well and everything works like it would in a genetic algorithm. Very simple. The problem with it, of course, is that it takes forever. Because let’s say that you’re a tadpole and you have a mutation. Mutations usually do multiple things. Let’s say this mutation makes your mouth be off kilter, but it also does something else somewhere else in the tail, something positive somewhere in the tail. Under the standard evolutionary paradigm, you would never get to experience the positive effects of that mutation because, with the mouth being off, you would die and that would be that. So selection would weed you out very quickly and you would have to wait for a new mutation that gives you the positive effects without the bad effects on the mouth. Right? So it’s very hard to make new changes without ripping up old gains and so on. So that’s some of the limitations of that kind of view.

But a much more realistic scenario is the fact that you don’t go straight from a genotype to the phenotype. You don’t go from the genes to the actual body. There’s this layer of development in the middle. And the thing about development is not just that it’s complex, it’s that it’s intelligent—meaning it has problem-solving competencies. So what actually happens in tadpoles is: if I move the mouth off to the side of the head, within a few weeks, it comes back to normal on its own. Meaning it can reach again that region of anatomical space where it wants to be.

So imagine what this means for evolution when you’re evolving a competent substrate, not a passive substrate. By the time a nice tadpole goes up for selection to see whether it gets to reproduce or not, selection can’t really see whether it looks good because the genome was great or because the genome was actually so-so but it fixed whatever issue it had. So that competency starts to hide information from selection. So selection finds it kind of hard to choose the best genome, because even the ones with problems look pretty good by the time it’s time to be selected. So what happens—and we did computational simulations of all this—what happens is that, when you do this, evolution ends up spending all of its effort ramping up the competency because it doesn’t see the structural genes, all it sees is the competency mechanism. And if you improve the competency mechanism, well, that makes it even harder to see the genome. Right? And so you have this ratchet; you have this positive feedback loop where the more competent the material is, the harder it is to evolve the actual genome. All the pressure is now on the competency. So you end up with kind of like a ladder, really an intelligence ratchet, right? And people like Steve Frank and others have pointed this out in other aspects of biology, and also in technology, right? Once RAID array technology came up, it became not as important to have really pristine and stable disk media, because the RAID takes care of it, right? So the pressure on having a really, really stable disk is off.

So what it means is that, in the case of the planaria, that positive feedback loop, that ratchet, went all the way to the end. That basically, what happened here is that what you’ve got is an organism where it is assumed that your hardware is crap. It’s assumed that you’re full of mutations, all the cells have different numbers of chromosomes. We already know that genetics are all over the place. But all of the effort went into developing an algorithm that can do the necessary error correction, and take that journey in morpho-space no matter what the hardware looks like. That is why they don’t age. That is why they’re resistant to cancer. And that’s why nobody can make a transgenic worm—because they really pay less attention to their genome, in that sense, than many other organisms.

So you can imagine a sort of continuum. So you’ve got something like C. elegans, the nematode, where they’re pretty cookie cutters. So, as far as we know, they don’t regenerate much. What the genome says is pretty much what you get. Then you’ve got some mammals, right? So mammals, at least in the embryonic stages, they’ve got some confidence. You can chop early mammalian embryos into pieces and you get twins and triplets and so on. Then you get salamanders. Salamanders, they’re quite good regenerators. They’re quite resistant to cancer. They are long lived. And then, when you run that spiral all the way to the end, you get planaria, which are these amazing things that have committed to the fact that the hardware is going to be noisy, and that all the effort is going to go into an amazing algorithm that lets them do their thing. And that’s why, if you’re going to make lines of weird planaria, targeting the structural genome is not helpful, but if you screw with the actual mechanisms that enable the error correction, AKA the bioelectricity, that’s when you can make lines of double-headed and so on, because now you’re targeting the actual problem-solving machinery.

1:46:06

Levin

It’s a good question. So nobody’s done that specifically, as far as I know. This is something that we’re just ramping up to do now, is to start—

1:46:29

Levin

Yeah, I think that—yes, I think that’s a prediction of this model that I just laid out. And so, yeah, so we can test that. I mean, well, part of it also is: there’s an ecological component. I mean, you can ask the question: so why doesn’t everything end up like planaria? And I think there’s an aspect of this that that ratchet obviously doesn’t run to the end in every scenario, because in some species there’s a better trade-off to be had somewhere else.

1:46:59

Levin

Which I think would probably be environment. It would probably be the ecology of how do you reproduce? How noisy, how dangerous, how unpredictable is your environment? I’m going to guess there’s something like that involved here, yeah. But I think that’s starting to kind of explain what’s going on with planaria. So we found the persistence of the two-headed phenotype, and then Nestor Oviedo and Junji Morokuma in my group wrote a nice paper on that, and so on.

And then the next kind of big advance there, in 2017, was by Fallon Durant, a grad student in my lab who also did something interesting. So when you take a bunch of worms and you treat them with, let’s say, this reagent that blocks the gap junctions, typically what you see is: okay, you treat a hundred worms, 70% of them go on to make two heads and 30% are unaffected. So we thought they were unaffected because they stay one-headed, and we always call them escapees because we thought that they somehow just escaped the action of the octenol. Maybe their skin was a little thicker or something—we never had a good explanation for it. But anyway, we had those 70% penetrance to the phenotype. And most things have not 100% phenotype, so it wasn’t.

1:48:17

Levin

Penetrance just means that when you apply some treatment to a population, not all of them have an effect, and not all of them have the same effect. That’s true for pretty much every drug, every mutation, and so on. So for years we called them escapees. And then, around 2015, when Fallon joined the lab, she recut some of those one-headed escapees and found that they also do 70/30. That 70% of them became double-headed and 30% didn’t. And so what we realized was that they’re not actually escapees, they’re not unaffected. They’re affected. But the way they’re affected is quite different. They are randomized. They can’t tell if they should be one-headed or two-headed, and they flip a coin with a 70/30 bias about what they should do in any given generation. In fact, we were able to show that, when you cut multiple pieces from the same worm—and we call them cryptic worms; cryptic because physically they look completely normal, one head, one tail, they look normal. But they’re not normal. Because if you recut them, they’re not sure what to do. Their memories are bistable. So what happens is that you can cut them into pieces, and every piece makes its own decision if it’s going to be one-headed or two-headed, even though they came from the same parent organism, with the same roughly 70/30 frequency. So that’s another kind of permanent line.

And the way we studied it more recently was the kind of perceptual bistability. So like the rabbit-duck illusion, right? You look at it and it looks like one thing, looks like something else. That’s kind of what’s happening here. There’s a bioelectrical pattern that can be interpreted in one of two ways, and that’s why they’re confused. They’re sort of bistable and they can fall in either direction. So that’s another thing we did in plenaria.

1:50:26

Levin

Yeah. Okay, so the way this works is basically that, when you’re finishing up your postdoc, you do what people call going out on the market, which means you interview at a bunch of places and see who will hire you as a brand new junior faculty member. So when you get a job, and so this is considered your first real independent job, because you are now in charge of all your successes and failures, it’s all up to you. And typically, yes, at that point, you get a lab. So one of the things you do is you negotiate the amount of space you have. Typically you start off pretty small. And then over time, if you bring in new grants, then you ask for more space and the lab grows. And when they say Levin Lab or Huberman Lab, what they’re just referring to is all of the research that is controlled by you, where you make the decisions, you know, where that particular person makes the decisions. So, for example, I have three different locations on this campus where my research is done. And that’s just because there isn’t one continuous space. I mean, I would be happier to have everything in one place, but they’re large spaces. And so there’s a few specific locations. All of them are considered part of the Levin Lab because, as the principal investigator, I’m the one whose job it is to bring in the external funding to support all the people that work there and to pay for the reagents. And also, I’m the PI of these labs. It’s their job to be responsible for the good and the bad of what they do. So that’s what it means. It just means you’re the principal investigator responsible for determining what research happens, and what happens to that research, who does it, you’re hiring, you’re recruiting people, you’re writing grants. That’s what it means.

1:52:28

Levin

Yeah. So some of the latest most exciting things that—oh, by the way, one thing I didn’t mention that happened between 2015 and 2020 or so was also Vaibhav Pai’s—he’s also a staff scientist in my lab—we were able to show that you can actually use reinforcement of bioelectric patterns to fix birth defects in frog. All this started in frog. And we were able to show that there’s a range of birth defects induced by other means that you can actually repair. So complex defects of the brain, face, heart—like these really complicated things—you could actually repair them by very simple changes in the bioelectrical pattern. So I think that’s a really important, actually, story, because not only is it a path to clinical repair of birth defects, but it actually shows how, again, you can take advantage of this highly modular aspect, and you don’t have to tattoo instructions on to every cell for something as complex as a brain or a heart. You can give pretty large-scale instructions, and then the system does what it does to fix it.

Okay, so the next couple of big things after that were, first of all, the discovery of xenobots. And this was Doug Blackiston, and my group did all the biology for it. And this was in collaboration with Josh Bongard, who’s a computer scientist at UVM, and his PhD student at the time, Sam Kriegman, who did all the computer simulations for the work. So this was the discovery of the xenobots and the idea of using epithelial cells from an early frog embryo—so prospective skin cells—to liberate them from the rest of the embryo, where they’re basically getting a bunch of signals that force them to be this like, you know, boring skin layer on the outside of the animal keeping out the bacteria.

1:54:29

Levin

They were comm—well, it’s hard because skin isn’t, you know, it’s not a real precise term. But it’s cells that, at the time that we took them off the embryo, were already committed to the fate of becoming that kind of outer epithelial covering. They hadn’t matured yet, but they already knew they were going down that direction. So we were able to show that when you liberate them from those cues, they actually take on a different lifestyle, and they become a motile sort of self-contained little construct that swims around and does some really interesting things, including making copies of itself from loose skin cells that you provide and so on.

Shortly thereafter, you know, a couple years later, we were able to make anthrobots, which is the same thing, but with adult human tracheal cells. And part of that—so there’s a couple of reasons why that’s important. Very simply, you know, some people saw xenobots, and they thought: well, amphibians are plastic, embryos are plastic. Maybe not shocking that cells from a frog embryo can reassemble into something else. But basically, this resembles a phenomenon in frog developmental biology known as an animal cap. And—

1:55:51

Levin

It’s called an animal cap. The animal cap is just basically that top layer of skin cells, of prospective, ectodermal cells. It’s called an animal cap. So some people thought about this as a unique feature of frog developmental biology. And so I wanted to get as far away from frog and embryo as possible, because I wanted to show that this was kind of general and a broader phenomenon. So what’s the furthest you can get from embryonic frog? Adult human. So Gizem Gumuskaya in my group, who just defended her PhD about a month ago, so she developed a protocol to take donated tracheal epithelial cells from human patients, often elderly patients, and let them assemble into a similar kind of thing, a self-motile little construct that swims around on its own.

And one of the most amazing things that these anthrobots do—and this is just the first thing we tried. So I’m guessing there’s a hundred other things that they can do. But one thing that they can do is: if you put them on a neural wound—so, in a pedri dish you can grow a bunch of human neurons, you take a scalp and you put a scratch down through that lawn of neurons. So there’s a there’s neurons here, there’s neurons here, there’s a big wound in the middle. When the anthrobots come in, they can settle down into a kind of what we call this superbot, which is the collection of anthrobots. And four days later, if lift them up, what you see is, what they do is they take the two sides of the neural wound and they knit them together. So they literally repaired across that gap.

So you can sort of start imagining. I mean, there’s a couple of things. On a practical level, you can imagine, in the future, personalized interventions where your own cells—no heterologous synthetic cells, no gene therapy, but your own cells—are behaviorally reprogrammed to go around your body and fix things in the form of these anthrobots. You don’t need immune suppression because they’re your own cells. Right? And so you can imagine those kinds of interventions. It’s interesting because a lot of the interventions we use now, we use drugs, and we use materials, you know, screws and bolts and things. And then occasionally, we use like a pacemaker or something. But generally, our interventions are very low agency, you know? And ultimately we’ll have like smart implants that make decisions for you. But mostly they’re very low agency kinds of things. And when you’re using low agency tools, the kinds of things you can do are typically very brittle. In other words, this is why it’s hard to discover drugs that work for all patients. You get side effects, you get very differential efficacy in different patients, because you’re trying to micromanage at the chemical level a very complex system.

1:58:28

Levin

Exactly, yeah. Oh yeah. Yeah, yeah, yeah.

1:58:32

Levin

Exactly. Which is why your own cells, coming from your own body, they share with you all the priors about health disease, stress, cancer, you know. They’re part of your own body. They already know all of this. And so part of this is understanding how to create these agential interventions that can have these positive effects. We didn’t teach the anthrobots to repair neural tissue. We had no idea. This is something they do on their own. Like, who would have ever thought that your tracheal cells, which sit there quietly for decades, if you let them have a little life of their own, they can actually go around and fix neural wounds, right?

1:59:22

Levin

True. True. Yes. This is true for a lot of stuff in our lab where people say, “Well, did you know that was going to happen?” And so, on the one hand, no, because it’s wild and it wasn’t predicted by any existing structures. On the other hand, yes, because we did the experiment. And that’s why I did it: because I had an intuition that that’s how this thing would work. So did I know that it was going to specifically repair peripheral ennervation? No. But I did think that among its behavioral repertoire would be to exert positive influence on human cells around it. And so this was a convenient assay to try. We have a hundred more that we’re going to try. There’s all kinds of other stuff.

1:59:55

Levin

Correct. You have to start somewhere, right? And so Gizem and I said: well, why don’t we try a nice, easy neural scar? There’s many other things to try. So that’s kind of a practical application. But the kind of bigger intellectual issue is, much like with the xenobots, what’s cool about making these sorts of synthetic constructs is that they don’t have a long evolutionary history in that form factor. There’s never been any xenobots. There’s never been any anthrobots in the evolutionary history. The anthrobots don’t look anything like stages of human development. And so the question arises, where do their form and behavior come from, then, right?

And so this is where you get back to this issue of the platonic space, right? If you can’t pin it on eons of specific selection for specific functions, where do these novel capabilities come from? And so I really view all of these synthetic constructs as exploration vehicles. They’re ways to look around in that platonic space and see what else is out there. We know normal development shows us one point in that space that says this is the form that’s there. But once you start making these synthetic things, you widen your view of that latent space as to what’s actually possible. And I see this research program as really investigating the structure of that platonic space and the way that mathematicians—you know, people make the map of mathematics, right? And this is sort of a structure of how the different pieces of math fit together. I think that’s actually what we’re doing here when we make these synthetic things, is: we’re making vehicles to observe what else is possible in that space that evolution has not shown us yet. Yeah.

And then you can do interesting things, like—and this is, you know, still unpublished, but you can ask questions like: what do their transcriptomes look like? You know, what genes do xenobots and anthrobots express? And, you know, without blowing any surprise, the paper should be out sometime this year. Massively new transcriptional profiles in these things. No drugs, no synthetic biology circuits, no genomic editing. Just by virtue of having a new lifestyle, they adapt their transcriptional profile. The genes that they express are quite different, quite different. So that’ll be an interesting study.

And then, you know, for the rest of it, I mean, what we’ve been doing in the last few years is trying to bring a lot of the work that we’ve done earlier into clinically relevant models. So the cancer stuff has moved from frog into human cells and organoids, spheroids. So human cancer spheroids, and glioblastoma, colon cancer, stuff like that. The regeneration work has moved from frog into mice, and it’s coming along. I’m not claiming any particular result yet. I should also say there’s an invention. What do you call it? A disclosure here I have to do, because we have a couple of companies now. So I have to do a disclosure that, in the case of regeneration, so Morphoceuticals is a company that Dave Kaplan and I have. So David is a bioengineer here at Tufts, and he and I have this company aiming at limb regeneration, and more broadly biorelectrics in regeneration.

So yeah, so the cancer, the limb regeneration stuff. You know, more experiments in trying to understand how to read and interpret the information that flows across levels. So we know cells exchange electrical signals to know how to make an embryo. Turns out, embryos actually communicate with each other. So that’s been a really exciting finding recently for Angela Tong in my group—we just got her PhD as well—where we studied this embryo-to-embryo communication, showing that groups of embryos actually are much better at resisting certain defects than individuals. And they have their own transcriptional profile. So I call it a hyper embryo, because it’s like the next level. They have an expression and transcriptome that is different from normal embryos developing alone, let’s say. So that’s pretty exciting. And yeah, that’s those are the kinds of things we’ve been focused on.

2:04:22

Levin

Boy! Well, step one is to ignore most people’s advice. So I don’t know how helpful that will be. But I actually have a whole thing about—maybe we can put up a link—I have a whole long description of this on my blog. So on my blog, I have a thing that basically talks about advice.

2:04:47

Levin

Yep, Gizem Gumuskaya, yeah.

2:05:14

Levin

I mean, the most important thing I’ll say is: I do not in any way feel like I could be giving anybody advice. I think that there are so many individual circumstances that I’m not going to claim I have any sort of—

2:05:32

Levin

Yeah. So this is pretty much the only thing I can say about any of this. That even very smart, successful people are only well-calibrated on their own field, their own things that they are passionate about. They’re not well-calibrated on your stuff. So what that means is that if somebody gives you—so it’s kind of meta-advice. It’s all about advice on advice. And the idea is that when somebody gives you a critique of a specific product—so let’s say you gave a talk or wrote a paper or you did an experiment, and somebody’s critiquing what you did, right—that’s gold. So squeeze the hell out of that for any way to improve your craft. What could I have done better? What could I have done better to do a better experiment? What could I have done better in my presentation so that they would understand what I want them to understand? That’s gold.

The part where everybody gives large-scale advice—oh, work on this, don’t work on that, focus, don’t think of it this way, think of it that way—all of that stuff is, generally speaking, better off ignored completely. So people are really not calibrated on you, your dreams in the field, your ideas. It does not pay to listen to anybody else about what you should be doing and how. Everybody needs to be developing their own intuition about what that is, and testing it out by doing things and seeing how they land.

And I think that most everything that we’ve done along the way that’s interesting—and certainly we’ve had plenty of dead ends and made plenty of mistakes—but most of the interesting things that our lab has done along the way, very, very good, very successful smart people said: “Don’t do this. There’s no way this is going to lead to anything.” And so the only thing I know is that nobody else a crystal ball. Paths in science are very hard to predict. And people should really be very circumspect about taking extremely seriously specific critiques of specific things that will help them improve their process versus these large-scale sort of career level things that, yeah, I don’t think you should be taking almost anybody’s advice about that.

2:08:05

Levin

Yeah, I mean, the minutiae happens every day, because every day we get comments on, let’s say, a paper submission and we see, and somebody says, “Well, you know, it would have been better if you included this control,” or “I don’t get it because…” You know, it’s clear that the reviewer didn’t understand what you were trying to get at. And so that’s on us to describe it better, to do a better experiment that forces them to accept the conclusion whether they like it or not, right? The best experiment is one that really forces the reader to a specific conclusion, whether or not they want it to go there. It’s irresistible. It’s so compelling, you know? It’s clean, it’s compelling. So that kind of stuff happens on a daily basis where you see what somebody wasn’t able to absorb from what you did, and you say: okay, how can I do this experiment better? What kind of a result would have gotten us to a better conclusion where everybody would have been able to see? So that stuff happens all the time.

The other kind of thing—I mean, I’ll give you an example from the tail regeneration kind of era. We showed that, when tadpoles normally regenerate their tail, there’s a particular proton pump that’s required for that to happen—a proton pump in the frog embryo. And so what we showed is that you can get rid of that proton pump, and then the tail stops regenerating, and then you can rescue it by putting in a proton pump from yeast that has no sequence or structural homology to the one you knocked out of the frog, but it has the same bioelectric effect, right? And that’s how you show that it really is bioelectricity, right? So we had two reviews on that paper, and the first reviewer said, “Oh, you found the gene for tail regeneration. That proton pump is the gene for tail regeneration. Get rid of all the electrical stuff. You don’t need it, you found the gene for tail regeneration.” The second reviewer said, “Oh, the gene obviously doesn’t matter, because you just replace it by the proton pump, by the one from yeast. Yeah, get rid of all of that and just do the bioelectrical stuff,” right? So that shows you right away that two different perspectives, right—each person had a particular way they wanted to look at it. They had exactly opposite suggestions for what to throw out of the paper. And only together do those two perspectives explain that what’s going on here is that, yes, the embryo naturally has a way of producing that bioelectrical state. But what actually matters is not the gene, it’s not how you got there, it’s the state itself, right? And so that kind of a thing, those kind of perspectives.

Or, you know, the people who are upset at, for example, calling xenobots bots, right? We call them bots because we think it’s a biorebotics platform. So one thing that happens is that you’ve got the people that are sort of from the organicist tradition and they’ll say, “It’s not a robot, it’s a living thing. How dare you call it a call it a robot!” And part of the issue is that all of these terms, much like the cognitive terms that we talked about, it’s not about is or isn’t a robot. It’s simply the idea that by using different terms, what you’re signaling is: what are some of the ways that you could have a relationship with it? So for example, we think that we might be able to program it, to use it for useful purposes. That’s what the terminology “bot” emphasizes. Do I think it’s only a xenobot? Absolutely not. I also think it’s a proto-organism with its own limited agency and its own things that we haven’t published yet, which we’re working on. It’s their learning capacity and so on. So you often run into that is that; is that people think everything should only be one thing, right? And that this is all a debate about which thing is it. And I don’t think that’s true at all.

There’s another—just, you know, kind of one last example. Again, having to do with terminology. Somebody said to me once—people are very resistant to the use of the word “memory” for some of the things that we study. And she said, “Why don’t you just come up with a new term, you know? Schmemory or something. And then nobody has to be mad.” You can say: okay, human learning is memory, and then this other thing where these other things learn, well, that’s schmemory. And that’s the kind of—

2:12:26

Levin

Yeah, yeah. Exactly. And that’s just an example of the kind of advice you might get from somebody. And in a certain sense, it’s true that if you do that, you will have fewer fights with people who are very purist about, you know, they want memory to be in a very particular box. You’ll have less fights with those. Those are true. But bigger picture, though—imagine: so there’s Isaac Newton and the apples—I mean, I know it didn’t really probably happen—but the apple’s falling from the tree. So gravity, I’m going to call gravity the thing that keeps the moon in orbit of the Earth, and then I’m going to call schmavity the thing that makes the apple fall. That way there won’t be any arguments, right? Like, yeah. But what you’ve done there is you’ve missed the biggest opportunity of the whole thing, which is the unification. The fact that actually the hill you want to die on is that it really is the same thing. You don’t want a new term for it. So that’s just an example of the kind of, you know, it’s good advice if you want to avoid arguments. But if your point is that what actually is, that’s the whole point, we need to have a better understanding of memory. This is, you know, I want those arguments, then that’s something else. And that’s the kind of strategic thing that you should decide on your own what you want to do.

2:13:53

Levin

Yeah, you could. My point is—

2:14:04

Levin

That’s correct. Yes. No, true, true, a priori, you don’t know. And so the question is: in your own research program, which road do you want to go down? Because if you commit to the fact that they’re separate, you don’t try the unification, right? You try the unification, you spend your time—I mean, it takes years, right? You spend time and effort in a particular direction because you feel it will pay off. If it were truly, you know, it could be this or that, are you going to spend ten years on one of these paths, right? You really need to, in science, there’s no do-overs. Like, you commit, those ten years are gone. So you need to have a feeling or you need to have an intuition of which way it’s going to go. And you definitely don’t need to declare ahead of time that I know how it’s going to turn out, because you don’t. But you do need to know, despite everybody telling me this or that, I’m going to commit. That’s really all you have, right? You don’t have any kind of crystal ball, you don’t have a monopoly on the truth. But what you do have is a responsibility to manage the limited time that you have. So how are you going to spend your ten years?

And it’s going to be hard, right? Lots of blood, sweat, and tears. It’s a hard job. There’s constant criticism. And that’s how science goes. Lots of stress. But so now the question is: are you willing to have that stress following somebody else’s research agenda or yours? You’ll still be old and stressed out by the end of it. But the question is: will you have tested out your own best ideas or somebody else’s view of what science should be? That’s my only hope.

2:15:34

Levin

Super. Thanks so much.

2:15:37

Levin

Thank you so much. Yeah, it’s great to see you again. Great discussion. I love talking to you. Thanks for having me so many times. It’s been really excellent.

2:16:09

Levin

Thank you so much. I really appreciate it. Thank you.