Watch Episode Here
Listen to Episode Here
Show Notes
This is a ~1 hour discussion with Adam Safron (https://www.adamsafron.com/) and Max Shen (https://www.maxkshen.com/home), and Michael Levinabout diverse intelligence and its relevance for pain.
CHAPTERS:
(00:00) Personal pain and prediction
(04:00) Physiological patterns as agents
(11:46) Multi-scale goal contagion
(22:36) Valence beyond neural circuits
(34:21) Therapy, placebo, regeneration
(44:10) Communicating with bodily intelligences
(56:56) Modeling emergent pain agents
PRODUCED BY:
SOCIAL LINKS:
Podcast Website: https://thoughtforms-life.aipodcast.ing
YouTube: https://www.youtube.com/channel/UC3pVafx6EZqXVI2V_Efu2uw
Apple Podcasts: https://podcasts.apple.com/us/podcast/thoughtforms-life/id1805908099
Spotify: https://open.spotify.com/show/7JCmtoeH53neYyZeOZ6ym5
Twitter: https://x.com/drmichaellevin
Blog: https://thoughtforms.life
The Levin Lab: https://drmichaellevin.org
Transcript
This transcript is automatically generated; we strive for accuracy, but errors in wording or speaker identification may occur. Please verify key details when needed.
[00:00] Max Shen: Maybe I can just jump into it and give some context. The context of this involves a personal story, but I'll keep it brief: how I came to this research. A few years ago, I was a grad student at MIT using model-based and model-free RL to study how humans solve puzzles. As I was working one winter, my hands started to really hurt and seize up. I went to MIT Medical; the doctor and the specialist he sent me to gave me an explanation in terms of my hardware: tendonitis, RSI, whatever. Things began to get worse until I basically couldn't functionally type or open doorknobs. For a couple of months I was confounded: what is happening here? It was confusing because I'm quite athletic, and only a couple months prior I was able to do pull-ups. I started to experiment a lot with different interoceptive things, like sensing into what the sensation is, getting more precise about that, and, to some extent, exploring the memories that came up. That was what effectively let me resolve the pain condition. I was so astounded by that that I reoriented my research. I really wanted to understand what was going on because it was bizarre to me that I didn't have an accessible explanation. As I started to talk to people like Professor Fan Wang at the McGovern, she said, "Yeah, of course that makes sense. Your nervous system is implicated in your pain." But I still felt that we don't have good, clear models besides the vague idea of biopsychosocial factors being implicated. Having come from cognition, I tended toward the predictive processing, active inference lens, and it's helpful to think about pain as a trapped prior, especially chronic pain when there's no structural damage. I'm mentioning this because it gives context for how I'm orienting towards this in a pragmatic and personal way, and to express that the ideas of both Friston and yourself transformed my Umwelt and my felt experience. Having clarity on what was happening changed this cognitive prior and many other things. I think it would be really cool to explore the intersection between somatic cognition and pain with you both. It was through my interactions with some of the active inference researchers that I was brought to Alexei Tolchinsky; after one of the blog posts you put up, I reached out. I met Adam at a conference; there was a clear synergy and I think we both enjoyed each other's energy. I have a bunch of things I'm interested in exploring, but I'll pause there and see if there's any resonance or comments.
[04:00] Michael Levin: A very interesting and important topic. I'm not an expert on pain or anything like that. The closest that we come to some of these ideas is our attempt to understand persistent physiological states as agents within the body, distinct from the material structures, the cells and tissues and subcellular components. I really think that there's a lot to study in terms of the kinds of issues that we deal with, which are the competencies and disorders of morphogenesis. Embryonic development, regeneration, aging, cancer suppression, metamorphosis. I think in all of these things, we still largely lack an understanding of agents that are persistent patterns, not physical structures, but patterns moving within the excitable medium of the body, which is more than neurons. Neurons are an excitable medium that store and process patterns, but so are all the other cells. They can do this at other timescales in another modality. We're very interested in some of those issues as far as what priors the tissues have and whether failure to regenerate is basically a persistent and in fact false belief among the collection of cells that they are not going to be able to regenerate properly. I think we're just beginning to get a grasp on those experimentally.
[05:34] Max Shen: And so that makes them legible, or my sense is that you have to look when you're studying another organism in terms of the physiological states from this pretty third-person objective perspective. But when it comes to trying to approach the affect of some of these states, how might you approach something like that?
[06:03] Michael Levin: The one thing that we're interested in, among other things, because we want to understand how to modulate and impact those kinds of states and how they limit what's possible in the body, oftentimes we try to think about things from the perspective of the state itself. We too are a metabolic pattern, a temporary metabolic pattern. We have mental states and memories and goals and preferences and thoughts. I think of all of these things as a continuum. James said that the thoughts are thinkers too. It's a spectrum. We are also patterns, but we spawn off other patterns. There's a continuum between fleeting thoughts, and then persistent, intrusive thoughts that are harder to get rid of, and maybe they do niche construction in your nervous system to keep themselves around longer. Much larger things, personality fragments, in dissociative identity disorder, then full human personalities, and then maybe some sort of transpersonal, bigger things after that. All of these things, I see them as a spectrum, which means that if you have persistent states, it might make sense to think about what degree of agency they have as a cohesive unit. What does the world look like from the perspective of the pattern itself as an active agent in the body?
[07:42] Max Shen: Do you want to share what you're typing in the comments, Adam?
[07:46] Michael Levin: Adam, he's got some good stuff there. Why don't you talk about this?
[07:51] Adam Safron: I was thinking in your work, Mike, talking about memories themselves as kinds of agents that are good at being there and they're basically constructing their niche to persist. So thinking of persistent pain from its perspective, as you're saying, as a kind of active inference agent and even somewhat more fearsomely in terms of the ability of pain to persist as a multi-scale competent active inference agent where the prediction can play out on multiple scales and be reconstituted on multiple scales. But that's not necessarily fearsome; it's also an opportunity for many interventions. I was just reading the paper you shared, Max — was it the "fighting monkey" method? And what you were describing earlier of different ways you can send your pain body or the message that maybe it can relax some of its priors.
[08:53] Max Shen: One thing that's really interesting is that, in my own experience with pain and in that of a lot of people, it's not spatially continuous. You don't have to have that to call it an agent. To what extent are we differentiating between whether the persistent pain pattern over your whole lifetime is the agent, or do we say at different stages of time? How do we map the continuity of that type of agent?
[09:30] Michael Levin: In my framework, the mapping of any agent at all whatsoever is an empirical matter. You formulate models of the agent, which include hypotheses about the size of its cognitive light cone, the biggest goal states that it can pursue, the space that you think it's working in, and what those goals are, and then you do experiments and see what that gets you. From there, you revise your model accordingly. I don't think we can say ahead of time, at least I certainly can't, what that's going to look like. I think we have to do experiments, which is quite difficult. I think that's the only way to get it optimized.
[10:16] Max Shen: One thing I remember from the team paper is the notion that what unifies cybernetic agents is their goals and also stressors from the goal. If you were to think about a persistent pain pattern as the agent, is the goal — it's hard to know exactly what goal this is, but some people suggest it's distraction from some traumatic memory, or maybe its goal is persisting over time, or it's multiple of those goals. Then would stressors be things that are removing the persistent pain pattern and it tries to not be eradicated.
[10:56] Michael Levin: It seems plausible to me. I don't think that mere persistence is as central or as unique a primary cause as standard evolutionary theory would say, that everything is about persistence. I'm not sure that that's really the primary biological imperative. But it's certainly a plausible one, and it seems reasonable that whatever degree of agency some of these physiological patterns have, not being wiped out by various attacks, which the various pain reduction techniques in effect are, if you're the pain pattern, trying to resist those and persist seems like a high-likelihood hypothesis to look at.
[11:46] Max Shen: One thing that also comes to mind with Adam, your messages is this pattern of the social contagion of certain pain responses. There's this book, "The Geography of Madness," that catalogs this. I think one case was bulimia in Hong Kong. In the 1980s, there was basically no one who reported it. Then there was a publicity campaign to expose it as a health problem. That actually induced a lot of the behaviors and patterns associated with it. You could also think about this as a super, larger-than-organism agent through time. But those seem much harder to empirically test.
[12:43] Michael Levin: Social contagion is interesting. I gave a talk just the other day about bioelectricity showing the way that bioelectric patterns can be convincing or not convincing and can spread. You can imagine the whole notion of an embryo is basically a kind of social contagion where all of the cells buy into the same vision of what they're going to do in anatomical space. They're committed to the same; they've got the same story about what they're going to do. If you have cells that are descending from that or defecting from that story, you're going to have various teratomas and tumors and embryonic defects. In a certain sense, all goals are a kind of contagion among their parts in terms of lateral. So there's going to be some top-down control where the collective, once it gets going, the collective itself can shape the option space for its parts, but also there are going to be lateral effects. We discovered about a year or two ago now this thing called the CEMA effect, C-E-M-A, stands for cross-embryo morphogenetic assistance. It's the idea that we found as a matter of experimental results that large groups of embryos resist teratogens way better than small groups or singletons. It's because the embryos are communicating and we showed how they do it, but they're communicating to strengthen each other's commitment to a particular morphogenetic outcome, and being in a large group and doing that for each other is very helpful. It can help them resist all kinds of teratogens and other things that would otherwise pull them off mission. In that talk I gave yesterday, I showed some movies where there's a group of cells and they have a particular voltage. Then there's another cell over here and it has a different voltage and it's crawling around. It comes and it just barely touches—bang. As soon as it touches, this cell acquires the same voltage as all the others. Then it walks over and joins them in a mass they are building. So this idea of—we have many examples of goal states, physiological goal states which end up being transcriptional and ultimately anatomical goal states—propagating across tissues and being moved. We have data. Memories do this too. If you have planarian flatworms and you train them to a task and then you cut off their heads, they will regrow a new brain. The memories migrate into the new brain and basically imprint on the new brain so that the animal can exhibit recall again. I think the whole social contagion thing goes across levels, goes all the way down.
[15:50] Max Shen: One thing I'm confused about, and this might be that I should familiarize myself more with some of your papers, is that it seems like the unit of bioelectricity is often at the level of cells communicating to each other. I wonder, because it's harder to do it at the level of entire organisms, or once you get bigger there's more chaotic effects, how you think about cells versus organs versus organ systems interacting in the interfaces between them?
[16:27] Michael Levin: I think the best model for this we have is polycomputing, which is this idea that there are multiple overlapping, interpenetrating agents within the collective at different levels, different scales, and they all have their own interpretation of what's going on. For the vast majority of what we do in bioelectricity, we are not talking to individual cells. The messages we send are meaningless to individual cells. They are interpreted by the collective as a large-scale, tissue-level scale of some size that we can experimentally derive.
[17:06] Max Shen: When you use optogenetics to change the sodium channels and the electric patterns, that is at the level of a cluster of cells or an individual cell, right?
[17:16] Michael Levin: Yes and no. Physically, no matter what anybody does, if you wanted to zoom in to the level of molecules or cells or anything else, you can. You and I are having a conversation right now, and if somebody wanted to describe it at the level of air molecules moving around, you could. And you would say, "What do you mean? It's all physics. There's nothing but physics going on here. I can see all the molecules move. It's just physics." You can always do that. But the real question is what is the level at which you can cash out a model of interpretation that gives you the most fertility as far as new experiments. If we had a conversation and somebody gave a physics description of the air movements.
[18:04] Max Shen: It's ridiculous, yeah.
[18:05] Michael Levin: But certainly not as fertile for understanding what's actually going on and making and thus doing new things like having a conversation themselves. It wouldn't be helpful. The same thing is true here. Yes, you can track down what individual cells do or what the visual proteins do when they get hit by light. But the interpretation, the useful interpretation here is actually at the level of the collective. When we modulate the pattern, it really doesn't matter which cells precisely we hit. It's very forgiving in terms of those low-level details. It's very coarse-grained over that. We know it doesn't matter which ions we use as long as we get the voltage right, and voltage is a very coarse-grained system-level parameter. More importantly, the message that we transmit is, for example, "build an eye." No individual cell knows what an eye is. It doesn't mean that message has zero meaning at the level of a single cell, but it does have a meaning to a large region that actually can make an organ. You could say that the message you gave was "there's a photon here," or the message was "be depolarized at this region of the cell membrane." My claim is those are not exactly wrong, but they're completely useless for figuring out how to design therapeutics like that. If you want to design therapeutics, you have to understand how the collective is interpreting that message, and why it is going to be an eye versus a limb versus a heart. That's where the specificity of the message is.
[19:41] Max Shen: Adam, do you want to say something about "The Body Keeps the Score" in terms of morphological predictions?
[19:55] Adam Safron: You could try. So it seems—I think I'm begging some questions in terms of what's higher and lower, but still the idea of a control hierarchy and moving across abstraction layers, levels of organization as you move from cells to tissues, but different ways that the prediction could play out. You mentioned this, for instance, earlier: communication about the prediction at the level of fascia or the vascular system, the way you have muscle tension in the body itself as a kind of prediction about where you would be in the world. You could also have not just the way the body is shaped and the physical mechanical system, but the way the reflex arcs are assembled—that the stance, are you guarding yourself or not? And you might not even know you're doing it. You might be bracing for a blow without realizing it, trying to protect yourself but cutting yourself off and not realizing that's what's happening. Or it might be explicit belief on some level. When I've lived with chronic pain for periods of my life, there's a certain sense that you almost become identified with it. It becomes part of your self-concept. You're almost attached to it. There's multiple levels where the pain engram can reconstitute itself. I was also wondering—you were mentioning social contagion—about the different levels at which we can couple with each other and transmit our emotional states. You have the bulimia example that could happen at a very high level—the way you schematize yourself and something you could talk about—but also the way we might couple just in the way we hold ourselves when we're together, subtle motor entrainment or chemosensing entrainment, even what's being shared, or to what degree we have potentially bioelectric field coupling. I was thinking about the different levels at which pain engrams can reconstitute themselves and be transmitted, and seeing how these different levels can play out when they're all coupled.
[22:36] Max Shen: That reminds me of what we call a lot of parts of culture. It reminds me of an anecdotal observation by some meditative teachers: if you go to different cultures and you ask people to point where they feel that their self is located, there's a different pattern. In Asia, the self tends to be located more in the heart or chest. In the West, people tend to locate themselves in their heads. I don't know how empirical that is, but it feels like an important and useful question to ask — the social element of this. One question I have is the degree to which "pain" is the right word to even talk about these things. What I really like in your work is how you try to move away from this consciousness question, and you say we probably can't understand consciousness before we understand cognition. I've had this experience, and I'm calling it pain, but it feels like such a nebulous term that people have really — it doesn't feel that useful in many ways. I wonder how you might think about the language or articulating different ways of thinking about the phenomenon I'm pointing to.
[24:12] Michael Levin: Language, in any case, is a very late-coming add-on to all of this. I would think that the earliest pain is the initial depolarization of a microbial cell being penetrated or damaged by something that causes the voltage gradient to collapse and the cell to depolarize, and I think that's probably the first painful event. As far as consciousness goes, my perspective is that it goes all the way down. I think that was actually a bona fide painful event.
[24:58] Max Shen: From the paramecium fleeing.
[25:00] Michael Levin: Paramecium to me is a no-brainer. I'm concerned about much, much lower levels than that, the harder cases, but the paramecium for sure is having a pain experience. I don't know what to say about language and how it gets put on top of that.
[25:24] Max Shen: I had this sense that the basic evolutionary story in my head was Paramecium and maybe a little bit before that, when there was this reflex arc to retreat from a potential source of damage. That is the foundation of everything we're calling pain and all of this circuitry. The Urbilateria added more complex motor responses to flee and to coordinate. And when animals were in a more social context, some of the machinery for affective response and theories of mind came to play. That added a dimension to this pain-like experience. But it sounds like you want to extend that even more before Paramecium. Would you alter or question any of the narrative that I just described?
[26:23] Michael Levin: The way I see it differently is that with a story you just told, which is certainly the plausible conventional story. It's very focused on embodiment in the three-dimensional world. So when you're saying it runs away from the pain, it moves physically through physical space, and that's fine. But I think biology was navigating many other spaces long before it could move through physical space. So there are desired and undesired regions of metabolic space, physiological state space, transcriptional state space. And I think that many things that we think of as completely immotile, disembodied, whatever, are actually navigating lots of other spaces that we just have a hard time visualizing. I'm interested to know what transcriptional pain looks like, what is happening in a gene regulatory network that is being chased into a region that has negative valence for it. Admittedly, these are all very much work in progress types of things to be able to cash that out. But I don't think it's just an issue of conventional physiology and then motility. I think those are just obvious instances of a much broader set of phenomena. So, what does it take to punish a molecular network? Well, when it's in paramecium shape, it's pretty clear what to do. But in other embodiments, those same molecular networks, we don't really know. And our tendency is to say that's impossible. Chemistry doesn't have valence; it doesn't care. But obviously some kinds of chemistry do. And I just think we have a very narrow view of what that is.
[28:13] Max Shen: And one thing that strikes me with what you're saying about the patterns themselves being agents is the first time there is a capability. We can go back to transcriptional series, but even with the paramecium, the ability to respond—can you say that that's the first instance of a new type of agent appearing in our evolutionary timescale? This is a pain-like agent. When it's out of the range of imminent danger, the agent goes away. Does that yield experiments?
[28:52] Michael Levin: Possibly, we got to do experiments. The only thing I'm sure of in all of this is that these things are not to be answered with linguistics or philosophy; we have to do experiments here. I don't know, but it feels like a reasonable hypothesis. The origins of valence in unconventional embodiments are a really critical and fascinating aspect. Some of our extremely minimal models, like sorting algorithms, look to be... There's something interesting that happens. We study these sorting algorithms that all CS students study in their first years. What's interesting is we're putting in the frustrated algorithms. I think geometric frustration of the kind you get in crystals and plane tilings and magnetic domains and icing models is real frustration. It's a minimal instance, but it's real frustration. In those algorithms it's interesting. There are things that it's doing because the algorithm made it do those things, namely sorting numbers. The algorithm tells it to sort numbers. But there are also some other weird things that it's doing that are nowhere in the algorithm. I'm sure it'll end up being a continuum and not binary, but there are things that it's doing because it wants to. I think we see there a minimal model of intrinsic wanting or intrinsic goals, which are goals that you have not because somebody forced you to have them mechanically, but for some other reason that we don't have a good vocabulary for. It struck me the first time I realized what was happening there, because in these sorting algorithms they have this curve. Eventually, because the physics of that minimal world makes you sort the numbers, by the time the numbers got sorted, whatever else you were doing gets ground into dust. You're going to the physics of it, much like in our universe. Whatever cool things you can do that are not prohibited by physics during your lifespan are going to come to an end, and you will eventually be a boring sorted array of numbers, and all of those patterns will be wiped away. But in the meantime, until that happens, you can do these strange little side quests that are not actually prescribed. They're neither prescribed nor prohibited by the physics. You get to do them. That is where I would start to look for the minimal origins of frustration, of valence, of goal seeking. For all these kinds of things, we don't know. I don't think we know where that kind of valence comes from. But I would look for it specifically in those kinds of emergent goals, not the ones that you were given because the biochemistry or the algorithm forced you to have these goals, but actually the ones that you developed that nobody saw coming that are intrinsic.
[32:40] Max Shen: And how might you address someone with a more conventional stance that you need certain neural or somatic circuitry to experience affect? What do you think might be convincing to the majority of people who believe this?
[32:58] Michael Levin: If I had a knockdown argument for that, it would be great, but we don't. However, my intuition would be simply this. I would say, in the greater universe the higher probability priors are that the universe is filled with all kinds of very weird and wild kinds of life, and it is completely implausible that in the whole universe the only creatures that actually feel pain are the ones that have our precise kind of neural architecture. That seems crazy. If some alien landed on your front lawn and came out and said, "Oh my God, this is a long trip, I got such an ache in my tentacles," do you say, "No, you don't have the same — you're just a philosophical zombie"? If that's the case, we can take the next step and say, if it's not the same neural structures that we have, what is it then? And as soon as you've cracked that door open, all bets are off because it's all about goal directedness and where do goals come from in the first place and all those questions that I've been poking at.
[34:21] Max Shen: To swap to a more pragmatic lens, these perspectives, it strikes me that Psychotherapies do something important. We don't seem to have tools to express or to try to capture what's going on. The RCT, the standard consensus for that, was that we used to try to determine.
[35:18] Michael Levin: What does RCT stand for?
[35:20] Max Shen: The randomized controlled trial.
[35:22] Michael Levin: Oh, I see.
[35:23] Max Shen: The randomized controlled trial that we use to test the efficacy of a particular drug does not seem to make very much sense when you think about the interaction between a therapist or a practitioner and the modality they're using and the patient. Given TAME and given your perspective, I know this is outside of the usual wheelhouse, but how might you think about understanding this at the level of human therapeutics, non-pharmaceutical therapeutics?
[35:57] Michael Levin: A lot of this is outside my expertise, but a couple of things come to mind. I'm just looking at Adam; he keeps putting interesting things in chat. I should shut up for a bit and let you talk about that. One is this notion of a therapeutic alliance. My understanding from talking to people is that one of the most predictive aspects for therapy and many different kinds of therapeutics is the alliance and the belief on the part of the patient.
[36:42] Max Shen: Therapist cares about them, yeah.
[36:44] Michael Levin: That leads to the other area, which is placebo and nocebo effect. Fabrizio Benedetti has this phrase that he uses. He says, "words and drugs have the same mechanism of action." I think that in his work and other people in that field, it's interesting and important that the mechanical biochemical story, much like we were talking about a minute ago, of why certain drugs work and don't work may be largely at the wrong level and that the actual explanation might be a much higher-level thing that has a lot more to do with the expectations and the beliefs, not necessarily of the patient at the human level, but of the tissues of the patient, of the physiological systems of the patient. We have a minimal example of placebo effects in molecular networks. You can do a placebo in the molecular network in terms of associative conditioning. I can do associative conditioning of a stimulus that actually triggers a particular response node together with a neutral stimulus. The neutral stimulus doesn't do anything. If you pair them, the system eventually learns that they're associated, and then you can induce the effect just with the neutral stimulus alone. That's a kind of placebo because in effect, you've got a stimulus that actually isn't a trigger response, but because of your past history, you now believe that it does, and therefore it does. That work, the papers on Pavlovian conditioning or associative conditioning in GRNs, I think are a minimal model of a placebo. If a five-node molecular pathway can do it, then humans should be subject to it.
[38:44] Max Shen: I think that the therapeutic alliance, I'm familiar with that literature. In my more anecdotal experience, there feels like a really big difference between how people approach it. There's a lot of stuff that's not just the therapeutic alliance. I'm wondering if you have heard of "Internal Family Systems," the therapeutic model. It's Minsky's "Society of Mind," but trying to heal people. To me, that seems like we're taking the level of agency, sub-unitary agents, and we are trying to have this discourse and unify them. That process of unification leads to more coherence and a greater level of settledness. That seems like one instance where you're applying tame-like thinking to this therapeutic interaction.
[39:43] Michael Levin: The first time I saw it, Alexei Tolchinsky was pointing out to me that it's not new and that this kind of idea has been around under other names for a while, which I don't know anything about. It looked like, "oh, this is polycomputing" — this is a polycomputing frame applied to your own behavior, which, like any good frame that captures what's going on, makes it more effective to do what you need to do on the system level. So it seems quite plausible.
[40:18] Max Shen: There's an interesting set of bodily modalities or somatic therapeutics. Have you ever heard of the work of Moshe Feldenkrais?
[40:32] Michael Levin: No, it doesn't ring a lot.
[40:33] Max Shen: He was a physicist, a cyberneticist, who later turned to motor learning and how to heal people. He was in the Francois-Joelle Curie's lab, and he was the main engineer for their experiments in radioactivity. He had a really bad knee injury, and in the '50s and '60s, he swapped from being a physics professor to studying motor learning and how people can heal and coordinate their limbs better. There's now a pretty famous, within certain communities, method called the Feldenkrais method, where it looks like you are lying on the ground or sitting in a chair, and you pay very close attention to gentle movements you make. The act of paying very close attention causes the body to reorganize around making those types of movements more easeful. What struck me as I was reading the TAME paper and the section on regenerative medicine was that that's a decades-old instance of this type of internal alignment or trying to create larger coherence in the system. We haven't really studied it; it feels like we don't have the tools to really study that. That's another example of TAME being useful.
[42:00] Michael Levin: I think that's reasonable. We have certain results that look like they might. For example, in the frog leg regeneration work, typically adult frogs don't regenerate their legs. We put on a wearable bioreactor that has a certain payload of drugs to convince them to do it. But one of the things is that even an empty biodome has some effect. I suspect that what's happening there is that what we're providing with an empty biodome is a degree of agency for the cells over their environment. It's a very small, contained, protected milieu where any molecule that cells put out, instead of floating off into the infinite bath of the frog tank, is actually constrained in that space and is still there for the cell to read. Over time, the cells can feel that this is a protected environment in which what I do matters. Therefore, it makes sense to put some effort toward regeneration; if I was just waving out there in the harsh world and I was going to get banged around on various objects, I'm never going to regenerate anyway. Why even try? Why spend the effort to regenerate? We haven't proven this yet, but I have a feeling that what's happening there is the degree of safety and control that you're providing for the cells is step one. You have to get them that it's even possible — it's feasible for them to regenerate. And then you might want to do some other things. I think those things are going to end up getting buy-in from the physiological networks, rather than trying to force them. Ultimately, to me, that's the road to therapeutics that actually work in a permanent way and don't exhibit so many weird off-target variable effects as the cells try to fight what's an obvious hacking attempt. I don't know exactly how that intersects with pain, but I think these are all related issues.
[44:10] Max Shen: I was going to go where Adam was commenting about joint actions and also this idea of 'tame' applied to health. One preprint that I wrote with my movement teacher was an exercise in what happens when we try to think more seriously about health as intelligence in different problem-solving spaces. I'm especially interested in this at the level of an individual and where you want to orient, go, and solve problems. I wonder if you would add any refinement or specificity to the types of problem solving and the types of spaces that we might call healthy. I think it's a very generative idea that how our vestibular system solves problems is really important, and so is how our liver solves problems in this metabolic space. How can that be brought forth and translated into something more practically guiding for the medical industry?
[45:44] Michael Levin: Martin Picard had a whole-day symposium at some point earlier in the year around what is health, trying to define the notion of health. That's non-trivial entirely. The way that we are addressing the biomedical aspects is as part of more broadly the diverse intelligence agenda, which is to develop translation tools to communicate with unconventional beings. I'm very interested in producing tools that help diverse minds communicate with each other; they inhabit different spaces.
[46:27] Max Shen: Like liver could be a diverse mind.
[46:29] Michael Levin: That's just the start of it. Talking to your organs, that's actually a project that's in the lab now, using AI to develop a way to communicate with your organs and ask and hopefully give instructions. So instead of, hey, Siri, it's going to be, hey, liver, why do I feel like crap? Let's talk about my potassium balance.
[46:53] Max Shen: How about improving interoceptive ability or the existing communication networks? Some people can have an incredible interoceptive ability; they can sense very specifically where there is tension or where their organs might be feeling different. That's worthy and interesting as an avenue to explore too.
[47:21] Michael Levin: We don't have any facilities here to study the natural human case, but what we are interested in is to develop tools to allow anybody to do that. The idea is there should be a scan that isn't scanning molecular states, it's scanning affect of your various body organs. How is the liver actually feeling. My suspicion is that there needs to be an impedance match between what you're looking for and the tool that you're using. The reason that physics doesn't see minds, it sees mechanics, is because it uses low-agency tools, voltmeters and rulers. If you want to see minds, you need to have a mind as a detector of it. That means that if we want to read molecular states, great, you'll have a molecular test and you'll learn about molecular things. If you want to learn about Pryor's beliefs, counterfactual thoughts, and to whatever extent the different organs have that, then you're going to have to have an agential sensor on the front end of it, which might be AI, it might be a biological. We're working on that now. Adam points out something interesting, that the first thing we found anthrobots to be doing is healing other cells. The first thing we found is that anthrobots will heal neural wounds in vitro. All of these things have the capabilities. These are your tracheal epithelial cells that have the capacity to form a motile little being that will heal other types of wounds. I think the degree of healing potential is probably enormous. We are just extremely mind blind in our hyper focus on chemistry, the level of chemistry. We're just totally bad at communicating with all these things. I think we can get better.
[49:33] Max Shen: That's very moving, actually, the brethren healing each other. What you've just said — that's an incredible thought. Thank you for sharing that. The impedance of the tool and the thing being measured, then, that's what you were saying in the TAME preprint about there being a blurring between first- and third-person science, right? Because third person is this pretense that the thing you're measuring doesn't have its own perspective, or the thing you're using to measure. It strikes me that this is pretty — I'm not sure how real this is, but one thing I'm recalling is that in certain traditional Chinese medicine circles there are people who reportedly can generate bioelectricity. There are a few cases I'm aware of of people talking about being able to affect tumors as they put their hands and cause the tumor to shrink. And it seems at least naively plausible that there is some kind of unifying thing or you're convincing those tumor cells to become part of the whole again.
[50:55] Michael Levin: We don't know. In those cases there are all kinds of data going back a long time: improved cardiac scars, immune system function, and all kinds of cancer—many other things being affected that way. In the case of cancer in that scenario, I actually don't know what's happening to those cells. Are they normalizing and joining the normal collective the way that we have done in some of our cancer therapeutics, or are they being killed off, cleared off? We don't know. What I've heard from people who do this kind of work, the healers, is that it is less effective to visualize "the bacteria are going to be dead, I'm going to kill the bacteria." That is less effective than visualizing, for the benefit of the entire system, "just do whatever is good for you." And that seems to be more effective. It matches our experience in the regeneration arena because we can, for example, try to micromanage genes, cells, tissues. That's a lot of effort and it usually goes wrong and you get all kinds of mismatches and so on. What works very well are very high-level signals that say, just build whatever normally goes here. I'm not gonna try to tell you what it is. The amazing thing is that the bioelectric trigger to regenerate a tail in the tadpole is exactly the same as to regenerate a leg in the adult frog. I've never seen a tail form where the leg goes or a leg form where the tail goes. We do not tell it what to build. The most effective intervention we found is a system that is a signal that just says build whatever goes here. I find that parallel, where you're leveraging as much as possible of the natural competencies of the system. You are not trying to force it. I think that's the direction we want to go from the molecular approaches, where you really try to be in charge of all the microstates, to the much higher levels where you can hopefully just let the system do what it already knows how to do. But of course, the devil's in the details of how to do that in the general case.
[53:32] Max Shen: Would you like to share some of what you're typing, Adam?
[53:37] Adam Safron: I was thinking of the case of cancer that you're bringing up. It seems like one model that's coming to mind is if for whatever reason, the cells and the tumor are getting the message that things are not copacetic, the cybernetic entropy of the collective is not being managed. They become untrustworthy and defect on the ancient alliance and say, "I'm not going to constrain my growth." I'm going to make a break for it, take as much as I can, or not even think it; their cognitive light cones constrict, and they start seeing the rest of the agent as an externality. But if you can somehow, by whatever means, help the cells to feel safe again, feel trusting, then maybe they can come back into the fold. Some sort of spiritual master comes to you with this benevolence and confidence. Maybe via the quality of their bioelectric field or by multiple levels of entrainment. We need to figure out how to scale that and create an army of spiritual masters to go around the world, laying their hands on people.
[54:54] Michael Levin: That's very interesting, right? In our approach, what we do is we functionally reconnect the cells to their neighbors. And in a way, it's what you're talking about, that as soon as we force that aspect of it, suddenly it's "I'm part of this mind meld again." Of course, I should be building a kidney. Why would I be trying to crawl off as an enemy? No, this is clearly what I should be doing. There's a danger here, which is that what I probably should have foreseen but didn't, is that a lot of people read my take on cancer. They immediately project it into the social sphere and e-mail me. I get one of these per day. Somebody says, "Oh, I got it, I got it. I saw your thing about cancer and I know what we need to do. We just need to borg ourselves together into this giant human syncytium and then life will be good. No more selfishness, no more antisocial." That's been tried repeatedly. It always leads to the same result, which is horror, and you can't just project these things into this human sphere, because you end up with a gap junction, exactly, some sort of human gap junction. People ask me this constantly. I get it, but the problem is that you will get a larger collective that does new things. But as we've seen, that larger collective really doesn't care about the welfare of the parts any more than our body does. I always use the same example: you go rock climbing, you have a fantastic day, which leaves a bunch of your skin cells on that wall. Who's worried about them? They weren't asked what was going to happen to them. And this is right. So I don't know whether there's a version of this that is relevant for the greater scale for humans, but it is not going to be by giving up the autonomy the way that cells do. I'm not afraid.
[56:56] Max Shen: It strikes me that there's probably new types of—you might call them parasitic or at least non-sympathetic agents that arise in the new interface and that can hijack the whole in a different way when there are gap junctions. You need a different type of immune system when you're scaling novel agents. Given we have a couple minutes left, I was wondering: my current research is very interested in what is this minimal agent that can, using active inference, characterize associative learning or this placebo effect and translates over time into what we might call chronic pain. I'm wondering if there are questions that you think are interesting to ask in this computational domain about this. One that you mentioned already is this physiological state—can we model pain as an agent itself?
[58:06] Michael Levin: To me, some of the most interesting things would be a model of agents doing active inference, but then looking at that system to see if we can identify novel emergent perspectives. Are there patterns? I'm going to put up the link shortly; something just got accepted from us on the interaction between economic game theory, prisoner's dilemma and morphogenesis. One of the things you see when you do these kinds of simulations is that there are large-scale patterns that show up that are not overlapping with the boundaries of the physical agents. You get these novel things. What is their perspective? I think that would be very interesting to check. I apologize. I've got to be at another meeting. Thank you. This has been fun to think about. I'm happy to talk again at some point and see what can be done in practice.
[59:07] Max Shen: Thank you so much both for your time. Have a great rest of your day.
[59:10] Michael Levin: Great to see you. Thanks.
