Conversation #1 with Alexey Tolchinsky, Anthony Weiss and Chris Fields

This transcript is automatically generated; we strive for accuracy, but errors in wording or speaker identification may occur. Please verify key details when needed.

Main Episode

[00:00] Alexey Tolchinsky: And

[00:00] Anthony Weiss: So yeah. Sure. Yeah. Well, it's just great to be here. And, again, I just wanna say how much I enjoy the content you're putting out, Mike. I think for me, it's helped restore some of the mystery, curiosity, and enthusiasm about biology that I haven't had since I was an undergrad, studying entomology at University of Wisconsin. So thank you for that. I think today, we wanted to talk a little bit about the frames within modern psychiatry and psychology, and to some degree, how they may be holding us back from further progress in those fields. And I'll just say a couple of words about myself, again, avoiding long introductions, but I think that it's a little bit relevant for the conversation, because I've been a psychiatrist now. I'm a psychiatrist, neuropsychiatrist, behavioral neurologist at Harvard, on faculty there, and I work at Beth Israel Deaconess Medical Center. And I've been a psychiatrist for 30 years. And so I entered the field in the heart of what was then called the decade of the brain. And there was extraordinary enthusiasm at that time around clinical neuroscience. The Huntington gene had just been fully decoded. We had MRI, for really the first time clinically, to allow us to see within the cranial vault, beyond what CAT scan was allowed us to do. And we had a number of new medications that were entering the field. And so for me, I was excited to enter this space thinking that the field, as I entered it, would change radically in my career. And what I've seen over the past 30 years is extraordinary advances in all of those domains, especially in the neuroimaging field, and in our understanding of the genetics. And yet despite amazing advances in neuroscience and in understanding the brain and the genetics as it's associated with brain function, I'd have to say our ability to help people with mental illness is actually worse. And I think if we judge this in any sort of objective way, I'd say that the outcomes of the work that psychiatry is doing as a whole today is probably worse than it was 30 years ago, when you look at suicide rates, levels of depression, levels of societal distress. So it's a conundrum. And what I've heard is, well, we gotta close, you know, bench to bedside. We gotta close that gap. And I think there's more that we can learn about the bench from the bench, but I think our approach at the bedside is holding us back. And I'll just say a couple words on that, and then I'm interested in certainly in Alexi and Kristen and Mike, your perspective. But what I'm seeing is that our approach to descriptive analysis of what people are coming to us with is still very rudimentary. It's still very checklist oriented. You know? I mean, my iPhone could diagnose someone with major depression because it's check the box. You know? Do you have this, this, this, and that? We also are focused way too much on the individual mind or brain, in a deeply interconnected world. And I think that when I am seeing someone clinically, I'm seeing someone who is already a hybrid. They're a hybrid of everyone that they're interacting with. They're a hybrid of their interactions with technology, and yet we're focused on checklist on this particular individual. And I think our descriptions are very static. You know, it's literally called the mental status exam, cross sectional, very static, and yet all that we see clinically is massively dynamic, and we're not really accounting for that. So, yes, I think there's much we can learn at the bench, but I think our approach at the bedside needs to change because, otherwise, what we're doing is trying to match very rudimentary descriptions of clinical phenomena with massively elegant pictures of the brain and the genetic aspects of that. And I think that's not going to move us forward as a field.

[04:54] Alexey Tolchinsky: I agree with everything, Anthony, you said. It was very eloquent and useful. And if I can pick it up and maybe add a perspective of just mathematics, because Chris is here. And I try to focus on just a few aspects of what you're saying. Because we have theories from which we develop models, and from there, we have therapeutics and diagnostic processes, it will look into how we model mental health and psychopathology. And, Mike, I know you like Mandelbrot. You showed his picture the time. I like his quote very much, that clouds are not spheres, mountains are not cones, coastlines are not circles, and bark is not smooth, nor does lightning travel in a straight line. That captures perfectly how we, I think, in graduate schools in psychology and perhaps medical school, psychiatric residents, we continue for the most part to model the brain and the mind as a car or a watch or a computer. And that I think is the heart of the issue. And there's three components of that: determinism, modularity, and lack of context. So with determinism, I think the idea there goes back to nineteenth century physics, as there's a law and there's a cause leading to a certain effect. So when the patient walks in the door and they present with some kind of distress, unpleasant feelings, ruminating thoughts, insomnia, and a whole bunch of other stuff, very intertwined, I think the first thing that happens is clinician tries to make sense of what's going on. And we call it a case formulation. Maybe a component of that will be diagnosis. And the idea there is there's a hidden cause. There's something discrete, clear, categorical that is leading to the symptoms. That goes back to Emil Kraepelin many years ago, and with DSM-III was revitalized, called neo-Kraepelinian tradition, that there is a condition that is causing the symptoms. And it's amazing that, you know, for major depressive disorder, there can be 227 combinations of symptoms. But we're gonna say the only thing responsible for all of that variety, and that's called MDD. And that's the thing. That's the clear-cut entity that is sort of universal. And that's the first thing we do: that's the cause already. The patients feel maybe a touch of relief, somebody name it for them. And we use static labels. I have MDD, which results in me feeling hopeless. That's the first explanation. And we usually go down. The hidden causes, the relationships are vertical underneath. But we don't stop there. We, research-wise, like, why MDD? Why do people feel depressed? And then we go even further down, maybe in brain regions. You know, Eric Kandel talked about how preoccupied we are with where things are on the imaging studies. And we say things like dorsolateral prefrontal cortex, with anterior insula, with lateral amygdala, is together contributing to depression. Or we go to molecules like serotonin and glutamate, or we go to the genes. And again, we look down to say that that thing is responsible for depression, which is responsible for the symptoms. And in so doing, I think we commit the myriological fallacy, where we take a phenomenon of the higher scale of the entire mind and the whole person, and we attribute it to a part. Just like, say, an amygdala is a fear center, or insula is involved in interoception, or default mode network is involved in looking within.

[08:08] Alexey Tolchinsky: And that’s how we model things vertically and with going down. And related part to that is modularity, where we carve the nature of its joints, and we do it anatomically, chemically, functionally, and diagnostically. I've already said that major depressive disorder is a clear-cut difference from generalized anxiety disorder from bipolar disorder. But also, I mean, we hear the stories about the amygdala. Where is the boundary of amygdala and other things? What does amygdala mean if you disconnect all the projections? Right? Is there border patrol there? And we use LEGO pieces, and we then assemble our theories from LEGO pieces. Sometimes we use a larger LEGO piece like default mode network as if it's the, you know, ontological entity with clear boundaries from cell in network and front of parietal network. It's not. I mean, it's the concept we created, and we may see boundaries on the MRI, but it's very averaged out, not just temporarily, but also with other people. But we talk about it as the thing, and then we start building a theory. So default mode network did this and cell length network did that. The two major issues with modularity is we don't talk about how the components interact. And also, in a very fundamental thing is what Chris, you did with James, separability. We assume that modules are separable, that their states are separable from one another, and then we can build these theories. By the way, none of what I'm saying is new. It has been said for decades by Walter Freeman, Society for Case Theory and Life Sciences, Luis Pesso's book in Henkel Brain. So modularity. Right? And the last thing is context. Depression is depression in Indonesia and Portugal. Right? Lexapro is Lexapro. Look at randomized clinical trials. Zero context. You assemble a thousand people that have the same clear universal thing called MIM DD, and then you give them Lexapro, then you do effects. There's no context. And anatomically, we're not saying, you know, insula, you know, is doing this under that condition of generalized arousal with this state of thalamus and that state of ventromedial prefrontal cortex. We're just saying insula is doing that. So all our models are built this way. Why is this important? It's because of Chris. Chris, your work was very influential on me where with James, maybe you can talk about that, you showed mathematically within quantum information theory that in order for us to separate the two systems, we must have context. And if we strip away the context, the separability is a flawed assumption. They're not necessarily separable. But we do both in DSM and elsewhere. We use separability and we strip away the context. So to me, I mean, I talk too much, but these are the foundational issues of how we model things. Essentially, we model them as a car engine that is decomposable into parts, fix the faulty part, put it back together, it works. We don't model things like waterfalls and tethered kites, which is more appropriate, and but I'll stop talking for a moment and if there's time, I'll talk about sort of what we could do instead.

[11:22] Alexey Tolchinsky: Yeah. So Yeah. Go for it, Chris.

[11:36] Chris Fields: Well, but I think the question we're faced with here is how can work in developmental biology and perhaps biophysics help this situation?

[11:51] Alexey Tolchinsky: Right.

[11:52] Chris Fields: Either from the point of view of modeling tools or from the point of view of concepts that work in morphogenesis or developmental biology in general better than these kinds of nineteenth-century metaphors that you said are still dominating thinking in mental health. And I have to say that I know nothing at all about mental health as a field or diagnosis or therapy or anything else. So I'm a complete newcomer to this conversation. I'm only familiar with some of the newer modeling tools based on things like the free energy principle, which allows us to get a grip on how systems interact with their environments, but at a very abstract level. And part of the problem that you're pointing to, I think, is a problem of abstraction, abstraction out of context, abstraction out of the history of a particular person, and abstraction out of the details of a particular body that functions in some particular way. So this makes me concerned about how helpful tools like the free energy principle sorts of tools can be. And certainly, we've done these recent experiments in modeling with the free energy principle that Daniel Friedman has led. And how they give us a result that in some sense is obvious in hindsight. So the question is, can we go past that, I think, using these sorts of very abstract constructions that make a lot of assumptions that one could argue biologically aren't valid. So that's I think that's my concern is whether the tools we have can be usefully brought to bear.

[14:29] Michael Levin: Can I ask probably a dumb question, but I also, I'm not in this field, so I don't know. What do you guys consider to be a what's the best case scenario? In other words, if everything were to go well, you had the tools, everything is great, what is your mission? So the patient comes in. The world is however the world is. What do you consider to be a success? What are you trying to do for them? I know what the worst case looks like. I mean, just dysfunction and disorders, we can recognize when things go terribly wrong. But what is the best case scenario look like?

[15:08] Anthony Weiss: I'll just make an initial pass at that. I think Alexey alluded to it before, but as a physician, as a psychiatrist, as a psychologist, essentially, we're faced with uncertainty. We as we all are as humans. Right? But what we're faced with is a patient coming to us with a load of uncertainty about what is affecting my brain, my mind, my life. How do I make sense of this? And then what we're doing, again, relevant to the free energy principle or to Tristan's work, is we're taking a mental map and applying it to what this person is with. And that mental map is maybe a diagnosis. What they're coming in with looks a lot like major depression. And if that's a good fit, then we should have greater predictive power. We should then be able to take action in the world and to be able to help this person. But our mental maps are not really good fits necessarily for what these individuals are coming in with. And as a result, our explanatory capability is starting to, I'd say, maybe it's losing or isn't really where it needs to be. So, yes, I can say this looks like this map, but the benefit of the medications I have to prescribe, the genetics of that, the prognostic factors associated with that are still very poor. And physicians, psychiatrists, psychologists, we're in a we have to take action in the face of uncertainty. And so a simplified model that works is generally better than a complex model that doesn't really work in the majority of patients. But what we're left with right now is a really simple model that kind of doesn't work a lot of the time. And so I think that if we could improve on that model, on that map, and give something to people that have a more explanatory... The other thing is that patients, I guess, are voting with their feet. The models that formal psychiatry or psychology is offering don't really, like, you've got a serotonin deficit, doesn't really cut it, you know, to the patient saying, well, that doesn't really make a lot of sense to me. I think I'm gonna go to, I don't know, the occult, or I'm gonna go to some other frame that might help explain this better. And so I think we're losing confidence of patients that might otherwise seek help from us.

[17:55] Michael Levin: Alex, that's okay. Sorry. Go ahead, go ahead, Alexi.

[17:58] Alexey Tolchinsky: I'll I'll

[17:59] Michael Levin: ask after.

[18:00] Alexey Tolchinsky: Well, I just wanted to add to that what I think nobody said it better than Mark Szolmes. If we table all the diagnostic labels and the chemistry and the inferences about the genes, the patient walks in in distress, and there's a specific quality of distress. Feeling lonely and unattached is one issue, and feeling bored is another, and feeling very anxious is a third one. And we think that the work is done when this feeling is gone, when this particular mental distress, the what the patient tells us through a narrative, I feel bad, when the patient's like, that issue is not with me anymore. And then we make sure it is in a stable fashion for a few months at least. They're saying, well, then we're stabilizing the treatment results and we're saying goodbye. But we deal with the mental level, not at the molecular level, not a genetic level, and the patient says, I'm better.

[18:47] Alexey Tolchinsky: There's not much quantum mechanics to that.

[18:52] Michael Levin: I guess I'm probing, and I don't know if this is where we wanted the conversation to go, but I'm really interested in the spectrum between what I've heard called organic disease versus psychological disease. And I'm interested in situations. So let's say a patient comes in and he says, look. I feel really hopeless. I've been reading a bunch of physics books, and there's this heat death of the universe thing. It's totally bumming me out. I've also been reading a bunch of neuroscience. Now I know that I don't have the free will I thought I had, all this stuff. I'm super upset about all this. What what's going on? And would it be considered a success if you did have some sort of, I don't know, serotonergic modulation where you say, oh, well, yeah, I guess whatever. I guess I'm not worried about it anymore? Like, in some sense, that's a success because you've resolved some anxiety. But one could also argue that it was a perfectly rational response. Right? And that what we've done is something quite different, is disconnect the thinker from the content, right? So I'm just trying to figure out, to what extent are we thinking about a hardware fix to certain things that are legitimate issues? I have no idea how to think about that, but I'm guessing you guys think about it.

[20:25] Anthony Weiss: I think about this all the time because the work I do is largely in the hospital, for people who are in the hospital for other reasons. So most of what they have, you know, they've come in for surgery and now they're confused afterwards, or they've come in for a stroke and now they've got depression or something like that. I think the organic, nonorganic labels, for me, I don't like them. I don't think that they're useful, and I also don't think, you know, materialist, nonmaterialist is a useful approach. I used to think that. I mean, I fell in love with neuroanatomy, and I still love neuroanatomy in the brain. But our understanding of the brain hasn't gotten us further in some of these things. And so what I've generally done is taken a much more functional approach, which is to say, each of us is trying to navigate the uncertainty in the world. That's what we're trying to do. We're trying to survive in an uncertain, infinitely chaotic world. Again, excuse my use of chaos there, Chris, but it's complicated. Right? And we're trying to survive. Now is what this person is presenting with inhibiting them from successfully navigating that world? And I would say that if this person that you described is so distraught by, you know, the heat death of the universe or of the earth or what have you, that they can't function, yeah, well, then I think that's a problem. Right? Just as much as it would be a problem if they couldn't focus or they couldn't calculate or they couldn't remember things, because they're gonna have difficulty in navigating the uncertainty of the world. I've increasingly thought about, we all go into the world with the brains we have, with the computational systems we have, and we can try to optimize those. And then if you stroke out part of that, well, you've gotta figure out, now how do I navigate the world and its complexity without that part of my brain? And so, yes, organic damage to the brain is important, but distinguishing between organic, nonorganic, I don't think is a good long-term strategy for understanding these concerns. But that's just my perspective on it.

[22:46] Alexey Tolchinsky: And if I can add to that, Anthony and Mike, the way I think, and it may be limited, is there's an external event. Let's say we read the news or the patient came up with thoughts, the stimulus came from within, that the world is dying. And then the final outcome of them feeling badly in some kind of specific way. But there's a mediating factor in the middle, the reaction, the perception of what happened in the world. And when I see patients in psychotherapy, we shift the focus from the world. I'm not a coach. I'm not going there with a Phillips screwdriver trying to fix the world. I'm working with how the patient deals with these things in the world. Sometimes nothing can be done. If the patient is chronically abused at home, the first order of business is to get them to safety. That's an issue with the world. That's an environmental problem. But if the patient is in a reasonably safe environment and they have very strong overreaction at a high tone of anxiety, then that's the therapeutic focus. And we try to get to the point that the turbulence within the emotional distress is getting better so that they are suffering less. I also wanted to go back to Chris, your question that you posed, how can developmental biology and Mike's work help? Look at the xenobot. Xenobot is a nonlinear dynamical system. If you decompose it into cells, each cell doesn't have any qualities of a xenobot. Right?

[24:06] Chris Fields: And

[24:07] Alexey Tolchinsky: the secret sauce is in the interaction, the communication, the layers, and the structure. And so I think that we learned so much. And free energy principle, of course. Carl Friston is fully on board with nonlinear dynamical systems. He's got papers with Chris Frith, with, you know, hierarchy of linear operators. And prediction error is nonlinear. You know? But, you know, I just wanna maybe zoom back for a moment and talk about mathematics, Chris, that this is a quote from Stanislav Ulam. Using a term like nonlinear science is referring to a bulk of zoology as the study of non-elephant animals. Right? We would be hard-pressed to find anything linear in any living system, anything modular in any living system, but we continue to model in this way. Right? So let me give you a little bit of data at the risk of being trivial. When I'm talking to you right now, if you do the scalp EEG, my neocortex is in a high gamma mode, which means chaotic. You cannot predict the future wavelets from looking at a bunch of them. But if I were having a seizure, it will be more orderly. If I'm in deep sleep, it is more orderly. If I'm in a coma, it will be quite orderly. And when I'm dead, it's gonna be perfect determinism, flatline, linear system. Right? Now take a heart, another organ. You know, heart rate variability is a good thing. When heart rate variability decreases, that's a predictor of heart failure or disease. Right? Lung has a fractal-type structure. You know, neuron and neuronal network has a fractal-type structure. Take patients with persistent vegetative state. The predictor of recovery is the level of order on the EEG, not the level of order. Order is a bad predictor. And we call our mental conditions disorder. We think order is good, chaos is bad. That's the bias in the field. Right? And we've known about it for a long time, but we continue. That continues to be a fringe, you know, kind of at the margins, nonlinear dynamical systems. And so I think if we shift away from the buckets and decomposing mental health into sort of clear things and look at, you know, how we can model nonlinear systems that exist that even exist in medicine, in neurology and pulmonology, in cardiology,

[26:24] Alexey Tolchinsky: Detection of seizure on an EEG uses chaos theory algorithms. We don't do that in psychiatry and psychology. We continue to put people in buckets and prescribe static treatments like ten milligrams of Lexapro. So when we go back there and we start creating clinical models like waterfalls or tethered kite, then we will do better. In a tethered kite, we have a string, which is a constraint. And then we have wind, which is chaotic. Remove the wind. What will happen to a kite? It's flat on the ground, dead. That's what happens when we remove chaos. And if you remove the string, the kite flies away. But in the middle of that, if you take a more nuanced condition, say a sudden gust of wind, if you try to control it tightly with a string, the kite will plummet. And that perhaps is a metaphor when your partner has an emotional outburst and you try to micromanage your partner. It's not gonna go well. Or an obsessive compulsive disorder trying to control tightly the chaotic world. It doesn't work well. We need both. It's a balancing act. It's a movement. It's a process. And instead of that, we have, like, you have MDD and here is ten milligrams of Lexapro. We diagnose statically. We treat statically, and we model statically. So to me that that's that's just the big appeal is to go back there. And in biology, I think it's front and center that it's nonlinear. Right?

[27:48] Anthony Weiss: There's some level of urgency here, I'd say, because all of what we've talked about, the challenges we're facing right now will get more complicated as we look at people who are literal hybrids. I think we're already hybrids. I'm already a composite of the people in my life and the technology, but I think we'll have actual literal hybrids. And our field has no capacity right now to understand how to help someone who, for example, people even with deep brain stimulators or certainly people that have are gonna have wearable brain-computer interfaces. We don't even have the language for it. And I worry because already the language being used is nineteenth century descriptors like AI psychosis or hallucinations, right? Already the field is embracing these 200-year-old categorizations instead of thinking about what is actually happening here with the AI agent and its interaction now with the human, in generating maps or generating predictions that don't seem to fit reality or just don't seem to be regulatory for the person. So I don't know. I feel like a sense of urgency about this, which gets me excited, but also makes me worried, like, where are the ideas gonna come from? Where how can we move this forward?

[29:32] Michael Levin: Yeah. I mean, I agree completely. I think about that a lot, and Mark Solms and I are writing this thing. I gave a talk a couple of weeks ago to some mental health professionals, basically just making the argument that your clientele is gonna get really, really weird very soon. Like, you're gonna have, right? At night, Jeremy gave me this amazing graphic for, which is this waiting room, and there's a couch in the back, but there's a waiting room. And the beings that are sitting in this waiting room are every kind of, you know, there's wings and tentacles and there's wires everywhere. And so, really, all of these things are gonna get very, very unusual. And I think it raises some fascinating questions about helping to adjust novel beings who are not the same as the evolutionary stream that standard humans come from, if there is even a standard. And how to figure out what is it going to mean to understand the symbols of their dreams and, you know, what's it like to try to help beings that are radically different.

[30:34] Alexey Tolchinsky: Yeah.

[30:36] Michael Levin: I think that's very interesting. And I think we're also heading for all kinds of situations, for example, in court, where we know what diminished is, and maybe there's a brain tumor or a Twinkie situation or something like that. We could get that, but we're gonna end up with people with enhanced capacity. Somebody's gonna walk in there with an extra hemisphere and a 280 IQ or something, and we're gonna have to figure out, do we really think there are only two types of responsibility? The sort of preadult human and post. Well, no. It turns out there's gonna be every variety of insight that you could expect from a being in the world, and what are we gonna do that? I think these are really important questions.

[31:21] Anthony Weiss: Have mentioned for you, Mike, and that is around energy. Because I feel like a lot of the conversation has been on information flow, entropy reduction, but not as much on the flows of energy and, you know, the competition for limited energy within even within the ecology of this human being. You know, and there's a lot going on inside of here, and there's not an infinite amount of energy. And in fact, I think our brains are energetically bounded. We only have so much blood flow going up to this computer system up here. There's only so much ATP that can be generated at any one time. There's not a lot of storage capacity of energy up here, and it's intensely energy utilizing. And so, again, I think a lot about energy, and how that flows between maybe the system subsystems of the brain, maybe between those subsystems and computational systems within the body, between my subsystems and yours, you know, the four of us here. But I'm curious how you've maybe thought about that as you think about very small entities or, you know, organisms that you're growing. How do they compete for energy, and how does that affect their ability to navigate in the world or in their space, in their environment?

[33:03] Michael Levin: So a few things that are relevant to this. First of all, I think more generally, the fact that we all start, life starts out as beings that are fundamentally in a scarcity regime for time and energy, right? The fact that it has an interesting consequence, I think, which is that you can't afford to track microstates. You can't afford to be a Laplacian demon. You're gonna be eaten and dead in no time. So from the very earliest points of life, you are forced to coarse grain and try to tell agentic stories about the world. Like, okay, I can't track all the details, but this set of things, I'm gonna call that a thing that does things and has certain properties. And so it kind of forces you to see agency in the world, so to speak. And then eventually, maybe you turn that on yourself. You say, wait a minute. I'm also a thing that does things. And so I think that scarcity pressure for energy really, to me, it suggests that there should be a drive against taking a very low-level view of the world and instead making world models of the outside and then also yourself of beings who do things. And then you can sort of start to work out the mechanics or the psychology of how do I predict what these things are going to do, and it sort of drives that kind of outlook. So I think that's interesting. You could even imagine some sort of a conjecture where you say, you know, any being that evolves under strong resource constraints is going to believe in free will. I'm not saying they have it. I'm saying they're going to believe it of themselves and of the things around them. You know? Everything looks like larger-scale entities doing things because you have to think that way otherwise, right? So I think the energy constraint is important in shaping that aspect of us. In our creatures, we've measured some metabolic stuff. We don't know terribly much about it. One thing that's always been interesting to me is competition for real estate. So this well-known thing that if you sort of include the vision, then those areas get taken over by, let's say, auditory processing algorithms or whatever. So this notion that there are patterns or algorithms within an excitable medium that are driven to take over real estate, that if something is, but the flip side of that is they're being kept out of there by active, like, you know, there's some kind of territory. Under normal circumstances, it doesn't happen because there's some sort of pushback. So whether that actually boils down to a competition for, as you said, blood flow or something else, it's interesting, I think, to think about algorithms as needing material resources and having this push. But it also makes me very, the fact that that happens makes me sanguine about all kinds of augmentation technologies because if our algorithms are already good at taking over new real estate, maybe if we slap on a couple of new hemispheres, which we can do during the embryonic, like, can make you a chicken or a frog with as much extra brain tissue as you want, maybe there's enough plasticity that these things would actually spread out into new tissue. If you give it new tissue to live in, maybe they're good at spreading out into those things. And I've talked to David Eagleman about testing some of this stuff. Like, maybe there's a lot of opportunity for augmenting the real estate that they have now that we're not stuck with the basic human metabolism. You can sort of augment it with things. And conversely, there are ideas I've thought about in terms of some negative patterns. So you could imagine, right, if you have positive algorithms that do processing and they're good at taking over, there might be decoy scenarios where you can take negative patterns and say, well, okay, I'm not gonna try to sort of wipe you out. I'm gonna give you some place to live, and you can be over here, and that's fine. Just don't bother this other partner. There may be some version. So that's something that I'm in particular interested in, this spectrum Chris and I have written about between thoughts and thinkers and the notion that some of these patterns have their own, some level of degree of agendas and problem-solving capacities. And so then the metabolic questions become very clear. Are they fighting for metabolism or for something else? And if they are competing for metabolism, that may be a whole new set of interventions where you can sort of resolve some of those conflicts by providing, you know, extraneously providing whatever it is that they're looking for in a way that doesn't let them take away from the main system.

[37:54] Anthony Weiss: I love it. Yeah. I think that the cranial vault is probably limited, in terms of, you know, and this is highly speculative, but there's probably some sort of bounds on how much processing it can do because it's probably insulated and probably would overheat if it got too active. And so, yeah, I think adding more hemispheres would need to be outside the cranial vault. And maybe to some degree, we're already doing that with this or in our connections with other individuals who are thinkers. But, yeah, I've just been thinking a lot about energy flow and trying to better understand how that might drive because one of the things that, you know, I think I've been very keen on, Peter Sterling's models of allostasis, or Bruce McEwen or, you know, this idea that we need to dynamically regulate energy. And so what you see breakdown in a lot of psychiatric conditions is that energy regulation. So insomnia, for example, the sleep, the beautiful sleep-wake cycle gets completely disrupted. It happens in almost every single neurologic and psychiatric illness. Dementia, anxiety, depression, schizophrenia, they all have sleep-wake cycle disturbances. But then also appetite regulation. You know, the fact that I'm hungry, I eat. When you're depressed, you either don't eat or you're eating a lot of crappy carbs. So the beautifully regulated allostatic mechanisms that we have as humans break down. And to some degree, I don't know if that's chicken or egg to the changes in thinking that then occur, the obsessionality that Alexey talked about or the ruminative anxiety or the delusions, right? But they seem to come together. And so, and we've spent a lot of time thinking about the latter, about that. And we just think, oh, well, the sleep issue, we'll just, you know, throw Ambien at that or whatever. But I think that the two are very much connected.

[40:13] Michael Levin: I mean, in general, that's something we're very interested in doing, is checking to see whether xenobots and anthrobot and those kinds of things, whether they sleep. And that becomes that.

[40:24] Alexey Tolchinsky: becomes a really

[40:26] Michael Levin: A difficult thing because we could perhaps show cycles of inactivity. We could even perhaps show because we can read calcium signaling out of these. And Thomas Marley in Bongard's lab has looked at various information theory metrics comparing to MRI, fMRI in brains. And the difference to null models is quite similar, actually. So you could imagine applying those kinds of things and saying, well, it looks like they sleep. And then some people will say, well, you've taken the metrics outside of context where they’re sensible. And then other people will say, well, if you believe the metrics capture something, then you have to let them teach you about new systems. And so it's not clear to me how we would resolve that, but I do think it would be interesting to figure out what kinds of systems actually sleep and what does that look like in unconventional sorts of systems. And what is it really? Like, there are very specific things to our well, there's a theory that it's supposed to keep you quiet at night so you don't wander outside the cave and get eaten. These things are very sort of specific to our evolution, but my feeling is it's probably much more fundamental than that. And then the question is where does it occur?

[41:51] Chris Fields: Now if I could go back briefly to a previous point in the conversation in terms of modified humans, I think that a human equipped with one of these, and this is a really old one, of course, is very different from a human not equipped with one of those.

[42:13] Alexey Tolchinsky: Mhmm.

[42:14] Chris Fields: And I think we're increasingly seeing that it's a two-way street in terms of offloading of computations or offloading of effort onto this device, which is helpful. And means that we have to use less energy and less time to do certain things. But the flip side is that the device offloads stuff onto us, and the device demands energetic expenditure and attention expenditure and computational cycle expenditure by using various sorts of incentives. And, as you know, the business models of the companies that put content onto these devices actually depend on offloading their computational needs onto us in terms of data collection. And we're all being given surveys about what we like and what we want and what we find interesting 24 hours a day now. So there's a lot of cognitive effort that's being extracted from humans by technology in the same way and often by the same tools and in the same work sessions as we're offloading cognitive effort onto them to do something that we want. So if we think of these technologies only as effort-saving devices or labor-saving devices or things that we offload onto, I think we're making a mistake because they're also offloading onto us and demanding more energy and more cycles, etcetera. So I wonder in thinking about these kind of enhanced beings, including us, with our technical add-ons, which is fairly recent. I mean, this is only the last 30 or 40 years that we've seen this level of offloading capability. Slide rules didn't do as much offloading in either direction as iPhones do. I wonder to what extent that is affecting metabolic demands, sleep demands, hormonal demands, etcetera, in people, and to what extent is the need to do all this extra work in service of the technology debilitating? Or is it an additional layer of debilitation on top of what you would see without it? I mean, I mentioned hormones here because that hasn't been part of the discussion, but I think as soon as we talk about energy, we're also talking about the whole chemical regulatory system that's associated with metabolism from the mitochondrial level up to the whole body level.

[46:08] Alexey Tolchinsky: But that made me think, Chris, is entanglement in acceptability of us on the phone. Harari makes that point that we're hackable creatures. We use Google to figure out how to get to places, Amazon on where to buy things, Facebook to what's new, and it's getting closer to us. It is getting our attention. But also now with AI, it's pretending to be a friend. The next frontier is intimacy. It wants to really just be a mental object inside our mind. So we become entangled, right, in your terms. And that I think is essential in how we model the us plus the phone. It's no longer a diet. We merge. But I want to also go back to Michael, what you said and your work with Michael's Mark Soames. He, when he gave a lecture at the neuropsych analysis congress in Barcelona, it was called "From Molecules to Mind." And one of the points was there isn't anything in between in how we model things. We have a lot about molecules and genes.

[47:07] Alexey Tolchinsky: Mhmm.

[47:07] Alexey Tolchinsky: Something about the mind, but nothing in the middle. That reminds me of another coauthor. You have Katrina Schleisman, where you have this paper on memory is not storage. She makes two essential points: that we're missing meso theories. We don't have much in the middle. We have a lot about micro and a lot about macro, but not much in the middle. She also uses this very useful question: why are computational metaphors sufficient? And she goes back to hydraulics and electricity and how the heat and cool physics was used to explain mind. Now we're all using computer metaphors. Is it warranted to say that hardware and software are as separable in your language, Chris, as wetware and the mind? It's really not clear. And Anil Seth makes that point as well, that if you take a millimeter of tissue from the temporal lobe, that's 14 petabytes of data. It's like that. Show me the boundary between the mind and the brain there.

[48:02] Alexey Tolchinsky: But when we build our theories, we assumed substrate independence. We think it's the word that you can have for an Apple computer or a Windows computer, but it is really not clear that in the brain it is so, and we can that substrate independence is a warranted assumption. Take that away, and half of computational models are gonna not be applicable anymore. So I think that holding on to the computer metaphor in modeling the brain is also an issue here, but I don't know what you think.

[48:33] Michael Levin: You know, I don't mean, I don't know if these are related, but they seem related to me. The issue of having models at the very lower level and then, you know, at this high mental level. Language, I think, really doesn't do us a lot of favors. I don't know if this is true in other languages, but let's say the ones I'm familiar with, we have a what and we have a who. And that's it. Those are the two options you get. You could be a what or you can be a who. And I think this is fundamentally extremely problematic because we already have and are certainly going to have more of all kinds of things that are somewhere in between. And so I've jokingly proposed that you need a little exponent. So you can say, I'm a WHO three. So, like, I'm a Roomba with some human neurons on it, or I'm a WHO 12, you know, which, like, I'm a full-on human with some extra stuff. So because because the language, as soon as you learn the language, you're sort of straightjacketed with this notion that, well, the world divides into what's, which presumably is the low-level stuff, and then there are the real who's, which but what happened in the middle? And, by the way, all of us were what's at one point, and somehow we got to be who's. So you would think that this mesoscale has to be filled in. I mean, I think, you know, Chris and I and various people have been trying to fill some of that in to give us some tools for that sort of thing. But there it is. In the grammar that we have.

[50:08] Anthony Weiss: Yeah. I think just going back to Chris, your comment is spot on, and it's about this bidirectional flow of control. And certainly, we are these tools to offload computational capacity. We've all I think we've always done that historically, whether that's a hammer or, you know. But this is even more now regulating us. And there's some really interesting books like Jonathan Crary's Twenty-Four/Seven or Bernard Stiegler's work, Automatic Society. Really thinking about, well, what is that doing to us as humans and as a society? I think fundamentally, it's very interesting, and it gets back to the energetics because if we need to sleep, and we do, we as humans do need to sleep, what if the machines don't need to sleep? What if the xenobots don't need to sleep? Right? Because what they're gonna be doing is they're gonna be driving us to a place where what about our sleep? What about our regulatory capacity? You know? And I think we are already being shaped in a way to become more twenty four seven. And I don't know that that's healthy for us as biological organisms. Right? I'm also curious about concepts. I'm working with Marlin Way, just, you know, thinking about some other things related to AI-human interaction. But one of the questions we were asking is why is listening to emotional content so energetically exhausting? You know, if you listen to, you know, as a therapist, for example, Alexi, patient after patient after patient, you need to take a break. And we've bounded those sessions. We've set boundaries on that because I think it's just intuitive that it's exhausting to listen to emotional content. But why? Why is emotionally laden information seemingly more taxing on our computational capacity? And is the same then true for this? Is this gonna get exhausted? Does this move these chatbots, these AI chatbots listening to people, you know, twenty four seven, three sixty five, is that going to affect them, the agents, the AI agents? And so at any rate, that's very speculative, but something that we're thinking about because we need to think about not just how we're using these tools to offload. You're absolutely right, Chris. We need to think about the impact it has on us, and to some degree, the impact that we have then on the agents that could help shape them in a negative way as well.

[53:08] Michael Levin: That related to and you can tell me if this isn't really a solid result or whatever, but I was reading this study where they had people take tests that were sort of ethics problems. Would you do this or that? And half the group was squeezing a spring right beforehand, which makes them physically tired. And then there's a measurable reduction, this was reported anyway, a measurable reduction of ability to sort of do delayed gratification and prosocial reasoning and things like that. So the argument was that it's the same pool of effort, right? That physical effort and exerting a will against that spring is the same as the effort of thinking through the ethics problems and so on. So, I mean, that seems relevant. Right? If all of that stuff comes from the same pool in some sense, maybe it's metabolic, maybe it's something else, then it makes sense that processing these kinds of things would be exhausting. What do?

[54:11] Anthony Weiss: Guys, I think there are a number of examples. Again, these are now, I don't know the sources of this, and they've become lore, but you see a judge, a judge who's at the end of a long day, you're probably gonna get the most lenient sentence. But I think that's, I think there's no computational free lunch. And yet we think about our minds as having the unlimited capacity to compute, especially for complexity or for emotional content. But I don't think that's realistic. And it has to come from somewhere. So, yeah, that's an interesting example. I hadn't heard that about the squeezing the bar, but it makes sense to me.

[55:03] Chris Fields: Well, we've talked a little bit about a need for mesoscale concepts. But, Alexei, you kind of started out this conversation talking about the unhelpfulness of trying to draw a boundary around the amygdala or something like that and around the default mode network and consider it a black box that has a particular kind of predictable input output. So those are mesoscale ideas, but they're mesoscale ideas that, in a sense, are making the wrong kind of approximation. And I think that one thing the free energy principle tells us is that we will make a mistake not thinking of those systems as agents embedded in their own environments.

[56:09] Alexey Tolchinsky: Mhmm.

[56:10] Chris Fields: That are having their own experiences, including experiences of stress or resource depletion or experiences of excitement or overstimulation. And so maybe part of the mesoscale modeling issue is not treating mesoscale systems as agents. And perhaps part of the solution could be actually better models of these mesoscale structures or functional units, functional systems as agents that have environments so that we're forced to talk about what that system is experiencing and forced to consider it as something that is not just doing information processing, but doing information processing in a stressful, resource-limited environment in the same way that the whole person is or the same way that some neuron is.

[57:38] Alexey Tolchinsky: Mindful of the time. I don't know if we have a few more minutes or we must say goodbye, Mike.

[57:42] Michael Levin: I've got four minutes.

[57:45] Alexey Tolchinsky: Four minutes. But I really like what you said, Chris. And, again, it's artificial because when we said micro, meso, macro, we've already separated the scale. We've already assumed separability. So perhaps we need to talk about how things flow and, perhaps in somewhat scale-invariant mode, what happens in the entire hierarchy up and down and left and right, and not just look at the different chunks vertically or horizontally. That's one of my comments. So it is also a metaphor to say we need mesoscale theories, but the message there is that we're really no explanation from serotonin to depression. There's none. Or from the genes to bipolar, there's nothing, and we need to fill in that gap conceptually through theory and model as well. And I think free energy principle is on the way there as one of the possibilities. So yeah.

[58:34] Anthony Weiss: And I agree. I love what you said, Chris. I think that this mezzo layer, even if it is imperfect and even if it is just yet another label or simplification, could be that translational piece that we need to go from bench to bedside to take some of the beautiful neuroscience and our understanding of the brain and genetics at the lower level and actually then be able to apply that clinically, because right now, we're really struggling.

[59:12] Michael Levin: Are you aware of any venues for publishing stuff at this intersection? What's a good, if and when we have things that contribute to this particular set of questions, is there a place you like to put that stuff?

[59:36] Anthony Weiss: I'm excited about the anti kythe project that MIT and others are starting to spawn. I think that's moving away from, I think, core biology and moving more toward the agentic side, but they're doing some cool stuff. I think Psalms and some of the work in neuropsychoanalysis, Aleksey? I think that's a space where you start to see a bit more dynamic understanding of mind, and I think they're a bit more open to ideas. But I don't have good suggestions otherwise. I don't know, Alexi, if you have any thoughts on that.

[1:00:18] Alexey Tolchinsky: Anywhere. We could do a preprint. I think that the message is subdued. I mean, I think the issue is maybe systemic or bureaucratic or something. It is peculiar why physics moved on from determinism, to Boltzmann, and we haven't. I don't know how come we continue to model waterfalls with buckets. I think that it takes some effort to do it.

[1:00:45] Anthony Weiss: But I think it's important not to the established journals will want to see DSM diagnoses, or they'll want to see, I would say don't modify the approaches to try to shoehorn them into those just to get them published in those journals. Right? I think that those models, and I don't want to be too cynical, and we have helped a lot of people. I don't want people to go off their medications or abandon their therapist or anything like that. But I do think we need to move the field forward, and it's going to take new ideas that aren't shoehorned into these existing frameworks or buckets.