Watch Episode Here
Listen to Episode Here
Show Notes
This is a ~50 minute conversation with Michael Johnson (https://t.co/YxAOZif0V2) on vasocomputation, stress as cognitive glue, and more generally computation and cognition in unconventional substrates.
CHAPTERS:
(00:00) Framing Distributed Stress Minimization
(03:16) Vascular Tension As Priors
(10:34) Neuronal Competition For Blood
(16:25) Trauma, Self, And Stress
(23:36) Stress, Valence, And Fascia
(31:34) Bodywide Bioelectric Field Architecture
(38:49) Stories, Delusion, And Flow
(45:32) Hydraulic Computing And Boundaries
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] Michael Johnson: I was thinking could have a conversation about intersections between your work and my work. I've been thinking so much about distributed stress minimization. It's just an absolutely beautiful frame and the various fields of the body. And so I'm coming from the perspective of formal research into consciousness. And then this frame of vasal computation, this idea that tension in the vascular system essentially holds Bayesian priors about the appropriate range of the neural system. To begin, I'd love to hear distributed stress minimization directly from you. What's going on there?
[01:08] Michael Levin: As you know, we're interested in various cognitive mechanisms in unconventional substrates. I've been thinking most recently about how to measure stress in unconventional systems. In particular, for us the use of stress reduction as a driving variable for anatomical homeostasis. The idea of cells and tissues having to adjust both anatomically, but also transcriptionally and physiologically to instantiate specific goal states that they have that then pull them along. Our latest framework, and much of this is still unpublished, what's coming out now is that we can use various stress markers to identify how these loops in embryogenesis proceed stage by stage, and we can actually see the stress. At the beginning, there's a particular target morphology, and we study the one that's set by bioelectrics. There may be others, but we study the bioelectric one. The target morphology doesn't match the current anatomy at all, and so the stress goes up, and it works like hell to get there. When it does, it goes down. By then, the bioelectric state has moved on again. It happens again and again and keeps pulling it along through all the stages, and then eventually it equalizes. Then we can talk about aging and things separately. This is the kind of thing we study. In particular, what I wanted to also discuss: everything that you've seen in the vascular system I want to hear about, but also how do we measure it in really weird systems that are not biological at all? What does stress look like in a gene regulatory network? What does it look like in a physiological circuit? What does it look like in other kinds of processes? Some generic metric that we can apply across substrates.
[03:16] Michael Johnson: I was thinking yesterday about how this distributed stress minimization that you talked about—if a cell isn't in the right place, there'll be stress and cells emergently coordinate to get the cell into the right place. The bound of a system is the empathy of the cells in some sense. I was thinking about how different systems may have different currencies of stress. So first of all, how does that work? What is stress in endothelial cells and what is stress in neurons and what is stress in glial cells? How do they communicate? I had the loose hypothesis that there's an inter-system stress exchange. It would be interesting to see if one of the core levers the body has to adjust, mood or strategy or Bayesian priors, is adjusting the exchange rate of stress between systems. Maybe in a rest and digest mode, stress in the stomach gets weighted 5x normal. Stress gets distributed, but the stress that the stomach is holding gets pushed into other tissues where maybe in a fight or flight mode that's reversed. I think that you're touching on something that feels extremely important when you talk about the cognitive glue and things that bind the system together. To bring in vessel computation, one core hypothesis that I'm trying to look at is that it feels like a very efficient compression to think of the vasculature as an agent. The body is an amalgamation of agents, all emergently cooperating. I think of the vasculature: to what degree can you ascribe different personalities to different agents of the body? I think it's a very interesting question. The vessel computation baseline is that vascular tension stabilizes local neural patterns. Or whatever the local substrate of compute is, that's the slightly more general brain. Vascular tension basically freezes the patterns. It's a neat way to hold certain things as constants where there's tension. Other things can be left as variables where there's no tension. Vascular muscle, of course, is a form of smooth muscle.
[06:54] Michael Johnson: Smooth muscle has this latch bridge mechanism where it can glue itself shut, the actin and myosin stop sliding and get chemically glued. This can persist. My expectation is that these latches, when vascular muscle engages this latch bridge mechanism and these may be actively defended or actively refreshed, or there's metadata from the body indicating we need that, can last anywhere from minutes to hours to potentially decades. A very long-term prior. Vascular tension is gatekeeping the body's central currency, blood flow. If neurons don't get blood flow, they don't have the resources to adapt, to rewire, et cetera. I've been thinking about that as a backup or almost like an RLHF system for the neurons: if some big prediction error happens, the vasculature can leap in and say, okay, that didn't work. We're not going to let that happen again if you get bitten by a dog. It can latch patterns into a known safe mode or bring in this active inference frame: tension as a prediction. You can say you latch a prediction that you will be safe in a certain situation. But that reduces overall system dynamism. You block out certain parts of your dynamic range, which has downsides. The story that I want to tell is that we're an amalgamation of systems. These systems are amalgamations of smart parts, as you've described. I want to say people think of intelligence as embedded in the neural system, and I think your critiques have been absolutely spot on. Neuroscience shouldn't be about neurons only. Neurons are specialized for long-range communication. I would propose looking at smooth muscle cells with fresh eyes: what are these cells? What are they doing? What is their functional role? I see them as playing neurotic protector to the neural system. If the neural system can't handle something, the vascular muscular system jumps in and tries to make it manageable. I want to pause there. Does that make sense?
[10:34] Michael Levin: I was thinking about something a little related recently. In terms of gene expression, you have 20,000 different genes and you don't have the metabolic resources to transcribe them all. One way to think about it is that the cell is making decisions about what to transcribe. Another way to think about it is that the genes could be in competition for the attention of the transcription machinery. If there is this obvious overseer that is going to decide who actually gets transcribed, you would think that there would be some forces to start to hack it, where the genes are trying to get the attention of the system. My question in your case is, do you think that neurons, in addition to doing whatever it is that they're doing, are actually trying to get the goodies from the vascular system? Is that a thing? Are they trying to hack the vascular cells at all?
[11:39] Michael Johnson: They'd have to be, right? I think the powerful thing here is the vascular system is holding the purse strings that if the neurons don't get blood, there's just a hard cap on what they can do. I expect that if we poke into neuron physiology, we'll find that, I'm not sure if it'll be continuous or discrete, but they'll drop into safe mode in a very clean, elegant way. They're designed to go through metabolic winter. We can say that there was this sequence of papers starting with the hemo-neural hypothesis by Moore and Kao, then moving into "Cognition Is Entangled with Metabolism," Jacob et al. There were a couple of crazy unexpected findings. One was that blood flow in healthy tissue can vary by over a factor of 20, not 10%, but 20x plus. So we should expect cells to have different modes of operation depending on that. The second finding was that changes in blood flow precede changes in neural activity. It's not that neurons fire and use a bunch of fuel. It's that the blood changes, the blood comes and then neurons change.
[13:28] Michael Levin: That would be really interesting. Maybe it's already been done, but it would be interesting to do some kind of multi-scale neural network model where the cells are agents, where the neurons are agents that have some pressure to compute whatever it is, but if you want to do the computation you're supposed to do, you also have to navigate your local environment and get the purse strings to feed you.
[14:02] Michael Johnson: I do think that a very interesting near proxy for attention is blood flow. Moronkow described it as turning up the sensitivity or the gain on networks. It'd be interesting to see how neurons compete for blood flow. How do the algorithms of the body punish defectors on that? The cooperation-defection dynamics get pretty deep pretty quickly.
[14:55] Michael Levin: An hour ago, I had a conversation with Frank Putnam and Alexi Tolchinsky, and they were talking about dissociative disorders and the different personalities that can live within one body and how they navigate their relationships. One of the most interesting things he said was that they have shifted clinically from trying to integrate; what they now do is provide a bulletin board where the different personalities can leave each other messages. And so this is a stigmergic medium where you get to leave these messages for each other and communicate that way. As a result, the collective becomes more functional and integrated towards goals, but it's not the thing where you hope the pieces disappear in favor of the whole. So I wonder, in these vascular networks, how many agents do you see? Do you see multiple agents at the same level? Of course there's multiple levels, but at the same level, are there regions? Is the whole thing tightly integrated? Has anybody done any causal emergence metrics on the data?
[16:25] Michael Johnson: I haven't seen that. That'd be fascinating. My expectation is that a lot of this logic is local, and may have local organs as the big attractor or the semi-sovereign domain, I guess I would say. And then the tissue trying to serve that domain. So the stomach being a domain and the heart being a domain, brain being a domain and so on. It'd be fascinating to dig into that. But the mechanisms of how the muscle cells are organized are like the vasculature is not one big muscle. It's a bunch of tiny muscles. And I guess part of this thesis is that a lot of the challenges of being a human living in modernity is that sometimes these vascular punches and latches, these crystallized priors, they're too sticky. An absolute central challenge for humans is how do we release this tension? Not all of it gets properly garbage collected, we can say. We can say, okay, no, I got to pick up my daughter from school. I'm going to remember that. And that's instantiated with a prediction / a tension in a certain part of my nervous system. The vasculature reaches everywhere. Anywhere there's neurons, there's blood flow.
[19:21] Michael Johnson: And anywhere there's blood flow, there's this muscle. But I suspect that as we go through our day making predictions, making active inference predictions, we naturally clench and release. And it's seamless, but sometimes cruft builds up. If we look at this as a form of side-channel memory — of the proper dynamic range, or not using a big part of the range, or remembering a preset pattern — then over time you'll naturally get situations where the resources aren't released to the system. They're just not released. Maybe you were interrupted in doing a task and your body was holding something and it never got the task-completed trigger so it could release. So my expectation is that this is a pretty lean description of, first of all, what we can call trauma. This is shards of information held by tension in your nervous system. I have a friend, Warren Winter, who explained what a Wilson affordance was. As we perceive something, we don't see it as it is. We see the object in terms of what I can do to the object and what that object can do to me. So if you see a dog, maybe you see, oh, I could put that at that dog. But also, that dog could bite me. And of course, if something bad happens, then the negative Wilson affordance might get latched into the system. I see it as a neat explanation of trauma, which would be pretty exciting if this is the home system where that lives. But also, in terms of the Buddhist frame, I've been trying to dig into it for a while. It feels like what is talked about — this self, this immutable set of constants that prevent a wider aperture on reality — might be a particularly good fit for explaining that as well. And that it's not that, if we had a magic wand and we could open all your vascular attention, good things would automatically happen. Probably some of it is very load bearing. But I do expect that if we were to track the fine vascular attention of meditators as they go through the path, we'd see an opening where networks that were latched when you were two or three may reopen and you get a much wider sense of possibility. So those are the two applications that I would be very excited about.
[22:19] Michael Levin: Eric Hall in our center has been developing some more recent new metrics for things like uncovering new levels of causality. Presumably there are lots of data that this could be applied to, to see what is actually going on in that system as far as integration.
[22:46] Michael Johnson: I've been a huge fan of Eric. I think his ECE 2.0 looks really good. I'll have to think about that.
[23:01] Michael Levin: Another interesting thing to look at: there are not that many of these patients, but there are some really interesting exceptional human cases where people have very diminished brain volume and yet normal cognition. I would be interested to know what's going on right in the vasculature and the muscle in those. In that case, is it like taking over some of the processing? Does it matter?
[23:36] Michael Johnson: One question that seems to be at the intersection of what I'm looking at and your work is this idea of distributed stress minimization between types of contractile tissue in the body and neurons. I do have the expectation, first of all, that this system is trying to minimize net expected stress. If you see a tiger and your stomach clenches up, it is a high stress state, but it's lower stress than being eaten. I think it's trying to multiplex stress across Bayesian features and minimize total stress.
[24:46] Michael Levin: Richard Watson and Chris Buckley and those folks have done some models. I don't know how much of it is published, but they've done some models on the computation in networks of springs and the physical stress dynamics and how they allow the network to compute. I think that's very interesting. This intersects with stress as it's understood in cognitive science, stress as it's understood in biology and physics, and also in mathematics, for example geometric frustration. It would be very interesting and helpful to connect all of those. The null hypothesis is that those are all different things that we just call stress. My suspicion is that it is all under the same phenomenon. I think geometric frustration is real frustration. On the very end of the cognitive spectrum, that's alignment — misalignment of your parts is probably a fundamental aspect of stress and frustration in composite cognitive systems.
[26:04] Michael Johnson: In 2016 I wrote a short book, "Principia Qualia", that laid out what I call the symmetry theory of balance. This idea — if we have a formalism for an experience, which is a hard thing to construct, and if we had a perfect mathematical representation of what it feels like to you or to me — the symmetry of this mathematical representation would exactly correspond to the pleasantness of the experience. It's a formal mathematical way of saying harmony in the mind is the thing that feels good. What you say about geometric frustration definitely moves a lot of bells.
[27:12] Michael Levin: What do you think about systems, simple systems in which everything is aligned and harmonious, a magnet or something where everything is nicely aligned? For that minimal system, can we say that it's somehow maximally — it has surveillance at that point?
[27:39] Michael Johnson: I would say that there's going to be a couple requirements. One being the system has to be conscious. The symmetry theory of valence is not a theory of consciousness. It's a theory of valence. If we can point to a conscious system, and construct a formalism for what it feels like to do that system, then I do think that symmetry and formalism correspond; there's an identity relation with valence. A simple magnet may or may not lead to consciousness. I would also say that Eric studied with Tononi, and one of Tononi's frames is that you need integrated information for consciousness. I do expect there to be an interesting trade-off between enough complexity such that you have some integrated information, and enough simplicity such that everything is nice and symmetrical and harmonious. That's maybe worth loving.
[29:08] Michael Levin: Yeah.
[29:09] Michael Johnson: Back in on the distributed stress minimization frame, I'm reminded of the different types of muscles. And I'm a big fan of Joe Bullock's work on muscles are a little bit different than we think they are. And I do suspect that there's going to be this interesting leakage of stress from one muscle type to another muscle type. If you have a lot of skeletal muscle stress, maybe some of your smooth muscle also picks up some of that tension. This may be computationally interesting in a lot of ways, especially insofar as contractile tissue that's not finally regulating the neural system bleeds into this vascular muscle that I think is finally regulating the neural system. There's also fascia, which is a wildcard topic. A lot of people ascribe very interesting properties to fascia. I think it's very interesting that it's so electrically active. It's very conductive. One thing that I've been wondering, and I'm really curious what your instincts say. So fascia can't exactly latch, but it can durably contract. It's much slower than VSMCs. VSMCs can operate on the scale of hundreds of milliseconds, whereas fascia maybe a minute or so to contract. But once contracted, they can rewire, and that's the new default. I'm wondering, I suspect that VSMCs might have this fine regulatory effect on neurons. But what do you think fascia regulate? And what do you think contractions in fascia might, quote unquote, latch?
[31:34] Michael Levin: That's an interesting question. I think an area that might be relevant to this is acupuncture. I've seen some good work — Helene Langevin in Vermont has these really interesting experiments. She's got this full-thickness skin tissue model where she puts the needle in and you get this lateral view of all the layers of the fibroblasts and everything, and you twiddle it. What she shows is that the fibroblasts grab onto the needle, and then by twisting it you're basically making these tensile forces that spread very long range. I'm not an expert on connective tissue, but there's almost certainly going to be some kind of mechanical computation there. There has to be. I would assume nature is using it. One thing that Josh Bongaard and I have been developing recently is this idea of polycomputation, where the body is a collection of observers that is interpreting each other and all the physical events in every which way. I would be shocked if a dynamic like that had no observers paying attention to it.
[32:53] Michael Johnson: Right. Fascinating.
[32:55] Michael Levin: I don't know which processes are tuned into it, but I'm sure something's watching it.
[33:03] Michael Johnson: One thing that comes to mind: I think you're one of the experts, if not the expert on the electric fields of the body. I appreciated your recent tweet about how it's really hard to measure fields and it's often possible to measure V_mem, the membrane potential, and so on. There are caveats: it's hard epistemologically to approach this topic. I'm really wondering: what's the basic electrical layout of the body? As I understand it, cells have a strong membrane where the charge density of the membrane is equivalent to lightning. It's very small but very potent for its size. Mitochondria have even stronger membranes. You also have fascia, which conduct electricity. I understand that bone is essentially calcified fascia, so that's also very electrically conductive. Wounds generate a stronger electrical field, which may help tissues reorganize and heal. From the perspective of consciousness research, electromagnetism is an interesting way to describe the organism and whether consciousness lives in the EM field. That's a rabbit hole, but I'm looking at this in terms of Wolfram's branchial space, where an object's true shape lives in this branching space of possible ways to decohere. I have the idea that a mind is a shape in branchial space. I'm very curious what the body's shape in branchial space could be. I expect that the EM fields of the body would say a lot about our true shape in branchial space. I don't exactly know the questions to ask, but that's the setup.
[36:17] Michael Levin: I haven't yet figured out the relationship between my model and Wolfram's model of that space. I think there is basically what used to be called a platonic space of forms. What we are building when we make cells, embryos, biobots, AIs is we're making interfaces. We're making interfaces to specific patterns in that space. The bioelectric circuits are very versatile in that way, and they can pull down lots of different patterns. In the body, you have a huge number of these things. You have static forces among the molecules. You have different distributions of static charges. You have voltage gradients across intracellular membranes. Golgi, ER, nuclear envelope, mitochondria, all of those things have a voltage gradient. The cell itself has a voltage gradient across the membrane. Even that isn't one gradient. A typical cell has many different voltage domains across its surface. It's like a soccer ball of different domains. Then the cells come together into tissues and epithelia. We have an epithelial, a trans-epithelial potential across them on top of that. Becker and other people measured this weird longitudinal electrical potential, which is body-scale voltage differences over very long range. Those are all the static electric things. On top of that, you've got the electromagnetic components, where some of the voltages change very slowly. The induced magnetic field is very low, but there are other events that produce natural EMFs coming off living tissue and going into the ultraweak photon range, including UV photons. It's an enormous amount of these kinds of things going on at different scales, at different frequencies, all the way from DC, which is what we studied, to light. All of those are interpenetrating at the same time. The cells and the other systems are trying to make sense of all of it. They're in a soup of signals.
[38:49] Michael Johnson: Fascinating. I guess I'm cognizant that I'm sitting with the expert here. And what to you has been the most surprising in studying this system?
[39:10] Michael Levin: Interesting question. The most surprising thing to me so far has been just how plastic it all is. The idea that evolution apparently has really spent most of its effort on creating a system that is able to creatively interpret the memories that it has, whether those are genetic memories or behavioral memories. I've been playing with this bow tie architecture thing where at any given moment, you don't have access to the past, but you're aimed forward in prediction. You have to take the prompts that you've been given, whether those are your genes or your engrams from previous experiences, and you have to construct the story. Every day we see these amazing things that have no evolutionary precedent, but what you're seeing is the evidence that living things are these amazing sense-making systems at multiple scales, and they're using all of these computational affordances, all the different layers of the body, to tell coherent stories that may or may not bear any relationship to the previous story they were given.
[40:33] Michael Johnson: Fascinating. Well, this reminds me of looking at the vasomuscular system and this idea that maybe in some idealized setup, the nervous system is this, I think of it like a set of wind chimes and as sensations come in and hit it, then by the presence or absence of certain hums, you can build a model of your environment. And then you have these latches, these tiny areas of chronic tension. And the resolution on these is between 100 to 400 micrometers. So that would say just in the brain, you could have somewhere between 21 million to 1.3 billion. I'm calling them vascular addressable units of differential tension. And so these can save patterns and over time, we accumulate these little points of tension, which become a predictive story. They not only encode what we expect from our environment descriptively, but what we expect prescriptively; they're very active-inferencey: I will make my environment into this, not just I'll expect this to happen. And so over time, it does seem like we begin to live in this story. And sometimes part of the Buddhist critique is that this is bad in some ways and that it's hard to not live in the story. And this is interesting hearing you say that you have all these systems doing something that feels very similar and that there are stories upon stories in a lot of systems.
[43:01] Michael Levin: I'm working on something I jokingly have provisionally titled Femto-Buddhism, where you ask the question. I've had this discussion with Buddhist scholars. All living beings are under delusion and must be liberated. Cells? Molecular networks inside of cells — what does that mean? Chemistry presumably doesn't make mistakes. Developmental biology definitely makes mistakes. People's typical intuition is that no, that stuff doesn't accrue karma, it doesn't get liberated. It just does what it does, but then you have this amazing living stuff and it has these issues over it. Can you take this all the way down? I'm interested in taking some of these concepts of what it means to be in delusion about your environment and what it means to be in a flow state, and have more or less direct access to this information as opposed to painstakingly and often mistakenly trying to work it out. And how does that relate to least action laws? When you have a photon that doesn't have to worry about calculating all the different paths, it always goes in the least action path. What's happening? So from there we dip into this place where we as living beings have to work really hard; we fight for information and try to figure out what to do. Some of us, some of the geniuses or exceptional people, get into the flow state and they're like the photon again. Does that curve look like that? I think there's a lot to be said about what chemistry is doing from that perspective. The other thing I'm thinking, having talked to you about this, is we have something coming out where we took xenobots, which are novel frog-based constructs. We introduced a nervous system and wanted to see what a nervous system looks like in a being that has never had evolution shape the structure of its nervous system. But we don't have a vasculature there. Now I'm thinking the next thing we have to do is introduce some vasculature and see what that looks like.
[45:32] Michael Johnson: Amazing.
[45:34] Michael Levin: There's something else I wanted to show you before we break. You want to see a really weird vascular hydraulic computer. I don't know if you've ever seen this. This is something Nirosha Murugan worked on in my lab. This is the slime mold. It does mazes and things like that. I want to show you one particular thing that you might appreciate. Here's a branch of the slime mold. Here's another branch. This thing has been injected with these little fluorescent beads. We're just going to track the fluorescent beads. What you can see here is right now it's just acting like a Y splitter in a hose. But each one of these little things is independently addressable because you can see that it shut this off. This thing's still going crazy. This one's completely shut off. This amazing fractal thing has many branch points. If all of them are independently addressable, first of all, what network is controlling which things get turned on and off? You can even see sometimes it goes backwards. I don't know if there's directionality to the actual synapses here, but they're synapses, because you can easily imagine a mechanism where, based on prior experience, this thing gets turned on or off.
[47:01] Michael Johnson: Right. That's beautiful.
[47:03] Michael Levin: So I thought that was cool. And what the heck are they computing, and there should be at least two overlapping systems where there's the hydraulic system, but on top of that something has to be guiding the opening of the branch points and the decision making there.
[47:23] Michael Johnson: Absolutely. Super relevant. It does also call into question: how do you figure out the causality between the systems? Everything is regulating everything else, but can we say, this is the dog and that's the tail? Or whether it's grandeur, causality, or different metrics? I think Eric's work would be relevant.
[47:45] Michael Levin: Thank you so much. I love your work. I think it really opens a whole new avenue of all this stuff. I think we really need to start just looking more carefully at the vasculature and the muscle, but also this broadens and helps us to try to define metrics of stress that in systems that are not directly mappable onto each other.
[48:16] Michael Johnson: Likewise. I'm such a Michael Levin fan. What you're doing is amazing. Thanks.
[48:25] Michael Levin: Thanks so much. At some point it would be cool. We're all full for this semester, but maybe in the fall if you could give a talk to our center, that would be really awesome.
[48:35] Michael Johnson: That'd be amazing.
[48:36] Michael Levin: I think people would love it. I think they need to know about this stuff. Thank you. Let's keep chatting. If you have any thoughts on general definitions of stress in diverse models, I would love to talk some more about it.
[48:54] Michael Johnson: To inject one more observation: the idea of boundary conditions in physics and biology is very important. I think you have a very fresh perspective on what defines the boundary, what defines the boundary of cooperation or morphological boundary. I guess one of the big core challenges in consciousness research is determining the boundary of a conscious system. I think your work definitely seems relevant. From my bias perspective, this branchial space view — viewing objects as shapes in branchial space, the true shape lives in branchial space — I expect that understanding your work in terms of what determines boundaries in organisms and biological systems, and then does that lead to a natural boundary condition in space? This seems very challenging; I don't know how to solve that, but maybe you do?
[50:23] Michael Levin: It sounds like we should have another conversation about the boundary, the whole boundary thing. I think it's super important.
