BI 216 Woodrow Shew and Keith Hengen: The Nature of Brain Criticality

[00:00:03] Speaker A: But the reality of real systems is that there are thousands of parameters that control its state, and once you account for all thousand of them, criticality is no longer a point, it's a large volume in that parameter space. [00:00:24] Speaker B: The realization that criticality is not just a sort of statistical fascination, but actually like an end point of evolution and development and homeostasis and truly like the kind of unifying mathematical principle that allows brains to work. [00:00:43] Speaker A: What Keith is pointing to is a potential very clear and concrete benefit of criticality that was not explicitly addressed in a lot of the earlier literature, which is learning. [00:00:57] Speaker B: There's a lot to be learned from how close you are to criticality and your ability to sort of modulate and move around that space as well. [00:01:13] Speaker C: This is Brain Inspired Powered by the Transmitter A few episodes ago on episode 212, I conversed with John Beggs about how criticality might be an important dynamical regime of brain function to optimize our cognition and behavior. Today we continue and extend that exploration with a few other folks in the criticality world. Woodrow Shue is a professor and runs the SHU Lab at the University of Arkansas, and Keith Hengen is an assistant professor and runs the Hengen Lab at Washoe Washington University in St. Louis, Missouri. Together they are Hengen and Shu 2025 Neuron. That is. They recently co authored a review paper in Neuron titled Is Criticality a Unified Set Point of Brain Function? In the review, they argue that criticality is a kind of homeostatic set point or goal of neural activity, describing multiple properties and signatures of criticality. They discuss multiple testable predictions of their thesis, and they address the historical and current controversies surrounding criticality in the brain. Surveying what Woody thinks is all of the past studies on criticality, which is over 300. In the review, they also offer an account of why many of these past studies did not find criticality, but looking through a modern lens with new analyses, they most likely would find criticality or near criticality. So we discuss some of the topics in their paper, but we also dance around their current thoughts about things like the nature and implications of being nearer and farther from critical dynamics, the relation between criticality and neural manifolds, and a lot more. And it's kind of fun because you get to experience Woody and Keith thinking in real time about these things, which I hope you appreciate. Find all of the relevant links and information in the show Notes there's quite a bit more in the full Patreon version of this episode. Consider supporting Brain Inspired if you want all the full episodes, the full Archive as well. Or just to express how you value this podcast, bringing you these kinds of thoughtful conversations. Go to patreon.com braininspired or find the link in the show notes. Here we go. Right when Keith logged in here to this, to this podcasting software, you guys just began talking about one of you, like a shared student or something who's going to be doing insane amounts of research at a very fast clip. So I didn't know that you were that intertwined. What is your relationship? Working relationship. And how far back does it go? [00:03:59] Speaker A: It's pretty intertwined. This meeting while Keith was in New Hampshire, I was meeting with three of his students. No, Yeah, I don't know. I guess Keith, I mean, what happened was a year or two ago, Keith contacted me because he was working on a grant that was related to Criticality, together with Ralph Wessel, who you may also know Paul. I don't know. He's another guy. [00:04:24] Speaker B: I don't know Ralph. [00:04:25] Speaker A: He's been in the Criticality community for some time, too. He's also at Washu, who I know him personally. So, anyway, they contacted me to see if I wanted to get into this grant. They were writing a big one, and I said yes, and also asked Keith if he wanted to help me write that review article that came out in Neuron yesterday. [00:04:47] Speaker C: Just yesterday? Yeah. Or two days ago. [00:04:50] Speaker B: Two days ago. [00:04:51] Speaker C: Two days ago. [00:04:51] Speaker A: Well, and then secondly, I had you. [00:04:54] Speaker B: Come to Washu a few years ago and give a talk. That's when we first met. But then in working on this very, very cool kind of visionary, RM1, I don't know if you've heard of this mechanism, Paul. It might be dead now in the current climate, I don't know. But it's a cool idea from the NIH perspective. It's like bring together a team of people and just aim for the moon. Like, no aim, literally no specific aims. But your overall aim should be singular. It should not be able to work. If you were to break it up into multiple parts, it couldn't work in any one lab. [00:05:27] Speaker C: So really, I want to be part of that. [00:05:30] Speaker B: It was super fun. And we just started sort of nerding out on our own rabbit holes on the side. And we started writing this paper together. And effectively, I've tried to recruit Woody to come from Arkansas to wash you. And he's happy in Fayetteville. And so the next best thing is to just pretend that we're down the hall from each other. So we basically have zoom calls every day and. Yeah, constantly. And sending People back and forth and. Okay, well, it's been a lot of fun. [00:05:58] Speaker C: This will be like a recorded daily zoom call for you guys then. [00:06:01] Speaker B: Totally. We have a whole shtick. [00:06:04] Speaker C: Okay. So I recently had John Beggs on the podcast Most well to talk all about criticality, but he, you know, he wrote the book the Critical now it's escaping me. The critical brain. [00:06:17] Speaker B: The brain and cortex point. That's right. [00:06:21] Speaker C: Quick text on the critical point. Yeah. Okay. And I've been studying criticality now, poking around in my own research. So I've been interested in this. And you guys have been part of the criticality community. I guess it's odd to think about that there does seem to be a criticality community. Am I right with that? Do you feel like you're part of a community or no? [00:06:43] Speaker A: A little bit, yeah. [00:06:45] Speaker B: I mean, no. [00:06:47] Speaker A: Oh, Keith's in it. [00:06:49] Speaker C: Oh, good, you're disagreeing already. [00:06:52] Speaker A: He's in the community. No, but there is a little bit of an old part of the community that is. That is not. I mean, Keith's new, so. Yeah. [00:07:04] Speaker C: Okay. Yeah. [00:07:06] Speaker A: There's the new guard and the old guard. [00:07:07] Speaker B: Yeah. Let me give you my vision on this, Paul. So I came to this from the perspective of homeostatic plasticity, set points and sort of robust functions. I did my postdoc with Gina Torriggiano, who's worked with Eve Marders. Is that Brandeis? That's kind of the epicenter to me of trying to understand how a neuron or a small network of neurons or a brain can, despite the inconsistency of any of its little bits and pieces, continue to do the same thing reliably over time. And the idea of criticality is a really attractive endpoint to that. And so I kind of just toyed with it and toyed with it, toyed it. And once I started my own lab, we did this project with Ralph Wessel on that. And so very much felt like I was dipping my toes into this very small and very aggressive community. But I want to make fun of them out of total affection for a moment of. It's just a bunch of physicists who love jargon, and they just like to yell at each other about that jargon and like, specific mathematical nuances that, honestly, the rest of the world does not care about. And the problem is they're sitting on this incredibly powerful pot of gold. And I think maybe we've lacked the kind of data acquisition methods or even the tools to measure criticality in the right contexts. And so I think it would be really easy in the last 20 years to sort of look at criticality and be like, that's an interesting mathematical argument and then walk away from it and no big deal. But I think that the tide is turning. And so to the extent that I'm part of the community, I'm trying to like, hitch my wagon to this, to this new direction within the field. And I think Woody is the leader of the pack on that. [00:08:57] Speaker C: Of the new direction field. [00:08:58] Speaker B: Yeah. I think of the, the realization that criticality is not just a sort of statistical fascination, but actually an end point of evolution and development and homeostasis and truly the kind of unifying mathematical principle that allows brains to work broadly writ. [00:09:20] Speaker C: Is that a fair assessment, Woody? [00:09:22] Speaker A: I don't know if I would read. [00:09:25] Speaker C: The leader of the pack, President, President Xu. [00:09:29] Speaker A: That sounds way bigger than my ego is. No, I think, you know, I want to give Duke credit. I trained in Dietmar Plentz's lab just like John did, and you know, he's also interested in the big picture. But. Yeah, so I would say that it's not quite as extreme as the picture Keith just painted, but it's not too far off. [00:10:00] Speaker B: This is coming from somebody who's, who doesn't plan to be part of that community. Like when I was trying to understand those early papers in Criticality, back when I discovered the idea in say, you know, 2013, 2014, what are the early. [00:10:14] Speaker C: Papers like the 2003, 2004, like, I. [00:10:17] Speaker B: Mean, all straight up to 2014, like, you know, the arguments about what, how do you, how do you show power law, ishness and what is a universality class? And you know, what are the different spike word sequences within these avalanches? And like, I think that there were people at the heart of that research who had a, had a vision and an idea of where it was going, but like, there's somebody who wasn't already in that family. It was not, it was not articulated in a way that I got it, like it took me 10 years to kind of realize what, what this might mean. [00:10:53] Speaker A: Yeah, and I get that especially from the papers themselves, the papers that were written early on, like you said, Paul, in the early 2000s and up to 2014 or so. I mean, a lot of the papers at that time period, you know, it faced a lot of resistance because if you try to make these kind of grand claims, it's, you know, you, it's, you can't do it without the backing it up a little bit. And over those years, you know, the last 20 years, essentially I feel like enough support has gradually built up that you can start to make those claims without sounding like a lunatic now. [00:11:26] Speaker B: But there's also the addition of the machine learning stuff too though. Woody. Right, absolutely. That it's sort of sacrosanct, it's like just carved in stone that. Yeah, you need to initiate deep models at critical phase transition or they're not going to learn. I think the connection to kind of optimal information processing for an unpredictable task, like that's what brains have to do. Right. And so I guess I don't, I don't want to like come across as accusing those early papers of sort of missing the plot, but suggest that in the, in the last 20 years, as they've developed the kind of science in cell cultures and sort of these simplified neural preparations, there's been a lot of math and CS that's built up along the side. And so we're at a place now where, where I think it's, it's much easier to clarify what this means. [00:12:12] Speaker C: Well, I mean, things take time for those bigger picture clarifications. Right. I mean, and, and like you said, with the advent of big data, big tools, better recordings, that those things can become clearer. I suppose. But so, okay, so you, Keith, you used the word aggressive earlier about the so called community. I mean, is there like, I know that there is debate because part of what your review addresses is like this historical debate on whether criticality is found or isn't found in brains, in various parts of the brains. And one of the things that you argue in the paper is that, well, you were just looking at it the wrong way and if you look at it the right way, you actually do indeed find criticality. And so there shouldn't be this debate, but is the internal debate among this, what I'm calling a community or whatever criticality, folks? Is that hot? Is it aggressive or is it friendly? How would you characterize it? Either. Either of you. [00:13:09] Speaker B: Woody, I'll let you. [00:13:10] Speaker A: Okay. [00:13:13] Speaker C: Without getting yourself into trouble, I'm sure. [00:13:16] Speaker A: No, no, I mean, I'd say it depends where you draw the boundary of the community. Right. There's been, I think John mentioned on his chat with you as well, that there are some folks who have had a fairly long run of being the critics. There's sort of a few of those groups who have consistently written papers from the critical point of view on the topic. And if you put them inside the boundaries of the community. [00:13:43] Speaker D: Then you could. [00:13:44] Speaker A: Probably say it was hot at times. But it's also cordial, like John said I mean, I've also been. I've had plenty of good conversations with those folks. Even one of the people. So John was mentioning Alenda Stacks and Jonathan Tabool, who were some of the early critics. But more recently there's been some papers by this group at Emory and Georgia Tech. So there's Ilya Nemenman and Audrey Sederberg and her student morale. But Audrey was writing this grant with Keith and I that we were just talking about. So. Yeah, yeah, we're not enemies by any means and we're very friendly. Yeah. [00:14:23] Speaker B: I think the reconciling factor on this though is when every single time we've looked at data, like other people said, Audrey's data, or our collaborator Jason up at Chicago, Jason McLean up at Chicago. He's at UChicago or he's at Chicago. Right? Yeah, yeah. People who are kind of maybe a little, a little skeptical. They're not quite drinking the Kool Aid yet. [00:14:45] Speaker C: What are they skeptical about? The grand story or I mean, the data? [00:14:48] Speaker B: So I think the strongest argument against criticality from an empirical perspective is the last 50 years of systems neuroscience. Because if you predicate your understanding of a brain on a highly repetitive task and sort of a trial structured design, objectively things don't look critical. So say you have a rat pressing a lever 25,000 times and you look at its motor cortex or wherever you want to. [00:15:16] Speaker C: Timing also is important. Right. [00:15:19] Speaker B: 100%. You are going to see something that has a very, very rigid spatio temporal structure. Both the behavior and the neurobiological dynamics have, have a lot of scale. They have the same spatial dynamics, the same temporal dynamics. And so by definition it's not scale free and so it's not critical. Right. That's very, very simple to conceive of, if that makes sense. Paul. Sorry, I know, I just got. [00:15:44] Speaker C: Yeah, no, that's okay. Because I wanted to actually get into this because I was complaining to Woody the last time I think we had a zoom that if you have to. So one of the things, the critical thing to do when measuring criticality is to make your time bins wide enough to have enough data within the time bin to be able to accurately assess criticality. But then, so, and you know, and the last, whatever 100 years of, of neuroscience is all about training animals to perform tasks that are very temporally structured. And if, you know, if we want to relate the neural activity to aspects of the task, let's say a lever press that happens on the millisecond timescale. Right. So you Just can't measure criticality there. [00:16:30] Speaker B: No, no, but I don't even want to get on to a technicality like this. Right. Let's just imagine you have an infinitely resolvable tool to tell you where you are with respect to criticality. I think that what we have found with Jason and with Audie and with these tools that Woody's, we'll get into who Sam Souter is in a few minutes. But a high school kid who's a genius, this kind of new mathematical tools that they're developing, let's just say you had infinite resolution, and you say this system is not at criticality. And I don't think it is on the 25,000 first lever press. And what we find, though, is as soon as animal walks away from that lever and it goes back into sort of an extemporaneous, unpredictable world, you're right. [00:17:14] Speaker C: Back at overtrained. [00:17:18] Speaker B: Or it's overturned. Or you could think about it this way. If you're Caitlin Clark and you go up for your 10,000th jump shot, you don't want to be integrating all the scales of information across all the regions of your brain. You don't want to be remembering the fight you had with your mom three days ago. You don't need to be reflecting on where your car is or the emotional state of your, you know, any of these things. You just need. You need to collapse onto a very, very simple manifold and repeat that thing. Right? And so I think one of the really interesting points that's coming out of the work we're doing right now, and by we, I mean this is a larger collective community, and Woody's in my labs working together, is that being near a critical bifurcation gives you a property called marginal stability. And in our little review, I don't mean that in a pejorative way, like, I just. I love our. I love this paper. Woody has this incredible analogy that just. Just resonated with me. Marginal stability is the difference between a 747. And an F16, like the 747 is immensely stable, but you cannot change. It would take a huge amount of force to change direction. And an F16, in contrast, is on the limit of uncontrollable and unfly. But by being one step back from that loss of control, you can flip that thing around on a dime. And so when Caitlin Clark presumably goes know. Gets to the line, she can shift that cognitive space and sort of heighten that space so that it is no longer scale free and shift into this simple manifold Execute the three pointer and then, you know, back into this kind of more dynamic range. And it's. It's by virtue of being at criticality that you can. Being. Sorry. It's by virtue of typically tuning the system to that set point that you have that type of flexibility in the first place. And, you know, we can scratch this out on the. On a chalkboard or the back of a napkin. We're like, oh, that sounds cool. And then we actually go look in the data, and we have Audie there and Jason there. We're like, holy shit. It worked better than we could have predicted. It's weird how well this stuff keeps lining up. Back me up on that, Woody. [00:19:17] Speaker A: This is happening a lot recently. Yeah, that's true, for sure. [00:19:20] Speaker C: But that's not good science, right? You don't want confirmations all the time. You want to falsify stuff. Right? I'm not saying you're not doing good science, by the way. That's not what I meant. [00:19:34] Speaker B: No, no, no. So I think. I think. I think. And. Sorry, I will shut up in a second. Like, I think that the math of it is that you can, you know, any. Any point has some distance from criticality, Right. [00:19:47] Speaker C: Like, any criticality itself is infinitesimally small. [00:19:50] Speaker B: Right. [00:19:52] Speaker C: It's a point that can never be achieved. [00:19:54] Speaker B: Sure. But it's just like saying every single. Well, hold on. But every single point on the globe has some distance from New York City. Right. That's just effect. And we're trying to ask, are we proximal to New York? Are we close to New York? And there are certainly cases where we find you are not even close to New York. You have moved away from there. But interestingly, it's in this case of the jump shot or the lever press, and then you'd actually expect that. So if you're going to have these repetitive dynamics that occur trial over trial over trial over trial, that shouldn't be near there. So I think that the math is. Not the math of that. The idea that it has an agenda is. I mean, there's no room for an agenda in there. Just it is what it is. And it's surprising what it keeps showing us about the kind of underlying organization of the system and that it keeps coming back to these. Sort of. Proximal to New York City, right? [00:20:50] Speaker A: Yeah. And I would add a few comments on here. Those last. The last few points you guys have been chatting about. Which one. One of them is that it's one of the misnomers or not misnomers. One of the misconceptions out there is that criticality is a point. This is. Is that it's a single, singular, infinitesimal point. So this is one of the ideas that has been on the list of criticisms of the topic for ages, is that, you know, how in the world could a sloppy system that's alive and biological be precisely at any mathematical point? And that would be a fair criticism if criticality was a point. It's not. So what I mean by that is that when people say that they've already kind of decided that they know what the important parameters are for the system that control its state, and then they'll tell you for that important parameter, sometimes called a control parameter, in physics, it only has one value where it's critical. But the reality of real systems is that there are thousands of parameters that control its state. And once you account for all thousand of them, criticality is no longer a point. It's a large volume in that parameter space of all thousand parameters. And people don't think of it that way because of the history of the physics behind this kind of topic, where there is a critical point on the phase diagram for water, but that's because you're looking at pressure versus temperature. You're not including volume and uncountable other parameters that affect water. They're just not the big important ones. So the point, the fact of the matter is that criticality is not a point. That's one of the really important things. But the other really important point I want to make that's related to this proximity to New York City stuff that Keith was just mentioning is that it's almost more important than the fact that it's not a point, actually, is that as you go slightly away from the ideal perfect critical, I won't say point critical state, as you go slightly away from New York City, all of the. [00:23:05] Speaker B: There's still traffic. [00:23:06] Speaker A: There's a lot of traffic. There you go. That doesn't sound right. I wanted to talk about the benefits. So a bunch of benefits associated with criticality, they don't disappear suddenly as you depart from criticality. As you step away further and further away from criticality, those benefits that come from the weird properties of criticality gradually dissipate. They gradually become attenuated. So the system is. It's quite a bit more robust than that old criticism would suggest that it's such a fragile state to be in is not especially the benefits of it. They're not fragile. You tune away slightly and you just have a slightly less large range of Scale invariance and things like that, you have a slightly less huge dynamic range. [00:23:52] Speaker B: But, Woody, this, this, the fact that this is even a point of concern is what I'm talking about, about the last 20 years of this kind of like involuted, like sort of naval gazy arguments. Because to anybody else in the world, it's like body temperature. Yeah. We say 98.6, but, like, we don't worry if you're at 98.7. Right. We just said. Yeah, there's. Yeah. You know, there's a point. Biology tunes to that. We're fine with it. Like, nobody, you know, me. No. Nobody's sitting here arguing that. No, you know, because. Because biology can't tune to a point that there is no body temperature regulation. Like, that's crazy. [00:24:25] Speaker A: No, that's true. That's true. [00:24:27] Speaker B: And I think that, like, intuitively, the further you are from an optimal set point, the worse things are going to get. I think a lot of people accept that. And so the questions that are more relevant to the world in terms of criticality is, does a disease state pull you further from that? Right. Does sleep deprivation pull you further from that? Does IQ inversely, is it sort of inversely correlated with proximity to criticality? Things like this, rather than pulling up phase diagrams. [00:25:03] Speaker C: Go ahead, Woody. [00:25:04] Speaker B: Sorry. [00:25:05] Speaker A: Okay, so that was just one last little point. I mean, part of the reason for that historical mess that we. That that second half or the last third of our paper talked about was there was because people didn't pose their scientific questions in this correct way that Keith is saying. They're not saying. Instead of asking, how far from New York City are you? They were asking, are you at New York City or not at New York City? [00:25:31] Speaker B: Right. [00:25:32] Speaker A: That is conceptually, you know, that's how the questions were asked initially. Just because that. I don't know, that's how people thought of it initially. But it's, It's a. It's a misguided way to go about it because the answer is always going to be no. Right. You're never going to be precisely there. [00:25:48] Speaker C: Right. The whole concept of what was critical, what was critical, what was at criticality was always fuzzy to me, like from those earlier papers. And I guess what you're saying is that that's what you're trying to address. That's the new. The new fun in criticality research is so. Okay, let me, Let me back up. My audience offered up some questions, and a kind of recurring theme in people thinking about criticality is, okay, fine, fine. If you're subcritical, the signal dampens out. Obviously, that's not good. You're super critical, Everything goes berserk and you're epileptic. That's not good. And. But then the question really is like, within the pretty good, close to New York, maybe you're in the suburbs. Zone of criticality. Right. So one of the benefits of criticality is information processing. And that's kind of a fuzzy thing, right? So is that you either have great information processing or not, but is that what you're starting to try to address is like when you are, when you deviate because of one thing, the nature of the information processing, its capacity, your deviates in a certain way. Is that what you're trying to start to get at? [00:26:58] Speaker A: Yeah, yeah, to some extent also, you know, yes. And information capacity is not a yes or no, or information processing is not a zero or one thing. You're not either able to process or not able to process information. There's a huge range from not to able, you know, and yeah, as you, as you tune closer or away, it's going to modulate that in some way. [00:27:24] Speaker B: And that's what Paul. I think the thing that felt like a light bulb for me a couple years ago, maybe as Woody and I were starting to work on this RM1 project, is the realization that if not everything, the vast majority of what a brain has to do is unpredictable. You could not have known your genetic code, could not have anticipated podcasting, but you don't have a hardwired circuit for almost anything you do other than sort of really basic cardiorespiratory function and maybe kind of bipedal motion. [00:28:02] Speaker C: Did you just say that I'm not born to do this? Come on. [00:28:06] Speaker B: Well played. So the point here, though, is that whether you're born into a cave fighting leopards in prehistoric wherever, or you're born flying jetpacks in Tokyo in the future, you effectively have the same brain and it has to optimize itself to either environment. And so you can't know which pairs of neurons, pairs obviously being a very, very small network, are going to become relevant and at what time lags, you'll need to learn information. And so if you think about like a. I don't want to say make this concrete because it's extremely abstract, but just imagine a simple grid network of nodes or neurons as you're building a model, you need to be able to somehow reliably drive covariation between any possible, any arbitrary pair of those neurons in there, because you don't know the solution, you don't know this. So the only place where you can explore the entire solution space is if you start out very close to New York City. And the point about Caitlin Clark is that once you've learned that solution, if that's all you need going forward, you don't need criticality. You just need to replay that solution over and over. It's a much simpler manifold, but given that every time you wake up and walk out your door, it's a new world, you might need to modify what you have learned. You might need to add things to it. It would make sense again from an axiomatic perspective that you'd like to return the kind of resting state of that network back to something near New York City so that you could continue to make these unpredictable connections and optimize your behavior and information processing for the future. [00:29:41] Speaker C: So this is directly. Seems directly related to explore exploit, dichotomy. [00:29:46] Speaker A: Absolutely. I. Actually, that's one of my favorite things. I would like to do an experiment on in a very direct and explicit way with a mouse that's exploring. Oh, interesting or not. You know, like. Yeah, I would. That's one of my pet hypotheses that I would like to test right now. But I don't have the experiment going right now to say, does criticality benefit foraging and exploration? I think it does. I think it will because of a bunch of reasons that I'm not going to go through, but it's a good idea. [00:30:23] Speaker B: I think I want to shout out Steve Van Hooser's data here that you analyzed. [00:30:26] Speaker A: Oh, yeah. [00:30:27] Speaker B: I feel like that's kind of such an incredible. To give you an example of one of these experiments where we take somebody else's experimental data a couple years after the fact and then run this analysis and walk away going, well, that worked better than we could have hoped. [00:30:42] Speaker C: So let me just pause there because this is. Criticality is one of those things where you have so many resources because there's so much experimental data just sitting there, you could just take it from anywhere. It's almost limitless. But anyway, that's just, you know, it's a good field to be in for that very reason. It's like such an open field. It seems like a fun spot to be in. [00:31:05] Speaker B: Totally agreed. So Steve Van Hooser is a visual cortical physiologist at Brandeis, and he was a sort of de facto mentor of mine during my postdoc. And just. I don't know if there's any trainees listening to this, but Steve is just an incredible mentor and a really, really good, humble guy. He worked with ferrets and he studies the experience dependent emergence of orientation and direction selectivity in the visual cortex. Because those aren't hardwired in the ferrets. It has to be driven by experience. And there's decades and decades of research on things like monocular deprivation and dark rearing and gabaergic treatments to kind of delay or advance the critical periods and you know, et cetera, et cetera. And so if we are right though, you would predict that before the animal is exposed to visual experience, the further it is from criticality, the less effective that experience should be in driving these plastic changes. That make sense? Sure. Okay. So Steve did these beautiful, what are the 8 or 12 hour long recordings where he puts the ferret into sort of a head mount and he's doing calcium imaging and displays repeatedly the same orientation and direction just for hours and hours and hours on end. And then he met, you know, before he does that and after he does that, he measures the selectivity of the neurons and visual cortex and you know, surprise, surprise, if you play a 45 degree angle on repeat for 12 hours into this animal's eyes, into the retina, you see a major shift towards that orientation after the thing. Right, okay. But if you regress the amount of change that you drove. So the efficacy of that plasticity against that animal's distance from New York City, before the experiment started, the R squared was like 0.99. It was bonkers. Just to restate that the further animals were from critical, the less effective this plastic paradigm was. And that was kind of our first hint about this, you know, this like our first sort of empirical, wet lab, neurobiological hint that these rules really do set the brain up for learning, modeling itself on the world. [00:33:27] Speaker C: Yeah, Well, I was going to say that's not really a learning paradigm, but it's like an entrainment paradigm or something, right? Yeah, but it is related. I mean, it directly would apply to learning. [00:33:40] Speaker B: Now, Woody and I have developed a protocol in my lab where we actually. So we think that the kind of organizing principle, the end point of sleep across phylogeny, is to restore critical dynamics. [00:33:54] Speaker C: That's a big claim. Because there's a big claim. [00:33:56] Speaker B: But I will back it up. I will die on this hill right now. I will, I will. And I'm open to counter evidence, but I have yet to find it convincingly. So anyway, rather than just sort of pushing animals away from criticality with pharmacology, or sleep deprivation or drugs. We have found a way to structure sleep and kind of drive a homeostatic super compensation. Sort of the same way, like when you lift weights a bunch, you get hypertrophy. We can enrich the robustness of the kind of near critical regime. And when we do that, these animals learn complex tasks in about 1/5 the time that a normal mouse does. [00:34:37] Speaker C: You do that while they're sleeping. [00:34:39] Speaker B: So we actually, I won't get into some of the entire details of this because I don't want to throw my postdoc James McGregor under the bus and preclude him from publishing it, but he has found a way to kind of manipulate the efficacy of sleep. So we'll just leave it at that. They're super mice. [00:34:58] Speaker A: They're super mice and they learn how to. This is not an entrainment task. This is a learning how to hunt cockroaches task. Super cool. [00:35:09] Speaker C: Oh, yeah, I think I've heard you say. I think, Keith, I've heard you say that you've gone through thousands of cockroaches or something. [00:35:14] Speaker B: Yeah. My grad student, Jacob Ami, realized that for a period of time, though, he was the world's number one commercial consumer of Turkestan Red runner cockroaches. And we use this just if there's any experimentalists listening to this. Chris Neal did really beautiful work back, and I think like 2015 or something, showing mice capturing crickets. And he demonstrates that it's a visually guided behavior. And it's very cool. It's a neat sort of visual motor spatial interceptive integration task. But the problem we were having is that crickets have a third dimension of movement. They can jump. And that makes the trial to trial variants just huge in an already complicated task. And so an ecology collaborator, this guy John Grady, said, hey, you should switch over to this fast, small cockroach 2D. And it worked. Yeah, it worked beautifully. And so to Woody's point, when you first introduced, these mice have been raised their entire lives in a laboratory environment at Charles River Jackson. Wherever they're from, they've only seen chow. And you throw a cockroach in and they're terrified. Like, they go to the other side of the cage and then you can see them kind of start to calculate. Maybe I could eat that. And they have no idea what they're doing. They kind of walk towards it. It takes them off and they get scared again. And it takes them like hours. It takes them multiple trials. Like, I don't, you know, they're fumbling it and after, you know, a couple days they, they become little, little killers. But the, we'll call them the mice that are sort of ultra critical, you know, right in downtown Manhattan. They figure it out after one or two trials. I mean it's just the very first trial, there's no difference. They're scared and then they just. I think James has done, you know, 20 or 30 animals at this point and it is a beyond robust effect. I mean it's startling. [00:37:02] Speaker A: Yeah. [00:37:04] Speaker C: So do you see. Sorry Woody, I was waiting if you wanted to jump in there. I couldn't tell if you wanted to say. [00:37:09] Speaker A: Well let me just summarize this last five minutes here, which is that what Keith is pointing to is a potential very clear and concrete benefit of criticality that was not explicitly addressed in a lot of the earlier literature, which is learning. So let me just kind of hit that with a little bit of a underline that point here, punctuate the end of that long bit there. That is criticality I think is important for learning, which is very similar to what we were saying. Learning is when your brain encounters something that it hasn't had to deal with before. You got to come up with a new solution. That kind of open ended computation or performance is where I think criticality is really most important. And that's where I think this data keeps describing is going to, going to be cool. It's going to open up a pretty new interesting result. [00:38:04] Speaker B: And in terms of just like immediate value application, James has been running this paradigm in multiple sort of mechanistically distinct animal models of degenerative diseases. And it provides an immediate cognitive benefit to them and extends the kind of effective. By that I mean sort of like no gross symptoms, it extends their sort of effective lifespan. [00:38:31] Speaker C: Do you see? So I'm about to back us way up. But before I do that, I've recently been to a few conferences in the neurotech world. Neuromodulation implantable devices to treat disorders. And people in those fields are very excited and I've seen some really cool stuff and it made me wonder are we going to be building devices as a sort of mental hygiene get you back to criticality External either like transcranial direct current stimulation kind of stuff, mild electroshock therapy to invoke critical states. Is that in the near future? [00:39:12] Speaker A: I would say it is to some extent in the near future. I think that using criticality as a clinically relevant guide is a yes in the near future. But exactly how one pushes it, pushes the brain around to get it closer or not is a little less obvious. I mean, Keith mentioned an idea that's pretty cool that has to do with messing with patterns of sleep in order to do that, which we think we can do in mice anyway. But whether or not it's going to happen by drugs or by electrical zappers or whatever, I don't know. I don't know. But my suspicion is that it will be useful. I mean, there's a lot of ways it could be super useful. Clinically. [00:39:56] Speaker C: I mean, that's probably. Go ahead. [00:39:58] Speaker B: Oh, I was just going to say. So I think that, like, before we jump to the kind of sci fi. [00:40:04] Speaker C: Yeah, I didn't mean. [00:40:05] Speaker B: I don't know. Let me just philosophically say this sort of human desire to just have a device or a pill that makes you better. Like if we just forget that for a moment, simply having an effective readout of where you are in terms of these dynamics would be immensely valuable without trying to manipulate it externally at all. Right. Like. [00:40:30] Speaker A: What you need is a quantitative and objective measure of cognition. And that's something that every biomarker for. [00:40:40] Speaker C: Cognition criticality measurements could be that biomarker. Okay, okay. All right. I'm going to bring us way back here. So I was. So scale freeness is everywhere, right? Like per box, sand piles, critical. I was. I know you're going to disagree with me on that. On the. But. But you know what I'm pointing at. I was in my basement two days ago waiting for the washing machine to finish and it has like a drain that goes into a big sink, you know, and it's on spin and I see the water kind of spitting out, sometimes a couple drops, then a little squirt. And I'm thinking if I measured that, it would be at criticality. It had a branching ratio of one, which I know is not the right way to assess criticality, but it looked very scale free. So criticality is not just in brains. But Woody, you were just talking about, you know, having a good metric for criticality essentially in brains. So I guess it's a two part question. Like if criticality is not something that is specific only to brains. Right. It's in other processes as well. Why do we care about it in brains? And then how do we define. Do we define criticality differently in a brain than we do a washing machine drip or whatever the. Do they need to be defined differently? Like how do we clearly define criticality? Sorry, that was a lot. I know. [00:42:07] Speaker A: No, no, this is my wheelhouse. How do you define criticality? So I can give you what I think of as a hard and fast definition of criticality. And this definition applies to things that are more general than the brain. So more things than the brain I don't know about. And I'm sure about your washer, maybe I'll measure it. [00:42:30] Speaker D: Yeah. [00:42:32] Speaker A: But the definition of criticality and my. From my point of view, which is almost. I mean, it's coming from physics really, but I think it's important to kind. [00:42:44] Speaker D: Of advance beyond some of the most. [00:42:46] Speaker A: The original places that criticality was studied in physics were not living systems. They were. I think John mentioned this too. They're equilibrium systems and they're not alive. And they're. [00:42:56] Speaker D: In some ways, dynamics is unimportant in. [00:42:58] Speaker A: Those systems, whereas the brain dynamics is at least. At least half the story. Space and time are both very important. But anyway, what's the definition of criticality? It has two pieces. It has two necessary and sufficient conditions. One of them is not going to surprise anyone who knows a little bit about criticality, which is that the system's dynamics have to be scale invariant if you're at criticality. So this is this peculiar property where many, many, many, many, many scales are important in the dynamics of the system. And those scales are arranged in a very peculiar way that's arranged according to a power law. So that's scale invariance. Sounds like a very technical thing, but from the point of view of brain function, it's again like your brain needs to have lots of scales of interactions both in time and space in order to do all the stuff that it does. So it should be very plausible at a very basic level like that. But anyway, that's condition A. Is that scale. [00:43:56] Speaker C: All right, can we just pause there and ask a question about. Okay, so come back. Scale invariant across a couple, what are called decades. Right. And the more decades. It scales, whatever. It just means, like across different orders of magnitude. Right. So, and I, I asked John about this also, and it's still lingering that it seems that different behaviors, different neural processing function best within different spatial temporal scale. Right. So one brain area, let's say keep it simple, like the early visual cortex. I'll bring this up later. Okay, so one brain area might be doing something that affects things on very short timescales, whereas another brain area might be doing something that affects things on longer time scales. Would you then expect that scale invariance, the set in which there's a nice power law, there's scale invariance, would that be then shifted, or do they all have to be the same does that make sense? [00:44:57] Speaker A: Yeah, it's a very good question. It's a very interesting and important question. I think that's being addressed right now in a lot of research. But I would say, yeah. Coming back to point Keith made earlier, like, if an animal is doing some very repetitive task on say a, they're repeating the task every half second, say, I bet you if you take a look at their motor cortex, you're going to see dynamics in there that's dominated by a 1 second time scale. So that is an example. That's where at least some neurons in there in the motor cortex are not going to be scale invariant in their dynamics. So that's the example Keith was talking about, this long tradition in computational systems neuroscience that will be a counter example to criticality, apparently. Whereas if you are doing a working memory task and you look in prefrontal cortex neurons where the monkey's got to like keep something in mind for the last three or four or five or six or ten seconds, now you're talking about this sort of persistent activity that has very long time scales that are important as well as short timescales that are probably important. Like the monkey had to see a brief stimulus and catch it and keep it in mind and maybe do something impressive with it. Most tasks they don't have to do too much impressive with it, but you just got to keep it in mind. But that's a very basic example where you need short timescales and long timescales. [00:46:19] Speaker D: At the same time. [00:46:20] Speaker A: That's a case where you probably are going to need criticality. But the short answer to your question is, I think absolutely there are, depending on the brain region and what it's doing, it may need to be closer to criticality where it has a wider range of timescales and spatial scales. Or it may need to be further from criticality where it has a narrower range. But that close and far that I'm talking about here is still like, you're either in the center of New York or you're in the suburbs. You're not in Milwaukee. [00:46:52] Speaker C: Okay, well, let me. [00:46:53] Speaker A: Sorry. Nothing against Milwaukee. [00:46:55] Speaker B: I like Milwaukee. [00:46:56] Speaker C: I like Milwaukee. [00:46:58] Speaker D: So I didn't ever give my second criterion. [00:46:59] Speaker C: No, no, we're still coming back to it, I promise. [00:47:01] Speaker A: Okay. [00:47:02] Speaker C: Yeah. Or you can do it now, but. Yeah, yeah, I haven't lost it. [00:47:05] Speaker D: Well, yeah, let me just throw it in there. So the second criterion is that you're at a. You're at a boundary in your phase diagram of your system. So that means if you tune the parameters that control the system. There's in your. At a boundary. That means you're at some point where the dynamics change abruptly. [00:47:22] Speaker C: Okay, so just to, just to recap scale invariance and add a boundary. [00:47:26] Speaker D: Exactly. Those are the two necessary and sufficient conditions, in my opinion, for criticality. Both of those conditions are met by most of those bifurcation type models. Part of the reason I said this is to come back to Keith's point here about temporal criticality or spatial criticality. That is, if you're talking about a one dimensional time series model or a low dimensional time series model, like most people who study bifurcations, you're talking about time domain only. If you're talking about a one dimensional hopf bifurcation or something like that, there's no population of neurons per se in that model. You might think of it as representing some population or something like that, but there's no explicit representation of multiple neurons in there. [00:48:12] Speaker A: Yet. [00:48:12] Speaker D: If you tune one of those models to be right at its tipping point, right of its boundary in a certain part of its phase space, and it has scale invariant dynamics, then that's a type of criticality. [00:48:24] Speaker B: Yeah, this is Leandro temporal. So John Beggs had a grad student, Leandro Fosque, and he's now a postdoc in my lab, working also with a dynamical systems guy, Xinon Ching. And he's put in the other paper now where he actually solves for criticality, basically using Woody's definitions in dynamical systems that don't have space. So purely, are they temporal? I mean. Yeah, these things are sort of activation. [00:48:53] Speaker C: Activation based. [00:48:54] Speaker B: Yeah, yeah. Because the sort of history of neuroscience is, I mean, the Hodgkin Huxley stuff, it starts out as a series of differential equations or a dynamical model that then predicts so much of what we've then gone on to discover experimentally. And we can go back now, or Leandra can do this, go back and show that all of those models come from the family of critical bifurcations. [00:49:21] Speaker A: Yeah. [00:49:22] Speaker D: And so this conversation that can be had about whether you're talking about temporal or spatial criticality, it's kind of getting into the weeds. So just that's kind of a disclaimer a little bit. But what I would point out is that most of the ways that people have assessed criticality and real experimental data from brains is on the temporal side. That is, you end up looking at a time series of fluctuations and looking at the nature of those fluctuations and asking if they're scale invariant. Most ways of studying criticality. Do that when you get down those. [00:50:01] Speaker C: Original or first few studies, the Beggs and Plin stuff, that was spatial, right? [00:50:06] Speaker D: It looks spatial at first glance. But the first step in avalanche analysis is to take all the spikes from a population, turn it into one population sum. But Woody, this is where count spikes in one time. [00:50:18] Speaker C: But that was across electrodes in that case, I think it is. [00:50:21] Speaker D: It is, but you turn it into a single time series. [00:50:24] Speaker B: But Woody, this is where first step. [00:50:26] Speaker D: Eliminates spatial correlations entirely from the picture. [00:50:29] Speaker B: But let's, let's get to the. What you were calling reticulo temporal in the supplemental section versus pure purely temporal. So this is like the avalanching versus D2. Because I think that's an important point. [00:50:40] Speaker D: Well, no, I mean avalanche is not assessing spatial correlations at all. Yes, it requires you. Well, it works better if you have lots of neurons, which requires a spatial. Spatially, you know, it requires a recording system that covers space. [00:51:00] Speaker C: Why would we care about space? Why, why would we even care about the spatial aspect? [00:51:05] Speaker B: Because it's like, like an ising model, right? Like you, you, you would expect that if you look at the, the correlation constant that it basically approaches what infinity in a purely critical system that you can move information across any. If you were to say, do a whole brain voltage imaging experiment, you would expect to see spatial like fractals, you'd expect to see spatial fractals. If you ignore the time series, you'd expect to see very, very, very small patterns that are self similar to medium patterns and large patterns and whole brain patterns. But that's not what we're doing. But Woody, I guess what I'm trying to get at here is even though avalanches are intrinsically temporal, there's still the trying to set you up for this man. He's not taking it the gap between avalanches thing. Right. And this is where we actually see a divergence of these two ways of thinking about criticality. [00:51:56] Speaker D: That's true. [00:51:56] Speaker C: Yeah, yeah. [00:51:58] Speaker D: The new method that's being described a couple times here, D2 is a way of assessing temporal scale invariance. That is, it's really the most direct and most. It's taking temporal scale invariance head on and you know, not kind of stepping around it, going at it from weird angles. But it's, that's the whole basis of this new analysis is that is assessing temporal scale invariance in a time series. Now avalanche analysis, like Keith was just saying, is, is a. This long tradition is that almost from the beginning of this Whole field. But like he said, by definition avalanches, they kind of chop out the quiet periods of activity and therefore, you know, you've sort of disrupted the full time series is not there. You've cut out the silent periods and you're just analyzing the periods where you have a lot of activity. And that gives a different answer slightly than what you'll get from this new method, D2 that assesses the full temporal scale invariance of the whole time series. [00:53:06] Speaker B: But what's interesting is that they usually point in the same direction when we take the same data and we sort of analyze them with D2 or with the avalanche based measure that Zhang Yuma, a grad student I worked with, she came up with. It's called DCC. It's based on avalanches. So D2 and DCC typically point at the same thing. For example, during disease, in most cases. [00:53:27] Speaker D: They'Re like in, in awake animals, they're, they're almost always identical or I mean like, not identical but like, there's no discrepancies. [00:53:37] Speaker B: But when we look at certain situations that are punctuated by regular silent periods, E2 will go away from criticality because sort of by definition it's losing some of that richness of timescale. So this is things like non REM sleep when you're doing single unit recordings. When Woody looks at non REM with D2, he says, oh, D2 bumps up. He still sees the same trend, that sleep is pulling the system back towards critical. But sleep itself looks further from criticality than wake. We don't see that with dcc because when you just look at the distribution of those bursts of activity, you still have this beautiful scaling relation. We saw recently the same thing in another collaborator, said a Gordon Smith, who's at Minnesota, also works in ferret visual cortex. These animals are born. Are you familiar with sort of ferret development, Paul? [00:54:31] Speaker C: No. [00:54:31] Speaker B: Are you a connoisseur? [00:54:32] Speaker C: Of course I am. No. I don't know. [00:54:34] Speaker B: It's amazing. They're born at the equip. When they come out, they're like a second trimester human or like a G18 rat. So they still have layer two, three neurons migrating while they're postnatal. And so, yeah, so you can record postnatally in animals and be getting retinal waves and a lot of stuff that's typically happening in utero. And you have these really, really long off periods where the system is just silent and then it's just punctuated by really beautiful spreading bursts of activity and then long silences. Again, and if you look at D2, it's not critical. Right before the eyes open, all of a sudden something changes and with this emergence of temporal scale freeness. But in those early punctuated periods. [00:55:22] Speaker A: Those. [00:55:22] Speaker B: Bursts of activity are scale free. They still show that same scaling relation. And so I'm not sure. To me, this is a really, really fascinating question of how do you make that conversion and what's the switch? We think it might have to do with inhibitory neurons and inhibitory strength. And Woody's been modeling a bunch of this. But when you get back to this question of which parts of the brain are closer to criticality or further, I suspect that. Would you call these different types of critical? Like how would you. [00:55:49] Speaker D: Yeah, I think so. I think so. That's the thing is that general definition that I offered you, Paul, over the past half hour, it's sort of an umbrella definition. So you were asking, like, how's the definition of criticality in the brain different from the one in sandpiles or your washing machine or whatever. The definition that I just gave is one of these umbrella definitions that covers quite a few things and a lot of different types of criticality. [00:56:16] Speaker C: Like what is the expand on what you mean by types of criticality? Because it's tricky. [00:56:22] Speaker D: Yeah, so I mean, so Keith just went through this one example is a pretty good example where early in development the activity of cortex looks very different than later in development for an awake animal. And it goes from this type of dynamics that looks like quiet periods punctuated by large bursts. That's what we see in these very young animals. Whereas in awake adults you see this more constant fluctuating spike rates that don't ever shut up. You know, it's just the brain never quiets down to like quiet. No, no activity. [00:56:58] Speaker C: But you, but you could say that those are two different manifestations of a single thing, AKA criticality. Or you could say that those are two different types of criticality. [00:57:08] Speaker D: Yeah, they're both from our data analysis. They both satisfy those two criteria I mentioned. But here's how they're different and why I would say they might be different types. Is that the precise kind of scale invariance that we would say each of those two states has is different? [00:57:27] Speaker C: Okay. [00:57:28] Speaker D: In one case it's this temporal scale invariance where you have all the time scales embedded together. That's the adult awake brain. In the other case, you don't have all the time scales. You have this very weird thing happening in time where it's perfectly quiet and then big burst of activity and then quiet in time. It's very disrupted and choppy and weird. But if you just cut out the active bits and throw away the silent periods and analyze those things like neuronal avalanches, then those things are power law distributed. And that is another. That is a type of scale invariance. [00:58:03] Speaker B: Woody, can I. I want to say something that I'm going to dumb this down and just tell me if I'm wrong. Okay, so, so Paul, you know, Woody was talking about like all of these thousands of parameters that can kind of influence a brain. And the definition of criticality is basically that if you were to take one of those knobs and just turn it, it would have very, very little effect. Just turn it, turn it, turn it. No effect, no effect. And then all of a sudden you make a small nudge and the system changes dramatically. Right? [00:58:25] Speaker C: That's this discontinuity, an emergent bifurcation, et cetera. [00:58:29] Speaker B: Well, it's like heating a pot of water, right? Like it's. Water is water. Yeah, sure, it's getting warmer, but it's still liquid. It still has the same specific gravity. And then all of a sudden you go from 50 degrees to 99 and then you cross this little line and all of a sudden it becomes a gas. [00:58:43] Speaker A: Right? [00:58:43] Speaker B: So there are presumably multiple boundaries that you can cross. You can turn other parameters and find a different boundary. And as long as you're at one of these boundaries and it supports some form of scale free brain activity, that is a critical point that should be capable of producing all of the different kind of solution space, pairwise connectivity of neurons and things that you need for computation, et cetera. And so is that sort of a too simplified way of saying no, I. [00:59:17] Speaker D: Agree with everything you just said, but maybe the last sentence I would modify a little bit. [00:59:22] Speaker B: Go for it. [00:59:23] Speaker D: I'm not sure that every one of those different types of criticality is, you know, good for all kinds of computation. You know, like, like that early development stuff. It, it's scale invariant and it's doing something. But is it like. It's probably not good for like processing visual input or something. You know, like these long periods of science seem like a brain that is not quite working yet. It's doing something, it's wiring up and that. [00:59:50] Speaker B: That you're completely right. [00:59:53] Speaker D: Yeah, I mean, it's very. Sorry, go ahead, Paul. [00:59:57] Speaker C: Yeah, well, no, no, I mean, this is kind of orthogonal, but this is. So I told you guys that I've been using Woody's new measurement D2 classic neural avalanche Measurements, et cetera, to look at criticality in brain activity that is recorded in my lab. And there's a certain amount of frustration at. You get the numbers and then you're left with like, all right, well, what does that, you know, like how to interpret. Like, you can come up with a kind of just so story if you want to. But then I guess what I'm wondering is, like, how far away are we from being able to definitively make claims about. As you move from the center of New York out to the suburbs, you know, like, when you're at the third stoplight, what does that mean versus when you're all the way out in the suburbs, how far away are we from that? [01:00:53] Speaker B: You just, you just knocked it out of the park. You just defined to me the transition from 15 or 20 years ago in the field to where I think it's going now. [01:01:01] Speaker C: Going, which is to sort of have. [01:01:02] Speaker B: These abstract sounding like, yeah, there's a power law. We're in New York or we're near it. [01:01:08] Speaker C: It has a slope. Look, it has a slope. What does. It's near. It's like within some range. What does that, you know, but it's on the upper end of that range. So is it more? Is it, you know, so there's all. [01:01:17] Speaker B: These three decades, five decades for power law. Like, okay, but I think that the more interesting question is to say, well, what if you start comparing conditions? So if you have motor, like a motor learning paradigm to say, do animals that learn more tend to have a lower D2 value? Right. Or they're after sort of extensive experience and learning before they sleep, does D2 drift up? Like, can we start to actually make predictions about the function of the system based on this measure? Because it has direct explanatory power in terms of information reach. Right? So in dynamical systems, what do they call it? Empowerment and reach. And those things go up at a critical bifurcation. And so I think you need to make comparisons. You need to look at, you know, different animals that have different sort of intrinsic IQ in the capacity for learning. You need to look at disease, you need to look at, well, rested versus fatigued. And all of a sudden, I think you're going to be sitting on a measure that can predict these things and actually make an accounting for them. So one of the things that leaves me wanting more in a lot of neuroscience is say, like LFP spectral power. We can say, for example, during sleep, slow wave sleep delta power goes up and it's more restorative. [01:02:37] Speaker C: Well, this is also you're pointing to another thing that has had years and years of controversy of whether. [01:02:43] Speaker B: But like you can, you can, it's very clear statistically speaking that if you, if you fatigue an animal, it's going to have larger, more, more slow wave energy. But my point is it doesn't mean anything. It could be, you could literally invert it and say ah, as animals are more fatigued, you have less slow wave power and it would be just as useful. It's it, it is a totally abstract, effectively arbitrary correlate that's reliable. It's like the sound of a heartbeat. Like I can listen to it, I know you're alive. It is an effective reporter. But if I don't understand the fundamental idea of valves and pumps and blood, it doesn't teach me anything. Whereas criticality, whether you like it or not, like it makes a very specific prediction. Like I would, I would have to close up shop on these projects if you could show me that kids with higher IQ are further from criticality than kids with low iq. Right. Like would you make a direct prediction ahead of time? That's, you know, that's my point is it's not just a, it's not just a, it's not just an abstracted correlate. [01:03:42] Speaker C: Wait, but you could do the same thing with, let's say, well, whatever gamma oscillations or something. And so how does it differ? Like you could use gamma oscillations in a predictive way like that. [01:03:52] Speaker B: Yeah, but my point is what is it about gamma that would tell? Like fundamentally, mathematically speaking. Don't fade this on prior knowledge of what people have seen in the past. But why should a system with more or less gamma be cognitively better or worse? [01:04:09] Speaker C: Because it correlates with spikes. More high gamma means high higher spiking correlation. [01:04:13] Speaker B: It doesn't have to and it doesn't in every situation. Like and right. Like I could design a different system that doesn't have that. That's just a byproduct of. It's an epiphenomena. That's it. [01:04:28] Speaker C: Okay, so you just gave away, you just showed your hand and what you think of as of oscillations. But the, but it's also a top down causal. But we can't go down this road about. [01:04:38] Speaker B: There are species that don't have theta, for example, and they still do just fine when it comes to that doesn't. [01:04:44] Speaker C: Mean that theta doesn't contribute anything in species that do have theta. Just because you can do it without theta. [01:04:50] Speaker B: Sure. But here's my point here. I don't think that a system, any system that's far from criticality can learn unpredictably complex tasks. [01:04:58] Speaker C: Okay. [01:04:59] Speaker B: Right. I think, I think it's a mathematical necessity. And so therefore you have. This is Occam's razor. You can either come up with all these situations. Well, that species has it and it matters there. But those guys didn't because they follow a different set of rules and they've evolved a different thing. And I'm saying. Or we can just go to one rule. [01:05:14] Speaker C: What about species, what about species that don't spike? C. Elegans. [01:05:20] Speaker B: You can still be critical in a non spiking network. [01:05:23] Speaker C: Plants? [01:05:25] Speaker B: No idea. I don't think so. No, actually, I actually let's go out on a limb here. In a system that can't do adaptive, you know, meaningfully adaptive learning that can't do things that aren't genetically pre programmed. I. Sorry, I'm like, I'm not prepared for this at all. But so I'm thinking on the seat of my pants. There's no evidence that there's been a selective pressure for them to adapt a flexible computational regime. Ergo, there should be no reason to expect those things to be critical. [01:05:57] Speaker D: So let me, let me, let me just. [01:05:59] Speaker B: Woody's like you are. [01:06:02] Speaker A: I'm going to push back. No. [01:06:03] Speaker C: Did you drop acid right before we hit record and is it just now kicking in? Just kidding. I'm sorry, sorry. What do you. [01:06:11] Speaker D: Here's an interesting one for you that I've been thinking about. We had. Somebody came and gave a talk in my department recently about bacteria and bacteria obviously do not have neurons and yet one of the things that bacteria do or like they're pretty simple as you know, but they do have sensory apparatus. [01:06:30] Speaker B: Yeah. [01:06:30] Speaker D: Like they have, they have certain receptors for chemical agents. [01:06:34] Speaker C: Was this E. Coli, specifically E. Coli. Okay. [01:06:37] Speaker D: Yeah. These guys that have a, have a flagella that they use to swim around. So they have a motor system, they have a sensory system and they have a weird chemical network inside their single cell body that does some computation. It takes some sensory input and turns it into motor output. Like they swim up the gradient to get food and stuff like that. Importantly, here's where I'm going with this is that in the absence of a food gradient, what these things do is they swim around according to a path that is, that has power, law, statistics. [01:07:12] Speaker C: This is their tumbling, their random tumbling thing. [01:07:15] Speaker A: Yeah, yeah. [01:07:15] Speaker D: The run and tumble thing they do is. Yeah, they, they move around according to what's called A Levy flight. It has these little path lengths that are in random directions, but the lengths of them are drawn from a power law. So in a sense their behavior is scale invariant. And the computation that their simple little body is doing is, is actually another example of criticality. I'm pretty sure I'm putting that at like 95% certainty because there's a chemical interaction network inside their body that has dynamics and those dynamics are generating these scale invariant motions. So I would put a lot of money on the gamble that says that the dynamical system that turns sensory input into motor action for E. Coli is at criticality. [01:08:07] Speaker C: But you almost can't not be at near criticality because if you're not, you're not going to have any information transformation. It's just, it's almost like defining. [01:08:18] Speaker B: No, no, no, no, no, no, no, no. [01:08:19] Speaker C: Okay, good. This is good. [01:08:20] Speaker B: So you can. So Viola Priestman is awesome and she has a really beautiful paper showing that when you take a neuromorphic chip and you're doing really simple things on it, let's say two plus two, there's no need to be near critical. You don't need multiple timescales, you don't need multiple spatial patterns for very simple computations. There is absolutely no benefit to criticality and it might be costly. There might be a reason that when Caitlin Clark takes a jump shot, you actually collapse onto a simpler manifold. It's an easier situation. Right. [01:08:50] Speaker C: Then why would E Coli be at criticality to do this? Very simple, repetitive. [01:08:56] Speaker D: Here's why I've got an answer for you. It's because when there's no sensory signal and they, that's when they, when their, their system turns to criticality and starts doing critical dynamics. That critical dynamics in their movement about. [01:09:10] Speaker C: This maximizes their range. [01:09:14] Speaker D: That maximizes their foraging. Yeah, efficacy. [01:09:17] Speaker C: Efficacy. I know but, but, but it's a very. It's. It's only one thing they have to do to do that. So I'm trying to, I'm trying to have what Keith was saying gel with what you're saying, Woody. You know, if you're kid. So the E. Coli is Caitlin Clark. That's all they do is run. And when they're, when they're not. [01:09:36] Speaker B: I don't think, I think it would be. [01:09:40] Speaker D: It would be Caitlin Kirk if that's all she ever had to do was take a jump shot. No, no, that's not a good analogy. That's a horrible. [01:09:48] Speaker A: Cut it, cut it. [01:09:50] Speaker B: But they're not like the E Coli have to take some strategy where they're sampling a large space and it's physical, right. If they just had to kind of always just move 10 microns to the left, then yes, they would not need to be. [01:10:04] Speaker D: If they always had a food rich environment, they would never need this, this behavior that they have. They even go anywhere. So they wouldn't need that. And they wouldn't have evolved it, right, if they were somehow always in food. So they evolved this critical sensory motor system so that they can find food when there's not much in their nearby environment, I guess. [01:10:28] Speaker C: So here's one of my issues that. So with the advent of dynamical systems, studying dynamical systems and doing low dimensional transitions on high dimensional data and then everything, absolutely everything is all of a sudden a manifold and manifolds are everywhere. And it gets to the point, well, like, well, it's almost kind of meaningless if it felt special at the beginning and now that everything, stupid manifolds are everywhere. But I worry that that criticality could go down that road in my head. [01:10:59] Speaker B: Okay, Paul, do you want to, do you want to like do acid for a moment as a group? Like I think, I think we can explain everything in existence by this sort of fundamental principle of what criticality is. Do you want to do that real fast? [01:11:12] Speaker C: See, you're going to. Oh wait, are you about to talk about how acid resets our critical point? [01:11:17] Speaker B: Yes. No, no, no, no. So I got, somebody was asleep at the wheel and invited me to give a TEDx talk in outside of Boston last year. And this guy, Mike Wong was talking about this idea of functional information, which he thinks is the sort of third law of thermodynamics. And the idea is that even though that systems sort of have increasing entropy constantly, they will always evolve towards more and more and more complicated sort of informational, informationally dense phenomena within them. And this got me to thinking, if you start out with nothing but simple particles in the universe, why do you end up, why do we end up with complicated systems that we have? And you could say, well, what exists, persists, and what persists, exists by definition. And so imagine an infinite amount of time and an infinite amount of iterations. Now you could say, okay, well if we just have a bunch of sand on a planet, let's start there, the planet's self organized, we just have a bunch of sand, it's flat, there's nothing happening. You're never going to get sandcastles, you're never going to build anything, right? So you add some wind to it, you have some activation energy to this thing. If there's too much wind. It'll destroy anything that starts to build. It was rip it down. But if there's too little wind, you'll just move one particle a little bit here and there. What you need is wind at some threshold where you can push sort of all of the different sizes of sand. Sometimes you move one grain, sometimes you move hundreds of grains. And if you do that infinitely eventually and those sand grains can interact with each other, right. They're a little bit sticky. You would by definition probability 1, you would generate every possible sandcastle out there of all scales. Well, hold on. If there's any selective pressure, if there's any selective pressure for one of those sandcastles over others, then that's going to fucking persist and you'll start to see more of those things. And then the same thing applies to that. So now all of a sudden, if those sandcastles can interact, you can build this out. So is it so surprising that you start to see these power law signatures throughout the universe? Not every system has them. Right. Obviously you could look at 8 billion people, you'll never find someone with 400 fingers. Right. We have a lot of definite scale in a lot of the universe, don't get me wrong. But the fact that we see people have criticized criticality by saying, oh, there's power laws everywhere. And I'm saying, yeah, no shit, we wouldn't be here if there weren't. [01:13:38] Speaker C: Well, that's what kind of what I'm saying is, how can you not be at criticality to exist? Right? [01:13:43] Speaker B: So maybe that's your point you're making. We can predict kids IQs by looking at how close they are to criticality on the day that they're born. Right? So. So it's not that. No, but it's. But I'm, I'm not. I shit you not. Like, this is. Sorry, I'm starting to swear because I'm. [01:13:58] Speaker C: It's kicking in, man. It's really. [01:13:59] Speaker B: I, I can't even. You guys just had potion. This is data from Deanna Bartch and Joan Luby at WashU. It's amazing. They scanned 250 babies at the moment of birth and then tracked them prospectively for the next six years. And I guess what I'm trying to say, and I think what Woody was getting at, is none of these systems, neurobiologically speaking, will work if they're not somewhere near New York. And so that's not the question. The question is, are there moments when you deviate from when you really. When you go out to Staten island when you go out to New Jersey. Right. And why would you need to go there? And then if you can't really get into Manhattan, does the, does the, does the, as you were saying, the number of stops that you are out on that train, does that predict something? Can that tell us when you're going to have a disease? Does that tell us how quickly you'll be able to learn something? And I think the answer to this resoundingly is yes. So to your point, yeah, if you're not even close to critical, if you're way off in Nebraska, you're dead. It's just not going to work. But there's a lot to be learned from how close you are to criticality and your ability to sort of modulate and move around that space as well. [01:15:06] Speaker C: Woody. [01:15:06] Speaker B: So I just monologed at you. [01:15:08] Speaker C: No, no, that's. I just. [01:15:09] Speaker D: I like that. I like that trajectory. That was like a. That was a big avalanche, Keith. That was like way out the outer space. And back to the very practical. One of the things that I'm really excited about right now in terms of the theory side of what my lab does, my lab actually does do experiments, but we do theory too. So one of the theory projects going on in my. Is about. I really would like to connect the dots here between what people who are taking the high dimensional brain and breaking it into all these different manifolds and what I've been thinking about for the last 15 years, criticality. And what we find. [01:15:52] Speaker A: In these, the. [01:15:54] Speaker D: Computational models that lots of people use is that one brain or one model that's high dimensional can do lots of different things simultaneously. That is, it can have a manifold that's critical. It can have another manifold that's completely desynchronized and stable, another manifold simultaneously, that is oscillating with gamma oscillations in different subspaces. Yes, yes, yes. You can break everything down. And we had a paper in 2024 about subspaces, but we didn't really. We barely scratched the surface on the topic, really, in that paper. But it's blowing my mind a little bit right now that I don't know, at least from my point of view. A large fraction of the debates in computational neuroscience center around questions of the type what is the dynamical state of the brain? And obviously Keith and I are sitting here promoting this notion that criticality is the dynamical state of the brain. And there are other folks who would promote the idea that some asynchronous regime is, is the most important state the brain is in when an animal's awake or inactive and stuff like that. And there are others oscillations, there's many. [01:17:03] Speaker A: Right. [01:17:03] Speaker D: And people argue about them forever. Are they epiphenomenal, Are they, Which one's the real one? Which one's the most important one? I think they all coexist at the same time. And at certain times one of them dominates over the other and that's why there's support for all of them in the experimentals data. [01:17:22] Speaker B: Right. [01:17:22] Speaker D: Like one experiment has a case where, where the critical subspace dominates, and that's what we see. Another case has a case where some oscillatory mode dominates. [01:17:33] Speaker C: How does one mode come to in and out of dominance? [01:17:38] Speaker D: It all has to do with how you wire up the interaction network between the neurons. And that question, though, that you just asked is a deep question. It's hard to answer that with. I mean, there's a lot of theory out there. There's this random matrix theory stuff that is all about that kind of thing really. But. [01:17:58] Speaker C: Well, I mean, going back to like some of the early recurrent neural network modeling from Monte and Cecilo, where you know, they had, they emulated a decision making task that had two different contexts. And if they put contexts as inputs to the system, then it, you know, if given one context, then the system would, would form like a low dimensional line attractor state or whatever. And given another context, it would form a different, its dynamics would form a different low dimensional state. And so these are like actually trained in as, as the inputs. So. [01:18:34] Speaker A: Yeah, I don't know. That's a different question. [01:18:35] Speaker D: I mean, that's an interesting question. But it's a different question of like, how could you control which, which mode? [01:18:41] Speaker C: Well, in that sense that they control it by. It's just the input. It's a different input. [01:18:45] Speaker A: Okay, okay. [01:18:46] Speaker D: Yeah, but that could be. A big part of it actually is the input determines which mode is dominant. But part of the reason I brought that up was just because you mentioned manifolds. But the other reason I brought it up was that this notion that Keith was going off about just a second ago, it's sort of like the inevitability of criticality in some ways. And it comes up in this context too. That is if you want to have any mode in a neural system or subspace in a neural system that has long timescales, that has persistence of any. [01:19:25] Speaker A: Kind. [01:19:27] Speaker D: That'S not just, yes, you could have it come in dictated by some input that's slow. If you have some input to the system that's driving it at a slow fluctuation, then that will trivially cause a. [01:19:39] Speaker C: Slow, let's say, intrinsic activity. [01:19:42] Speaker D: If you want an intrinsic generated mode that has long timescales, it has to be a critical mode. This is what this kind of theory will tell you. [01:19:52] Speaker A: And it's. [01:19:55] Speaker D: To me, that's profound because there's tons of. The list of things that the brain does that are intrinsically generated and have persistence is. You know, you can't. That list would be hundreds of things. It's like emotions. It's everything. It's. It's almost everything we do has persistence in long time scales in it. And I saw it. I saw a talk at Cosine last spring where they were. They were talking about. They had these really cool experiments. This was. Who is this? David Anderson's group. And they had these really cool experiments where they're trying to get to the bottom of aggression in mice. [01:20:30] Speaker C: Yeah. [01:20:30] Speaker A: And. [01:20:31] Speaker D: And they. They found a manifold, right. They found a manifold in a certain nucleus in the mouse's brain where there is persistent activity. So they have. They found this slow mode that ramps up when the mouse gets angry and slowly goes down when they cool off. And. And they have some cool methods that they. They have for identifying what is this slow manifold. [01:20:55] Speaker C: Was that in Lisbon? I saw that talk, I believe, but I'm not sure if it was that cosine or not. That's the most recent one. Yeah. Sorry. All right. [01:21:03] Speaker D: But I mean, he's. He's pinpointed a slow mode, a slow manifold. And I'm telling you, like, a brain cannot create such slow manifolds without being close to criticality. [01:21:19] Speaker B: The other point that I want to tack on, your point that you tacked on my weird monologue, is this idea that, like, we got critiqued on something we wrote, that learning only requires very fast pairing of stimuli. And like, that's just crazy. I mean, so much of everything that we do, you. You would think, like, you know, rage and aggression is a fast, hot, immediate response to a thing. But there are slow modes to. That's what he was saying. Almost every aspect of what we're doing, even with immediate sensory processing, when you close your eyes, those systems don't just. You know, your visual system doesn't just stop like you. There's persistence, there is continuity, and there is reverberation through all of this. And so I think the kind of baseline fabric to that is a system self organized around one of these. You know, what did you call not Discontinuity. What'd you call it, Woody? [01:22:24] Speaker D: The what? The boundaries? [01:22:26] Speaker B: Yeah, some boundary in parameter space. And to make it concrete, Woody, when you showed me that you could separate out these modes, the first thing to point out is that it's very easy to build, you know, to take a data set, shuffle things. You will not have a critical mode. You can't. So it's not, it's not an inevitability. Right. You can't just kind of like shake a bunch of spikes and be like, aha, I can select a subset that look critical. That's not true. Right, that's, let's, let's lead with that. Right. So it's not, it's not a foregone conclusion, but when Woody showed me the critical subspace and we looked at the avalanche power laws, they were the most beautiful linear, long stretched power laws I've ever seen in real data. So I think that this layering of modes on the backbone of this kind of reverberating critical framework is, I think it's getting somewhere. [01:23:24] Speaker D: At the risk of sounding like, Paul, you object sometimes to being like, we're always going to find what we're looking for and this is going to sound like that. But I think it's actually correct, which is that I think the brain has got a bunch of mixed up modes. So like, it's not always the case that one single mode is going to dominate such that it's clean and looks like its own thing. Most often what you're going to have is a mix of modes that are apparent. And any given neuron, actually each neuron participates in modes, multiple modes. [01:23:58] Speaker B: Maybe this is what it means to get it right though, right? So Paul, like you're saying, you know, you try to falsify yourself over and over and you could throw up your hands like, oh, we see this here all the time. But it's, you know, it's not like we just see criticality, power laws, and that's it, walk away, case closed. But you start to realize, oh, there are times when it looks muddier. Why is that? And then you start to say, oh, there's different modes and you're shifting between this and you have, you know, quick flexibility around these things in a context dependent fashion. And you know, you see drift as a function of waking time, for example. And so it's, it's the, I think that John Beggs and Dietmar and all those guys were where they were. Right. I think that they were right axiomatically and on first principles and I think they were right in terms of their data. And a lot of the controversy has arisen around figuring out the more effective high resolution ways to look at this and then the kind of the contexts. Yeah, yeah, but I think, Yeah, I don't know. I think, I think, I think that it's. I'm rambling now. I think you're right. [01:24:58] Speaker C: Is there anything that. So we're getting kind of close and I'm not going to spend all of your day here, but so it sounds very exciting. Criticality is everywhere. So wait, before I do this, Woody, based on what you're saying, so is criticality then, criticality necessary for the mixed modes? Was that the claim? [01:25:16] Speaker D: No, but it's necessary to have slow modes. Okay. The closer to criticality you get, the longer the timescales you'll get. And in order to get the scale invariance that we sometimes see in the experiments, it requires a mode to be very close to. [01:25:33] Speaker B: Okay, got it, Paul. That's what I guess I was trying to say is that like even things that people classically assume do not involve a slow mode. I don't necessarily agree with that. So even sort of fast social interactions, even learning and sort of, you know, sort of stimulus driven responses, I think have slow modes in a real brain. [01:25:52] Speaker D: Yeah, maybe, maybe I should be more specific. By slow, I mean like seconds, right? [01:25:57] Speaker C: Yeah. [01:25:58] Speaker D: Slow compared to the timescale of neural interaction. [01:26:01] Speaker C: Neural firings. Yeah, yeah. [01:26:02] Speaker D: Like slow compared to 10 milliseconds or 5 milliseconds. [01:26:07] Speaker C: What's holding you guys back these days? What do you worry about when you're approaching these problems? [01:26:12] Speaker B: The nih. [01:26:13] Speaker C: Okay, all right, That's a given. That's a given. [01:26:16] Speaker B: Yeah. [01:26:17] Speaker A: That's number one, Keith. [01:26:19] Speaker C: It's certainly not access to drugs. We can tell that. I like this theme though. That's. I don't know, it's good. [01:26:29] Speaker B: I think that the thing that I want more than anything now is to really aggressively validate D2 in human data sets. Because all of this again, as we said, started out with just kind of observational statistical descriptions. And now, you know, I, I, Woody, I'm again, say you're driving, you're driving the field towards these really interesting predictions and comparisons of conditions. But it's still all, maybe I'm shallow. I want to, I want to see this play out in people. Yeah. You know, and, and it's been hard like without D2. Like, is it Meisel or measle? Do you know how to say his last name? Christian. That's like really beautiful work with like. And what is it, Gustavo? Deco, like these long range temporal correlations. But all of these things are just ever so slightly adjacent to kind of getting at the core of it. And so to. To be able to measure proximity to criticality in a mathematically rigorous fashion on a fast timescale in human data is going to. I think that's going to move us forward like a freight train. [01:27:37] Speaker C: What, what is the. So mice are. We think of humans as smarter than mice. Their brains are different. Is it a different kind of criticality? What I'm asking about is how criticality differs between species. Because it seems like a universal single kind of thing, even though it's not a set point. But it seems like a principle that is required for brains to function the way they do. But there are differences amongst brains between species. [01:28:04] Speaker B: Simplistically speaking, I think it's the same thing. So think about building a convolutional network with a thousand neurons versus ten thousand neurons. In either case, tuning them to a transition, a phase transition or bifurcation in the small network and large network is going to maximize information processing in that network. You're just going to be able to do bigger, more complicated tests. So we did a really cool study with Audi, Cederberg, showing that in. What are they called? The networks we used to. [01:28:35] Speaker D: Reservoir computers. [01:28:36] Speaker B: Yeah, reservoir computers. That if you took a network and tuned it to criticality, but you gave it a task that was just way beyond it, it didn't matter. It didn't do a better job, it just failed. So that'd be like trying to teach a bullfrog to speak Latin. It's just not. It doesn't have the machinery. It doesn't matter if it's near criticality. And then we replicated what Viola showed that for a really, really trivial task, it didn't need to be near criticality. It could still solve it, sort of. But sweet, sweet spot. Yeah, but so when you kind of match the complexity of a task for the size of that network, you will maximize your performance at critical. And so I think that fundamental principle. People have seen this in leeches, right. [01:29:22] Speaker D: There's some evidence of scale invariant fluctuations in leech neural systems. Yeah. [01:29:26] Speaker B: Okay, so leeches all the way up through people, there's evidence of it there. But to make those fundamental predictions about cognition and disease and, you know, etc in. In people is kind of. Yeah, you know, yeah, that's where we got to do it. [01:29:42] Speaker D: I'll go ahead and just send a brief plug out there. If you ask What. What's holding my lab back the most? I need some. I need some better grad students. So if there's anybody listening. [01:29:52] Speaker C: Oh, my God. You just threw your own grad students under the bus, huh? You need some better. [01:29:57] Speaker D: I should have said more, actually. [01:29:58] Speaker C: Yeah, you said better. [01:30:01] Speaker A: I didn't mean better. [01:30:05] Speaker D: No, I need more manpower. Yeah. I feel like I'm drowning in the projects that I need to finish, and there's so many more on the docket. [01:30:13] Speaker B: That are all exciting I can vouch for. The people in Woody's lab are awesome. It's a cool culture. And those guys. You have a smart team. [01:30:20] Speaker D: Yeah, they're smart. [01:30:21] Speaker B: Okay, well, I actually have to run to meet with. [01:30:25] Speaker C: Yeah, yeah, we're going to say goodbye here. Yeah, but you can log off, too, and Woody and I can finish up as well, if that's fine. But I was just going to ask if there. There was anything that we didn't cover that you guys wanted to highlight. And we didn't really even talk about, like, all the stuff in the review, which is crazy since. But I'll talk about that in the intro and I'll point people to it. [01:30:45] Speaker B: But. [01:30:45] Speaker C: But if. Is there anything. Woody or Keith, if you have to go, go for it. [01:30:49] Speaker B: Yeah. Paul, just real quickly say, thank you so much. This was a blast, and I. Yeah, I really appreciate the chance. [01:30:54] Speaker C: Okay, thanks. Thanks, Keith. I just wanted to ask if there were things that you wanted to highlight or discuss that we haven't. [01:31:00] Speaker D: I think we hit most of the ones that I had emailed you, and. [01:31:04] Speaker C: I'm more or less. Yeah, yeah, yeah. [01:31:06] Speaker D: I mean, I would maybe mention that if you're going to talk about the review. Like, one of the things I'm proud of, I kind of. This was my sabbatical project where I. I did this. What I am pretty sure is an exhaustive search for experimental evidence. [01:31:23] Speaker C: Over 300 papers, right? [01:31:25] Speaker D: Yeah. And it's. [01:31:26] Speaker B: And I. [01:31:26] Speaker D: And I did this in like a systematic review in a way, you know, where I had a couple of search criteria and got like 3,000 papers and then excluded them one by one manually until I was down to 320 or something like that. So I'm pretty sure I have every experimental paper. Every paper that reports experimental evidence for criticality, and they're not. I couldn't, of course, talk about 320 papers. [01:31:56] Speaker C: But you graphed them. They're in that beautiful figure. [01:31:58] Speaker D: They're in the graph in. But more importantly, this is something I would like to point to. Maybe it's not appropriate for your podcast, but how about just for you, if you want to. And it's that there's a downloadable spreadsheet of all 300 papers. [01:32:12] Speaker C: Oh, that's great. Yeah. [01:32:13] Speaker B: Okay. [01:32:13] Speaker D: Which is. I really want people to make use of it because it's like, if you need to cite something, that's a great resource. Yeah, yeah, the spreadsheet has. [01:32:22] Speaker C: So you did like that huge lead legwork that someone would be frustrated attempting to do if they're searching for something related to the criticality that they're interested in. Right. That's awesome. Yeah. [01:32:32] Speaker D: And it's like, I hope that people use it and are like, oh, now I can be sure I'm not missing things. [01:32:38] Speaker C: But yeah, I'm going to directly go and download that myself, so thank you for doing that work. And I mean, the review is great also. And I'll mention again in the intro, but. Yeah, but I had read it a long time ago before it was published. All right, well, I will be in touch with you in the near future. Thanks, Woody. I appreciate the time. And this was. This was fun. We were all over the place, but it was fun. [01:33:02] Speaker D: I know that was a little chaotic. I feel like John is more in control of his. [01:33:07] Speaker C: Well, we had two people too, which was. Which was. Which was good. Okay, so this has been fun. Thanks. And have a wonderful weekend, sir. [01:33:15] Speaker D: Yep, you too. [01:33:16] Speaker B: See ya. Bye. [01:33:25] Speaker C: Brain Inspired is powered by the Transmitter, an online publication that aims to deliver useful information, insights and tools to build bridges across neuroscience and advance research. Visit thetransmitter.org to explore the latest neuroscience news and perspectives written by journalists and scientists. If you value Brain Inspired, support it through Patreon to access full length episodes, join our Discord community and even influence who I invite to the podcast. Go to BrainInspired Co to learn more. The music you hear is a little slow, jazzy blues performed by my friend Kyle Donovan. Thank you for your support. See you next time.