Transcript

Paul King: [00:00:00] Today I'm with Jaan Aru. Jaan is joining us from Estonia. Great to have you on our neuroscience, AI and consciousness conversation series.

What got my attention is that you have a paper titled Cellular Mechanisms of Conscious Processing, and you propose a neural mechanism of consciousness. And that paper was suggested to me by both Blake Richards and Dileep George. So I was interested to chat with you about it. We've been having ongoing conversations about consciousness and neuroscience here on this forum. So glad to have you here.

Jaan Aru: [00:00:29] Thank you very much for the invitations. It's great to be here an honor.

Paul King: [00:00:34] So one of the things I want to start with is just to find out a little bit about what motivated your original interest in this problem. And what problem it is that you're interested in. Is it a neuroscience? Is it consciousness? Is it something else? So how did this journey start for you?

Jaan Aru: [00:00:47] So, hi everyone. So I'm originally from Estonia. This is a tiny, tiny country. So many people in Europe even don't know about it, but for example, Skype was coded here.

But anyway, I went to study psychology in Berlin and then there in my second month or, and the first month of undergrads studies there, I was in the library and I saw a book by John Eccles. John Eccles is a Nobel prize winner for  medicine and physiology. He worked on the synapses, but he was an interesting guy because he believed that consciousness is something different from the brain. So it's, he was a strong believer and supporter of dualism. Anyway, that was the book got me, that got me interested in consciousness because I read it and I realized there that, Hey we have invented vaccines. We have sent the people, the moon. We humans, we do all kinds of amazing things, but we don't know how the brain generates or is related to consciousness.

And that was just a boom moment for me. And after that, I was just so caught up in this problem that ever since I've been kind of working on it a bit, at least.

Paul King: [00:02:16] That's great. And I think that you did, you mentioned that you spent some time with Wolf Singer and  studying gamma oscillations?

Jaan Aru: [00:02:21] Yeah, so, I mean, I started working and studying consciousness myself. So I read all the books that I had that I could find. And then I got really interested also in gamma oscillations. I was still a undergrad student. And then I was in some seminar where a PI from Wolf Singer's department was presenting and I asked some questions and then she invited me to do a PhD there. And I mean, to those who don't know, so Wolf Singer is one of the most important neuroscientists in Europe, but now he's already getting old, but I mean, back then, he was really one of these key people who invented really and showed that, gamma oscillations are important for perception and in the brain. So I was really happy to go to do my PhD in his department.

Paul King: [00:03:16] And what's your view on gamma oscillations? Do they have a role in consciousness?

Jaan Aru: [00:03:20] I was super fascinated by oscillations, but then then I mean, I just started to study them and I worked with intracranial signals, so signals that are measured from inside the human brain.

And then Wolf Singer told me, look at gamma oscillations. And I started to look at gamma oscillations, but I saw that, wait a minute, these, these look like something else. So they cover a large part of the spectrum. So not only like 40  Hertz or 60 Hertz, but they go from 40 to 200 and people in the department were saying "hey, hey, ignore that, ignore that. Yeah, these are oscillations," but I was not so convinced. So I started to talk to many people in the world. And in the end I was the one who convinced Wolf Singer also that these are not gamma oscillations that we saw that this is a different phenomenon that this is related to neural spiking.

And the fact that I, a really stupid PhD student could show something new to the whole department kind of made me think of, wait a minute. Okay. How much do we then really know about gammas oscillations? And the more I looked into it, the more I looked at them, the more problems I saw. So in the end, although my PhD started with the big hope of contributing something to understanding that synchrony is the mechanism of consciousness, it ended without it. It ended with a completely different contribution.

Paul King: [00:04:55] I see, and just to complete that topic so these higher frequency oscillations call them gamma or let's say, everything  over 40 Hertz, do you, do you sort of see a role for them in anything some people say it's solving the binding problem. It's it's part of amplifying signals between cortical areas. Other things.

Jaan Aru: [00:05:12] I mean, definitely after my, my work on that I do think that there is a lot of hype around these oscillations and this was even even more intense like 10 years ago, like maybe people are still saying, oh, all these amazing things about gamma oscillations, but then there was this new generation of researchers, including me who said, look, many of the results that have been attributed to gamma oscillations are actually reflecting some kind of a broadband signal that is indexing the spiking activity.

So it's actually the opposite of what oscillations were saying. So there might be still a role in some situations, we know that there is a role, however, this role is definitely not so broad. And, and I mean, currently I don't think that saying that gamma oscillations is kind of a viable mechanism of consciousness makes a lot of sense. There's simply not a lot of clear cut proof for that.

Paul King: [00:06:20] I see. Well, let's go to your paper Cellular Mechanisms of Consciousness. This came out last year and Trends in Cognitive Sciences or TICS as people call it. And can you just maybe give an outline of, of what your proposal there is? What is dendritic information theory or what mechanism do you think is going on and why you think it relates to consciousness?

Jaan Aru: [00:06:41] Let me backtrack a bit because what was interesting for me was also that, after my PhD, I was, I was quite frustrated with the field of consciousness research because meaning I wanted to solve the problem of consciousness, but in the end of the PhD, I just showed that the it's very hard to study it.

So we had one paper that is relatively famous for showing that you cannot study consciousness the way people did. So then from 2013 to like 2018, I had a, kind of an off relationship off phase of a relationship with consciousness where I didn't do so much. And I wanted to actually to, so I, I studied neuroscience, virtual reality, artificial intelligence, and then I wanted to do more hardcore neuroscience.

So I went to Berlin to the lab of Matthew Larkum to actually study a kind of a neural computations underlying predictive coding. I didn't want to study consciousness, but then when I was doing there, my first round talking to post-docs, then one post-doc showed me something amazing. And this post-doc was Mototaka Suzuki, and he showed me his recent data, which demonstrated that if you apply an aesthetics, so, drug that brings about anesthesia, then this has a very specific cellular effect.

So the cells we study the pyramidal neurons, which are the main class of excitatory neurons in the brain. They have kind of two compartments. One is the classic compartment around taxon that this, the basal compartment, the other of however is like  top of the cortex, the surface of the cortex is called apical compartment. Maybe will maybe clarify these terms a bit later, but the point that he was showing to me then was that anesthesia. decoupled, or basically functionally cut these two compartments of a single neuron.

Paul King: [00:09:07] And for folks listening, a compartment is like a section of the dendritic tree.

Jaan Aru: [00:09:10] So basically you could say that the neuron before it, it sends any output, then it spikes, it has two compartments. It has to integrate between these two compartments or the information we will go into that later.

But what Mototaka was showing me that is that anesthesia cut off, this integration, it basically cut off these two parts and did this was so amazing. I mean, to think about that, wow. This, there is a cellular effect that anesthesia has, and that is probably related to consciousness. So although I didn't want to work on consciousness, I had no other way.

This was the way I had to do this because this finding was simply so amazing. And Paul, if you think that I didn't explain it well, why it was amazing then ask me something and I can try to maybe rephrase something or, or explain my excitement better.

Paul King: [00:10:15] Well, and, and just about, you mentioned general anesthesia. So these are, I assume the, the GABAergic anesthesia agents, like propofol.

Jaan Aru: [00:10:23] Well, the interesting thing was that he, he did  different ones, so it  also ketamine and isoflurane and urethane. And that was exactly the point. So Paul was saying that, ketamine has a completely different effect on the, on the level of circuits. However, they're on the level of a single neuron. We saw the same effect that I stated decoupled the single neuron. And that was really like a cellular effect of anesthesia that immediately told me that look, there is a cellular mechanism of consciousness.

Paul King: [00:11:10] So, so the theory is that the what's causing these anesthesias is to eliminate consciousness, is that they decouple the two parts of the neurons and so the signals can't be combined the apical and the basal signals?

Jaan Aru: [00:11:24] If you say it like that, then it sounds like a really boring theory, right?

Paul King: [00:11:30] Well, consciousness is an unsolved problem. So anything that makes progress is big.

Jaan Aru: [00:11:34] Yeah. Well actually two things. So one thing why I was actually struggling and frustrated with consciousness over, over several years before, before working with Mototaka and Matthew Larkum was that, there is a lot, there are a lot of papers that show that, consciousness is some kind of global integration in the brain, so that it's some people even write in, in, in top review papers that, what is a consensus is it is that consciousness relies on distributed processing along the thalamocortical networks. So, which is like a brain part involving the cortex and the thalamus. Basically that it is some kind of a global phenomenon and many theories go along that way. But that was kind of a boring for me that I thought, well, it's global, but what does it really mean?

And now here, when we have this local mechanism, I mean, very local, like inside a particular cell, then we saw that, we can explain all of the results that are in the literature of this global dynamics. We can explain that by this local mechanism, because if you cut this processing within the single pyramidal neurons, then actually you cut it in the whole network because these cells, the so-called layer 5 pyramidal cells, they are the cells that send information to thalamus. They are the cells that send long range information in cortex. If you cut these cells, then you will cut all the global dynamics. So we can actually explain why people be it fMRI or MEG, et cetera had observed this global signals and why psychedelics or anesthesia or sleep had affected this global dynamics.

It's not because consciousness is something global, but it's because there is a local mechanism within each cell, each pyramidal  cell that is decoupled, for example, by anesthesia, probably in sleep. And this coupling is changed through psychedelics, for example, and this affects and and explains their global dynamics.

So that was the first point why we thought that, this is actually something interesting because we can explain it these data with a local biophysical, very specific mechanism.

Paul King: [00:14:16] But I guess what you're saying seems consistent with the idea that consciousness is a global phenomenon. If, if let's say global workspace theory or global neuronal workspace, this idea that, I think what Daniel Dennett calls fame in the brain that information is sort of widely known throughout the cortex and is shared among critical areas, for that to work the context information from one part of the brain needs to be available to other parts of the brain.

And if that context information is coming in through these apical dendrites then wouldn't this just be identifying the integration mechanism for this global phenomenon to work?

Jaan Aru: [00:14:47] Yeah, yeah, exactly. I mean, I was, I was not saying that though, these global findings about the global signatures of consciousness are crap, right? I mean, I was just saying that, there is actually a local mechanism behind it. So we indeed think that, global neuronal workspace theory would benefit from actually considering these findings and our proposal, because this could be a mechanism how you integrate information. And now talking about the information integration, we actually also think that, the integrated information theory would benefit from this because we have a very specific mechanism that can cut the information integration within the thalamocortical circuits. So our, what we propose is not really in conflict with these theories, but we hope that it kind of complements them and, and gives a very specific mechanism. So that is our point, in fact, that, t His global phenomena are interesting.  They do exist.

However, if you want to understand consciousness, then perhaps we have to acknowledge that, each neuron consists of two parts, the basal and apical compartments, and the mechanism of controlling the global dynamics comes from this local switch between these two compartments.

Paul King: [00:16:17] Sure. Well, if you, if you look at for example, the history of genetics, the idea of genes were around for, a century.

But then when DNA was found, and then finally we had in this case, a molecular mechanism for how the gene model could be happening. So maybe we have these higher high level, sometimes vague  models for how consciousness might be working neurally. And if you've identified a specific neural mechanism, it's a, it's an important piece of the puzzle.

So there are some different views on consciousness out there and, and anyone who's listening, if you want to raise your hand and we'll take some questions and kind of explore kind of neural mechanisms of consciousness generally. But so, one idea out there of course is information integration theory.

What this mechanism you're proposing would seem to be aligned, anyway, with information integration theory as proposing one mechanism by which information could be integrated. Would you agree or not?

Jaan Aru: [00:17:11] Yeah, that's all I, what I said also that in principle, I mean, we would be happy if, people from, from different theories could be saying that, wow, this is indeed exciting mechanism that we could understand as to, specific switch point for that they can explain that global dynamics that can explain how the global workspace is set up, or that can explain how the information integration happens and these kinds of controls in the brain. Right? So something like that.

However, I mean, probably it's also important to stress the other aspect of the theory, which goes a bit beyond this first, maybe more biophysical point. And the other aspect is that we, we then look at okay, we know that there is this decoupling of within the pyramidal cell, what does it mean, functionally?

And then when we start looking at it, then within the brain, you, you see this beautiful structure that the top town context information. So prior knowledge information about the previous things information coming from the model goes to the apical dendrite, goes to the apical compartment. Whereas the information from the world, from the senses from the external goes to the basal compartment. So this is also very important for us to say that, look, it's not just that you decouple two parts of the neuron. You actually decouple the internal from the external. You actually decouple the context from the sensory. You decouple the model from data. The there is this functional thing that happens that is important, and this is central for our theory, because now we say that, look, conscious experience is always affected by prior knowledge. I can bring some examples later. Many theories, I mean, the, for example, the global neural workspace theory or the integrated information theory, they do not explain it naturally.

They do not explain simply with the basic principles of why it is like that, why consciousness is affected by, by prior knowledge. The theory of Anil Seth, for example, does that, and the theories that based on predictive coding do that, however here we say that look, here's also our idea that also naturally inherently explains this why go conscious experience is always affected by prior knowledge.

This is because this apical compartment, the top compartment carries this context. It carries this prior knowledge.

Paul King: [00:20:17] Yeah. And I was going to ask about that because obviously one main idea out there in mechanisms of consciousness is the idea of the, what we're experiencing the perceptual content of consciousness are really those top-down predictions or top-down interpretation. if you will, of the outside world, that's being aligned with sensory input. So aligning that with your model, it sounds like you would say. The apical compartment, the context integrating compartment is the, is the carrying that top-down signal. And the basal compartment is carrying the bottom up signal.

And the integration at the pyramidal cell level is what's combining the top down and the bottom up together and allowing the top-down to be the context for interpreting the bottom up. Is that, does that sound right?

Jaan Aru: [00:21:00] Exactly. Exactly. And when I was kind of, away from consciousness a bit, then I did some work that reverse studying like a very simple phenomenon. So you're doing an experiment of relatively boring experiment where you have to attend to stimuli and then, sometimes attend to some stimuli, sometimes the other. And you're asked all, how well did you perceive them?

And we do this task, and then suddenly we remove the stimulus completely. There's no stimulus, but we still ask the subject, how well did you perceive the stimulus?  It turns out that subjects, they claim sometimes that they perceive the stimulus very well.

And so the point is that in these kinds of setups where people have a strong expectation about stimuli, then they can perceive stimuli, even if none is there. And other people have also shown this effect. And the problem for me was that I said, the big theories of consciousness could not really explain this phenomenon of why, this prior knowledge can create something out of nothing.

So that was one of the key issues of why. I mean, I, I was especially happy also about finding out this mechanism and, putting this also into the paper that this is important for the science of consciousness, because sometimes we mainly perceive as you say it, what the prior tells us. Not, not, what's really out there in the world.

Paul King: [00:22:33] And in some let's say, strong forms of that model. Some would argue that we only see the prior that the sensory input is being used to modulate our, our fiction of what's out there, a hallucination, if you will.

Jaan Aru: [00:22:45] yeah. And in principle, you could say that our cellular mechanism is completely compatible with that, and would be a mechanism of how this is done, actually that, the, the, the prior or the model is primary, this is the main thing. This is what carries consciousness, but then through this dendritic integration mechanism that you propose, you can integrate the external world into, into this model and into this conscious experience. Yes.

Paul King: [00:23:14] I see. Now another idea out there is that consciousness is partly facilitated by recurrent feedback loops. Douglas Hofstadter would call these strange loops. I think a Victor Lamme has talked about a top-down and bottom-up pathways, which I think would be consistent with the mechanism you're identifying. Do you have a view on the role, if any of recurrent processing in consciousness?

Jaan Aru: [00:23:36] Yeah. Yeah. I mean, this is, I mean, and then what is going to the apical compartment, right? It is in some sense, recurrent, it is in some sense, a loop, right? So, I mean, I actually wanted to name the paper at first, the loop of consciousness.

Because it is it makes a loop. And if you look into the figures of our paper, then we actually make these loops because this is, this is very important. The main difference with Doug Hofstadter is, of course, that he was, he doesn't really care about what exactly are the neural mechanisms. He pioneered the idea of that there are loops, processing loops, but he doesn't care about what is the really implementation. Whereas here we care really a lot about this implementation, right? This is the whole point of the paper that this is important, which cell types are they, where are these loops going to that they're going to thalamus for example. And which kinds of inputs go, do they be car compartment? They could go come from cortex. They could come from thalamus. They could come from amygdala, from different cortical areas from the medial temporal lobe. And this is the main difference to Victor Lamme, right? Who first of all, it was not also really didn't care or didn't know back then about the cellular mechanisms.

And secondly he was really interested or is interested in the recurrent loops between visual cortical areas or sensory cortical areas. Right. So that it's a higher sensory cortical area projecting back to a lower cortical area. And we say that it's not the only cortical areas. It's also thalamic areas, it's medial temporal lobe and all of these areas carry kind of different information, but all of them contribute to consciousness and to conscious experience in particular.

Paul King: [00:25:34] It sounds like you would be fairly agnostic about how the content of consciousness comes about. You're mostly focused on just the mechanism by which the different information streams are brought together.

Jaan Aru: [00:25:46] Well, I mean, in some sense, I would say that we care more about the content  than many other theories, because for us it's very important that this content is often dominated and always modulated by prior experience. Right. Which is a claim that other theories do not make, for example, in the global neural workspace theory, I mean, you could have a basically processing without any prior knowledge at all. Basically it doesn't say anything about the, that you have to have prior knowledge involved. So in some sense, Paul, I would say that we care more.

Paul King: [00:26:24] So, I see Anil raised his hand. And I think maybe Anil, we might go with you first, since you've thought about this problem pretty deeply, if you had a comment or question.

Anil Seth: [00:26:32] Oh, hi Paul. Hi Jaan. It's a really cool summary of your stuff here, and I really enjoyed just that, that you know, the summary.

I, I yeah, it's, it's, it's when I've read your stuff before I have indeed, I thought this is really wonderful because it is a, it's almost like an implementation level that might sit underneath a whole range of other theories. And so I guess my comment sort, the comment in the question, the comment would be, I'm thinking about it, I think, as you mentioned, like my own ideas on, on consciousness tend to be framed in terms of predictive coding. The idea that our conscious content is the brain's best guess, as, as Paul just said, the joint content of top-down predictions about the causes of current sensory input. And so there's a lot of work in predictive coding and predictive processing, trying to flesh out what the actual neural circuitry underlying that might be.

There are things like canonical circuits or predictive coding, and so on Andrea Bastos has been developing. And this might be something you've you've actually it is turning into question. This might be something you've thought about. It might be something I'm missing the paper, but it's super interesting to try and compare those sorts of circuits to the the mechanisms that you identify on your dendritic integration theory.

Because in that way, I think you could really, for instance, make some specific predictions about yeah, exactly how exactly how now prior expectations shaped conscious content. And also why, for instance, like predictive coding processing is not a theory of consciousness. That was something at least for me. I keep banging on about that point. It's a general theory of how the brain does stuff and it's useful for framing questions about consciousness. And most of the time it's used as framing questions about conscious content, what we perceive, whether we perceive it or what its relation to the sensory stimulus is. And it's actually very, not many people talk about predictive coding in terms of global changes in conscious level for anesthesia, sleep and so on.

But I think your work provides a really nice avenue to do that, right. You can say, okay, exactly what disruption of canonical predictive coding leads to global loss of consciousness. So that was the first thing. And the second thing I think I just wanted to push you a bit more because it's, as a consequence of your ideas being compatible with basically every other theory and being a bit provocative here. But if you're, if your theory is compatible with every other theory of consciousness, what, what besides giving us an additional hints about the implementation, is it really a theory of consciousness or is it just a theory of, for instance, how anesthetics might work, which is incredible, but it's not really a theory of consciousness.

Jaan Aru: [00:29:09] Thanks, Anil. Of course, thank you for being here and really listening in, I mean, I'm a great fan of your work. We'll start with the last question. So, I mean, indeed I, myself, I'm, I'm relatively humble. So when we were writing the paper, I mean, I didn't say that it's a theory. I mean, and I talked about the dendritic integration mechanism and that I thought, this is compatible with all of them. Maybe it's just, we should say that it's a mechanism, right. Then my, my supervisor, Matthew Larkum was saying, no, no, yeah, no, no, look, you have to, you have to put your head out and say that it's a, it's a theory. So he pushed, pushed us to really have this put in. And I, I'm not really confident until today in that, because I really also myself think that, do we really need that other theory of consciousness? Right. Do we have enough. Could we just say that this is our mechanism. Anyway, that's paper turned out this way. Let's see how it, how it goes. But I'm just saying that I, I myself have had the stops of whether it's really worth calling a theory of consciousness. But let's see how it goes. Right.

Paul King: [00:30:24] It does seem to be a theme in the world of consciousness research that every proposed aspect that's elucidated is presented as a master theory of consciousness.

Jaan Aru: [00:30:33] Sorry sorry for that. I mean, I feel bad.

Anil Seth: [00:30:35] Oh, but no, don't worry, don't apologize.  It goes to that mechanistic level and actually, you know, it could be, I think as you point out in some of the outstanding questions, it is, unlike some other theories out there, it is. If you take it as a theory of consciousness, it is at least testable in the sense that, you could look at some conditions like as, I think, as you mentioned, if you're dreaming, if, if, for instance that dreaming the dream state, you find this disruption of dendritic integration yet we're still having conscious experiences that would, presumably, that would refute the theory. Yeah. And it's always nice to have a theory that's refutable. So I think there are, it's good that there are these testable predictions, but I'm just thinking, like we have all these things in terms of these adversarial collaborations, like trying to find predictions that would disambiguate between, let's say global workspace theory and integrated information theory.

And it's kind of difficult for me anyway, to think of what experiment would disambiguate them. And of course I presume your response would be well, of course it is because they're compatible

Paul King: [00:31:41] And it always seemed to me that global workspace was compatible with IIT as well.

Jaan Aru: [00:31:46] Yeah. Some people have even pointed that out that, you could integrate them these two theories and have a master theory or something like that.

However, I think thanks Anil for also helping me out here because the other main point or the third main point, you tried to make is exactly that, we tried to be very subtle not to kind of make anyone angry, but the point we are making is that it's very hard to test global neural workspace theory.

I mean, of course you can do an experiment that is compatible with it, but it's very hard to refute the main claims because it's so unspecific and as a Anil has shown in, in his works, it's impossible. To refute the IIT or integrated information theory because we simply cannot even compute it for systems like the brain.

So Tononi can always say, Hey, yeah, okay. You kind of have this toy example, but you know, yeah. It is like that it is what it is and, for the brain, it, my theory still holds. So you cannot do really much about it, but our theory is very different in that sense. And that's why I think that was one of the main points of Matthew also that, now currently today, people can actually go out and test these ideas because of all the amazing work that is done in mouse neuroscience, you can actually go and do the next experiment, which is about deep sleep.

Is it so that in deep sleep, that these neurons are also decoupled as in anesthesia. That's a key prediction of the theory. If it fails, then the theory fails, but this is an experiment that can be done. It's not an easy experiment. It will take several years years to do it. But basically this is, it's a doable experiment that this can test this.

And he also pointed out, that you can, we can test the effect of dreaming. Well, that's harder in mice, but they could at least look what happens in REM sleep. That's an experiment that's even harder to do, but we propose that basically during REM sleep, these neurons are not decoupled. Right.

And I mean, I won't go into many details, but the point is that people in neuroscience labs that have mouse mice in the lab, they can go and check different steps. They can, for example, check what are the neurons that control this coupling? What are the specific molecular mechanisms? What are the specific circuit mechanisms that control this coupling between these circuits?

So for me, this theory stands out from this others because it can be tested right here right now with tools we have available today.

Paul King: [00:34:36] Go ahead, Anil.

Anil Seth: [00:34:37] Okay, so to add one more to that first question, I was just wondering if, if it would be a fair characterization to say that, the processes of dendritic integration that you talk about are necessary for conscious states, but not sufficient for conscious states.

Jaan Aru: [00:34:55] Yeah. I mean, this is a question we kind of deliberately avoid in the, in the paper. And I think we had a whole discussion on Twitter with Joscha Bach also said like, look, you're, I mean, you could have also consciousness, let's say in insects or in some other organisms that do not have cortex or layer 5 pyramidal cells at all. So what do you say?

And my claim, my point was that look for consciousness science, it's really important to figure out one mechanism, one mechanism that is necessary. And if you have figured that out, that will be a great leap forward. So I'm not claiming, and they're not claiming that this is a sufficient mechanism. They're claiming that this is one necessary mechanism in our brains in mammalian brains.

Paul King: [00:35:51] One thing that is interesting is of course when Francis Crick and Christoph Koch started the, the great thaw of the neuroscience study of consciousness, they propose a search for the NCCs, the neural correlates of consciousness, which suggested that, there might be special neurons or maybe neurons that are active in a special way at a special time that would be associated with consciousness.

But with this model, you're basically saying it's it's not actually particular neuron behavior, but rather how those neurons are combining information. So, in a way, this is a proposed mechanism that doesn't actually fit the NCC model.

Jaan Aru: [00:36:25] Well, I do think it, it is, it is a bit a tiny, more interesting because Francis Crick with his great vision, he actually together with, Christof Koch they have some papers and some chapters where they write stuff like most likely this will be the deep layer, 5 pyramidal neurons projecting outside of the cortex.

And they have a sentence. Like it would be marvelous if that were to, and now we are in a position to say that this hypothesis that they wrote in 1992, and then a chapter in 1994, this has merit. It actually could be like that. And this is crazy. Of course they didn't speculate about apical dendrites. But they do speculate about this particular cell types that are central for our theory, that deep layer 5 pyramid cells projecting to  thalamus and other subcortical structures.

Paul King: [00:37:23] So perhaps you're fulfilling their prophecy.

Jaan Aru: [00:37:26] In some sense, in some sense, oh, let me also add here. The funny thing, I, I ha we haven't had the chance to discuss it with Christof Koch, but in the 2016 Nature Reviews, Neuroscience review paper with Tononi, the write that most likely the cell, the cells that are important for consciousness are not the layer 5 cells because the Tononi together now, they don't like this idea of broadcasting outside the cortex so much. So Koch has reversed his position. He's now saying that layer 2/3 cells, so the superficial layer cells are important for consciousness, but ironically enough, we do think that we kind of see strong evidence for their original idea.

Paul King: [00:38:16] We're fortunate to be joined by Kevin O'Regan, who was our guest last week and who has for 20 years been putting forth a sensory motor contingency theory of consciousness really focused on how the perceptual content of consciousness comes about and Anil, as it happens, had built on that theory to propose a way that neural predictive coding could possibly be a computational mechanism that might conform to this sensory motor contingency model. So I wanted to invite Kevin to make a comment or ask a question. Kevin, your, your thoughts.

Kevin O'Regan: [00:38:51] Yeah. Thank you very much for offering me the possibility to ask two questions. So my first question is why do you think it's specifically consciousness rather than just overall cognitive processing that's affected by this decoupling? If you knock out cognitive processing, it seems inevitable that you should also knock out consciousness. And as Anil pointed out, the interaction of the top down with bottom up information is a general aspect of cognitive processing. So that's my first question. Why do you think it's specifically consciousness rather than just overall cognitive processing?

 And then a more specific question to test your theory, I was wondering what are the brain mechanisms that do not require coupling between the two compartments of the pyramidal cells? Are these precisely those mechanisms that remain functional during anesthesia?

Jaan Aru: [00:39:46] Yeah. Thank you very much. Of course, Kevin, and thank you for the nice clubhouse also last week, I mean, I enjoyed also your clubhouse and all the papers and the books you have written. So thank you. It's it's an honor to have you here and ask these questions.

So indeed I do think that in principle, in principle, we could have also tried to make the claim that look, this is a very general mechanism, underlying cognition. However now the interesting part is that it's directly related to your second question also because we do think that processing that happens within the cortex that stays within the cortex.

For example, that mainly requires the processing of the superficial layers, layer 2/3. This can happen. This can happen even when the layer 5 pyramidal cells are decoupled. And this is kind of, as you see, I mean, you could say that this is why does it have to be so specific the hypothesis. It's stupid. Why, why do we make it so specific? But exactly because we can, we can make a neuroscientifically very precise hypothesis which can be wrong. But we did do that, that, layer 2/3 processing can continue. So in fact, a lot of cognition that relies on, on feed-forward processing can happen because you do not need necessarily to project out of the cortex. You do not need the project to the subcortical structures when you're doing some computation within the cortex.

And I think a primary example could be something like object recognition. So I would not be surprised if we like with some agents, we drugs, we decoupled the layer, pyramidal neurons, for example, let's say now in the monkey and we now show images of a picture, like it's done today with, with, this massive datasets that we would still, I see in a higher order visual cortex selective responses to different classes.

So when that was now back back to the first question, also, the reason why we think that this is not simple recognition that the is consciousness, and of course, again, I acknowledge that maybe it's stupid. Maybe we made they both, this is that these too precise, we could have made a more diffused. I both, as you said, that this is relevant for cognition, et cetera, but here we want that the really contrast with, for example, the global neural workspace theory make a hypothesis that is specific enough that it kind of tickles the neuroscientists who can actually go ahead and test it and show that Jaan Aru and Matthew Larkum, they were wrong.

Their hypothesis is not true because actually layer 2/3 cells are relevant for consciousness, et cetera. I mean, I would be super happy. So I think our hypothesis is very specific. Maybe should not have been so specific, but we wanted to contrast ourselves and show that, you know, today, you can make this very specific claims about specific neurons doing specific computations and other labs can go ahead and just test it.

Paul King: [00:43:36] Regarding, regarding this question of what types of processing would be possible when this integration is not happening. I'm imagining that Simon Thorpe would probably say that basic object recognition could happen via a feed-forward pathway only. So perhaps that might be an example. But maybe that would be object recognition to drive action without any awareness of the object. If you wanted to make the argument that that awareness required that top-down pathway.

Jaan Aru: [00:44:01] No, no. I mean, I think that most people agree that, you can have a, if, if, unless you show some really new objects, you can have a unconscious responses as to make the unconscious responses to various object classes. So that feed forward processing as we see  also in deep neural networks can happen unconsciously and mainly relies on this processing in superficial cortical layers.

And according to our hypothesis, is unconscious cognition.

Paul King: [00:44:35] And I guess Kevin raises an interesting point about the relationship between consciousness and cognition. One view, I guess, could be that cognition is something that happens underneath consciousness that just the ability to sort of plan action and execute behavior is a form of cognition and consciousness of some sort of extra sauce layered on top that, can happen or not happen.

Another view, I guess, might reverse the order and say that consciousness is the base foundation for awareness and for, integrating information in this sort of global workspace theory sense. And that cognition is a special type of activity on top of that for doing procedural systematic action and rational thought.

And Kevin, I don't know how you think of the relationship between cognition and consciousness or, Anil. I'd be curious on your views and also Jaan.

Jaan Aru: [00:45:20] Yeah, I mean, if I go first and I mean, this is also related to the question about the compatibility of, of consciousness, right? And there have been now a really kind of important arguments from Tononi from Christof  and also from Anil saying that, consciousness,  is different from cognition. It is different from intelligence.

And I tend to agree with these views or that, we can see them as independent and moreover, we can already propose some independent neural mechanisms. The two interact. And that's fine. That's the good thing that they do interact as do layer 2/3 neurons and layer 5 neurons, but fundamentally that they are different. And when you study the one, you are not necessarily studying the other.

Kevin O'Regan: [00:46:18] The trouble with that view, though, is that if, if you want to claim that consciousness is something over and above just cognition then you have to really define what aspect this is. I mean, how would you test it? It seems very hard to test it unless you can somehow say, well, what exactly are you looking for? Because most measures of consciousness proxies for most measures of consciousness measured using cognition, reports, reports. So how can you ever prove that consciousness is something over and above just the simple reports or ability to report.

Jaan Aru: [00:47:00] Yeah, this is one of the fundamental problems of consciousness research and, this TICS paper, really new paper, and let's see how that goes and how people feel about it. But the paper people know me for in consciousness research, is a paper, saying that, if you cannot understand the neural basis of consciousness by doing these kinds of cognitive tasks, because you will always always confound the prerequisites and the consequences of consciousness, which is basically the same thing that, when you have reports or, or you could try it with eye movements, you could try different things,then you will have also processes that kind of confound the search for the neural correlates of consciousness. So it's a fundamental problem.

I think I hope that there are many people who are actually looking at data and try to find out new, clever ways for dissociating these. It Is not an easy task, but I mean, we will do it eventually, but for now, one view, is that what, for example, Tononi and Koch have taken  is that they basically, they define consciousness as integrated information. And they show that, look, you can have this measure independent of cognition. You can have the system that is good in cognition in some cognitive tasks, but has low phi, et cetera.

So that's one way, I mean, I'm not necessarily agreeing with them. I'm just saying that, this is one way of doing that, that you basically do it kind of, in some information theory of math terms, right.

And the other way, how, how how we do it, that we try to propose that, there are specific neural mechanisms and we try to dissect it in the brain with on the level of, of, of, of neurons circuits, and molecular mechanisms, but we will need it and we will need to rely on new tasks that kind of better differentiate with the, between these two.

And this is a hard task, but it will also be a very fun task because it's a real challenge that will propel the field forward a lot.

Anil Seth: [00:49:16] I think, yeah, you're exactly right to point out. I think Jaan, your earlier papers on, on prerequisites and consequences, very influential. And but if there's been, as I think there's actually a lot of quite innovative work, trying to get around that as far as we can, it's not easy, but they're always now no report paradigms that try to distinguish between the correlates that underlie reportability as compared to the true as you would call it the true NCC.

I mean, these are not guaranteed to work, but I think that there are useful approaches there that go some way to solving this problem. But yeah, I think this distinction between consciousness cognition is, is so, is so difficult and it's kind of awkward. I, I think you summarized, I do think there are different things that, that defined differently. Cognition is, is all about function, mapping of input to output, computation of some sort or other. Consciousness is fundamentally about phenomenology. And so, in a sense, it's really a difference of emphasis. When you ask questions and, and do experiments, and this is where, for instance, I've been very influenced over the years, both by Kevin's work to back from 20 years ago, his famous paper with Alva Noë, the sensory motor contingency theory, because it really gets at phenomenology.

And you're trying to explain why particular experiences have the phenomenal character that they have. And if you try and understand the mechanisms that, that account for that phenomenal character, then you're asking questions that are about consciousness, primarily. About you know what, what can people do with certain input information, and so on. Can they do arithmetic unconsciously or not then that's, that's, that's a question about cognition rather than phenomenology, I think in practice, there's a lot of it's bound to be a lot of overlapping mechanisms.

So I'm not entirely comfortable with this idea of putting a particular order. But I do think it's important to recognize that they're picking out fundamentally different properties of assistance. So we could, as, as, as evidence and artificial systems, we could have systems and computers that we could all probably come to concept some consensus that they're implementing some form of cognition, but without any phenomenology.

One, I think that's one part of the public discourse on this, where it gets a little bit, I have a few issues with, which is this assumption that as artificial intelligence or artificial cognition, if you are getting increasingly sophisticated, then as things come on and there's artificial consciousness as well. There's an actual conflation of phenomenology and intelligence.

Paul King: [00:51:51] I wanted to return just for a moment to the topic of the sensory motor view on consciousness and its relationship to the work that you're doing on context integration,  Jaan, just because we have the opportunity to have both Kevin and Anil here, who've thought about the sensory motor version and Anil who's also tried to connect that to some of the neuroscience. So I'm guessing that Kevin would take the view that consciousness, or at least a conscious perception, isn't so much something that's happening inside the brain, but rather a circuit with the environment. That by engaging in trying to seek out answers and the environment, providing it through sensory input, that that circuit is creates the experience of conscious perception.

Whereas in your dendritic integration view, I think you would say that it's just really the combination of the feed forward pathway with the feedback pathway that applying the context that sensory input is what is producing conscious experience. I, I don't know if he would say that, but it's sort of implied by your paper.

Any thoughts on that or Anil, as someone who probably understands both these views. I'm curious how you combine them.

Jaan Aru: [00:52:54] Thanks, Paul. I mean, that's a really interesting question and obviously we didn't really think about that, but I mean, thanks to this clubhouse now I'm seeing that, we could also be saying that look, when a theory is mainly about the cortex and how things are computed, within the cortex, like for example, integrated information theory of consciousness or global neural workspace theory, then it's very kind of opposite to what Kevin is proposing in the sensory motor theory.

However, in our theory, the important thing about layer 5 pyramidal cells is that they are the cells that project out of the cortex and that project to the motor system, also.  They project to cerebellar nuclei. In many cases they can control movement. So in that sense, I feel that there is actually a strong connection between our theory and what Kevin and Anil have been proposing, because exactly our theory, is inherently tied also to these sensory motor loops because these cell layer 5 neurons are the ones that control action.

They are the ones that control eye movement. They are the ones that control our, our different movements. They are their neurons that couple us to the environment. So there is this coupling between the different compartments of the, within these neurons. And the, Kevin has and Anil have been stressing, the coupling to the, of the, between the environment and the brain.

But what we are, what I'm saying now is that look because, these neurons are involved directly involved in the motor aspects. Then there is this coupling between this specific biophysical mechanism and the action. And therefore there is this coupling between our view and then the what Kevin and Anil are proposing.

Of course, as you can understand, I made this up. Right now right here. So it might be that there are some significant holes in this idea, but at least my first impression was that it's not in opposition, but rather it stands hand in hand.

Anil Seth: [00:55:18] Yeah, I think that's sounds more or less, right to me.

It's still, I think it's, it's an idea that would need to be developed actually gets back to, I think one of the comments, stroke questions I had at the top, which was how to think about the circuitry that Jaan is talking about in terms of canonical circuitry for predictive processing or active inference.

Can we try and map those to sort of, two architectures or circuit level descriptions? Together. I mean, from, from my point of view, I just to clarify, I think that's at any given time what's going on in your brain is sufficient for conscious experience. So, I think for instance, when you dream, you're not intrinsic, but you know, interacting with the environment in that in that moment, but you might well have needed it, a history of interaction over evolution and development, but that the conscious experience is sufficiently produced by activity in your, in your brain. And so what I've been trying to do is inspired by, by Kevin is, is, think it's basically connect this idea, these ideas of predictive coding and predictive perception, which at least in the beginning were largely passive and largely about how the brain can interpret ambiguous sensory data through minimization of prediction, error to something that's now like Karl Friston would call it active inference and so on, where we think about predictions about the sensory consequences of actions, conditional or counterfactual predictions, which I think have an interesting relationship to what Kevin and Alva were calling sensory motor contingencies, the rules by which motor actions and sensory signals behave.

And I think, and I hope I'm not mis-characterizing you, Kevin, I think this is super important and shaping phenomenal experience. And then for instance, explaining the difference between different kinds of conscious experience, why vision is the way it is, and is different from emotion or from olfaction.

And so in a sense you might think of a three level description here that you've got a sort of philosophical framework for considering these differences in phenomenology. And you've got some almost algorithmic level about conditional predictions as an extension of predictive coding. And then you've got circuit level description of canonical circuits for predictive coding and of this dendritic integration theory and figuring out how all these things relate together. I think it's quite interesting, but it just gets complicated. Right. But it gets back to this idea that Jaan's theory is at the implementation level. And if you really unpack the consequences of that, that's maybe where you might find how it makes distinct predictions about other different theories that are phrased in the first instance at a slightly higher level.

Wisam Reid: [00:57:59] I really like that, that you're pointing that out. Jaan, correct me if I'm I wrong on this, but what, what your, the, the, this paper that we've been talking about, it's super in line with this dendritic predictive of coding theories. In my view, it was Walter Sen and Mathew Larkum on that were the first to really robustly, formalize this dendritic predict coding, predictive coding framework.

And I think you're right to point out that this is like where the rubber meets the road on this idea between the level and the content of consciousness. This the way the, the surrounding local circuitry is interacting with dendrites with these, these really compartmentalize dendrites and that relationship between behavior, the content of the environment, the brain states and the compartmentalization and the, and the neurons.

I think that all of these things are very tied together and fit in with what you've been talking about Jaan.

Paul King: [00:58:53] I want to make sure we get Kevin's  framing.

Kevin O'Regan: [00:58:55] Thank you. I was just going to add, I have nothing against everything that's been said and I support what Anil said and, and so on, but there's just one thing which the sensory motor approach adds to all this. And it's something which I would say is kind of metaphysical because most people are working on consciousness, they believe that the searching for a thing that they call consciousness and there they'll always say, well, what in the brain could be generating this consciousness? What is it? What mechanisms in the brain might it be the generate consciousness? But what the sensory motor theory suggests is that we shouldn't look for anything in the brain that generates consciousness because consciousness is not the kind of thing that can be generated.

So it's like life, you wouldn't say metabolism generates life. You wouldn't say DNA, replication generates life. These metabolism and DNA replication. They contribute to what we mean by life. And at the beginning of the 20th century, the vitalists were looking for a thing called life and they postulated a vital essence.

But since then, modern biology has changed to, to take a kind of constituent to, to a constitutive view of what life is instead of looking for what causes life. They look more for what constitutes life. And this is what gave rise to the revolution in modern biology. And I suggest, and that's what sensory motor theory suggests that we should look for a similar revolution as concerns, the notion of consciousness.

We should not consider consciousness to be a thing that's generated by the brain. We should do what I would call is divide and conquer. We should divide up consciousness into a whole bunch of capacities that we call consciousness. And we look then for the mechanisms in the brain that enable those capacities.

So for example, being able to report on, on a stimulus, that's been flashed up, the ability to change your, your decisions or planning, concern concerning some incoming sensory event. These are all mechanisms, which you can easily explain by brain processes that don't require you to imagine that these brain processes might be generating some magical, extra thing that you call consciousness.

So, so, I would, I think that the idea that the, this communication between the compartments of the  pyramidal cells could be very important, obviously for, for, for, for example enabling motor reactions, enabling all sorts of computations that the brain must do that constitute what we call consciousness.

It's a mistake to think that they somehow generate consciousness. They're not generating any special essence that constitutes consciousness. They, they enable what we call consciousness. So this is a metaphysical difference. It requires a different way of thinking about what consciousness is, and it suggests a different way of going about neurophysiological research.

Paul King: [01:02:09] Ramzy, I think you were wanting to introduce potentially a new topic. Ramzy, go ahead.

Ramzy Abueita: [01:02:15] Yeah. And I'm not necessarily trying to change the topic, but rather try and process the topics that we've been talking about through a unique lens. And, in this conversation I've been picking up on different levels of focus in regards to how we interpret consciousness. And I think that there are certain levels of focus that have been brought up such as this interface between these two compartments the apical and the basal?

And and this whole, like layer 5 pyramidal neurons, and, these are more of like a local level of focus. And then, recently in the conversation I heard brought up the topic of these corticothalamic loops. And this is a subject that I find very recurring whenever, I read about consciousness and to discuss it, it's this notion of, what is the involvement of these feedback loops between cortical and thalamic regions and and all these different nested layers of feedback loops.

And I kind of wanted to tie this to how a lot of times in neuroscience, some of the greatest discoveries that elucidate, some of the toughest topics come from people who are very atypical case studies where a person's physiological disposition, for example, is quite different than the norm. And this allows us to kind of like, examine how their idiosyncrasies come into play in regards to informing, which structures and physiological properties are responsible for which phenomenon.

And I wanted to tie this to this phenomenon where there is in Canada there are twins who are conjoined by the head and their brains are actually conjoined as well. And these twins their names are Krista and Tatiana Hogan. So they're conjoined by the head and their brains are also conjoined and they claim to be able to see through each other's eyes. They're able to feel each other's thoughts and they're able to actually control the body movements of the other. And through PET scans, they found that these conjoined twins in their brains, there is a thalamic bridge that is linking together the the thalamus of each of these girls.

So this leads me to the question, so, based on their subjective reports they seem to have consciousness of the contents of the other's mind. And this makes me, beg the question of, so is this, process of experiencing the phenomenology of the the mind of the other through this thalamic bridge, is this process entailed, through some sort of feedback mechanism or information  transmission mechanism that is contingent upon activity within like this thalamic pathway that links the their two brains is this, experience of consciousness, done at the more local dendritic apical, like layer 5 pyramidal neuron level in the individual brains while the thalamus, this thalamic bridge, merely serves as an information relay that kind of like enables information to be transmitted into the individual compartments in their own brains where it is then perhaps I'll use the word processed into consciousness, or is this thalamic loop something that is necessary for this, or, or, are we in a situation where, you know, this, thalamic loop is a necessary, but not sufficient.

And in these more local level activities, are, are more necessary for the consciousness? I just wanted to toss this on a table and see how we parse it. You know what can we learn about consciousness from this very atypical example of these twins who have a conjoined brain with a thalamic bridge that allows them to see through each other's eyes and, and hear each other's thoughts.

  Paul King: [01:06:14] So it sounds like each has a separately operating language areas if they can share the report separately. And I'm reminded also the split brain patients of Gazzaniga, but I'm curious, Jaan, your perspective and also Kevin, how you might explain the phenomenology of the situation.

Jaan Aru: [01:06:29] Yeah. Yeah. I mean, it would be really funny if I could give out kind of a straight answer for like the the at, at the answer A's the right one or, you offer three different perspectives. Right? Of course. I don't know. Of course disease are a very cool phenomenon that needs to be processed

But at the least this case shows that all the theories that only rely on cortical processing have have a problem, right? Because here, as you say, the thalamus acts as a bridge. It could be consciousness, it could be information, but regardless, these. It's a very central bridge if you can process the contents of the core consciousness of the other.

So at the least it shows that thalamus has to be a central layer for any theory of consciousness. And that these are strong points we also make. We do not make it so much in the Trends in Cognitive Sciences paper, but we have two kind of a companion papers, one in Frontiers in 2019. And the other is a smaller paper in philosophy of mind sciences that you will find from my Google scholar profile.

And if you're interested in this thalamic aspect, and definitely I suggest to check it out because we do think thalamus was, it was really important in the middle of the 20th century and towards the end of the 20th century, but then cortex and cortical chauvinism took over in consciousness studies and in mainly related to cognition, et cetera.

But the cases like this show that, we have to take thalamus into the picture. So thanks very much.

Paul King: [01:08:22] Well, certainly in the case of these twins if each brain is essentially projecting context information to the other brain, then each brain has some information for both brains to work with. It'd be interesting to know whether the perceptual experience is similar or different for the twin looking out through their own eyes presumably that's a much more vivid experience than maybe they're vaguely aware of what the other person is seeing. Not necessarily seeing what they're seeing, but it'd be interesting to see how that's reported by them. I don't know the information on them so well.

Ramzy Abueita: [01:08:53] Yeah, and it really is an interesting case because one wonders, how much can we kind of like. Use this opportunity to study this really bizarre phenomenon without kind of like overzealously like making these two girls scientific spectacles.

And so there is that aspect of, of the ethical dilemma, but I will just add that, there's an interesting thing where one of the girls can see the contents of both of the eyes of the other girl, whereas the other girl can only see the contents of one eye of the other. So there's some really interesting, asymmetrical dynamics at hand and, yeah, Paul, you kind of summarized it quite well. Is, is this, process, is this mechanism where the, the consciousness is kind of like shared by both is this, done by mere virtue of the thalamus serving as an information relay or is this, that is thalamus sort of like a crucial hub where integration happens where their, their conscious experiences are bound together.

And would this similar situation where conjoined twins at, at the brain have access to each other's phenomenology, would this still hold if a different region let's just say, like, they just have like layers of the cortex, kind of overlapping would this overlap of the layers of the cortex, entail the same phenomenon or is this explicitly contingent upon the thalamus being this relay?

And I think these are really fascinating question.

Paul King: [01:10:23] I'd be skeptical that the thalamus is doing any integrating. I think that the thalamus is mostly regarded as a relay center, but it's certainly controlling the flow of information. And it could be directing the flow of information to more than one brain. But Jaan, your perspective?

Jaan Aru: [01:10:36] I was just saying that most likely or most likely, but you could say that or predict that if the cortex would be conjoined, then there would be only one consciousness. Right. I think that's what IIT would predict and probably we also, right. But it's an interesting, interesting thing to figure out.

Paul King: [01:11:00] But what about when the corpus callosum is severed? Do you still just have one consciousness or two or is it one and a half consciousnesses?

Jaan Aru: [01:11:08] Yeah, probably one and a half, because that's the coolest version. I mean, I really don't know.

Kevin O'Regan: [01:11:14] What about dissociative identity disorder, where you have just one cortex, one brain and several people in there.

Paul King: [01:11:22] Taking turns, I guess.

Jaan Aru: [01:11:24] Well, yeah, I mean, as is there are a lot of I mean, it's a blunt statement, but there are a lot of neurons then in principle, you could have ah, let's call them different configurations, right. That, different set of layer 5 pyramidal neurons in, in prefrontal parietal cortex is active and these generating this integrated loop system. Right. And it is like one person. And then, some contents can be shared, but you know, some essential parts of the brain have completely non-overlapping neurons. So I do think that there's something, this is a, this, these dissociative disorders are very interesting case, but they can be explained by, by, by our theory, or also by Tononi's theory, for example.

Kevin O'Regan: [01:12:12] You could make a prediction that could, if you took somebody under dissociative identity disorder, that when they were, depending on the different person that was currently present, different brain areas would be active. That would be very interesting to test. I personally doubt that very much, I think.

Jaan Aru: [01:12:30] Yeah. Yeah, no, but I mean, I'm not saying different areas, but like different neuro, neural coalitions within that area. So perhaps perhaps with some very specific measure, maybe with the seven tesla MRI and specific analyses, you could show it perhaps. I mean, I would not be surprised, but it's not that different areas, but it's more specific. So it's on the level of neural populations, I guess. I mean, let's see.

Paul King: [01:12:58] I would say something more like that, that each identity in dissociative identity disorder is activating a different dynamic circuit that's using the same brain areas, but configuring them in different ways and accessing different episodic memories, if I was to speculate. So in that sense, dissociative disorder could be sequential consciousness. There's sort of one consciousness present at a time, but it switches between person A and person B.

And maybe split brain patients are parallel consciousness. You've got two consciousness that are mostly aligned, but they can drift apart a little bit if they happen to get different sensory input.

Divya, I know you've been waiting. Go ahead.

Divya Chander: [01:13:31] Yeah. I have so many things to say. The conversation has gone in many directions. I happen to be one of those neuroscientists who's an anesthesiologist who studies consciousness. We recently published a paper in which we shared that changes in levels of consciousness are actually reflective in complexity measures in the brain.

There are things if you could imagine basically looking at the brain at any given time point and then using a sort of a time delay mechanism to sort of trace out its pathway using for physicists in the room who might understand this time delayed embeddings, for those who are not just imagine that we were actually tracing out chaotic attractor shapes for different states of the brain.

And what we found is that brains that are qualified as more conscious, so awake and aware, basically our patients were interacting with us. They had these attractors that traced out this relatively  spherical shape which, it sort of indicates that there is multi potentiality in that particular brain, as a complex system.

And as our subjects began to lose consciousness, get drowsy as we pushed drugs and then finally lost consciousness altogether. That level of complexity began to slowly shrink and change its attractors shape. There's by the way, there's hysteresis in this process that we find with so many different metrics, which means basically going down is not the same as coming back.

But one of the things that sort of struck me about this entire conversation is that we we're sort of conflating two things where we're considering consciousness to be like a single, a single access system. And I think it has at least two dimensions. Stephen Laureys who's a a coma neurologist over in Belgium tried to capture some of this by describing a two dimensional graph.

On one axis you have a level of consciousness and the other is content. And the nice thing about trying to break apart, these two things is number one, a lot of what everybody's talking about, these sort of surrogate, neural correlates of consciousness and indirect measures of cognition, those would fall along the content access.

Whereas the kinds of things that those of us who measure things like anesthesia, sleep and coma do. is we actually look at what happens when you toggle through the level axis. And I think this is an incredibly important distinction to make because that level access is actually conserved phylogenetically through most of evolution.

As a good example, fruit flies can be anesthetized. They do not have layer 5 pyramidal neurons, but we seem to be able to remove something to do with their level of consciousness. And what's even more interesting is that fruit flies actually have a, a knockout gene, which is analogous to that in humans, essentially.

If you knock out a gene that produces something called orexin or hypocretin, which is sort of this positive, neuromodulator it kind of, it's not as strong and excitatory neurotransmitter in the brain as something like acetylcholine that kind of helps to maintain the system at a slightly higher, more revved up excitatory level.

If you knock that out in humans, you get narcolepsy and in humans who have narcolepsy. If I try to anesthetize them, their dynamics of going down of losing consciousness is the same as anyone else. But when I wash anesthetics out of their brain, they don't come back and it takes forever to sort of reassess the brain to sort of reassert itself and reestablish all these connections within these neural networks.

Fruit flies do the same thing. And if you almost look at the effect of anesthetics through not just the animal kingdom, but we're now even finding in the plant kingdom, those that actually have a motor aspects to them, we can actually knock out very similar systems.

So what I think is kind of important to understand then  is that if you just look at consciousness as this uni dimensional thing we're gonna have a lot of problems and conflicts in sort of defining what that shape looks like, but even within a single human brain we can toggle that content access.

I think Jaan, you mentioned the difference between the connections between the apical and the basal compartment during deep sleep versus REM or at least a prediction that you had. That's a really great example of where that same one, same brain may actually it's content may be fluctuating between high levels of content to complexity and low levels.

Whereas the level of consciousnesses actually kind of the same. It's quite diminished. And so I wanted to just at least bring up this idea that we have at least a two axis system that we are dealing with. Another interesting thing is that during anesthesia feed forward connections are largely intact.

And sensory processing itself is largely intact, whether it's the visual, the auditory and somatic sensory system. What seems to have been disrupted is the lateral connections and the feedback connections there, connections between association cortices rather than primary visual and primary auditory cortex.

I find that interesting because actually Jaan your theory about disruption in layer 5, pyramidal neurons actually supports that, right? Because these are the neurons that are making these long range projections, these different lateral cortical spaces. And I, I tend to agree with someone else who said that, I think that this is more of a, sort of a mechanistic description of what happens as consciousness is sort of slowly removed and, and what needs to happen for it to, to come back. And it's almost like these more global theories and your more cellular biophysical mechanisms. These are all part and parcel of almost like the fractal nature of the question that we're asking.

 Have you been yet. Have you ever thought about Stu Hameroff and Roger Penrose have that interesting sort of quantum microtubule theory of consciousness, which is actually quite difficult to test . But your proposal in terms of disruption of these two compartments.

This is actually one of the first things that I have seen that may actually connect up there sort of quantum level theory, cause these microtubules are in the dendrites, right? This sort of cellular biophysical mechanism. And I'm wondering if you've given any thought to that.

Jaan Aru: [01:19:55] Thanks for the contribution. I mean, a lot of, a lot to unpack here. So let me start from the easy from the, the one, which easy indeed, that as we commented, the feed forward processing is intact feed forward processing happening mainly layers 2/3. So you can actually compute the image content, you can classify what is on the image.

For example, however, indeed feedback coming from higher, higher areas to apical compartment of the deep pyramidal neurons is the one where you get the problem. You as, and as, as we show basically might be activating those apical compartments. The information does not get through. It does not get to the base or compartment because it's cut in the middle.

Now, why is it cut? And this here, you very nicely point out the Hameroff and Penrose, which means that you really, you really understand how this thing works in the brain in the sense that it has not escaped us that indeed microtubules might be the key mechanism that control coupling in the between these two compartments that we find super ironic.

I mean, crazy that people dismiss these theories of Penrose and Hameroff. Although they might actually have something to do something, some truth in it now, to be clear, we are not, of course supporting the quantum aspect of the theory. We're not saying that Hameroff and Penrose were completely right, but we are saying that strangely enough, ironically enough, these damn microtubules might be a way how you could actually control the flow of information between these two compartments.

The microtubules if they contract, basically they can block the communication between the apical and the basal compartment. So that's a really fun thing. And we are looking forward to basically looking into that or testing that. And it will be really, really ironic if that turns out to be the case.

And finally, the first main question about the dimensions of consciousness. So level versus content, this is very important for us and I, we didn't have the chance to really specify here enough, but we do think, and if you have time, please look into our also into our 2019 Frontiers paper, we do think that it's interesting and important that in the mammalian brain, both the level system, so the system that controls the level of consciousness, and the content system, the system that is responsible for computing, the contents, are intertwined. They both kind of intersect at the level of layer 5 from pyramidaln neuros because the level of consciousness has been related to thaamus and to subcortical structures.

But now we show in our papers that look, the subcortical structures related to a level of consciousness. They project, for example, do the apical compartment of these pyramidal neurons, right? And so the pyramidal neurons, so, they project to thalamus, so that these loops contents of consciousness, they are different how the levels of, and the contents of consciousness are kind of, generated in the brain.

But they crucially interact on the level of layer 5 pyramidal neurons, and we find it also quite interesting. Yeah. But thanks for these questions and for these contributions.

Divya Chander: [01:23:58] And Jaan, just really quickly, have you looked at the relationship also between other biophysical phenomenon that occurred during anesthesia and sleep like hyperpolarization and activation of certain types of currents, like, potassium currents, et cetera, that we also see at a global level give rise to phenomenon such as bursting. And since you can entrain the fellows in the reticular nucleus of the thalamus to the cortex to start bursting in lock steps, simply by changing their level of polarization.

Jaan Aru: [01:24:27] Yeah. Yeah. I mean, we are aware of that because we, we worked on the thalamus. However, we did not manage to do those experiments. We started the set, but we didn't actually actually do that. But we are aware that there are this very interesting mechanisms that tie these loops and and that there are many out there really beautiful biophysical mechanisms besides the one that we are proposing in this one, small TICS paper

Wisam Reid: [01:24:59] About the microtubules.  I think it's important to note that the big difference between say an axon and a dendrite, is that you're getting bi-directional microtubules. In other words, these microtubules facilitate, molecular pathways the movement of proteins and such up and down the dendrites where as an axon, they go in one direction.

But Jaan, just for some clarity here, wouldn't you agree that ion channels inhibition dendritic morphology, these kinds of things are much more important for compartmentalization at timescales that are relevant to behavior, then microtubules.

Jaan Aru: [01:25:40] Yeah. Yeah. Obviously, I mean, obviously we, we, that's why we focus on dendrites. That's why we've focused on the  apical dendrites, right. That's why we in the paper, we do not mention microtubules this idea simply that was suggested by the previous contributor occurred to us and we thought of it. We really, we didn't start the project because indeed in, in the lab of Matthew Larkum do think that the process is happening in the dendrites and the processes that are for example, controlled by the metabotropic receptors are more fundamental.

So that's why we simply think that, it's possible that somebody will show that you know, this decoupling mechanism is at least partly controlled by microtubules, because we are not really following up that hypothesis.

Paul King: [01:26:30] Go ahead. Wisam,

Wisam Reid: [01:26:31] First of all, thank you, Paul, for doing the room and Jaan for, for being here.  This is super in line with all the stuff, maybe some of you have heard me ranting about in rooms. But I'll give you a little bit of background and just so you know where I'm coming from. I study dendrites for working on my thesis now. I've done some work. I've been working with Mark Harnett and Anne Tekesian at well Mark's at MIT and, and Anne's at Harvard medical school.

So I just have just kind of a general question about consciousness and I'm just kind of curious what your working definition of consciousness is and more directly, I'm curious what your thoughts are about the computability of consciousness itself. I also wanted to, just as a comment, when I just said, I couldn't agree with you more about in particular, the high gamma frequency oscillation stuff, but in general, I think the people who got their causal arrows mixed up when it comes to oscillations in general.

The other question I had, and Paul, this is admittedly more about learning and intelligence, which is what I'm more focused on at the time. So, if this is off topic, that's fine, but I had a, I heard you were talking earlier about that. And, and just as a little bit of background, I'm, I'm very familiar with Larkum lab work.

I haven't had a chance to discuss work from the lab with a member of the lab before. So this is cool for me. So, but I am very familiar with the history of the, of the lab's work. So I just, as a point of clarification you, weren't talking about the pyramidal neurons as kind of like, I think we would at least agree to a first order approximation to two compartments and obviously relevant to the paper at hand two compartments, but where do you stand on the level of compartmentalization in neurons in vivo in the, in the living brain in general.

Jaan Aru: [01:28:30] Yeah, thanks.  I, I'm very a big fan of the work done in Harnett lab. So that's, that's, that's really great. And I'm just starting to answer the questions. I hope. My memory serves me well that I can, I do remember most of them, but so yeah.

The definition of consciousness is something that, there's always the funniest story that, Francis Crick answered questions of philosophers like, when  philosophers ask, what is your definition of consciousness? Then Francis Crick said that, come on, guys, you have worked for two millennia on this problem without any success, let us now work on it. And we don't want to kind of define it prematurely because if you don't know really what the, what the phenomenon needs, then why should we define it at first? And they always brought the example of achieving that, it's very hard to define it, then you didn't start with defining it anyway.

So we proceed by discussing it and, and the kind of having gotten an understanding of the phenomena we are interested, then we could kind of explain that consciousness is something gear you gain when you wake up from a dreamless sleep or when you wake up from anesthesia, right. Consciousness is the feeling of being you. Consciousness is are the different contents of, of, of, of, of consciousness that you have. And when necessary then of course, we go through the pain and try to define it better.

But for, for, for the present purposes, these kinds of general definitions seem to work now. I already see that I, I forgot some of the questions. I do remember the lot question about the compartmentalization  and this is the fun thing. If you have also of course, read many papers from a lack of bath, and you see that in our papers and in the papers of others also the number of compartments is different, right?

So sometimes we go very detailed. Sometimes we, we go only these two. And of course, as most of the people, most of the folks in artificial intelligence, for example, they only have one compartment. They don't care about any other compartments, only one. Right. But for us, it seems that, you have to probably have, I think about three compartments.

So two are the ones we mentioned, apical being the one where top-down information is sent, basal being the one where the external data information is sent. And we think that probably we have to also think about the middle compartment, which we call the coupling compartment that, because we have to have a specific mechanisms and we kind of know some of these mechanisms already that control this coupling between the other two compartments.

There are specific neurons targeting, this coupling compartment. There are specific receptors there on this coupling compartment. And we definitely need that. But then as we could go deeper or we could say that, we have the compartments that compute the NMDA spikes, et cetera, et cetera.

But for these present purposes, we do think that two, or even by the three compartments or the ones that are necessary and not really sorry, can you again, remind me, what was the question in the middle? There was something interesting. I remember.

Paul King: [01:31:53] Wisam, was there something else that you had there, or did that address your question?

Wisam Reid: [01:31:58] I was interested about I totally get where you're coming from with the definition of consciousness, but I am curious just as if you have a general feeling about whether this is this phenomenon is computable or not. The last question I'll ask question. I wanted to throw in there, it was a while ago, since I looked at the paper and I just got, I would have been more prepared for further for your discussion today, but I I'm moving in such can you remind me what was the metric that you used to to measure the amount of compartmentalization in these neurons? Can you remind me?

Jaan Aru: [01:32:35] Yeah, yeah, yeah. Okay. So, I mean, for this paper and for the paper of of Suzuki and Larkum, I mean, they did not use a particular measure right. In the sense we, we say that, look, if we measure the activity from near the cell bodies in layer 5, then it will be the base or compartment. If we stimulate the neurons in the, in layer one, it will be, the apical compartment.

And in the middle, around the oblique dendrites when we do the manipulations of the  metabotropic receptors, then this will be the coupling compartment. So we did not go any deeper than that in these studies. But of course, as when the other papers of the Larkum lab, people do look at more specific dendritic compartments, for example, et cetera, but this was the level. 

Paul King: [01:33:31] It's the third compartment in the middle that integrates the other two compartments.

Jaan Aru: [01:33:34] Well, yeah, I think, I think, yeah, if you, if you kind of think about these two compartments and you say, as we do that, the, that coupling or integration of these compartments is important, then you also have to explain what is the mechanism controlling it actually, there we say that, you, then you, you focus on this middle thing, what is in the middle? And for convenience, we say that there's this coupling compartment that we do know that it's, it's a specific thing because it's controlled by this metabotropic glutamate and metabotropic acetylcholine receptors. A And metabotropic glutamate receptors probably come from the higher thalamus axons . So it's very specific neural mechanism, a molecular mechanisms that we are kind of testing in the further studies. Yeah.

Wisam Reid: [01:34:26] In my mind, I think of it, I know there, it's very cell type specific region specific, but I, I think about a lot of the stuff that's been done in the in the hippocampus between Markum and, and McGee lab and this kind of stuff, you know this cholinergic, muscarinic acetylcholine receptor potentiating, calcium channels, yada yada   The readout was burst probability?

Jaan Aru: [01:34:52] Yeah. I mean, yeah, I mean, in, in this paper, so if you, anesthetize the mouse and you activate the apical compartment, then basically you do not see spikes, or maybe you might see a few in the layer 5 in the output. So you could do a burst probability, you could simply do spike count. The effect was very robust. So you could do different things there to actually quantify this effect, right? 

Paul King: [01:35:28] Okay, Amruth, go ahead.

Amruth BR: [01:35:30] Hi Jaan, fascinating paper. I wanted to quickly get your thoughts on a very interesting model that seems to me is a natural outcome of combining your work with the experimental work done by Sawtell and others in the mormyrid electric fish.

Just for the benefit of those who may not be familiar with it. They essentially looked at how the fish cancels out electrical inputs that is coming from their own EODs when perceiving signals from their electro receptors.  And they found that the medium ganglion cells, which receive inputs both from the corollary discharge, as well as the sensory inputs from electro receptors, they essentially create inverted signal of the generalized corollary discharge, which gets passed on to the next layer in perception.

And they subtract it out from the overall signal in order to generate a noise free version of the clean signal. If in your model in your experimental work, we find that the signal from top-down layers are actually getting blocked under anesthesia, then in their paper, what they are in the model that they argue, then the output of the medium ganglion cells would be the negative of the signals coming from the sensory inputs.

And therefore, if that gets canceled out from the total same sensory input, then the resultant is close to zero, which might explain the lack of perception, which might give the cellular mechanism for the lack of perception under anesthesia. So I just wanted to get your input on this.

Jaan Aru: [01:36:58] Yeah, very nice. Thank you. Thank you for this contribution. And of course, I mean, this is something we didn't touch upon very much today, but you could see this mechanism as amplifying some inputs from the environment, but given that these apical dendrites are on the heavy, under heavy control by inhibition, you could also see them as canceling out or, or subtracting something that comes from the environment.

This is related also to the question that Anil asked, but we didn't have the chance to answer about the relationship to predictive coding and at its core our mechanism is an amplificatory one, right? If you have a match between the top down and the bottom up, then the neurons will fire strongly.

And this is something that runs counter to predictive coding. And this is a real biophysical problem to my mind for the predictive coding theory. However, of course there is this inhibition. So you can actually, by adding an inhibitory neuron, you get could, get something like the predictive coding theory required.

If anyone is interesting, interested, then actually in our paper in section computational theories of a top-down bottom-up integration, we do address this and we are a bit more specific about that question, but in general, we, the theory we propose is consistent with these kinds of experiments that you discussed and is more broadly, also consistent with predictive coding but provide some very clear neurobiological implementation of this phenomenon. 

Paul King: [01:38:56] Well, I think we should probably let Jaan go, 'cause I think it's what time is it where you are?

Jaan Aru: [01:39:01] So it's, it's 11 and I really, I mean, I have to end the consciousness of my kids for today. Yeah. So, not by doing a molecular change, not by decoupling their neurons, but by reading a bedtime story. And if I don't do that, if I don't go now, then my wife will end my consciousness.

But I am very grateful to Paul for putting this together. I mean, it was such a blast to have Kevin and Anil here. I mean, I, I'm a big fan of their work. Thank you to everyone who asked questions. I mean, such amazing questions. It was really a pleasure to be here and thanks to everyone who listened and are here now. So thank you very much.

Paul King: [01:39:48] Yes. And thanks to everyone for contributing, thanks to Jaan for staying up late for us, dialing in from Estonia.  We'll be keeping this weekly series going with new topics. And there's also a survey that you can take as part of that where you can volunteer topics that you would interested in, or guest speakers you'd like to see.  So, wonderful Sunday with everyone. Until next time.