BI 210 Dean Buonomano: Consciousness, Time, and Organotypic Dynamics

[00:00:03] Speaker A: And there's this ongoing tension that exists between physics and neuroscience where we sort of like to blame each other. So I think neuroscientists like to sort of say, listen, we live in a presentist universe. Time is flowing, time is changing, and if that's not what physics is telling you, figure out where the physics went wrong. IIT is not a neuroscience theory. IIT is a theory of fundamental physics that proposes a new ontology to the very structure of the universe, to the very properties of the universe. IIT proposes that certain configurations of matter are conscious. Timing should not be seen as a specialized function in the brain because timing is so important to everything we do that it is sort of a universal property of neural circuits. [00:01:15] Speaker B: This is Brain Inspired, powered by the transmitter. I just hit record there and then totally forgot what I was supposed to be doing. Hello everyone. Now I remember. It's Paul. Welcome to Brain Inspired. Dean Buonamano runs the Buonomono lab at University of California, Los Angeles, ucla. Dean was a guest on Brain Inspired way back on episode 18 where we talked about his book your Brain is a time machine, the neuroscience and physics of time. In that book he details much of his thought and research, and others thoughts and research as well, about how centrally important time is for virtually everything that we do. He also describes different conceptions of time and philosophy and how brains might tell time. That was almost seven years ago. And his work on time and dynamics in computational neuroscience has continued to this day. One thing that we discussed today, which actually comes later in the episode, is his recent work using what are called organotypic brain slices to test the idea that cortical circuits implement timing as a computational primitive. It's something they do by their very nature. Organotypic brain slices are somewhere in between what I think of as traditional brain slices and full on organoids. Traditional brain slices are extracted from an organism and maintained in some brain like solution while you perform experiments on them. Organoids on the other end start with a small amount of cells that you then culture outside of the organism, let them divide and grow and specialize until you have a mass of cells that have grown into organ of some sort to then perform experiments on. Organotypic brain slices are extracted from an organism like the traditional brain slices, but then they are also cultured for some time to let them settle back into some sort of homeostasis set point to get them as close as you can to what they're like in the intact brain and then you perform experiments on them. So Dean and His colleagues used optogenetics to train their brain slices to predict the timing of the light pulse stimuli that they're using. And they find that populations of neurons do indeed learn to predict the timing of the stimuli. And they also found that these populations of neurons exhibit replaying of those sequences similar to replay seen in brain areas like the hippocampus. That's, you know, rough and tumble. Summary and we go into a lot more detail in the episode, but we begin our conversation today talking about Dean's recent piece in the Transmitter that I will point to in the show Notes called the Brain holds no exclusive rights on how to create intelligence. So there he argues that modern AI is likely to continue its recent successes despite the ongoing divergence between AI and neuroscience. So this is in contrast to what folks in neuroai believe or what many have been arguing for. Okay, we then talk about his recent chapter with physicist Carlo Rovelli titled Bridging the Neuroscience and Physics of Time. In that chapter, Dean and Carlo examine where neuroscience and physics disagree and where they agree about the nature of time. Okay. And finally we discussed Dean's thoughts on the integrated Information theory of Consciousness, or iit. IIT has seen a little controversy lately. It's fairly well known, at least amongst my community, that over a hundred scientists and a large part of that group calling themselves IIT Concerned. IIT Concerned. They have expressed concern that IIT is actually unscientific and this has caused backlash and anti backlash and all sorts of fun expressions from many interested people. Dean explains his own views about why he thinks IIT is not in the purview of science, namely that it doesn't play well with the existing ontology of what physics says science is all about. It doesn't fit within all the pieces. It can't be examined in the same way that we examine all other science. So what I just said, of course doesn't do justice to his arguments. And again we hash it out in the episode. He articulates it much better in a few moments. So lots of topics today and I hope that you enjoyed the conversation as much as I did. I want to say thank you to my Patreon supporters. If you support Brain Inspired on Patreon, you have access to all full length episodes, the full archive of all past and present Brain Inspired episodes and I have started posting the complexity discussion group meeting recordings there as well. We just, I've mentioned this a few times on the on the podcast. We just started a discussion group journal club to cover the foundations of complexity papers that we Discussed in my recent episode with David Krakauer. It's a big group of us, over 350 now as of today. Anyway, we just had another one today and I am so happy that I put that together. And there are so many people as interested as I am to learn about many of the foundational papers in complexity science. Okay, look for all the links and information in the show notes. Here's Dean, can you. Speaking of time, you probably hear time jokes all the time. Right, right. Anyway, speaking of time, do you know how long it's been since you were on Brain Inspired? [00:07:25] Speaker A: I'm afraid I do know. I think I actually looked that up. Was it. [00:07:29] Speaker B: I just looked it up. [00:07:30] Speaker A: Was it 2018? [00:07:32] Speaker B: Yeah. Six and a half. Almost seven years ago. Yeah. [00:07:35] Speaker A: Good job. Way to go. [00:07:36] Speaker B: I think you're in the same office though, I think, aren't you? [00:07:40] Speaker A: I can't remember actually. It's quite likely, but you had. [00:07:43] Speaker B: You had time sensitive lighting where if there wasn't motion, the lights would go off. [00:07:48] Speaker A: So I'm good memory that I don't remember. But yeah, you were on the forefront of the podcast and you hung in there. So congratulations. [00:07:56] Speaker B: I'm still hanging in there. I fondly remember reading your brain as a time machine. This was back when my family, we all quit our jobs and moved into an rv. And that's when I started the podcast. [00:08:08] Speaker A: I remember that because I think you were actually in an RV when we first interviewed. Is that possible? [00:08:13] Speaker B: Probably so. Yeah. I remember sitting on the couch and reading. Reading your book. Actually listening to your other book. [00:08:20] Speaker A: So you were also prescient with the RV because you were obviously preparing for Covid. [00:08:25] Speaker B: Yeah, but we coveted happened after the RV broke down and we moved into a house. And so we were. Our time. Our timing has been off. [00:08:33] Speaker A: Timing was bad. [00:08:35] Speaker B: Yeah. Okay. So we have a lot to discuss here. First of all, though, welcome back. Almost seven years. That's. That's. It seems so long. But at that is so long, it does not seem so long. [00:08:46] Speaker A: Yeah. Thank you. It's a pleasure to be back. [00:08:50] Speaker B: I thought so. We have a host of topics to discuss today and I thought maybe we could actually start with the AI side with the piece that you wrote in the Transmitter recently. Somewhat. I couldn't tell if the tone was pessimistic or if you. Well, maybe you can tell me. Right. Because the point in the transmitter piece is that AI has disregarded brains and more recently has begun advancing without any regard to dynamics and what are found in our brains. And then you suggest. I Think it's in the title, actually that that trend will continue and there's no reason to expect that it wouldn't continue. But one of the points that you make in your talks recently is that timing has never, you know, really been a factor in AI and especially recently, dynamics. Even though you can, you can study dynamics and recurrent neural networks, which is a lot of your history. Real time is not important in AI and continues to not be important. Do I have that correct? [00:10:05] Speaker A: Yeah, I would. I think you got the spirit correct. I would certainly inject a few comments there. So, first, the AI and neuroscience have been sister fields from the dawn of AI. So I never implied in any way that AI has disregarded neuroscience. Indeed, AI is based on fundamental tenets of neuroscience. If you look at one of the most fundamental tenets in neuroscience, it's that information memories are stored in the strength of synaptic connections, in the strength of connection weights, and that learning relies on changes in the weights. That tenet is in many ways anchored in all of AI based on artificial neural networks. And then many, many other aspects, from convolutional neural networks to things like regularization, which in my mind have aspects of homeostatic plasticity, things like dropout, which have aspects of synaptic failures, things like convolutional neural networks obviously are based on architecture of V1. So AI and neuroscience have had a very synergistic and intimate relationship. But I make the analogy that much like if you go back and I give this example of von Neumann when he was writing down the first code of the first architectures for what we now call the digital computers and von Neumann computers, that he was also inspired by the brain. But obviously computer science and neuroscience diverged very, very rapidly. My point is, is that we are already reaching the point where neuroscience and AI will continue to diverge. I think they're already diverging and they will continue to diverge. And as an indication of that, I gave the example of time. So up into the 2010s, 2011, 2012, when computers, when AI started dealing with time varying problems for real, in earnest, in terms of speech recognition, motion recognition, interaction with external world, the view was that RNNs would be the way for the future. That's right, because RNNs are how the brain seems to tell time and it seems to be based on the internal dynamics. But then something dramatic happened, which was the 2017 transformer paper, attention is all you need. And that switched very rapidly from RNN approaches to transformer approaches and to Me, what's absolutely amazing about Transformers is how good they are without any ability to tell time whatsoever. [00:12:50] Speaker B: But they can tell sequence. [00:12:51] Speaker A: They are timeless, but they can tell order. They can tell sequences, but in a way that's not really time dependent. Right, so this is a good point. And so maybe we should clarify. So everybody knows transforms can absolutely tell sequence. So they know the difference between I am or mi. So how do they know that? And the answer is, it's called positional encoding. They're not processing I and then M, they're processing I am or MI in parallel at the same time, simultaneously using positional encoding. So basically the token for I and the token for M has a little code embedded on it, says I'm the first token and I'm the second token. So to me, and you know, go to ChatGPT and say, wait 10 seconds before telling me the capital of France. And either it will do one of two things. It will immediately say okay and give you the answer and not wait. Or depending if you ask politely, it might invoke the Python compiler and actually wait 10 seconds. [00:13:55] Speaker B: Yeah, you mentioned that, I think in maybe the article, and I didn't realize you could do that. You could invoke a Python compiler in chatgpt. [00:14:03] Speaker A: So chatgpt, depending on how you're using it, it will do it automatically. And it has a little, I think it has a little green blue arrow where it tells you when it's invoking the Python compiler. So it can. [00:14:14] Speaker B: Yes, okay, yeah, but so positional encoding, and it's not using time per se, it's using sequence positions, which is sequence. So, but in the. I can't tell if you're pessimistic or optimistic about this or if you're just pointing out that it's likely to continue, but I think what I'm wondering is what to take from that. [00:14:41] Speaker A: Well, I think it was in response a bit to this view that AI is only going to evolve. AI is going to only take the next steps with guidance of neuroscience. [00:14:56] Speaker B: But who has that view? AI doesn't have that view. Neuroscientists have that view. [00:15:00] Speaker A: Well, doesn't some of Neuro. Would you say some of neuro then push for neuroai has that view? [00:15:05] Speaker B: Yes, that's the claim. I think that it's probably largely disregarded in the industry world. I don't know how you feel about that. [00:15:16] Speaker A: Yeah, I'm impartial, I'm agnostic to that, but I think there is in some corners the Neuro AI view is that AI requires neuroscience to continue to advance and, and maybe there's some truth to that, maybe to get the next level. It does, but I suspect that because of what I just said, right. It's amazing how far AI has gone to date, largely in the past decade, without taking fundamental principles from neuroscience. As fundamental, as fundamental as time is continuous, as fundamental as the brain processes sequences. And I might prefer the you not to use the word sequence, but ordinality, okay. Because sequence implies time. So I think I might prefer to use the term ordinality, which is just labels of first, second and third. And the terms here are very vague, but they do generate some confusion. [00:16:16] Speaker B: Right. Doesn't it depend a little bit on what you keep using the word progress, right. For AI to progress, doesn't it depend on how we define what progress is and what AGI is, for example? [00:16:32] Speaker A: Absolutely, Paul. And I don't think anybody knows. Some people's progress might be other people's regress, right? I don't know. So it depends if you view sort of eventually achieving AGI or what does that mean? [00:16:50] Speaker B: Does that mean something to you? I know what singularity means better than I know what AGI means. [00:16:56] Speaker A: Yeah. So. Yeah. So again, but the irony there is, right, AGI for artificial general intelligence, forget artificial intelligent intelligence, just try to define intelligence. There's no ingredient, there's no agreement on what caused intelligence. So it's not very intelligent to talk about something without defining intelligence. So, so, but that's, that's true in a lot of fields, right, Paul? I mean, we just have to accept, you know, you know, you know this, whether it's the consciousness field or the free will field or even what a gene means. Yes. We have to deal with, with abstract comp sets sometimes that aren't not very well defined. So. Yeah, and that's okay. I mean, AGI is ill defined, but so is intelligence. I mean, you know, are humans smart? Is that a high bar or a low bar? [00:17:46] Speaker B: Personally, I think depends on the human. [00:17:48] Speaker A: I think human intelligence is a very low bar for intelligence. [00:17:52] Speaker B: Okay, we're in agreement there for sure. Yeah. So, and okay, so maybe we can move on here. And I don't know the best route through these topics that we want to discuss because on the one hand, you have teamed up with a physicist to talk about how physics and neuroscience sees time differently. Maybe we should start with that. Okay. [00:18:17] Speaker A: And I'll add there that we were just talking about transformers, and transformers might be a good lead into that because I also make this analogy, right, because in Transformer the past and present are all there at the same time because you're feeding in the whole paragraph to the transformer simultaneously in parallel. So there's a slight analogy there with what Carlo Rovelli and I talk about the distinction between two views of the nature of time then. And I know we talked about this a mere seven years ago in terms of presentism and internalism. And so transformers are very much sort of eternalistic or sort of block universe in that they. The past and present is there at the same moment, if you will, and the language here gets a bit fuzzy to feed into the system. A recurrent network doesn't work that way. A recurrent network is inherently presentist. So what Carlo and I, in this piece bridging the neuroscience and physics of time, were attempting to bridge is the two views in philosophy, physics and neuroscience of the nature of time. One view is presentism, which is our intuitive view as neuroscientists in terms that we all take for granted, right? The present is real. The past was real. It's no longer real. We have residues, memories of what happened, but that's all it is. And the future is not yet real. And presentism is the block universe view in which the universe is a 4. In the extreme view. This sort of what we call static eternalism is a block universe with four dimensions in which time in some sense has already transpired. All of time, past, present and future are all equally real, if you will. And so one way I find convenient to think about this is that now is to time as here is to space, meaning that just as you and I realize that we're in our own location, we're in our here, but other locations are perfectly valid. We accept that. But in presentism we don't accept that I could be in other nows, but in the present eternalism, yes, you could be in other nows, because all presents are sort of equally valid. [00:20:44] Speaker B: I always think of the block universe. And maybe this is from your book. I don't remember whether you did this, but I always think of it as like a block of wax that you could like sort of very thinly slice through any slice you want and then visit, you know, the different slices. I'm not sure if you had a graph in your figure in your book that made me. [00:21:05] Speaker A: Yeah, yeah, I know I didn't mention wax, but, you know, I think now that you're mentioning, I think there is a Greek philosopher that somehow makes the analogy with the cutting edge technology of the time, which was probably wax. So. [00:21:21] Speaker B: Well, there was Descartes, ball of wax. But that was epistemology, not block universe stuff. So. [00:21:26] Speaker A: Yeah, well, but they did go back to like some Greek philosophers, like anything, it was paraminities. Paraminities had this view that it was sort of unchanging. [00:21:37] Speaker B: Right? Yeah, Everything statistical. Yeah. I found this chapter difficult to read and I would also just love to hear how this collaboration came about with Carlo Rovelli because it was new to me. I didn't know if you guys knew each other. How did it come about? [00:21:56] Speaker A: Carlo and I had been to a couple of meetings together on time, and there's a couple of philosophical symposia on time. [00:22:05] Speaker B: It's the philosophy that binds you together then. [00:22:08] Speaker A: Yeah, absolutely, absolutely. And there's this ongoing tension that exists between physics and neuroscience where we sort of like to blame each other. So I think neuroscientists like to sort of say, listen, we live in a presentist universe. Time is flowing, time is changing. And if that's not what physics is telling you, figure out where the physics went wrong. And whereas neuroscience, where some physicists, and I'm exaggerating here, I don't want to get. [00:22:43] Speaker B: Well, this is how you set it up in the chapter as well as like a dialogue kind of of the physicist blaming the neuroscientist, the neuroscientist blaming the physicist. [00:22:52] Speaker A: And yeah, with the physicist saying, listen, some physicists, not all, some physicists do view the eternalist view as dogma and that just take it for granted. And in that view, sometimes it's hard to imagine where the flow of time comes from, where the we have this subjective feeling of the flow of time. And so some neuros, some physicists view that as an illusion of the mind. So they say, listen, you neuroscientists figure this out why we have this illusion of time flowing, because it's not. And I'm exact sort of dichotomizing things here. But that's. [00:23:33] Speaker B: And that's because math equations work forward and backwards basically. Right. [00:23:38] Speaker A: So that's one of the reasons. So that's exactly right. So why. So this is very counterintuitive to most neuroscience. So why would the physicists take this view? And there's a couple of reasons. One is that the equations, the fundamental equations of physics do, are so called time reversible. Doesn't matter if you want to run them in the forward direction, you can use that, predict the future, or you can use them in the backward direction to retrodict the past, but also because of relativity. So that's independent of relativity, although the equations of relativity are of course time reversible. But in relativity, because time and space are sort of a trade off. And there's something called Minkowski space in which the best way to visualize this is that if there's no absolute present and you're going at one speed and I'm going at another speed, our clocks are moving at different speeds. So one way to do this, to go back to your block of wax, is you can imagine all of us in a block, and if we're cutting that either orthogonally or diagonally, you can sort of see that we might be on different, similar planes of simultaneity. So that suggests this spatialization of time. So that. But that's an interpretation. But as I think hopefully came out in the chapter, you know, we're both 100% in agreement that there's no strong empirical evidence either way. So these are fundamentally open questions. And indeed there's not many ways to even distinguish between those empirically, except one. [00:25:16] Speaker B: Well, there. So, okay, so the goal of, from what I gathered, the goal of the chapter in this case was to figure out what you guys, what physicists and neuroscientists agree on, what you still disagree on. And one of the things that it seems to me it hinges on you already mentioned, which is our subjective experience of the flow of time. And then the other thing that it seemed that I took from it was okay, in the mesoscopic temporal scale that we humans live in. It's sort of like, okay, that neuroscientists would have the presentism view because that's all we can study because we have only a limited capacity to experience fundamental lower and upper limits of time anyway. And outside of that, where the philosophical, the physicists, philosophical underpinnings of time would come into play, is even outside our epistemological domain, if you will, because we can't measure anything like that, we can't observe it. Does that make sense? [00:26:34] Speaker A: Yeah, I don't. I mean, certainly that we live in the mesoscopic scale in which relativistic speeds or quantum phenomena aren't on our day to day existence. And this makes sense in terms of why our intuitions are absolutely so horrendous and inappropriate for understanding those levels. But I don't think the lesson there is that our view of presentism is an artifact of our mesoscopic existence because the block universe would operate on that exact time scale. The block universe is scale independent for time. No, I don't think I would take much from that. And I would say that that's precisely the part of the problem. Unlike gravity near black holes or time at relativistic speeds, the flow of time should occur basically across all scales. So my argument is that because the brain evolved to make sense out of the universe in which we live, our mesoscopic universe and our mesoscopic world is governed by the laws of physics. So the fact that we see it flow, I would argue, reflects reality. [00:27:56] Speaker B: So you mentioned in the chapter, like it's governed by approximately, good enough, Newtonian physics. And so that's the world we're living in. And therefore that's what our sense of time is, right? [00:28:11] Speaker A: Well, that's what our sense of physics is and it's what our sense of time is. But the sense of time, again, even in the block universe view, applies to the mesoscopic scale. So I would say that no, our perception of time is not a consequence of the fact that we live in a world sort of operating on the mesoscopic scale. The argument we made, and then I think Carlo and I disagree on this point, is that we perceive it. I would argue that we perceive it because it is actually happening. It is a flow of time, which is not. Which is. It's important to be clear here, which is absolutely not inconsistent with the laws of physics. The laws of physics allow for presentism and they allow for eternalism. As long as you're talking about local presentism, not some absolute time a la Newton. [00:29:10] Speaker B: But I've stopped believing in laws. Are the laws. Is that a real thing? Is a law a real thing? [00:29:17] Speaker A: So have you been getting many traffic tickets since you stopped believing in laws? [00:29:20] Speaker B: I've never believed in those laws. [00:29:22] Speaker A: But what do you mean you stop believing in laws? [00:29:26] Speaker B: I jumped in there because I was going to ask you what you think about, like the. You talk in the paper also or in the chapter about mathematics and laws, as far as I can tell, are mathematics. But let's say. Let's say the. Let's say a constant can change with the evolution of the universe. Would that still count as a law? [00:29:53] Speaker A: I don't know. I think that gets. You're picking something that's on the philosophical edge there. And sounds like you may have read Lee Smolin recently. And because he. That's precisely what he argues, that the constants of the universe are changing. [00:30:08] Speaker B: How could they not be? It's all dynamics. [00:30:11] Speaker A: Well, I think they could not be because the constants are indeed constant. So I think they could not be. But they could be too. I know I wouldn't say either way. But the bottom line is that again, going back to our mesoscopic corner of the universe, those laws are pretty goddamn, really goddamn good. [00:30:31] Speaker B: Yeah. [00:30:31] Speaker A: I mean, they are. It's. We have no right to have figured. We unusually smart apes have no right to have really figured out those laws. I quantum scale and the cosmological scale avoiding, you know, assuming you're not living beside a black hole or a collider, you know. Yeah. It's, it's absolutely astounding how good those laws are. So, so no, I, I don't know why you don't believe in laws, but I, but yeah, I do. And that was going to come up in our next topics. [00:31:05] Speaker B: Right. Well, I think what I actually mean, I just, I just self elucidated this to myself is I think I don't believe the same way that I used to believe about what a law is. Maybe that's the right way to put it because I, you know, I remember Yael Niv saying one of the beautiful things about mathematics is that because I guess I rail on this too frequently. But she says, well, you can take your experiment, you can do the math on it, you can pick up the math, you can move the math, put it down in a different thing and it still works. And that's the beautiful part about it. And I like that description of it. And that speaks to laws, I suppose. [00:31:46] Speaker A: Yeah, I would say that. The way, and you've probably seen me say that, is that the brain is full of cognitive biases. We make wrong decisions. It didn't evolve to do what we're asking it to do. Whether it's understanding the laws of physics or understanding consciousness. It didn't evolve. That's not what it's primarily. [00:32:06] Speaker B: Some people would actually disagree with that as well. [00:32:09] Speaker A: Well, yeah, we can go there, but, but I would maintain that, that the brain clearly didn't evolve to understand the laws of physics or to understand the nature of consciousness, not talking about minds. And so one way around that, to me, I view mathematics as the best debugging device ever invented. So once you, once somebody comes up with a set of equations that describe the incomprehensible, whether it's general relativity or Schrodinger's equations in quantum physics that make no sense, it doesn't matter that we can't understand them, that we can't interpret them, the bottom line is if we use them, we can predict the world around us. So, but I, so yes, I view the equations as our device that allows us to understand those Laws are those equations. [00:33:05] Speaker B: Where are you on the Platonic side? Right. What? You're shaking your head. [00:33:12] Speaker A: Yeah, no, no, I'm not a Platonist in any way. I don't think they're reflecting their objects out there in the universe that. Why is it so perfect? [00:33:20] Speaker B: The math is kind of perfect in that way, so it's tempting to ascribe. [00:33:26] Speaker A: No, it's. Math is not perfect. Math can be perfect in certain configurations. So math can capture truths. But there's a lot of bad math out there. Some things you can write down mathematically that's. That's just applicable or applied math. And again, we might come back to this topic as well. In terms of, you know, mathematics that is not particularly accurate in some sense, in terms of reflecting reality. So, no. Mathematics is agnostic to whether it's good or bad. Mathematics sometimes captures things that reflect how the universe works and other times not. [00:34:04] Speaker B: All right, so you've alluded to it twice. Are we. Are you ready to talk about integrated information theory? Is that what you were. [00:34:11] Speaker A: You're. I did allude to it, but you're the boss. [00:34:14] Speaker B: Oh, yeah, I'm the boss. Okay. Well, I mean, we can be all over the place, so that's fine. Another reason why we're talking today is because you are part of the IIT Concerned Consortium. Is it a consortium? Is it a group? [00:34:31] Speaker A: I don't know. I think it's just a concerned bunch of people. [00:34:35] Speaker B: Yeah. Okay, so I don't think. [00:34:36] Speaker A: No, I don't think we're organized enough to be a consortium. [00:34:39] Speaker B: Okay. All right. We have to define consortium. All right, so I just had on. And the problem with this is I don't know if our episode is going to be released before or after the episode I'm about to speak about. I just had on people from a group called Cogitate, whose purpose from the Templeton foundation is to be an unbiased third party to test various theories of consciousness in an adversarial collaboration manner. Which means that the proponents of, in this case, two theories of consciousness have to agree on experimental questions that can be asked where the answers will provide evidence disconfirming one of their respective. One of the respective theories. And so the people I just had on were. I guess they're all postdocs now, I think maybe I can't remember. But they have been essentially running these experiments that were pre registered to test the consciousness. I guess it's theory. Integrated information theory versus the global neuronal workspace theory. And so listeners to our episode will either have heard that already or not. I'm not. I'm not sure. [00:36:08] Speaker A: Depending if we live in a presentation. [00:36:11] Speaker B: Yeah, it's all wax anyway, so. All right, so you were part of a group. There's, I guess it is controversy in integrated information theory world, where. So integrated information theory is a kind of. Has gained a reputation as being one of the. I don't want to say leading theories. Well, you would say that because you're part of IIT concerned, right? [00:36:39] Speaker A: Yeah, I think it's considered a leading. [00:36:42] Speaker B: Theory in popular press, let's say. Right. [00:36:44] Speaker A: Yeah. [00:36:45] Speaker B: So therefore it is a leading theory. A lot of people know about it. Seems really cool. Internal causal structure identified with consciousness. And then there was this backlash. And the word that is causing all of the fuss is pseudoscience. So there's backlash that a big group of people, not a consortium, suggested that IIT, we can say from now on, is pseudoscience. And you wrote a piece recently, you and many others, sort of hashing out why you do consider it pseudoscience. Why you. You justify why IIT is pseudoscience. I think part of the problem, before we start talking about this is that in my mind, probably in yours, too, the word pseudoscience is a slander against something. It is bad, Right. To call something pseudoscience. It is your fake news, your. It's like the worst kind of thing to say about anything scientific. Would you agree to that? [00:37:56] Speaker A: Yeah. I mean, it's a slander. Absolutely. I mean, it's not good. I agree with that. But I think, by the way, I mean, this was obviously talked about a lot in the letter, and I think we ended up not using the word pseudoscience. [00:38:08] Speaker B: I can't remember in the original letter. [00:38:10] Speaker A: I think we used the term unscientific. [00:38:12] Speaker B: Oh, okay. So that's a little less bad. I was gonna say a little better. [00:38:17] Speaker A: But they're just words. I mean, they're just words. [00:38:19] Speaker B: Yes, but they carry connotations, right? [00:38:21] Speaker A: Yeah, yeah. [00:38:22] Speaker B: We're pseudo. Like using the word pseudoscience in my mind, and I don't know, now I have to reassess it because. Semantic drift. Right, but you use the word pseudoscience. You're trying to, like, just cancel any validity of, you know, anything. But that's maybe just in my head. And. And in the original piece, I. I didn't revisit this, but I think it said. I think the phrase was like, could be considered pseudoscience. It wasn't like, this is pseudoscience. It was like, this could be considered pseudoscience. From this point of view? [00:39:00] Speaker A: Yeah, I don't think, I don't know how useful that direction is. I mean, what direction? I think focusing on the words to describe it, I think it's much more fruitful to describe, yes, the theory itself, but, but to build on that, it's important. Words are important. I get that. I think the, the conclusion was that the word unscientific might be less. You know, emotions cloud judgment. So as soon as you use certain types of words and then people become defensive, then people shift into a different mode where they're using less reasoning. And this is, this is the classic think fast, think slow. This is classic systems 1, system 2. Once you bring in an emotional valence to something, it tends to cloud reason. So I don't, I don't think focusing on the words is particularly important, but I understand why people debate on it. But, but I think it's much, much more useful than to focus on why 100 people signed a letter concerned that IIT is unscientific. [00:40:14] Speaker B: Yeah, I think that that would be fine to start there, but so then there's that original letter and then there's this most recent letter which, I mean, I guess it alludes to the cogitate work where. So the whole point of cogitate is. Right. IIT has you can make predictions based on its theory and maybe I'll let you say a word about what IIT is and then you know why it's unscientific. But it makes predictions and then they tested those predictions when that sounds like science. [00:40:47] Speaker A: Yeah, okay, so that's great start to that. First place. First place. Let me just make put a couple of caveats. So this is so I don't think this is a consortium. This is just probably better looked at as a bunch of cats people herded. So, and the fact that they did manage to herd carnage hundred cats to sign something I think is something that people should reflect on as opposed to why that came out. And I should also say that all that if you ask a hundred people why astrology is a pseudoscience, maybe they're all going to agree that astrology is pseudoscience or unscientific. But you're not going to, all those people are not going to agree on why. So if I asked you why astrology, you may think astrology is pseudoscience, you probably going to give a different answer than me. So I can't speak for the group, Paul. All I can do is share my opinion and that's all I can do. [00:41:56] Speaker B: That's what you're here for. [00:41:58] Speaker A: In no which way, shape or form am I representing any group whatsoever. [00:42:03] Speaker B: And it's important you said opinion right there as well. [00:42:06] Speaker A: Yes, yes, absolutely. So. And you brought up cogitate and maybe we can come back to that. But just, you know, cogitate is collecting data. And collecting data is always a good thing. You know, high quality data that's going to be open source. That's wonderful. So, and that, and I think everybody supports that. [00:42:28] Speaker B: So I think, I think, let me just say like, so they, they, it was like 250 subjects and they collected magnetoencephalography, MEG, EEG, FMRI and then they had three predictions that they were testing and it pre registered it all. I already said that. Anyway. Great. [00:42:48] Speaker A: Okay, so, so I don't. All my concerns with IIT have nothing to do with cogitane. Yeah, they have to do with IIT and the laws of physics. So I think it's important for people to understand, and this is a misconception a lot of people have, that IIT is not a neuroscience theory. IIT is a theory of fundamental physics that proposes a new ontology to the very structure of the universe, to the very properties of the universe. IIT proposes that certain configurations of matter are conscious. That, and this is why some people consider it to be panpsychist, is because under it many forms of matter, whether they're the neural sort or not, can be conscious. So you have. So that's the first thing as to why I'm going to focus on the laws of physics here. So the most succinct way I can say why I am concerned that IIT is not. Is unscientific, is that you can't propose new laws of physics without integrating those laws into the existing laws. Or else you have a free floating, unmoored, unintegrated rule or law that you can't sanity check. You don't know if it's consistent with the existing laws or not. Physics is an incomplete puzzle, but as we just mentioned, it's a beautiful puzzle with a lot of pieces already embedded. You can't make up new pieces willy nilly without trying to see if that piece you just made up can be integrated to the existing pieces. Because then you don't know if you're violating all those other pieces. You don't know if you're violating the laws of physics. So my main objection with IIT is I don't think it fits within the normal scientific method to make up entirely new laws, integrate them with existing, or else it's a free for all. Right? What prevents you and me from making up a new law, writing down the equations that describe it, and not integrating it with existing laws, which is required in order just to know if there's a violation? And in my opinion, and that's just me, in my opinion, yes, I do think IIT may be violating some of the laws of physics. [00:45:22] Speaker B: Is the crux here because of the way IIT is developed from the axioms that are supposed to be from phenomenology, from the phenomenological perspective. Is that where it goes unscientific or no? [00:45:40] Speaker A: To me, maybe not. To me, it's really the fact that it's an island. Science isn't integrated. So here there's a bit of irony, right? Because IIT is integrated information, but it's unintegrated with the rest of science. So you can't have freely floating theories that don't have any other connection with the other laws of physics. So perhaps it would be helpful if I try to give an example of what IIT is and where I think it may or may not be in violation of existing laws. [00:46:18] Speaker B: Right, that'd be great. [00:46:20] Speaker A: Okay, so IIT is typically expressed as something systems that have internal causal power, which those three words don't necessarily jump out in terms of what they. Intuitively, we're lost already. [00:46:32] Speaker B: Yeah, yeah. [00:46:34] Speaker A: So let's try to do. I'll do the best I can to actually give a real example now, because it is not a neuroscience theory, it doesn't require neurons. We can just think of logic gates. So let's think of two logic gates, A and B. And those are such. They're just called copy gates or threshold. [00:46:56] Speaker B: They can either be on or off. [00:46:58] Speaker A: They're on or off. And these gates, if they receive a zero, they output a zero. If they receive a one, they output a one. That's all they do. So one thing that IIT requires is knowing if the current state of the system has information about past and future states. Now, in reality, there's all this partitioning, sub networks, purviews, mechanisms, and so forth. But for our simple system composed of A and B, it's sufficient to just think about it in the following way. If B is in state one, does that constrain the past of the system? And the answer is yes, it does. Because if B is in state one now, A had to be in state one in the previous time step. That's the only way B could be in state one. So that's good. So that means as the potential to be conscious has the potential to have a positive phi value, but you also have to look in the future. So does B constrain anything about the future? And the answer is no, it does not, because B doesn't circle back on itself or onto A. [00:48:05] Speaker B: It's not predicting anything. [00:48:07] Speaker A: It has no constraints. And these constraints are really in relation to sort of above and beyond what a random state would have predicted. But the answer is, as you just said, no, it's not predicting anything for the future. Now here a neuroscientist, by the way, should ask, okay, but what do you mean past and future? When in the past and when in the future? And that is a bit of problematic with it because it actually requires you in some cases to look at all points in the past and all points in the future within some range. So it's not. So it's a discrete. It's like well written for a computer because you know what the time steps are. But the brain doesn't work like. Like that. But let's. Let me. Let me proceed here. So, but keep that in mind because problems do arise there. So now if we connect B back to A. So now we have a recurrent neural network. And IIT is about recurrency in many ways, like many theories of consciousness, by the way, which is something I'm totally. [00:49:08] Speaker B: I can totally get behind, like the recurrence as well. They just didn't know how to handle it. [00:49:13] Speaker A: I'm sorry, who did? [00:49:14] Speaker B: McCulloch and Pitts. [00:49:15] Speaker A: Absolutely, absolutely, absolutely. And there's even a theory called recurrent process theory of consciousness. Okay, so now B, the state of B will influence the future. It does constrain what can happen in the future. Okay, so now it turns out that the system's conscious. We added a connection. And now the system's conscious, has a phi value of one. Okay, so that's fine. Now one of the things. So maybe it will be useful if we try to do some thought experiments here. So one of the things about IIT is that it's space and time independent. So if those units. If we get those units and separate them by a huge difference, it doesn't really change. The system is conscious. The second thing that's a bit strange about IIT is it's conscious even if both units are off. So they're both zero. So let's get our units and cut the wires between them and use optical beams to connect them. [00:50:19] Speaker B: Let's take it to absolute zero Kelvin. [00:50:22] Speaker A: And let's take it to absolute zero Kelvin. Although maybe not, because I don't know if it would be responsive. Then I don't know if you can. All right, all right, I tried, but I, like, I just don't want to. Are you trying to trap me, Paul? So then. So we bring them out to whatever opposite ends of the solar system. So IIT says that, yeah, it still has a phi value of 1. As long as the connection is good, that's really important. And so where is consciousness? You know, is it somewhere. It's a bit mystical here, right, because is it on Earth? Is it in Neptune? Is it in between? And then, you know, that's just the way it is. So. Okay, but that's fine. That's what IIT predicts. But now let me ask you, here's the next step in our little thought experiment. Let's block the optical path. So now we're going to block the optical path. [00:51:21] Speaker B: No communication, remember? [00:51:23] Speaker A: The optical path is not transmitting anything because they're in zero states. [00:51:26] Speaker B: That's right. Yeah. But we're blocking it without it being active. [00:51:29] Speaker A: Okay, Right. So you're blocking it without it being active. So most our conventional. In our conventional thought process as physicists, nothing should happen. Right. There's no precedent for this. [00:51:42] Speaker B: Right. You're going to say the only thing that does happen is that phi goes to zero. [00:51:46] Speaker A: Yes, it goes. So very good. It goes to zero. How long does it take to go to zero? By most accounts, it goes to zero instantaneously. So even though these two gates are very, very far apart, it seems to go to zero instantaneously. Okay? Now, when things happen instantaneously in physics, you have to raise some flags at least, because you could start running into trouble because there's something called faster than light transmission is a huge, huge. No, no, in physics, you don't want to get into that. You don't want to get. You. You don't want to. Yes, and I know you don't like laws, Paul, but it's wormholes. [00:52:34] Speaker B: I like wormholes. [00:52:37] Speaker A: Those are fictitious objects for now, Paul, though. So you don't want to violate that in terms of faster than light transmission. Okay, but I don't think that's happening yet. So, you know, phi drops instantaneously, but you're not really transmitting any information there. So that might be okay, I don't know. But let me now take the final step in the little thought experiment, and let's call it the Middlebrook split brain experiment. And we're going to get, with your permission, one of your hemispheres and keep it on Earth. Other hemisphere. [00:53:17] Speaker B: Take one of my kids, do this with one. [00:53:20] Speaker A: Do you have extra. [00:53:21] Speaker B: Do you have backups? Too many. [00:53:25] Speaker A: No, let's stick with your hemisphere. I don't want to get the kids involved. I don't want to get the kids involved. And so we'll take one hemisphere, put it on Earth, and one hemisphere, put it on Neptune. And we have to have these beams connecting them, right? So by most accounts you'd still be conscious. So the thing about conscious entities is you can communicate changes in consciousness, right? So now if we interfere with some of those connections and if we knew what we were doing. And this is a question, I'm literally asking a question, is that now if we block those connections, that will change your conscious states. And you could communicate, let's say your left hemisphere. The hemisphere that can communicate is over there in Neptune. Or if we want to go old school, we can put it on Pluto. Up to you. [00:54:20] Speaker B: Yeah, I'll do it. [00:54:22] Speaker A: So then the question is, is now do we have a way to instantaneously transmit information? Because now you can communicate those changes in your phi in your conscious level. And here I should. I want to make a couple of caveats here. I think an IIT proponent would make a couple of points here. One is that by adding the delays, that changes this partitioning. And maybe your consciousness is just in Neptune or an Earth and it wouldn't work. But I think we can get around that because it's a thought experiment and we can just change the delays in the inter. We can match the delays of the inter atmospheric connections with the intra hemisphere connections. Number two, I think a proponent of it would point out, well, because I'm interfering with the pathway here on Earth, I become part of the cause and effect structure. That gets really complicated now, because now do I. I think the argument might go that the phi is already not goes to zero or is changed by the fact that. That the system knows that at some point in the future I'm going to mess with those connections. So it gets a bit complicated. It's a bit uncomfortably close to clairvoyance for me. But I. But my point is the following. And this is. Sorry to give you this whole complex. [00:55:50] Speaker B: It's a lot. It's a lot. It's okay. [00:55:52] Speaker A: I know. Sorry to give you this complex, but. But it is a lot. I'm sorry. And I think that's why sometimes it's hard to find why it may be violating or not to the laws of physics. So my point is simply the following. That because it is free floating, because it's unmoored with the rest of physics. I can't know the answer to this question I'm posing. I don't know if it's violating the laws of physics because it's not sufficiently well defined, it's not sufficiently well integrated with the rest of physics to know if those laws are being violated. That's my main point, is that if you have a system, a scientific system in which people can make up new laws, new ontologies without integrating them into the existing, I don't see how that allows for the normal scientific process to unfold because we don't know if they're violating the existing laws. That's my only point. [00:56:53] Speaker B: That's your only point. Okay. [00:56:56] Speaker A: It was a long run. Thanks for hanging in there, Paul. Thanks for hanging in. [00:56:59] Speaker B: It's been a great show. [00:57:00] Speaker A: I don't know if your listeners will hang in there, but. [00:57:02] Speaker B: Well, all right, I'm gonna. I'll turn the tables here real quick, like. So you said all that and yet it could be correct. Right. I guess your point is there's no way to know whether it could be correct because it doesn't. [00:57:24] Speaker A: But that's not how science works. It could be correct. Question is, how would you determine if it's correct if you don't know? And what would prevent you and me from you or me going home tonight and writing down some equations that are incalculable? Right. [00:57:38] Speaker B: Well, but if we agreed to a new ontology, it could be correct, right? [00:57:45] Speaker A: If there was, I don't think it's a question of we agreeing to the new ontology is if the new ontology is supported, then we'd agree to that. But the first step in that, before we agree to a new ontology, I want to know if it's violating the existing ontologies. To me, that's the first step of the scientific process. I'm willing to say, as I said before, the laws of physics are not immutable. They have room for change. But I need to know if the new ontology proposed. New ontology proposed in the absence of any empirical data whatsoever, is consistent with the current ontologies. [00:58:24] Speaker B: And it's worth saying here that in a pauperian, Karl Popper sense, the whole business of science is the capacity to actually be wrong according to the integrated scientific knowledge base that we have. The other thing that I wanted to say that's very brief, which I find. I don't know why I find this humorous, but. So after your whole spiel, and we're not talking about Cogitate, but it's interesting that. So we're talking about Neptune and Pluto and cutting an optic line and instantaneousness or not. And then when it all boils down to it, the prediction is it's in the posterior parietal cortex. [00:59:12] Speaker A: Yeah, so that's the other. Okay, so again I said the Cogitate project is great. I think collecting data is good, but I didn't say that the way it's being interpreted or used is ideal. And I think that is non ideal because it's being posed as a test of two theories and as you discussed, that are highly under constrained. And if you read the conclusion statement by Stan Dehaene, he correctly points out that IIT is not really being tested. What's being tested is a loose interpretation of IIT that predicts, for reasons, honestly, I don't fully understand that it's in the posterior parietal cortex. [01:00:03] Speaker B: Structural reasons, right? [01:00:05] Speaker A: I guess. But remember that as Dan states, IIT is the mathematical backbone. IIT is, and that's not being tested. So IIT lies on a mathematical backbone which everybody admits can't be calculated because it's incalculable because the math, by the time you calculated heat death of the universe, the phi for a c. Elegans with 321 neurons, and you start those calculations today, by the time you finish, the sun would have literally expired. So it's of limited use that mathematical structure. So. And again, this is something else I want to make clear. If IIT proponents are arguing that phi is correlated with consciousness or a measure of consciousness, I don't think the IIT letter concern letter would exist. [01:01:01] Speaker B: Oh, it's just the identity. Identity aspect. [01:01:05] Speaker A: It's abs to me. Absolutely. I think so. When they say so, how can. So they're not really testing IIT because what's applauded in IIT is that it has a mathematical background. But they're not testing the mathematical background because by definition phi can't be calculated. So at best it's estimates. So I think, yeah. [01:01:35] Speaker B: Can we spend a minute. And if you don't want to, that's okay, but. Okay, so in the Cogitate case, they ended up testing IIT versus global neuronal workspace theory. And are we okay as a scientific community about global neuronal workspace theory? Why. Why would that be okay? Whereas IIT isn't. And what you're going to say is that because you can epistemologically test things. [01:02:02] Speaker A: Well, but no, I'd like to back up on that. Can I? [01:02:05] Speaker B: Yeah, yeah, please. [01:02:06] Speaker A: I mean, the answer is quite simple, right. If you have two theories, Paul, and one requires changing the laws of physics and the other doesn't. [01:02:15] Speaker B: I know, but. But my. [01:02:16] Speaker A: But that's. But what? [01:02:18] Speaker B: But my point. Okay, so when. When let's describe global neuronal workspace theory. I don't know if we want to do the Bernie Bars, Bernard Baars version or the Stanislaus Dehaen version or whatever. I think the Bernard Bars was more abstract philosophically, and then Dehaene sort of made it more neuroscientific. Would you agree with that? [01:02:42] Speaker A: Yeah. [01:02:42] Speaker B: Okay, but the idea when you write it down on paper is like, well, you have something, some brain activity that ignites and then is available informationally is available that is broadcasted to the rest of the brain, the parts that are necessary for your perceptual sensation of sandpaper or something. Right. When you're running your finger on sandpaper and it's required that there's an ignition and then it's broadcast. And in terms of neurons, I guess it's easy to say, well, then there should be a lot of spiking activity and then there should be the proper connections that you can measure with spiking activity in those touch sensorial brain areas. Right. Is it simple as that? [01:03:38] Speaker A: Well, I don't think it's as simple as that, but I think it's a fair summary of that. But I think the point you're trying to make is that it's also very vague and amorphous and severely under constraint. And I think this is what's causing the puzzle. So I'm glad you're bringing it up because I've talked to people and they say, well, you know, you guys are picking on Iit the IIT concerned letter. [01:04:04] Speaker B: And you don't want to be known as someone who's just picking on a certain group. Right? Yeah. [01:04:08] Speaker A: And people are a bit puzzled by that, but. Because they say, you know, but GNW is a bit fuzzy and vague. [01:04:17] Speaker B: Sure as hell is. Yeah. [01:04:18] Speaker A: And hand wavy as is higher order. [01:04:22] Speaker B: Theories and reentrant processing. [01:04:25] Speaker A: So it all is. So people are puzzled and I couldn't see a simpler answer or a simple dichotomy. Simpler dichotomy is that if your theories. Theories. And we should probably use theories in air quotes because they're not really developed yet. [01:04:43] Speaker B: Now we have to define theory. Yeah, so. [01:04:45] Speaker A: So. Right. So they're, you know, they're pre theories or ideas that we're calling theories. And that's Fine. But if you suggest a theory of globe of what causes climate change and I. That's consistent with the laws of physics, that doesn't require modifying laws of physics, and I suggest a theory of climate change that causes climate change, that requires changing the laws of physics. Those two theories are not on the same footing. That's a huge responsibility. So you should. What I think bothers many of us is precisely what you're getting at is somehow comparing a set of theories that, as imperfect as they are, are not requiring changes in the most successful ontology is scientific humankind has ever achieved, which is the laws of physics and another theory that does. So if cogitate were to come to the conclusion, which it sort of does, that, you know, GNW and it are close to 50, 50, whatever are those two theories on. Are those two theories on the same level? [01:05:51] Speaker B: Of course it was going to come out like that. How else could it come out? [01:05:54] Speaker A: You know, but, but, but my point is, is if they come out 50, 50, but one of them is right, requires changing the laws of physics. Who's ahead? It's the one that doesn't require changing the laws of physics. [01:06:06] Speaker B: How very Bayesian of you. [01:06:09] Speaker A: It is. It should be Bayesian. That's exactly right. [01:06:12] Speaker B: Oh, actually, like in my conversation with them, we were talking about like, well, how do you. How would. Given the results, like, how would you even measure. Like what. Which one you believe more. And it is a Bayesian method. I think Carl Friston suggested this maybe, which is funny. [01:06:31] Speaker A: Yeah. And the idea there was if you have a lot of things that are graded, you have many measures. How do you do that? So you can have a Bayesian approach, but as we know, in the Bayesian approach, you should probably take in the laws of physics. [01:06:41] Speaker B: That's a strong prior. [01:06:43] Speaker A: That would be the strong prior. I cannot emphasize how strong of a prior that is. [01:06:49] Speaker B: Yeah, but I mean, so people say it got us to the moon, but we all know we haven't been to the moon. Right. [01:06:57] Speaker A: Okay, fair enough. Fair enough. Good point, Paul. [01:07:01] Speaker B: Okay. All right, let's see. Is this a. I just said Carl Friston and that maybe that's a segue to talk about what you've been up to lately. Have we left anything on the table with. You did a great job, by the way, of being diplomatic. And I don't think you're going to get in trouble. That's good. [01:07:20] Speaker A: How. I probably am. I probably am, but thank you for saying that. [01:07:24] Speaker B: Yeah. Okay. Well, Carl Friston, I mentioned him because he suggested this Bayesian approach, which when I was talking with the Cogitate people, I was like, oh, yeah, obviously that's the way to do it. And then everyone's response was, yeah, everything is obvious in retrospect, you know, like when you're in the thick of it. Science is hard. Science is hard. It's never obvious what to do. Anyway, Carl Friston wrote a. I don't know if it's a commentary on your recent organo. It's not an organoid. Organotypic. What's the difference between an organoid and organotypic? [01:07:58] Speaker A: So organotypic slices have been around for many, many decades, since the 80s in which you get. [01:08:03] Speaker B: Brain slice. [01:08:05] Speaker A: Yeah, so. Sorry. So, yeah. So in brain slice electrophysiology, people are more familiar with acute brain slices where you extract a slice of the brain tissue of a rodent and then you keep it alive under acsf, but you can also culture it and it can sort of stay. It's a brain tissue slice and it can stay alive for weeks and months. And people now, so there's some papers by the Allen Institute where they've done human organotypics. So the tissue is there. You have a lot of. You maintain a lot of the structural connectivity, the cell types and so it's, it's close to, you know, we look at it as a sort of little VLSI chip of the brain. So it's sort of there. And can we teach it stuff? Organoids are much. [01:08:53] Speaker B: Go ahead. Sorry, no, no. [01:08:54] Speaker A: Organoids are much more recent development in which you're getting. Starting from cells and cell cultures, you. That have pluripotent properties. You can sort of coax them into the early stages of neurodevelopment. That technology is still ongoing. It's a very promising one. But you can't really study learning yet or how adult cortical tissue is processing because those are still very, very early development. The synapses aren't fully adult functional. So it's very early development. I think they'll get there, but it's a bit early. [01:09:35] Speaker B: So it's just not similar enough to in vivo brain structure. [01:09:40] Speaker A: Neither. I mean. No, no, no, no, no, no. Very, very different still. [01:09:44] Speaker B: Yeah, okay. [01:09:45] Speaker A: They're sort of. They're small 3D spheres that, that. [01:09:48] Speaker B: Yeah, there's. [01:09:49] Speaker A: Yeah, yeah. [01:09:49] Speaker B: The connectivity. Hard to get it right. Basically in an organoid. [01:09:53] Speaker A: Mad as you can imagine, it's extremely. Imagine. Yes. [01:09:56] Speaker B: Yeah. Okay. So I just want to back up and say, so I used to extract brain slices from mice. When I was first a tech, this is my introduction to neuroscience world where I would go in and you'd have to sacrifice a mouse and then you would extract brain. Extract part of the brain, put it on a fine mesh slicer, diamond, sorry, vibratone, thank you, vibratome. And you'd just get like just the tiniest slivers of the brain. You'd put them in a solution that would keep the cells happy and alive. Meanwhile, you've severed many of the connections. The longer ranges range connections among the neurons, and then you let them recover. And then you could do things like patch clamp where you put little pipettes down and listen to the single neurons. You could record local field potentials, you could inject currents and so on. And the difference with an, with organotypic is that organotypic, you do that, but then keep growing cells. Like what is the. [01:11:13] Speaker A: So what happens is. So as you just said, you sever a lot of the long range connections. So this tissue is just. Each neuron has lost much of its input, so it's in a different dynamic regime. So what happens then? Our interpretation is then homeostatic plasticity kicks in and it sort of adjusts its thresholds, adjust synaptic strength to return to some ontogenetically programmed level of activity and dynamics in which you start seeing neurons that act like neurons where they can networks, I should say that act like networks in which activity can propagate. They can have upstates, they can have dynamics. [01:11:55] Speaker B: But what's the difference then between a brain slice and an organo typic. [01:11:59] Speaker A: Organotypic. So in a brain slice, this types that you worked in, generally they're much more silent. They're not in a dynamic regime in which they can self propagate activity because they've just lost most of activity. [01:12:14] Speaker B: We wait like an hour to record from them or something. [01:12:17] Speaker A: No, this is probably days, Many days. Yeah, yeah. [01:12:21] Speaker B: Okay. All right, so that was a crash course there in brain slice and organoids and organotypic slices. But so why did you want to use organotypic slices? And. And you used optogenetics. And maybe we should say briefly what optogenetics is. Actually, let's just say what the question was first. How about. [01:12:43] Speaker A: Okay, so. Well, the question is if neurons are computational devices, if neural circuits perform computations, why do we have to study those computations in the intact brain? So neuroscience has benefited from reductionist approaches, whether it's from aplysia or from zebrafish or from C elegans our take on that is just that the best reduced system for the mammalian neocortex is the mammalian neocortex. So we just look at this as our reduced system, and that makes it much more tractable. So we know the inputs, we know the outputs, we know the experience of the tissue. So then we can control it if we can find ways to give it activity patterns, reproducible activity patterns. And that's where the optogenetics comes in. So, yeah, the optogenetics is such that we can stimulate cells or subsets of cells with different optical patterns to see if they can learn and do prediction. And I think that's what sort of was the connection you were just thinking of in terms of maybe Friston's commentary or something. [01:13:58] Speaker B: Yeah, well, I found his commentary interesting because it focused very much in a confirmatory manner for his ideas. But that's not everything. That was almost. It's just one of many things that you guys found when you did this work. All right, so you have the organotypic brain and you're going to be able to shoot different light patterns of different, different light frequencies and different light patterns of those frequencies onto different subpopulations of the neuronal cells. And basically you're. You tried to emulate conditioning, learning, timing of these light patterns in a, in a traditional way and, and allow me to do this, and then you can correct me. And you took about 24 hours to just over and over bombard this cultured organotypic slice with these patterns that paired a conditioned stimulus with an unconditioned stimulus. I should say that more elegantly. But anyway, it's like a classic reinforcement learning paradigm where you're trying to entrain a pattern into the cells. And one of my comments was going to be like, well, man, 24 hours at such short latencies, you entrained the hell out of it. So I was going to ask you also, maybe, perhaps after we clarify what I just said about how it could not. How you could not how you could fail to see the result that you got. [01:15:41] Speaker A: Well, fail is always easy, Paul. There's many ways you can fail. [01:15:44] Speaker B: I'm doing it right now. [01:15:49] Speaker A: Speaking for myself, I have no trouble failing at any experiment. [01:15:53] Speaker B: We agree on that. [01:15:54] Speaker A: So. So, no, that's not our, that's not quite our take home message. But yeah. So, I mean, if you want me to explain that very briefly, the quickest way you can sort of understand it is we have one stimulus, we'll call it the red light. And Then just a train of stimuli that lasts whatever, close to 400 milliseconds. Then we have another stimulus. We'll call it blue light because that's what it is. And the blue light comes on either at the beginning of the red light or at the end. So you have two different groups, an early group and a late group, in which. Only difference between the experience now of the two groups is the temporal relationship. [01:16:33] Speaker B: And they're subpopulations. Although I know the red light. No, I think the. Yeah, the red light also slightly stimulates the blue cell. Guys, I don't. I might have it backwards, but. But otherwise they're. They're separate populations where when you stimulate with one, you get cell firing. That was about 400 milliseconds. And then the blue light was. Yeah, something like that. [01:16:59] Speaker A: Something like that. So you're just training it with different temporal relationships. I mean, it's no different. I mean, remember the cortex, the auditory cortex doesn't hear, the visual cortex doesn't. See, all cortex just receives patterns of action potentials. Right? There's no. Is there anything magical about the auditory cortex that makes it better for auditory processing? I don't know if you remember these incredibly popular experiments going back to Miragaka Sur, in which auditory stimuli, I'm sorry, visual stimuli, was reoriented into the auditory cortex and sure enough, shockingly, visual, those cells became orientation selective. So the point of that view is that the cortex is sort of a universal computational device that can process spatial temporal patterns of action potentials. So, yeah, we're just tapping into that logic. And by the way, please, nobody quote me saying that. The saying that the visual cortex is identical to the auditory cortex. Cortex. That's not what I'm saying. I'm just saying that ultimately both sensory or motor or auditory or visual cortex are processing spatiotemporal patterns of action potentials because that's all neurons see is spatiotemporal patterns of action potentials. So what we're simply testing is whether you can capture some sort of computation or some sort of experience dependent learning in these isolated autonomous chunks of neural circuits. And so after this training, in which, yes, you're just doing it over and over again every 20 or 30 seconds for 24 hours. You want to know is, does the tissue learn? Does the tissue change its structure, its dynamics according to its experience? This is just an early step, I hope, and I'm glad to you it seems obvious that it will, because it does, because we were happy with that result. And so what the result is Is then you test with just red light. Now you just give the train of red light and you want to know, do the cells in different groups behave differently? Is the dynamics of the system different, meaning that it adapted to its experience as you would hope it would. [01:19:17] Speaker B: Another way to say that is, does it miss the blue light? Then when, when it's been, well, let's. [01:19:22] Speaker A: Let'S just back off just one second and first ask, is it different? Okay, so, so one. And the answer is that it was different. Yeah, in that the blue and the red light tend to elicit activity earlier in the early group and later in the late group. So one interpretation of that late activity is that it was predicting something would happen or that it's a prediction app error that it predicted was expecting blue light, but it didn't arrive. And I think that's what Carl Friston was sort of focusing on, which was not our main point of this, by the way. Our main point was that circuits can learn in experimenting way and to capture the temple structure. One of the things that we were most surprised by, and by the way I should say this is work by an amazing graduate student in the lab was Ben Liu, who's now at ucsf. One of the surprising things that I never predicted and was surprised when Ben came to tell me is that there seemed to be replay in the slices in terms of spontaneous activity. He noticed the structure of the spontaneous activity seemed to be different in the early and late group, in which in the early group spontaneous activity often went on quickly and decayed. And in the late group, that spontaneous activity tended to sort of grow in time and have peaks after the expected time of the blue. So the overall overall conclusion is, as you may or may not know, a long line of our research is that timing should not be seen as a specialized function in the brain because timing is so important to everything we do that it is sort of a universal property of neural circuits. And one way to test that is to say, hey, if what we're saying is true, that neural circuits are intrinsically able to tell time, this is one way to demonstrate that. This is again something that might come up in a minute if, depending on the next questions you have to me, but this shows that yes, neural circuits, even without a body, can tell time. And I'll tie this back up to the iit, because by iit, these, these circuits are clearly conscious as well, because they. Because that's what it would predict is that you have the recurrence in the circuit. So. So it would also predict that These. That was one of the ethical implications of the letter in that it clearly predicts that these, a lot of organoids, fetal tissue and stuff, would be conscious. [01:22:09] Speaker B: So there's time and there's ordinality. Like you said earlier, what is the difference in a neuronal circuit? Like, why would it be telling time instead of ordinality? [01:22:18] Speaker A: Well, in neural circuits, ordinality, time are much more closely connected to each other because, as we talked about earlier, sequentiality requires something flowing in time. So when it comes to the brain, I don't normally use the term ordinality because ordinality refers to something that's discrete. It's one which is discrete from each other. [01:22:47] Speaker B: I was going to say sequence, but I wanted to avoid it because it. It brings to your mind, it brings in time. So. [01:22:53] Speaker A: But that was applied to transformers, not applied to the brain. [01:22:56] Speaker B: Yeah, yeah, but so then how do you distinguish, I guess, sequence from time? [01:23:04] Speaker A: Well, I don't think sometimes you do. I mean, I think they can be. Time is one way. So think of neural sequences. So neural sequences, like A la Birdsong. [01:23:16] Speaker B: Studies or timing is very important. Yeah. [01:23:19] Speaker A: So there you have sequences. So the sequentiality. So you can have neuron A, B, C, D, E, and that whole pattern can take a second, or you can have A, B, C, D, E, and the whole pattern can take a half a second. So the sequentiality is the same, but the timing has changed and that's fine. It's just speed of the dynamical system. There's nothing particularly deep about that comment, but it comes up, of course, in that you're using the dynamics. The dynamics is by definition changing in time and that it provides by changing the speed of dynamics. Now we have a way to change the speed of our temporal. Of our timing and our. I can speak very slowly or I can speak very quickly. Right. So that's the idea, is that you have the dynamics, but you can change the speed of the dynamics. There's nothing particularly profound about that. It's just a property of most dynamical systems, or many dynamical systems. [01:24:15] Speaker B: But also. All right, so I can speak slowly too, but that slowdown was uniform. Right. And sometimes I speak slowly, sometimes I speak fast. Then that's Birdsong. Right. Where the patterning of the temporal internal intervals is very important. And so that's where time. Maybe that's a distinction between time and sequence. Right. Because then you have timing within the sequence. [01:24:50] Speaker A: Yeah, Normally I think of timing as the container. So I would say you have sequentiality within the temporal flow and you can control the speed of that. But, but I don't think it really is fundamental here to understanding of the dynamics. So yes, some things I would say that everything is flowing in time and that flow of time can be captured as sequentiality irrespective of the timing or as truly in the timing in which now you're paying attention. So the way to think about this is if I ask you a order discrimination task. So people study this, I give you a B, I give you flash of lights and I give you a green light and a red light and I ask which came first, that's an order task. But if I asked, well, which one lasted 100 milliseconds or was the interval between those 100 milliseconds? Now that becomes a timing task. They both require order and sequentiality. But one of them is phrased as order, a temporal task in that the question requires units of milliseconds. The other sequentiality doesn't require units of milliseconds. It just requires ordinality. [01:26:05] Speaker B: Yeah, okay. Okay. So I had David Robe on recently. He's actually going to be visiting Pittsburgh in a couple weeks. I've never met him physically but should be able to say hello to him. He studies basal ganglia. He has come to the temporary perhaps conclusion. All conclusions are temporary, that brains do not measure time. That in fact the way organisms measure time is through behavioral processes. So there's the recent past few years have seen a lot of studies with where we're recording lots and lots of neurons and we can record fine tuned behaviors from animals while they perform tasks. And then, and he studies timing tasks specifically he studies the task that we talked about is rats on a little treadmill. Where the treadmill. So there's a, there's a treadmill. They're like in a little box and treadmill. There's a little pedestal on the back they can hang out on and then there's a reward port up at the front of it. And the whole task is just, all right, you have to time 20 seconds and then there's going to be a reward. Right. And the rats figure out these behavioral patterns while they're on the treadmill. They don't. He thought they would just hang out up at the reward port, basically running at the same speed and then wait 20 seconds and then get the reward. But instead what they did is that they sort of went through a series of attempts. That's anthropomorphizing, but they went through a, they learned a sequence of behaviors that just so happened to match the Timing of when the reward would occur. Okay, so then he needed an alternative explanation for this. He had a history of reading Henri Bergson. Elaine vital person who has a different conception of time ontologically than Einstein, as famous debates with Einstein and David has come to the conclusion that this is more closely matches what he thinks about what time. Is that in any way that time is not measured in brains, but that organisms use their own behaviors as proxies to estimate the time. Okay, so here his. Sorry for that long winded thing, but. [01:28:38] Speaker A: Can I interrupt you a bit? Is that you started this by saying that he argues that the brain doesn't tell time. Is that what you said? [01:28:46] Speaker B: Let me play you his question and perhaps. [01:28:49] Speaker A: Okay, okay. [01:28:54] Speaker B: Let me play you his question and I can play it again if need be. [01:28:57] Speaker A: Hello, Dean. Thank you so much for your inspiring work. My question for you is going to be extremely simple. In fact, if there are population clocks in the brain who read them. So I'm looking forward to hear your answer and your discussion with Poly, I'm sure is going to be very interesting and hopefully we'll have the occasion to meet and chat. All right, bye. Bye. [01:29:28] Speaker B: So who's decoding these population clocks? Oh, I just. I said decoding. I shouldn't have said that. We'll just stick to this question. [01:29:35] Speaker A: Yeah. So first place, I mean, the question of who's decoding is valid to any form or most forms of processing, right? So whether it's space or color. So I think the question is. [01:29:51] Speaker B: That's true. [01:29:51] Speaker A: Coming from a place where people assume there's something special about timing, but it's general of any code is that who is asked who's reading the code? This goes back to like little homunculi in the brain. So. So I don't think it's a particularly unique question to timing. And I also think it's a question that's fairly well answered, by the way. So, I mean, you don't need a reader, right? That's the whole idea of a computation. The whole idea that the computation, the code generates a pattern of activity that generates the motor pattern. So I think sometimes in trying to understand what Ravi's saying, and you phrased it initially as the brain is not telling time. That's a very extreme view. I don't know if he'd really. Sometimes I think he's saying that. Sometimes I don't know if he's saying that. But it's also a view that we know is sort of incorrect. Right? Because we know that if I give you A Morse code task. And you're tapping out Morse code, or you're processing Morse code, or we're doing the musical task. We know that the brain is doing that, and that was just an ex. That's why we're doing this in in vitro. [01:31:04] Speaker B: Right, but the brain's doing it. But. But it's constrained by time. But I think his point. And I'm. You know, I wish he was here because I don't want to speak for him anyway. But. [01:31:12] Speaker A: Yeah, no, it's hard. I understand. [01:31:13] Speaker B: Yeah, yeah. But I think his point is, like, let's say I'm tapping Morse code. Right? [01:31:18] Speaker A: Mm. [01:31:18] Speaker B: So then, of course, time exists. We all agree on that. But the circuits, the temporal aspects, are constrained by my body learning the sequences within the constraints of the neuron pathways, the muscular, the musculature, etc. And then I can get that in order in a rhythm. And I don't need to know. I'm not measuring. This is 200 milliseconds. 200 milliseconds. But it's constrained by my mesoscopic ability to move through the world, my behavioral brain to behavior output. [01:31:57] Speaker A: But you, in your podcast with Robby gave an example. So I think I know the answer to the confusion, but let me just get there. So you gave the example of one Mississippi, two Mississippi. We count. Right. So we use the body in the podcast. [01:32:15] Speaker B: You listen to the podcast. [01:32:17] Speaker A: Of course I did. And so you have this situation in which we use our motor commands, our motor lip movement, to time something. Okay. But there's no doubt whether who's controlling the lips. I think nobody's arguing that the lips control the brain. I think we all acknowledge that the brain controls the lips, even if there's feedback. But we can control that feedback. We can cut that feedback. And there's still. So. Thank you. I could read your mind where you were going. But we know, because we can. There's many neurological disorders in which that feedback is corrupted and performance drops, but there's no doubt as who's controlling who. So the fact that you have the. So the. So there's no argument there that the brain is generating the dynamics. What I think the problem is, what I think the misconception may be, is what time means to the most of the field. When most of the field says timing, what we mean is essentially dynamics. So. And I think sometimes I think his comment is, is that the brain doesn't have access to time. The brain can't tell time, but clocks can't tell time either. So clocks what is a clock. A clock is just a mechanical system. It's just a dynamical system that changes in time. Time is an abstraction that we use to standardize change. There's a great quote by Ernst Mach, the physicist, that says it's utterly impossible to measure change using time, because time is an abstraction we arrive at by measuring change. So when he says that time is not in the brain, I mean, that's a bit of semantics, but fine. But time is not in a clock either. So time is just an abstraction for change. And when you have a temporal task, what it means is when we say 200 milliseconds or 700 milliseconds, then what you really mean is that as measured by a clock, this amount of time has elapsed, this amount of change has lapped and we need to match that. And sometimes we match that using the body, sometimes we don't. And I can give you an examples of some embodied timing or not, but let me give you. So here, you see, we have an elevator and we have that high technology pole. So, and the elevator is very annoying because you have to flash your card and within something like 905 milliseconds, something I haven't mastered yet, you have to press the button of the floor you want to go to. [01:35:09] Speaker B: Sure. [01:35:10] Speaker A: How do we do that? So what we do is sometimes change the speed of our arm in order to match that. So the brain uses, sometimes the brain uses the body to help us do timing. So there's no doubt about that. I don't think anybody would disagree with that. But, but it's incorrect to imply that the brain is not ultimately controlling the body and doing the dynamics, but the. [01:35:35] Speaker B: Brain is part of the body also. [01:35:37] Speaker A: Yes, fair enough, thank you for that correction. And, but either way, yes, the brain and the rest of the body is controlling that. So like let's say I'm counting and if I count in binary, so 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 15, 16. [01:35:54] Speaker B: Show off. What a show off. [01:35:55] Speaker A: Show off. Sorry, sorry, sorry, sorry. To do that then Is that embodied counting? No, my brain is controlling my body to help me count. Do you want to call that embodied counting? I don't know, I guess you could if you wanted. But let's never get cause and effect changed mixed up there. So it's clearly the body that's tapping in. I'm sorry, the brain that's tapping into the body, even though the body, brain is part of the body. Sorry, that that's controlling that. So, so I, I still struggle to to understand the point. [01:36:30] Speaker B: I think. I mean, could this be a case where you're both right? Is that a. But that's often the case. So thinking to your. You have single neurons, right, in your organotypic slices, and they can tell time, right? They. With an approximation. [01:36:51] Speaker A: Yeah, absolutely, absolutely. And again, I think what Robby would object to, that saying is what they're doing is changing. And so time is a word we use to quantify change. So what my view is, is Robby is using the word time in the way most people don't use it. And that's generating a bit of tension because he thinks that maybe clocks have access to time. But no, nobody has access to time because time is an abstraction that we arrive at by measuring change. In this case, in physics, we're going to avoid getting into the physics for a moment. [01:37:29] Speaker B: Okay? Yeah. Yeah. All right. Yeah. I've done enough speaking for David here, so I won't. Because I can hear him in the back of my head sort of melting about what I said. [01:37:40] Speaker A: Someday I hope I'll get to chat with him. One try to address. [01:37:44] Speaker B: Yeah, but anyway, I mean, my thought was like, all right, a single neuron, right? So it has all these changes, processes, those changes take time. The synaptic vesicles, when you're entraining it, right, you're in. You're in training, in training it. [01:38:02] Speaker A: I don't know. [01:38:03] Speaker B: So I get entrained by my children's music all the time, right? And then later I hear the same stupid song that I heard that I don't enjoy, right? But it gets stuck in my head and it has become entrained in my head absent stimulus. [01:38:20] Speaker A: Okay? [01:38:21] Speaker B: A single neuron, its own. It is a living body, right? An identity. Not an identity, but, you know, an autonomous thing. So you could make the analogy to, well, the human body as a whole. So if I'm timing things as I drum or something, right, I have all these internal processes that are contributing to my eventual behavior. I have a question here. So a single neuron, then flashing the lights on it. It's getting used to recycling the synaptic vesicles in my example at a certain rate, because it needs to. So that entrains the ATP production, etc. And that rhythm happens, and there is no time there. It's literal changes in physical processes happening within the living cell. So there is no time there. And you said, let's avoid physics, but then we have to talk about physics. [01:39:28] Speaker A: What do you mean, there is no time there? So I think that Sentence. [01:39:31] Speaker B: The timing is the change. Right. So I could. I'm pushing. Let's say I'm pushing the limit. So if I'm doing this for 24 hours, I'm probably pushing the limits of the cell. And you guys could have. You stimulated in the sweet spot and you could have stimulated too much. Right. And it wouldn't be able to keep up and it wouldn't be able to learn. Or you could have stimulated too little and it wouldn't. So there's a sweet spot of the stimulation within the regime that's possible of a single cell, and then populations of cells to keep up and then to be entrained. That's not a question, but that's my question. [01:40:03] Speaker A: Okay, so just to clarify, is the first place these phenomena are circuit properties. So they're not single. They're circuit properties that require neural dynamics. Neural circuit dynamics. And it's a bit confusing. Sometimes we use the difference between neuronal dynamics to mean single cell properties and neural dynamics, which now means circuit properties. So this is. [01:40:26] Speaker B: Apologies for that. Yeah. [01:40:28] Speaker A: Neural dynamics. So now what is neural dynamics? Neural dynamics is simply any dynamical system, whether it's a computer or a ball falling down a hill, that's governed by the laws of physics that plays out in time. So I think the word time generates a lot of confusion, and this is my concern in this debate is that some people take time as being something that's out there, and this generates confusion. So when we say it lasts 500 milliseconds. Yes. We say that our stimulation protocol was set up so the time difference between this stimulus and that stimulus will be 500 milliseconds. What is 500 milliseconds? Well, 500 milliseconds is the time it takes a quartz crystal to oscillate 17,000 times. So. But the. So this is. I think this is why these debates are so confusing and because time is such an important concept that we tend to forget that it's not out there in the external world. And we've talked about this in the past, Paul. Time is the most common noun in the English language. [01:41:46] Speaker B: I was gonna ask you that again, because I referenced that and I remember you saying that, or maybe it was in your book and people don't believe me that that's the case. [01:41:56] Speaker A: Well, there is this thing called the Internet. People can look up to double check. [01:42:00] Speaker B: That that's where you find all the facts. [01:42:01] Speaker A: Yeah, it's the common noun. So I don't want to say it's the common word. Hopefully I didn't say the word, I say noun. [01:42:07] Speaker B: I'm careful to say noun when I do it. [01:42:09] Speaker A: I wouldn't say surprise me if I messed that up at one point. But anyways, it's generally stated as the most common noun, so it's a bit ironic to say that, well, it doesn't exist because hopefully it does. But I would push back to say that it does have a purpose. And the purpose is that it really anchors much of our lives and the changes that occurring. And so when I say that the brain, there's no doubt that the brain tells time. What I mean is there's no doubt that the brain has dynamical processes that allow it to synchronize, to anticipate, to match, to decode changes happening in the external world. And in both those cases, in the internal case and the external case, we happen to use the word time to quantify those changes. So I get it that sometimes it causes a certain sense of confusion, but I think a lot, and this goes back to Bergson too, I think he was just using the word time in a way that didn't match how most people use the word time, particularly nowadays. And that's why it's not a particularly fruitful line of inquiry. [01:43:29] Speaker B: When's your next book going to come out? Are you writing a book right now? You're a book writer, you're an author? [01:43:33] Speaker A: I'm a scientist, but I hope to write another book. But I'll keep you posted. You'll be the first to know. [01:43:41] Speaker B: Keep me posted. Okay, last thing. So we started off talking about AI and how time is irrelevant, basically in AI, but of course in robotics, timing is important. I mean, does I need time? [01:43:55] Speaker A: Okay, so I just want to again, avoid confusion. I never said time is irrelevant. And I. What I said is that it's amazing to me how successful Transformers are and that even though their architecture doesn't allow them to tell time. So, and I think this is sort of your point is that obviously in robotic self driving car, time is important. But not only that by the way, is when you use speech processing. So if you ask ChatGPT, but rather than typing in the prompt, you're now using the speech recognition system and you say something like great eyes or gray ties or they gave her cat food versus they gave her cat food where the temporal structure is there. So the ChatGPT can pick up some of those temporal differences, but it's not really the transformer, there's a front end that's doing the speech recognition. So obviously to do speech recognition, well you need to be able to look at the intervals, the temporal structure there. It's more than just the ordinality, but when you type it in, it's primarily just the ordinality. So I'm just backing, I want to make it clear, I'm never saying time is, is not important there. Okay, so your question is, is time important? [01:45:22] Speaker B: Well, the interesting thing is, you know, we use the term dynamics and you can we talk about dynamics of systems that are not biological. Right. So you can talk about the dynamics of a neural, artificial neural network, but then it's not really dynamics, it's you're imposed dynamics. [01:45:41] Speaker A: Simulated dynamics. [01:45:42] Speaker B: Yeah, simulated dynamics. Yeah. So it's just kind of an interesting. Time is so important to everything that we do. And as you say in the transmitter article, you turn off a computer, you turn it back on and it's fine. [01:45:56] Speaker A: Yeah, it's a very different, definite different type of computation. And the answer to your question is AI is doing incredibly well without sort of self driving. Cars obviously need to have temporal structure and have to do speed and stuff. And I think they have dynamics in one way, but it's mostly sort of discrete processing. So the answer is there's many ways to cope with time. There's many strategies to deal with that, including just joining sort of a feed forward network with delays. But I don't know if we're wrapping up as soon. I think we are, we are. But you know, I would maybe to wrap that around and go back to consciousness. So to me, consciousness is a biological process. And I like it. [01:46:53] Speaker B: I mean I'm. I agree. [01:46:55] Speaker A: Okay, now, now, Paul. So. But it's a biological process that's defined by how it evolves in time. So to me consciousness is like music. It only exists in the flow of time or like life itself. Right. You look at life, it doesn't make sense to ask if a frozen organism is alive because life is defined by change, by metabolism, by reproduction, by entropy. Entropy. Thank you. And so forth. So my guess. And that's all it is. This is not a theory. [01:47:31] Speaker B: Oh come on, put theory on it. It'll make it. [01:47:34] Speaker A: I'm not going to do that. So is that the one thing we do know is this consciousness is a biological process that at least the only example of conscious we know and biological processes are things that flow in time. So I think it's helpful to from the get go as the consciousness field attempts to go through all the different channels. Theories here is to try to ground itself on the role of time. So IIT as I understand it is not a process, it's a state theory. That's true. So it doesn't, as I understand it, Iit doesn't really require change or time and that if everything is static, it still be conscious, as I understand it. But again, it's hard to really know because it's not very well defined. There's no other units there. So now then, in the Transmitter article. Yes. I speculate that to me, AIs implemented on standard digital computers which don't have continuous time, which are discrete in terms of their computations. I speculate and that's all it is is speculation that those would not support consciousness because they don't support continuous time. [01:48:58] Speaker B: Pono mono theory, everyone. [01:49:01] Speaker A: Damn it, Paul. [01:49:02] Speaker B: Okay, so Dean, I guess I'm going to thank you very much for coming back on. I guess I'll see you in 2032. [01:49:10] Speaker A: Sounds good, Paul. It's a date. [01:49:11] Speaker B: Oh my God. Thanks, Dean. [01:49:14] Speaker A: Thank you, Paul. It's been a pleasure. [01:49:21] Speaker B: 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 Brain Inspired to learn more. The music you're hearing is Little Wing, performed by Kyle Donovan. Thank you for your support. See you next time.