WEBVTT 00:00:03.497 --> 00:00:10.497 This is the Convergent Science Network podcast. Leading researchers in the domain 00:00:10.497 --> 00:00:16.777 of neuroscience, brain theory and technology are interviewed by Paul Verschure and Tony Prescott. 00:00:17.257 --> 00:00:21.237 Paul Verschure with the Convergent Science Network podcast. 00:00:22.217 --> 00:00:23.857 I'm here with my colleague Tony 00:00:23.857 --> 00:00:28.957 Prescott at our Barcelona Cognition Brain Technology Summer School 2018. 00:00:30.017 --> 00:00:33.677 Where our guest Bjorn Merker, welcome to the podcast, Bjorn. 00:00:35.357 --> 00:00:40.517 Has given a tutorial on the different systems of the brain and how to organize, 00:00:40.797 --> 00:00:44.977 especially from an anatomical perspective, right? 00:00:45.017 --> 00:00:49.977 And then also what possibly the functional consequences of this organization would be. 00:00:50.817 --> 00:00:58.337 So Bjorn, how many systems would you distinguish in the mammalian brain? 00:00:58.337 --> 00:01:02.357 How many subsystems would you consider as being relevant? 00:01:02.757 --> 00:01:08.417 Well, what I did in the tutorial was try to, first of all, raise some problems 00:01:08.417 --> 00:01:11.537 about what we mean by systems and how to define them. 00:01:11.617 --> 00:01:16.277 Because anatomically, I made one example. For example, take midbrain and diencephalon. 00:01:16.497 --> 00:01:21.277 In our anatomical nomenclature, we divide them up into two clear compartments, 00:01:21.337 --> 00:01:23.397 diencephalon and midbrain. brain. 00:01:23.437 --> 00:01:28.557 But if you look, I showed horizontal sections, and it's an absolutely uniform 00:01:28.557 --> 00:01:35.637 griseum, you know, gray matter with various nuclei, and you can't see any distinction between them. 00:01:35.857 --> 00:01:42.577 And then you look at embryology and molecular markers for developmental processes, 00:01:42.777 --> 00:01:49.177 and you see that midbrain and diencephalon in its origin is one unit, 00:01:49.337 --> 00:01:53.297 and it's the unit that receives terminals from the optic tract, 00:01:53.457 --> 00:01:56.277 the midbrain does, and so I call it the optic brain. 00:01:56.697 --> 00:02:02.177 And so that was to try to make a little problem out of having to find a system. 00:02:04.477 --> 00:02:10.137 And anatomically, when I was looking at brains myself, you took the posterior 00:02:10.137 --> 00:02:11.737 commissure, was the dividing line. 00:02:11.897 --> 00:02:14.077 Behind that is the midbrain, in front of that. 00:02:14.297 --> 00:02:19.557 But functionally and in terms of circuitry, you can't see a border between them. 00:02:19.557 --> 00:02:22.857 So I tried to make a little problem out of defining systems. 00:02:23.037 --> 00:02:27.537 And then I said, well, we can use an approach from, for example... 00:02:29.359 --> 00:02:32.459 Systems theory or cellular analysis of cell function 00:02:32.459 --> 00:02:35.479 there you have clear division of labor between the different 00:02:35.479 --> 00:02:38.639 organelles they're each doing a specific thing the 00:02:38.639 --> 00:02:44.959 mitochondria are doing metabolism and the endoplasmic reticulus reticulum is 00:02:44.959 --> 00:02:51.899 doing is doing dna work and goldie apparatus and so each has its function and 00:02:51.899 --> 00:02:56.919 they all work together and when they stop working together the cell is dead, the system is gone. 00:02:57.319 --> 00:03:03.539 So if you look at this in terms of the division of labor, what would you like 00:03:03.539 --> 00:03:08.639 the parts, the system parts of the total system, what would you like them to do? 00:03:08.699 --> 00:03:12.519 Well, you would like them to have rather clear-cut generic functions. 00:03:13.899 --> 00:03:19.819 And if you think about that, you would also expect that if there is a system 00:03:19.819 --> 00:03:24.919 and it's anatomically visible, it would have a pretty redundant structure. 00:03:25.339 --> 00:03:29.299 And the classical example of course is cerebellum which 00:03:29.299 --> 00:03:33.139 has this crystalline structure redundant throughout and you 00:03:33.139 --> 00:03:37.239 know you would be have to be pretty much out of your mind not to think that 00:03:37.239 --> 00:03:41.639 wherever in the cerebellum you are that circuit is doing something similar across 00:03:41.639 --> 00:03:51.479 the whole cerebellum so a generic function and reflected in in a in a clear-cut structure the cortex, 00:03:52.259 --> 00:03:55.239 Neocortex is similarly very uniform in its structure. 00:03:56.159 --> 00:03:59.859 The basal ganglia is also, but they are all very different in their structures. 00:04:00.139 --> 00:04:05.499 So by that kind of reasoning, by showing lots of pictures of these things, 00:04:05.659 --> 00:04:09.499 you can start saying, well, maybe there is a reason to divide this thing up 00:04:09.499 --> 00:04:16.019 into the higher functions, into systems, division of labor between basal ganglia, 00:04:16.079 --> 00:04:19.319 cerebellum, and neocortex, Or actually, cortex as a whole, 00:04:19.479 --> 00:04:24.019 because the hippocampus, for example, is simply the top of the hierarchy, 00:04:24.219 --> 00:04:26.579 in a sense, of the cortical hierarchies. 00:04:27.046 --> 00:04:32.206 Then as you go down, midbrain, you have this clear-cut, very well-defined, 00:04:32.446 --> 00:04:35.806 which I wouldn't hesitate to call the system, which is the colliculus. 00:04:36.246 --> 00:04:39.786 You have the extensions of the basal ganglia down into the midbrain. 00:04:39.966 --> 00:04:44.386 And then, of course, you have the brainstem, which is just this massive collection 00:04:44.386 --> 00:04:49.006 of very, very specific circuitry having to do with all the basic functions, 00:04:49.106 --> 00:04:54.626 from respiration and locomotion to vocal control and so on. 00:04:54.626 --> 00:05:05.166 So once you're up in the higher reaches, I find it easy to give at least a conjectural notion of a system. 00:05:05.606 --> 00:05:09.746 How many systems do you divide the brainstem up into? That's your free choice. 00:05:10.146 --> 00:05:14.906 So I wouldn't hesitate to give an explicit number, but I would say that there 00:05:14.906 --> 00:05:16.486 are very clearly defined systems. 00:05:16.826 --> 00:05:20.526 And the challenge would be two challenges. 00:05:21.406 --> 00:05:26.306 What is their generic function, each one of them? can we define a global function 00:05:26.306 --> 00:05:30.266 for each one of them that is reflected in the circuitry and secondly how do 00:05:30.266 --> 00:05:33.686 they work together and but we could also as a heuristic, 00:05:34.306 --> 00:05:40.466 go to ethology or we could go to psychology and say well systems are defined around motivation, 00:05:42.086 --> 00:05:45.866 attention perception yeah memory right would you would you find it a helpful 00:05:45.866 --> 00:05:49.246 heuristic to Yes, absolutely. It would come in. 00:05:51.466 --> 00:05:57.886 For the brainstem, you would obviously have to deal with motivational systems 00:05:57.886 --> 00:06:01.066 built in at a very basic level, 00:06:01.566 --> 00:06:06.346 survival-related on all kinds of dimensions, and psychology, 00:06:06.806 --> 00:06:13.426 attention, volition, and so on, abstraction, cognition, surely. 00:06:13.426 --> 00:06:18.606 And then now the question comes, how do you interface them with this other way, 00:06:18.746 --> 00:06:20.166 this division of labor notion? 00:06:20.446 --> 00:06:27.566 And much of what psychology looks at, of course, is consciously accessible information. 00:06:27.746 --> 00:06:33.586 You can ask people in psychophysical experiments, or you can do behavioral experiments, and. 00:06:35.046 --> 00:06:40.746 You extract some kind of functional conception from that, 00:06:40.826 --> 00:06:43.906 and so you get a class of your attentional mechanisms play across 00:06:43.906 --> 00:06:47.766 the different modalities so that's probably some system 00:06:47.766 --> 00:06:55.026 that is doing that and so on uh now it would play it could conceivably cut across 00:06:55.026 --> 00:07:03.346 several of these division of labor defining systems attention there is a theory 00:07:03.346 --> 00:07:06.266 of attention that actually involves the basal ganglia isn't there, 00:07:06.946 --> 00:07:10.186 Somebody wrote something about the base of ganglia and attention. 00:07:10.486 --> 00:07:14.386 Now, normally we don't think of it that way, but where did I see that? 00:07:14.566 --> 00:07:16.566 Anyway, absolutely, yeah. 00:07:16.946 --> 00:07:22.186 So, this sort of overall systems decomposition, can we find some. 00:07:24.859 --> 00:07:30.199 General principles that are even above the level of modules of subsystems. 00:07:30.459 --> 00:07:35.099 For example, sort of the Jacksonian principle of layered control. 00:07:35.399 --> 00:07:42.019 Would you agree with that? Is there you have redundancy, or maybe not redundancy, 00:07:42.079 --> 00:07:45.759 but you have similar functionality at different levels of the brain, 00:07:45.879 --> 00:07:48.299 but the differences? Very much appealing. 00:07:49.879 --> 00:07:56.739 I saw one example this afternoon with Giovanni's presentation where the eye-bling 00:07:56.739 --> 00:08:01.719 conditioning with cerebellar involvement, the eye-bling reflex, 00:08:02.099 --> 00:08:07.699 which is the one without the cue, when just an earpuff goes to your eye, that one keeps working. 00:08:08.479 --> 00:08:12.459 That one keeps working. And what you get is you move the beginning of that up. 00:08:13.279 --> 00:08:19.379 So when you saw the detailed videos of the eye closing, eye-bling conditioning, 00:08:19.619 --> 00:08:23.499 when when the animal was conditioned, you see this slower thing, 00:08:23.699 --> 00:08:31.779 which is the anticipatory blinking, and then a real squeezing of the eye, which is the old reflex. 00:08:32.099 --> 00:08:35.739 But here they are both two levels, two Jacksonian levels, if you please, 00:08:35.879 --> 00:08:38.099 both working at the same time in parallel. 00:08:38.479 --> 00:08:42.739 In that case, there is no reason whatsoever to eliminate the lower one. 00:08:42.859 --> 00:08:48.839 Let it run. And that to me, by the way, brings up the very appealing notion 00:08:48.839 --> 00:08:54.219 that sometimes we think of the systems working together that you have to switch 00:08:54.219 --> 00:08:55.659 back and forth between them. 00:08:55.899 --> 00:09:04.479 But the default procedure ought to be that every system does its generic thing 00:09:04.479 --> 00:09:06.739 all the time in parallel. 00:09:06.739 --> 00:09:13.199 And when no other system is sending it input, it idles. 00:09:14.065 --> 00:09:17.825 And when it gets something significant, it does its thing on it and sends it 00:09:17.825 --> 00:09:19.645 back wherever it's going to send it. 00:09:19.865 --> 00:09:22.645 So that means that they are running in parallel all the time, 00:09:22.765 --> 00:09:26.605 and that there would then be some behavioral situations in which you would have 00:09:26.605 --> 00:09:32.125 to turn off actively the lower level because it interferes with something, 00:09:32.285 --> 00:09:34.165 but those would be special cases. 00:09:34.425 --> 00:09:39.565 The default would be multilevel, parallel, running, each one doing their own 00:09:39.565 --> 00:09:45.045 thing, and like there is the old notion that visual function moves from the 00:09:45.045 --> 00:09:46.705 colliculus up to cortex. 00:09:46.865 --> 00:09:52.405 No, a new layer is added with completely new capacities, object vision, 00:09:52.725 --> 00:10:02.665 and the colliculus keeps doing its old orienting thing and actually some salient things and so on. 00:10:03.325 --> 00:10:05.245 So let it do its thing. Why? 00:10:06.005 --> 00:10:09.085 It's perfectly home from billions of years of evolution. 00:10:09.885 --> 00:10:13.965 Let it do its thing and just add new things. And then the new things have to 00:10:13.965 --> 00:10:15.225 be tied into the old ones. 00:10:15.425 --> 00:10:18.565 And they are usually tied in by some kind of fiber projection. 00:10:19.245 --> 00:10:22.165 So absolutely, yeah, air levels thing, I have no problem. 00:10:22.485 --> 00:10:25.345 But now we have a bit of a mess, right? 00:10:25.445 --> 00:10:31.105 Because we start with anatomy, and it looks like, although you already said, 00:10:31.225 --> 00:10:33.585 be careful with drawing borders, because it's not so obvious. 00:10:34.105 --> 00:10:37.725 But we could think about, okay, but there are structural divisions, right? 00:10:37.845 --> 00:10:42.805 I have a brain and brain is my midbrain, forebrain. but now if we take this 00:10:42.805 --> 00:10:47.945 Jacksonian view together with this more ethological psychological functional perspective, 00:10:48.705 --> 00:10:53.345 it starts to look like at each of these structural systems. 00:10:54.285 --> 00:10:59.685 All these function elements are actually represented in some form but as I move 00:10:59.685 --> 00:11:01.945 forward along the neorexis. 00:11:03.305 --> 00:11:07.265 It is starting to become more dependent on let's say on memory it starts to 00:11:07.265 --> 00:11:13.765 become more dependent on learning properties of the control of these different 00:11:13.765 --> 00:11:14.665 psychological functions. 00:11:15.465 --> 00:11:21.385 So how should we now square that circle? How do we get the structure and the function aligned? 00:11:21.685 --> 00:11:24.645 Because the standard heuristic was always, well, function follows structure. 00:11:24.985 --> 00:11:28.245 If you understand the structure, you have to head around the function. 00:11:28.805 --> 00:11:30.745 But maybe it's not that straightforward. 00:11:31.965 --> 00:11:37.345 In other words, your question is, at each of these levels, even at the most 00:11:37.345 --> 00:11:43.205 primitive, is there an instantiation at that level of each of the fundamental things? 00:11:43.345 --> 00:11:46.985 And then you replicate them, but they get more and more sophisticated and more 00:11:46.985 --> 00:11:51.165 and more dependent on advanced kind of operations. 00:11:52.205 --> 00:12:00.665 And naturally, in a lamprey, you know, you have, yeah, but yeah, that's, 00:12:01.777 --> 00:12:06.637 Maybe we don't have to go that far because what Grillner is showing is that 00:12:06.637 --> 00:12:14.437 lamprey in this, you know, compared to our brain or a mouse brain is quite primitive. 00:12:14.557 --> 00:12:18.617 But it has a basal ganglia and it's divided up pretty much. 00:12:18.637 --> 00:12:24.257 It sends its abendular for avoidance, projection and so on. And when he takes 00:12:24.257 --> 00:12:31.757 a lamprey and legions the basal ganglia, they get tardive dyskinesia. 00:12:31.837 --> 00:12:37.397 They suck on to rocks and they won't let go, like a Parkinson's patient who 00:12:37.397 --> 00:12:38.917 can't get going. Now they get stuck. 00:12:41.537 --> 00:12:46.537 So even a lamprey has, of course, a telencephalon. on. 00:12:47.257 --> 00:12:55.857 So there is presumably never a stage at which only a brainstem works by itself. 00:12:56.597 --> 00:13:00.477 The brainstem itself is a bit hierarchical. 00:13:01.037 --> 00:13:04.777 And I think the big idea... Perhaps a developmental state. Yeah. 00:13:05.677 --> 00:13:17.097 Something like the Monodelchus domesticus opossum, where the infants are born extremely juvenile, 00:13:17.857 --> 00:13:23.797 essentially just of brain stem and what they can do at that age is just cling 00:13:23.797 --> 00:13:31.257 to the mother and find and they would be very different from the beginning. 00:13:31.257 --> 00:13:35.057 I think they're born around two weeks after gestation. 00:13:35.097 --> 00:13:41.457 And you can watch the other parts of the brain begin to kick in. 00:13:41.557 --> 00:13:48.057 And I think you can probably see some similar things in rats in development. 00:13:48.657 --> 00:13:56.557 There is a mutant mouse where there's a... 00:13:57.826 --> 00:14:01.606 A gene which controls dopamine and 00:14:01.606 --> 00:14:04.446 if this this this gene 00:14:04.446 --> 00:14:07.486 is absent then dopamine cells don't develop 00:14:07.486 --> 00:14:14.526 properly and those mice are finally their first form but by by two weeks so 00:14:14.526 --> 00:14:21.306 the dopamine system and the basal ganglia presumably which you know which has 00:14:21.306 --> 00:14:26.906 a veto on any kind of voluntary movement if once that comes in if there's no dopamine it won't work. 00:14:27.826 --> 00:14:31.206 To that they will be they can generate behavior deficits yeah 00:14:31.206 --> 00:14:34.286 that stage but not their own yeah which means that 00:14:34.286 --> 00:14:37.786 in a sense that system comes it comes in a bit late yeah 00:14:37.786 --> 00:14:41.086 so maybe in a guinea pig which which 00:14:41.086 --> 00:14:44.706 is born and runs around uh with incredible 00:14:44.706 --> 00:14:49.886 competence because i was said to watch a guinea guinea pig birth the woman was 00:14:49.886 --> 00:14:54.566 away and i had them in a cage and i didn't realize that the newborns could get 00:14:54.566 --> 00:15:02.186 through the meshes of the cage within one hour of birth, 00:15:02.726 --> 00:15:07.186 they were running around the place and hiding in places, and I had to chase 00:15:07.186 --> 00:15:12.526 them down, and it was just as bad as chasing a full-grown rat. 00:15:12.786 --> 00:15:14.906 So they would be very competent. 00:15:15.206 --> 00:15:20.566 So there may be developmental schedules that keep these things in for precocity 00:15:20.566 --> 00:15:23.786 versus altriciality. Yeah. 00:15:24.186 --> 00:15:29.226 Interesting. And it's interesting in humans that we are precocial in terms of 00:15:29.226 --> 00:15:34.806 our sensory development before with our eyes open, but our motor systems are really altricial. 00:15:34.886 --> 00:15:37.686 So really altricial. What is interesting in your mouse example, 00:15:37.686 --> 00:15:44.526 Tony, maybe that's a riddle also you're going to solve, if you sort of perturb 00:15:44.526 --> 00:15:48.786 the neural mental trajectory, you get severe deficits, right? 00:15:48.886 --> 00:15:52.606 But if you take an adult animal and you would lesion the forebrain. 00:15:53.514 --> 00:15:58.494 Then often from the outside, they're pretty much functional, right? 00:15:58.574 --> 00:16:02.834 If they live in a cage, they can take care of themselves, they can take care of their young, right? 00:16:02.874 --> 00:16:06.314 So it's interesting to see that if you disrupt the developmental pathway, 00:16:06.414 --> 00:16:11.494 the system apparently is really going towards some attractor that it cannot 00:16:11.494 --> 00:16:13.894 reach and it collapses, it's pathological, dies. 00:16:14.734 --> 00:16:19.194 Well, it wasn't reached that developmental stage, and then you just remove it 00:16:19.194 --> 00:16:23.674 discreetly, it has to fall back on these earlier layers in its architecture. 00:16:25.054 --> 00:16:34.434 So why is this bootstrap system to me so sensitive to perturbation as the adult 00:16:34.434 --> 00:16:38.774 system is actually very robust to massive lesions? How would you explain that? 00:16:39.054 --> 00:16:43.214 Well, I thought sometimes about the possibility that there are certain structures 00:16:43.214 --> 00:16:47.634 whose essential function is to educate the system early on. 00:16:47.634 --> 00:16:54.434 And once I have done their work the deficits will be less and that's approximately. 00:16:56.614 --> 00:17:03.474 Rephrasing what you're saying now I have thought about whether the cerebellum 00:17:03.474 --> 00:17:05.354 is like that that it's real, 00:17:06.894 --> 00:17:13.354 use is very early of course later on we know that you have to recalibrate your BOR when your, 00:17:14.554 --> 00:17:19.594 eyeballs grow and so on So there are always adjustments that are needed for 00:17:19.594 --> 00:17:22.214 fine-tuning during adult, 00:17:22.534 --> 00:17:32.394 so it has a function then also, but maybe it's real work that's being done early. Exactly. 00:17:33.414 --> 00:17:38.154 So, okay, so we haven't really solved the problem yet, right? 00:17:38.214 --> 00:17:43.094 Because we have a structure, we have a brain structure, there is heterogeneity, little uniform. 00:17:43.094 --> 00:17:49.874 Form but we cannot really draw borders very clearly necessarily right it's not clearly modular. 00:17:51.274 --> 00:17:55.514 But we cannot necessarily map it in some clean way to the to different functions 00:17:55.514 --> 00:17:58.754 you might want to see either and the possibility might be that all functions 00:17:58.754 --> 00:18:04.154 exist at multiple stages of the system but still when we when we progress along 00:18:04.154 --> 00:18:07.194 anorexias and we go to the telencephalum the forebrain, 00:18:08.566 --> 00:18:12.706 There are distinct features that you learn. Would it be fair to say that actually 00:18:12.706 --> 00:18:16.406 the key thing that's happening is just we're building more advanced memory systems. 00:18:16.586 --> 00:18:20.646 As we move forward, we build more advanced memory systems, and we have to add 00:18:20.646 --> 00:18:22.586 features to just control these memory systems. 00:18:22.866 --> 00:18:25.486 Would that be enough? I think that's absolutely right. Anyway, 00:18:25.486 --> 00:18:32.626 I said in my tutorial that as soon as you see a structure in the brain that 00:18:32.626 --> 00:18:36.766 has big volume with millions and millions and maybe billions of neurons, 00:18:37.106 --> 00:18:40.686 you should think learning structure, memory structure, something that stores. 00:18:41.046 --> 00:18:48.266 Because I have this heretical view that in every learning system there is this 00:18:48.266 --> 00:18:53.326 tradeoff between stability and plasticity. 00:18:55.546 --> 00:19:00.246 So, if you make the synaptic weight change, it's very rigid, 00:19:00.386 --> 00:19:03.866 you're going to not be able to use those for anything else. 00:19:04.646 --> 00:19:16.846 Well, that has been solved in neural network context by adding new units to the network. work. 00:19:17.526 --> 00:19:25.086 And my alternative to that is you start out with a huge information storage 00:19:25.086 --> 00:19:29.106 capacity and you use it up sequentially as you learn. 00:19:29.406 --> 00:19:35.206 So there would be some kind of a learning front going through it, using up the tissue. 00:19:35.346 --> 00:19:41.446 It would be equivalent to the ovaries stocked with all the eggs a woman is going 00:19:41.446 --> 00:19:45.906 to produce in her entire lifetime, and they're successfully released. 00:19:47.486 --> 00:19:51.326 Here you would start with your full learning capacity and use it up, 00:19:51.386 --> 00:19:57.966 so it would sort of burn up as you learn and as you commit more and more synapses 00:19:57.966 --> 00:20:00.906 in various ways to specific content. 00:20:04.101 --> 00:20:11.661 When you see, so I have a chapter in my 2004 paper in Cortex, 00:20:11.761 --> 00:20:16.441 a long introduction about the volumetrics of information storage. 00:20:16.781 --> 00:20:21.141 And the best evidence is in bird vocal learning, where you have this very, 00:20:21.181 --> 00:20:25.141 very clean correlation between the size of these nuclei and the learning capacity, 00:20:25.321 --> 00:20:27.541 the complexity of the song that they will acquire. required. 00:20:27.681 --> 00:20:32.421 And this goes within species for individual with different capacities of different 00:20:32.421 --> 00:20:34.081 size, and it goes across species. 00:20:34.661 --> 00:20:38.901 Species with very complex repertoires have big nuclei and a big system, 00:20:39.181 --> 00:20:40.641 and the other ones have smaller. 00:20:40.861 --> 00:20:44.941 So if you extend that to all learning, you would expect... 00:20:44.941 --> 00:20:48.581 So if you extend that to all learning, you would say, looking at the brain, 00:20:48.741 --> 00:20:51.361 just like that, you would say, where are the learning structures, 00:20:51.721 --> 00:20:56.521 cerebellum, basal ganglia, neocortex? Those are the three voluminous, big chunks. 00:20:56.861 --> 00:21:00.581 And the nice thing in evolution is that they go hand in hand. 00:21:00.801 --> 00:21:04.721 They are sharply, the regression of volume between cerebellum, 00:21:04.721 --> 00:21:08.321 neocortex, and basal ganglia should raise a sharp correlation. 00:21:08.741 --> 00:21:12.221 A bigger cortex requires bigger basal ganglia, bigger cerebellum. 00:21:12.281 --> 00:21:15.761 Well, the colliculus stays pretty much the same, you know? 00:21:15.961 --> 00:21:21.841 So if you assume that they are storing information continually through a lifetime, 00:21:21.981 --> 00:21:26.741 and for everything thing that you have to adjust to and using up their storage 00:21:26.741 --> 00:21:30.541 group, but they have to start out big and sequentially use it up. 00:21:30.801 --> 00:21:35.661 But in the story, you didn't include hippocampus, why is that? Dr. No, it's on the top. 00:21:35.901 --> 00:21:40.761 I include that with the neocortex, not neocortex, but with cortex. 00:21:41.321 --> 00:21:46.301 In my tutorial yesterday, I just said cortex general, and I simply put the hippocampus 00:21:46.301 --> 00:21:52.061 at the top of the hierarchy. When you have an alternative learning system that 00:21:52.061 --> 00:21:54.581 accounts for the rest of the cortex. Yeah, it's specialized. 00:21:55.041 --> 00:22:02.741 One of the reasons is it's at the top of the hierarchy and it actually converts, 00:22:03.041 --> 00:22:06.461 it's very elegant, it converts the. 00:22:10.208 --> 00:22:16.488 Supragranularly dominated feed-forward path to the infragranularly dominated feedback path. 00:22:16.708 --> 00:22:22.648 So from entorhinal up into the hippocampus, you get the feed-forward path and 00:22:22.648 --> 00:22:26.348 it comes back in the deep path, in the feedback path. 00:22:26.688 --> 00:22:31.848 So there's this nice hinge, but of course, the hippocampus preceded neocortex 00:22:31.848 --> 00:22:35.388 by millions of years, by a long evolutionary history. 00:22:35.508 --> 00:22:40.048 So it had a separate function long before it became the hinge of the cortical 00:22:40.048 --> 00:22:42.888 counter-curve, which I… Are you sure about that? 00:22:43.108 --> 00:22:47.828 I mean, I think… What are you saying about the illusion of hippocampus? 00:22:47.968 --> 00:22:56.868 It preceded that only mammals have neocortex, and a lot of vertebrates have… 00:22:56.868 --> 00:23:01.488 The medial pallium is some hippocampal catalogue. 00:23:01.848 --> 00:23:06.588 It's the unlogger of the hippocampus. It grows is big only in the neocortex 00:23:06.588 --> 00:23:09.988 because it serves it in very intimate ways. 00:23:10.408 --> 00:23:19.528 But there is hippocampus and an unlogged amygdala prior to man. 00:23:19.728 --> 00:23:23.968 Well, there is a sort of three-layer sort of cortex in reptiles, 00:23:24.128 --> 00:23:29.488 and the hippocampus is more sort of, I guess, reptilian. 00:23:29.748 --> 00:23:37.948 Yeah, so it's really cortex, so it's a three-layer, and then you get this mammalian 00:23:37.948 --> 00:23:43.268 invention of five-layer cortex, which birds don't have, though their forebrain expands vastly. 00:23:44.568 --> 00:23:47.028 And they all have a hippocampus. 00:23:47.968 --> 00:23:53.128 In fact, hippocampus is one of the very good examples in birds, 00:23:53.428 --> 00:24:01.228 not only vocal learning, but these food caching jays and scrub jays and stuff 00:24:01.228 --> 00:24:08.768 somebody up in Siberia they've studied them they hide seeds in tens of thousands 00:24:08.768 --> 00:24:10.568 of different places one animal, 00:24:11.672 --> 00:24:15.472 tens of thousands of places. They remember, they know where they are. 00:24:15.732 --> 00:24:20.712 And the ones who have different species, the ones that hide in fewer places 00:24:20.712 --> 00:24:22.112 have smaller hippocampus. 00:24:22.212 --> 00:24:26.572 The ones that hide in more thousands and thousands of places have bigger hippocampus. 00:24:26.612 --> 00:24:30.372 Very nice volumetrics. Otherwise, the animals are pretty similar. 00:24:31.672 --> 00:24:35.512 But there's something interesting about that, Bjorn, also relative to the architecture 00:24:35.512 --> 00:24:39.652 of the neocortex, because you indeed see then the hippocampal hinge. 00:24:39.652 --> 00:24:46.192 This is where feed-forward switches back to feedback, but there's another pathway 00:24:46.192 --> 00:24:50.672 like that which runs over frontal cortex, which is receiving more attention in some sense. 00:24:50.732 --> 00:24:55.452 You say, okay, I'm at forward to frontal cortex, and then the feedback projections come back as well. 00:24:55.692 --> 00:24:59.972 So now I have two orthogonal, if you want, hinges in some sense. 00:25:00.092 --> 00:25:06.552 So how should I think about that? They are enmeshed. If you look at the connectivity 00:25:06.552 --> 00:25:09.392 instead of where they are physically, 00:25:10.232 --> 00:25:18.312 and you look back at Malcolm Young's beautiful graph theory plot from the macaque, Kokomak, in 1995. 00:25:20.192 --> 00:25:29.652 He shows there is this beautiful visual part of the graph, one big thing hanging out to the left. 00:25:29.652 --> 00:25:33.852 There is the somatosensory in the middle and the auditory coming off on the right. 00:25:33.972 --> 00:25:36.172 And then there is the gustatory. 00:25:36.452 --> 00:25:40.692 And they all converge in what he calls the frontal limbic cap. 00:25:43.639 --> 00:25:52.899 And when you look at the cortical areas, they are completely enmeshed with reciprocal connections. 00:25:53.319 --> 00:25:58.599 The hippocampus sits in the middle of the frontal things like a spider in a web. 00:25:58.799 --> 00:26:04.759 When I saw that, I said, now I understand finally. They are all the highest level of the system. 00:26:04.899 --> 00:26:08.359 So that hinge is a broader hinge than hippocampus. 00:26:08.439 --> 00:26:13.539 It's enmeshed. And when simply looking at this graph, you see the densest number 00:26:13.539 --> 00:26:15.899 of connections are up there. They're all crowded together. 00:26:16.479 --> 00:26:20.219 Limbic frontal and hippocampal is all meshed together. 00:26:20.379 --> 00:26:26.959 And then the modality-specific streams, the hierarchies, are hanging off it. 00:26:27.319 --> 00:26:33.759 So that's the head of the system. And the hippocampus is connectively very close 00:26:33.759 --> 00:26:38.239 to frontal area, much more distant to visual, for example, than frontal. 00:26:38.239 --> 00:26:41.439 So it's the next-known neighbor to those guys. 00:26:41.579 --> 00:26:45.119 So if you just collapse, if you just shrink the fiber lengths. 00:26:45.899 --> 00:26:51.679 So sometimes the physical positioning of an area can be very deceptive. 00:26:51.759 --> 00:26:56.679 Here you have at the tip of the temporal, middle of the temporal lobe is one 00:26:56.679 --> 00:27:01.759 of them, and the other one is in frontal cortex, and they are connected just like that. 00:27:02.219 --> 00:27:07.279 And the other deceptive one is the frontal eye fields. from the light fields 00:27:07.279 --> 00:27:12.379 is placed frontally but connectively in the young diagram, it's in the visual system. 00:27:12.479 --> 00:27:18.519 It's much farther back because his connections, of course, are like rubber bands. 00:27:18.759 --> 00:27:24.779 So they pull and push till they settle to an equilibrium in his modeling of this. 00:27:25.356 --> 00:27:30.316 So there you see that really frontal life films is high in the visual dorsal stream. 00:27:30.436 --> 00:27:34.356 Right. But now more recently, there was another proposal by David Van Essen 00:27:34.356 --> 00:27:38.536 and Henry Kennedy where they spoke of this sort of bowtie structure. 00:27:38.816 --> 00:27:42.816 I think in classical architecture, it means there's a core at the center of 00:27:42.816 --> 00:27:44.776 the bowtie that is very densely interconnected. 00:27:45.236 --> 00:27:49.896 And then there are these two wings of the bowtie that are feeding into it from 00:27:49.896 --> 00:27:51.916 the periphery or have received inputs from it. 00:27:51.976 --> 00:27:55.716 And that will give you the bowtie structure. Is it that your mind is still compatible 00:27:55.716 --> 00:28:00.176 with the older Malcolm Young picture, or is there any change? 00:28:00.276 --> 00:28:01.676 Did anything change in that picture? 00:28:01.916 --> 00:28:09.636 What has changed is that at the time of Malcolm Young, they interpreted it as a small world system. 00:28:09.956 --> 00:28:13.696 And it was by the evidence they had then. What has happened since, 00:28:13.796 --> 00:28:16.636 which was updated in this Vanessa, I love that paper. 00:28:16.756 --> 00:28:19.916 I mean, it's a tremendous synthesis they had done there. and what 00:28:19.916 --> 00:28:23.436 they they found so many more connections that hadn't 00:28:23.436 --> 00:28:26.216 been reported back then and when you add those 00:28:26.216 --> 00:28:29.356 connections it's it's it's more connected than 00:28:29.356 --> 00:28:32.236 a small world network should be and that's 00:28:32.236 --> 00:28:36.476 how they got the bow tie structure and it doesn't conflict with the old view 00:28:36.476 --> 00:28:41.616 it's simply another version of having an efficient network and it's to have 00:28:41.616 --> 00:28:48.156 have this sort of super hub of in the middle and having sensory things feed 00:28:48.156 --> 00:28:51.496 into it and things where it projects coming out. 00:28:51.596 --> 00:28:55.556 But if you look at their diagram, their sort of. 00:28:58.259 --> 00:29:03.699 Conceptual sketch of the boat type you see one double huge arrow you're going 00:29:03.699 --> 00:29:08.379 from the feed forward and the feedback is intact that whole and that's the thing 00:29:08.379 --> 00:29:09.819 that's essential about about, 00:29:10.579 --> 00:29:16.239 malcolm young and all the old versions of this is that feed forward and feedback are streaming. 00:29:17.099 --> 00:29:24.259 Against each other all the time that's sort of the basic right and and but now 00:29:24.259 --> 00:29:28.239 what you also did yesterday which i found very interesting uh so so we have 00:29:28.239 --> 00:29:30.099 now this outline of of the anatomy. 00:29:30.739 --> 00:29:33.559 We have a feeling maybe for for how this mapping to 00:29:33.559 --> 00:29:37.959 function but you went a step further we're really formulating at 00:29:37.959 --> 00:29:45.219 least as a hypothesis distinct functions of them sort of major subdivisions 00:29:45.219 --> 00:29:49.879 of the architecture right yeah so could you could you step us through those 00:29:49.879 --> 00:29:56.279 and I only did my guess work on, 00:29:56.399 --> 00:29:59.519 because it is, as far as I can tell, 00:29:59.679 --> 00:30:06.299 there is no consensus on what these things do in the sense of how to describe 00:30:06.299 --> 00:30:09.819 them at the very abstract generic functional level. 00:30:09.819 --> 00:30:21.339 For neocortex, my guess is that it's doing veridical source reconstruction across all afferents. 00:30:21.959 --> 00:30:24.119 You should remember things like 00:30:24.119 --> 00:30:28.999 the cellular system has a cortical representation. Everything is up there. 00:30:29.619 --> 00:30:36.659 It wants to make sure it has access, rather direct access, to all the basic 00:30:36.659 --> 00:30:41.599 sensory systems, including the stimulus, which is quite spectacular, actually. 00:30:42.179 --> 00:30:49.139 So what do I mean by virilical source reconstruction across all afferents? 00:30:49.359 --> 00:30:54.099 Its task is to tell us in a sense what reality is, and to do that, 00:30:54.119 --> 00:30:58.739 it needs help from the base of ganglia, mostly in terms of how to deal with 00:30:58.739 --> 00:31:02.999 the world, and from the cerebellum in terms of maybe. 00:31:04.119 --> 00:31:10.859 Decorrelation, maybe something else, but there are these very strong anatomical 00:31:10.859 --> 00:31:14.759 connections from cortex into cerebellum, from cerebellum back, 00:31:14.999 --> 00:31:16.719 same thing with basal ganglia. 00:31:18.099 --> 00:31:23.099 So, and of course, basal ganglia is serving much more than motor cortex, 00:31:23.299 --> 00:31:28.699 it's serving vast areas of the cortical mantle or projecting into the basal 00:31:28.699 --> 00:31:33.939 ganglia and then and they get down through the funnel and shit back through 00:31:33.939 --> 00:31:35.839 the thalamus up to cortex. 00:31:37.599 --> 00:31:39.379 So division of labor... 00:31:40.588 --> 00:31:47.588 Source reconstruction means because sensory afference is not only noisy, 00:31:47.688 --> 00:31:53.468 but loaded with so much ambiguity in terms of ill-posed and inverse problems. 00:31:53.848 --> 00:31:58.928 In vision, for example, there is a whole catalogue. Computer vision has been 00:31:58.928 --> 00:32:04.028 a catalogue of working through all the inverse problems that you have to solve 00:32:04.028 --> 00:32:06.028 to make sense of what's on the retina. 00:32:06.028 --> 00:32:10.888 If you have a circle on the retina, it could be an ellipse at the tilt, 00:32:11.048 --> 00:32:15.808 it could be a close ellipse at the tilt, or a small circle close to you, 00:32:15.828 --> 00:32:18.788 or a big circle far away from you. It could be any number of things. 00:32:18.988 --> 00:32:23.668 It could be an infinity of stimuli that generate the same retinal image. 00:32:24.048 --> 00:32:28.688 So how do you tell? Well, in a monocular view, you can't tell. 00:32:29.428 --> 00:32:34.948 You need some additional cues. fuse. So you open two eyes or you move your head 00:32:34.948 --> 00:32:42.188 and suddenly you see, well, it can't be far away because parallax is little and parallax is much. 00:32:42.388 --> 00:32:45.808 There's lots of parallax when I move my head, so it must be close. 00:32:46.048 --> 00:32:47.568 And so you disambiguate. 00:32:47.948 --> 00:32:50.768 And then you take the help of other systems. 00:32:51.508 --> 00:32:54.448 If you're really unsure, you might try to touch 00:32:54.448 --> 00:32:58.248 the thing and see what what's actually there uh not 00:32:58.248 --> 00:33:01.728 not as an adult but early in development um 00:33:01.728 --> 00:33:04.688 so so it has 00:33:04.688 --> 00:33:11.868 to reconstruct the the my my bet is that cortex is there in order to give us 00:33:11.868 --> 00:33:18.828 a realistic model of reality it has to reconstruct from the senses the auditory 00:33:18.828 --> 00:33:23.808 system of course forget about it you know All this noise is binging on the cochlea, 00:33:23.808 --> 00:33:26.688 and the first thing it does is a huge Fourier transform. 00:33:27.288 --> 00:33:32.568 So it has to reconstruct what's out there from this very indirect, 00:33:32.908 --> 00:33:39.008 very noisy, very ill-posed and inverse problem riddle sensory afference. 00:33:39.168 --> 00:33:40.428 And how does it come to the brain? 00:33:40.748 --> 00:33:45.088 As chattering of action potentials in millions of fibers. 00:33:45.948 --> 00:33:50.488 So it's a huge task. and, of course, the lower levels that we talked about, 00:33:50.628 --> 00:33:54.068 they solve these things by simply throwing away information. 00:33:55.248 --> 00:34:01.608 The colliculus doesn't reconstruct objects. It just knows whereabouts they are 00:34:01.608 --> 00:34:04.528 when they move and when they shine brightly over there. 00:34:04.728 --> 00:34:08.568 That's all it cares about. It doesn't have to reconstruct anything that way. 00:34:09.668 --> 00:34:11.308 Location in space is enough for it. 00:34:11.853 --> 00:34:16.033 So once it has that, it can direct the eyes there. Its job is done. 00:34:16.353 --> 00:34:22.013 The cortex wants to know what is the object like, and therefore it is confronted 00:34:22.013 --> 00:34:28.613 with all these ambiguities and inverse problems. 00:34:28.873 --> 00:34:33.273 So it has a huge job. And for me, the hierarchy, the hierarchical organization 00:34:33.273 --> 00:34:39.173 of cortex with feedback and feedforward is to extract priors from its experience 00:34:39.173 --> 00:34:42.753 with the world to make the next experience clearer, 00:34:43.613 --> 00:34:44.873 to clarify the view. 00:34:45.093 --> 00:34:54.253 And so our perception of the world goes from, it's sort of an irreversible gain in clarity. 00:34:54.433 --> 00:34:57.653 We see the world clearer and clearer from infancy and up. 00:34:59.073 --> 00:35:04.733 And we understand more and more about it. By our age, the perceptual lessons 00:35:04.733 --> 00:35:07.513 are long gone. They have been handled long ago. 00:35:07.633 --> 00:35:11.353 We just go about the world. That's the most self-evident thing. 00:35:11.693 --> 00:35:17.313 So it's so easy to overlook how difficult those problems are. Size constancy. 00:35:17.453 --> 00:35:22.553 The thing that's small in the retina can be just as big as this if it's far 00:35:22.553 --> 00:35:24.193 away. That has to be acquired. 00:35:24.433 --> 00:35:28.353 And it's acquired first for the proximal space and then extends out. 00:35:28.453 --> 00:35:33.333 Size constancy, the progress of size constancy is simply extending the constancy 00:35:33.333 --> 00:35:34.373 farther and farther out. 00:35:34.373 --> 00:35:40.653 So some people living evidently in rainforests with dense foliage all around 00:35:40.653 --> 00:35:45.953 them, with never open spaces, they have deficient, distant fact size constants. 00:35:46.173 --> 00:35:52.753 They have very good clothes. But if you take them out from the rainforest to 00:35:52.753 --> 00:35:58.213 the plains and they see a gazelle, they've never seen such a small animal. They're surprised. 00:35:59.633 --> 00:36:03.033 I was making this one up. No, sure, but how does that scale? 00:36:03.153 --> 00:36:07.773 Because, okay, I can imagine that this is a really good summary if I will be 00:36:07.773 --> 00:36:13.113 just a passive observer of the world and I want to have accuracy about this world. 00:36:14.239 --> 00:36:18.519 But in some sense, the brain wants to act. Yes. Otherwise, you're toast, right? 00:36:18.639 --> 00:36:27.179 We want that. So how does this model now of neocortex map into the ability of 00:36:27.179 --> 00:36:31.519 making a decision and matching your goals and building strategies? 00:36:31.539 --> 00:36:34.079 So how does it generalize to this 00:36:34.079 --> 00:36:38.619 more active, goal-oriented part of the system? Tony and others come in. 00:36:38.899 --> 00:36:41.999 But you have to answer the question. Frontal, frontal, basal, 00:36:42.019 --> 00:36:44.219 ganglia, circuitry, going back to cortex. 00:36:44.339 --> 00:36:51.099 So first of all, the desirability of an accurate view of accuracy in the reconstruction, 00:36:51.339 --> 00:36:54.219 the vertical, the vertical part of it is a big thing. 00:36:54.579 --> 00:37:00.699 And cortex, instead of like colliculus or other subcortical things throwing 00:37:00.699 --> 00:37:04.739 information away and saying, I don't care, I just want a fast solution, 00:37:04.899 --> 00:37:09.019 they get instant certainty at the price of permanent ignorance. 00:37:09.359 --> 00:37:15.379 Okay? Okay, so if you really want to know what's out there, you've got to be accurate. 00:37:15.559 --> 00:37:18.819 So cortex steps back and says, I'm neutral. 00:37:19.499 --> 00:37:26.419 I'm not making any commitments initially. I simply make it on a probabilistic basis. 00:37:27.759 --> 00:37:33.159 It's probable that this is a slanted line oriented 45 degrees to the right. 00:37:33.359 --> 00:37:37.659 Okay, it's a probability. So a tuning curve for orientation in the visual cortex 00:37:37.659 --> 00:37:40.239 is like a probability density distribution. 00:37:40.579 --> 00:37:45.859 So the cortex keeps doing that and then it starts merging in the hierarchy with 00:37:45.859 --> 00:37:50.459 feedforward and feedback, the incoming information with the priors, 00:37:50.539 --> 00:37:54.539 and at some point it has to collapse these things to an estimate. 00:37:54.719 --> 00:37:59.179 And that's when we get the best estimate we can get of what's out there. 00:37:59.899 --> 00:38:02.739 But we have to act like you say what do we do about it we're 00:38:02.739 --> 00:38:06.899 not just some big eye viewing the world contemplatively 00:38:06.899 --> 00:38:12.879 gazing at our navel so how do we get about it then you have what you said goals 00:38:12.879 --> 00:38:19.379 motivation we have to have we have to accomplish the things we have to accomplish 00:38:19.379 --> 00:38:26.459 we are we are basically a metabolic engine that converts food into energy and in dividing up that 00:38:26.579 --> 00:38:30.979 for all the tasks that need to be accomplished, we have trade-offs. 00:38:31.079 --> 00:38:33.979 So you can spend all your time searching for a mate. 00:38:34.019 --> 00:38:38.679 If you don't eat along the way and allocate enough time for eating, 00:38:38.899 --> 00:38:41.819 you will starve to death before you find a mate. 00:38:41.959 --> 00:38:46.559 So trade-offs, protecting your body, finding food, finding shelter, 00:38:46.839 --> 00:38:50.999 finding the right environmental circumstances, finding the right social environment. 00:38:51.979 --> 00:38:57.399 So trade-offs constantly, dividing up the energy to different tasks, 00:38:57.579 --> 00:39:01.679 and evolution has taken care of a lot of that, saying there are some fundamentals 00:39:01.679 --> 00:39:04.199 that you have to accomplish. 00:39:04.679 --> 00:39:08.579 You have to eat, you have to protect your body integrity, pain system, 00:39:08.779 --> 00:39:15.959 you have social motives, you have fear, protect yourself from danger, 00:39:16.119 --> 00:39:17.699 and you have curiosity to explore. 00:39:17.699 --> 00:39:19.939 Yeah, but I'm not sure you're solving the problem. 00:39:20.199 --> 00:39:24.979 Well, you're putting the solution to that part of the problem goes back into 00:39:24.979 --> 00:39:28.079 the brainstem, areas like the hypothalamus. Yeah, yeah. 00:39:28.979 --> 00:39:37.099 But then there's also a sort of led architecture of motivational systems that 00:39:37.099 --> 00:39:38.859 you would then trace up into cortex. 00:39:39.219 --> 00:39:44.199 Yep. And how does that link to areas like the hypothalamus? I would say that 00:39:44.199 --> 00:39:47.499 it's just like you say. They are up there. 00:39:47.739 --> 00:39:53.879 The larynx system is the motivational layer, in a sense, the motivational compartment 00:39:53.879 --> 00:39:59.139 of the cortex for some of the motivational systems. 00:40:00.119 --> 00:40:04.979 Not only that, but the frontal cortex, the orbitofrontal system is shot through 00:40:04.979 --> 00:40:13.439 with motivational things like social and so on and it's all gonna now the problem 00:40:13.439 --> 00:40:15.219 about cortex and action is, 00:40:15.819 --> 00:40:18.679 that the cortex is doing this big sort of 00:40:18.679 --> 00:40:27.079 objective reconstruction of the world but it's doing it in parallel and action 00:40:27.079 --> 00:40:32.699 goes serially one action at the time essentially and how do you convert this 00:40:32.699 --> 00:40:36.379 huge parallel display of of incredible 00:40:36.579 --> 00:40:40.659 amounts of information, including motivational information, in frontal and limbic, 00:40:40.759 --> 00:40:43.379 how do you get it down to behavior? 00:40:43.579 --> 00:40:49.079 You have to have something that converts that, in a. 00:40:52.535 --> 00:40:59.815 Sequences, action, based on the best evidence that all those parallel areas are supplying to it. 00:41:00.235 --> 00:41:05.915 It seems to me that that's done by piping it into the striatum and into the 00:41:05.915 --> 00:41:11.035 basal ganglia, which looks to me like a big funnel, which whittles things down 00:41:11.035 --> 00:41:12.855 to a narrower and narrower compass. 00:41:16.055 --> 00:41:21.375 That's the final one, but the globus pallidus externa is bigger than the interna. 00:41:21.375 --> 00:41:24.375 And Niagara is smaller than that. 00:41:24.635 --> 00:41:29.375 So it's a big funnel. It gets narrower and narrower because parallel things 00:41:29.375 --> 00:41:35.215 are now competing and you're selecting the most urgent thing along the way until 00:41:35.215 --> 00:41:40.695 finally there is a Niagara signal that says, okay, that one, release it. 00:41:41.015 --> 00:41:45.855 It has won a competition along the way in a sense through the base of Yangon. 00:41:45.975 --> 00:41:47.575 It's like an obstacle race. 00:41:48.555 --> 00:41:51.395 There are all these other, I think at the core of entertainment, 00:41:51.655 --> 00:41:53.355 there are all these other guys competing. 00:41:53.775 --> 00:41:57.395 You know, there's a topography there from Cortex. All these other guys competing. 00:41:57.755 --> 00:42:01.795 And the further down you get, the more they have knocked out the competitors. 00:42:01.895 --> 00:42:06.075 And one is going to win. That's the one that right now is going to be acted on. 00:42:06.535 --> 00:42:11.215 And next comes the next runner-up who is busy. 00:42:11.275 --> 00:42:14.975 He still wants to get to control the system. 00:42:15.195 --> 00:42:17.655 And that's the next one. So now that this one is accomplished, 00:42:18.015 --> 00:42:21.715 urgency for this one is lessening, the next guy wins, and so on. 00:42:21.995 --> 00:42:25.155 How that's worked out in circuitry, I can't tell you. 00:42:25.235 --> 00:42:31.355 But that's my intuitive sense of this funnel, this narrowing thing down to the little naiva in the... 00:42:32.155 --> 00:42:36.615 So that means, in your mind, the whole process of actually comparing, 00:42:37.015 --> 00:42:43.735 deciding, evaluating, so on, is a process playing out at the Bezo-Gange level. 00:42:44.929 --> 00:42:49.589 Frontal basal ganglia yeah but the cortex is only dealing with building little 00:42:49.589 --> 00:42:54.249 models yeah of whatever states you might care about but it's also hierarchical 00:42:54.249 --> 00:43:01.189 so so uh super ordered goals are in frontal and so they are gonna and and if 00:43:01.189 --> 00:43:04.089 you look at the weighting of input from cortex, 00:43:04.769 --> 00:43:10.329 to the to the striatum you have the the tail of the car date that's where visuals 00:43:10.329 --> 00:43:12.449 you know Where sensory stuff gets in. 00:43:12.609 --> 00:43:18.229 And the farther frontal you get, the more massive is the input to the striatum. 00:43:18.289 --> 00:43:22.209 So the striatum is biased towards frontal input. 00:43:23.129 --> 00:43:27.829 And that's because it's high level. It's a big thing. It cares about things 00:43:27.829 --> 00:43:29.229 that are talking to the motor system. 00:43:29.429 --> 00:43:34.549 Yeah, it cares. The bits of cortex that directly or indirectly talk to the motor 00:43:34.549 --> 00:43:36.649 system. So only visual cortex. 00:43:36.929 --> 00:43:40.849 It doesn't speak to the cortex. It doesn't matter much. But there still is a 00:43:40.849 --> 00:43:46.709 little bit, it still keeps an eye open a little bit, but massively for the motor-related 00:43:46.709 --> 00:43:50.049 things, and which is… We just talked about sort of early and late attention. 00:43:50.409 --> 00:43:54.569 So in a sense, the basal ganglia is late attention, 00:43:54.889 --> 00:44:00.629 it's you already identified some candidate action plans and you're now down 00:44:00.629 --> 00:44:04.829 to choosing, you know, which of a small number of things to do, 00:44:04.929 --> 00:44:09.929 rather than cortex calculating everything that you could possibly do and having 00:44:09.929 --> 00:44:10.969 to choose between the two. 00:44:11.689 --> 00:44:18.049 So there's a lot of ways of resolving those competitions that don't require 00:44:18.049 --> 00:44:22.649 maybe secretive basic anglia, but operate through maybe a tractor, 00:44:22.789 --> 00:44:24.869 like dynamics and then cortex. 00:44:25.289 --> 00:44:31.269 Yeah, I perfectly agree. It makes eminent sense. And there was something that it reminded me of. 00:44:32.425 --> 00:44:40.485 Which is that the reinforcement learning thing makes perfect sense in that context 00:44:40.485 --> 00:44:44.665 of the whittling down of competitors from a parallel thing to a serial thing. 00:44:44.905 --> 00:44:53.145 Because now, if there is this successive obstacle race across narrower and narrower competition. 00:44:54.105 --> 00:44:58.725 Then you can go in with a reinforcement signal, the outcomes that were successful, 00:44:58.725 --> 00:45:03.805 you can say mark that one that's a good one let that one slip by faster next 00:45:03.805 --> 00:45:09.925 time that was a good one so you can start marking in the base of ganglia marking 00:45:09.925 --> 00:45:14.205 synapses or whatever constellation of activity it is, 00:45:14.725 --> 00:45:21.725 by reinforcement blurring outcomes because that's what action tells you I know 00:45:21.725 --> 00:45:27.785 you threw in the base of ganglia to pacify Tony and it seems to be working so that's problematic 00:45:28.025 --> 00:45:31.485 because you swing 00:45:31.485 --> 00:45:34.325 something else that you haven't accounted for because 00:45:34.325 --> 00:45:37.465 the starting point is and it's very meditative brain 00:45:37.465 --> 00:45:40.945 i care about having very accurate 00:45:40.945 --> 00:45:46.745 reconstructions of the the sources of my center stimulation in the aphid world 00:45:46.745 --> 00:45:52.725 and i use a form of predictive coding for that right okay fine i can buy that 00:45:52.725 --> 00:45:58.725 for for let's say everything up to the central of sulcus from the occipital cortex. Yes, exactly. 00:45:59.385 --> 00:46:03.785 So you have delta in that part. Yeah. Which is not talking about the basic area. 00:46:04.305 --> 00:46:07.965 Now you say, okay, if you go more frontal, then goals come. 00:46:08.585 --> 00:46:13.165 But you haven't already taught me how I use that same hardware, 00:46:13.485 --> 00:46:17.105 that cortical hardware, to get this goal representation. 00:46:18.465 --> 00:46:21.285 So how do the goals come in? Where do they come from? 00:46:22.325 --> 00:46:26.905 High levels on the sensory hierarchy are ideas and concepts, 00:46:27.185 --> 00:46:33.605 high levels of the motor hierarchy are plans, goals, and intentions. Mm-hmm . 00:46:36.341 --> 00:46:42.881 But can't we say everything that is more, let's say, parietal looks at the outside world. 00:46:43.081 --> 00:46:45.621 Everything more frontal looks at the inside world. 00:46:47.881 --> 00:46:52.021 It looks towards action. No, but if you go intention. 00:46:53.281 --> 00:46:58.281 No, but that's what are they for? They are tying you up. 00:46:58.381 --> 00:47:05.501 They are tying up your trade-offs tremendously. As soon as you set out the goal 00:47:05.501 --> 00:47:10.201 to be the greatest guitarist in the world. That's Tony. Tony is. 00:47:11.001 --> 00:47:15.601 Oh, I know you. So that wasn't by a pick. 00:47:16.181 --> 00:47:21.001 You try again, right? Now guitar. What else? Soon there'll be Sheffield. 00:47:21.121 --> 00:47:22.201 World's greatest guitars. 00:47:23.941 --> 00:47:26.841 That eliminates a lot of stuff right off. 00:47:27.041 --> 00:47:30.021 You're now going to spend eight hours a day. you know 00:47:30.021 --> 00:47:33.041 you're narrowing your options by setting up 00:47:33.041 --> 00:47:35.941 such a goal so what was that an answer to 00:47:35.941 --> 00:47:38.661 something you just said i was saying would it 00:47:38.661 --> 00:47:41.681 be fair to say that all the everything in front of the central circus 00:47:41.681 --> 00:47:45.201 characters in all states of the internal they are yeah and they you were not 00:47:45.201 --> 00:47:50.801 convinced by that yeah i objected and that the reason for that is that that 00:47:50.801 --> 00:47:56.821 goal you're setting up is really not so much you feel it and in that sense it's 00:47:56.821 --> 00:48:00.441 internal but it really has to do with your faith in the world, 00:48:00.481 --> 00:48:03.461 your trajectory through this world that you have built up objectively. 00:48:03.741 --> 00:48:05.501 How am I going to get to that place? 00:48:05.841 --> 00:48:09.781 And it now eliminates, once you have set up the future goal, 00:48:09.821 --> 00:48:14.581 and goals are always in the future, once that's set up, you are constraining 00:48:14.581 --> 00:48:19.581 action in very, very concrete and basic ways. 00:48:20.401 --> 00:48:23.541 Because now you can't play all day. 00:48:23.681 --> 00:48:26.681 Now you have to practice and so on. so it 00:48:26.681 --> 00:48:29.721 so even so though they are internal in 00:48:29.721 --> 00:48:32.621 the sense that they're that you you feel an 00:48:32.621 --> 00:48:35.601 impulse to do something or a desire to be 00:48:35.601 --> 00:48:40.361 that great artist or an ambition that's the internal part and of course that's 00:48:40.361 --> 00:48:44.701 related to what I said about motivational systems being meshed in meshed with 00:48:44.701 --> 00:48:49.401 the trompo that's why young calls it frontal limbic frontal limbic is a very 00:48:49.401 --> 00:48:55.361 nice concept so so So naturally there is this and there is the insulin for proprioceptive, 00:48:55.361 --> 00:48:57.041 for interoceptive. 00:48:57.961 --> 00:49:05.861 So naturally, but ultimately I would say goals and planning is the big topic, 00:49:05.981 --> 00:49:07.681 the overriding topic is action. 00:49:09.401 --> 00:49:12.561 Okay, so now we have an idea of the core hierarchy. 00:49:14.961 --> 00:49:19.061 And now in some sense we have a more, let's say at the microscopic level, 00:49:19.261 --> 00:49:23.501 we have to now Now start to think about how all this information is represented 00:49:23.501 --> 00:49:26.061 and passed along between all these circuits. 00:49:27.753 --> 00:49:35.013 And some time ago, you wrote a fantastic review on gamma responses in the brain, 00:49:35.193 --> 00:49:41.153 which I would think you wouldn't also see as a possible substrate for this kind 00:49:41.153 --> 00:49:41.993 of information exchange. 00:49:42.653 --> 00:49:44.133 This is baiting me. 00:49:45.013 --> 00:49:49.653 Oh, you're off the main line. I'm on to you. So, I said that easy. 00:49:49.653 --> 00:49:56.273 But the point is, because now we talked in abstracto about these cortical hierarchies, 00:49:56.293 --> 00:49:59.733 but now signals have to pass between all these different stages. 00:50:00.653 --> 00:50:05.853 So how should I think about that? How does that happen? My answer to that is, 00:50:05.973 --> 00:50:08.993 what passes is signal energy, not messages. 00:50:09.773 --> 00:50:14.793 And how does it pass? it always passes through fibro-bundles that are coherently organized, 00:50:15.933 --> 00:50:20.233 essentially most of the time in point-to-point topographies they may be broken 00:50:20.233 --> 00:50:27.013 up by interleaved columns but across the columns you're still maintaining the topography, 00:50:28.533 --> 00:50:34.533 from map to map there are these transformations from map to map and there are 00:50:34.533 --> 00:50:40.793 examples of actually twisted fibro-bundles like from cortex apparently ground 00:50:40.793 --> 00:50:46.493 it down to the pre-survival nuclear in the ponds, 00:50:46.613 --> 00:50:50.253 there's this weird topology and then it pipes up to the Ceylon. 00:50:54.093 --> 00:50:59.393 My take on the interaction, how they work together, would be that each system 00:50:59.393 --> 00:51:02.813 is doing its specialty all the time. 00:51:05.158 --> 00:51:10.238 But the signal exchange was the question. How do they exchange signals? By their projections. 00:51:10.798 --> 00:51:13.698 What is the signal? Where is the message? 00:51:16.478 --> 00:51:23.518 There isn't any message. Well, you're setting traps for me. Rate coding, yes. 00:51:24.978 --> 00:51:29.958 Really? Yes. That's a fundamental question. Because you said it last year, 00:51:29.998 --> 00:51:31.358 what the computational issues are. 00:51:31.978 --> 00:51:37.858 It's energy, not messages. But you've also talked very much in sort of representational 00:51:37.858 --> 00:51:40.318 terms about what's happening in cortex. 00:51:40.518 --> 00:51:45.198 These are two things which are quite hard to square, because if you want to 00:51:45.198 --> 00:51:50.578 go down a sort of really hardcore dynamicist route, you wouldn't mention the 00:51:50.578 --> 00:51:52.578 word representation in your language either. 00:51:52.818 --> 00:51:57.498 You would say the brain is just in tuned with its body and the environment in 00:51:57.498 --> 00:52:01.658 such a way that, you know, it balances between appropriate protractors so that 00:52:01.658 --> 00:52:03.698 the animal does the right thing in the right time. 00:52:04.038 --> 00:52:08.258 But all of your language is actually about modularity and decomposition. 00:52:08.838 --> 00:52:13.078 It all sounds like the kind of thing that a computationalist would love. 00:52:13.638 --> 00:52:21.358 I am definitely not in the dynamicist camp and the dynamic systems camp, 00:52:21.598 --> 00:52:28.118 but in terms of my representational talk, it's because what I see 00:52:28.378 --> 00:52:35.898 those cortical areas as are two-dimensional maps, and those maps are topographically organized. 00:52:36.338 --> 00:52:41.238 Neighborhood relations are preserved in projections from area to area. 00:52:41.398 --> 00:52:45.718 They form a hierarchy, and there is a pattern. 00:52:47.063 --> 00:52:50.203 On each map a certain as a certain you've 00:52:50.203 --> 00:52:53.023 seen that the famous to tell the picture of when 00:52:53.023 --> 00:52:57.783 they showed when they showed a macabre bullseye with 2d oxyglucose and they 00:52:57.783 --> 00:53:02.443 developed the films and they showed it the beautiful transformed image and visual 00:53:02.443 --> 00:53:07.403 cortex a lot of people misunderstood you by a weasel when they said that the 00:53:07.403 --> 00:53:12.483 stimulus is taken apart into orientation color and so on They said that locally, 00:53:12.623 --> 00:53:14.783 and people started thinking globally, 00:53:15.003 --> 00:53:19.583 but they also said equally emphatically that as you move the electrode across 00:53:19.583 --> 00:53:23.183 V1, you are crossing the columns and. 00:53:23.963 --> 00:53:27.523 Over distance you are maintaining the retinal topography. 00:53:27.783 --> 00:53:34.503 So there is an image in V1, there is a physical pattern of different activation 00:53:34.503 --> 00:53:39.483 states, and I treat the neurons as pixels in a picture. 00:53:39.483 --> 00:53:43.303 Essentially, that's my metaphor, a very abstract metaphor. 00:53:43.623 --> 00:53:49.763 There are pixels in a picture, on a screen, and what happens across the hierarchy 00:53:49.763 --> 00:53:53.163 is it gets transformed, the screens get smaller and smaller, 00:53:53.323 --> 00:53:56.503 they get more and more abstract, and they are interacting. 00:53:57.023 --> 00:54:01.663 This sounds a little bit like what people have described as morphological computation. 00:54:02.063 --> 00:54:06.203 Morphological computation in the context of the brain, you're taking advantage 00:54:06.203 --> 00:54:12.823 of the the geometries and the physical hardware that they end up to do computation for you. 00:54:12.843 --> 00:54:15.203 Are you trying to describe me as a complete disaster now? 00:54:15.443 --> 00:54:20.343 I mean, we were doing so well, yes. It sounded so convincing. 00:54:20.903 --> 00:54:26.143 And we could agree. What you're saying, the morphological computation fits very 00:54:26.143 --> 00:54:29.403 well with how I think, and I call it analog computing. Right, 00:54:29.463 --> 00:54:31.023 but that would not pacify me. 00:54:33.143 --> 00:54:39.663 Because now suddenly we completely switched perspectives, and we talk about 00:54:39.663 --> 00:54:47.543 maps, 2D maps that are wired, conserving some sort of topography between them. 00:54:48.443 --> 00:54:52.983 On the other hand, you talk also about hierarchical feed-forward top-down relations. 00:54:53.363 --> 00:54:59.223 Yeah. So that means I have a forward and backward topographically preserving 00:54:59.223 --> 00:55:03.043 projections between all these maps. So you're going to run out of wires very quickly. 00:55:03.523 --> 00:55:06.203 Not at all. The wires are fixed, 00:55:07.183 --> 00:55:11.763 That doesn't solve my problem. The interaction between feedforward and feedback 00:55:11.763 --> 00:55:18.203 simply changes the synaptic weights and fills those areas in the algorithm. How can it scale, Bjorn? 00:55:18.443 --> 00:55:23.443 How can it scale? Because here you have this sort of Buddha brain that just 00:55:23.443 --> 00:55:28.383 contemplates the world and tries to estimate the sources of its sensory stimulation. 00:55:29.123 --> 00:55:32.403 The world is all practically infinitely variable. 00:55:32.403 --> 00:55:39.903 And in your case I have to capture all that content by really labeling individual 00:55:39.903 --> 00:55:45.003 connections with a certain meaning right? 00:55:46.543 --> 00:55:50.643 There's no way to escape from that there are no labels you have labeled lines, 00:55:50.803 --> 00:55:57.943 you must because you have topography preserving maps so if you want to encode 00:55:57.943 --> 00:56:04.943 any combination of features It is a combination of features expressed by activities 00:56:04.943 --> 00:56:06.803 and locations in such a map. 00:56:07.263 --> 00:56:13.643 And locations in the map must be conserved as you move through such a hierarchy. 00:56:13.963 --> 00:56:19.383 So if I now want to have, let's say, a location invariant representation of 00:56:19.383 --> 00:56:24.603 your hat, and I want to also have a rotation invariant representation of your 00:56:24.603 --> 00:56:27.543 hat, there are a lot of wires going between these maps. 00:56:27.543 --> 00:56:31.903 That's way up in the hippocampus, or way up next to the hippocampus, 00:56:31.903 --> 00:56:37.783 in the perihippocampal, and that's where those recognition things are. 00:56:39.263 --> 00:56:44.603 Your knowledge of a rotation and abstract hat is not this hat. 00:56:44.683 --> 00:56:49.263 This hat is located in a specific place on your maps, all the way up through 00:56:49.263 --> 00:56:50.963 the infratemporal cortex. 00:56:51.123 --> 00:56:54.663 They still maintain topography all the way down through infratemporal cortex. 00:56:54.663 --> 00:57:00.263 When you get up to HT and parahippocampal and entorhinal, that's when you start 00:57:00.263 --> 00:57:01.483 breaking down topography. 00:57:01.723 --> 00:57:07.083 And once you are in hippocampus, you are in sparse mass, which is, 00:57:07.123 --> 00:57:10.483 you know, people, a lot of people don't understand that. 00:57:11.352 --> 00:57:15.032 That the place cells that are responding to 00:57:15.032 --> 00:57:17.992 a single location in a rat's cage 00:57:17.992 --> 00:57:21.692 are spread all over the hippocampus through 00:57:21.692 --> 00:57:26.952 the cross-section and scattering it looks like pepper you know like scattered 00:57:26.952 --> 00:57:33.992 you know and the guys next to them are is another subset of hippocampal cells 00:57:33.992 --> 00:57:38.252 so there is no spatial map in the colliculus or the visual one sense in the 00:57:38.252 --> 00:57:39.492 hippocampus that's an abstract 00:57:39.592 --> 00:57:45.492 map that's for storing the maximum amount of information in a finite number of cells. 00:57:46.612 --> 00:57:52.412 But I'm not running out of pixels or wires because these screens, 00:57:52.612 --> 00:58:00.492 the maps, the two-dimensional maps, just metaphorically think of them like a TV screen. 00:58:00.732 --> 00:58:05.112 You're not running out of TV pixels because you're showing a movie on them. 00:58:05.232 --> 00:58:10.692 They are scintillating and changing all the time. and what they are displaying 00:58:10.692 --> 00:58:16.372 is every one of the areas has a different content depending on where it is in 00:58:16.372 --> 00:58:19.052 the hierarchy and that has to do with the priors. 00:58:19.312 --> 00:58:26.052 For instance, okay, an example. And atomically, it's a fact that inter-aerial 00:58:26.052 --> 00:58:29.112 synapses landing into a volume of cortex, 00:58:29.732 --> 00:58:35.792 but the process coming from another part of cortex or the long-range interactions is. 00:58:37.032 --> 00:58:38.832 Less than 3-2%. 00:58:38.832 --> 00:58:41.452 It's minimal. It's very, very small. 00:58:42.412 --> 00:58:48.372 I love it. But in your proposal, in your proposal now, I think you will need 00:58:48.372 --> 00:58:55.112 way more than this 2% to wire your system together to be able to be this meditating 00:58:55.112 --> 00:58:56.972 Buddha and say, ah, it's a box here. 00:58:58.912 --> 00:59:03.632 I don't quite see why. We may be. 00:59:07.272 --> 00:59:10.312 Kawato did a very nice well that was 00:59:10.312 --> 00:59:18.312 this is a test so we should make a prediction what kind of wiring ratios you 00:59:18.312 --> 00:59:21.652 would get local global and whether they match what you find in the real thing 00:59:21.652 --> 00:59:28.812 I like when I read Young back then his numbers. 00:59:29.672 --> 00:59:34.392 For density of interconnection but remember those two, three percent, 00:59:34.512 --> 00:59:37.652 they are not scattered randomly. They are into part. They are in register. 00:59:37.952 --> 00:59:42.432 They're always in register. Even though they're... We will finish this over dinner. 00:59:42.612 --> 00:59:47.732 I think that because we need maybe some wine to go along with this, 00:59:47.752 --> 00:59:49.232 because we're speculating. 00:59:49.792 --> 00:59:52.552 And we know you don't like that. But, 00:59:53.321 --> 00:59:57.741 So, Björn, with your long career in many fields, which is amazing, right? 00:59:57.781 --> 01:00:04.421 You've been in many fields. You have impact in many fields from music to neuroscience 01:00:04.421 --> 01:00:07.101 to humans' understanding of mushrooms. 01:00:07.421 --> 01:00:10.161 You've been all over the place, right? And it is really astonishing. 01:00:10.321 --> 01:00:17.221 Vedic exegesis, I have solved puzzles that Vedic translators have not managed to solve in the Riveda. 01:00:17.221 --> 01:00:20.281 And I published a paper in Mongolian studies 01:00:20.281 --> 01:00:23.641 where I solved four puzzles 01:00:23.641 --> 01:00:26.721 in the translation of nobody had solved and the 01:00:26.721 --> 01:00:34.201 reason was I had studied the step nomads religious ideas and there's the step 01:00:34.201 --> 01:00:38.601 has a common culture there no mass are going back and forth the ring made over 01:00:38.601 --> 01:00:42.401 step people came down into India and they carried some of their lore from the 01:00:42.401 --> 01:00:44.721 step with them and you knowing what 01:00:44.741 --> 01:00:47.741 they thought about the stories and so on, I could solve problems. 01:00:48.001 --> 01:00:50.541 So, re-veila exegesis without knowing Sanskrit? 01:00:51.761 --> 01:00:56.341 But now you do know Sanskrit. No, I don't. No, okay. Never did I read it. Oh, okay. 01:00:58.021 --> 01:01:01.321 But I heard you sing the girl from Ipinema once in Sanskrit. 01:01:01.921 --> 01:01:05.781 Oh, yeah. But that was, I un-memorized it. Oh, okay. Memorized again. 01:01:06.961 --> 01:01:09.901 So, it was summertime. Oh, it was summertime, sorry. 01:01:10.981 --> 01:01:15.141 But, okay, look, Look, this broad experience that you bring to the table here, 01:01:15.281 --> 01:01:23.341 what is Bjorn's law that we should follow in order to understand how the brain works? 01:01:26.890 --> 01:01:30.790 First of all forget about language that's the 01:01:30.790 --> 01:01:33.710 last thing you want to look at once you have solved all 01:01:33.710 --> 01:01:37.570 the rest of it you can add language probably easily so forget about language 01:01:37.570 --> 01:01:44.570 and messages and so on are language based metaphors look at analog computers 01:01:44.570 --> 01:01:47.270 how you would solve these things analog 01:01:47.270 --> 01:01:53.510 wise which would be morphological computation in a sense and third one, 01:01:55.030 --> 01:02:02.510 I think that's enough okay so then um totally convinced you soon in Christian's thought, 01:02:03.610 --> 01:02:10.490 about three years from now to to check whether you actually have falsified or 01:02:10.490 --> 01:02:17.430 verified your key hypothesis on how the brain works so what hypothesis would 01:02:17.430 --> 01:02:20.750 you like to see really tested in a three year time frame, 01:02:21.690 --> 01:02:24.890 that's a good question That's a nice question. 01:02:26.650 --> 01:02:30.810 Finally, we have a question he liked. Well, I liked all of them. 01:02:30.930 --> 01:02:39.250 You tried to trick me on Ghana and the what was the other one? 01:02:41.010 --> 01:02:47.350 Yeah, you tried to trick me on Ghana and the one on the No, you weren't trying 01:02:47.350 --> 01:02:49.630 to trick me, you just wanted to ask a question. 01:02:49.910 --> 01:02:51.810 Let's see. Prediction. 01:02:53.510 --> 01:02:54.610 Prediction for the future. 01:02:57.530 --> 01:03:00.910 And it's also one that Tony can appreciate. He's going to come to Christian 01:03:00.910 --> 01:03:02.570 and start to check whether it actually happened. 01:03:03.530 --> 01:03:05.870 In winter, it shouldn't be too pleasant. 01:03:10.690 --> 01:03:16.030 I'm trying to get out of this field. So you're going to prediction and walk away, right? 01:03:17.910 --> 01:03:21.490 Not me alone, but Tony rings the doorbell. Let's see. 01:03:22.670 --> 01:03:34.470 My dream experiment my dream research line three years from now with an answer what would that be, 01:03:36.030 --> 01:03:44.710 hell I can't think well I think given our discussion an obvious one would be 01:03:44.710 --> 01:03:50.330 that you're going to prove that you're topographically conserving architecture 01:03:50.330 --> 01:03:52.390 that you can scale. I know now. 01:03:53.490 --> 01:03:57.030 I mean to support you. I know now. 01:03:57.170 --> 01:04:02.750 It's not that. I likely touched upon the fact that cortex works probabilistically, 01:04:02.930 --> 01:04:06.110 but you need an estimate. And. 01:04:10.137 --> 01:04:15.457 There is work that shows that as long as cortex works on the probabilistic basis. 01:04:16.297 --> 01:04:22.357 Its transmission is fast its its operations are fast as soon as you want try 01:04:22.357 --> 01:04:28.117 to constrain cortex to precipitate an estimate things not crash but slow down 01:04:28.117 --> 01:04:33.117 and i forget now the name of the people who suggested that. 01:04:33.557 --> 01:04:37.517 So probabilistically cortex is fast and 01:04:37.517 --> 01:04:40.377 Mumford back in the 90s had a 01:04:40.377 --> 01:04:44.057 couple of papers on corticothalamic relations and in 01:04:44.057 --> 01:04:46.837 the abstract he says estimates are made 01:04:46.837 --> 01:04:54.237 in the thalamus and I am on that line I think that cortex is working on a probabilistic 01:04:54.237 --> 01:04:59.877 basis and when it is time to make an estimate for actual action things these 01:04:59.877 --> 01:05:05.857 are these those are made in in subcortical locations and like basal ganglia for action, 01:05:05.977 --> 01:05:11.077 colliculus for prioritization and orienting. 01:05:11.237 --> 01:05:20.897 And I think the thing I would like to be tested is is there a global best estimate of the whole. 01:05:21.577 --> 01:05:28.497 Where the whole probabilistic panoply of cortical areas where all that information 01:05:28.497 --> 01:05:34.437 is collapsed to one single estimate of my best now in the moment estimate of 01:05:34.437 --> 01:05:35.937 what's happening around me. 01:05:35.997 --> 01:05:41.657 But this sounds like your brain, your cortex, which already had a problem with 01:05:41.657 --> 01:05:47.717 representing all these topographic maps, now has to support a full panoply of Bayesian. 01:05:48.457 --> 01:05:53.277 Probabilistic estimates of different interpretations of the world and maintain 01:05:53.277 --> 01:05:54.677 those at the same time. Yeah. 01:05:55.057 --> 01:05:58.617 So, I mean, do you really want to say that that's what the brain is doing? 01:05:58.617 --> 01:06:03.197 I want to test whether that happens in the pool or not. 01:06:04.682 --> 01:06:09.442 That it collapses there is massive convergence of multiple higher order cortical areas, 01:06:10.042 --> 01:06:12.982 uh most of the the v1 projects involved now but 01:06:12.982 --> 01:06:15.662 but the the dorsal pulmonary which is what 01:06:15.662 --> 01:06:18.942 i'm looking at it has this massive conversion from 01:06:18.942 --> 01:06:24.482 a lot of higher areas both frontal and parietal and temporal okay so there is 01:06:24.482 --> 01:06:29.702 this converted place and it has it's invested with a special inhibitory internal 01:06:29.702 --> 01:06:34.482 that exists no other place in the thalamus and their long-range inhibition in 01:06:34.482 --> 01:06:36.902 the dorsal discovered by Kathy Rockland. 01:06:37.102 --> 01:06:42.822 And that place to me looks like a place that might be ideally placed to make 01:06:42.822 --> 01:06:48.742 a global best estimate in the moment of what holy cortical information, 01:06:50.702 --> 01:06:53.362 amounts to in terms of one global estimate. 01:06:54.182 --> 01:06:58.762 And you need sort of a global estimate because some of the things are not disambiguated 01:06:58.762 --> 01:07:00.482 until you have put them all together. 01:07:00.822 --> 01:07:05.322 McGurk effect, you know, You look at somebody's mouth and they say one syllable 01:07:05.322 --> 01:07:10.122 and another one comes into your earphones and so on. Actually, 01:07:10.182 --> 01:07:10.902 it's all the time with Tony. 01:07:14.462 --> 01:07:18.002 But that's also in your conscious sea. The conscious sea is that. 01:07:18.062 --> 01:07:19.262 That's the conscious sea. 01:07:19.422 --> 01:07:24.802 So if that could be tested in three years, I would be very happy. 01:07:25.842 --> 01:07:29.162 But your prediction would be that it will be true. It will be consistent. 01:07:29.162 --> 01:07:32.562 There will be a global-based estimate in the dorsal pulmonary. 01:07:32.842 --> 01:07:36.242 That's my prediction. A clearer prediction you can't get. 01:07:36.482 --> 01:07:39.122 Very good. John Marker, thank you very much for this conversation. 01:07:39.402 --> 01:07:41.182 Wonderful. I enjoyed it. 01:07:42.742 --> 01:07:48.402 The CSN podcast was produced by the Convergent Science Network of Biometrics 01:07:48.402 --> 01:07:54.822 and Biohybrid Systems, a project funded by the European Sevens Research Framework Program. 01:07:56.382 --> 01:08:01.642 For more interviews, recorded lectures, or upcoming conferences in the field 01:08:01.642 --> 01:08:07.922 of biometrics and bio-hybrid systems, go to csnnetwork.eu. 01:08:08.242 --> 01:08:10.082 And thank you for listening.