WEBVTT 00:00:00.017 --> 00:00:00.717 Huh? Okay. 00:00:04.057 --> 00:00:06.877 This is the Convergent Science Network podcast. 00:00:08.897 --> 00:00:13.337 Leading researchers in the domain of neuroscience, brain theory and technology 00:00:13.337 --> 00:00:17.097 are interviewed by Paul Verschoor and Tony Prescott. 00:00:18.937 --> 00:00:22.957 This is Paul Verschoor with the Convergent Science Network podcast together 00:00:22.957 --> 00:00:27.597 with my colleague Tony Prescott and with Murray Shanahan who was speaking this 00:00:27.597 --> 00:00:29.697 morning in our summer school here in Barcelona. 00:00:30.017 --> 00:00:37.057 And Murray, you presented, let's say, a bit more an abstract view on cortical 00:00:37.057 --> 00:00:39.337 dynamics and how we could think about cortical dynamics. 00:00:39.517 --> 00:00:45.237 So what are the key features of this model of cortex you're presenting and what 00:00:45.237 --> 00:00:45.937 are you trying to explain? 00:00:46.597 --> 00:00:52.717 So I think the key feature of the model, which reflects my current interests 00:00:52.717 --> 00:00:56.777 really, is the richness of the dynamics that it produces. 00:00:56.777 --> 00:00:59.857 Produces so um metastability is 00:00:59.857 --> 00:01:03.017 the was one of the main themes that i talked about this morning so 00:01:03.017 --> 00:01:05.897 that's one important feature uh which 00:01:05.897 --> 00:01:08.977 is related to um to the size 00:01:08.977 --> 00:01:11.737 of the of the repertoire of different states that the 00:01:11.737 --> 00:01:14.877 system can produce as well so metastability just to 00:01:14.877 --> 00:01:17.557 to spell that out so uh so a 00:01:17.557 --> 00:01:20.677 system is is metastable uh basically if 00:01:20.677 --> 00:01:23.657 it's instead of um instead of falling 00:01:23.657 --> 00:01:26.697 into a stable attractor it rather it 00:01:26.697 --> 00:01:29.877 sort of lingers in the vicinity of an attractor like state 00:01:29.877 --> 00:01:33.117 um without you know stabilizing and 00:01:33.117 --> 00:01:36.357 falling straight right into that attractor um and 00:01:36.357 --> 00:01:39.537 then maybe uh but then maybe moves on to a different attractor 00:01:39.537 --> 00:01:42.877 so um so for uh 00:01:42.877 --> 00:01:45.717 for listeners who who for whom that isn't very 00:01:45.717 --> 00:01:49.137 familiar uh terminology so uh so 00:01:49.137 --> 00:01:51.877 the idea would be uh or one kind of 00:01:51.877 --> 00:01:55.357 attractor state would would be where um where 00:01:55.357 --> 00:01:59.857 all the parts of the brain were all oscillating in synchrony all at the same 00:01:59.857 --> 00:02:03.937 time and of course this never happens and it would be it would be very very 00:02:03.937 --> 00:02:07.317 bad if it did but that would be that's one kind of attractor state that you 00:02:07.317 --> 00:02:11.277 get you can get in dynamical systems is where you've got a whole load of oscillators 00:02:11.277 --> 00:02:13.757 they're all completely synchronized with each other. 00:02:14.658 --> 00:02:21.318 In the brain, I think the kind of dynamics we see is where there's lots of oscillations going on, 00:02:21.378 --> 00:02:25.958 there's lots of patterns of synchrony, but what we're really interested in is 00:02:25.958 --> 00:02:29.738 we're interested in states of partial synchrony, where some parts of the brain 00:02:29.738 --> 00:02:31.438 are synchronized with other parts of the brain. 00:02:31.438 --> 00:02:33.878 That often represents that they're talking to each other. 00:02:34.238 --> 00:02:40.298 And we're also interested in metastable states of synchronization, 00:02:40.438 --> 00:02:45.278 so where it doesn't stay in that exact combination of synchronized parts, 00:02:45.558 --> 00:02:51.058 but that sort of coalition of synchronized parts is like that for a while, 00:02:51.158 --> 00:02:54.138 and then the coalition breaks apart and it finds another state. 00:02:54.418 --> 00:03:00.198 But now you could argue that this metastable state, which is defined in a state 00:03:00.198 --> 00:03:03.838 space, is in some sense arbitrary because as observer, 00:03:04.198 --> 00:03:09.458 I can decide at what level I define my state space and what's metastable at 00:03:09.458 --> 00:03:13.778 one level of description can just be a fixed point attractor at another level of description. 00:03:14.478 --> 00:03:18.798 So do you see that as a problem, this arbitrariness of how we define such a state space? 00:03:20.018 --> 00:03:25.438 Well, it certainly doesn't have to be so arbitrary and there are various ways 00:03:25.438 --> 00:03:28.038 that you can try to quantify it precisely. 00:03:28.038 --> 00:03:31.718 Precisely so uh so one way for example uh 00:03:31.718 --> 00:03:35.278 is to is to look at um uh 00:03:35.278 --> 00:03:37.878 in the case of these coupled oscillators if you 00:03:37.878 --> 00:03:41.098 have a whole collection of oscillators that are say that are that are decoupled 00:03:41.098 --> 00:03:46.538 um then um then there's no connection between them at all really and in that 00:03:46.538 --> 00:03:52.198 case that's that's the sort of um uh the trivial case where where there's not 00:03:52.198 --> 00:03:55.698 going to be any a any um you know, 00:03:55.698 --> 00:03:58.038 synchronization except coincidental synchronization. 00:03:58.198 --> 00:04:00.138 So that's the statistical baseline. 00:04:00.438 --> 00:04:05.158 So if you know, if you're looking at your system and things fall into synchrony. 00:04:06.233 --> 00:04:11.773 A lot more a lot more than you would expect them to uh in in the case of that kind of um, 00:04:12.373 --> 00:04:17.913 that kind of baseline statistics then you know that that's that's serious synchronization 00:04:17.913 --> 00:04:24.473 a serious sort of uh are you are you saying with this meta stability concept that let's say between. 00:04:25.333 --> 00:04:30.713 Stability and complete chaos it's in this intermediate zone that interesting 00:04:30.713 --> 00:04:34.453 things are happening yes that's certainly certainly like an edge of chaos kind 00:04:34.453 --> 00:04:37.693 of it's very much Much related to that, yeah, very much related to that. 00:04:37.773 --> 00:04:43.233 So there is a kind of criticality phenomenon there where the sort of systems 00:04:43.233 --> 00:04:47.813 that you're interested in or the dynamics that you're interested in is poised 00:04:47.813 --> 00:04:49.013 between order and disorder. 00:04:49.373 --> 00:04:55.713 So a highly ordered system will be one that was just, say, completely synchronized, 00:04:55.713 --> 00:04:58.453 and a highly disordered system will be one like the one I just described, 00:04:58.553 --> 00:05:01.313 completely decoupled, there's no relationship between the different oscillators. 00:05:01.453 --> 00:05:06.593 What you're interested in is this state which is between the two where there's 00:05:06.593 --> 00:05:09.013 a certain amount of order and a certain amount of disorder. 00:05:09.273 --> 00:05:16.193 So the parts are sort of interacting with each other, but there isn't one dominant state that persists. 00:05:16.473 --> 00:05:23.433 So now you were defining metastability also in relation to the physiology of 00:05:23.433 --> 00:05:26.073 people like Pascal Friess, who comes from the school of Wolf Singer, 00:05:26.333 --> 00:05:32.633 who sort of ascribe a lot of importance to, let's say, synchronization phenomena in the brain. 00:05:32.753 --> 00:05:34.873 Yeah. So how are these two things now related? 00:05:35.073 --> 00:05:39.333 Yeah, so that's very much the backdrop for the discussion we've just had, 00:05:39.413 --> 00:05:46.973 really, is, so Pascal, according to Pascal Friese's communication through coherence hypothesis. 00:05:47.993 --> 00:05:52.873 Which makes a lot of sense to me, you can think of two populations of neurons 00:05:52.873 --> 00:05:57.633 that are oscillating in synchrony as basically in a position to communicate 00:05:57.633 --> 00:05:59.413 with each other and cooperate with 00:05:59.413 --> 00:06:02.813 each other to influence each other and exchange information and so on. 00:06:03.253 --> 00:06:07.793 Because when the troughs and peaks of their activity coincide, 00:06:07.793 --> 00:06:13.493 then they're in a good position to exchange spikes, so long as all the delays work out and so on. 00:06:15.353 --> 00:06:21.693 So that's the reason why synchronisation is one hypothesis, but it's one reason 00:06:21.693 --> 00:06:26.413 why synchronised collections of neurons might be of interest if you're looking 00:06:26.413 --> 00:06:28.313 at populations on different parts of the brain. 00:06:28.313 --> 00:06:31.113 Counterintuitive because in some sense 00:06:31.113 --> 00:06:34.073 if i would look at these two neurons and i would plot 00:06:34.073 --> 00:06:37.153 their activity over time but against each 00:06:37.153 --> 00:06:39.933 other they might actually be in some sort 00:06:39.933 --> 00:06:44.573 of fixed point they might be in a very limit a limit cycle but it would not 00:06:44.573 --> 00:06:49.993 look like a very variable metastable state so how is then the coherence in gamma 00:06:49.993 --> 00:06:55.793 reflecting this metastability or this criticality phenomena yeah well first 00:06:55.793 --> 00:06:58.733 of all you've got to look at different time scales um. 00:06:59.949 --> 00:07:03.829 So, um, uh, and, and you're also looking at whole populations of neurons. 00:07:03.909 --> 00:07:07.009 So you're certainly not looking at just single neurons, but you're looking at 00:07:07.009 --> 00:07:07.849 populations of neurons. 00:07:07.969 --> 00:07:12.609 So, so you might look at one population of neurons. You might see that the overall 00:07:12.609 --> 00:07:16.449 firing rate, the mean firing rate is oscillating in a, you know, 00:07:16.449 --> 00:07:19.749 for, for a while is oscillating in a really, in a regular kind of way, 00:07:19.769 --> 00:07:22.429 say a gamma type frequency, 40 Hertz or something. 00:07:23.129 --> 00:07:29.209 And, um, and then you've got another population that maybe is also oscillating in the same kind of way. 00:07:29.209 --> 00:07:32.069 And the oscillations are in are in synchrony and then 00:07:32.069 --> 00:07:34.989 then those two populations are in a position to 00:07:34.989 --> 00:07:37.989 exchange information um and so 00:07:37.989 --> 00:07:41.089 that's the kind of situation that that that you're interested in 00:07:41.089 --> 00:07:44.609 now the information is going to be exchanged during those 00:07:44.609 --> 00:07:48.409 little peaks of of of activation and 00:07:48.409 --> 00:07:51.589 during a peak of activation um then that 00:07:51.589 --> 00:07:54.769 whole population is going to be most receptive to incoming spikes 00:07:54.769 --> 00:07:58.089 and is going to be the individual neurons going to be more likely statistically 00:07:58.089 --> 00:08:00.969 more likely to fire than in a trough of 00:08:00.969 --> 00:08:04.289 of activation and so that's why the communication 00:08:04.289 --> 00:08:07.349 can take place you know in that in that time but surely that 00:08:07.349 --> 00:08:12.349 if you're thinking about maximizing the opportunity to fire the postsynaptic 00:08:12.349 --> 00:08:16.529 neuron you don't want to come when it's firing you want to come just a bit before 00:08:16.529 --> 00:08:21.269 that yeah okay that's absolutely true but but but you've uh in within 40 hertz 00:08:21.269 --> 00:08:24.589 you've got quite a few milliseconds uh it's quite a it's a 00:08:24.669 --> 00:08:27.389 it's a little window of opportunity so you want you know the 00:08:27.389 --> 00:08:30.389 timing's got to work out that's that's for sure uh you've 00:08:30.389 --> 00:08:33.729 got a window of opportunity you know um uh so quite 00:08:33.729 --> 00:08:39.369 a lot can happen in that in that little window um uh and uh and also you you've 00:08:39.369 --> 00:08:42.909 also got to worry about the delays so so you know the there are going to be 00:08:42.909 --> 00:08:46.929 delays between uh depending on the length of the connection and the type of 00:08:46.929 --> 00:08:50.749 connection and so on they're going to be delays as well so all of that's got to work out. 00:08:51.749 --> 00:08:54.269 But the point is that it's getting, 00:08:55.043 --> 00:08:59.083 the right relationship in the phase to work out is going to maximize the opportunity 00:08:59.083 --> 00:09:01.243 for information to be exchanged. 00:09:01.543 --> 00:09:03.863 And also, allows for competition. 00:09:04.323 --> 00:09:10.803 So you might have three populations. Two populations are trying to influence 00:09:10.803 --> 00:09:17.343 a third, and one is going to win out by entraining the downstream population. 00:09:17.363 --> 00:09:21.563 Once it's become entrained, then it can shut out the other guy who will end 00:09:21.563 --> 00:09:22.703 end up in the opposite phase of 00:09:22.703 --> 00:09:25.943 the relationship and that's when the information sort of gets exchanged. 00:09:26.823 --> 00:09:34.503 The dominant idea about gamma is that it's very much a locally generated response, right? 00:09:34.563 --> 00:09:39.243 So you have let's say cortical circuits or the hippocampus, it's actually the 00:09:39.243 --> 00:09:44.583 tight coupling of excitatory neurons with local fast GABA-A mediated inhibitory neurons. 00:09:44.863 --> 00:09:48.703 So as soon as you start to drive the excitatory cells they will drive the local 00:09:48.703 --> 00:09:50.323 inhibition that will shut down 00:09:50.323 --> 00:09:53.783 the whole population. So now you get the rhythmicity in the gamma range. 00:09:53.923 --> 00:09:58.343 So you could then first argue, okay, if I have a gamma in two areas, 00:09:58.483 --> 00:10:01.583 it means, okay, there's a drive onto these cells. This is the first thing. 00:10:01.863 --> 00:10:05.623 And then you can say, well, now if these two areas talk to each other, 00:10:05.763 --> 00:10:10.723 okay, by necessity it will happen within this gamma rhythm because they cannot 00:10:10.723 --> 00:10:12.763 fire at any other rhythm, but they become driven. 00:10:13.003 --> 00:10:17.323 Yeah. So then how meaningful is it really to talk about, let's say, 00:10:17.343 --> 00:10:24.523 enhanced communication between areas when gamma becomes more or less coherent? 00:10:25.023 --> 00:10:28.923 Yeah. Well, I think it's particularly meaningful in the context of this kind 00:10:28.923 --> 00:10:32.863 of competition. So where, indeed, you may have…. 00:10:33.951 --> 00:10:37.311 You know several populations several neuronal populations are 00:10:37.311 --> 00:10:40.391 all as it were relevant to the ongoing situation so 00:10:40.391 --> 00:10:43.451 if you've got you've got uh uh well 00:10:43.451 --> 00:10:46.471 actually actually an example a potential example is 00:10:46.471 --> 00:10:49.211 binocular rivalry maybe you know about binocular rivalry so where 00:10:49.211 --> 00:10:53.151 you've got two um two gratings one 00:10:53.151 --> 00:10:57.911 horizontal one vertical uh presented to each eye and uh so the right eye can 00:10:57.911 --> 00:11:01.911 see the say the horizontal grating the left eye the vertical grating and uh 00:11:01.911 --> 00:11:06.671 and then there's this binocular rivalry phenomenon whereby you don't see the 00:11:06.671 --> 00:11:10.351 two gratings overlaid on each other, you don't see a crisscross pattern, 00:11:10.511 --> 00:11:14.611 but rather you tend to see consciously or become aware of either one or the other. 00:11:15.171 --> 00:11:19.571 The one will fade in and out while the other one fades in and out. 00:11:19.811 --> 00:11:22.391 So that's known as binocular rivalry. 00:11:24.211 --> 00:11:27.631 That's an example of where there's clearly some competition petition going on 00:11:27.631 --> 00:11:30.911 between rival neuronal populations. 00:11:31.671 --> 00:11:39.051 One way possibly to account for that is in terms of this communication through 00:11:39.051 --> 00:11:45.871 coherence idea, where one population temporarily entrains another population further downstream. 00:11:46.331 --> 00:11:52.511 That dynamics goes on for a little while, but then But then it can sort of burn 00:11:52.511 --> 00:11:57.231 itself out for statistical reasons, basically, because they're equally salient stimuli. 00:11:57.531 --> 00:12:00.451 And then the other population becomes entrained instead. 00:12:00.671 --> 00:12:04.231 And so you have this flickering in and out. And we had a paper in Frontiers 00:12:04.231 --> 00:12:05.631 in Computational Neuroscience paper. 00:12:05.711 --> 00:12:10.231 I had a paper with one of my PhD students, Mark Wildey, where we showed that 00:12:10.231 --> 00:12:13.611 kind of phenomenon in the context of Pascal Frese's ideas. 00:12:13.611 --> 00:12:19.051 So now we know a little bit, let's say, the physiological phenomena we would 00:12:19.051 --> 00:12:20.871 like to understand and explain, right? 00:12:21.331 --> 00:12:24.371 And what you told us is that actually you start to model. 00:12:25.192 --> 00:12:28.212 These large-scale network phenomena with fairly 00:12:28.212 --> 00:12:31.472 abstract oscillator models yeah so so 00:12:31.472 --> 00:12:34.992 why did you make that step yes um i 00:12:34.992 --> 00:12:40.592 think i made that step just because i just fell into it it wasn't really that 00:12:40.592 --> 00:12:46.792 i uh that i i took a uh you know a very carefully um a very careful decision 00:12:46.792 --> 00:12:52.212 about what i should look into next um uh but i mean my you know my background 00:12:52.212 --> 00:12:54.072 is in computer computer modeling. 00:12:54.272 --> 00:12:59.472 So, so my natural, my natural tendency is to tinker with my computer model. 00:12:59.612 --> 00:13:03.472 So I'm sitting there with MATLAB fiddling around and, uh, playing with my computer 00:13:03.472 --> 00:13:04.992 models. And that's what I, that's what I do. 00:13:05.412 --> 00:13:12.732 Um, and then, yeah, but, uh, but, but I'm always vulnerable to, 00:13:12.832 --> 00:13:16.752 you know, there's, so, so a lot of people are interested in that kind of thing 00:13:16.752 --> 00:13:18.012 in its own, in its own right. 00:13:18.292 --> 00:13:22.552 Physicists in particular, you know, kind of quite interested in that sort of thing in its own right. 00:13:22.832 --> 00:13:26.592 But then if you're trying to make a claim about its relevance to brain dynamics 00:13:26.592 --> 00:13:30.012 or cognition or anything like that, then of course you're instantly vulnerable 00:13:30.012 --> 00:13:34.032 to the criticism that, well, where's the data? How does it relate to data? 00:13:34.212 --> 00:13:37.852 And so of course it's very important to try and relate it to real data. 00:13:38.032 --> 00:13:41.432 But in my case it's been a bit back to front, so I was tinkering around with 00:13:41.432 --> 00:13:45.992 the models for a lot and only a little bit later did the opportunity to make 00:13:45.992 --> 00:13:47.872 it match up with real data come in. 00:13:47.972 --> 00:13:50.992 So what are the key parameters that you then fiddle with when you're sitting 00:13:51.112 --> 00:13:55.112 behind MATLAB on your desktop computer in your office? 00:13:55.252 --> 00:14:01.252 Ah, well, so in coming up with that particular model, the metastable chimera 00:14:01.252 --> 00:14:07.612 states, I tinkered with so many things, but particularly with the network construction. 00:14:10.352 --> 00:14:16.952 So, yeah, the sparsity of the network, lots of parameters to do with the network construction. Yeah. 00:14:17.946 --> 00:14:20.646 Yeah, the parameters of the oscillator models, the coupling. 00:14:20.986 --> 00:14:22.486 I mean, there are hundreds of them. 00:14:22.546 --> 00:14:27.566 And of course, if I was a good mathematician, I mean, I'm handicapped by being 00:14:27.566 --> 00:14:33.606 a mediocre mathematician, by not dealing with real data, not being a proper scientist. 00:14:33.826 --> 00:14:39.626 The only thing I'm any good at is programming. So, you know, so you end up tackling, 00:14:39.806 --> 00:14:46.486 it's like Dorothy just now was talking about, you know, if you're a capuchin 00:14:46.486 --> 00:14:50.706 and you see you've got a hammer stone, then you see everything is a nut. 00:14:50.846 --> 00:14:52.826 Well, it's a little bit like that with programming, you know, 00:14:52.866 --> 00:14:55.226 I see everything as a programming problem. Right. 00:14:55.526 --> 00:15:02.726 So, I tend to… But for the model to close that, I think the key parameters you're 00:15:02.726 --> 00:15:07.206 controlling is, let's say, the phase of your oscillators, I guess the transduction 00:15:07.206 --> 00:15:08.606 delay in their interactions, 00:15:08.826 --> 00:15:16.046 these would be roughly the key parameters that would be dominating the properties of your network. 00:15:16.406 --> 00:15:20.666 Yeah, so the coupling, so the two main parameters that we end up with fiddling 00:15:20.666 --> 00:15:23.986 around with are the coupling and the delays. 00:15:26.406 --> 00:15:30.166 I'd just like to go into those models a bit more because one of the things that 00:15:30.166 --> 00:15:35.146 surprised me in your talk was you described some of the history of the study 00:15:35.146 --> 00:15:38.446 of oscillators, which has been going on for hundreds of years. 00:15:39.026 --> 00:15:44.486 But then there's some fairly fundamental discoveries that you have made, 00:15:44.566 --> 00:15:51.766 and I think you mentioned a 2002 paper about properties of oscillators that we didn't know about. 00:15:51.766 --> 00:15:55.546 And uh can you say a bit more about those discoveries why 00:15:55.546 --> 00:15:58.866 didn't we know those things before and uh you 00:15:58.866 --> 00:16:02.706 know is there a lot much more to know about these simple oscillator systems 00:16:02.706 --> 00:16:08.186 i know there are large numbers of oscillators but there seems to be um potentially 00:16:08.186 --> 00:16:13.046 for areas like brain science lots of scope for more uh discoveries in this yeah 00:16:13.046 --> 00:16:16.206 i think i think it's because the so So it's physicists, 00:16:16.366 --> 00:16:18.586 you know, do all the serious, 00:16:18.926 --> 00:16:23.526 you know, legwork with all the heavy lifting with all of these oscillator models. 00:16:24.326 --> 00:16:29.746 And they have a particular mindset which tends to focus in on things like, 00:16:29.826 --> 00:16:32.446 oh, let's look at, you know, let's look at the limiting case where there's an 00:16:32.446 --> 00:16:34.126 infinite number of oscillators, for example. 00:16:34.166 --> 00:16:37.686 Or, you know, let's look at the total, you know, where everything is completely connected. 00:16:37.686 --> 00:16:40.766 And um and and also 00:16:40.766 --> 00:16:44.066 let's look at stability conditions let's look at all these bifurcation diagrams 00:16:44.066 --> 00:16:49.106 and the stability conditions and in fact there's kinds of um dynamics which 00:16:49.106 --> 00:16:52.526 i'm interested in i i don't think they're actually very surprised that they 00:16:52.526 --> 00:16:55.566 exist i just don't think they particularly thought that they were interesting 00:16:55.566 --> 00:17:00.006 uh as is what i i suspect and it's only um. 00:17:00.896 --> 00:17:03.496 But I mean, they seem to be getting interested, because that particular paper 00:17:03.496 --> 00:17:09.156 of mine has been quite well cited by physicists in looking at it, relatively speaking. 00:17:10.436 --> 00:17:14.856 But I think it's partly because they probably knew that these sorts of things 00:17:14.856 --> 00:17:19.836 went on, but they weren't interested in irregular networks that are not homogenous. 00:17:21.576 --> 00:17:24.616 But could you then elaborate a bit what you discovered with this model, 00:17:24.696 --> 00:17:28.236 the so-called Chimera states, and why they are interesting? And perhaps also 00:17:28.236 --> 00:17:31.756 how they relate to different network configurations, because that's obviously 00:17:31.756 --> 00:17:32.976 a key parameter. Yeah, sure. 00:17:33.256 --> 00:17:36.236 So why they're interesting is… Well, maybe first define what they are. 00:17:36.236 --> 00:17:44.056 So these chimera states is where you have a set of oscillators that basically 00:17:44.056 --> 00:17:47.556 partitions into two or more subsets, 00:17:47.636 --> 00:17:51.276 well, two subsets really, 00:17:51.376 --> 00:17:55.396 where one of them is synchronized and the other is desynchronized. 00:17:55.496 --> 00:17:56.796 So that's a chimera state. 00:17:57.976 --> 00:18:01.356 When these were first discovered by the physicists, I think they found it quite 00:18:01.356 --> 00:18:06.136 surprising that these things could exist, especially because they had very homogenous 00:18:06.136 --> 00:18:10.776 connectivity in those ones that they were looking at. So they were a bit surprised by that. 00:18:12.156 --> 00:18:15.416 This is certainly of interest from the perspective of brain dynamics, 00:18:15.516 --> 00:18:20.476 because you're not really interested in states either where there's no synchrony, 00:18:20.476 --> 00:18:23.596 because nothing's happening of interest at all there in the brain. 00:18:23.896 --> 00:18:26.536 Nor are you interested in the case where everything is synchronized because 00:18:26.536 --> 00:18:30.936 that's basically seizure and nothing interesting is happening either. 00:18:31.056 --> 00:18:35.476 You're interested in the case where some populations are synchronized and some 00:18:35.476 --> 00:18:39.136 are not, and the ones that are synchronized are the ones that are governing behavior. 00:18:40.650 --> 00:18:48.430 So we should try and relate this maybe a little bit to the behavior of animals, 00:18:48.750 --> 00:18:50.830 right? Because we've been talking very abstract terms so far. 00:18:51.230 --> 00:18:54.950 I think the key thing about your Chimera states was also you see there's a signature 00:18:54.950 --> 00:18:59.390 of, let's say, a network property, right? 00:18:59.430 --> 00:19:02.930 Where you would say, well, now I start to get minorities, if you want, 00:19:03.110 --> 00:19:06.790 in the whole population of oscillators that start to, let's say, 00:19:06.830 --> 00:19:10.250 be out of phase with the majority. Yeah, that's right. 00:19:10.750 --> 00:19:13.990 Well, actually, I think it's the minority that are going to be synchronized, 00:19:14.030 --> 00:19:17.890 and the majority are going to be… Oh, your data looks a bit different, I must say. 00:19:17.930 --> 00:19:22.150 The data, in that particular model, it's usually about half and half, actually. 00:19:22.330 --> 00:19:26.430 So in that particular model, about half the oscillators end up synchronized 00:19:26.430 --> 00:19:29.430 with each other, and about half end up desynchronized. 00:19:30.090 --> 00:19:34.390 But I think in real data, you probably expect a smaller set to be synchronized 00:19:34.390 --> 00:19:36.970 and to be governing the behavior of the animal at that time. 00:19:37.090 --> 00:19:39.710 That's also the link to your meta-stability, right? Where you say, 00:19:39.750 --> 00:19:44.430 look, we want to be in between order and chaos, which would roughly be expressed 00:19:44.430 --> 00:19:46.470 in such a network as these chimera states. 00:19:46.630 --> 00:19:50.050 You say, look, now I'm at this critical point where interesting stuff can happen. 00:19:50.230 --> 00:19:52.590 Yeah, but it's not just the chimera states, it's the metastability, 00:19:52.770 --> 00:19:57.330 because you don't want to be stuck in one chimera state either. Sure, absolutely. 00:19:57.510 --> 00:20:03.550 You want that particular coalition of synchronized oscillators, 00:20:03.670 --> 00:20:05.770 you don't want that to be the same one forever. 00:20:05.770 --> 00:20:08.710 It's doing its thing for a while and then you're expected 00:20:08.710 --> 00:20:11.470 to break apart and then another coalition of oscillators so a different 00:20:11.470 --> 00:20:14.990 chimera state to arise exactly but you've got this 00:20:14.990 --> 00:20:20.910 uh minority group of coupled uh oscillators and then you say these ones synchronize 00:20:20.910 --> 00:20:25.310 these ones that are desynchronized what do you think they're doing are they 00:20:25.310 --> 00:20:28.870 just well they're along in the background but not having any influence well 00:20:28.870 --> 00:20:32.990 that's i think that's exactly what they're what they're doing so So any evidence for that? 00:20:34.470 --> 00:20:44.230 Well, I mean, things that are not synchronized or not even exhibiting regular oscillatory behavior, 00:20:44.530 --> 00:20:50.350 I mean, that's mostly what's going on in the brain, right? Right. 00:20:50.830 --> 00:20:55.590 But I mean, I wouldn't want to go as far and say that things that aren't oscillatory 00:20:55.590 --> 00:20:58.790 aren't of importance. Most of the sensory events aren't oscillatory. 00:20:59.010 --> 00:21:02.250 No, no, sure. Okay, sure, sure. Sure, sure, that's true. 00:21:04.710 --> 00:21:10.490 Yes, I think, I mean, I see it as, I mean, this is a hypothesis that, 00:21:10.510 --> 00:21:16.610 you know, demands empirical validation, but I see it as a mechanism for communication, 00:21:17.490 --> 00:21:22.470 and cooperation among populations that are anatomically distributed around the brain. 00:21:22.970 --> 00:21:26.510 So I think we should talk a bit about behavior, right? Because all this has been very abstract. 00:21:26.690 --> 00:21:29.890 Before we come to behavior, though, when you're saying, okay, 00:21:29.950 --> 00:21:35.390 the interesting stuff, happens in the synchronized oscillators yeah is that 00:21:35.390 --> 00:21:39.790 because you have some fundamental theory about brain hardware that you know. 00:21:40.551 --> 00:21:45.691 Perhaps goes beyond this information transmission thing i mean so what's happening 00:21:45.691 --> 00:21:49.611 in the oscillation is information is being passed across the network but these 00:21:49.611 --> 00:21:53.811 other decoupled ones are still processing information perhaps and passing it 00:21:53.811 --> 00:21:58.031 around and i'm just not clear why that wouldn't be important or. 00:21:58.411 --> 00:22:01.431 I'm not saying it's not important i mean i i i'm very 00:22:01.431 --> 00:22:04.431 much um uh you know against the idea 00:22:04.431 --> 00:22:07.391 of of saying there's one mechanism to 00:22:07.391 --> 00:22:10.551 rule them all in the brain because everything i mean 00:22:10.551 --> 00:22:13.231 i i come to the brain as an engineer and i'm 00:22:13.231 --> 00:22:16.111 constantly trying to do that and that's because that's 00:22:16.111 --> 00:22:19.131 my background and training you want some kind of set of engineering principles that 00:22:19.131 --> 00:22:22.171 you want to squeeze the brain into and i'm constantly being 00:22:22.171 --> 00:22:24.851 brought up short and made to realize that no in fact what you 00:22:24.851 --> 00:22:27.591 thought was a principle is only you know is only a 00:22:27.591 --> 00:22:30.251 statistical tendency and in fact there's much more of a 00:22:30.251 --> 00:22:33.551 mess there so uh so i think this is one mechanism that 00:22:33.551 --> 00:22:36.191 may be quite important you know in 00:22:36.191 --> 00:22:39.111 the brain but I wouldn't want to say it's the only one on by any 00:22:39.111 --> 00:22:42.091 means right I was just wondering if there's some kind of thing about clock speed 00:22:42.091 --> 00:22:46.211 or whatever going on in the brain here that you think might be interesting oh 00:22:46.211 --> 00:22:50.591 well I think there is actually as as Paul was saying earlier on there is actually 00:22:50.591 --> 00:22:55.571 an inherent tendency for these systems to enter oscillatory regimes when they're 00:22:55.571 --> 00:23:00.231 driven and if you when you build models you see this pretty quickly you build a 00:23:00.471 --> 00:23:02.591 model, say, with spiking neurons. 00:23:02.931 --> 00:23:09.671 And there's a lot of excitatory connections there, and you drive it, 00:23:09.731 --> 00:23:11.471 it's going to naturally start to oscillate. 00:23:11.551 --> 00:23:17.171 And if you've got broadly biologically accurate parameters for your spiking 00:23:17.171 --> 00:23:20.991 neurons, it does tend to oscillate in around about the gamma frequency. 00:23:21.211 --> 00:23:27.691 It just does that. So that probably falls a little bit out of the neurobiology, 00:23:27.691 --> 00:23:33.951 out of the biology and then I suspect that the brain is going to fight I think 00:23:33.951 --> 00:23:34.871 whatever rich technology. 00:23:35.384 --> 00:23:38.264 Possibilities for dynamics you've got in the brain the brain 00:23:38.264 --> 00:23:41.464 and evolution are going to find some way of exploiting them so so 00:23:41.464 --> 00:23:44.764 you've got these oscillating things going on and 00:23:44.764 --> 00:23:47.904 the brain is going to find a way of exploiting these potential the potential 00:23:47.904 --> 00:23:51.144 of this and i think one way is is that it facilitates 00:23:51.144 --> 00:23:55.264 competition and cooperation among anatomically 00:23:55.264 --> 00:23:58.004 distributed populations that raises on the next question right whether these 00:23:58.004 --> 00:24:01.104 chimera states would also just drop out 00:24:01.104 --> 00:24:04.244 of the kind of local circuitry of 00:24:04.244 --> 00:24:07.264 cortical networks in the same way gamma just you get 00:24:07.264 --> 00:24:10.064 it for free or whether something else is required 00:24:10.064 --> 00:24:15.744 for that yeah yeah yeah um that's a very good question i suspect that they probably 00:24:15.744 --> 00:24:22.124 do drop out but i would uh but i would that would that's a very good question 00:24:22.124 --> 00:24:27.044 it would be good to construct models that prove that prove that i mean one reason 00:24:27.044 --> 00:24:28.564 why they might drop out is is because, 00:24:28.664 --> 00:24:34.084 I mean, another thing I've noticed is that in competitive mechanisms, 00:24:34.224 --> 00:24:38.084 you try and build some kind of competitive winner-takes-all mechanism. 00:24:38.224 --> 00:24:41.444 I mean, Tony might well know about this as well. But you've got to build it 00:24:41.444 --> 00:24:44.624 in a very particular kind of way to make the winner persist, 00:24:44.924 --> 00:24:46.644 really persist indefinitely. 00:24:46.864 --> 00:24:49.864 And it doesn't take much, especially if you've got small populations of neurons, 00:24:49.944 --> 00:24:54.844 for a bit of statistical variation to flip the winner to the other one, 00:24:55.244 --> 00:24:57.324 if you've got two, you know, you've got two rivals. 00:24:58.224 --> 00:25:01.444 And so that's, again, I think is a natural property. 00:25:01.584 --> 00:25:10.204 So that does lead pretty naturally to these kind of chimera states and flipping between them. 00:25:10.564 --> 00:25:15.024 But now the next thing you did is essentially you took like the state of the 00:25:15.024 --> 00:25:21.244 art, if you want, on human brain connectivity, like the human connectome and 00:25:21.244 --> 00:25:22.564 its different statistical properties. 00:25:22.564 --> 00:25:26.344 And we had an interesting discussion about how to interpret that this morning, 00:25:26.444 --> 00:25:28.124 but we don't have to get into that. 00:25:30.424 --> 00:25:37.744 If you now combine these dynamics and the notions of criticality with our understanding 00:25:37.744 --> 00:25:41.164 of the human brain, right? So you build these kinds of models. What happened? 00:25:42.144 --> 00:25:49.504 Well, so the credit really has to go to the people from here in Barcelona who did this first of all. 00:25:49.724 --> 00:25:54.924 Gustavo Deco's group and Joanna Cabral, who built a model based on the Hagman 00:25:54.924 --> 00:25:58.104 dataset, where they used the Hagman connectivity dataset. 00:25:58.104 --> 00:26:02.284 Set and basically put one of these oscillators, Kuramoto oscillators of the 00:26:02.284 --> 00:26:06.844 type that I was playing with, one on each of the nodes in Hagman's network. 00:26:07.184 --> 00:26:11.124 So it's very, very similar to the thing that I built, but the thing that I built 00:26:11.124 --> 00:26:13.884 used a completely synthetic network. 00:26:14.144 --> 00:26:19.224 It was a synthetic network that had some brain-like features such as modularity, 00:26:19.304 --> 00:26:21.804 but it was just a little constructed by an algorithm. 00:26:22.024 --> 00:26:26.904 Now, they did it with a real DTI-based dataset. 00:26:27.644 --> 00:26:33.084 Then the interesting thing was that they showed that they were able to produce 00:26:33.084 --> 00:26:39.784 a strong correlation with resting-state fMRI data. 00:26:41.504 --> 00:26:46.044 I have to say that it did rather surprise me, and it pleased me enormously that 00:26:46.124 --> 00:26:50.324 this only happens when it's in this metastable chimera state, 00:26:51.304 --> 00:26:53.324 exhibiting this metastable chimera state dynamics. 00:26:53.644 --> 00:26:59.544 So the kind of dynamics that I had found in my model, and nobody really had 00:26:59.544 --> 00:27:02.784 published anything showing that Kuramoto models could produce this. 00:27:02.884 --> 00:27:05.164 Although, as I say, I think the physicists really knew it. They just didn't 00:27:05.164 --> 00:27:05.904 think it was interesting enough. 00:27:06.164 --> 00:27:09.984 But the amazing thing was just to find that somebody could show that it could 00:27:09.984 --> 00:27:13.204 be used to model real data, and only if it was in that regime. 00:27:13.204 --> 00:27:18.444 I would like to push you on that a little bit more, because you could argue, 00:27:18.504 --> 00:27:22.324 look, to get to something like the dynamics of a resting state network. 00:27:23.538 --> 00:27:27.838 The key thing is that you get your connectivity defined. So then in this case, 00:27:27.858 --> 00:27:32.318 you take this notion of functional connectivity, which is a statistical structure of interactions. 00:27:33.978 --> 00:27:39.918 You exploit that to seed the topology of your network. 00:27:40.238 --> 00:27:42.178 But in this case, it's structural connectivity. 00:27:42.518 --> 00:27:45.358 So they were using a structural connectivity. So you have your structural connectivity, 00:27:45.658 --> 00:27:48.538 and you start to drive it now. 00:27:49.318 --> 00:27:54.538 Couldn't you argue that any kind of activity model, Even a rate-based model, 00:27:54.698 --> 00:27:57.318 a linear threshold unit, 00:27:57.538 --> 00:28:02.298 would be able to give you a resting state kind of pattern because the secret 00:28:02.298 --> 00:28:04.978 is in the connectivity and not in the dynamics of the nodes. 00:28:04.978 --> 00:28:07.278 Yeah, that may well be the case. 00:28:07.398 --> 00:28:12.538 And indeed, other people, including, again, Gustavo Deco's group, 00:28:12.598 --> 00:28:15.758 have used different kinds of dynamical models. 00:28:17.318 --> 00:28:21.038 But few of them are simpler than this Kuramoto model. 00:28:21.318 --> 00:28:28.138 I mean, it's about as simple as you can get and produce interesting kinds of results, I think. 00:28:28.218 --> 00:28:31.058 So that's what strikes me. so the the data 00:28:31.058 --> 00:28:33.858 set that you're looking at there is fmri which of course 00:28:33.858 --> 00:28:36.798 doesn't really have the temporal resolution to look at 00:28:36.798 --> 00:28:39.538 things like oscillations so i mean can 00:28:39.538 --> 00:28:43.578 you impact unpack for us a little bit how 00:28:43.578 --> 00:28:46.338 that data really validates the model because obviously you're looking 00:28:46.338 --> 00:28:49.178 indirectly for evidence for the kind of states that 00:28:49.178 --> 00:28:52.118 your model is creating yeah so so so 00:28:52.118 --> 00:28:55.478 we're looking at so in this kind of this kind of work um 00:28:55.478 --> 00:28:58.298 you're looking at how the you know the model over 00:28:58.298 --> 00:29:01.438 a long run over quite a long run so the 00:29:01.438 --> 00:29:04.258 model itself is oscillating at a gamma a gamma 00:29:04.258 --> 00:29:07.858 frequency but it's producing a phenomena at 00:29:07.858 --> 00:29:11.158 a much slower um you know it's producing dynamical 00:29:11.158 --> 00:29:14.238 phenomena at a much slower rate as well where where large you 00:29:14.238 --> 00:29:18.178 know groups will synchronize for a while and then that synchrony will then go 00:29:18.178 --> 00:29:23.198 away so that is happening at a much slower rate so there's a slower dynamics 00:29:23.198 --> 00:29:27.058 which you can then look so there's slower dynamics yeah so so so so the actual 00:29:27.058 --> 00:29:31.158 model is constructed at the level of this faster dynamics, 00:29:31.338 --> 00:29:33.398 but the results that you produce. 00:29:34.158 --> 00:29:37.498 Are at the level of the slower dynamics. So that's where you're making the matches. 00:29:38.257 --> 00:29:42.837 And what are the phenomena at this slower rate that you're matching on? 00:29:43.157 --> 00:29:47.397 So, well, you'd have to ask my co-authors exactly what's going on. 00:29:47.397 --> 00:29:48.597 But there's an issue here, right? 00:29:48.617 --> 00:29:52.177 Because, as you know, there's quite a discussion again now about what are we 00:29:52.177 --> 00:29:53.857 really measuring with fMRI. Yeah, sure. 00:29:54.077 --> 00:29:57.077 And there's some correlation with blood flow. 00:29:57.197 --> 00:30:00.557 But actually, the link to neural activity is still being debated. 00:30:00.557 --> 00:30:05.977 So now, at best, with your oscillator model, you capture, let's say, 00:30:06.037 --> 00:30:11.197 the hemodynamics that are sort of indirectly reflecting something neural that 00:30:11.197 --> 00:30:12.677 we don't really understand yet. 00:30:13.337 --> 00:30:16.677 On the other hand, the origin of your model was neurophysiology, 00:30:16.937 --> 00:30:21.217 a very different level of description, both spatially and temporally, 00:30:21.257 --> 00:30:24.317 because now we're talking about single cells doing things. 00:30:24.317 --> 00:30:27.377 Things right so isn't it quite a stretch to actually 00:30:27.377 --> 00:30:33.117 first say ah and i capture pascal frees's synchronization gamma range communication 00:30:33.117 --> 00:30:38.217 data and explain it yeah and on top of that i also give you now here the whole 00:30:38.217 --> 00:30:41.877 human fmri yeah well actually the origin of the model is a little bit higher 00:30:41.877 --> 00:30:46.797 level than that so you're really talking about populations of of neurons so if you're, 00:30:47.357 --> 00:30:50.477 so what you what you might maintain that one oscillator 00:30:50.477 --> 00:30:53.217 represents would be the activity of a quite a 00:30:53.217 --> 00:30:56.537 large population of of neurons exhibiting a 00:30:56.537 --> 00:30:59.437 gamma type uh uh bit uh you 00:30:59.437 --> 00:31:02.137 know dynamics right but but uh you know i have to say 00:31:02.137 --> 00:31:06.737 i have to i completely agree with you i i i it came as a total surprise to me 00:31:06.737 --> 00:31:10.997 that you could use this kind of model in this sort of way and that it would 00:31:10.997 --> 00:31:17.737 actually um uh match uh the data you know um so but your closest link with the 00:31:17.737 --> 00:31:19.597 data is more at this fMRI level. 00:31:20.277 --> 00:31:24.177 Than at the neurophysiological level where we look at direct cell... 00:31:24.177 --> 00:31:28.637 Well, indeed, but that's only because those are the only experiments that people 00:31:28.637 --> 00:31:32.357 have, where they've tried to make the match, as far as I'm aware. 00:31:32.777 --> 00:31:36.997 I'd be very interested indeed to do it at a lower level. 00:31:38.337 --> 00:31:41.657 Right, but now the other thing is that if you go for the whole, 00:31:41.717 --> 00:31:46.417 let's say, populations, where we talk about, let's say, cubic millimeters of 00:31:46.417 --> 00:31:51.077 cell volume that we're measuring with fMRI. 00:31:52.537 --> 00:31:57.177 So if we describe that as an oscillator model, then what we're capturing is 00:31:57.177 --> 00:32:00.017 the activity of millions of neurons of very heterogeneous types. 00:32:03.589 --> 00:32:07.569 Also, the dynamics of these neurons might be much more heterogeneous than your model captures. 00:32:07.709 --> 00:32:11.709 Absolutely, yeah. Like if you look at gamma, and say, okay, gamma is locally driven, right? 00:32:11.749 --> 00:32:14.969 You might actually have all sorts of subpopulations of cells and all sorts of 00:32:14.969 --> 00:32:20.849 complex dynamical relations, metastable or not, that you're completely blind to. Yeah, absolutely. 00:32:21.249 --> 00:32:24.529 So how are we going to cross that bridge now? Well, I think by, 00:32:24.669 --> 00:32:29.429 I mean, as I said quite early in my talk today, that I did actually start off 00:32:29.429 --> 00:32:34.649 with spiking neuron models and showed how the spiking neuron models produced 00:32:34.649 --> 00:32:37.249 a kind of dynamics that I was interested in, 00:32:37.289 --> 00:32:40.009 or complex dynamics that I was interested in. 00:32:40.089 --> 00:32:44.229 And people were saying to me, oh, why do you need all the complexity of these spiking neurons? 00:32:44.369 --> 00:32:48.149 Why not do it with a simpler model? So I thought, okay, I'll give that a go. 00:32:48.309 --> 00:32:53.589 And so I moved, I went up a layer of abstraction and moved to these oscillator 00:32:53.589 --> 00:32:57.389 models and kind of got drawn into that because it's a whole world in itself. 00:32:57.949 --> 00:33:03.489 But I do very much think it's a good idea to go back down again and to build 00:33:03.489 --> 00:33:06.829 larger scale spiking models than the ones I was doing before. 00:33:07.009 --> 00:33:10.169 And I have a PhD student, David Baumig, who's done a lot of work. 00:33:10.389 --> 00:33:13.989 And there's a PLOS paper last year that describes exactly that. 00:33:14.069 --> 00:33:17.669 So you've got a much richer variety of phenomena, in fact, with lots of different 00:33:17.669 --> 00:33:21.769 frequencies interrelating in different ways. Right, exactly. 00:33:22.009 --> 00:33:26.169 Is there some possibility that you might see metastability at different spatial 00:33:26.169 --> 00:33:28.189 scales in the brain? Oh, absolutely. 00:33:28.469 --> 00:33:31.169 In fact, I would be astonished if you didn't. 00:33:31.849 --> 00:33:36.889 Right. So here we are. You still have a model that's actually pretty powerful 00:33:36.889 --> 00:33:41.249 in capturing data and describing and interpreting some of this data on the brain. 00:33:41.409 --> 00:33:45.329 And now you're also applying that to sort of pretty extreme states in which 00:33:45.329 --> 00:33:49.629 we can push your brain like under a cello bison. 00:33:50.089 --> 00:33:52.589 Celocibin, yeah. Celocibin, sorry. Yeah. 00:33:54.263 --> 00:33:58.343 So how has the model helped you to understand the dynamics that you would induce 00:33:58.343 --> 00:34:00.423 with magic mushrooms? Yeah. 00:34:01.183 --> 00:34:04.223 Can I give a little bit of background there? 00:34:04.403 --> 00:34:09.323 One of my colleagues at Imperial College, Robin Carr-Harris, 00:34:09.443 --> 00:34:15.303 has done some pioneering work with psilocybin, which is the active ingredient of magic mushrooms, 00:34:15.583 --> 00:34:21.363 where subjects are administered with a psilocybin preparation, 00:34:21.503 --> 00:34:25.263 and then you do functional MRI on the subjects. 00:34:25.263 --> 00:34:35.363 And you can interpret the results as an increase in metastability, 00:34:35.723 --> 00:34:45.643 which in turn you can interpret in terms of a larger variety of states being produced, 00:34:45.963 --> 00:34:50.743 flipping between a larger variety of states. 00:34:50.743 --> 00:34:54.883 So you can kind of explain some of these results, some of the dynamics that 00:34:54.883 --> 00:35:00.303 you see under psilocybin as moving to a more extreme version of the metastable 00:35:00.303 --> 00:35:03.003 regime that was identified in this model. 00:35:03.843 --> 00:35:10.223 And Robin Carhart-Harris is quite amenable to this kind of interpretation, 00:35:10.423 --> 00:35:14.443 and we had a joint paper in Frontiers describing that kind of interpretation. 00:35:14.443 --> 00:35:17.983 Interpretation and um um so but 00:35:17.983 --> 00:35:20.663 i think the uh i think a lot of and so we're trying to 00:35:20.663 --> 00:35:24.403 apply the model to that as well but i think that it's it's very much work in 00:35:24.403 --> 00:35:28.523 progress and we have to in fact i think the application of this whole methodology 00:35:28.523 --> 00:35:33.763 is very much a work in progress i mean we have to uh uh you know um refine it 00:35:33.763 --> 00:35:38.943 a good deal before it becomes really accepted okay you said before 00:35:39.063 --> 00:35:43.163 you wanted to bring this back to behavior and one of the ways that i expect 00:35:43.163 --> 00:35:47.183 you will do that is through the work that you've done on on cognitive architecture 00:35:47.183 --> 00:35:52.623 which is thinking much more about function of brain circuits particularly cortical 00:35:52.623 --> 00:35:57.283 circuits so could you say a bit more about first of all. 00:35:57.945 --> 00:36:01.705 The ideas of cognitive architecture that you're interested in and how that might 00:36:01.705 --> 00:36:03.545 mesh with this work on oscillators. 00:36:03.765 --> 00:36:07.905 Yeah. Well, one of the really fundamental problems that I've been interested 00:36:07.905 --> 00:36:16.405 in for a while is how it is that a novel coalition of processes can form, 00:36:16.745 --> 00:36:18.225 can sort of coalesce out of, 00:36:18.825 --> 00:36:23.845 nowhere to deal with a situation that's never been encountered before. 00:36:24.345 --> 00:36:30.945 And that's clearly, Clearly, I think that's clearly at the root of human level cognition. 00:36:31.225 --> 00:36:36.985 Well, actually, maybe even some animals, some non-human animals can do this as well. 00:36:37.805 --> 00:36:46.065 And so I see the rich dynamics that I've been talking about here as a way of 00:36:46.065 --> 00:36:47.345 maybe addressing that kind of problem. 00:36:47.505 --> 00:36:51.365 And it also relates to the kind of cognitive architecture that I've been interested in. 00:36:52.465 --> 00:36:56.265 So I've been interested in global workspace architectures, 00:36:56.285 --> 00:37:00.045 you can relate global workspace architectures to a certain kind of connectivity 00:37:00.045 --> 00:37:06.765 in the brain where you have a connective core of hub regions which perhaps can 00:37:06.765 --> 00:37:10.125 facilitate coalition formation, 00:37:10.405 --> 00:37:14.885 in particular the formation of novel coalitions to deal with a new kind of situation. 00:37:15.205 --> 00:37:19.825 So it's this formation of novel coalitions which I 00:37:19.825 --> 00:37:22.565 i think is quite difficult to account for and that's the kind of thing 00:37:22.565 --> 00:37:25.285 that i'm that ultimately i'd really like to be able 00:37:25.285 --> 00:37:28.245 to model so that the connective core that you're talking about 00:37:28.245 --> 00:37:30.945 there is is the bit that's actually active when we're 00:37:30.945 --> 00:37:33.865 not doing anything which has you know external where 00:37:33.865 --> 00:37:36.585 we seem to be thinking rather than attending yeah and is 00:37:36.585 --> 00:37:41.705 that right so what kinds of activities that humans do do you think we're going 00:37:41.705 --> 00:37:45.385 to understand better through this well it probably Probably the kinds of activities 00:37:45.385 --> 00:37:51.225 where you're confronted with a novel situation and you have to pause for a second 00:37:51.225 --> 00:37:56.045 and sit back and think and stare at it and then suddenly, aha, 00:37:56.385 --> 00:38:01.525 a solution has come as if from nowhere. 00:38:01.525 --> 00:38:14.045 And so I think that probably is engaging a sort of mode, a dynamical mode, that enables. 00:38:15.945 --> 00:38:19.785 Processes to talk to each other and form a novel coalition that otherwise they 00:38:19.785 --> 00:38:20.685 wouldn't have been able to do. 00:38:20.865 --> 00:38:27.265 And I suspect that they can do this via these sort of centralized regions in 00:38:27.265 --> 00:38:32.685 the human human brain but then we still don't have behavior so what's it we have a kind of a. 00:38:33.729 --> 00:38:36.849 An unobservable behavior so you've got reflection so 00:38:36.849 --> 00:38:40.009 that's the point we want to get to what's the overt behavior 00:38:40.009 --> 00:38:42.969 we're going to then explain with that right yeah well i mean uh so 00:38:42.969 --> 00:38:45.969 so i i say you know i was caricaturing a 00:38:45.969 --> 00:38:49.049 little a little bit by saying you take you sit back and you say 00:38:49.049 --> 00:38:51.929 aha but i think if i mean you've probably seen 00:38:51.929 --> 00:38:55.589 this famous uh video of betty the crow uh bending 00:38:55.589 --> 00:38:59.089 the hook for the first time to retrieve the bucket and 00:38:59.089 --> 00:39:01.909 i you know of course you can ask questions about 00:39:01.909 --> 00:39:04.629 that experiment because it's hard to reproduce that kind of experiment but 00:39:04.629 --> 00:39:07.909 if we take that that bit of behavior at face value i think 00:39:07.909 --> 00:39:10.669 something interesting is going on there or when 00:39:10.669 --> 00:39:14.589 children solve the same problem because you can reproduce that that they'll 00:39:14.589 --> 00:39:18.129 they'll play around they'll play around and maybe there's a you know maybe you'll 00:39:18.129 --> 00:39:21.369 see this physically but maybe i'm not maybe you won't but there'll be a little 00:39:21.369 --> 00:39:26.389 moment when uh when their brain is going to when maybe they actually pause but 00:39:26.389 --> 00:39:29.329 their brain is going to go into a um a mode where. 00:39:30.029 --> 00:39:37.089 Where a new combination of processes can come together and then it's going to lock onto that. 00:39:37.509 --> 00:39:40.669 Somehow the brain, as it were, knows that that's the right solution. 00:39:40.689 --> 00:39:44.229 That's the sort of aha moment. It locks onto that possibility and of course 00:39:44.229 --> 00:39:45.369 it's going to do it straight away. 00:39:46.009 --> 00:39:53.329 And that phenomenon in the brain, I think, is key, absolutely key to understanding 00:39:53.329 --> 00:39:55.209 human-level cognition. 00:39:55.689 --> 00:40:00.129 I think it's as important as insight. It's as important as language, 00:40:00.329 --> 00:40:04.229 I think, and if not more important than language. 00:40:04.369 --> 00:40:09.189 And that's what I'd really like to be able to understand and capture in a model. Okay. 00:40:09.909 --> 00:40:16.029 So then, Murray, being an engineer and trying to understand the brain using 00:40:16.029 --> 00:40:21.889 these kinds of models and actually covering quite a range of phenomena, phenomena, 00:40:22.009 --> 00:40:29.109 what would be Murray's law that we have to adhere to to study the brain? 00:40:30.084 --> 00:40:38.464 I don't think i'm entitled to coin you are now you're now uh i think i think it would be, 00:40:39.204 --> 00:40:42.284 um play one word law 00:40:42.284 --> 00:40:45.564 play play with your models and make 00:40:45.564 --> 00:40:48.804 your models play okay i think playfulness in 00:40:48.804 --> 00:40:51.744 in as as a scientist uh and engineer 00:40:51.744 --> 00:40:54.644 um and playfulness in the models they're 00:40:54.644 --> 00:40:58.964 in a sense they're the same thing right we're being creative and 00:40:58.964 --> 00:41:02.144 create the root of creativity and the creative things 00:41:02.144 --> 00:41:05.604 we want to create is playfulness now the 00:41:05.604 --> 00:41:08.464 other thing is that tony and i trying to control the 00:41:08.464 --> 00:41:14.464 future so um that's why we have tony check the predictions people make and since 00:41:14.464 --> 00:41:18.164 he can just take a train from sheffield to london so we can save money that 00:41:18.164 --> 00:41:21.724 way the question is four years from now tony's tony's going to come visit your 00:41:21.724 --> 00:41:24.944 lab there at imperial and he's going to say look Look, you know, 00:41:24.964 --> 00:41:26.544 four years ago on this podcast interview, 00:41:26.724 --> 00:41:29.664 you made this prediction to them checking whether it was validated. 00:41:29.924 --> 00:41:33.024 What's the one prediction, concrete prediction you could make? 00:41:33.164 --> 00:41:37.444 I think it's that I'll have no more funding to address this issue than I do now. 00:41:40.884 --> 00:41:45.364 We hope that isn't true. All right, Ray Shannon, thank you very much for this conversation. 00:41:48.764 --> 00:41:54.604 I was queued up for that question. The CSN podcast was produced by the Convergent 00:41:54.604 --> 00:42:00.724 Science Network of Biometrics and Biohybrid Systems, a project funded by the 00:42:00.724 --> 00:42:03.324 European Sevens Research Framework Program. 00:42:04.904 --> 00:42:10.204 For more interviews, recorded lectures, or upcoming conferences in the field 00:42:10.204 --> 00:42:16.444 of biometrics and biohybrid systems, go to csnnetwork.eu. 00:42:16.400 --> 00:42:25.360 Music.