Support the show to get full episodes and join the Discord community.
Michael L. Anderson is a professor at the Rotman Institute of Philosophy, at Western University. His book, After Phrenology: Neural Reuse and the Interactive Brain, calls for a re-conceptualization of how we understand and study brains and minds. Neural reuse is the phenomenon that any given brain area is active for multiple cognitive functions, and partners with different sets of brain areas to carry out different cognitive functions. We discuss the implications for this, and other topics in Michael’s research and the book, like evolution, embodied cognition, and Gibsonian perception. Michael also fields guest questions from John Krakauer and Alex Gomez-Marin, about representations and metaphysics, respectively.
- Michael’s website.
- Twitter: @mljanderson.
- Related papers
0:00 – Intro
3:02 – After Phrenology
13:18 – Typical neuroscience experiment
16:29 – Neural reuse
18:37 – 4E cognition and representations
22:48 – John Krakauer question
27:38 – Gibsonian perception
36:17 – Autoencoders without representations
49:22 – Pluralism
52:42 – Alex Gomez-Marin question – metaphysics
1:01:26 – Stimulus-response historical neuroscience
1:10:59 – After Phrenology influence
1:19:24 – Origins of neural reuse
1:35:25 – The way forward
Michael 00:00:03 That’s a really important thing to notice. And what you see instead is that each bit of the brain is involved in multiple cognitive functions across multiple categories of cognitive function. Um, and, and now you’ve got a very different thing to explain. Hm. It’s a mistake to suppose that just because you can get a neural network to do something that that means it’s a hierarchical representational model, blah, blah, blah, in the brain that does something similar. There’s vast mystery here that remains for anybody in this view. Right? And if you deny the mystery, you’re, you’re simply closing your eyes to the reality of where we are in the science.
Speaker 3 00:00:50 This is brain inspired.
Paul 00:01:04 <silence> Hello, it’s Paul. And that was Michael Anderson, a professor at the Rotman Institute of Philosophy at Western University. In 2014, Michael released his book After Phenology, Neural Reuse, and the Interactive Brain, which among many other things, the idea of neural reuse, roughly. Uh, that is that any given brain area is involved in multiple different cognitive functions, and different cognitive functions happen as a result of the coordinated activity among different sets of brain areas. So, a given brain area might be used during one cognitive function, like attention, for example, and have a set of partnering brain areas to implement attention. And that same brain area, uh, is reused during another cognitive function, like, say, working memory, but have a different set of partnering areas to implement working memory in the book and related works. Michael details plenty of evidence for neural reuse, and he argues that this is strong evidence against the modular computational view of the brain that still dominates cognitive neuroscience, even though we got rid of that, uh, phenology kind of explanation, uh, or view long ago.
Paul 00:02:18 And he also writes, uh, about how it relates to evolution, inactive and embodied cognition, and many related topics. So we talk about all that, and Michael takes a couple guest questions from friends of the podcast, uh, John Crackower and Alex Gomez Madine, which deal with mental representations and metaphysical views of the mind. So you can learn more about Michael in the show notes at brand inspired.co/podcast/ 1 52. If you like this podcast, consider supporting it on Patreon where you can join our Discord community and get all the full episodes. So, for example, in this episode, we end up talking about, uh, language and transformer models during that extra time. So thank you to all my Patreon supporters, and thank you for listening. All right. Here’s Michael.
Paul 00:03:07 Michael, I’m, I’m gonna actually start by reading the last paragraph, uh, from your 2014 book after phonology, before the appendix where you list a bunch of open questions. Um, the official end of the book, I suppose. So here, here’s that, uh, last paragraph, and then we’ll rewind and talk about what it’s about, et cetera. You wrote, I very much look forward to walking the path that lies ahead. This is, I believe, the most exciting time in the history of the neurosciences. A time when major technological advances are allowing us to measure and analyze brain function in ways rit of even just a few years ago. If we can manage to sustain a level of conceptual progress that keeps pace with the rapid march of technology, then the future of our science is astonishingly bright. And then you ask, Shall we get to work? So <laugh>. So, um, this is, again, from your, I would say it’s a, like a modern neuroscience classic after phenology. I see it cited so much. Would you accept that it’s a, would you call it a classic?
Michael 00:04:11 Well, I hope so. It’s, it, it was, it was a monumental effort on my part, and it does seem to have been, uh, usefully influential in a lot of, in a lot of, uh, uh, corners of the discipline. And, uh, yeah. And, and certainly it’s gotten high praise in various places. So, um, you know, the word classic is a bit, it feels a bit much,
Paul 00:04:35 Um, okay, well, you’re maybe being humble,
Michael 00:04:38 Maybe. Um, but no, I, but I do think that, you know, there’s, there’s a lot there. And, and I also think that, um, although it has been influential in various ways across various, uh, parts of the discipline, it, the, I think, if I’m being honest, I think the depth of that work has yet to be fully understood.
Paul 00:05:03 Yeah.
Michael 00:05:05 And so, so there is definitely, as it were, as I, as I allude to, as you allude to in that paragraph you read, there is more conceptual work to be done. And I think the book can help. It still has work to do to help with that conceptual work that needs to, needs to unfold in the neurosciences.
Paul 00:05:21 I
Michael 00:05:22 Mean, uh, in, so I have a cognitive science reading group here at Western, and we read a very nice article, and it exemplifies exactly what I was, uh, pointing at in that paragraph, which is, it introduces a brilliant new method of measuring interest, objective correlations and neurodynamics, um, uh, it just really great stuff technologically. And then in the discussion, it falls back on localization of function because, and, and the, their method does not naturally match with that approach to interpretation. And so there’s a, there’s a gap there that, that needs filling. And I think, you know, Aerology has part of the back fill, right? The conceptual apparatus you need to put in place to make full use to, to be able to fully appreciate what you’re learning from these advanced techniques. But clearly that part has not really registered yet. Hmm. Um, and so, so that’s, that’s the conceptual advances that still need to need to come, that need to be absorbed, that need to change.
Michael 00:06:33 Not technological practices. I mean, people much more brilliant, much more mathematically talented than I are pushing those frontiers every day, but they’re still doing it as if they’re working within a framework that we developed for the general linear model for the, for the, well, this thing, the amplitude of this change in this particular region indexes this particular part of the stimulus that’s a really flat footed and simplistic way of looking at brain function. And, um, and we have the conceptual tools to move beyond, but those things haven’t penetrated to the degree that they need to. Let’s put it that way. You know, that they need to, And I’m hopeful that, Look, it’s a young
Paul 00:07:19 Science. Well, yeah. So, okay, so you’re still hopeful. We’re gonna have to step back and, and just talk about the high level conceptual, um, basis and premise, uh, of, of your work, of your research agenda, because they’re actually multiple moving parts, Right. And then, but they all kind of fit together. Um, but before we do that, so, so you’re still hopeful, you’re still optimistic. They haven’t dragged you down reading the conclusions of a paper that, that fall back onto the, you know, um, the,
Michael 00:07:44 Uh, No, no, no. And this is what I said to my grad students who were in that, in that room with me yesterday, which is, look, this is exactly the opportunity you want, Right? Right. This is, you can say, Look, here’s a really interesting paper, right? Look at, look at the new method they introduce, Look at the cool findings, and then look at what they do when they interpret it.
Paul 00:08:07 What, what paper is this, by the way? So I’ll,
Michael 00:08:09 I don’t know. I, I think I, I don’t, I don’t wanna be critical in this way of, um,
Paul 00:08:14 But you, you must see this frequently in the literature.
Michael 00:08:16 I, Well, absolutely. And that’s, that’s, But this is, this is, Yes. Uh, you know, this is, you know, crisis opportunity, right? So, so this is like, okay, that’s great. Um, and, and this particular author I respect quite a bit, um, the senior author, you know, and, and he’s been one of the, the prominent, most prominent neuroscientists, really taking seriously the notion that we have to grapple towards a different conceptual framework.
Paul 00:08:40 Oh, man, you’re gonna have to make me guess now, who the author is. No. <laugh>. Maybe people can try and <crosstalk> you come
Michael 00:08:46 Up with it. Yeah. <laugh>. Um, and, uh, but so my, my, my message to the students was not, Oh, this is gross. We should walk away. No, this is, now let’s look at the success of this method, right? Mm-hmm. <affirmative>. Now, if you were to sort of, uh, op devise a, a view of brain function for which this method was the natural answer, what would that view look like? Like what induction can you do from the method, the success of this method to a conceptual framework for understanding the brain? That’s the gap that has to be filled. And it’s gonna take a lot of people doing a lot of thinking to fill it. But those gaps, when those gaps show up in literature, the last thing to do is to get frustrated by it. Instead, you say, Aha, look. Right. We can show, we can demonstrate.
Michael 00:09:42 It’s obvious there’s a gap here. So let’s start doing the conceptual work that would be needed to say, Okay, this, You know, so the general linear model, right? This is just straightforward, you know, basically statistical co variance. It was based on the notion that the brain is modular. It’s got dedicated regions that do a thing. And all you need to do is isolate that thing. And if you can do that and say, a display, a simple display or a simple task, you know, you can do it sort of in a visual display, or you can do it with a, with a kind of a so-called, so called pure task, right? A task that only activates one particular brain function. Um, if you can find that thing, if you can find that display, then when you, when you turn that thing on in the display, or you have the person do the task, you’ll see, ah, this is the bit of the brain that responds most strongly to that. Okay? Right? That’s
Paul 00:10:38 The, that’s the phenology
Michael 00:10:39 Part. That’s the phenology part. And if the brain were like that, and by the way, people, you know, came up with this notion that the brain, well, it’s a very old notion, hence my reference to phenology. But, but even the modern version of it, uh, which is sort of more computational and modular and whatnot. So it’s, it’s sort of, you know, the, the, the old, old dressed up in the new, um, the general linear model is a perfect fit for that notion of what, how the brain is organized, Right? That’s an adequate empirical approach to the brain. Cause it, it’s a, it allows you to discover the bits that appear to be dedicated to particular functions,
Paul 00:11:19 And you can get significantly statistically significant models, Right. That would support that sort of view as well.
Michael 00:11:28 Absolutely. So it wasn’t like it was a, it was a, a mistake or a dumb move or anything? Not at all. There was a, a view of, of, of function. Then people matched the, um, the, the, the, the, an analytic methods to that view of function. And of course, we’ve got 50 years of neuroscience based on this that has produced all kinds of interesting results. Now, you know, as, as you know, having read the book, you know, much of my, my career for the past 10, 15 years has been dedicated to showing that the data gathered under this assumption actually undermines the assumption. Mm-hmm.
Paul 00:12:06 <affirmative>,
Michael 00:12:07 If you look at it the right way. But that brings us back to this paper and the opening, which is that <laugh>, Okay. So now we’ve got this really brilliant new multivariate method that allows us to, you know, do not just individual, but interest, subjective measurements of synchrony, um, you know, under various conditions, really cool stuff. And then they fall back interpreting it as if they’re still within the glm, you know, modular view, even though they clearly reject that. And that’s the
Paul 00:12:37 Gap. Could, could this have been reviewers? Could this have been reviewers that shaped that, that, uh, outgoing, those outgoing remarks?
Michael 00:12:45 Oh, guaranteed. Yeah.
Paul 00:12:46 Yeah.
Michael 00:12:47 Okay. Yeah. So one of the stories I told in this meeting yesterday where we discussed this paper was that, um, I was on a, a master’s, uh, a neuroscience master’s defense recently, uh, in which, and in other, it’s really cool work, um, you know, pushing the envelope, methodologically, um, you know, trying to, maybe this is a little off topic, but, but basically the, the real innovation, So a typical experiment in, in neuroscience and cognitive neuroscience in particular, uh, is what we call open loop. That is, um, you have a stimulus, maybe you make a response to it, maybe it’s just passive viewing, as in like the movie paradigms that have become very popular mm-hmm. <affirmative>. Um, but in any case, whatever your response is, it doesn’t affect anything in the future,
Paul 00:13:36 Right?
Michael 00:13:38 So it’s open loop. The, the circuit, the usual, you know, sensory motor, sensory circuit is broken in these cases. We, we closed the loop. And so this is a, a student of mine and, uh, who’s, who’s, you know, now moved to her PhD program, and, and she managed to figure out a way to play Pacman in the scanner.
Paul 00:14:00 Okay. With, I mean, it’s kinda like just a joystick down by your side, right?
Michael 00:14:05 Yeah. I think with the track wall is what they ended up using just because of the, you know, what, what, either what we had or, or, because that was, you need less movement of the arm. Yeah. Right? So you’re wiggling the head less, and so it’s just a track ball anyway, Um, that doesn’t matter, I don’t think. But, um, but here, but the notion, the important notion here is that your actions change your perceptions inside the scanner.
Paul 00:14:33 This is the inactive viewpoint.
Michael 00:14:35 This is the inactive view, this is the closed loop view. This is, you know, this is, uh, you know, fixing Dewey’s reflex arc problem. Um, and, uh, and, you know, lo and behold, it took a lot of technological, uh, grit and perseverance and, and, and changing, but got really cool data that showed that, yeah, indeed, the brain looks very different in closed loop versus open loop, Oh, circumstances
Paul 00:14:59 Interesting.
Michael 00:14:59 To me, that was enough. Like, okay, we proved this could be done. This opens up a whole new avenue for research, um, in closed loop contexts. But the, some of the other examiners are like, Well, no, you have to interpret. Oh, like, what does that activation mean? I’m like, We’re not there yet. Have
Paul 00:15:17 You read my book? That’s
Michael 00:15:19 <laugh> <laugh>. They have not. Um, but, but, you know, so, so this notion that somehow, even though you’re, you’re pushing things in a very different direction than you’re operating under a very different set of assumptions. You’ve gotta turn around and interpret the local activations, and if you don’t do that, you’re not done. Mm-hmm. <affirmative> is pervasive. And I think it’s a, it’s a, it’s a massive mistake. Um, and, and if you’re gonna do it, you’ve gotta do it, honestly. So you’ve gotta go to something like neurons, you know, that giant database that, that tele or, uh, Yeah. Telecon put together and, and say, Okay, well, if you really wanna know what this activation means, well let’s, let’s put it into neurons and let’s see all the things that, uh, cause activation in that area, and now interpret it.
Paul 00:16:02 Hmm.
Michael 00:16:03 Don’t just interpret it as if it’s in the context, right. That you think you know about in, in the context of Pacman and what could be going on here. Um, let’s look at all the things that could be going on here that have been already observed to be going on there. And of course, nobody takes that step, right? They’re, they’re interpreting things in a vacuum. And I think that’s really unconstrained, and, and there’s no reason to do it. There’s interesting things to be learned without taking that step.
Paul 00:16:29 Well, um, when you take it in a vacuum though, I mean, you’re addressing the cognitive function, for instance, that you are particularly interested in, in your lab. Right? So, but what, what you’re saying in the neural, the neural reuse story, I guess this is a good time to bring it up, is that a particular region, um, can be active in multiple different contexts. Um, and that the way that regions interact with each other, there are different sets of regions that are interacting. And so one particular region, if it’s active during Pacman, doesn’t mean it’s the Pacman region,
Michael 00:17:05 Or, or, or the, or the ghost region, or the pellet region, Right. Or the Right. That’s just a flat footed way of looking at things. Um, and, uh, so that’s right. That, that’s, that’s, that’s why I say that, you know, if you analyze the, the stuff gathered under the GLM model, under the modular model, under the dedicated region model, if you look at all the data that’s been, then it’s, you know, just vast TROs of this data that have been gathered. Um, and then you, you, you go at the next level and you analyze it, you can show that the assumptions on under which the data were gathered get undermined by the data itself, right? Yeah. And that’s a really important thing to notice. And what you see instead is that each bit of the brain is involved in multiple cognitive functions across multiple categories of cognitive function.
Michael 00:17:56 Um, and, and now you’ve got a very different thing to explain. Hmm. Uh, and so one of the conclusions that I came to as a result of this, uh, kind of analysis as, as you just said, that differential function, differential behavior is not a matter matter of differential activation of brain regions. It’s a matter of differential cooperation across different brain regions, many of which will be the same across different kinds of behaviors. And again, the data supports this. Um, now that’s not a deductive argument, right? But that’s what the data, uh, show. And now you’ve gotta either question the data or question the methods, or if you want to get out from under that, that conclusion.
Paul 00:18:37 But the, So I wanted to ask you how you came to this view, because, I mean, was it something where you were looking at a bunch of data and then, you know, induced, uh, neural reuse and, and, and inactive, uh, neuroscience or, or, you know, is, was it more something from first principles, just thinking about the complexity and the way that, um, complex neural networks interact? Like how did you formulate the idea, the, these ideas over, you know, the few years leading up to, Cuz you’ve been working on this stuff a lot, you know, for a lot longer than you published before you published the book.
Michael 00:19:14 Yeah. Um, so, you know, I, I come out of, um, body cognition. I, I, I was working on body cognition before that was a thing.
Paul 00:19:23 <laugh>, is it a thing yet? I guess it’s a thing now, isn’t it? It’s,
Michael 00:19:27 It’s definitely a thing. It’s a thing now. It’s, it’s for sure a thing. You know, there’s, there’s interesting kinds of infighting and, you know, the in activists versus the embodied cognition versus the, uh, ecological psychologists,
Paul 00:19:38 You guys need to all get along, right? The four E’s they need to get along, right.
Michael 00:19:42 <laugh>, well, certainly, well, so, so certainly in activists and, uh, ecological psychologists need to get along. <laugh>, I think there’s no question about that. The question is what to do about the embodi cognition folks? And, and I did come out of antibody cognition and, and here’s the distinction. So both on activism and, um, ecological psychology, uh, don’t start with representations. Now, depending on who you are, you might think representations come in, come online in various ways later. You know what? Your pg e is like this, Um, uh, uh, but that’s, that’s a, that’s, that’s not really a, a dispute that’s worth, uh, you know, breaking up over. The difference between that and the embodied cognition theorists is the embodied cognition theorists are still within the computational framework where representations are central. Ah,
Michael 00:20:32 Right. The difference is that the way they understand the nature of the representations is quite different from, say, the a modal symbol theory, which was the, is the extreme of the computational view, but like the, a modal symbol theory of the, of the physical symbol systems hypothesis, for instance. Um, you know, instead, you know, Larry Barcelo is, is, uh, you know, James Martin, uh, people like this say, No, no, no. There are representations and they’re, they’re computed in, in kind of the standard way, but the content of those things are grounded in the body in various ways. Uh, and that, that could mean they’re grounded in visualization. It could mean they’re grounded in, in action. So Andy Clark is, is for me, you know, he’s one of my intellectual heroes, but we disagree on this point. We do now, we didn’t back in 1992, um, but we do now.
Michael 00:21:21 Um, and because the embodi cognition theorist stick to the representational theory in mind, it’s just they understand representation as grounded in the body or action oriented or, you know, there’s various stories you can tell about how to understand representations, uh, in an embodied sort of way. And I think that marks a really significant divide in the four E world, um, between the embodied folks in the inactive, um, and ecological folks. Uh, and that’s, that’s, that’s I think, a real divide worth fighting about. Whereas I think the differences between activism and ecological psychology, I think those are the interesting differences of focus, um, right. Where differences in starting point, but I think there’s really, uh, convergence in those views.
Paul 00:22:13 So I’m gonna, I’m gonna go ahead and play you the first guest question because you, you dropped the representation word there, and,
Michael 00:22:19 Okay. Do you want me to finish though? The, the, We’ll
Paul 00:22:22 It, we’ll come back to it. Okay. If that, if, if we can keep that thread totally fine.
Michael 00:22:26 Yep.
Paul 00:22:26 So this is from John Crackower and, um, Ah, yeah. <laugh>. Oh, yeah. That <laugh>, You’ll appreciate that.
Michael 00:22:33 No, I love his work. I, I just, Yeah, I can, I, I can probably just say the question myself.
Paul 00:22:38 Yeah. Oh, that’d be, that’d be awesome if you say the question. It’s actually kind, it’s kind of long. So, um, it’s, it’s almost two minutes. So, uh, so be, be a little patient, I suppose. Both, both my guest questions are a little bit long.
Speaker 6 00:22:51 Hi, Michael. Um, it’s John Crackow here. My question pertains to your recent article, uh, in reply to Russ Pra on representation. Um, where I think you touch on a number of your favorite topics, um, you know, including sort of Gibson in affordances versus representations. Um, and I’m extremely sympathetic to your take on artificial representations, neuro representations, and
Speaker 6 00:23:25 How they don’t really map onto mental representations, um, as positive by most people who believe in them. Um, and, you know, I also wrote a review of Nick Shea’s book, where I felt one didn’t even have to invoke representations way he does, or someone like, well, Sierra Perini does in motor cortex sensory cortex virtual stream. I don’t think the representation needs to be evoked at all. Um, it doesn’t solve Ramsey’s job description problem, um, or meet, I should say, uh, those criteria. But what I struggle with, with all the anti representational, this is when it comes to the conversation that you are now having with Paul, or, you know, true cognitive content, ideas, beliefs, um, locked in. Patients can write novels through eye blinks, electrodes can evoke forced thoughts in frontal cortex. Um, I simply do not see how we can do without the mental representation notion when it comes to those phenomena. Um, and I always feel like they get skirted around by sub personal processes, uh, in sensory and motor areas and other such things. Um, and citing d in 1981, in my view, does not get rid of the problem <laugh>. So I was wondering if you could just tell me how do you come up with a story about this question I’m answering you and how you are going to, asking you and how you are gonna answer it without having to invoke mental representations.
Paul 00:25:02 All right. Sorry, that was a mouthful and long, but I think you got it all right.
Michael 00:25:08 Yes. Yeah. And he, you know, this is all familiar territory.
Paul 00:25:13 Yep.
Michael 00:25:14 Um,
Paul 00:25:16 <laugh>,
Michael 00:25:17 So, well, no, a couple things I wanna say at the outset. So, I, I, I, by the way, I totally agree, uh, with John about Nick Shea’s book and Gal Arrow’s recent work, you know, they use this word representation, and that’s fine. You people can use it however they want. But the, the trouble with that word, one trouble with that word is that it’s got such an historic bit of a historical baggage that comes with it. And, and I think this is part of, you know, this is, I say this explicitly, uh, by the way, the, the other author on that paper. So it’s not just me, It’s Heather Champion, a grad student of mine. And so we say explicitly in that paper that, you know, neuroscientists use that word in such a way that they’re kind of borrowing the notion that they’re somehow speaking to mental representations.
Michael 00:26:09 Because that’s what eventually we do wanna, uh, understand in psychology. Mm-hmm. <affirmative>, that’s psychology. Right? At least, at least there’s, there’s those folks who say, psychology is the science of behavior. I’m just quoting textbook Right. Titles now. Right. But the other is, is science of the mind. Right? Let’s understand how that that bit works. And so, at least on the latter interpretation, that’s what we’re asking, right? How do beliefs and desires and intentions, um, how do ideas, uh, get expressed and communicated between people? He mentioned writing a novel. I don’t, I’m not, Anyway, writing a novel. Like, how does that happen without mentor representations? So the, the, the main point of the, the piece on PRA is to point out that the kinds of things that neuroscientists and I think he’s right to loop in, in, and, uh, and, and shay into this camp, the main thing that they’re pointing to aren’t the kinds of things that have an obvious relationship to mentor representations.
Michael 00:27:10 And if what we care about is mental representations, this kind of work, at least in its current incarnation, is not gonna get us there. It’s interesting in its own right. These are clearly in, you know, they’re part of the mechanism of behavior. These are in, you know, information carrying, um, you know, neural, uh, activations, uh, that clearly are, you know, involved in, um, in, in behavior guidance, in light of perceptual, uh, uh, phenomenon. All that seems right to me. But as I say in that article, that’s totally compatible with the Gibson take on all of this mm-hmm. <affirmative>, right? Even Gibson would never, and, and did not, in fact deny that there are neural representation that carry information about, uh, aspects of the
Paul 00:27:53 World. You, you should clarify what the Gibson take is for the audience, perhaps, uh, in a,
Michael 00:27:58 In a nutshell long story here, but no.
Paul 00:28:00 Yeah.
Michael 00:28:02 But the basic idea is that, um, so in the, on the classic view, uh, that’s just focused on visual perception cause that’s where Gibson focused. But the same case can be made in other sensory modalities and, and across different kinds of perceptual experiences. On the classic view, um, visual stimulation, uh, creates a two dimensional retinal image, which clearly cannot capture all the aspects of the actual world. So it’s impoverished with respect to the actual world. And therefore, um, you have this two dimensional, um, impoverished retinal image from which you have to reconstruct the, um, the physical properties and, and layout of the actual world. And so you need a whole bunch of what my friend Tony Humira calls mental gymnastics mm-hmm. <affirmative> to achieve that. And so you build a model, a speculative model of what the world might be like given the impressions that it’s putting on your retina.
Michael 00:29:14 Um, and, you know, and once you’re at, in, in that space in cognitive science, that just brings in all the old notions of model building and built in assumptions. And where do the assumptions come from? Are they evolutionarily derived? Uh, do they come from experience in some way? Right? And so you’ve got these assumptions that structure, it’s very NeoCon, right? You’ve got these assumptions that structure behavior that’s excuse that structure of the model, um, that allow you to interpret the incoming impulses from, from the retina and so on. Gibson just rejects all that. First of all, he says the retina image is a fiction. There’s nothing in the retina on which an image could be projected, Right? It’s a fiction that was derived from experiments, uh, where people dissected cow eyes and, and such things, uh, and put them in windows and notice that, Right. The light was refracted through the lens, and then if you put a piece of paper behind it, you got an upside down image of the, of the world.
Paul 00:30:11 Right. But we can decode based on the two dimensional retinal act activations, Right. Can
Michael 00:30:16 Decode that is, that’s the thing you need to decode. Yeah. And so there’s a whole kerfuffle, like, Oh my gosh, it’s upside down. How do we turn it right side up again,
Paul 00:30:25 <laugh>.
Michael 00:30:25 Yeah. Right. And, and that’s, that’s actually a question that it’s funny, right? But it’s actually a question that makes sense from within this, this, if you take this stance, that’s a real question. Yeah. Like, you know, the model’s upside down, you’ve gotta turn it back upside, right? Right side up somewhere somehow. Um, so Gibson just rejects all that, and he says, Look, first of all, there’s no retinal image. Um, furthermore, there’s no poverty of the stimulus, Right? Think about it from the standpoint and not of, of, of what he calls ecological optics or ecological physics, not from the standpoint of say, you know, Newtonian or Einsteinian physics, but, um, light in the environment bounces off of and is thus shaped by every surface in the environment. So at any point in an environment, with the exception of things that are fully occluded from incoming light, the light itself is structured in a way such that it carries information about all the stuff that it’s bounced off of. He calls this the optic array. And his claim is that the optic array in and of itself carries sufficient information about the layout of the environment. There’s no po there’s no lack of information there.
Paul 00:31:46 This is direct perception, right? Is that,
Michael 00:31:48 Yeah. This, this is, this is what leads up to the view of direct perception. Yeah. Now that, that’s, Well, maybe that’s too subtle a thing to, to care about. Yes. This is what leads to the view of direct perception. Um, and so the notion is that what you’re not doing is taking the impingement of the world in an istic way and using that to build a model. What you’re doing is you’re learning to coordinate your behavior with the structure that’s actually in the light. Right? And so, whether it’s upside down right set up, this is, this is not a relevant question anymore. There’s in sufficient information in the optic array to allow you to coordinate your behavior with respect to the environment. It specifies, as he says, the layout of the
Paul 00:32:33 Environment, right? Yeah.
Michael 00:32:35 And so there don’t need, this is the direct perception view. There don’t need to be epistemic mediators. They don’t need to be representations, They don’t need to be models. Right. That then you use to, you know, infer what you ought to do in the environment. You can directly perceive the layout of the environment by coordinating with the structure and the light. So that’s, that’s the Gibson view, and it’s a non-representational list view. It’s important to note, back to John’s question, It’s important to note that, uh, Gibson myself, um, he was talking about visual perception. Mm-hmm. <affirmative>, he wasn’t talking about novel writing. Right? Right.
Michael 00:33:16 Um, and so, and what follows from the Gibson view, I think, uh, at least if you generalize it to the kind of account you want to give of, you know, a cognition wri large, is that, um, there’s no necessity to base or to center your explanation of, of mental phenomena of, of, of, of, of successful behavior on representations, on models, Right? What you need to do is understand coordinative, um, uh, dynamics, coordinative structures, How, how does that process happen? And so, you know, um, uh, James Gibson’s wife, Eleanor Gibson, she’s, she took over the developmental side of this view. And, and, you know, part of her story and, and, and some of the people that she, uh, trained and worked with, um, uh, Karen Adol is one instance at nyu where what they’re understanding is what does that development look like? That that, by which you, you learn to coordinate with various kinds of aspects of your environment. What’s the information that’s available? How do you learn to coordinate with it? What’s that developmental trajectory look like? So those are the questions you start asking from the gian standpoint. Now, novel writing <laugh>. Um, so one thing, I I, you know, this is not meant to be a sidestep, cuz I’m not gonna sidestep, but I gotta tell you, I find questions like this really unfair.
Michael 00:34:50 Cause I want you to show me the cognitive psychologist that has explained novel writing. I don’t care how many representations they throw at it. Nobody understands what it takes to write a novel to be creative.
Paul 00:35:04 But I suppose John’s bet would be that, um, the path forward to explaining would be, uh, we would be better served with a representation viewpoint. Right? Um, but it’s just a more likely path. Well,
Michael 00:35:19 Yeah. So I, I just wanted to register my objection before I said something about that. Okay. Right, Because cause this is, but this is because honestly right, this is kind of the classic move that the folks within the tra the tradition take with respect to upstarts like ecological psychology. Oh, sure. That’s cool. For motor control. And in fact, Peck says this explicitly in that paper that I respond to, Oh yeah, that’s cool for motor control, um, you know, uh, and, and visually guided motor, uh, activity, but
Paul 00:35:52 Right. Need it for perceptions. But, but he also like in, in pulled direct’s paper, and I, I think one of the things that John is impressed with, and I’m of course impressed with, I’m sure you’re impressed with, uh, are artificial neural networks, Right. And that we can, you know, get some sort of behavior, uh, out of an artificial network. And presumably, Yeah. It’s through these, the representations in the nodes and weights of that artificial network, Right? So just throwing that out there.
Michael 00:36:18 So, so remind me of this. Cause I’ll come back to that after I, so well actually know what, it’s better to lead up this way. So, so I’ve got a paper hopefully coming out relatively soon, we gotta revise and resub bid on it, writing it with some collaborators on auto encoders print
Paul 00:36:37 Already that, um, Is this auto en
Michael 00:36:39 Oh, you might have seen it. Yeah, yeah, yeah, yeah, yeah.
Paul 00:36:41 This was auto encoders. Um, Oh, oh, this was a question I was gonna ask you how auto encoders are in line with the Gibson and Oh,
Michael 00:36:48 Is that, is that why that question came up? Yeah. Because that Preprint Yes. So it’s that, and, and, um, we revise it and resubmitted it and, and you know, we’ll see what the, hmm. <affirmative> what the verdict is there. But, but the basic, um, the basic conclusion of that paper is that you still don’t have to use representational talk. You can talk about adjustments to the structure of the inputs. Right? And so that’s, it’s at least compatible with a kind of gib sonian perspective, not in detail, Right? You know, Gibon didn’t talk anything about neural nets and whatever. But, um, but this notion that, and, and this gets to, to, um, this interesting paper by, uh, well, Hassan Yu is one of the authors
Paul 00:37:38 Direct fit,
Michael 00:37:39 The direct fit paper. And we kind of follow his line there suggesting that, you know, yes, it’s an option to, to, uh, talk about these things in terms of, of models and representations, but you’re not obligated to do so.
Paul 00:37:52 Mm-hmm. <affirmative>,
Michael 00:37:54 Um, and, and auto coders is the, the, the vehicle we choose to illustrate that however successfully you think we did. But, um, back to the, the question, look, So <laugh>, my own view is that, um, yes, there are mental representations, and now I give a kind of origin story of them. I think that, um, that, that language began. So I’m, I’m Vic Jenny on this point, that language began as, um, a, a kind of a complex method of coordination. So just a straightforward of our ability to coordinate with information in the environment, in the, in the optic array, in in other, uh, uh, other, um, modal sensory modals. Um, and, you know, and so, you know, Vic and Stein’s builders, uh, is, is a kind of an exemplar for me, right? You just say slab, and that’s not a name for a thing. It’s a request for a particular item, and your coworker goes and grabs you a slab, which you then lie on the wall in the, in the correct orientation. Um, and, you know, uh, j Austin’s how to do things with words is, is in the, in a similar spirit mm-hmm.
Paul 00:39:15 <affirmative>,
Michael 00:39:16 Right? That, that we, we tend to think of words as representational, uh, as fundamentally representational. And what Austin says is, you know, they’re fun, They’re fundamentally performative. And what I suspect is that the representational aspect comes later. It comes from a, who knows where this came from. Again, I, I do wanna emphasize that, that despite this often being used as a kind of cudl against these non-representational list views of cognition, there’s vast mystery here that remains for anybody in this view. Right? And if you deny the mystery, you’re, you’re simply closing your eyes to the reality of where we are in the science.
Paul 00:39:59 But I just wanna clarify, are you’re talking about mental representation, not neural representation, right? At the
Michael 00:40:06 Moment, I do not have an account of how neural representation and mentor representation line up at this point.
Paul 00:40:10 <laugh>, no one does <laugh>. Yeah. That’s the issue. Yeah.
Michael 00:40:14 Yeah. But, but to, but to John’s question, look, I won’t admit. Yeah, I think that’s right. That there are representation, hungry problems. Now my my, I would love to get Tony online here for 10 minutes and see what he has to say about this. Cause he’s a bit a bit more aggressive on this than, than I am. Um, but, uh, I think it’s right that there are representation, hungry problems, that writing a novel is clearly one of them. It would be. And so I’m happy to let representations in at some level of our mental architecture
Paul 00:40:52 Detached. So John wants them detached and, you know, a lot of
Michael 00:40:56 What does detached mean?
Paul 00:40:57 Uncoupled rather, um, uncoupled from like the sensor motor, um, performative aspect of, of embodied and active aspect.
Michael 00:41:06 Yeah. No, that they would have to be that. And that’s, he mentioned, he mentioned, um, Bill Ramsey, and that would have to be that for it to meet the job description challenge, and presumably they’d have to be that to write a novel with them, Right? It’s not like you’re acting out the every action of your characters.
Paul 00:41:23 Right.
Michael 00:41:24 Right. As as you, Right. And so what I wanna say is that the, you know, the, I’m perfectly happy to allow in the cognitive economy where I wanna draw the line, is the notion that they’re at the core of that cognitive economy, right? What they are is add-ons. Um, and, and where they came in and exactly how they came in, These are all all very open questions, right? You can, you can, you know, do the tomicello thing. Um, but yes. So at some point, we acquired the ability to use storage, sensory motor traces to do things that didn’t involve sensory, overt sensory motor action. Mm-hmm. <affirmative>. And that was probably the origin of mental representations. Uh, and once you’ve got that, then you can elaborate language in a picture cook kind of way. And once you’ve got language elaborate in a particular kind way, now you’ve gotta, you got another resource to scaffold a level of mental representation, right? That allows you to do, you know, even more cognitive work than you’re able to do. Right? So, so I, I wanna insist that the core of the cognitive system is the sensory motor coupling, but that Yeah, sure. There’s, there’s an ability to step back, as it were from that, uh, and to utilize the kinds of stored traces that those processes, uh, um, create or, or Right.
Michael 00:42:54 Enable.
Paul 00:42:54 Well, I, I wanna, like, I’m not gonna articulate this well, I’m afraid, but in the back of my head, you know, I have this, So, so what you just articulated was that sure, the representations could be there and they could be uncoupled, but the challenges to get from where we are now too, to that story, Right? And I suppose your way of, Well, so I, I suppose someone like John’s way of thinking of it would be that we have these internal models and that, you know, when they’re active, those are the representations. And it’s just totally uncoupled from our sense remote, um, uh, faculties. But then if I, if I was gonna speak for you, what I say that, um, in a Gibson in direct perception kind of way, right? That comes into our sensory epi eye, uh, and then gets processed in a direct manner, let’s say like that. I mean, as you go up the hierarchical structures in the brain, is there a story to be told about the constraints of the brain architecture where it’s essentially the same process where you don’t have to speak of representations then, but it’s just, you know, I mean, now I’m tripping over my own words because, you know, I’m, I’m thinking of like, become fit, uh, activations becoming more abstracted, more abstracted, right? And then somehow that counts as a representation, but to fit it within a Gibson story, Sorry, that was just a incoherent blabber, perhaps.
Michael 00:44:15 Well, and, and this is the, the trouble we get into when we start thinking about this, right? So you said hierarchy, right? Yeah. Hierarchy already brings a bunch of baggage with it. And you said more and more abstract. Yeah. Uh, that already brings in a particular view of what’s going on here. Uh, I, I haven’t thought a lot about abstraction, I’ll be honest. Um, but I, you know, it makes my skin crawl a little bit. So I, I, like, I have the sense that there’s something wrong with that way of talking, but, but the hierarchy thing, I wanna push back against hard. Okay. Brain is not hierarchical, except in very limited cases it’s hierarchical.
Paul 00:44:54 We could say layered.
Michael 00:44:57 Well, the brain itself is, you know, anatomically layered, but that’s very different if we’re saying it’s functionally layered.
Paul 00:45:03 Hmm. Okay. Hierarchy implies function to you then, and I suppose that Well,
Michael 00:45:07 Yeah. Yeah. Well, and you said move as things move up the hierarchy, <laugh>,
Paul 00:45:11 I know, I know. It’s hard to avoid the language.
Michael 00:45:14 So Right. Implies a kind of feed forward thing where we’re moving up. Maybe it’s a anatomic hierarchy, but that, that somehow maps on to, uh, and despite his dissing of Dan there, I think, I think that basic Dan dantian point is right? That, that the notion that our psychological categories and are, are, are way of breaking up psychological function should map in a really neat, direct way to anatomical, um, regularities was a red herring from the beginning, and we should have noticed
Paul 00:45:42 It, that that isomorphism is a red,
Michael 00:45:44 That’s that isomorphism thing. That’s what I take from there. So there’s no moving up the hierarchy because, because, you know, first of all, and this is where this is already oversimplified, right? And, and we get into the reflex arc, uh, phenomenon, right? It’s not like there’s a, a thing that happens,
Paul 00:46:05 Right? Right,
Michael 00:46:07 Right. And then it moves up a hierarchy. That’s what happened in neural networks in
Paul 00:46:11 Isolation.
Michael 00:46:11 That’s, that’s one of the dangers of neural nets as a model for the brain. Cause that is exactly what happens in neural nets. There’s a thing that happens that’s presented to the input notes, and then a thing, a bunch of things happen along the way going a hierarchy, often a real hierarchy. Uh, and there’s real abstraction because right. You get features and, and and, and then higher level mathematical representations of those things,
Paul 00:46:35 And then you turn the computer off
Michael 00:46:37 <laugh>. Yeah. Um, to, to, to use that same kind of, uh, viewpoint in talking about the brain is, is I think, fundamentally misleading. Because first of all, there’s ongoing activity and not just ongoing activity of the organism. Even if you throw somebody in a scanner and they can’t do anything else, but sit there and look at whatever you’re showing them, right? There’s background activation, right? Um, you know, there’s just the natural o frequencies of various parts of the brain. So there’s no walking up a hierarchy as if, as if that’s all that happens, right? There’s stuff going on already that modulates anything that’s coming in. Um, in, certainly in naturalistic, uh, uh, scenarios, there’s ongoing activity that’s, that’s affecting what happened. This is what’s right about the multiple drafts theory, which is otherwise crazy
Paul 00:47:25 Denniss Daniel Dennis’s multiple drafts theory.
Michael 00:47:27 Another more, I’m just doing that to, to Neil John right now, <laugh>. Um, uh, but, but, uh, which is that like, there’s stuff going on all the time, and all that stuff affects the stuff. That’s the new stuff that’s happening, right? Right. The old stuff is still there and it’s exerting its influence. So the notion that there’s some kind of strict hierarchy you could identify just seems a mistake. Um, instead we’ve got to understand, you know, in the book I say, you know, top down, bottom up side, side, right? Just right. But this makes it very hard to study. Uh, you know, I, I do recognize that this is why, you know, you alluded to the appendix of the book. That’s why I put the appendix in. Like, I, I do recognize that the negative part of my work showing, Well, that can’t be right, that can’t be right, that can’t be right. You, you can’t just leave it there. You’ve gotta say, Okay, what do we do instead?
Paul 00:48:22 Yeah.
Michael 00:48:23 Um, and, and you know, among the things we do instead is look at dynamics. We look at dynamics directly. We don’t, we don’t, we get rid of this whole notion that there can be is stimuli or pure tasks. And we look at the modulatory effect of ongoing activation, right? And, and along with inputs and outputs, uh, and how all that works together. And we have the mathematical tools to do it. It’s hard. It’s hard from an experimental standpoint, and it’s hard from an analytical standpoint, but we have the tools to do it. Um, it, it’s, but, but it, it, it means, you know, getting beyond and it’s hard as you just demonstrated, right? Talking about things like hierarchies and, and you know, individual inputs that can be isolated and you can follow its processing in any kind of simple way. That’s, that’s not, that’s not the successful future of the neurosciences.
Paul 00:49:22 Is there room though, and sorry that this isn’t aside. I was talking with John about this also, uh, cuz he considers himself a, a pluralist, right? So if is there room in a pluralistic approach to understanding brains and minds for us to say, Yeah, there is higher, that looks like a hierarchy, fine, but you can also slice it this other way and you can also slice it this other way. And, and so, so everything’s okay to talk about in that manner and useful in some respect, but maybe it’s not that the brain is a hierarchy, a full stop or something like that, right? Is there use in, in continuing to use terms like hierarchy and, you know, representation and so on?
Michael 00:50:04 So, um, so this gets to the, the question of sort of realism in models, and I don’t mean inner mental models, I mean the models that scientists use to
Paul 00:50:16 Study things. Yeah. Right?
Michael 00:50:18 And so absolutely correct that, Yeah. Look, if I’ve got a particular purpose, say I want to predict, um, the onset of Alzheimer’s, and, uh, it turns out that a particular model that makes assumptions about neural hierarchies is really useful in predicting early onset of Alzheimer’s and maybe allowing us to intervene at a point in the process that allows better, say, clinical outcomes, uh, than is typically possible, then heck yeah. I mean, do that, right? Um, but then don’t make the follow on assumption that you’ve thereby discovered something fundamental about the architecture of the brain that can be generalized to tell a story about the whole thing. Hmm.
Michael 00:51:07 And so if you’re a pluralist, uh, uh, then you, you’ve really gotta resist that latter step and that latter step is proves to be hard to resist if you look at the literature, right? Mm. Um, you know, there’s this just very recent curfuffle over this cool paper about zebra fish reflexes and how they’re innate and, and people extrapolating from that, right? To, you know, ah, well this means that, you know, in fact, you know, the, there’s genetically specified behavioral circuits across. It’s like, no, it doesn’t mean that. It’s like, yeah, that’s cool and it’s important for the survival of this particular species in a geological niche, right? This is why giraffes, the baby giraffes can walk within whatever, a few hours mm-hmm. <affirmative>, you know, that’s not learned, that’s built in and, and it dang well better weed because there are lions on that, Serengeti and <laugh> if they, if they can’t be mobile quite quick, right? They’re, they’re not gonna, they’re not gonna survive that. So, um, so I would say I’m a pluralist as well. But, but then if you’re a plural, I think you also have to be an anti-real, or at least a mild anti-real.
Paul 00:52:12 Okay. Mild. I’ll accept mild there because I think there is room for realism in Well, oh man. Now I’m gonna have to play you the second guest question, and we haven’t fully gone through, uh, you know what, I’m just gonna go ahead and do it because this is, this is getting into, um, you know, realism and what the world is really like. So, okay. If you don’t mind, uh, this one also, it’s about a minute and a half or so. Uh, this is from Alex Gomez Mad and, uh, I’ll just play it here. It’s about, well, I’ll just
Speaker 7 00:52:41 Play it. Hello Michael and hello Paul again. I’m Alex Gomez Marin from the Institute of the In Staying, and thank you for asking me to contribute with a question or, or is it a comment? Okay, so here it goes, Michael, your work is really a garden of stimulating reflections as you have engaged in the discussions about the future of psychology. It’s problematic taxonomies, the evolution of the brain, the pitfalls of the stimulus response paradigm, uh, plea for radical embodied cognitive neuroscience, et cetera, et cetera. And generally speaking, I agree with virtually all of those, but I wonder, and this is the question, what is your metaphysics? Namely, could you make explicit the philosophical commitments underlying your scientific thinking? For instance, some are proud physicists and there are others who are crypto dualists, and currently there’s a revival of panpsychism and also of idealism. And many neuroscientists still believe that one can do science without adopting consciously or not any philosophical doctrine. So I’m very curious to know yours, especially given the fact that I’m currently undergoing a kind of four e crisis <laugh> since even when one puts the brain back in the body acting in the world, it seems to me that materialism broadly construed still remains a desperate attempt to understand that four letter work we call the mind. So thank you, and sorry for this rather abstract question. Um, warm regards from sunny ante. Bye bye.
Paul 00:54:28 See, he used, he, he used his hierarchical brain processing to, uh, ask an abstract question. There you go.
Michael 00:54:35 <laugh>. Um,
Paul 00:54:40 Is this gonna derail us too much, you think?
Michael 00:54:44 I, I’m not sure I’m gonna have a satisfying answer. And, and I’ll, I’ll tell you in part why, Um, and this is something that I’m, I’m working on. I’ve really largely either completely avoided the question of consciousness.
Paul 00:55:01 Oh.
Michael 00:55:03 Um, or talked about a really deflationary view, right? So kind of the sub personal, you know, how basically what charmel Charmel calls the problems
Paul 00:55:18 Mm-hmm. <affirmative>.
Michael 00:55:19 So I have a little on self-awareness, but it’s, it’s clearly an easy problem approach to that, to question. Uh, I’m pretty sure I’m not a doist. Um,
Paul 00:55:33 Well, are you, are you a realist? The realism thing set me off to play that question, so,
Michael 00:55:39 So I I I am a realist. Absolutely. Um, you know, and I think you, you, you know, direct perception leads straight to kind of scientific realism, right? Um, and I also, you know, I’m a big fan of Ian Hacking and his representing and intervening and even way back in my dissertation, which, you know, getting back to the original question is, you know, is sort of the pre embodied cognition and body cognition view of things. You know, the fact that we’re situated in the world in a way that we are, you know, within the causal nexus of that world and can intervene in that world, is one of the important resources we have for getting a realistic, uh, uh, getting realistic knowledge of that, of that world. So I’m definitely in the real camp. Uh, I don’t think, uh, I’m not a realist because I believe we have accurate models. I’m a realist because we have the capacity for successful intervention
Paul 00:56:36 Control and intervention Yeah.
Michael 00:56:37 Control and intervention, right? And that’s the hacking line on scientific realism. Uh, and I adopt that. Um, now, but the, but you know, the question of what is the mind and I, I take that to be an allied question about, about consciousness and experience, and how do we fit those things into a physicalist view. Um, I, you know, I, I don’t have settled answers to that. I mean, these are important questions, so I, I don’t want to be dismissive of the question at all, but, but I don’t myself have settled answers to that question. I’m, I’m attracted to kind of a, I once attracted to Sayre and Thompson’s, uh, in activism on this, but they can kind of shade into idealism in various ways. I mean, there’s a, there’s a slope there that they’re, that they’re trying to avoid sliding down, and it’s not clear that, that they can, um, on the other side of the coin, I’m attracted to kind of jamesy and radical empiricism, but as I argued even back in my dissertation, same kind of slippery slope there. So there’s a, there’s a, there’s something precarious about this, and, and that’s fine. I mean, that’s, you know, it’s again, I, I said the word mystery before. I’ll, I’ll, I’ll repeat it here. There’s an important mystery here. Um, and it’s a mystery that that should interest not just philosophers, but, but any intellectual, um, and maybe, you know, scientists who work on the brain especially, um, and <laugh>.
Michael 00:58:15 So I’m attracted to James notion that the mind is a kind of selection process, and that, that the nature of the mind is in its selectivity
Paul 00:58:28 Conscious mind. You’re, you’re, you’re, you’re completing consciousness in mind right now,
Michael 00:58:32 And it’s hard to understand, and it’s even harder to articulate. He kind of, James builds an experience from the get go, and he, he tries to build it in from the get go while rejecting both the doctrine of ideas mm-hmm. <affirmative>, you know, from Locke and, and so on. And, and also the kind of cont notion of, of, you know, sensory intuitions. And so he wants to both get rid of empiricism and, uh, idealism of, of, you know, in its various incarnations, but, but nevertheless, without getting rid of experience
Paul 00:59:09 Hmm. <affirmative>
Michael 00:59:10 And I don’t understand that view yet. I, I, I can say I’m attracted to that view, and, and that’s if I had to pin myself to a metaphysics right at the, um, for, for Alex, that’s where I pin myself and see how it came out. But I, I don’t have a way of articulating that yet. And, you know, I mentioned to you that I, I think I said it was before we recorded, so I’ll just say it now. I, I think the title of my next book, or one of my next books is gonna be James e and Neuroscience. Mm-hmm.
Paul 00:59:42 <affirmative>.
Michael 00:59:43 And clearly, um, you know, a lot of it has to do with things we’ve already talked about, you know, rejection of the stimulus, uh, the reflex arc concept, um, the rejection of the notion that, you know, there basically an adoption of a, a stream of consciousness versus the notion of that there are states of consciousness.
Paul 01:00:01 Oh yeah. The word state is really, I’ve come to really, it rubs me the wrong way. The word state in, in terms of anything having to do with brain or mind.
Michael 01:00:12 Yep. Yeah. Yeah. So we’ve gotta have, you know, we’ve gotta talk about processes over over states. Um, all that has to be worked out, you know, And, but if I were gonna, you know, uh, jump on somebody’s ship, that would be the one I’d jump on. Um, and we’ll see where that, where that leads. Um, but yeah, no, it’s an important question. And is
Paul 01:00:36 It important, I’m sorry, I, I sorry to interrupt, but I mean, is it necessarily important? Does it, would it, does it,
Michael 01:00:41 It’s humanistically, it’s humanistically important
Paul 01:00:45 Scientifically, is it important?
Michael 01:00:47 Well, I think so because, uh, the notion might be that, well, certainly aspects of it are clearly scientifically important. So, so if you think you can, you can isolate states of consciousness. This is what the structuralists looker, and, and these guys, if you think you can isolate states of consciousness, then the way you design experiments is gonna be very different than if you’re a jamesian and you think it’s just this, Right. The stream of consciousness. And that once you scoop something outta that stream, it’s no longer part of the stream and it’s a different thing. Right? So studying this bucket of the stream is just a very different enterprise than trying to get a handle on the stream itself.
Paul 01:01:26 So, So do you think that the stimulus response approach to neuroscience that’s dominated, you know, over the past, I don’t know, 8,000 something, you know, however many decades, do you think that’s been one of the biggest misleading approaches, mistakes scientifically?
Michael 01:01:46 Um, so I wanna resist calling any development in the history of science a mistake. Okay. Because everything that we do leads somewhere,
Paul 01:01:56 Right? But maybe less efficiently. Some, some roads are more efficient than others, perhaps,
Michael 01:02:00 But there’s no way to know that, right? And, and so in this context, I wanna reserve the word mistake for things you should have known better
Paul 01:02:12 <laugh>, okay? But if your metaphysics was right from the beginning, right, then maybe you should have known better
Michael 01:02:17 The, the history could have different,
Paul 01:02:19 Yeah. Okay.
Michael 01:02:20 Right. Um, and there certainly were resources throughout the entire history to have taken a different path. But look, um, excuse me, I don’t, I don’t think that, and this is back to the word mistake. The, the other thing that we, the notion that we didn’t learn anything,
Paul 01:02:44 I didn’t say that
Michael 01:02:46 Well, but I, I’m, I’m, and I, I wanna reject that too. I mean, I think we learned a great deal. Yeah. And you know, in my own view is that one of the things we learned is that the assumptions under which we gathered the, the data undermine the assumptions. That the data undermine the assumptions. And so now we know we need to do something different.
Paul 01:03:06 I was just gonna say, did, did we need to wait this long, or was the evidence in was enough evidence already in a few decades ago?
Michael 01:03:16 I mean, will you forgive me if I throw coon at you right now? Well,
Paul 01:03:19 I was gonna bring up Coon, so please throw Coon at me. Yeah,
Michael 01:03:23 I mean, just, that’s one of the, I don’t think it was Coon said this, it might have been Ernest McMullen, one of my old teachers.
Paul 01:03:34 This is Thomas Coon that we’re referring to who wrote the, uh, structure of scientific revolutions. And I was gonna ask you about paradigm changes and, and if you think we’re on the cusp of one, and, and, but anyway, that might not have been what you were gonna throw at me.
Michael 01:03:46 No, what, what I was gonna suggest, so I can’t remember who said it to me. Well, I think it was Ernest McMullen who said it to me. I don’t know if he coined this, this phrase or not. He may have gotten it from somewhere else, But he, he said that if you actually look at the, the history of the caper revolution, and this is back to Coon now mm-hmm. <affirmative>, right? So you, I think you just mentioned structure of scientific revolutions. His, his earlier book was the, uh, analysis of the Copernican, uh, revolution. Uh, um, and now we look at the church’s persecution of Galileo for adopted Copernican views. What, what he said, and he was a priest, by the way. What what he said was, um, well, the church was morally wrong to have persecuted Galileo.
Paul 01:04:34 Okay?
Michael 01:04:35 But epistemological epistemologically, they were probably right in the sense that at that stage, there really wasn’t enough evidence to compel anyone to adopt the Copernican, uh, over the Tian view of the universe mm-hmm.
Paul 01:04:50 <affirmative>
Michael 01:04:51 Universe. Right? And so, uh, there’s a very hard question that people have spent, you know, obviously the whole logical positivist movement was about this. Like, what is it when something is confirmed or disconfirmed, right? This, this is an impossible question to answer in a definitive way. Like, when is there enough evidence? How does that work? And so now you get back to Coon, and, and his claim is that, yeah, that’s the wrong question to ask. There’s no point at which the evidence has he committed efficiently to make it rational to do one thing rather than another. There’s always a rational way to continue within the same, right, within the same, uh, uh, framework by elaborating the framework. And it’s, it’s not, you can’t say it’s an epistemic, uh, or rational mistake to elaborate the framework rather than jumping to a new or developing a new framework.
Michael 01:05:54 That’s a decision that’s made, not based strictly gone evidence, but based on other considerations. Some of them are aesthetic, some of them are, you know, some of them frankly, are personal. They’re maybe ambition based or, or whatnot. Um, and so the reason I, I bring KU up here is to your question, there’s no point in the recent history of the neurosciences that you could say, ah, this is here, in, in 2001, we had sufficient evidence to reject X or Y <laugh>. Yeah. That’s not how it works, right? Um, uh, there’s always evidence, there’s always, you know, that supports, there’s, there are always anomalies. Um, you know, things that, eh, don’t quite fit, but you can kind of squint at ’em and make ’em kind of, well, okay, maybe, you know, maybe that was just noise. Or, or, or maybe, um, uh, maybe we need to, you know, tweak our model a little bit to capture that.
Michael 01:06:54 Um, and, you know, it’s, it’s, I won’t say never, but it, it’s, it’s the rare case where it’s irrational to make one choice over the other. Um, and so the way that science progresses is not that way. It’s not, uh, some kind of accumulation of evidence that other than we say, Oh, yeah, we, we nailed it, or, No, we’ve gotta reject this and move to something else. It’s that people come along and say, You know what? That, I don’t know why, but that just, that doesn’t make sense to me. Is there a different way of looking at it? And then, you know, Gibson, you know, since we brought him up already, that’s, you know, he, he was a traditional visual scientist for a long time, but he started noticing, you know, that his participants in, in, in his experiments, which were highly constrained, right? You know, this is how you did psychophysics back in the day in visual psychophysics, right? You put people in an ophthalmoscope, you know, with a bite bar. So they’re, they’re, they’re fixed. You have a focus on a distant fixation point, so they’re not secting or anything, and then you show things to them, and, and you have them in whatever you do. And you, they, they weren’t perceiving they were having hallucinations, and, Right. So what he came to, to decide, what he came to realize, I would say, is that that setup had actually destroyed the phenomenon of visual perception
Michael 01:08:18 By, by oversimplifying it to the point where vision was no longer what was happening.
Paul 01:08:23 It’s completely out of context for themic behavior.
Michael 01:08:26 Yeah. And so he said, Okay, well then what do we need? What, And this is back to the very beginning here, right? Like, okay, then what are the assumptions about the organism and the environment, uh, that make this a nonsense thing to do? Um, uh, that then replace the prevailing view of what’s going on in visual perception, which is that you start with these low level features that you slowly build up into more and more complex pictures of, of the world. And so what you need to do to study vision is isolate the lable features, right? And learn how those get processed and then go to more complex things or whatever that that’s the hierarchal view. Um, he, he just wanted said, No, that’s, that’s, that’s, that framework is mistaken, so we need a new one. And he said about creating a new one. Um, and, and so that’s, I think that’s the, the position we’re in now in neuroscience, right?
Michael 01:09:24 We’ve been operating under a particular kind of structuralist assumption where, where we, we get individual, we initially are res or receptives to features, very simple features. Mm-hmm. <affirmative>, we build those features up into models of the world. We test those models in, in various ways through behavior. Um, again, that leads to all the kind of, uh, notions of uncertainty and potential skepticism. Like how do we know our models actually fit the world and blah. And, uh, and, and in the neurosciences in particular, right? Uh, then we’ve got a vision of the brain that is oriented towards that. Ah, we’ve got feature processors in the brain, and then we’ve got things that do conjunctions for us and, and so on. And then we get object files. I’m, I’m now channeling tremen here mm-hmm. <affirmative>, and we get up to object files, and then the object files end up into concepts in some way and con and so on and so forth. Um, and, and again, you know, the kind of straightforward, generally new model, you know, straightforward, um, uh, uh, you know, statistical dependence between aspects of the task or the, or the stimulus, um, and, and activity in the brain. That’s a sufficient way to think about it.
Michael 01:10:37 But I think we’re at the point, you know, that Gibson hit in visual science, which is no, actually we’ve oversimplified things. That’s not how the brain works. We did a different kind of experiment and we did a different model, um, that, uh, that sort of better matches the more sophisticated methods, uh, that, that we have.
Paul 01:11:00 But it sounds like you are comfortable with the current quote unquote Coney and paradigm in neuroscience, comfortable working alongside it without shaking your finger at it. Right. And saying, You’re doing it wrong, and I’m, I’m doing it right. And my paradigm will come through one day soon, I hope. You know, and, and I think, I don’t know if we were talking about this before I, I hit record, but you’re, um, because, uh, after Phenology was, uh, 2014, which is now, you know, eight years ago, and it’s based on all the work that you’ve done before, but you’re still optimistic. I, I, I wanted to get your viewpoint on, on current, on the current paradigm in neuroscience and, and reflecting on how, how your ideas or your approach has been accepted and or rejected. Like, how have, how you felt you have fit, um, as the years have have gone by. That’s a lot to throw you there. But,
Michael 01:11:54 Um, <laugh>, so, uh, the first neural, the first neural reuse stuff that appeared in print was like 2006, 2007, something like
Paul 01:12:08 That. Mm-hmm. <affirmative>.
Michael 01:12:10 Um, so it’s been a while.
Paul 01:12:13 Yeah. So, so what, 16 was my math. Yeah. 16 years.
Michael 01:12:18 Um, and, and when I started this, uh, I had more of the attitude that you caricatured there, which is I’ve discovered something that’s really important to, y’all are wrong, <laugh>. And I quickly discovered that was a dumb way to go about things.
Paul 01:12:39 Okay.
Michael 01:12:40 Um, because it implied, and I didn’t, I didn’t think it through, Um, but it implied, and this is back, this is why I reacted the way I did to your word mistake.
Paul 01:12:52 Mistake. Yeah.
Michael 01:12:54 Because, and I didn’t mean to be implying this, but I recognize now that I was, um, that the work that had been done up till that moment was a mistake that was somehow, um, not valuable. And I actually, I gave a talk in, in New York and, um, you know, had a very prominent neuroscientist, pull me aside after the talk, quite upset with me,
Paul 01:13:26 Can’t name drop again.
Michael 01:13:27 Mm-hmm.
Paul 01:13:28 <affirmative>. Okay.
Michael 01:13:29 And, um, and she says, I, you know, I, I’m, I’m really upset about the way you presented that. Like, my grad students are now wondering whether what they’re doing is worth
Paul 01:13:45 It. How dare you upset Anne Treman, like that
Michael 01:13:48 <laugh> and not her. Okay. Um, and, uh, and my first reaction to that was, that’s not science. Like you can question my data and methods or whatever, but, you know, you can’t accuse me of emotional abuse.
Paul 01:14:03 Oh.
Michael 01:14:04 But, but upon reflection, I, I think she was probably right. Like, and, and, and here’s why. Because in fact, the discoveries I’ve made, let’s assume there are discoveries. Those discoveries I’ve made were based on other people’s work.
Paul 01:14:18 Yeah. Of course. Or other people’s mistakes. That’s
Michael 01:14:21 Super valuable. It was, what I did was impossible without that work. Right? Yeah. So wasn’t a mistake. Well, I think the assumptions under which those data were gathered is flawed, but we can use the very same data mm-hmm.
Paul 01:14:38 <affirmative> what you do
Michael 01:14:39 To heal the assumptions, to, to move the assumptions to a better place. So was it a mistake? No. Are they wasting their time doing it? No. Do they need to reflect in a bit more concerted way on what the assumptions under which they’ve been gathering their data and, and maybe they want to change some of them, or tweak some of them, or, or do something different? Definitely.
Michael 01:15:00 Right. And so, you know, I I, I pushed on the straightforward empirical stuff for, for a long time. And, and, but the, and then the book is really the, my attempt, um, to then fill in at least partially fill in the conceptual background. Right. That makes sense of the anomalies, because what I, you know, the analyses I did on their data mm-hmm. <affirmative> revealed that actually the data they gathered were full of anomalous information from the standpoint of their framework. And so, Right. You know, I don’t know if reuse is a paradigm shift or, or, or what paradigm is itself, Afro word, but, but it does point to the need for a new framework within which, um, uh, data that have been gathered should be interpreted. And that should lead to particular ways of, of gathering and analyzing data in the future. Um, and so, you know, Yeah. I’m the, I just, the reality is that these enterprises have to be going on in parallel, right. Both for practical and even epistemic reasons. Like why would I, we, I mentioned Alzheimer’s earlier, Why do I derail anybody’s Alzheimer’s
Paul 01:16:29 Research? Right.
Michael 01:16:30 Right. That’s just, you know, there, there are, there are ways, there are things, Why would I derail anybody’s work in molecular neuroscience? No. In these cases, maybe, you know, the prevailing assumptions about brain structure actually, or functional structure aren’t really relevant to the importance of that work.
Paul 01:16:53 Right. But that you’re talking about people who would wanna collect, um, connect molecular biology to mind or something. Right. That’s too far a leap. Is that what you’re hinting at with that? And, and their, their target of explanation is not mind, for example.
Michael 01:17:07 Right? Yeah. Yeah, yeah. Yeah. And so, um, and, and with, with those folks, you know, I’ll, I’ll this, I’ll name drop here, John Bickle, cause I know he can handle it. <laugh>, um, you know, his, his way of doing kind of direct connections from molecular mechanisms to, to mental phenomena. I mean, the data are interesting, right? But the framework, um, with, within which he’s operating, I think is mistaken not because, and, and his data don’t show that. Right. And the stuff he does with Antonio Silva, um, in particular, his data don’t show that it’s mistaken. That’s really interesting work. Um, but, you know, especially if you’re a philosopher of a neuroscience or a theoretical neuroscience of a particular stripe, you don’t have the excuse of saying, Well, look, I’m just, I just care about how this neuro transcription factor works in this particular circuit.
Paul 01:18:13 Right? Right.
Michael 01:18:15 Um, and that’s fine. That’s great. It’s important and it’s gonna lead to, you know, really important insights. But if, if you’re not in, in that kind of camp, we’re not involved, We’re involved in a giant game of inference to the best explanation mm-hmm.
Paul 01:18:32 <affirmative>.
Michael 01:18:33 Right. And that means you’ve gotta pull in evidence from all kinds of places and try to fit it all together. That’s why that that book is so fricking long. And the longest part is the bibliography <laugh>,
Paul 01:18:48 Right? Oh, yeah. Well, I think the language chapter might be the little, we’re we’re gonna come back to language,
Michael 01:18:53 But that might be the longest chapter. But, but actually if you, if you break it into parts,
Paul 01:18:57 I didn’t read the bibliography.
Michael 01:18:59 I’ve got a copy over there. I think I did. No, actually, Don, I must have given it to somebody. Um, uh, um, but I think if you just break it into parts and, and take the bibliography as a part, I think the biography is the longest. Because, because, you know, I, I felt it, you know, I had to gather information, gather evidence from as many places as I was competent to interpret, and then try to fit that into a picture. Um,
Paul 01:19:24 Let’s go back to the very, very beginning of our conversation then, because I think that we, we’ve sort of ratcheted to various topics, but, you know, when you were early on, when you were formulating these ideas, and I was asking, was it the data that was showing you the way some first principles a just a rethinking, Like, did you just have a hunch that the current, um, state of affairs and neuroscience was on a less efficient track?
Michael 01:19:47 No, It, it was, it was, it was, it was a genuine moment of, of, uh, Eureka or Aha. Oh. Or, or as, as, um, shoot, I forget who, who said this. Someone in who watches the podcast will know and will put it in the comments or something. Um, uh, but the quote is, The most productive words in science are not Eureka. I have found it, but, Huh. That’s weird.
Paul 01:20:15 Isn’t that Einstein? Or isn’t it attributed to Einstein? I,
Michael 01:20:18 Everything’s attributed to Einstein. I always mistrust things attributed to Einstein or, Yeah. D
Paul 01:20:25 William J. James. Let’s say it was William James.
Michael 01:20:27 Why not? Yeah, it wasn’t, But, um, it doesn’t matter who it was, but, so it was, it was that kind of moment. So I’ll tell you what, what, how it worked out. So, so I, I mentioned in body cognition, we went on a tangent, uh, on that as a result of me mentioning that, um, that phrase. And so, um, you know, I was a postdoc in AI at the time, and, um, and I was doing this side project on embodied cognition, and in particular on the neural basis of embodied cognition. And so I was very influenced by Larry Barcelo and Alex Martin. Um, and, uh, and I said, Okay, well, let’s, let’s, I’ve got, uh, I gather up a bunch of, of neuroimaging data. I, I have these AI techniques I can use to do pattern recognition and analysis and things like this. So let’s, you know, they claim that, um, as a generalizable principle, um, higher order cognitive functions like language, um, uh, you know, planning things like this. Mm-hmm. <affirmative> are built on lower level sensory motor circuits. So I said, Okay, well, so they have their individual experiments that show this. Can we see if this is true more generally?
Michael 01:21:45 Um, and, uh, and so I put together, I forget how big my database was at the time, but I basically put together a database of neuro imaging experiments, and I did some pattern analysis where I, you know, looked at what I classified as higher level functions and lower level. And Yeah. If you look at things that way, yeah, indeed, they’re right. It’s generally the case that, you know, higher level functions, um, mathematics, you know, language stuff, um, you know, planning, things like that. Oh yeah. They’re, they’re, they’re using, they’re built on the circuits that are, you know, also involved in sensory motor kinds of functions. Okay, cool. But then I realized, Huh, what I’m actually engaged in is a kind of confirmation bias, right. Because all I’m doing is looking for evidence for the claim.
Paul 01:22:36 You weren’t trying to falsify
Michael 01:22:37 Anything. I wasn’t trying to falsify it. So I said, Okay, well then let’s look it for the reverse. Cuz the reverse all shouldn’t Sure. Shouldn’t be true.
Paul 01:22:48 Right,
Michael 01:22:49 Right. On this model. And guess what? If you look for the reverse, you find that too.
Paul 01:22:55 Yeah.
Michael 01:22:56 Uh, and then I’m like, well, that’s weird. Uh, and so then now you start just kind of looking, uh, and that’s where Neuro reus was born, because it, you one possible conclusion from this sort of analysis, Well, everything depends on everything, so let’s just go back to lashley.
Paul 01:23:15 Right? Right.
Michael 01:23:16 But that’s not what you observe. If you actually dive into the data, what you observe is all the stuff I’ve reported over the years, which is that, um, you know, well, the base finding is indeed that every bit of the brain is involved in lots of different stuff. Okay. So that’s, um, that’s consistent with the embodied view, but it’s not the same as the claim they’re making. And then you say, and, and again, it’s not just everything depends on everything in kind of mush. No, There are patterns. So newer cognitive functions, language, mathematics, whatever, rely on more and more widely spread brain areas, um, than older ones, which are more
Paul 01:24:00 Not response inhibition, because that’s a pretty low level. Uh, and response inhibition is uses a lot of different, or
Michael 01:24:07 Response, I’ll confess, I’ve not investigated that in particular. So,
Paul 01:24:10 Oh, I thought, I thought in one of your plots you showed that it was response inhibition in particular was one of the test cases where there were lots and lots of different, um, it relied on lots and lots of different Oh, that’s,
Michael 01:24:22 That’s in the taxonomy stuff. So that’s, that’s a whole different analytic technique. And that’s not what we use thing per
Paul 01:24:28 Se. Okay. Sorry to derail you there.
Michael 01:24:30 No, no. That, that’s, that’s an interesting discovery as well. I think that leads to a particular view of how, what happens when these brain regions interact mm-hmm. <affirmative>. But that’s a different, different point. Okay.
Paul 01:24:41 Yeah. But you’re, you’re saying that the more recently evolved cognitive functions, as far as we can tell, rely on lots more,
Michael 01:24:49 Or things that develop later,
Paul 01:24:50 Or things that develop
Michael 01:24:51 Later. So Evo divo is hard to disentangle with the data I
Paul 01:24:54 Have. Right? Oh.
Michael 01:24:57 Um, so, and, and you find that, you know, um, older older, if you can identify them, you know, uh, evolutionary brain regions are involved in more things than newer brain regions. Um, uh, and both of those point to, uh, the reuse architecture, right? That, that what, what, um, what’s happening in the brain is not a matter of sculpting regions to serve particular individual functions or behavioral outcomes. Instead, it’s a matter of putting the pieces together in the right configuration mm-hmm. <affirmative>. And the later something comes online, the more potentially useful stuff there is in the brain. And no particular reason it should be in one spot. So localized. And so this, this is the basic picture of reuse that it’s, it’s it’s functional cooperation. That’s, that’s where the game is. And not localization.
Paul 01:25:56 Is, is that hard for them to accept? That seems like a straightforward, acceptable story that wouldn’t, I mean, has the neuroscience community just kind of accepted that? Or is there pushback on the neural reuse part of the story? Because it, you know, like I said, there’s a bunch of different moving parts.
Michael 01:26:12 This is, this is, this is, this is hard to to say. I mean, certainly the, the neuroscience community is fully embraced network thinking.
Paul 01:26:20 Yeah, yeah.
Michael 01:26:22 Right?
Paul 01:26:23 Yep.
Michael 01:26:24 But as in the example that we began with, even though they’re doing things at the level of networks and even dynamic networks, they really wanna fall back on interpreting the nodes.
Paul 01:26:34 Mm-hmm. <affirmative>,
Michael 01:26:36 Right? So there’s some, some kind of, you know, uh, stickiness, Right. Asis in the system that Right. There’s momentum behind that way of understanding things.
Paul 01:26:46 Not me. I’m free of all of that as you’ve, as you’ve, uh,
Michael 01:26:49 Witnessed <laugh>, No doubt. Yeah. You’re amazing. Um, uh, anyway, just the last one you’re finding. So then, you know, so that’s, that’s what that implies, right? That, that the, the actions at the level of, of, uh, of network configuration and changing network configurations, then if you go looking for that, yeah, you can find it, you know, like the same node. So, you know, in a, in the kind of standard talk I’ll give on this stuff, you know, uh, you, you pick a node. In this particular case, it’s left angular gyres, I think now it doesn’t matter what it is anyway. And you say, Okay, here, here’s a note. Now it’s involved in a motion tasks, it’s involved in language tasks, it’s involved in attention tasks, it’s involved in, Right? So on, well, guess what, it has different partners in every one of those cases. Mm-hmm.
Michael 01:27:33 <affirmative>. So that same note is involved in, in all these different, uh, you know, cognitive, uh, functions, cognitive capacities, but it’s got different partners in each, Um, and, and that’s, that’s the story to me. Uh, and that’s where I think, and it’s, I think that particular point, both methodologically and conceptually has, has yet to really be picked up in any, uh, significant way in neuroscience. So networks. Yeah. Um, um, but the, the, you, you know, again, back to hierarchy, hierarchy is still very much part of the notion. Whereas I think if you’ve got shared nodes between these various networks, which you demonstratively do at least demonstratively based on the data we have, it can’t be hierarchical in a strict way. It’s gotta be hierarchical, right? Because each bit can be a member of multiple different networks.
Paul 01:28:28 But you could talk about functional hierarchies within that domain, right? If, if you really wanted to,
Michael 01:28:33 Right? So, so this is what, you know, you said a lot of moving parts, and this is where things get a little bit delicate, right? So yeah, if you want to fix the behavioral or experimental context and take a snapshot of things, okay, what’s this configuration like for this purpose at this moment? Yeah. You may well be able to discover, By the way, I would love for people to do this. Uh, and maybe someone has, I mean, the, the literature is so vast, it’s, it’s impossible to to track it all. Um, but so entirely impossible someone has done this, but yeah, so if you, if you, um, if you take a snapshot of a network in a particular context and hold all that fixed, and then you wanna describe that network, Yeah. I wouldn’t be surprised to discover that there’s, uh, a discernible hierarchy there,
Paul 01:29:27 But it’s a ghost, you, you would say, within the larger context of other functions. And
Michael 01:29:33 It’s transient, if that’s what you mean by ghost.
Paul 01:29:35 Sure. Yeah. That works. That’s better.
Michael 01:29:38 Yeah. I would say it’s transient and hopefully repeatable, right? Because we do, we’re able to return to behavioral context and elicit similar behaviors. So this is what, you know, this, this acronym I came up with for the book TAs, right? Trans assembled local neural subsystems. And that’s the idea that, and, and I should have, but I couldn’t fit our end that, um, tars or something. <laugh>,
Paul 01:30:03 Yeah.
Michael 01:30:04 Um, uh, but repeatable is also important,
Paul 01:30:07 But not perseveration, perhaps. I’m just thinking in terms of like, when we perseverate, we’re going back to those, those same patterns and, and we’re not as plastic as we think we are. Maybe, maybe we’ll come back to that. Was there more to what you were, um, saying about the, the neural reuse aspect?
Michael 01:30:27 No, I mean, those, those are the, you know, again, so it came out of trying not to engage in confirmation
Paul 01:30:33 Bias, right? Yeah.
Michael 01:30:35 I
Paul 01:30:35 Forgot where you’re back on the origin story. Yeah, yeah.
Michael 01:30:37 And then, and then the patterns that emerged were the things that I’ve just suggested. And then, then that’s okay, that’s cool. Uh, and then you continue to, you know, develop, uh, both analytical and conceptual tools to, to probe that, you know, so the functional fingerprinting stuff that, that I did with Louis OA and, and, and Josh Kenon, um, was, was part of that, was part of that story. Like, okay, so we have this qualitative measure that individual bits the brain, um, uh, respond under various circumstances. Can we quantify that? Can we actually do, um, uh, and yes, we can, right? You basically turn every, uh, part of the brain into a, a functional vector, and then you can quantify its diversity, and you can look at the ways in which these things, how, how, how that affects, say, the assort of characteristics of that node.
Michael 01:31:33 Um, and, and, uh, yeah. So, so we’ve got that paper and no image that kind of details all that showing. Oh, no. And, and an important thing that came out of that conceptually is, um, well, so, you know, uh, you don’t need to identify the function of a particular region of the brain to do good neuros scientific work with an roi. You can say, ah, well here’s its vector, right? Here’s its activity budget, right? This is, its the set of functional characteristics. And yeah, I suppose you could say, Oh, well that must mean it’s a X. But the only kinds of things that people have come up with that I’m thinking of price in Triton now in particular, cuz they confront this problem, is you get these very general things. Well, it’s just an integrator.
Paul 01:32:24 Yeah, okay. You gotta give it a name, right?
Michael 01:32:26 Um, you know, or it’s, you know, this is, this is a, a, you only come up with these really generic descriptions. Um, and that’s because when you look at the functional profiles of most regions of the brain, it’s really fricking complex. And it’s hard to say that it only does one thing. And, and I used to think that, like earlier work, I said, Well, okay, let’s imagine it does one thing, but that one thing is abstract in a particular kind of way, or it’s, or it’s right. You know, you know, I, I was a programmer before I became a professor, and, uh, you know, so maybe it’s like object oriented programming, right? Where each of these things is a little object, um, uh, and it provides a service, and that service can be useful under multiple different, in multiple different programs. So maybe it’s like that. And that was the first model for reuse. And I’ve since given that up, because I just think it’s more interesting and more complicated than that. And I, I think this is back to your, the thing you brought up earlier about response inhibition. Um, and so, and actually the better example is, is attention, but it doesn’t matter. The I idea is that, say the attention specific, uh, network, and let’s assume it’s attention specific. When you look at the nodes of the network, none of those nodes are actually attention specific. So somehow something,
Paul 01:33:48 The collection of the configuration
Michael 01:33:50 Emerges outta that configuration.
Paul 01:33:52 Oh, man. You can set
Michael 01:33:53 Emerges, I don’t wanna say each individual bit does a specialized thing. What I wanna say instead is that there’s mutual constraint between the nodes and that mutual constraint selects out one of the functional possibilities of each of the nodes in the service of achieving whatever the behavioral orco function is that that network serves.
Paul 01:34:17 And that’s what you mean by emergent. You don’t mean a, uh, uh, spooky property of
Michael 01:34:22 No, I mean, I mean, uh, you know, something that, uh, that, and I’m trying to think of a word other than emergent,
Paul 01:34:30 Um, you could say emergent property, you could, you know, but that kind of takes some of the flame out of it, I suppose. I don’t know.
Michael 01:34:36 Yeah. Who knows. But whatever. But no, that, that, the, let’s just say the, um, the transient specificity of that network is a function of the way the constraints operate between the nodes. So each of the nodes is, is multifunctional in an important way, but when it’s in a network formed for a particular purpose, behaviorally speaking, then each of those, uh, nodes is constrained by its participation in that network to only express some subset or one instance of its functional possibilities.
Paul 01:35:19 So do we, Oh, go
Michael 01:35:20 Ahead. That sentence, by the way, that implies a very different neuroscience than the one we’re doing right now.
Paul 01:35:26 It does. Um, and I’m, So you originally, you know, were thinking of like an abstract way to talk about brain area X, right? In this, in this domain, but then it struck me, is it, do we need to just give up language and ontological terms or, and, and start talking about coordinates in a high dimensional functional cognitive space because that’s less satisfying than to be able to, Or do we just need to stop talking about brain area X and start talking more about configurations or what, what’s the, what’s the way forward is my question? You know, like how to discuss these things or,
Michael 01:36:00 Right. So, um, on the first question, do we need to stop, you know, bringing in, uh, psychology talk? You said language that’s, that’s too strong. <laugh>, right? <laugh>. Okay. But, but, um, look, the, the, you know, I, in, in work I’ve done, I, I think we’ve shown that you can actually do interesting science without bringing psychology talk into it and just talking about the location of, you know, networks and or nodes of networks in a high dimensional functional space. Now, you’ve still gotta characterize the dimensions, I think mm-hmm.
Michael 01:36:33 <affirmative>, but that puts things at sort of one step removed, right? From, from the typical way of doing experiment where you think you’re directly testing, um, you know, response inhibition, uh, something you mentioned already or, or work memory or something like this. Um, so yeah, you can do, uh, as it were, work that’s sort of maybe merely predictive. So back to the Alzheimer’s example, maybe it’s the case that, um, uh, or epilepsy, something I’ve actually published. One thing on, um, is, uh, well, what if you were able to track the, the way a particular region or a set of regions moved, given its activity in a big, high level functional space, And what if it turned out that when it enters this region of that functional space, that’s a danger signal.
Paul 01:37:31 Psychotic break or something.
Michael 01:37:33 Yeah. Or, or append, impending lept. Seizure
Paul 01:37:36 Seizure. Sure, Yeah. Whatever. It’s
Michael 01:37:38 Right. Um, well now you haven’t had to use psychology talk at all, right? You’re just doing a high level mathematical model. You’re tracing these things over, over behavioral time, and when it enters a particular region of that space that that’s meaningful, uh, uh, in a particular way. So it’s merely predictive. It’s probably not explanatory,
Paul 01:38:01 Right?
Michael 01:38:01 Uh, and there’s, there’s room in the world for merely predictive sciences,
Paul 01:38:07 But it, but it’s, it’s somewhat unsatisfying. Yeah, yeah,
Michael 01:38:10 Of course. So that’s, that’s the second half of your question. Question. Um, I do think that the computer metaphor inspired vocabulary that we tend to use is likely going to have to be superseded by something else. And so it’s just not likely to me for all kinds of reasons that even assuming, let’s, let’s just stick with this high dimensional notion, right? I I I talk about neural personalities, right? In the book. Yeah.
Paul 01:38:45 Yep.
Michael 01:38:46 Um, right. Where what we’re not, what we’re not trying to do is say, Well, this is the function of that thing instead, Well, there’s a set of tendencies that this region has
Paul 01:38:54 Dispositions. Yeah, yeah.
Michael 01:38:56 It’s got a set of, of dispositions. Um, now either for the individual dispositions or for the, the high level characterization of the, the dimensions of that space, we are probably gonna want a vocabulary and one that is gonna lead us to, uh, some kind of satisfying explanation of how all this works. Uh, I don’t think that the computational metaphor is gonna provide that vocabulary for us for all kinds of reasons. I think it’s probably fundamentally mistaken, even just from a kind of evolutionary standpoint, the, the notion that that’s the sort of thing that would’ve evolved is a, a computer of a particular kind. I, I can’t deny that it’s been fruitful. Right?
Paul 01:39:43 That’s so useful. Yeah.
Michael 01:39:44 It’s, it’s been driving, uh, cognitive psychology, cognitive neuroscience and, and other subfields for a very, for, you know, since 1950. Um, but it’s also not clear to me that it’s, well, it’s, it’s, these are useful models, right? Back to the pluralist question. Um, but it’s not clear to me that they’re explanatory in the strong sense, just because the trouble with computational modeling is it’s too powerful, <laugh>,
Paul 01:40:25 What, what do you mean?
Michael 01:40:27 It’s un totally unconstrained? I can model
Paul 01:40:30 Anything. Oh, yeah. Right. Given enough compute and data, et cetera. Yeah.
Michael 01:40:33 Giving enough data, given enough, uh, um, uh, parts given enough, right? I can model, uh, anything that, you know, this is another, you know, the embodied cognition folks, One of the ways that they get pushed back on is exactly this. Well, sure, of course. The body’s important. I can just throw it into my model <laugh>. It’s like, yes, yes, you can, right? Yes, you can. You can just model, you know, uh, you know, I dunno if you saw that interesting octopus paper that came out some time ago talking about, you know, the, the decentralized control of, of octopus, uh, uh, tentacles and
Paul 01:41:11 The each, each tentacle has its own mind, right?
Michael 01:41:14 Has its own mind, but also also the, you know, it imposes its own constraints on its own movement as a way to, to kind of do dimensional reduction. And, and so the idea was this is supposed to be a, an example, an exemplar of fully embodied, right, Right. Kinds of cognition. Because, because the computations, if that you wanna use that word or it’s doing, are not with neurons, it’s with the body itself,
Paul 01:41:37 Right? As a concern,
Michael 01:41:39 Strength, and computation. People came along and said, Well, I can just model that, Right? I’ll just throw that into some. And it’s like, Yeah, yes, you can.
Paul 01:41:48 That sucks, but you can <laugh>.
Michael 01:41:49 Well, that doesn’t suck. But that to me is, uh, that’s a clue that that framework is not actually a theory.
Paul 01:42:02 It’s a tool. Right.
Michael 01:42:02 It’s just a set of tools. Yeah. And they, they, people often take it to be a theory of the brain, but it’s not, it’s a set of tools that can be used to understand the brain that wants a theory.
Paul 01:42:18 It’s also a metaphor, and it’s been a, like you just mentioned, a powerful metaphor.
Michael 01:42:22 Yeah, absolutely. Powerful and fruitful. And, but, but it’s, it’s not a theory of brain function. It’s not a theory of mental function. It’s a, it’s a, it’s a set of tools that you can use to model it. And you need to constrain that modeling with a theory to, to figure out how you put those models together. And, and that’s the step that has not typically occurred. The notion that you can use, say, I can throw them into my model, somehow undermines any particular theoretical approach to the brain is, is a mistake. That’s not, that’s not how that works. Uh, even though it’s largely taken to be a good argument against certain ways of going, Oh, I can just throw that in my model. It’s like, Okay, but that’s because your model is unconstrained by a theory.
Paul 01:43:07 I, man, there was so much that I still wanted to ask you, but, uh, I really enjoyed the conversation. I really enjoyed the book and, and just reading your more recent works as well. So thanks Michael, for coming on, and, um, I, I just, uh, appreciate what you do and good luck with your paradigm shifting, uh, work
Michael 01:43:22 Moving forward. Well, we’ll see, you know, the, as as coon will be the first to point out, no individual author or even group can be responsible for that shift. That shift is a historical phenomenon that occurs or doesn’t, depending on lots of factors.
Paul 01:43:37 Who, Here’s another quote. I don’t, it’s not really a quote, but who, who said that paradigm shifts happen when the old people die. That’s what
Michael 01:43:46 <laugh>. Uh, so, so, uh, the, I think the quote you’re, you’re referring to, and I also don’t know who it is, is that, uh, science progresses one death, one funeral at a time.
Paul 01:43:56 One funeral at a time. Yeah. One generational funeral. Yeah. Yeah. Okay. Anyway, thanks a lot, Michael. Thank
Michael 01:44:01 You. This has been great.
Paul 01:44:17 I alone produce brain inspired. If you value this podcast, consider supporting it through Patreon to access full versions of all the episodes and to join our Discord community. Or if you wanna learn more about the intersection of neuroscience and ai, consider signing up for my online course, Neuro ai, the quest to explain intelligence. Go to brand inspired.co. To learn more, to get in touch with me, email Paul brand inspired.co. You’re hearing music by the new year. Find email@example.com. Thank you. Thank you for your support. See you next time.