This is the first episode of a book club series on Peter Gärdenfors's book Conceptual Spaces. In this episode, we will discuss chapters 1 and 2, which provide an overview over the book, and a discussion of the three kinds of representation: subconceptual, conceptual, and symbolic.
For this series, I'm joined by Koen Frolichcs, who was already my cohost for the books club series on Lee Child's Killing Floor. Koen and I are PhD students in the same lab.
Podcast links
Website: https://geni.us/bjks-pod
Twitter: https://geni.us/bjks-pod-twt
Koen's links
Google Scholar: https://geni.us/frolichs-scholar
Twitter: https://geni.us/frolichs-twt
Ben's links
Website: https://geni.us/bjks-web
Google Scholar: https://geni.us/bjks-scholar
Twitter: https://geni.us/bjks-twt
My interview with Jacob Bellmund: https://bjks.buzzsprout.com/1390924/9275803-37-jacob-bellmund-deformed-cognitive-maps-abstract-cognitive-spaces-and-how-many-dimensions-can-grid-cells-encode
Bringing up Marr on a first date: https://twitter.com/wavyphd/status/1456038544250638341/photo/1
References
Banino, A., Barry, C., Uria, B., Blundell, C., Lillicrap, T., Mirowski, P., ... & Kumaran, D. (2018). Vector-based navigation using grid-like representations in artificial agents. Nature, 557(7705), 429-433.
Bellmund, J. L., Gärdenfors, P., Moser, E. I., & Doeller, C. F. (2018). Navigating cognition: Spatial codes for human thinking. Science, 362(6415).
Gärdenfors, P. (2004). Conceptual spaces: The geometry of thought. MIT press.
Hafting, T., Fyhn, M., Molden, S., Moser, M. B., & Moser, E. I. (2005). Microstructure of a spatial map in the entorhinal cortex. Nature, 436(7052), 801-806.
Kriegeskorte, N., Mur, M., & Bandettini, P. A. (2008). Representational similarity analysis-connecting the branches of systems neuroscience. Frontiers in systems neuroscience, 2, 4.
Kuper-Smith, B. J., & Korn, C. (2021, October 28). Decomposed 2*2 games - a conceptual review. PsyArXiv. https://doi.org/10.31234/osf.io/5jxrf
O'Keefe, J., & Dostrovsky, J. (1971). The hippocampus as a spatial map: preliminary evidence from unit activity in the freely-moving rat. Brain research.
Poldrack, R. A. (2020). The physics of representation. Synthese, 1-19.
-
[This is an automated transcript with many errors]
Benjamin James Kuper-Smith: [00:00:00] So this is the first episode of our discussion of everyone for the first problem. I dunno how to pronounce his surname. Uh, Peter Gfo. Yeah, me neither. Whatever. Peter Ganden force's book, conceptual spaces. The geometry of thought. Yeah. Maybe if I, I'll just start straight with the co reader. I still don't know how to call the person.
I think co reader isn't a word, but the co reader of this book. Again, I'm John Va Foles, who already was there for the killing floor for club and distinguished himself through. Uh, questionable Google searches and in the name of the podcast, um, and, uh, also relived some chartered experiences on the podcast.
Yeah. So it was a very eventful episode or like series
Koen Frolichs: emotions and, and, and
Benjamin James Kuper-Smith: everything. Yeah. Yeah. I dunno whether I have that this time. [00:01:00] Uh, because it's a very different kind of book. Today we'll be discussing the first two chapters. Usually I did a kind of summary so far, always of the chapters, but that doesn't really make sense now if we just have two chapters.
Um, other than maybe to say chapter one is a kind of introduction, overview of the entire book. Mm-hmm. Um, and kind of prepares you for what's to come. And then chapter two compares the, uh, not actually sure how you, does you call it the three levels of, or the three types of representation
Koen Frolichs: here in the preface?
Yes. Three kinds of representations
Benjamin James Kuper-Smith: kinds. Yeah. So the three kinds, symbolic, conceptual and sub conceptual representation. Yeah. And those are the two chapters we're talking about today. Uh, I thought maybe it might be interesting to start by saying like, why we're reading this, because I guess it's not compared to the other books I've read so far on the podcast, which were Crime and Punishment Killing Flaw and the Invention of Nature, the kind of BOL biography.
This one is a much more niche book and. I [00:02:00] don't think most people have even heard of this.
Koen Frolichs: It's in, it's been in my wishlist, my Amazon wishlist. I would've, or like, you know, just my, my,
Benjamin James Kuper-Smith: yeah, yeah.
Koen Frolichs: For a year later. Yeah. Safe later. Because shely of like, of the title, right? Like the conceptual spaces. So,
Benjamin James Kuper-Smith: but I'm assuming also that, uh, it was the same for you as for me, which is that I read Yako Bemans review paper in science where Peter Fuss was one of the authors and.
They cite the book in there and that, and that's the first time I really heard of this guy. And Or was it different for you or.
Koen Frolichs: Yeah, I have to like, I I wasn't aware of that, that, that he was on that paper. Um
Benjamin James Kuper-Smith: Oh, okay. No.
Koen Frolichs: Oh,
Benjamin James Kuper-Smith: okay. I think Yako, uh, collected a good selection of co-authors with Yeah.
Koen Frolichs: Sounds
Benjamin James Kuper-Smith: like Peter again, first Edward Moza and, um, Cassela.
He
Koen Frolichs: did well. Yeah. Damn.
Benjamin James Kuper-Smith: Yeah, and no, that's, that's the, the first time I heard about him. Just because, uh, I was interested in the whole abstract cognitive maps topic. And [00:03:00] yeah, Peter was one of the co-authors and they cited this book in there and then I thought, got this kind of sounds interesting. Mm-hmm.
And here we are.
Koen Frolichs: Here we are. I mean, I
Benjamin James Kuper-Smith: guess, but wait,
Koen Frolichs: sorry. Go.
Benjamin James Kuper-Smith: You didn't hear about this from the paper or?
Koen Frolichs: No? No.
Benjamin James Kuper-Smith: I just, you just. Did Amazon just know that you would like this or,
Koen Frolichs: yeah, yeah, yeah. Yeah. I think the, the, the Amazon AI is, is pretty well trained on me. Um, I've got a big wishlist of books on Amazon.
No, I mean, and I, you know, and then you read, read some reviews, look on good reads or whatever, and they're like, ah, that sounds kind of interesting. Right. You know, they completely passed me. Um, I, I'll have another look at your paper. Okay. Because I did look him up, Peter g for, and because I was like, oh, he must be cited in like the RSA paper, the, the initial one.
Um, but he isn't. So, but I guess we can talk about a little bit more later.
Benjamin James Kuper-Smith: I mean, one thing that I found interesting is that this also just. Barely overlaps in terms of citations with the whole spatial navigation field, which is weird to me.
Koen Frolichs: Yeah,
Benjamin James Kuper-Smith: I mean, maybe for for slight context, so I guess I [00:04:00] have an entire episode with Yako Mont on this part or half an episode on this of how this applies to space navigation.
But I guess the general idea is just that, for those that don't know, there's this whole research area in neuroscience about spatial navigations. Navigations navigation and basically where they found cells that represent where you are in space and a kind of metric that helps you move through space. And that kind of research.
In the last, I'd say roughly five to 10 years has been now applied to kind of more abstract things. So where the two dimensions of space, let's say you're moving around in a room, in a like square room, the the, the two. You basically have like two spatial axes in which you're moving, and then people started asking, well, what if that?
These axis are not spatial, but anything else, basically. And that's also what KO's review paper is about. And that's kind of like the the, from how I got to this mm-hmm. [00:05:00] Um, and why I found this interesting. But what I found really interesting is that first caveat, this book was published in 2000, so this was also published five years before the first grid cell paper.
Koen Frolichs: Yeah.
Benjamin James Kuper-Smith: But what I still find interesting is that I just flick through the references. John O'Keefe is incited once and. Toman is cited. Okay. So there is at least one link to spa navigation. Yeah. But it was kind of fascinating to me that just the, the, the stuff he's talked about in the first two chapters also
Koen Frolichs: mm-hmm.
Benjamin James Kuper-Smith: Just, it's completely different. Right.
Koen Frolichs: Yeah.
Benjamin James Kuper-Smith: It, it doesn't overlap with that. And even his, you know, he has this section on how this relates to neuroscience, but it's completely different research, which Yeah, I just found really interesting and yeah.
Koen Frolichs: Yeah, totally. Yeah. No, I agree. I mean, I had a look as well, like at some papers 'cause it's like, oh this must be, you know, these papers, you know, the RSA stuff and stuff seemed to me like also like to be influenced by this.
But, you know, it it, it wasn't, but even like in his citations, he seems to miss some important, [00:06:00] like, us, like mastery levels for instance, right? Like even as, 'cause that seems such a.
Benjamin James Kuper-Smith: I think that that will be in there though.
Koen Frolichs: Oh, will be in there. Okay. I didn't look, I only,
Benjamin James Kuper-Smith: I think he side smile. Lemme
Koen Frolichs: check.
Oh, okay. Okay. I can have it. Yeah.
Benjamin James Kuper-Smith: Um, yeah. Yeah. He does cite, yeah. Okay. Okay. Cite. Okay, then I
Koen Frolichs: shut up.
Benjamin James Kuper-Smith: That's probably gonna, I think that's gonna come in at some point. Okay. Um, but yeah, just, just the, yeah, as I said this, this, this does precede, uh, the huge boost in space navigation research that came, especially with the moles.
Yeah. But there's no, you know, there's no head direction cells or grid cells or anything like that, uh, play cells or anything like that. Uh, so that was just a slight. Surprise.
Koen Frolichs: Mm-hmm. Yeah, totally. So what did you kind of expect of this book,
Benjamin James Kuper-Smith: to be fair, I had, I had very few expectations. I mean, I think I knew that he, I think he's described as a philosopher on his website or something like that.
Koen Frolichs: Mm-hmm. So
Benjamin James Kuper-Smith: I did expect. This to [00:07:00] be? Well, I mean, it is a theoretical work that, that's kind of what I expected. I mean, it's also a book, I guess like back in, I mean, I have read some books in which they present original data, especially like in the fifties. There's some of the game theory books where they present first results in a book.
Cool. Which is always vaguely annoying because that's not really why Iberg by book for, yeah,
Koen Frolichs: yeah.
Benjamin James Kuper-Smith: Sure. So I did expect it to be like in the, uh, in, in. Yeah, like if you buy a book now from a philosophy, you expect it to be theoretical, right? Mm-hmm. Um, or even a scientific book, you don't really expect new results in there.
So that's, that's kind of something I expected. Um, yeah, I mean, I guess I, because I've read Corp's paper, they asked that, you know, they, they kind of discuss how you have. How you can, you know, you, let's say you just have two axes and then you, you have kind of these, these almost stereotypical representatives of certain things.
Mm-hmm. So they, for example, have like in their paper, the two axes are So talking about cars and they have the weight and the horsepowers. Yeah. And the power of the [00:08:00] cars, something like that. And they say, you know, you, you basically. This, this continuous variables. But at some point, if a car is light enough and strong enough, it becomes like a sports car.
Whereas at some point it becomes like a lorry, that kind of thing. Mm-hmm. And so you have this kind of prototypical, stereotypical exemplar in there. And then depending on, basically if you have something that's in this 2D space, then whatever stereotype is closest to what you. To what you're seeing, you're associated with that.
Mm-hmm. So I expect something like that to be in there, but. Yeah, no, I had very few expectations. I mean,
Koen Frolichs: okay.
Benjamin James Kuper-Smith: What did you think?
Koen Frolichs: I asked? Didn't have many expectations. I was just kind of curious to, to hear what he had to say. 'cause I think representations in general are like a big topic in neuroscience.
Right? Like how,
Benjamin James Kuper-Smith: which is, sorry. I agree. Like, but the funny thing is I remember that I still don't. Exactly know what the problem is. It's some of this topic that's completely like gone by me [00:09:00] for some reason. I remember we once had a lab meeting about this, this one paper right by, was it by Paul Drug? I can't remember.
Koen Frolichs: Mm, yeah.
Benjamin James Kuper-Smith: Yeah. The physics of
Koen Frolichs: representations.
Benjamin James Kuper-Smith: Yeah. Yeah. Yeah. And I mean, to be fair, like when I wanted to read that paper I had like before the lab meeting on the same day, I had this complete. Mental, not mental breakdown, but like my energy just drained me entirely for some reason that I don't think was related to the favor.
So I didn't exactly get a lot from that discussion or from reading that. Got it. Um, but yeah, I don't really know much about representation. It was just, I. It seems like everyone thinks it's a super important thing and I still don't quite understand it. Um,
Koen Frolichs: okay, and that's interesting. Yeah. I mean, to me, I think one of the questions is just like, you know, how does it happen?
Right. You know, we know it's like neurons firing and then some kind of. Way Is there, you know, there combined firing or whatever it is? Or, or, yeah. I mean, I don't know much about it, but, [00:10:00] but makes it into like a thing or whatever, right? Yeah. It's very vague. I'm very, very vague now. Sorry. Um, anyways, like That's great.
Yeah.
Benjamin James Kuper-Smith: That's why you're here. It's,
Koen Frolichs: I'm the philosopher here. Um, but what I really, what I hope to get from the book is just a different viewpoint, I think. Yeah,
Benjamin James Kuper-Smith: definitely.
Koen Frolichs: You know, like another mental model to add to the list or a way of thinking. Um,
Benjamin James Kuper-Smith: and I think that's definitely gonna ha I mean,
Koen Frolichs: yeah,
Benjamin James Kuper-Smith: for me, I guess it's like even just the introduction of these three levels is, I guess, a kind of discussion about these things that I haven't really read about, which feels like I sht.
Should have in the sense like it would've been very helpful.
Koen Frolichs: Mm-hmm.
Benjamin James Kuper-Smith: Because I guess like a lot of the, the things I think about and that confuse me are exactly how these different concepts we read about relate to each other. And you know, for example, you have, let's say, as you mentioned, like the MAR three levels.
And then you think of something like this and you go like, okay, but how do they relate to each other? Like what, what kind of relationship is between these two other competing theories? Does one fit into the other or,
Koen Frolichs: yeah.
Benjamin James Kuper-Smith: And so it was kind of nice to see [00:11:00] that, by the way, did you, uh, did you see The Onion posted something, uh, like a week ago or something?
Koen Frolichs: No.
Benjamin James Kuper-Smith: Which was something, which is one of the greatest things I've seen. And it was, uh, like by them at least, and it was. Interdisciplinary in, in quotes, linguist brings up MA 1982 on first date and in this matter of these lecture people on a date. And then the scripture's like, yeah, it's really interesting how like the first level was the computation.
Koen Frolichs: So good. Yeah.
Benjamin James Kuper-Smith: Yeah. I'll put that, I'll, I'll put that in the description. Along with everything else we talk about. Um, but yeah, so now every time I hear Mark, I just have to think, think about that. Yeah. That anyway. Yeah. Very
Koen Frolichs: good. Should we
Benjamin James Kuper-Smith: get
into
Koen Frolichs: chapter one? Oh, sorry.
Benjamin James Kuper-Smith: Yeah. I mean, maybe it still as a kind of slight overview of what we thought of the first two chapters as we [00:12:00] discussed before we started recording.
Yes. I think we both had a similar kind of idea that I thought chapter one was really cool. Mm-hmm. Um, I, I really read it and I went like this. I mean, and it still might be true, this might be a chapter in my thesis, even though I'm not really doing this, it felt like suddenly like, oh, this is super cool. I already have vague, but I have several, like some ideas of what I might want to do with studies with some of these things and how things fit together.
And I was like really amazed. Then chapter two was a bit of like, yeah, slightly boring.
Koen Frolichs: Yeah, yeah, yeah. No, definitely. I, I, chapter one I also like, because I realized like, I don't know, like level, level of difficulty wise, I think this book is right on my level, like chapter one. Like I have to pay attention, but I can do it kind of, right.
Yeah. It's not that I have to like look up words or look up stuff like con constantly, and for chapter two I was like, or chapter one I was fully focused. Chapter two, I started to like, you know, drift a little bit. Um,
Benjamin James Kuper-Smith: yeah,
Koen Frolichs: yeah. 'cause it's just a little bit more example comparisons. Yeah. [00:13:00]
Benjamin James Kuper-Smith: I maybe one, one concern I might have had before is that, you know, philosophers don't necessarily have the best reputation when it comes to accessible writing, and especially in Germany.
And I thought this is really simply written, I mean, yeah, sometimes sure. You maybe drift off a little bit because it's something you don't really care about too much. But, um, yeah, I was really surprised how excessively written this is. Yeah, consider that this is like an academic textbook, but philosopher or philosopher of mind.
Koen Frolichs: Sure.
Benjamin James Kuper-Smith: To make it even worse. Um, but yeah. Anyway, so one thing I guess we're kind of gonna somewhat randomly maybe jump around like points we found interesting in the first two chapters. Mm-hmm. Um, I mean, I guess like, I guess we don't really mo or slash we have already motivated why we find the overlay of conceptual spaces interesting.
But one thing that I found really interesting and. It's one of those things, right? Like once you read it, it's one of those things where you go like, oh yeah, that's super obvious. Mm-hmm. And then you might have [00:14:00]had the thought before, but I never quite put it together that way. And that's for example, how, you know, how, how the, the mental representation of.
Certain sensory modalities doesn't match at all the representation of the actual stimulus that you perceive. So the example he gives, and the one that was most interesting to me is the one with color or with with vision in general, where the electromagnetic spectrum, you know, it goes from one end to the other, but we perceive it as a circle when it comes to colors.
Yeah. And that was super interesting to me because, you know, I mean. I take photos, although I seem to be saying that more than I actually do it recently, but I take photos and I edit them. Right. And I use Adobe Lightroom to edit photos. And the way he describes it here is pretty much exactly the way you edit photos, or at least you can edit, edit the color stuff in Lightroom, because you have a pretty much Exactly the color circle that it presents here.
Yeah. Looks pretty much exactly like that. You can then kind of select, you know, then it will crease the, the. How much orange, let's [00:15:00] say that's in it or something.
Koen Frolichs: Mm-hmm.
Benjamin James Kuper-Smith: So you have the hue kind of when you go around, you have the intensity when you go from the middle to the outside of the circle. Yeah. Um, and then you can change the brightness separately.
So it's, that's pretty much like one function that you have in photo editing tools. And I even, so that was one part. And the, the other thing I really also thought was really interesting that he mentioned that, because it's also. You know, it is also true once you also, maybe not that surprising, but is that the brightness and the, what it call it, chromatic of light.
Is not independent of each other. I mean, I had this recently, like when I walk home, I walk through this like, uh, there's like this passage where I walk for like 10 minutes basically through the pitch dark basically. And there's these gardens on each side. Yeah. And you know when after about like five minutes, your eyes get used to the darkness quite a lot.
Yeah. And then you see quite a lot. But everything is black and white.
Koen Frolichs: Oh, really?
Benjamin James Kuper-Smith: And
Koen Frolichs: yeah.
Benjamin James Kuper-Smith: Yeah. Because, [00:16:00] because when it's dark, you don't really perceive color that much. Sure. Yeah.
Koen Frolichs: Yeah,
Benjamin James Kuper-Smith: yeah. And, um, I find it always interesting that, you know, black and white is always made to be like an unrealistic kind of, uh, like kind of maybe even random feature that came from black and white photography.
Yeah. Or when you have like, uh, shadows or something like that. Right. But if you actually go through the dark, it's pretty similar. It, it, it is very much just a gray scale variation when you walk through it. Um, yeah, I guess. That was just really interesting to me, seeing like this stuff that I knew about from taking photos and that kinda stuff, just seeing that like in this book, suddenly appear in the beginning, like,
Koen Frolichs: yeah,
Benjamin James Kuper-Smith: oh yeah, I never put the two together.
Really?
Koen Frolichs: Yeah. Yeah. That's kind of also what I meant with like these different, you know, giving me new mental models. Right. It's just another way of thinking about like, even, you know, what he said, like. Time can be, you know, time is a line, right? But even for some cultures you said like time is like, you know, a circular thing, right?
I dunno what you think about Well, do you know what the [00:17:00] cultures are? I, I didn't
look
Benjamin James Kuper-Smith: it
Koen Frolichs: up and I, I've forgot to look it up now that I'm talking about it. Yeah, yeah. Me do. I mean, it's like that, that, that was means a little bit thing. The thing with this book is like. I kind of write things down for myself to look about, look at later instead of like, 'cause I don't wanna get out of it, right?
Because I feel like if you get out of it, you have to reread a lot. Um,
Benjamin James Kuper-Smith: that's what I don't like about his footnotes. I feel like they don't add much. And then every time I look, I've forgotten what the original sentence was.
Koen Frolichs: Yeah. No. Yeah, you're right. Yeah. So only some of them are, um, are, are interesting.
Benjamin James Kuper-Smith: Yes. A select few.
Koen Frolichs: A select few.
Benjamin James Kuper-Smith: But yeah, I guess it's, and that, that is something where, I mean, I, you know, I haven't thought a great deal about this slash basically not at all. But I am wondering like, are there other features, you know, where for some reason. Humans just mentally transform something into a kind of data structure that it actually isn't once you perceive it or once you do that originally.
And I mean, I haven't been able to think of one like [00:18:00] immediately and I haven't thought about it much, but it does make me wonder like, you know, why is, for example, sound. Perceived as a line that goes from zero to until your ears burst. Yeah. And, but the color is perceived as a circle, like,
Koen Frolichs: yeah. Yeah. No, it's very interesting.
And I, and I, I wonder if there's like, you know, I guess it comes in with the representations, like what is the best way of like com computing this or computing on this, right? You have this representation of color. But what does my brain, like, what does it need to do to passe this information? Right? Is there like a circular representation or does my brain transform it to two dimensional space to then do you know, because we kind of, I guess in, in nowadays, we kind of expect a lot of things to be transformed or in two dimensional space.
Right? If we, if we believe like current literature a little bit, right? With these conceptual spaces, but we you
Benjamin James Kuper-Smith: mean as scientists?
Koen Frolichs: As scientists? Yeah. Or as, as neuroscientists, I guess. Um, um, specifically. Yeah. But not like
Benjamin James Kuper-Smith: as a first off, you meant like as a general person?
Koen Frolichs: [00:19:00] No, no, no. Sorry. I, as, as, as neuroscience, like kind of like there's this little bit of a hype field about the grid cells.
Right. Or I mean, maybe not a hype, but like, I guess a lot of people are doing it 'cause, and we kind of expect, and we see a lot of times that there is this two dimensional representation of like, a lot of like abstract things. Right. And I've wondered, and I, and I talk with people about this like. One thing, of course, why we find these two dimensional representation is because that's easy for us to understand.
Right. But like what would a circular As scientists, as human beings. Yeah. Right. We can imagine a two dimensional plane and how things are projected on it. And the fact, the fact that we find that in neurons is, doesn't mean there's any other representation. It's just that this is two dimensional, is really easy to find for us.
Right. Like as
Benjamin James Kuper-Smith: three dimension. Yeah. That's what we're, yeah. I mean, I guess there's always this question like, are you finding it because it's the way it's done or just because it's what you're looking for? Yeah. Yeah. And um, yeah, it makes me wonder too, I mean, uh. [00:20:00] Yeah, I mean, I, I'm just trying to think of this patient navigation stuff.
I mean, I think. It does seem as if those were found fairly exploratory by, you know, you put a yes, you put an electrode and a rat or a mouse and let it run around. Yeah. And then you see wherever it fires. So I think that probably is something that's, it would surprise me if that wasn't actually the way it's coded.
You know? I don't think anyone planned and said like, oh yeah, it's probably acceleration, the click for that.
Koen Frolichs: Sure.
Benjamin James Kuper-Smith: But yeah, for most of the other stuff that's following it and is going to follow it, um, yeah, that's the question. You know, whenever you make. Do a grid cell analysis over something, what, you know, what does that exactly mean?
Maybe it does mean that it, the brain codes for it, but I guess even with grid cells, I'm not even sure whether it's clear whether they actually do anything.
Koen Frolichs: Yeah,
Benjamin James Kuper-Smith: because I think, I'm not sure, I dunno, there must, someone must have done a causal study about this, but I don't know.
I
Koen Frolichs: mean that's like reward cells and stuff like that.
Right? So that basically. Fire force?
Benjamin James Kuper-Smith: No, I mean, what I mean is with grid sales is that, [00:21:00] I don't know whether they're actually causally involved or, you know, it could be like some sort of artifact that just happens or something or you know.
Koen Frolichs: Sure.
Benjamin James Kuper-Smith: Like whether, again, I, I dunno this field that well, but all I know is that when a rat is there in a certain space, then a neuro on fires.
But like what exactly that means? I'm not even sure. Um,
Koen Frolichs: sure. Yeah.
Benjamin James Kuper-Smith: You know? Does it use that information or does it
Koen Frolichs: Yeah, yeah, yeah. Is this doing whatever? What is this doing? Yeah. Yeah. What's the information
Benjamin James Kuper-Smith: process? Just kind of some weird phenomenon that just happens and doesn't really matter much.
Koen Frolichs: Yeah, yeah.
Benjamin James Kuper-Smith: Someone must have done a causal like manipulation there.
Koen Frolichs: Yeah. There was this talk with Tim Burn, but I couldn't understand it is some kind of simulations where they were like, oh, this is. You know. Yeah. But, but the thing is, I couldn't hear the author name that he mentioned, and I never found a paper. Yeah.
So,
Benjamin James Kuper-Smith: so you'll just never know.
Koen Frolichs: You'll never know. 'cause you know, there's no way of finding this out. Exactly. Yeah.
Yeah. [00:22:00]
I mean, what, what I, so what I like about this, what I like about the first chapter is that he basically breaks it down. You know, like representations, right? He breaks it down, like, and he talks about, um, um, topographic maps that we find in the, in the cortex a lot, right?
Um, retinopathy, pathy and somato, that
Benjamin James Kuper-Smith: I'll just say Yes, yes.
Koen Frolichs: Okay's what
Benjamin James Kuper-Smith: it's called
Koen Frolichs: thank, but basically means that there's like. Coding, like where, um, similar items in like in visual space or on, on your body or in tones are also like next to each other in the, in the brain, right. In the cortex. And he kind of uses that, right?
I think in chapter two he talks about, um, um, pornography, right, as a, that it's actually a line. I guess we talked about this a little bit already. Yeah. I'm sorry.
Benjamin James Kuper-Smith: I mean, like, one thing I do find interesting, and maybe this is also just some, like the kind of math that I like is this kind of geometric stuff and topological stuff where, you know, you, you [00:23:00] can talk like, okay, so you can be in, you know, I mean like, yeah, you can be like in Euclidean space and then these kind of things are the case.
Or you can be in topological space or like as in like a network or whatever, right? And then you have all these other different rules. And I just found it really interesting to see like, you know, these, these, both of those are used widely. In neuroscience, right? Mm-hmm. I mean, this is super common, um, for lots and lots of stuff, but I don't know, for me, again, like I think one of the reasons why I found this so interesting and why I immediately felt like this might be important just for the kind of research I do in the future, is because I immediately thought, okay.
Just because he mentions those different kinds of, the, the way you can arrange things, it immediately gives you more than just one way of thinking about it. Right? You, you don't just go for example, okay, now I'm, I'm gonna have my two axes that are independent of each other, and now I'm gonna run like a grid cell study or something like that.
But you can immediately say like. That's one way of doing it. Mm-hmm. You could also say, okay, let's just assume, like what would it look like if this was not [00:24:00] continuous two dimensional space, but you had, for example, stereotypes, which you might want to arrange as a network or something like that.
Koen Frolichs: Sure.
Benjamin James Kuper-Smith: You know, how would that change the analysis?
And in a way, this is. Nothing new. But I don't know, somehow just the way he presented all of that, at least gave me the impression that I might do something interesting in the future, whether I will as an entirely different question. But yesterday evening for a brief, brief moment, I had hope. Yes. So that was nice.
Koen Frolichs: You had a feeling of, of importance in the future. Very good.
Benjamin James Kuper-Smith: Yeah. I mean, it will, it will all come crashing down tomorrow, but
Koen Frolichs: yeah,
Benjamin James Kuper-Smith: you know, right now it's fantastic.
Koen Frolichs: Actually, what, what, what surprised me about this to, to go on a little bit longer also like about the mathematics is like the difference in distance metrics, right?
Mm-hmm. You have like, you know, 'cause he gives two examples, right? The regular Euclidean distance, right. Just, um.
Benjamin James Kuper-Smith: Felt like he should have said that earlier. Feel like some was [00:25:00] like he's just describing Euclidean space.
Koen Frolichs: Yeah, yeah, totally.
Benjamin James Kuper-Smith: Yeah. But he did mention this at the beginning or something. Like why is he, yeah, I mean, just say that at the beginning then we can Yeah,
Koen Frolichs: true.
But then also like the, the C block metrics and stuff like that. Right. It's like even that.
Benjamin James Kuper-Smith: Yeah. Yeah. Actually that was, yeah, exactly. That was something I thought was really interesting. Yeah,
Koen Frolichs: because I've seen this used before in like some kind of paper or you know, why you're like, why would you do that?
And they didn't really explain it to us, oh, that we did this. Um, and I actually, I don't think he really explains why you would do it. I think in much detail, and this also something I forgot to look up, but it's like, it's interesting, right? Because like that it kind of shows you also to be, you know, you can be creative, right?
Or like, it's not just, you know, if you're in Euclidean space, you don't need to just get these, you know, Euclidean metrics, right? Or like, you know, you can use different stuff for, for, for, um, um, and stuff like that. Just, and, and related stuff, right? So.
Benjamin James Kuper-Smith: Yeah, yeah, yeah, that's, yeah, that's exactly like the kind of stuff that I found really interesting [00:26:00] here.
Where, where I just went like, I mean, the thing is also like for the stuff that, some of the stuff I do, some of this stuff is kind of like inherently kind of already in there that you can have like different kinds of metrics or whatever. But it was always discussed like in this one case, you know, like if you, you know, you have options that bring you money and someone else money.
Yeah. You can calculate like the joint payoff, how much both of them get together, what's the difference, that kind of stuff. But it was always described like in, in just those terms. And it seems to me, seemed to me that when I kind of read this part and especially the mathematical part that I did feel like, ah.
Yeah, just maybe there's just many more options that maybe I haven't considered yet that will be just as interesting or new ways of formulating the stuff I already know.
Koen Frolichs: Yeah.
Benjamin James Kuper-Smith: Um, that, you know, might lead to a dead end and not help at all, but might actually do something.
Koen Frolichs: Totally. I mean, it's, it's, and I think you did it a little bit with like to decompose game stuff, right?
It's. I mean, you, you know, you didn't invent this, but like, it's [00:27:00] looking at stuff that's there. It exists, but like in a different, you know, like decomposing it, looking at it a different way is already like, can change.
Benjamin James Kuper-Smith: Yeah. I mean there, the interesting thing for me is just to, to generalize the whole thing.
To have like a systematic way of looking at it. Sure. To not just say like, okay, so there are these things that we can. Do, I mean, this is, yeah, a very general description, but, uh, you know, there's, it's not, it's not just to say like, okay, these are the things that people have done, but just to show like what the space of possibilities is in which you're moving.
Okay. Yeah. And show like, yeah, you're, you're not just, it's not like these are the three options you can choose between. It's more like, here's the entire split. It's of options that you can choose between, and for some reason we've always focused on those. Yeah. Three or whatever. Sure. And I felt like here, yeah, it was, yeah, it was actually.
Yeah. It's a very good analogy then that I didn't realize myself until you mentioned it, that it's kind of similar here where you go like, yeah, this is just this. Using metrics space to calculate different differences between payoffs is one way of doing that. [00:28:00] Yes. There are many others.
Koen Frolichs: Yeah. Totally.
Totally.
Benjamin James Kuper-Smith: Yeah. And I think that just makes. Well, creating theories and, and experimental predictions. Just so much more flexible. Um,
Koen Frolichs: yeah. Yeah.
Benjamin James Kuper-Smith: I mean, yeah, I guess we'll see like how, how much this is actually gonna change anything I do. But again, the hope is there
Koen Frolichs: from everyone, by the way. Yeah.
Benjamin James Kuper-Smith: Yeah.
Koen Frolichs: Yeah, for
Benjamin James Kuper-Smith: sure.
Yeah. So if you want some hope, read chapter one. Read Chapter one.
Koen Frolichs: Yeah. Quit. Quit after chapter one. No, I mean. Okay. Can we go into chapter two?
Benjamin James Kuper-Smith: Yes.
Koen Frolichs: Um, so in a way I like chapter two, but I feel like, you know, basically every philosopher, and this is a different generalization, has there three levels or has there three?
You know, it, it, it,
Benjamin James Kuper-Smith: yeah. Two is not enough and falls too many.
Koen Frolichs: Yeah. Even though he mentioned, he did mention one guy who had four. Um,
Benjamin James Kuper-Smith: but
Koen Frolichs: yeah, towards the end. But yeah, I mean, I'm sorry, Mr. But did
Benjamin James Kuper-Smith: you memorize those four levels?
Koen Frolichs: No. [00:29:00] Yeah. I'm sorry. This doesn't make sense. Um, and the funny thing is, even he mentions kind of like everyone kind of uses the same, even the same names for it.
Yeah.
Benjamin James Kuper-Smith: Yeah. It
Koen Frolichs: just kind of like, yeah, guys, like, I, I, I, I saw this on Twitter the other time. They call it like the toothbrush theory. It's like you don't wanna use. Another guys or another scientist, um, toothbrush, right? You wanna create your own theory, give it your own name, and you know your name usually.
Yeah. Yeah. Um, and I was like, yeah, that makes sense because you know, there's been a lot of three levels. Maybe we can kind of like put them together. And
Benjamin James Kuper-Smith: I'm looking forward to KO's three levels.
Koen Frolichs: Its three levels. It's
Benjamin James Kuper-Smith: gonna be, I dunno what it's gonna be, but it's gonna be three and that's gonna be great.
Koen Frolichs: It's gonna be three. Yeah. Who knows? One, one chapter might, uh, motivate me as much as chapter one did two. I mean, so, so he talks about these three levels, the symbolic, the sub conceptual, and the conceptual.
Benjamin James Kuper-Smith: Mm-hmm.
Koen Frolichs: Let's, let's, that's my memory. Sub conceptual is basically on the neuron level, right? If I'm not mistaken.
Benjamin James Kuper-Smith: Yeah. I guess like he describes it in two different ways. Right? The, the first is kind of. [00:30:00] The history behind it and who's kind of been doing it. Mm-hmm. And then, yeah, there's kind of connectionist model and then he also, I guess
Koen Frolichs: that's it. Yeah.
Benjamin James Kuper-Smith: Also describes later that Yeah, I mean, you're right. He, he does kind of, I think, explain this in initially that this was often modeled at the neuron level, but also what I found interesting afterwards is that he said like, it's probably is actually best more like in actual physical systems.
This probably like the neuro level is probably the most appropriate for this kind of
Koen Frolichs: thing. Mm mm-hmm.
Benjamin James Kuper-Smith: Um,
Koen Frolichs: yeah.
Benjamin James Kuper-Smith: Yeah.
Koen Frolichs: So you have sub conceptual and that's, and then you have the symbolic representation. I'm doing these two now because that's basically already exists and he adds the conceptual level kind of, right.
That's his kind of addition as new thing, symbolic representations kind of. Highest level, almost human reasoning, I would say. Like, right, like how would you say it? Like, like being able to have symbols, things that have rules attached to them and you can use them and these rules to create [00:31:00] new things or, or understand or learn, stuff like that.
Right?
Benjamin James Kuper-Smith: Yeah. I mean, I, I don't know much. This is probably. Even though he, you know, introduces this conceptual level, I feel like I know the conceptual level better than the symbolic one. Yeah. It's like all the stuff he cites was always this kind of vaguely boring linguistic stuff to me. Like I didn't ever really get like, yeah, whatever.
Um, but yeah, you're right. I mean, there's, yeah, I think that's the kind of, uh, explicitly reasoning about, uh, how things relate to each other and yeah. Mm-hmm. Using logic and rules,
Koen Frolichs: language, logic. Yeah. You're right. You're right. Yeah. Yeah. And then the conceptual of, that's kind of what he. Says he introduces right?
I, I think he, he might, might do, and that's kind of like this in-between thing, between these, like this very low and very high level, um mm-hmm. Type of representation. And he used this jungle, in the beginning it made sense, but towards Sienna was like kind of lost in a jungle. 'cause like, it's like, it's like, yeah.
How do you say it? So it's like ways of. Like basically how it's very like granularity. [00:32:00] I think that's, that's the word he used. It's basically how zoomed in you are and then, right. That's,
Benjamin James Kuper-Smith: I mean, he literally says towards the end somewhere, or not towards the end, but like in chapter two he mentions that, um, these three levels match to different levels of granularity where Yeah.
The one, the, the, the sub conceptual is like the, the, the finest grained
Koen Frolichs: mm-hmm.
Benjamin James Kuper-Smith: Level. And then Yeah. You move up and up.
Koen Frolichs: Yeah.
Benjamin James Kuper-Smith: Um. It kind of made sense to me though, the analogy, right? Like at first you go to the jungle and the only thing that really influences your thinking is like, can I go over there? Can I go there to dig next to you?
Yeah. Yeah. And then suddenly you have like, okay, like, you know, these paths already exist and then suddenly you, and I found, I actually really liked the analogy that he, then, I thought it also really explained very well the, the symbolic level in saying like. Okay, so first you, you say like, okay, I can walk like along that line for that much, that much, but then in the symbol level you actually just name it, walk to that, to tree, whatever, [00:33:00] and then turn around or whatever.
Yeah, and I thought that was a very neat way of, yeah. I mean, I thought it worked, but I guess he himself says it's not, it doesn't, it's not perfect. I dunno, I guess maybe just added that so people don't criticize 'em for it.
Koen Frolichs: Yeah. That's an easy way. Yeah.
Benjamin James Kuper-Smith: Yeah. It's not perfect, but then you just write your
Koen Frolichs: theory.
It's pretty perfect. Yeah. Yeah. Um, uh, what I, oh yeah. Uh, uh, um, so, so in the beginning he talks about, sorry, I'm going a little bit back now. It just, it just popped in my head. Um, in the beginning he talks about Connectionism, um, and, um, assert, socialism. Association. Association. Geez, Christ. My, my English is not, not the best today.
Um, so he talks about these two as in like, those are the dominant series now, right? And then his. Representation is like the new thing. And I feel like, like you know, now, 20 years later, 21 years later, it is almost switched because to me, like yeah, his way of like rep representing like this Euclidean geometrical representation is kind [00:34:00] of the thing we use right now.
Like, I mean, I don't know about you, but we barely talk about Connectionism. I guess Connectionism is like. Depending on, I
Benjamin James Kuper-Smith: think like one thing that's important to bear in mind is that, or like to, to consider when, when you say that, is that it really depends on the discipline.
Koen Frolichs: Yeah, that's probably
Benjamin James Kuper-Smith: true because.
What does deep learning, right?
Koen Frolichs: Sure. Yeah, yeah. No, it's
Benjamin James Kuper-Smith: purely, I mean, all that stuff, I think, I mean, I, again, I'm very much an outsider to that topic, but as far as I can tell, basically this whole like boost in AI recently that came is in large part using the kind of deep learning, uh, the, the, um. Uh, connectionism with like back propagation and a few like add-ons.
Koen Frolichs: Sure.
Benjamin James Kuper-Smith: With lots of data that they now have. I mean, like what like one, like aside here is that I remember once I kind of somewhat randomly talked to a very well established professor in AI basically.
Koen Frolichs: Yeah.
Benjamin James Kuper-Smith: Um, I had like a interview for a like PhD thing and you could talk to like different professors and I [00:35:00] talked to 'em and I said like, oh yeah, there's this you huge boost like.
This was like in 2016 when I talked to him. Like huge boost in like deep learning, that kind of stuff. Like, you know, all these like advances. And he said like, well, not really, like most of the idea is that
Koen Frolichs: yeah, they're back from,
Benjamin James Kuper-Smith: they're kind of quite old. It's just now we have data to test them and to do stuff with it.
Koen Frolichs: Yeah.
Benjamin James Kuper-Smith: And the idea of using games as DeepMind does to test all this stuff and Yeah, like some papers of his from like the nineties or early two thousands suddenly got, you know, just shot of in citations with
Koen Frolichs: Yeah.
Benjamin James Kuper-Smith: All this stuff that came up. So I think. If you are in computer science, I think Sure. Yeah.
Koen Frolichs: Okay.
It's
Benjamin James Kuper-Smith: very much still. Although I guess even though there are some, you know, I mean, DeepMind is very much collaborating with, you know, several cognitive computational neuroscientists and there is this one paper that came out, it's called something like vector based navigation that were, where a lot of Okay.
Um, DeepMind people in, in involved in it. So I think they're also gradually. Adding this, but, okay.
Koen Frolichs: Okay. Yeah.
Benjamin James Kuper-Smith: Yeah. I think they're [00:36:00] probably associated, whatever it's called, association isn, um, is uh, I think still very well, especially predominant now. Yeah. I think I agree. Like for the stuff that we do, cognitive neuroscience, but then again, it, it isn't, but then again, I think this relates to the point I mentioned very briefly earlier, is that.
He mentions that, you know, the neurons actually are also probably the best, the most appropriate way of modeling that kind of stuff. Whereas kind of the higher you go, basically the, you know, talking again about the granularity as you go from sub conceptual to conceptual to symbolic, the higher you should go in your kind of stuff.
And I think mm-hmm. It is. Maybe, you know, what do we do? Right? We do kind of cognitive neuroscience that has Yeah, yeah. We are a bit of reinforcement learning in it, whatever. Mm-hmm. Which is kind of between. It is probably mainly conceptual with like a bit of symbolic Yeah. Maybe in there or with a bit of sub conceptual or whatever.
Yeah,
Koen Frolichs: yeah, yeah.
Benjamin James Kuper-Smith: I was, but we are actually, we are pretty much firmly in this conceptual level. You're
Koen Frolichs: right. [00:37:00] Yeah. I was just about when you were talking about, I was like, yeah, you're right. That that is actually true. Um, we are, and I think also like there is people, actually, there is plenty of neurosciences on the, on the, um, sub conceptual level, you know, the, the connectome and, and that kind of stuff.
Right?
Benjamin James Kuper-Smith: Yeah. And if, I mean, anyone who. Maybe this is wrong, but I feel like most people who do, well, I guess you said like molecular neuroscience with like, you know, ion channels is even lower than that, but mm-hmm. You know, lots of cellular neuroscience and like the very basic neuroscience is probably all of that.
Koen Frolichs: Yeah.
Benjamin James Kuper-Smith: Um, yeah, I think I, I agree though. I think for the stuff, I mean this is also why I think I found chapter one so interesting. Yeah. This, I think really applies to the stuff we do.
Koen Frolichs: Yeah. Yeah. Probably you're right. Yeah. I mean, he had this section connections to neuroscience, but I can't really, after, I mean, I can't really remember much from that.
Benjamin James Kuper-Smith: It didn't have any of the neuroscience I expected.
Koen Frolichs: Yeah. Okay. Yeah, no, that, that might be the same for me though. Yeah. So,
Benjamin James Kuper-Smith: but that's just because I thought like, oh yeah, now he's gonna mention, well, I guess it was before grid cells, but, uh, I mean, I think that's when he mentions like the [00:38:00] somatotopic mapping.
Oh,
Koen Frolichs: you're right. Stuff yes, yes, yes. Yeah, he's a bit more the Yeah, you're right. It um,
Benjamin James Kuper-Smith: I guess it's like many people who are from a different field, the stuff that they know about from other fields is like 20 years old
Koen Frolichs: probably. Yeah,
Benjamin James Kuper-Smith: right. Because you're not like reading papers that came out yesterday, but you're looking at what the classics are basically.
And I think it's kind of the case here too.
Koen Frolichs: Yeah, you're right. Um. I mean in anyways. I mean, yeah. I, I, I'm still still excited to read it. Um, I'm curious to see what's gonna come Actually,
Benjamin James Kuper-Smith: yeah, I'm not entirely sure. I think, I think if I remember correctly, chapter three and four are supposed to be linked somehow.
Oh. I think this is gonna be like the language stuff, which I'm probably gonna bit stupid bored by only a hundred oh pages. Chapter five is semantics.
Koen Frolichs: Yeah.
Benjamin James Kuper-Smith: Oh yeah, there's gonna be the interestingly named OID Tessellations. I have no idea what that is, but that's like a SubT in Sounds very interesting.
Koen Frolichs: I mean, just to to mention on your second date, right, if you impress her with Mar, you can then go, you know, second date OID Tess.
Yeah.
Benjamin James Kuper-Smith: Second date.[00:39:00]
I thought you were gonna mention PET again for. Well, that was actually put again, first three levels, which,
Koen Frolichs: yeah, yeah, yeah, yeah, yeah. You can do that as well. And then there's this one guy who came, put four say,
Benjamin James Kuper-Smith: I think like there's, I think day three and beyond is just, you're gonna run out of scientific theories to talk about Very quickly.
Koen Frolichs: Yeah.
Benjamin James Kuper-Smith: To think of something else to do. Anyway, I, do you have anything else to add to either dating life or this book?
Koen Frolichs: Um, no, I don't think so. I'm just curious to see. I mean, I'm, I'm kind of, I'm open, um, open to what's coming. Um, and I hope, like, yeah, as I said, I, I really hope it'll kind of keep expanding my horizons.
And yours?
Benjamin James Kuper-Smith: No, I, I have the hope. Don't change anything. Just leave me with the hope for a moment.
Koen Frolichs: Okay, I'll, I'll,
Benjamin James Kuper-Smith: okay. Um, yeah, I guess next time what'll be you talking about. Chapters three and four.
Koen Frolichs: Yeah,
Benjamin James Kuper-Smith: which is a bit longer than chapters one and two [00:40:00] and probably a bit more technical. We're always talking about colors anyway, so I guess, yeah, see you next time.
And if you're bored in the meantime, you can listen to the interview with Yako, uh, which talks about some of this.