Episode Transcript
[00:00:04] Speaker A: Does everyone know what people mean by representation in the cognitive sciences? Do neuroscientists even know what they mean by representation? And so, you know, we're looking at various literatures within the neurosciences and cognitive sciences and thinking, wow, it just seems like there's like very different uses of the term.
[00:00:23] Speaker B: What's at stake are the sorts of conclusions that philosophers, among others, draw in thinking that we've made a lot of progress in understanding these mental capacities and understanding consciousness and rational decision making and so on.
[00:00:40] Speaker C: It seems to me that those people should be allowed to keep on using the term representation for these things that don't have very much to do with mental representations. And the only bad part is equivocating between the two and pretending like we have an explanation when we don't.
[00:00:59] Speaker D: I think neuroscientists do actually distinguish these two types of terminology and they're just not using one as a way of talking about the other.
[00:01:12] Speaker E: I think the idea that you're going to ask them, who basically are saying there's nothing at stake, let's just play around, is going to hold it, hold.
[00:01:20] Speaker F: It, John, hold it for this is.
[00:01:22] Speaker D: Not what we want to say. John, that's fair.
[00:01:32] Speaker F: This is brain inspired, powered by the transmitter. What do neuroscientists mean when they use the term representation? That's part of what Luis Favela and Edouard Macher set out to answer a couple years ago when they surveyed a bunch of folks in the cognitive sciences and they concluded that as a field the term representation is used in a confused and unclear way.
Confused and unclear are technical terms here. And Louis and Edouard explain what they mean, what those terms mean in the episode In a moment. More recently, Louise and Edouard wrote a follow up piece arguing that maybe it's okay for everyone to use the term in slightly different ways. Maybe it helps communication across disciplines, perhaps.
My three other guests today, Francis Egan, Rosa, Sao and John Krakauer, wrote responses to that argument and on today's episode, all those folks are here to discuss that issue and why it matters.
Luis is part philosopher, part cognitive scientist at Indiana University Bloomington. Eduard is a philosopher and director of the director of the center for Philosophy of Science at the University of Pittsburgh. Francis is a philosopher from Rutgers University.
Rosa is a neuroscientist turned philosopher at Stanford University. And John is a neuroscientist among other things, and co runs the Brain Learning, Animation and Movement Lab at Johns Hopkins. I link to all of their information and all the papers that I mentioned, plus some other papers that are mentioned.
Throughout the episode. Also some books that Luis, Edouard and Francis have written. Francis in particular has written a recent book called Deflating Mental Representation, which deals with a lot of what we discuss in the episode.
So anyway, you can find all that jazz in the show notes@braininspire co podcast 213.
I am Paul Middlebrooks. I represent Brain Inspired. I had to do it. I'm sorry.
And now I present our discussion, which picks up right where those intro clips left off. Enjoy.
Well, that's. Yeah. Okay, so.
All right, it's already gotten ugly. Here we are. Thanks everyone for joining me.
So we're here to discuss a sort of back and forth that everyone here has had in the literature on the topic of representations.
And so I thought we could just start maybe Louis, with a high level view of what we're talking about and then we can talk about why we're talking about it, what's at stake, why does this matter?
[00:04:29] Speaker A: Yeah, great. Yeah, thanks for having us, Paul. It's great to see you and everyone else as well. So Eduard and I, a number of years ago, we're chatting one day and I was expressing discontent, which is really surprising for a philosopher to be discontent intent about anything.
So I'll say, you know, Edward, you know, I hear my, my peer philosophers say things like, everyone knows what neuroscientists mean by a representation, or there's a really well thought out sense of representation in the cognitive sciences. Let's just take that for granted and move on. Or here are my key examples that I like to go to that represent, no pun intended, my view on representation in the sciences. And so we're going to work with that because, you know, what the scientists do should ground how philosophers of science and philosophers of mind think about representation. And Edward and I were kind of looking into the puzzles and thought, does everyone know what people mean by representation in the cognitive sciences? Do neuroscientists even know what they mean by representation? And so, you know, we're looking at various literatures within the neurosciences and cognitive sciences and thinking. Wow, it just seems like there's like very different uses of the term representation. So there's, you know, in our publication, came out a couple years ago now on this, we decided to do a sort of empirical project on this in which we would try to find some, some evidence, systematic evidence of how the term representation is being employed or deployed, whatever your favorite term is in actual practice. In the actual practice. And so we sent out a survey to a few thousand, maybe more than a few thousand people around the world. And we ended up getting about 800 ish responses.
Philosophers of mine, from cognitive scientists, psychologists, neuroscientists.
And the survey included questions like, you know, do you think cognition involves representation, kind of a binary? Yes. No. We asked some questions later on about, you know, some of the foundational issues in cognitive science and neuroscience. You know, do you think representations can be embodied? You know, all these other kinds of questions. But the. The meat of the project was to present these vignettes that were supposed to be depictive of kind of your old school kind of neuroscience research that's looking at neural activity in relationship to some sort of stimulus. So we presented these vignettes that might be something like a face.
We have someone sitting in an fmri, and they're presented with an alternating kind of stimulation between the blank slides, but then also pictures of faces, pictures of cars, you know, all these sorts of stimulation.
Here's the part of the brain that's lighting up, and we present this picture. Here's a time series recorded.
Now, here are some options in terms of your responses. Would you describe what's happening with the neural activity in relation to the stimulus as being about the stimulus, or conveying information about the stimulus or representing the stimulus, and then standard Likert scale 1 through 7 of how confident they were in their responses. So we have these different vignettes that tried to look at kind of sub questions like do people tend to describe with more confidence that there is activity localized in a single neuron, or that there is activity distributed in a network, or that that network needs to be embedded within other networks to have the kind of relevant kind of activity that we're interested in?
And so that was the basic kind of overall vignette that we presented to them. Eduard, do you want to add anything?
[00:08:23] Speaker E: No.
[00:08:23] Speaker D: I mean, I think the only thing that's worth mentioning is actually say a few things about the three aspects you just described. One was about the scale at which representations are supposed to be found in the brain. One might wonder, well, are representation found at the level of neurons or small number of neurons, you know, a dozen of neurons connected in some way? Or are representations supposed to be found at the level of brain areas, hundreds of thousands of neurons, or maybe even distributed throughout the brain, millions of neurons? So when neuroscientists use a word representation, do they have any expectations at the scale of organization at which representations are to be found? Or actually, are they totally ignorant of that question or have no commitment about that question? The second one was about the causal relation between a Stimulus and brain activation. So does this causal relation, must this causal relation have some specific properties for it to card as a representation? Must it be fully reliable or just somewhat reliable for the activation to carte as a representation? The third one was the embedding in a broader network which we conceptualize as of having a function being used by the rest of the brain.
So that's a little bit the three question that were driving the project and that Louis was alluding to in his description.
[00:09:54] Speaker A: Yeah, thanks for elaborating on that. So what did we find? So of course listeners can check out the article, the primary article, and then Edward and I wrote a target article version of that for a philosophy venue and Drs. Egan, Kao and Krakauer kindly presented their views on that as well. And then we were able to give a response. So what did we find? What was the hotel the hot conclusion from Edwards and my work? Well, to put it gently, everyone uses.
[00:10:30] Speaker F: Representation in exactly the same way. That's what you found, right?
[00:10:33] Speaker A: Exactly. What we found was that this work was utterly unnecessary, that there is conceptual clarity and consistency across all the sciences. We are, we are inching towards solving the mind body problem. So that was it. So really Frankie, Rosa and John had nothing to comment on other than shower us with phrase.
Unfortunately, that's not what happened. So what happened was that Edward and I concluded that the concept is, to put it gently, unclear and confused in its application across the various relevant sciences. And we interpret a lack of clarity and a state of confusion by an interpretation of the results from our experiments. So one kind of result we found was a kind of hesitancy to take a strong stance on how this activity was described. So do you think the neural activity is about a face, for example, and people would pick the kind of middle scores, you know, the four or five, that they were not super confident that it did. They weren't unconfident that it was. So we took these kind of findings as indicative of a sort of, there's not clarity on how to deploy these terms as well. And so there were other findings. We found some confusion as well. I don't know, Edouard, if you want to elaborate a little bit on the lack of clarity and confusion interpretation we gave.
[00:11:59] Speaker D: Yeah, so by lack of clarity we somewhat have a semi technical notion.
So we mean that when people use the words, they don't really have many commitments at what follows from that. Right. So when they use a word representation, you know, you might think, well if something is a representation, then it follows that blah, blah, blah, it follows that we know at the scale at which it is to be found, or we know that it stands in a specific causal relation with something in the world, or we know that it has a specific function. And what we found is that because of this ambivalence that Louis was describing, neuroscientists by and large really don't seem to have any commitments about how the word is used. They just use it in a somewhat free floating manner, one might say. And so that's what we mean by unclear. So confusion was somewhat different. So we were interested in whether neuroscientists and psychologists are willing to say that the brain misrepresents or an activity in the brain is a misrepresentation of something. And what we found out is that neuroscientists are extremely unwilling to do that.
And they seem to be thinking, no, that's actually not the kind of things you can really say. Even if the brain fires, so to speak, when someone sees a house instead of a face, it's not a misrepresentation of a face as a house.
And that led us to the conclusion that neuroscientists confuse, and it's not necessarily a criticism, it's just somehow it's not a distinction that matters for them, but confuse representations and what philosophers call natural signs. So natural signs is smoke. It's a sign of fire.
So when you see smoke, it indicates there is fire.
But the sign itself cannot misrepresent fire.
By contrast, a map, a representation, can misrepresent a city. It can be a very poor map.
So it's two different types of symbols, a sign and a representation.
And our conclusion was that that distinction does not play much of a role in actually neuroscience. When neuroscientists use the word representation, they don't care about the distinction between signs on the one hand and representations on the other. We call that a confusion, because confusion just means you use a single word to talk about two different things.
That's the idea that we used.
[00:14:27] Speaker A: Great. And to finish up, so what do we do moving forward? So that's kind of our descriptive project, which is again the motivation. We're wondering, do we have any systematic evidence for the use of these terms in the relevant sciences? So that's one part. Next part is, you know, how do we interpret the results? But then it's the prescriptive part, which is how do we move forward? And Edward and I, at least to get the conversation started, suggested three possible ways to move forward. One is we can do, and I think Edward can speak on this more clarity, we could reform the concepts.
Edward, do you want to say a little bit about concept preparation?
[00:15:08] Speaker D: Yeah. So there's three options. Reforming, eliminating and understanding. One might say reforming is you take a notion that's not fully clear in science or in everyday speak, and you try to make it more precise. So you try to specify the commitments one should have when one uses the word, or you try to draw distinction between different things in the world. So make it clear that a sign is not a representation. That could be one project, the other one is just eliminating. And I know that at least John is sympathetic to that view for neuroscience.
And the third one, which at the beginning was sort of an afterthought in the original paper, but became more important in our exchange in mind and language, was to try to understand why a notion that's actually imprecise, unclear and maybe confused happened to be playing such a role in science.
And part of the thought is it's actually not an unusual situation that scientists use this kind of notion. And it's not bad. I mean, the crucial idea is that there's actually a virtue of the lack of clarity and of confusion. It is in some sense functional for how science works. And that's a project we so much in a speculative manner. I think that it's more of a gambit, one might say, rather than a full answer.
But we suggested, well, maybe that's the right way to think about the concept of representation. It's unclear and confused, but that's functional maybe. And so we need to understand why that's functional. And there could be a lot of interesting work there, both about contemporary neuroscience and the history of neuroscience, to try to understand why this notion has the features that it has. And so we suggested, maybe there's a very interesting project there to be fulfilled.
[00:16:59] Speaker A: The last thing I'll just, just say is that, you know, listeners might, it might be helpful to think of the. The term gene in biology is one of those concepts that has been, you know, potentially unclear, confused by our standard. But through this kind of, you know, where imprecision is a virtue, as Edward was talking about, maybe the lack of clarity and strict definitions is helping conversations happen across different kinds of biologies where, you know, the molecular biologist can, you know, behavioral biologists or something like that, and they have a loose sense of what gene means and that's good. It helps them kind of understand each other. If it was too precise, they might not be able to have those conversations that are really fruitful for multi scale investigation. And the last thing I want to say Edward said it in passing. The option of elimination. This is one of my favorites.
In the history of science. We see this move to just eliminate, to like just murder a term.
Just put it in a bag and throw it in the dump.
Terms I don't refer to either. Anything real in the world, so empirically supported or maybe that are so conceptually confused that it's not worth using. And so my sense, I think I hold it maybe stronger, less ecumenical view than Edouard. He's much kinder than I am. My sense is let's just get rid of this term altogether. And that's kind of. I think that's my last slide.
[00:18:23] Speaker D: It's the first time anyone has said I was kind in an academic setting.
Takes a compliment.
[00:18:32] Speaker F: Are there examples that come to mind, and anyone can chime in here of the first two options in the history of science, of reform or elimination?
[00:18:45] Speaker D: Phlogiston is an example, obviously is one.
[00:18:50] Speaker F: Yeah, but those are. Those terms still exist, I guess. It's just they've been discredited as referring to something real. Well, so what about reform then? I mean, the reason I ask is because language has semantic drift, right? And trying to reform anything is a fool's errand, perhaps. And so was that just off the table because it's just impossible to do, or are there examples where reform has been a successful project?
[00:19:20] Speaker D: It's an excellent question.
So I'm very interested in that very question about scientific concepts and how they work in science. And I used to be a very supporter, a fan of reforming projects because I think there was a job for the philosopher there. So it was a way to give me something to do.
I take a notion, I clarify it, I propose an amelioration. But I think exactly for the reason you're mentioning, I've become actually a little bit concerned about that kind of project.
You propose something and then there's a semantic drift, and your proposal gets ignored or transformed by users very quickly. And so it's actually not so easy to find very successful interventions of reform where the original intention has stuck throughout the years after its introduction. And I've been looking at a bunch of case studies, and either they don't work, they're not taken on, or actually they get very quickly corrupted. And I think that's exactly for the reason you're mentioning of semantic drift as unavoidable in language use, even in more formal contexts like science. I think that's just totally unavoidable. So I've become a little bit skeptical that reforming is actually a very like this normative top down approach is a very successful approach for concepts in science.
[00:20:43] Speaker C: I think John mentioned in his response article the invention of temperature. So that seems like a case where maybe there was bottom up reform based on people who are actually using the concept and thinking about it in science rather than top down or from outside by.
In for fear of intervention from philosophers.
[00:21:02] Speaker D: Yeah, that's a great example. But notice also it took, according to Shang, hundreds of years for the process to work. I mean, if you believe Chang, it was just such a very long process.
But that's a nice, very nice example. Yeah.
[00:21:20] Speaker F: Okay, so maybe we should just give overviews then of the responses to this. I will say something and then you guys, please correct me.
So John doesn't like the idea of, of being kind and accepting various uses of the term representation, doesn't think it's a useful way to proceed and wants to keep the concept of mental representation alive and to I guess, reform it. And then Rosa doesn't accept the results because she pushes back and says that people can use representation confidently in various ways. But, but the, the survey didn't necessarily get to those ways. And then, and Francis more or less says representation is. It's okay to use it because it's a.
And I. You'll have to clarify for me what, what a causal thin gloss is in terms of relating some measure of brain activity to a mental.
Mental representation.
Yeah, so, so where did I get, where did I go wrong there? Please correct me and anyone can, whoever is more vehemently against what I. Dad can jump in first.
[00:22:31] Speaker E: I'm going to, just because I genuinely find this whole thing startlingly irritating. Okay.
And the reason is that things will get better once we actually think about the science properly. It's not about terminology, semantics or anything like that. It's a failure to think hard about the phenomenology. What's going on.
Okay.
Right. So in other words, as I've said to you before, Paul, spend one week with me on a neurology ward and you'll never question mental representations for the rest of your life.
Okay? They exist. There is representation, rich behavior. This conversation would not be happening without representation.
[00:23:21] Speaker F: But does anyone here disagree with that statement?
[00:23:24] Speaker C: Yeah, I don't think anyone disagrees with that. We just disagree about whether other people who aren't dealing with those.
[00:23:30] Speaker E: Well, I'll keep going.
So represent mental representation, representation rich behavior cannot be disputed. It gets lost selectively. That's why everyone's so terrified of Alzheimer's, why everyone's so terrified of schizophrenia. Is because that representational rich, meaning rich capacity that we have gets cruelly and selectively targeted. Okay? So that's undeniable in my view.
Okay?
So the real question is, if you have that amazing capacity, you know, bats fly, mice don't. Flying is amazing. What's amazing about humans is this rich mental representational capacity which can be selectively targeted by injury and disease.
Now, what happened? That's just as difficult to explain as consciousness is. Now, there was a time in the past where consciousness and that form of cognition were put together. Like Tyler Burges, for example, when he talks about mind, talks about the ability to do a particular form of representation and consciousness.
Now, because consciousness was so hard and had first person ontology, it got shunted aside and said, oh, we can do the cognitive part because that's more third person. It's more algorithmic. Maybe we can deal with them. Okay? So that divorce was imposed, okay? And then ironically, what was done to consciousness until recently was done now to this particular form of cognition. Let's take it down a peg or two. Let's sensory motorize it, let's embody it. Let's find a way to get rid of its special qualities.
Now, totally in parallel, the word representation was used by the neuroscientists who go all the way back to Hubert and Weasel, where informational content, sometimes causal, mainly correlational, but sometimes causal, got called representation. It has absolutely nothing to do with mental representation in the kind that we're talking about, but it just took on a life of its own. So I think, you know, Louise and Eduardo are right. That representation just got its second existence for informational content in neurons that somehow correlates with external stimuli states, blah, blah, blah.
Fine. I would even be willing to say you call one representation star and you call the other one representation real. Whatever you decide. Okay? What happened though?
Maybe we can tell a neural story with the word representation about mental representations.
Okay? That's what happened, right? Is the word was allowed to overlap. Because maybe the answer to how mental representations happen is neural representations.
So you basically place the same property happening at the holistic psychological level and you stick it on the neural level and go, we're on the way to an explanation.
Right?
Complete disaster, right?
So my objection is don't use the word representation for just informational content, because that is not at all going to explain mental representation of the rich. Kind of like Frankie says in her talk in her book, if you're going to talk about mental representations, they are going to have to have the properties of External ones. Right. And she makes a list of those properties, you know, that maps and pictures have. And if a mental representation is going to be worth the name, it's going to have to have those properties.
Fine. That's all fine. I agree with her about that. There are real mental representations. You can really lose them. You don't want to get Alzheimer's. And they have the same properties, mysteriously in our brains that external ones have. All true neural representation. The word should never be used for information content. And then the final problem then is can we use neurons to explain mental representations? If we talk about structural representations in the brain, some kind of isomorphism between the neurons that are sort of mapping on to the real world, and that is allowed neural structural representations, now that's also a non starter in my view.
Okay, so in other words, the two neural stories that borrow the word representation for the true reality of mental representations are just not going to do it.
So then we're left with a very interesting situation is we don't know what the neural story will look like for mental representation any more than we know it for consciousness.
It may turn out not to have any easy intuitive basis that information and structure have, and we should just accept it.
But what doesn't help is A, to deny mental representations, which some people do, and B, to prematurely use the word for neural data, where it's actually not going to help whatsoever. And that's where we're at.
So I believe in mental representations, obviously. I believe that neurons and populations of neurons are ultimately causal for mental representation behavior, but not the way the neural data is talked about today.
[00:29:11] Speaker F: I'm going to let anyone jump in because, I mean, I have things I can say, but this is your argument.
[00:29:17] Speaker C: Rosa and Frankie, I almost agree with everything that John just said, except for the part where he wants to get rid of people using the term in neuroscience, the term representation in neuroscience, because I feel like neuroscience is actually quite fragmented and people go into it from different backgrounds and there are different subcultures, and especially once you include computational neuroscience in it, it seems to me that those people should be allowed to keep on using the term representation for these things that don't have very much to do with mental representations. And the only bad part is equivocating between the two and pretending like we have an explanation when we don't.
So as long as people keep their indices distinct or keep their representations distinct from their representation, stars from their representation with a Catholic Rs, then I think it should be fine. You might think practically it's Hard to be disciplined about the use of the term when we're all using the same word. But I think in principle there's nothing wrong with having a diversity of uses for this word in a diversity, just.
[00:30:29] Speaker E: A little code as that, Rosa. But we now know from the survey that Luis and Edouard did that there is a consequence to it. We wouldn't even be having Paul's podcast if the same word wasn't oscillating between these two qualitatively distinct meanings.
There would be less confusion. I mean, I think the mere podcast happening is indicative that having the same word being used in utterly different ways is hugely confusing. And I've had students come up to me saying they are confused by it. Right, but what, what does it matter?
[00:31:00] Speaker F: What does it matter that people are confused? Like, seriously, what's at stake here moving forward if people are using the terms in the ways in which they understand them?
[00:31:11] Speaker E: Well, I think philosophers have distinction. I made most people don't understand, but what is.
[00:31:17] Speaker F: But why would it matter to understand it?
[00:31:19] Speaker C: Because there's a huge consequence. For example, philosophers who are reading the neuroscience literature and think, oh, of course there are neural representations. And that licenses our talk of cognitive representations or first person accessible representation of neuroscience tells us that, like, that's a problem.
[00:31:38] Speaker D: Sorry, Rosa. I think that's not just an issue of philosophy. At least it is an issue for philosophers relating to neuroscience. But I also think it's also an issue for the kind of explanations that we are getting from neuroscience and the kind of models that neuroscientists are actually gravitating toward. There's really different ways of explaining neural processing or neural dynamics, and using the concept of representation pushes people toward some specific model and away from other models. I think Louis is probably someone, given his commitment to dynamic models of cognition and of the brain, is very sensitive to that kind of question.
But I do think using the word, even if you use it in a very empty way without much content, prime people toward a specific form of explanation of behavior. And indeed, I actually somewhat sympathetically to what John was doing toward assuming that they can get a very easy explanation between neural processing and psychological dynamics or psychological processing.
And I think that's actually really what's at stake is a kind of explanation. We should look for neurodynamics and for the relation between neuroscience and psychology. So I think the stakes are actually not just trivial here and not just for philosophers. We don't matter a whole lot.
[00:33:03] Speaker B: I'm going to jump in here. I agree with a lot of what John says, but I don't think that this shows that neuroscientists are confused about the notion of representation. So I have a different diagnosis for why. I think that John is right, that the notion that they have in mind is a purely information, theoretic correlation, causal notion.
So then the question is, why do they persist in using representational talk? And I have a different diagnosis for that. It's going to take a little bit of setup, but I think in general, my view of mental representation, whether it be what John's calling mental representation or the use of representation in neuroscience, is that such talk is always motivated by pragmatic concerns.
So representational talk is always pragmatically motivated, whether it be in our everyday lives talking about people believing this or that, or neuroscientists characterizing a structure that they're, that they're positing as representing maybe an edge in the world.
So what are some of the purposes that might be served by neuroscientists using representational talk, intentional talk with a commitment to misrepresentation, all of that stuff that philosophers are really committed to.
Why might neuroscientists, if they really mean information theoretic or causal or correlation, why are they using that loaded term? And here's, I think the main point of their using that loaded term. It's because characterizing a structure or a process as representational allows us to evaluate it for accuracy.
If you say that something's a representation, then you can ask, well, is it accurately representing whatever it's representing, or is it misrepresenting whatever it's representing? And I think there's a big benefit to that, to being able to evaluate the structures and processes for accuracy. Now it's not really a benefit that's intrinsic to the neuroscience.
I think it's not really something that neuroscientists care about. I don't think they care that much about evaluating these states and processes as being accurate or not.
But the explanatory targets of the project are cognitive capacities. They're not just arbitrary sets of behaviors. They're manifestations of cognitive or rational capacities.
We've got the normativity, the intentional characterization built in from the beginning.
Given that explanatory target, then the neuroscientist is interested in characterizing how it's possible.
How can we detect three dimensional structure in the scene? How can somebody reach for grasp a coffee cup without knocking it over?
So the neuroscientists are interested in explicating the causal processes underlying target capacities that are Pre, theoretically characterized in normative or intentional terms.
So one kind of maybe crass reason why they might do that is because they have to justify their research proposals, their research and grant proposals. And so everybody recognizes that what they're supposed to be doing is character. Or rather the relevant committees recognize that the explanatory targets are capacities, rational capacities, cognitive capacities. And so by characterizing these structures that are explicated by neuroscientists in purely information theoretic terms, by characterizing these causal processes in representational terms, by attributing content to them, then that's kind of a connective tissue between their causal mechanical accounts, that's what they're trying to give, and the pre theoretic targets that are, that are characterized as successes. I mean, what neuroscientist has to do is explain our successes and our occasional failures. And that gets done by attributing content, by construing the states and structures that are causally explicated in the theory as being representing this and that, and occasionally getting it right generally, but occasionally getting it wrong. So I think that's the explanation for why neuroscientists might seem to be confused or unclear about the notion of representation. They're kind of.
They don't really care about it that much. I think John's right about that.
But they're trying to bridge what they're doing with mental representation in the sense that John's happy with it.
These rational capacities that organisms have to succeed at various things and occasionally fail at others. So just one point to conclude here. Wrap it up.
Neural science, unlike, say, the science of digestive systems, is answerable to what we might think of as an intelligibility constraint.
At the end of the day, the process that the neuroscientist characterizes, it has to be that the inputs that the outputs, look, we can see them as being rational given the inputs to the process.
And digestive science doesn't have such a constraint.
So I think that's why we find representational talk in neuroscience and not in other branches of physiology.
[00:38:49] Speaker D: Yeah, I mean, I like a lot of what Frankie was saying. I don't think I agree with a lot or even. I'm not sure I agree with everything or even much. I don't know that much, but I definitely don't agree with everything. I mean, one thing, just a few small points.
The first one is it's not quite right that neuroscience does not care about assessment.
There's a very long tradition within neuroscience of very influential approach that develops Optimality models of what brain neural processing is.
You can't do a Bayesian model of the activation in, let's say V4 if you don't assume that actually it's optimal in representing whatever it is that V4 is about and so on and so forth.
And that's really one of the most important tradition in neuroscience. So assessment is actually crucial to some part of neuroscience. Yeah, go for it.
[00:39:52] Speaker B: So yeah, I'm not denying that neuroscientists are concerned with assessment, but I am denying, yeah, it's a particular kind of assessment. I think that they do need, they do need to explain, take the frog and the fly. They do need, they do need to. The frog typically catches flies when it, when there's something moving in it, when a fly moves across its visual field. They do need to explain the success of that process.
So optimality considerations, they bear on explaining the success of what's going on, of these behaviors. That's different from semantic evaluability.
That's the characteristic notion that comes as a package with representation.
That's why I mean, to characterize something as a representation is thereby to admit the possibility of misrepresentation.
That's a very particular, this intentional notion is a very particular way of assessing.
[00:40:53] Speaker D: I agree with that. But if you have a Bayesian model of, for example, color perception or a Bayesian model for some visual illusions, the assessment is not just at the level of the whole process. You're going to say, for example, that there's a transformation that's accurate or not. So I think the assessments will be much more fine grained than the one you have in mind there. So in that specific way of representing neuroscientific models, neuroscientific processing, I don't think this is just the kind of.
Are you getting the fly that gets to be assessed? It's a somewhat different form of assessment as well.
[00:41:32] Speaker E: There was one point, just about.
[00:41:36] Speaker D: John, give me a sec, I just want to make a second point, then I'll give you the thing. So one concern I have with your proposal, Frank, in light of our data is I think it's clearly part of the story about why people use the word representation that they want to be connecting it with this. Maybe pre theoretical or maybe pragmatic, as you nicely put it, explanatory goals. I think this is no doubt about that. I'm actually quite convinced by that. But I don't quite think that when people use the word representations they just mean that because our data seems to suggest that people actually don't treat causal relations the way they treat representation talk.
Is it totally happy to use puzzle terminology to describe what the brain is actually doing? They have no issue with that. They say it's processing a stimulus, it's reacting to a stimulus, and that's totally fine. However, when it's this intentional vocabulary representing being about, somehow they become really ambivalent. So I think neuroscientists do actually distinguish these two types of terminology, and they're just not using one as a way of talking about the other.
So that's actually the place where I want to push back. And I'm not sure if that's an objection to you, but I think that's a little bit of a wrinkle. Right. So it's not really what they mean when they use representations. They just don't mean causally connected to.
[00:43:13] Speaker B: I want to address that, but I don't. But I know John's trying to get in here, so if I could kind of come back on that point.
[00:43:18] Speaker D: Go ahead. Thank you. Okay. Okay, John.
[00:43:23] Speaker E: I mean, I just fundamentally disagree with you. It's obviously the case if you ask people and make them suddenly go, ooh, maybe there's a distinction here that in my everyday life, I don't actually entertain myself. I agree completely with Frankie that, that when they use that word, they are basically hedging their bets. They're using it in the informational content sense, and then they've just been infected by the everyday mental representation language as well.
They just don't. I can tell you. I mean, you know, I spend a lot of time with neuroscientists. Right. Being one.
They do it all the time.
Okay, so in other words, you're giving them far too much credit. Now, what happens is that when the neural data go from being sensory motor to being more cognitive, in other words, the actual neural work being done is on a more cognitive topic.
There's more danger of that infestation happening because now you're dealing with more representation rich tasks.
So it just comes more naturally to take the information content job and give it the word representation.
It's just a consequence of what Frankie's talking about. You can't help yourself. But the idea that they have a qualitatively distinct idea what the neurons are doing when they're representing versus when they're just correlating or representing a stimulus. There's the word. They don't, Eduard. They don't. There is no. There is no.
But that's my part, Josh.
[00:45:06] Speaker D: That's my point.
I don't think we're disagreeing.
[00:45:10] Speaker B: Could I jump in here with my point I was going to make to Edouard? I think that's right, but I think that there's another, you know, there's more to the story about why they use representational talk and one important aspect of it. And this is something that I think Rosa highlights in her really great paper putting representations to use. And that is that characterizing a structure or a state in representational terms, saying that it represents, say an edge in the world or a fly for the frog that has the, that has the result of high. I mean the causal process is really super complicated. It's very complex. It start there's a bunch of things going out kind of going on distally. Then there's, you know, proximal stimulations and a bunch of things going on in the brain. Really complex causal process and saying that this structure represents a bug. Stick with the simple example of the frog and fly selects or highlights a particular aspect of that complex causal story. That's, that's important, that salient given the explanatory target, given that the point is to explain how, how this mechanism, the prey catching mechanism works. So the point I'm making here is that it's more than just addressing these kind of pre theoretic explananda, but it's, it's still the content attribution or representational talk is still motivated by pragmatic concerns of the theorist. And that is to in a sense highlight certain aspects of this really complicated causal process.
Highlight them because they're important or salient given the explanatory target. So we're back again to motivating or justifying representational talk in terms of what is trying to be explained.
[00:47:05] Speaker D: My sense is that John was saying I was giving too much credit to the neuroscientist. My sense that you are giving too much credit to the neuroscientists.
My sense is actually the notion of representation is much vaguer and looser and it's much more like a free will that actually used with so little content that I do feel you have a somewhat really restricted and very functional, if not referential, but functional use of the word representation or the concept representation in neuroscience.
And I do think that might be part of the story, but I think it exaggerates its regiment. One might say, in fact much loser and vaguer.
[00:47:55] Speaker E: All I can say to reconcile the two of you is if you read Sharon on the stretch reflex, it's very unlikely the word representing, you know, it doesn't get Used a lot. You don't say that the muscle spindle represents position and velocity. Okay, but what happens if you now do Sherringtonian light work?
Same kind of work, but you do it in the prefrontal cortex. And again, you're just finding a correlation between something in the world and something in the brain because you're in the prefrontal cortex and probably doing a more cognitive task than the stretch reflex.
The word representation is used for the same kind of result, which is just a correlation between the outside world and the inside. But because it's a cognitive task, to Frankie's point, it gets infected by mental representation talk in much more likely than when you're down in the spinal cord. Even though the neuroscience being done is exactly the same.
[00:48:59] Speaker D: That's a really interesting empirical question which I want to flag.
And here's an empirical question. The frequency of representations of the use of the word representation is going to vary across either experimental traditions or parts of the neural system that's examined. And actually we could easily test that by using text analytic methods. I would get really interesting results.
[00:49:26] Speaker E: You see my point? You see my point that.
[00:49:28] Speaker D: No, I see your point. Yeah.
[00:49:29] Speaker E: The nature of the work, the nature of the neuroscience work itself has not changed in the spinal cord or in.
[00:49:37] Speaker D: The prefrontal cord yet.
[00:49:39] Speaker E: The language has changed and in neither case has the genuine entity realism of mental representation been addressed.
[00:49:49] Speaker D: I'm happy with that. I just want to flag out as an empirical prediction here, which I think would be worth examining whether it's true or not about how the concept happened to be used. But I. I think that's a plausible empirical claim. So I'll take that.
[00:50:06] Speaker C: I'm trying to think about the disagreement between Frankie and John here because it sounds like you agree about what neuroscientists are generally doing. It's just that Franke thinks that it's mostly harmless or somewhat helpful because it relevance is and John thinks that it's pathological original.
[00:50:30] Speaker E: I don't disagree with Frankie about where it might be coming from. The only I think where we seem to is I believe that mental representations are real entities. It's a genuine thing that we need to explain that gets lost in disease and one day there will be a neural story. But the neural representation story, whether informational or structural, will not be the story.
[00:50:52] Speaker C: But does Frankie actually disagree with that? I don't believe in first person representations. Right.
[00:50:58] Speaker B: Yeah. But I think it always content is pragmatically attributed. Even in the personal. At the personal level. Yeah, but that's A whole other story.
[00:51:11] Speaker F: Can I ask Louis and Eduard, I don't remember in your study did it distinguish between different stages of researchers careers, their ages?
I know that it's kind of underpowered to start asking those kind of fine grained questions. I mean, the reason why I ask is because I was going to make the comment. So I'm a neuroscientist and the people that surround me and because of that, because of the culture, I probably do this too often. Neuroscientists, when they use the term representation, are merely talking about the shape of whatever neural activity they're measuring. Like when you talk about a manifold that is a representation of the neural data and they're not trying to, to connect mental representation with the neural representation. They're literally talking about the neural activity and the kinds of different shapes that you can make out of it, whether you're reducing dimensionality, et cetera. And I think, and, but I'm not sure if that's kind of a younger generation that's doing that.
So where would that sit? I mean, if we're just talking about neural activity, the shape of neural activity, I think everyone would be okay with that.
[00:52:21] Speaker E: Right?
[00:52:22] Speaker F: As in the activity of deep learning networks as well. Those representations are just saying something about the shape of the neural activity.
[00:52:32] Speaker D: Yeah, so we didn't have.
So we have the data, but our sample size, as you mentioned, is really way too small to look at variation for age as is. Actually it was part of the original project that we would be looking at sub, sub, sub disciplines within neuroscience because Rosa mentioned computational neuroscience and you might think that it has a different use of the word representation compared, let's say, to system neuroscience. So, you know, I was actually, you know, we are very excited, Louis and I, by the possibility of actually doing much more fine grained empirical work. It turns out that neuroscientists aren't the most easy sample.
You know, it's a very busy creatures.
[00:53:15] Speaker A: And.
[00:53:18] Speaker D: You know, we sent, I don't know, 12,000, 15,000 email and just to get a sample of 800, which is not unusual in this kind of work. It's actually what we should be expecting as a response rate. But that was really so we couldn't look. I think that's a very interesting question to examine whether there's variation as a function of age or as a function of cohort.
I don't have a strong intuition. I mean, you have one, maybe John has one, or Rosa, I'm not sure.
[00:53:48] Speaker A: Yeah, can I. Yeah, I agree.
[00:53:49] Speaker D: Yeah.
[00:53:50] Speaker A: Oh, go ahead.
[00:53:51] Speaker D: Yeah.
[00:53:52] Speaker C: Oh, sorry, yeah, just like to the particular question, I think I would speculate that it doesn't have to do with age directly, but it does have to do with the subfield, whether people came into it from computer science or from like a more computer science tradition.
You know, if you go back to David Marr, he had a use of representation that was pretty different from what you see in systems neuroscience. I think in systems neuroscience representation often means whatever the thing is, a current neural activation, the shape of neural activity.
But from computer science it can be anything that carries information about the stimulus.
It doesn't even matter what the further context is, whether it's being used downstream and so on. Everything is called a representation. Any activation in a deep neural network that is called a representation.
And I think there the motivation is not the one that Frankie highlights in. We want to explain some kind of representation hungry activity. It's rather just a way of talking about the way that information is transformed at different stages in some processing system.
[00:55:03] Speaker A: Yeah, I agree, but Rosa points out some equal features, especially the kind of Marian tradition. But I also, I agree with Edward.
You know, the sample size is too small. You know, it would be great because, you know, to say neuroscientists think, I mean a molecular neuroscientist to some kinds of computational neuroscientists. I mean, why do they even go to the same conference? Right. So there's, you know, a lot of variation there. So follow up study, we'd love to see more variation and maybe generational issues. So, you know, I kind of wanted to respond to that as well, but also tie it together. Paul? Yes. Earlier, what, what's at stake here, you know, with, with these debates and I think also to speak to, you know, the theme of your podcast, you know, neuroscience and AI and that kind of relationship. You know, one of my worries is that if people get too comfortable using, you know, I'm putting up scare quotes with my fingers using mental kinds of words to describe non mental phenomenon. I think that's a bit of a problem. And one kind of concrete case where that's a problem is what we're seeing with contemporary. Again, I'm putting up scare quotes. AI right. And the public at large and the grant funding agencies, they love hearing these terms like the machine, the system is creative, the system is hallucinating. And they're using these kind of words that we only use in application to minded things. Right. Maybe the dog, dog is creative, the human is hallucinating, or you know, something like that. But we're so you know, we use these terms like representation.
It's imported from, you know, the cognitive sciences and cognitive psychology and stuff that use it in relation to mental representations. The kind of juicy stuff that John is, it doesn't want us to ignore.
We, we slip back and forth and then we start using that as a computer scientist to describe, you know, nodes in a network. But we're so used to applying these terms in the way that it conveys some sort of mental activity. Then we slip into this quickly wanting to describe these systems that hot take here by Louis Favela. They're not mental, they're not thinking, they're just input output, pattern finding and all this kinds of good stuff. Sorry if I offend anyone, but to describe them as hallucinating, it's just ridiculous.
And you might say, well, but I just mean a technical term. Yeah, but the public doesn't know that, that the grant funding bodies don't necessarily know that. So I see that as one kind of, you know what's at stake here. It's the importing of this mental talk and applying it to things that are not mental.
[00:57:40] Speaker E: And also at that point, just one thing to your point, Paul. Oh, us younger sophisticated generation, we're just using the word representation for the shape. It's just false. I know all those people. Let me give you a quote.
[00:57:53] Speaker F: I just had someone on my podcast I can point to. Of course there are, there's variety, but there are people, people who use it in that way.
[00:58:00] Speaker E: Here's a quote. Where neurons represent the formation of decisions. Okay, that's from a manifold paper. I will not say who it is. So in other words, where the psychological thing, decision making is attributed to the manhole. It happens all the time.
[00:58:19] Speaker F: And I agree that that's. That that's a problem. You're com. You're. I completely agree with you that that's a problem.
And to Louis point also, yeah, it bristles me still when people call an artificial unit a neuron. It's something as simple as that.
And AI, I feel like has essentially co opted and because AI is so popular, has essentially co opted these terms like hallucination or whatever mental capacities you have. But even at that neuron, I really hate it. But we can't control that. That we can just about it.
[00:58:52] Speaker D: Yeah, I will just make a point. I think mostly for the listeners, we are here a somewhat biased sample of opinions about neural representations.
I think it's quite striking that despite our disagreement, we all agree that there aren't neural representations in a very realistic, hardcore sense that we can collect to psychological processing. I think there's no one here on this panel who makes that claim them. There are actually plenty of philosophers who actually are committed to that view. That we can actually easily move from a psychological story about reading and then look at changes in the brain or steps in the neurodynamics and say, oh, this is a detection of the phoneme. Oh, this is a detection of gravitical structure. Right. So here, you know, it's, it's somewhat a select sample of individuals. You know, all these agreements are among friends in some sense, people who are in various ways skeptic of the notion of neural representation. So that's maybe just for the readership. There are actually, there's another set of philosophers and neuroscientists. The second thing I want to say just. And then frankly will follow is just also a small comment about the sociology.
I do think neuroscientists do care extensively about what representation means, about what the word representation means, about what representations are. You can just go on social media on Blue sky, and every third week you're going to have a long thread with 75, 100, 200 neuroscientists going on and going on about what representation are.
And every third week it is the same thread with some different words that happen to be used. So there is actually a lot of demand, I think, in neuroscience to clarify these notions, to, to help somehow build some form of consensus about the way we should use this, we should have this terminology or the way we should use this term. So I think there's actually quite, you know, it's not just philosophers who are just trying to just look a little bit. I think neuroscientists themselves are actually really interested in understanding what are the foundations. And so what do we mean by that? Frankie? Sorry, I interrupted.
I went on.
[01:01:09] Speaker B: Thanks, Edouard. So just to clarify my position, I agree that there's lots of misuse of representational talk by neuroscientists. What I'm trying to do is I think it can play a useful role. And so I'm trying to characterize what's going on in the cases where there's some point to it, where it is playing a useful role.
The other connected with what. So John's cases are. There's lots of cases like that of egregious misuse of representational dogs. And there it's not serving any of the functions that I try to characterize in my book to the point of so what's at stake?
I agree with both of you with everybody. I think here, or at least with Eduard and John, that what's at stake are the sorts of conclusions that philosophers, among others draw in thinking that we've made a lot of progress in understanding, understanding these mental capacities and understanding consciousness and rational decision making and so on. And using the same vocabulary could prop up that idea, can support that idea that there's much more. They're talking about the same thing. So there's much more progress. This connective tissue. I mean, there's an upside and a downside that you need to connect the work with the explanatory target. But you shouldn't make it look like, oh, we've solved the problem of intentionality. Now we know how it is that brain states can refer to. I'm not saying that at all. And that's involved in. I think philosophers are guilty in appealing to neuroscientific work in support of their particular theories of content thinking. That these views are solved. They that the neuroscientific work is actually solving the problem of intentionality. I think that's probably the most flagrant misuse of the work.
[01:03:09] Speaker E: I mean, I was just at a talk at a meeting on representation that Ken Ozawa just at your alma mater, basically. Frankie. Right. And it was interesting that it's not. I mean, it's not that we shouldn't be doing neuroscience on these hard questions of mental health representations. I mean, I'm. No, I'm not a nihilist.
Right. In other words, there has to be a way of using neural evidence to update our theories, algorithmic theories of what's going on in representation, rich behavior. I mean, I've always said you should send out all your hounds at any level of evidence.
The critical issue is what is the explanation that is constructed out of these different forms of evidence.
And what I'm saying is, is that the explanation using all forms of evidence isn't itself going to be articulated in neural language.
Right. This is the mistake is to hope that you see a sort of homunculus of the phenomenon you want to explain in the neural data.
Right. That's what's not going to happen. It's going to be a much more. Just like when you close your computer, there's no word document in your computer. There isn't.
Right.
It's a word document when you open it and you use it. And then it takes on that true representational format for use.
Now, that transformation that occurs when you open your commuter and you get a genuine representation that you can look at that neural data may One day tell us how we open up the representation in our heads to use like we use an external representation.
Just like computer science one. We will, we will need such an explanation. But it's not going. You don't go looking for a Word document in the computer before you open it. That's you could.
[01:05:09] Speaker C: Finding words. There's definitely a document in your computer.
[01:05:13] Speaker D: Exactly, exactly. This is so bizarre.
[01:05:18] Speaker E: It's not an external representational format. When Frankie says these, of course there's information.
There's information in your computer that can be turned into a representation. But you're doing it now. You're saying it's a representation. I mean, you have stuff in your head right now, Rosa. Right. That you're not saying like phone numbers from past homes and things that you've seen and you're not using them right now.
[01:05:43] Speaker C: You're talking about they're not a current. But I think I still have them.
[01:05:46] Speaker E: Them they're not, but they're not in that external representational format that is used when it's represented. It's in a stored format. What I'm saying is information through transformation can be turned into representation. But just because transformation of information can turn into representation doesn't mean you call the initial information representation before the transformation.
[01:06:11] Speaker C: I agree with that. But I think if it's sitting in a system where there's like a, where there's a well established mechanism for are consistently turning it into a representation every time, turning it into an occurrent representation every time, then you can say that it's a non occurrent representation. That's. Well, that's fine.
[01:06:26] Speaker E: But then we're just back to representation R and representation star. And what I'm saying is, is I'm just saying that what we need to do with the neuroscience is how do you get information plus transformation or process into a representation for use? And that's what the neuroscience should be after. Not prematurely providing the property to be explained on the information.
[01:06:50] Speaker C: But I think there's a difference between totally unstructured information that could never be used by the system versus something that is already structured. It's in like a standard format for retrieval and you could retrieve it at any time. Those just seem like very different things and I don't think we should mush them together. I think that the thing that is in a standard format that can regularly be retrieved is much closer to the current representation than it is to the.
[01:07:11] Speaker E: Unstructured structured information is a squirrely word. All I'm saying is there is a difference that makes a Difference before and after retrieval.
[01:07:19] Speaker D: I do think you really don't want to use computers as your case study to argue that there is no representation at the level of the hardware. What's remarkable about computers is that you're compiling high level language into lower level language up to machine language.
And that means literally that a high level language, when you've got an instruction in a high level language, that instruction.
[01:07:50] Speaker E: Exists at the level of the hardware, exists as information.
[01:07:55] Speaker D: It exists as a representation.
This is the difference between the brain and the computer. Computers have representation, brains don't, because brains, there's no compiling.
It's the main distinction between a brain and a computer.
[01:08:10] Speaker E: Looking at the representation in the hardware, it's not very possible.
What you're not understanding my point? This is the same mistake again.
All I'm saying is that.
[01:08:22] Speaker C: I know it sounds like you're saying it doesn't exist unless you're looking at it.
[01:08:26] Speaker E: You're just stipulating.
[01:08:28] Speaker D: But that's such a strange idea. Of course it exists.
[01:08:32] Speaker E: It's a process. No, of course the information exists. But representation, representation is a process. It's a thing that you do. It happens at the moment of use.
[01:08:42] Speaker D: You are just stipulating it. You're just stipulating the use. And you're basically out of thin air stipulating a use here of the word representation that you know, there's no reason that anyone would be granting. You know, look, I write, I have a map. No one's using the map. It's hidden. It's actually, I've buried it in my garden. No one will ever use it ever again. It's still a map of Paris, even if it's your cues right now.
[01:09:12] Speaker E: That's true, but I would just make the point. The point there is that what happened? I mean, I gave this example. If I. Let's say I'm trying to draw a map for all of you guys right now for how to get from my apartment in Lisbon to a really nice cafe. And I'm drawing it and half of it comes out of my pen. And then you distract me and I stop drawing. So half of it has come out onto the paper and the other half is about to be drawn by me onto the paper. Okay, What I'm saying is, is that I'm in the act of drawing it and I'm halfway through drawing it, it's in representational format as I'm getting it coming out of my pen onto the paper to join the representational format on the piece of paper. Now when I Draw that whole piece of thing. And I expressed it as a representation and I drew it. I completely agree with you, Edouard. It froze. It's like a frozen accident. That representational behavior is now locked onto a piece of paper. What I'm saying is that if I was distracted from writing that final half of the map and I thought about something else, I went, oh my God, I have to do this. And it's now gone.
It's no longer in my head in that representational working memory like format that it was being used to complete the map. And all I'm saying is the format it was in when I was finishing before interruption versus now no longer doing it, and I could do it, complete it the next day. Those are two very different things. And I'm saying one is representational at time of use, gets frozen on the paper. But when it's in my head before I conjure up again, it's no longer map like. Like the map stored in your cupboard. It's in a different.
[01:10:55] Speaker C: You think, it sounds like you think that memories, when they're not actively being recalled, are not representations.
[01:11:00] Speaker E: I think that they're in a format that can be transformed into a representation.
[01:11:05] Speaker C: But are you conflating stipulate that they don't count as representations?
[01:11:08] Speaker F: Because how is your representation, how is a Krakauer representation different from just a mental phenomenon? Because it has to be.
It sounds like when you think it, that's the representation. And so. And so how's it different from.
[01:11:23] Speaker E: That's right. I think you're right. I think representation is a conscious overt moment of use.
[01:11:29] Speaker F: So that's a very particular, specific definition of representation.
[01:11:33] Speaker E: It is, because it's the one that we all kind of are implicitly hinting at when we talk about thinking, planning, creating an AGI.
[01:11:43] Speaker C: But I don't think it's the only kind of first person representation that we care about.
[01:11:48] Speaker E: Right?
[01:11:48] Speaker C: Like if we're worried about Alzheimer's and not being able to remember things, we worry about them not being there to be retrieved, not just.
[01:11:54] Speaker E: Again, I'm not concerned about that. I'm not just saying that implicitly, what we are genuinely interested in when we talk about AGI, for example, is conscious overt representation, rich behavior that's happening. Of course we're interested in all the things that go wrong in the substrate and how it makes that transformational ability go away. Of course we are. I'm just saying that I have a problem using representational language, which is at least tacitly a lot of the time. Time referring to that kind of behavior. And then we imbue subpersonal processes with the same property.
[01:12:32] Speaker D: I think Louis wanted to jump in earlier and I cut him. So maybe we should give an opportunity to. To jump in.
[01:12:40] Speaker A: No, no, that's okay. I was sitting here. I don't know if any audiences are going to see this video, but I was doing like a Mr. Burr. And it's like, good.
I'm just enjoying watching the hate flowing through people.
But I'm just kidding. So if I had to describe John as a meal, I would say it's a meal that has a label that says very spicy and hot. But then when you take a bite into it, it's actually cool and creamy. So John says things like, I totally disagree with all of you.
And then he'll say things that are very sympathetic or along the same lines. But. But one thing that came across as spicy, but I think ended up being cool was.
I'm still not quite clear what John thinks the role of a neuroscientist is for explaining these kinds of phenomena. So John sounds to me like a methodological solitary.
So someone who thinks we can theorize about mental states on their own. And they are this kind of domain in which we need to explain that kind of phenomenology or those kinds of relationships to other ideas and conscious states and things like that. And then there's like, you know, and he said, you know, but of course there's neural activity that's kind of related to it. And so maybe the job of the neuroscientist is to kind of look at. For those correlations or the strongest term that I picked up from him was look at the transformations from the neural activity to those higher order mental states. But that just seems like extra crumpet on the side for him, but doesn't really tell us anything about what mental states are like or what representational mental states are like. Are like. So I'm still not quite sure what neuroscientists are doing for explaining that psychological scale. I think they're. They're useless for John Krakauer. We should all be psychologists, we should be Fedorians and just theorize about the mind.
[01:14:34] Speaker E: So why. So why can't I say the same thing to you with why aren't. When I. When Sherrington was doing the stretch reflex, why wasn't he doing particle physics to explain it?
So in other words, why. Why is it that you're. That circuit. Neuroscientists are not forced to be particle physicists when they do their work.
So why do psychologists have to be neuroscientists?
So. So are neuroscientists also methodological solipsists because they're not doing particle physics when they do their circuit neuroscience?
[01:15:09] Speaker A: I don't think it came across as hot and spicy like you. Yeah, probably, but.
[01:15:13] Speaker E: Yeah, but I'm just saying that that's what emergence is. I find it ironic that the best thing is effective theories. And basically psychology is an effective theory for mental phenomena, just like economics and sociology are effective theories. We don't ask people.
And that was and Anderson's main point in Moore's different in 1972.
[01:15:38] Speaker D: I think there's really interesting questions here about levels of explanation and the role of the notion of representation in tying. Or maybe actually we should be untying these levels of explanation. And I think that we would learn a lot people who are interested in neuroscience and psychology by looking at how levels of explanations are articulated in other domains of science. For example, in physics. Right. So in physics we also have a different level at different scales of energy, different models that can be related to one another.
And it's very unusually the case that one notion, I think that's going to be very much in line with what John was saying, that one notion at a higher level is actually found at a lower level. So when you have a different model at different scales of energy, usually the explanatory primitives stay at one scale. But it's not that we don't have any understanding of how the scales are related to one another. In fact, we often have very precise understanding in at least some domains of physics. Not everywhere, but in some domains of physics we have very precise understanding about how model at one scale of energy can emerge or can be the result of what's happening at lower energy.
[01:17:05] Speaker E: Which is why I said transformation. You know, in a complexity science, one has to distinguish three distant disciplines. There's discipline on level X with its vocabulary and explanatory primitives. And then there's the discipline with its explanatory principles at level X minus one.
[01:17:23] Speaker D: Yeah, no, that's right.
[01:17:24] Speaker E: Primitives remain at their level. There can be domain of science where you look for a congruence, a transformation between level X minus 1 and X. Yes.
What people mustn't misconstrue is that if you have that transformational understanding, as you correctly say, exists in some areas of physics, that that transformation will make the vocabulary of X no longer necessary, that it would be explained away. And that is what I think Louis is actually trying to Infer is oh, on and other philosophers. And if we just had the transformation and we had the lower level X must 1, we could change our primitives for level X. And that is what I think.
[01:18:11] Speaker D: Yeah, I don't think anyone here wants to say that.
I think another attitude which is not the one you're describing is actually hoping that you can get. Get, so to speak, a reduction of the higher level primitive. Let's say representation. You can say, ah, representation just happened to be this quiet complex things that I can describe in lower level primitive terms.
[01:18:40] Speaker E: That was what Louise was saying because he's saying I'm a methodological solar system. So neuroscientists are methodological solicists when it comes to physics?
Yep.
[01:18:48] Speaker D: I mean, I don't know Louise.
[01:18:54] Speaker F: Sure.
[01:18:57] Speaker A: Why not?
Yeah, I just, it just wasn't.
Yeah, it still wasn't clear to me.
It, you know, if, you know, if we're doing quote unquote, higher order.
Your study, higher order phenomena like what John's calling cognitive states with rich phenomenology.
I still don't know what the job of the neuroscientist is. Is it too. Is the. The neuroscientist supposed to elicit the transformations or illuminate the transformations that are related to the psychological states, or is the psychological researcher supposed to illuminate those transformations? Or is it supposed to be both sides?
[01:19:37] Speaker E: I said send out all your hounds. I said you can have confirmatory evidence at the neural level that can. We wrote this in the behavior paper in 2017. That can break the tie on psychological theories and help you update psychological theories. But when the update occurs, the language of the update will still be in the original X primitives that borrowed updating from the neural data. So confirmatory updating evidence for neuroscience to break the ties of psychological theories, rule out others and update them is fantastic, but the explanations themselves will not be in neural primitives.
[01:20:16] Speaker A: Does the updating go the other way too for you, John?
[01:20:19] Speaker E: I think so.
[01:20:20] Speaker A: So we can update our neuroscience theories based on psychological scale.
[01:20:25] Speaker E: Well, that's where temporal difference learning came from. I mean, temporal. Temporal difference learning with a mathematical theory long before we went looking for the neural data for it.
[01:20:34] Speaker A: Yeah, so I don't know that literature, but you know.
Yeah, that we have these levels that have not been explained by a lower level, of course.
[01:20:50] Speaker E: Looked at those traces, you know, and went, oh, that looks like temporal difference learning basically. I mean, I'm bastardizing the story a little bit, but was able to use mathematical theory and abstraction and see it in the neural data. And it was, you know, one of the most fruitful moments in neuroscience, the recent data. But it was. But if it hadn't been for his psychological mathematical insight, it would not have been as obvious when the neural traces were looked at.
[01:21:18] Speaker F: I'm missing how that updated the neuroscience primitives.
[01:21:26] Speaker E: I'm not saying the primitives, I'm saying it was. I think the question was what happens when psychological level theories can provide insight on how to look and be and interpret neural data. And I'm just saying I was making.
[01:21:39] Speaker A: A stronger claim, which is how I interpreted what you were saying, that you can arbitrate psychological theories by looking at neural data. And I was wondering, can we do that the other way as well?
[01:21:51] Speaker E: I mean, look, I thought that would maybe that one doesn't meet what you want. I would say I would be very surprised if it can't be a two way street, but I would be surprised.
[01:22:01] Speaker A: Now that's hard for me to reconcile with this emergentist effective theory approach.
Right. Because it seems like there's some sort of unidirectional relationship.
[01:22:13] Speaker E: I don't think that's implicit infective theories at all.
Reading off and there are, you know, we, you know, economists. It'd be very odd to say to economists that you really would be much better if you were at, you were chemists. And it'd be very odd to say to a basketball coach, you know, you did a good job bringing them to the NBA finals, but if you'd done some brain imaging, you'd been even better. I mean it was just be very odd. Right. And all I'm saying is every area, every discipline has its primitives, its ontology and we should allow that even though there can be another discipline that moves between them, that can help them. But that's a different science and they're just different disciplines.
[01:22:55] Speaker F: But isn't the grand goal to connect different disciplines? And yes, you use the same language like at your own level, but what you called earlier, a, I think the phrase was a complete disaster in terms of relating neural activity to mental functions.
I would call an ongoing project because it still is a goal to relate neural activity to mental functions, whether we're. Whatever language we're using in whichever level it is.
[01:23:26] Speaker E: But it's relates a bit of a filler to term. What does relate actually mean? Correlate.
[01:23:32] Speaker F: Okay, yeah, yeah. So have some explanatory power in terms.
[01:23:37] Speaker E: What does that actually mean? That's what I'm saying. In other words, that's what I'm saying. Is that it's very easy to utter such sentences.
[01:23:45] Speaker F: But what I'm saying is correlation, just correlation.
[01:23:48] Speaker C: I think you want to look for difference makers, right? You want to see like what changes at the psychological level if I make these manipulators, correlations at the neural level. Like that's one kind of, that's one kind of explanation that's stronger than correlation.
[01:24:01] Speaker E: Well, it's causal. It's causal, right? A stroke, you know, we know, I'm a stroke neurologist and we know exactly where the lesions are to make you aphasic. The different kinds of aphasia. And I can tell you all of that exactly where you will have a causal consequence. But that in itself is not going to tell you me much about language and how it's computed at all. It's a start, right? But it's hardly very richly explanatory for me to say that I can make you a phasing by putting a lesion in the inferior.
[01:24:30] Speaker C: I mean, that seems like a non sequitur. It seems like it helps explain strokes. If you want to explain language, then you want to know what kinds of manipulations do you do that allow you to change particular things about, you know, a stroke, more specific, more fine grained things.
[01:24:45] Speaker E: Okay, but it's. But I think in the end, lesion.
[01:24:48] Speaker D: Analysis, I think that's only so. I agree with Rosa. That's definitively more than correlations. It's of the right kind. But also if we look at again, I mean, I'm pushing back for having this not a myopic view about relationships in neuroscience and psychology and what's happening in other areas of science. Here's a common relation between models at different levels. Derivation under assumptions. So you take a model at a specific scale of energy and you assume, for example, that if there's an infinite number of particles, that model can be derived from a model at a lower scale of energy. And by derived, it's mathematical derivation.
So it's in some other parts of science more successful, I would say, at least at this point than neuroscience and psychology, because they have a longer history, maybe because their object is simpler, easier to understand.
You have literally formal derivation of models at different scales under various idealizations. So you need always to idealize. For example, you need to assume an infinite number of particles. You need to assume that some quantity goes to infinite, and so on and so forth.
But this is really what gives you the understanding in these other areas of science. It's not just a manipulation here, give you some outcome there, there it's literally, oh, that model is true at this level.
[01:26:15] Speaker E: Higher.
I totally agree with that. Sometimes that works, right? As my brother says, you know, sometimes you look under the hood and then you don't. I do find it kind of interesting though, that whenever these conversations are had, it's always statistical cyanodynamics and the ideal gas laws and the kinetic theory.
[01:26:32] Speaker D: No, that's not true.
[01:26:34] Speaker E: Where's the one that comes up? Give me another one then.
[01:26:39] Speaker D: So, for example, you can explain the kind of work that Bob Betterman has been doing about various emergence phenomena where they have a critical point and that's also of the same kind. Right. So we have a mathematical understanding about why the higher level description depends on the lower description.
[01:27:08] Speaker E: I'm not again, I mean, you know, I hear these arguments all the time.
[01:27:12] Speaker D: No, of course.
[01:27:14] Speaker E: All I'm saying is, is that I think we just have a very interesting discussion here, which is, what do we think in 20 years time for psychological language and things like mental representations and all the cognitive science of people like Chaz Firestone make all. All the people who do lots of great work in psychophysics and psychology who don't refer to neurons. Are we saying that other than this transformational knowledge of how you get one from the other, that when you construct explanations of these higher level cognitive acts that you are going to now start uttering sentences like you do for the stretch reflex with neurons in the sentences? That's the question is, will we put neurons in the sentences naturally, like we do now for stretch reflex?
[01:28:00] Speaker C: I think this happens in popular culture now, actually.
[01:28:03] Speaker E: Right.
[01:28:03] Speaker C: Like people talk about their dopamine levels, maybe not very accurately, but they do. Not very accurately. But it seems like, you know, they are happy to expand their everyday language to include terms that had previously only been in.
[01:28:17] Speaker D: But.
[01:28:17] Speaker F: But that's actually a case, I think, where people are imputing, like dopamine, giving dopamine. I mean, they're talking about dopamine on a psychological level and they're kind of sneaking it in. I don't think they're talking about dopamine as a brain process in most of these popular examples, but there is like. I'm kind of surprised, John, that you're not. So you're saying in the language of neurons, but we're beyond neurons. Right. We're in manifold land. We're in topology land. We're in shapes of neural activity that are getting closer as explanations. Line attractor land, like a line, you know, if you can map a decision along a line attractor in a Dynamical regime. That's a closer explanation.
[01:29:04] Speaker D: That's true.
[01:29:04] Speaker E: I mean, I, you know, when I wrote the paper with David Barratt. But again, the line attractor is not making a decision. That's a mere logical fallacy.
[01:29:11] Speaker F: I don't think anyone. It just sounds like you're actually.
[01:29:13] Speaker E: It actually does get used. I can again give you quotations.
What I'm saying is, I think that the idea that there will be primitives, dynamical objects, as David Barak called them, that you can combine to do a computation and that we get intuitive insight like the beautiful work that David Sicilo and XJ and others are doing, where you have some idea of these dynamical primitives. Yes, but we are looking from the outside in going, oh, that's a delayed match to sample task.
That's an alternative voice choice task, which we understand qualitatively. And then we see how you would do combinatorials with those dynamical primitives to construct that. But when I asked David Tassilo, who knows that, to combine them in that way, who frames the task to then build it? Ah, Stefano Fousey, beautiful paper in Nature late last year where there's a geometry for doing two contextual, you know, switching between two contexts. But the amazing thing in that paper from late last year was in one minute you could instruct a human. These are two contexts and one, rather than thousands of trials of training to construct that geometry, it was instantly configured. But the question of what's the upstream cognitive process that allowed the construction of that geometric structure? Completely unknown. So in other words, the actual stuff that we're interested in, this cognitive understanding is upstream of the phenomena that you're talking about.
That's what's puzzling is we don't have a neuroscience of that upstream conscious understanding bit. We see its products downstream, which are fascinating, don't get me wrong. I love that shit. Right. I'm just saying it's downstream of what we're talking about when we talk about mental representation.
[01:31:04] Speaker C: So.
[01:31:05] Speaker A: So that's as something that Edward pointed out earlier towards John. That's another kind of stipulation of that relationship. So you could have an identity relationship. And so this is, I think with Rosa, the example of dopamine. Right. The details might not be right, but let's tell a story that is right.
You know, Patricia Churchland, neurophilosopher, has this story that she talks about where she says, I came home and if I use my folk psychology, I would talk to my husband and say, I had a really stressful day.
I Need to have a glass of wine to relax. But she said, but there's no real principal reason why I couldn't say to my husband, my cortisol levels are up. I really need like some, you know, alcohol in me right now to lower the dopamine. And that's just. Isn't that the same thing? It's an identity relationship, right? It's identity between stresses, the cortisol levels or whatever is the. Right.
[01:31:55] Speaker E: But as John, as pointed out a long time ago, you know, saying pain is just C fibers.
Well, no, right. The C fiber side has certain properties that a third person and you can characterize. And pain is something else.
They're not equivalent. Right? They're not. Pain has the ontological, subjective, first person part. And yes, it can be causally related to C fibers. But to say that C fibers and pain are the same thing is beyond bizarre. Right? It's beyond.
[01:32:23] Speaker A: Because you're not a philosopher, John. This is not bizarre.
[01:32:26] Speaker E: To think he was the one that's a philosopher, pointed out the fallacy of that.
[01:32:31] Speaker C: John, now you're starting to sound a lot like some kind of mysterian, right? Like you. You say we want to treat, we ought to treat representation the way we treat consciousness. But it also seems like you're saying we couldn't even in principle make sense of representation in neural terms in the same way that, you know, hard problem people in consciousness say we can't make. We can't even in principle make sense of consciousness.
[01:32:53] Speaker A: Rosa made my point better than me.
[01:32:54] Speaker E: We'll come up with very interesting necessity insufficiency conditions. We'll update our psychological theories with neural data, of course. But just like we don't go around saying we should have physics explanation for stretch reflexes. Why don't we have this, why don't we have a podcast on that? Rose?
[01:33:12] Speaker F: Why does it need to be. Why is it always down to quarks? Like with. With levels, Right? So think about, like think about a neuron.
[01:33:22] Speaker E: All I'm saying is that what's happened here is that whether it's consciousness, mental representations, or stretch reflex, they're all part of the nervous system and they're all made up of neurons. So the idea is because it's made up of the same stuff, and because we've done quite well thinking about the stuff at the level of individual neurons and then populations, we should get the whole story up the neuro axis. In other words, the explanation should cash out in the same way because it's made of the same stuff. But that's Just what I'm saying is that that's not true. It's not true in physics. Right. Astrophysics and particle physics are different disciplines with different objects. And we have great trouble, as you all know right now, reconciling gravity with quantum mechanics. So in other words, all I'm saying is, is that the way in which neural data will be used to think about higher level cognition will just be used in a different way to the way that neural data is used to explain CPGs, saccadic eye movements and stretch reflexes. It's just going to be a different way we use the data. And I'm just saying right now we don't know how we're going to use it.
[01:34:27] Speaker F: Agreed.
Gravity versus.
What did you say? Gravity versus some other.
[01:34:35] Speaker E: We haven't yet got a unifying theory of the very large and the very.
[01:34:38] Speaker F: Small and the very small. That is will be part two of our conversation today.
[01:34:45] Speaker D: We're going to have to bring new hosts, I'm afraid.
So I also wanted to push back a little bit against. So there's been in the discussion and maybe accidentally, somehow, I think that's partly the way John frames the issues at the personal level.
There's a bit of an alignment of a bunch of different notions, personal, psychological and representational.
And I think these are different notions. Much of psychology is actually not partly concerned with the conscious or what people are aware of. Many of the representations that cognitive scientists postulate are not representations that people are aware of having. They're very different. So if you learn, if you work for example on reading, so transformation of graphemes onto phonemes, you know, there's literally 40 years of work in cognitive science on that topic. You're going to be postulating a bunch of representations that explains how a grapheme can be produced, can result in a phoneme. But none of that is meant to be. They're all representation, but none of that is meant to be connected to in a way of conscious life. Right. So much of psychology, when they're talking about representation, it's really not what John seems to think they're talking about. It's not this first personal experience of the world. It's not imagination, it's not memory, first person experience. It's really something quite different.
[01:36:22] Speaker E: Just to be not mischaracterized. When I go to Tyler Burge talking about mind, he talked about a particular form of representation and consciousness and it could be either or. I have no problem with there being unconscious representations. I very much. You know, I'm writing a paper With J. Quilty Dunn right now on language of thought, which can be implicit. All I'm saying is that. But. But it's very important to distinguish between those cases where there seems to be implicit psychological representation. Tyler Burge makes the same case versus the existence of implicit policies. And what I'm saying is a lot of what is called representational that is implicit can just be explained in terms of policies. They're not the same thing.
[01:37:07] Speaker D: That's a completely different question.
[01:37:09] Speaker E: Please don't say that. I don't believe that.
[01:37:12] Speaker D: I didn't say anything about you, John. I just say it's important to distinguish.
It's important to distinguish issues such as your experience of the world that involve, let's say, a first person grant and the kind of things cognitive scientists are doing where they postulate representation. The two are quite different from one another. And it's not quite clear to me what.
Allow us. What justifies postulation of this second type of representations. Right. You know, it's on your view, John. For example, it's very clear why we want to postulate imagination. We have first person experience.
[01:37:53] Speaker E: Just to be clear, I just want to say is it's absolutely true that when we talk about reflective mental representations of the kind that everyone would like to go after intuitively they have that form. But I have no problem.
I mean, you know, with Steve Fleming, we're writing something now on perceptual representation and how that might have been the base for conscious reflective representation. So I'm. Representation is a term that should have. Have properties that survive them being conscious or not. What I'm saying is that the kind of stipulations of the lot framework of representations, the perceptual stuff that Tyler Burge, Ned Block have done, neuroscientists, I can say those distinctions made by those philosophers, they never even talk about those very much. That's what I'm saying. They just go straight from V1 through to prefrontal cortex and use representational language all the way through. And I'm saying that there has to be a point where you go from sensation unconscious to perception and representation, which can also be unconscious, like blindsight. But there is a division there. Even in the unconscious realm when you use the word representation.
[01:39:09] Speaker F: I want to just jump in here. Sorry, sorry, Edward. But Frankie, I don't feel like I know you well enough to be calling you Frankie yet.
[01:39:16] Speaker B: That's okay, go ahead.
[01:39:17] Speaker F: Everyone else is. So I will also. I just. You've been fairly quiet and I just wanted to tell you to jump in if you have comments or what you've been thinking in the past 10 minutes of conversation. And of course, if you don't want to chime in, that's totally fine. I cut. I cut Edward off there.
[01:39:34] Speaker B: Okay.
Yeah. So, I mean, we've moved on now to personal level.
Let me just use that umbrella term for the sorts of representational talk that we use in or in our ordinary lives. I'm not talking about the science talking because I think the science.
I think if you look at the science, the point that I want to beat is that that representational talk and content attribution is always pragmatically motivated. So the issue then is what are the various purposes or functions that representational talk can serve in these various domains? And I think I've talked a little bit about the sciences, but I think that in belief attribution and explaining behavior of ourselves and of others, in thinking about perceptual experience, we talk in representational terms. What we're doing there is we're modeling.
Not sure how. I mean, this doesn't necessarily connect with the dispute that we've just had, but I think that we're modeling mental processes in the various domains in terms of representational notions that apply primarily, I think, in language.
And what we do is we model inaccessible mental states and processes in terms of these publications, uses of representation that we understand. So I think that belief processes, we think about them as inferential processes. They're not. They're not inferential processes. They're causal processes that can be modeled by relations among sentences. Those are inferential processes.
With respect to perceptual experience, I think we do talk about representing the world. World.
I've got a completely different account of that.
[01:41:28] Speaker E: It's.
[01:41:28] Speaker B: It's a type of adverbialism. I don't. I think that we model our internal experiences, our perceptual experiences in terms of what's going on out there.
We use the external world model, our ways of conceptualizing the external world to talk about our perceptual experiences. And that serves our purposes pretty well, but it doesn't mean that. And again, one of the points, general points of. Of content attribution or representational talk is allowing evaluation for accuracy.
So with respect to modeling mental processes that are involved that underlie kind of action and the explanation of behavior, we can model those processes in linguistic terms. We can model perceptual processes, perceptual inferences, what we call perceptual inferences in terms of. Of. So let me put it this way we can characterize our experiences. What kinds of experiences are like what, what kinds of experiences lead to other experiences and so on in terms of categories that are based on our, our experience, our, our understanding of external objects and properties. So again, representational talk. We move from what's out there, we model it in terms, we model our, our inner lives in terms of what happens to our external lives.
With respect to the question how does the science hook up with either our ordinary ways of understanding high level experience, high level thinking, or how does it hook up with the sciences, the kind of fairly primitive sciences of those things?
I think that's an open question, but I think I'm pretty convinced that representation is going to play the same sorts of roles as we make progress on understanding kind of the level of, at the level of personal experience, at the level of thought and action, the same sorts of roles that it plays in the sciences, namely simplification, being able to talk to each other about it, about private experiences, in terms of the, our understanding of the public shared world, evaluating our experiences and our thoughts for whether they're accurate or not, and so on. So I see kind of representation as playing the same sort of role that it plays in the sciences, in everyday life, in thought about personal level experience and so on.
I don't know if that's, it's just kind of my two cents on how the sciences and these higher level.
[01:44:20] Speaker F: Just wanted to bring this up in case so many of the ideas that you were just discussing likely be in your book. So I'll talk about it in the intro also. But I just thought I'd do a fancy screen share real quick to show.
[01:44:33] Speaker B: People the bottom's cut off. So it's mental representation. Thank you for that.
Anyway, that's all kind of chewed over in the book.
[01:44:42] Speaker D: Quick question for you Frankie, following on what you say.
It's really interesting when you say that in everyday life, in a non scientific context, we explain behavior by means of presentation.
What do you exactly mean? So of course we provide intentional explanations of what people do in terms of beliefs, desires, in terms of what they want, what they need, what emotions they might have, which are self intentional entities.
But I would have thought this is not quite the same as explaining in terms of representations. I mean, you know, one might have a view that of course there's intentional explanations, but that's another step to just say that and thereby you're committed to anything, like for presentations. I mean my, my colleague, now retired, John McDowell had that kind of view. Right. So of course Intentional explanations are just fine. People have beliefs, desires, in some sense, you explain them. But this is not representational. When you say it's representational, you're actually bringing something else.
So, any thought about that, or do you just want to collapse intentional and representational?
[01:45:58] Speaker B: Well, I think that we explain behavior by positing in terms of beliefs and desires. So, yes, and I think that those explanations are typically true, but I think that what we're explaining behavior in terms of are either complex causes or dispositions, sets of complex sets of dispositions. And we model those in terms of the sorts of relationships that hold among sentences.
So I don't think common sense psychology is representational at its base. I think that what we attribute are certain very complex causes, and we characterize those. We can't get in the brain and discover those causes, but we. We characterize them in linguistic terms, in terms of that clause. Attributions, content attributions. And we model causal relationships that hold among beliefs and desires in practical syllogism, for example, by the relationships that hold among linguistic objects.
[01:46:55] Speaker D: Objects.
[01:46:56] Speaker E: So, Frankie, just so I understand, so when you know when you're drawing those fantastical ibex on your farm, right. And you draw one foreign. Sorry, my bad. Sorry.
[01:47:09] Speaker B: Very important.
[01:47:11] Speaker E: No, I know, I know.
[01:47:12] Speaker F: Jesus, John, that's.
[01:47:13] Speaker E: I know.
So when you're drawing an orange for me, is that not representational behavior?
[01:47:20] Speaker B: Yes, behavior is representational, for sure.
[01:47:23] Speaker E: So you. So the act of drawing a picture for me to correct my egregious error is. I mean, that seems to be very psychological behavior to me.
[01:47:35] Speaker B: So certainly my drawing of the Oryx is a public representation.
[01:47:40] Speaker E: But you did it, right?
[01:47:42] Speaker B: I did it. So I produced a public representation. I produced it because I've got that. I mean, how did I produce it? Well, now we're in the realm of, like, what processes were going on when I produced it. And we can characterize those internal processes in representational terms in a way that I've been talking about. But there's no question that I produced a public representation, right? So I don't.
[01:48:07] Speaker E: But that's like.
[01:48:09] Speaker B: I guess. I think representation has its home in kind of public representation in language. And we use that. So that's kind of the home base. And we use that to talk about all kinds of things that are. That are not public.
We can talk to each other about our perceptual experience, and we can talk about them in terms of stuff that's out there.
[01:48:31] Speaker E: People with brain injury can selectively lose those kind of representational behaviors.
[01:48:35] Speaker B: Right, right, right.
[01:48:38] Speaker E: So, I mean, I.
I can't believe, you know, I think we're going to have to have some neural explanation for the tissue that is responsible for being able to do that remarkable construction of a public representation. And we've done work showing that you can lose the ability to represent shapes that you can subsequently draw.
Right.
[01:48:58] Speaker B: I mean, I agree with all that. Yep, that's right.
[01:49:02] Speaker E: That it's a unique representational ability. Just unique in just the same way languages. And it's got nothing to do with language. And we. And I, you know, just to, you know, kick back at Louise, you know, I believe in a neural story one day for that ability, but I think that the word representation should be for that ability, and you're not. You shouldn't ascribe that ability to the neural explanation itself. It's going to have to come in different language.
[01:49:28] Speaker B: Yeah, I agree with that. That we agree on that.
[01:49:32] Speaker D: Quick question. Why do you think it's important to describe it in representational terms? Isn't that just an intentional activity? So I can imagine. I'm not sure what it adds to sets of representational activity. I'm not sure what the word representational here.
[01:49:47] Speaker E: It's what Frankie said in her book. Right. It's that the thing about representations, and I can't remember your list of three attributes, Frankie, I'm not going to say it. You're here. But there are three things that. That they have these properties. And what I'm saying is that those mental representations have exactly those properties. You can imagine a shape. You can operate on that shape. You can operate on it just the way that you can amend a drawing. So it is literally a representation that you're operating on. It just happens to be in your head rather than on the page. It's literally. Eduard, a representation.
[01:50:21] Speaker B: Yeah.
[01:50:22] Speaker D: It's really not a representation.
It is.
[01:50:26] Speaker B: I agree that it's not a picture.
Agree that it's not a picture in the head, but it is not a picture.
[01:50:31] Speaker E: What I'm saying is it you. You. We are able to open the computer, the word document in our head. We can look at the word document in our head in the same way that we can look at it on the screen. And that is odd that we have that ability that we can literally conjure something up, operate on it the way that we do on a drawing.
Now, we don't know how that works, but that's what I mean by representational behavior is that's our superpower.
You know, Gaudi imagined the Sagrada Familia in his head before it ever laid the first stone.
Now how is it that that's possible, that we can do that?
Right. That is what I'm saying.
[01:51:12] Speaker D: So I agree this is a very interesting set of capacities. I agree that it has some interesting similarities with the use of external representation. So I'm actually quite happy with that. I think that's really interesting. I'm not quite sure that everything you want to call representational has those features. So I think the imagination here might actually be somewhat an unusual case. Probably a lot of different things that don't quite have all these nice properties that you find in imagination where I can move things around in my head.
[01:51:47] Speaker E: People like Dan Dennett, when I would have arguments with their language, got all the credit for why we have zoom and books and arguments like this. And language was like an infection in a regular primate brain that made it suddenly do all this work. And I'm saying the representation is as powerful a concept, I agree with Bill Ramsey, as languages. In other words, we really do have this ability to represent abstractions and operate on them. And I think that, that, that is why I defend it because it seems to get less credit as a concept that we have that language does. Language is something we use as an excuse for our abilities far more than our unique representational ability.
And I'm just trying to address that balance. It doesn't mean that there aren't unconscious ones. Absolutely. That's a fascinating topic. But I'm just saying that representation, because it's become just neural correlates and information, it's lost. That weird thing that you've just acknowledged that which I can see disappear in patients with everything with language remaining completely intact.
[01:52:57] Speaker D: I think Louis wanted to add. To join.
[01:52:59] Speaker A: Yeah. So what maybe what an aspect of what Edward was getting. That was a question that, that I had, which is, you know, I just wonder, you know, people like John do is. Is all mental life expressible in representational terms? Right. Are there non representational features of our mental life going from the conscious to the non conscious to the cognitive to the non cognitive, the imaginary, the non imaginary, the vertical, the non vertical. Are all of these to be described in representational terms? I'm happy to hear. Frankie just put a hand up. I'm happy to hear what you have to say, then I can finish up later.
[01:53:36] Speaker F: And then let me just say we're coming up on two hours and so maybe in a minute we'll do like a wrap up. And Rosa, I don't. This is probably way too much for A new mother to bear for so long here. But. So we'll ask you first, but. Sorry, Frankie, go ahead.
[01:53:50] Speaker B: So, just a really quick point. I don't want to leave the impression that I think that all of these capacities and abilities, what we might think of as the personal level, are linguistic. The point rather was that we understand them in.
In theorizing about them. We model them as linguistic processes. But it's a different question.
It's an important question. What properties of mental states are actually getting modeled as linguistic properties? And that's a big issue.
[01:54:22] Speaker F: John, you were going to respond, I think, to this also.
[01:54:26] Speaker E: I was going to say that I.
I think you're right. And Edward made the point, too, that if we're going to talk about external representations as our model for internal ones, whether they're words, drawings, pictures, algebra, numbers, to the degree that a lot of our thinking operates on numbers, symbols, pictures, sentences, I think representation is a useful, you know, language of thought. Notion of representation is a useful, universal way of thinking about it.
So, yes, in a way, to the degree that there are external clues to the representational work that we do in our heads, whether it's pictures or math or poems, I would like to think that the generative process that led to those external representations were themselves representational.
I don't know whether there are things, you know, off the top of my head where they're never going to have an external version of themselves. Whether those should be called representational or not. I haven't actually. I don't want to be a completist about it, but I think to the degree that many thoughts can be expressed in different types of external representations, they have their analogues inside the head.
[01:55:53] Speaker F: Does anyone want to comment on that before I demand that we. Okay, Rosa, I'm going to put you on the spot first here, closing thoughts in general. But did we make any progress here? Did any of us move our opinions? Did we learn anything?
Are we going to agree to all disagreements?
What are your thoughts? That's a lot to ask of you. You can pick and choose among any of those things. Some closing thoughts.
[01:56:21] Speaker C: I mean, I feel like I got more clear on what John's view was. I think I agree with what Louis said, that it sounded sort of outrageous and authoritarian at first, but actually it's quite reasonable once he articulates it in more detail.
Yeah, I guess what I took away from this is I'm still quite resistant to the idea of trying to legislate how people use the term representation, even though there are these sort of knock on consequences of equivocation and confusion. Like it just seems to me that there's such a diversity of enterprises where people use the term representation that it really makes sense to. I think it really makes sense to try to get clear on how it's used in each of those domains. Domains and to try not to equivocate between them, to not offer a promissory note for a future explanation. As an explanation that we already have, that seems bad. I agree with John there. But really trying to get clear on how it's used in different domains is really useful. And I guess I'm just going to use this opportunity to advertise.
I was part of this generative adversarial collaboration at CCN a few years ago where a bunch of people, neuroscientists and computer scientists and cognitive scientists and philosophers got together and tried to come up with a menagerie of different ways that the concept is used and that collaboration is finally publishing its thing. And we're trying to give a list of a taxonomy of different uses of representation, why they're useful in different ways. And the one thing that they all have in common is that, that they have to be used and they have to be usable. And I think that relates to the thing that John was saying about the interesting process as sort of being upstream of the stuff that you find in neuroscience. Because what makes something used or usable is that it's part of this larger system that we can think of as doing these interesting things, these representation hungry activities. And it's only in that context that it makes sense to talk about representation.
[01:58:25] Speaker F: Frankie, I might ask you to go next because that sounds very pragmatic and I don't know if you have anything to add to that and, or closing thoughts. Have you changed your mind about anything? Did we make any progress, et cetera?
[01:58:40] Speaker B: I don't really have a lot to add to what Rosa said. I agree pretty much with everything that she said. I'm looking forward to seeing the results, results of the work she mentions.
And I guess what I've. I've learned a lot and I've. I'm interested in kind of just continuing to find out more about what's going on in the empirical sciences. And thanks for this opportunity. It's been a lot of fun.
[01:59:05] Speaker F: Well, thank you for being here, John. Did you learn anything today except, you know, the correct animal on Frankie's ranch or farm?
[01:59:12] Speaker E: Oh, I did know it. Don't ever. I'm never going to be forgiven and I'd never get the Invite that I was waiting for now.
I mean, what I learned is actually that my concerns are somewhat valid, right. That I think that there's a usefulness to the notion of mental representations which can be got at with effective theories of cognitive science.
You know, my friend Chaz Faistin always says, don't call me a cognitive neuroscience, I'm a cognitive scientist.
That you can do cognitive science without referring to neurons because you can have effective theories of representational behavior. And I certainly have done work on that. But I don't want anyone to think I'm anti neuroscience. I love neuroscience and I think that neural evidence, mainly confirmatory, will have a lot to say about mental representational behavior.
My only concern is a misunderstanding when it comes to manifolds connectionism, single neurons, where somehow by using the representation word on those neural data, that somehow you're much closer than you actually are to a true neuroscience of mental representation. And that's why I feel like in fact the use of the word, which I would never police. In fact, we wrote an article for the transmitter on this. I never would suggest policing it. I'm just saying, saying that I think I'm correct, that it leads to deep conceptual confusions about how we're going to in the end properly link neuroscience and psychology.
[02:00:48] Speaker F: Edward, that's. That was surprising to hear John say he thinks he's correct, wasn't it?
[02:00:52] Speaker E: It was.
[02:00:53] Speaker D: No, it's not.
[02:00:53] Speaker A: Right.
[02:00:55] Speaker D: So something I really, you know, I said earlier and I want to, to highlight again is how much agreement there is among us about some of the crucial issues. Right. So we should not be naive about neural representation. We should just not assume there's very trivial link between the neuroscience or what are called neural representations and psychology. And I think there's a great consensus here which I don't think is reflected in the whole of philosophy and I suspect also in the whole of science. So I think that's, that's really quite remarkable here and I think that's worth highlighting. We all agree about that much. I think the place where we might disagree, but I'm not entirely sure is about the value of having imprecise and loose concepts. I think there's some of us who, and I think John might be one of them, Rosa might be on the other side, and I'm not sure about Frankie.
I think loose concepts have a fundamental place in science. I think they allow knowledge, information to circulate in a somewhat unregulated manner.
It can breed mistakes, confusion.
But in the grand scheme of things, this is actually the way science tends to make progress by ideas jumping from one area to the other earth by means of equivocation. I think that's extremely important for the good Marshall science. So I tend to be here a bit more of, yeah, let's 1000 flower bloom less confusion, do its nice trick, and that actually might be for the greater good.
And I think that there's tons of cases in science where that has happened. So I think that's a place where there might be a disagreement about this normative take on confusion and ambiguity in, in science.
[02:02:54] Speaker F: I could be wrong, but that's. That's a surprising thing for me to hear from a philosopher. It's. It tends to be the other way. Right?
[02:03:02] Speaker D: That's right. I've changed my mind actually over the last five to six years on that. On that very matter. So I used to think restricting and regimenting was actually what philosophers should be doing. But I've become actually confused, convinced that it's actually not the case. That actually has a great value for or I know, semantic drift and actually words being used in somewhat vague manners. And I think that's both in everyday speak and in science too.
[02:03:32] Speaker F: Louis, I hope you're happy with yourself bringing all of us together here since this was your inspiration you're doing. Closing thoughts from you.
[02:03:42] Speaker A: Well, thanks again, Paul. I'm an emotional vampire, so when I see people get worked up, it just really feeds me.
I've seen people get upset. I just, I feel like I've been buffeting today.
So thank you for this opportunity.
You know, I'm gonna, I'm gonna go an opposite view of Edouard, since we're a team on this. He can be good cop with the Kumbaya and everyone, you know, we can let a thousand flowers bloom. I just want to let the good flowers bloom. I think some of them might actually be weeds that we're thinking are flowers. And so I want to weed out the weeds and really facilitate, make room for the flowers. So, you know, I think Eduard and my. Our project, you know, we're the embodied people who did it. But I think there has been want and concern about trying to get some systematic evidence for how these terms are used.
This conversation has further illustrated how people working on these topics continually talk past each other. There's like a lot of work that needs to be done with like saying like, you know, what do you mean by your terms in this case? That doesn't mean we have to pull it police all universal uses of the term. But when we're trying to have a debate Getting clear on the terms is essential. And I further learned that, you know, I say we go scorched earth and get rid of representations. Maybe just talk about other things. I'd like to plug Rosa's paper. That was a reply to Edward and my paper, where she lays out this very nice list of different kinds of uses of the term representation. And as I said to Rose in the past, fine representation can mean this, representation can mean that. Why not just say this or that?
That's, you know, my. My kind of view to take from it. But, yeah, just happy to spend this time chatting with smart people about this.
[02:05:24] Speaker F: All right, guys, we did it.
Thank you so much for joining me. And this was easier than I thought it would be, so that was. That was nice for me. But Rosa and you, you hung in there, so thanks for hanging in there for so long. I know you've got some other things going on right now, so congrats, Rosa.
[02:05:41] Speaker E: That's awesome, by the way.
[02:05:43] Speaker F: Okay, take care, everyone.
[02:05:44] Speaker D: Thanks again and thanks, Paul, for having us. That was actually really, really, really nice.
[02:05:51] Speaker E: Thank you, Frankie. And you, Louie. Yes, and we should.
[02:05:54] Speaker A: We should.
[02:05:55] Speaker D: We should get lunch. We should get lunch, Paul. After we're neighbors.
[02:05:57] Speaker F: Sounds great. Sounds great. I want to hear who's been talking about me behind my back, so.
[02:06:03] Speaker D: Only next week. All right, take care.
[02:06:05] Speaker B: Bye Bye.
[02:06:13] Speaker F: Brain Inspired is powered by the Transmitter, an online publication that aims to deliver useful information, insights, and tools to build bridges across neuroscience and advance research. Visit thetransmitter.org to explore the latest neuroscience news and perspectives written by journalists, journalists and scientists. If you value Brain Inspired, support it through Patreon to access full length episodes, join our Discord community and even influence who I invite to the podcast. Go to BrainInspired Co to learn more. The music you hear is a little slow, jazzy blues performed by my friend Kyle Donovan. Thank you for your support. See you next time.
[02:06:55] Speaker C: Sam.
