BI 197 Karen Adolph: How Babies Learn to Move and Think

[00:00:04] Speaker A: My diabolical scheme is to use simulated robots to understand behavior, but their diabolical scheme is to use developing behavior to build better robots. You know, like we saw children do things in hammering the peg that no one would ever have imagined that a child would do. We actually, right now, we as a field, because if Kieran Adolf can do it, anyone can do it. We have all the tools and technology to do the whole enchilada. [00:00:43] Speaker B: This is brain inspired, powered by the Transmitter. Hey everyone, I'm Paul. I'm a neuroscience researcher at Carnegie Mellon University. I also have children, so I was witness early on to their tiny bodies fumbling about the world as they learned to crawl and then walk and then run and jump. But I haven't seen nearly as many babies do that as my guest today, Karen Adolph. Karen runs the Infant Action Lab at New York University nyu, where she studies how our motor behaviors develop from infancy onward. We discuss how observing babies at different stages of development illuminates how movement and cognition develop in humans, how variability and embodiment are key to that development, and the importance of studying behavior in real world settings as opposed to more restricted laboratory settings. We also explore how these principles, together with simulations, can inspire advances in intelligent robots. Karen has a long standing interest in ecological psychology, which we have discussed multiple times over the past month or two on this podcast. And so she shares some stories of her time studying under Eleanor Gibson, the developmental ecological psychologist, who was a colleague of the well known ecological psychologist, the, what would you say, Originator perhaps of ecological psychology, James Gibson. So she talks about her time with Eleanor and some of her other mentors. And finally we get a surprise visit from Karen's partner, Mark Blumberg, with whom she co authored an opinion piece arguing that quote unquote, motor cortex doesn't actually start off with a motor function, oddly enough, but instead processes sensory information during the first period of Animals Lives. Okay, I linked to the papers associated with the topics I just mentioned and which we discuss in this episode at BrainInspired Co podcast 197. Those church bells in the background tell me that it's time to quit this introduction and get into the episode. So thank you to all my Patreon supporters. Learn more@braininspired co how to do that and join our Discord community and even influence who comes on this podcast. And thank you as always to the Transmitter for their awesome support of this show. All right, here's Karen. Karen, you're busy. You've been busy for a long time. Do you feel just overwhelmed all the Time. [00:03:15] Speaker A: No, I have the best. [00:03:16] Speaker B: I mean, I know you were just talking. [00:03:18] Speaker A: I have the best job in the world. [00:03:20] Speaker B: Is that right? Why? Why humans? [00:03:24] Speaker A: Why human? Why? Why do I study human behavior? [00:03:27] Speaker B: Yeah. As opposed to some other animal model? [00:03:33] Speaker A: It was why? Because you have to know your organism. And so I worked my way through college working with three and four year olds. And I really know young children and babies. But I'm about to do a piglet study in September. Pigs at the edge of a drop off. And we got some chick stuff on the horizon. It could be any animal. You know, pretty much anything I've learned about any animal, they all produce amazing beautiful behaviors that learn and develop. So could have been any animal. It's just I have a better feeling for the human animal. [00:04:18] Speaker B: And has it always been development in particular that you've been interested in and the behavior? [00:04:25] Speaker A: Yeah, I'll tell you why. So I was, you know, I was one of those children that was your parents worst nightmare. Well, among them. So I went to four colleges. By the time I was at Sarah Lawrence my fourth, I was paying my own way. And I ended up. It's so convoluted, but I ended up at Sarah Lawrence because I wanted to study printmaking with this famous amazing Japanese woodcut printmaker on Sayu Chima. And he told me that everything he had done to date at that point I believe him. I mean he was right, was like self indulgent drivel. And that I needed to become a better draftsman. And so while everyone else in the class was doing other things, he would bring me things to draw. And they were usually like some branch, you know, like some twig of something, that sprig of something that he plucked on his way into the studio. And he would put it down and say draw it. And, and then, and I always came up short and he said, you need to just still yourself and really look so that you can understand how it grows. And. And I got really kind of pretty good at doing that. And. And then through another convoluted story, I ended up doing my doctoral work in part with Eleanor Gibson. So I considered primary advisor and her main advice was actually the same, but only about behavior. Just shut up, sits still, clear, open your mind and you know, let the behavior speak to you. [00:06:24] Speaker B: But she wasn't telling you that your work was drivel. [00:06:27] Speaker A: Oh, she sure did. Yeah. [00:06:29] Speaker B: Did she? [00:06:31] Speaker A: She mostly, she had these things. She would say that our work was fine, just fine or punk and drove me crazy because I never knew whether fine was better than just Fine or just fine was better than fine? And I'd say, jackie, Jackie, which one's better? Am I striving for just fine or fine? I don't care, you know. So I'll tell you a funny story. Like I'm, I think I was my second faculty job, I moved to nyu, you know, then that means I was tenured. And so I was doing fine. I was doing just fine or fine. I know, yeah, one of those, one of those. And it was the olden days, so we had answering machines and the light was flashing and I pressed play and it was Eleanor Gibson saying, you know, Karen, dear, thank you so much for sending me the chapter article. Whatever you wrote, I think it's terrific. And it was so fun to read. And I'm like, terrific, terrific. And you know, she said a few other things. Then I pressed the thing again. Karen, dear, I took another look at your article and while it's fine, there's a few things we need to discuss. And then literally there was a third message and she said, we really need to talk before you send this. [00:08:03] Speaker B: That's funny. [00:08:03] Speaker A: I don't know that I was a disappointment to her. I think she loved me. That I know. I don't think she considered me her best student, but in the end I might turn out to have been our best student. We'll see. [00:08:18] Speaker B: Yeah. Well, how did you did that kind of criticism? Because some people would wilt under that kind of criticism early in their careers. [00:08:28] Speaker A: Not a wilter. You know what? When I wrote my dissertation, my three advisors were Ulrich Neisser, Dick Neisser, the father of cognitive psychology, Eleanor Gibbs, Jackie Gibson, whatever, the mother of all of behavioral, you know, whatever, and Esther Thielen, who invented dynamic systems. And all three hated it, but for different reasons. Connor Gibson said, karen, I had to start, I had to stop reading it. The title, it just won't do. I couldn't read past the title. So that was pretty bad. Dignicer. It was the way I presented the results. And Esther Thielen kept saying, but I've written things with you. What's wrong? Did you write this? I mean, did you write because I've written things with you. And like what was that? Did you really write that? Like, why would you. Karen, why would you write that? So, wow, you know, it was the olden days. We had the 5 inch floppy disk. I just took the whole thing and all the paper cut, threw it in the trash, started over, said Jackie. How about this title? Okay, there you can continue. Esther, I'm going to show it to you section by section. She would give me the thumbs up and then got to the results and dick nicer. [00:09:52] Speaker B: But that's intense. Beneficial training in the long run. [00:09:56] Speaker A: Yeah, but I'm not a welter. No, I'm just get back on and keep going person. [00:10:05] Speaker B: So did your. So I haven't published these episodes yet, but I have a few episodes specifically on ecological psychology. [00:10:12] Speaker A: Yeah. [00:10:12] Speaker B: With a few people. And did your interest in ecological psychology begin with Eleanor? [00:10:18] Speaker A: No. So there I was at Sarah Lawrence, working in the Early Childhood center, paying my way through college, and I took a class on perception because I thought it would help me to understand. And this is the kind of thing that all the. Okay, so Sarah Lawrence had just gone co ed and there's like five men, all gay dancers, whatever. And all these women, we all look the same, you know, hair to our waist, black clothes wafting through the campus. [00:10:51] Speaker B: What year is this? [00:10:54] Speaker A: What decade is this? You know, and meeting in the pub, smoking cigarettes, discussing like. But how do I know it's a rock? And deep questions like that and Wait, what did you ask me? [00:11:13] Speaker B: Just about your interest in ecological psychology. [00:11:18] Speaker A: And I was thinking perception with this guy named Bob Becklin, who had been a doctoral student at Cornell. So he knew James Gibson very well, and Ulrich Neisser was his dissertation advisor. And then he knew Eleanor Gibson, like as a bipartisan. She wasn't on their official faculty then because of nepotism. So it's a whole other amazing story. But anyway, knew the Gibsons, but he. So. So the first semester was a traditional perception class and it was supposed to be a two semester, like, you know, fall and then spring semester, long class, and I went to him and the end of the fall semester, after we'd learned, you know, about the chambered eye and the blind spot and all these optical illusions, whatever the hell you teach in traditional perception. And I went to him in tears and I said, bob, I have to drop your class. I just know this is not how we see. It's just everything about it feels wrong. This is not for me and it's not helping me to. I wanted to understand how you get light into a painting. And so, you know, like, it's not helping me to understand my painting, you know, Bye. Bye. And he said, well, all right, before you leave, just read this book over the Christmas break and then see if you still want to drop the class. And he gave me James Gibson's the Senses Considered as Perceptual Systems. And it was like I found religion copied most of the book into my journals, like, because I had to give the book back. And you know, and so he was. He died in 79 and Honor Gibson was still alive. And so I wanted to, in the end, when I decided to do a doctoral program, I wanted to work with her. [00:13:15] Speaker B: One of the things I've asked a recent guest about is. So Louis Favela is the guest I'm talking about right now. He's written this book that's trying to connect ecological psychology and the neurosciences, which is all well and good. One of the interesting things to me though is that, yes, I know Eleanor continued the Gibsonian work, but ecological psychology, I guess Gibsonian, ecological psychology, it's like from one person and has essentially remained the same over time, correct? [00:13:51] Speaker A: I don't think so, no. [00:13:54] Speaker B: I mean, because James Gibson is always, you know, the person mentioned and then just the concepts, you know, the affordances, continuous perception, action cycle, direct perception, those have all been core tenets. And it's just odd to me that from my perception, and you please correct me, that those core tenants have remain, remained in place. Whereas I'm so, I'm thinking about it in comparison to neuroscience, which has had many, many, you know, different influences, but ecological psychology all like, seems to go back to one person. And I know Eleanor doesn't get as much credit as she deserves. [00:14:34] Speaker C: Right, but. [00:14:35] Speaker A: Yeah, so you're sort of correct and you're sort of wrong in an orthogonal way. I mean, they were married, she read everything he wrote. They published only two things together, and she has publications spanning 70 something years. She was not his partner. She was not. And she was not his muse. She was not anything. She was just her own separate person. And he was really, I mean, I never met the man. I only heard about him through her. And I've read his stuff. He was more interested in kind of epistemological questions. And she, she was really like, shut up. Look at the behaviors and let the data speak to you. Like, that was her, you know, she started as a comparative psychologist and she studying. She would have, she would have been studying non human primates if they'd let her do that at Yale. And she ended up starting out, you know, in a rat lab with, you know, so. So some of James Gibson's ideas are now so enmeshed into the literature that people don't even know that it's him. You know, like optic flow. Yeah, but what's really behind that is you can only understand a process which is a thing that takes place over time. If you study things that take Place over time. And so all his notions are really these time based notions and you know, about patterns over time, not stopping time in some fictional way, like as if it's a photograph, it's continuous. Some of his ideas like Affordance are also like kind of out there in the literature, but in a much sloppier way. Yeah, so when I talk with young roboticists, for example, well, there's a generation that I just get like free, free revering from them because you know, you do the, you know, Eleanor Gibson, you know, but, but, but they all, you know, beyond the term affordance around, they have no idea what they're saying, where it came from and they mostly it's meaningless the way they're saying it, you. [00:17:12] Speaker B: Know, because it's gotten too watered down. [00:17:16] Speaker A: It just could mean anything. It's like a helping verb, you know. [00:17:19] Speaker B: Okay, what does it mean? [00:17:22] Speaker A: Well, I think I know what James Gibson thought it means. I think I know what Eleanor Gibson thought it means. And then I think I know what I think it means. And those are three different things. [00:17:34] Speaker B: Wow. See that's, that's a problem, isn't it? [00:17:37] Speaker A: Why? [00:17:37] Speaker B: I mean the neuroscience is, for example, I'm a neuroscientist and we're sort of famously bad at using words that we all mean different things by them. Like the word representation, right. That has a thousand different meanings. And so what gets lost, what happens is people are just talking past each other all the time. And I thought that that would not be the case in ecological psychology because I know this is not fair, but you know, it all kind of originates in this very narrow, narrowly defined way. And I know that, I know that James Gibson like changed his notions of things over time as well. [00:18:17] Speaker A: But I mean, I guess I would say sure, I'm an ecological psychologist, but no, I'm just myself. And if anything, I'm first and foremost a developmental scientist and it's not in conflict with ecological psychology. And so like the sort of pre theoretical assumptions all come from there. Something like direct perception. I don't give, you know, I don't care. I'm not, I'm not committed to. [00:18:58] Speaker B: I. [00:18:58] Speaker A: Guess the only things I'm committed to I think are absolutely facts. And anyone who's not committed to those things is an idiot. Like we live in real time. That's the way it happens. And if you don't, if you want to understand a process, you have to understand how change happens over time. The end. It's a fact. You know, an affordance to my mind is a fact. It's the fit between a set of characteristics in the environment, a set of characteristics in a particular animal that make a particular action possible or not or possible with some probability. That's a fact. Like, I don't think anyone who thinks about it would argue with that. Whether people can perceive affordances for action and how they perceive it, that's still up in the air. And I don't know the answer to that. Probably we have to perceive affordances because otherwise I don't see how any animal can, you know, get around and fit through a hole, you know, be like a functional animal in an environment. But affordances are a fact. Define the way I said it. You know, okay, if you want to say an afford, you know, like, like, like, like the zoom call affords learning for listeners. Well, you've not said anything, but if you want to say, like to perform a motor action, stuff has to be true about the environment, stuff has to be true about the animal, and they have to merge in some way. That's a fact. That's just biomechanics done. [00:20:26] Speaker B: So something like the term mind, right? [00:20:29] Speaker A: Yeah. [00:20:31] Speaker B: So neuroscientists think, you know, a lot of. A lot of us think that the mind is what brains do. An ecological psychologist thinks that mind is more relational. Right. And not located in brain so much as the brain, body, environment system. Where. What is your conception of mind? Or do you. Is it. Is this another thing that doesn't matter? [00:20:59] Speaker A: I don't think about it like that. I think about processes like thinking and remembering and hoping and planning and believing and desiring and whatever. And sure, people and human babies and lots of animals do lots of those kinds of things. But those are processes, and they are processes that have to do with, you know, all the body parts, so brain, but also like eyes and feet and skin and whatever and other people and the environment and all the rest of it. [00:21:45] Speaker B: So, yeah, you are pro embodiment for sure. [00:21:52] Speaker A: Well, I think there's a lot of ease. You know, I wrote a paper with a lot of ease before I realized that other people were also into that. Of course, behavior is embodied. How else could it happen? But of course, it's also embedded in the environment. Where else could it happen? There's always an environment. It's also enculturated because at least in people and in many kinds of animals, because there's other creatures around that have minimally affected the environment, but also, you know, influence behaviors and, you know, and it's also enabling so that one thing can lead to another thing. You know, we can affect change on our environment, the environment can change on our body and behavior. You know, links all these. So I would say the behavior is embodied, embedded and culturated and enabling and probably some other use, but at least there's data about it. [00:23:02] Speaker B: So you're referring to the. Was it the annual reviews in the Developmental Psychology? Maybe. Yeah, that's a. Yeah, it's a really nice piece. And I'll link to these things that we're talking about. Let's take them kind of a drastic turn, because I want to hear your thoughts about modern artificial intelligence and how it relates to the way that you think about cognition and natural intelligence, because then I want to bring it back to development. [00:23:29] Speaker A: Yeah, I think that. So at some point, someone had posed the ultimate AI challenge, beating a grand chess master at chess. And then Deep Blue did it. But when you watch it, it's hilarious because there's like a computer and then there's like a guy who is translating the moves, you know, to tell the computer where the. Where the pieces are and then. And then taking the computer's output and moving the pieces for the human player. So, all right, Kasparov lost the game, but the guy is, like moving his own freaking chess pieces around and knowing where Deep Blue moved his chess pieces, and Deep Blue couldn't do any of that. So then that challenge got replaced with, you know, a kind of RoboCop challenge. Like, you know, if we can, like, true intelligence would be building robots that could be, you know, the best human World cup champions or whatever. And, you know, so one of the originators and also one of the people who's been most successful at winning RoboCop and various simulated and real robot divisions is Peter Stone. And I've very much been blessed to work with him and other roboticists, Alan Fern's group at Oregon State and Laura Pinto at. At nyu. I think that. Sorry, what was your question? AI in mind? Something like that. [00:25:16] Speaker B: Yeah. Just your take on AI and because of the disembodiment. Right. And what I really want to ask. [00:25:23] Speaker A: So, you know, the. I think the most helpful part about computer science in general for behavioral scientists, neuroscientists, etcetera, Cognitive scientists is, you know, make your ideas explicit in some ways. Now, I think there's great power in metaphor, and the most truly powerful theories in developmental science are all at the metaphor level, like Piaget's. Whatever. I mean, he's wrong, but truly powerful. But it's really helpful when you can boil it all down. And that's my Carnegie Mellon background, you know, coming out and put it in a formal model. But why not put it in a formal model in a real thing like a robot? That's the ultimate formal model. Now real robots are not very good. They're really clunky and an 18 month old human toddler can run circles around the best one of those. But a robot in simulation has a lot to offer behavioral scientists and neuroscientists at this point. Computer vision has a lot to offer, but I think developmental science has a lot to offer to AI. Like one easy example, back to the robots is that like a human infant, for example, they learn so much so fast. I think in part because what they need to learn is constrained by their bodies and their bodies constrained their effective environment. So it's not a blooming, buzzing confusion because most of the environment is impinging on them at all. Like it's, it's like it doesn't exist because they don't have any access to it. They don't see it or hear it or whatever. So who cares? Whatever, right? And then as they get new behavioral skills, their environment expands and they're continually pushing their own skills in the expansion of their own environment. And so think about learning to walk. Literally. A baby can wake up and find themselves 2cm taller than they were when they went to sleep. Their body is different and yesterday couldn't stand up at all. Now they can stand up a week from now. They can take steps. So that baby is never going to get stuck learning any fixed fact at all. It's never going to learn I'm a crap walker or I'm a wonderful walker because what he was yesterday isn't true anymore today. It's never going to learn, you know, don't move this or don't step on the carpet because you'll fall because turning his head made him fall. Like he doesn't, you know. So the system is geared toward, and you know, that's true for motor behaviors, that's true for language, it's true for social interaction. The system is geared toward flexible, generative, creative learning. And that, and that would be a great lesson for AI. And I think that some people in AI are hearing this, you know, loud and clear. It's hard to implement right now in a robot, but totally doable in simulation. And you know, and there's some really smart people out there that are really interested in these questions. [00:29:21] Speaker B: But, and so for example, well, any organism, let's take humans, one could argue that those developmental changes that are happening that enable the context to Change, because the environmental access changes, where a baby can even position its head and look changes and it's growing and its coordination is changing. That could be a bug instead of a feature. It's a necessary thing. Just because we have to be born small and our bodies go through these. [00:29:55] Speaker A: Changes that could be bugs or they could be features. And it's totally possible that all the animals in the animal kingdom and they all develop, you know, this was a kind of a, you know, like a bug, but something that just like a byproduct. And you can gloss over it, but you can test that and you can ask, is it actually a feature or a bug? And it turns out all the variability is absolutely a feature that leads to, you know, better, faster learning, more generative learning, better transfer, you know, everything at least to winning more games than robocop. [00:30:36] Speaker B: But is there something about development specifically that makes it a necessary way forward to that high amount of flexibility and rapid learning? Couldn't you just figure out the algorithms necessary for it and build it in a computer? I'm playing devil's advocate here. [00:31:01] Speaker A: I think you'd have to also structure the learning curriculum just right. And so, you know, like currently with Peter Stone's lab and Alan Fern's lab, we're trying to do exactly that, you. [00:31:16] Speaker B: Know, like in simulation. [00:31:18] Speaker A: Yeah, like they don't want to necessarily have a robot that develops, but if they could understand what the optimal curriculum is, then, you know, then they could have their, their, their algorithms be exposed to the right environments at the right time so that the algorithms become as powerful as a human child. If they could do that. I mean, forget getting to like, you know, our level, adult level or whatever. You don't need wisdom, you just need a robot that's like a 2 year old. And then, you know, man, that'd be amazing. [00:32:06] Speaker B: So you don't think that there is something special about the way our bodies change and the fact that we are embodied and it's always a relation between our body and the environment and affordances, I'll say, that would be necessarily missing in a robot that wouldn't go through those changes since because it's always had the same body. [00:32:34] Speaker A: It might need to change its body. I don't know. I mean, my diabolical scheme is to use simulated robots to understand behavior, but their diabolical scheme is to use developing behavior to build better robots. And it might turn out you've got to have a robot whose body can change and you can't do it all with, you know, sort of meeting out the environment and you know, adjusting the algorithms. That's their problem. They'll figure it out. [00:33:10] Speaker B: You know, maybe, maybe with your help. How much does embodiment matter? Once we're fully developed? [00:33:19] Speaker A: It always matters. It always matters. [00:33:21] Speaker B: But how much is the, you know, because it seems to maybe one could argue that it might matter more as we're developing because we are like all of our cognition develops as you've studied over many years in congruence with our bodily changes. [00:33:36] Speaker A: My dear, you're in that happy point of life. I can tell you from the other side. We are always developing and our bodies are always developing. Gravity might be your pal right now, but gravity is not your long term friend. [00:33:50] Speaker B: No, I mean, I love watching, you. [00:33:52] Speaker A: Know what I mean? No, seriously, we're always developing. Your body is always changing and your body is changing when you put on your sneakers or take them off in your barefoot or in your socks. Your body's changing when you pick up your backpack. Your functional body is always changing because you're always doing things. And so every plan you make, every action you take has to take the current status of your body into account. And even, you know, raising your arms changes your body and changes what you can do next. So yeah, you're always embodied and you actually are always developing. It may not be, it's diff, you know, like development is different across the lifespan. Which things grow and which things shrink or sag or whatever, but yours changing. [00:34:53] Speaker B: I remember when I was younger and falling was actually kind of fun. I mean, first of all, you'd be fine if you fell. And I remember actually enjoying it. I used to fall on purpose to make my friends laugh because it was just hilarious. I don't look forward to falling anymore. But this is one of the things that you've studied just how much developing infants and babies just fall and just get up and are fine and they're just tumbling around the world and enjoying it. [00:35:26] Speaker A: Yeah, maybe what you're referring to is like the three to six year old period when children twirl in circles and stand in the snow drift or a leaf pile and you know, oh, I. [00:35:39] Speaker B: Used to be able to stand straight up and this was still when I was like a teenager and just fall face forward on concrete and just put my hands out and be. And be fine. I know, I'm not going to try that today. [00:35:50] Speaker A: No, I don't think that babies enjoy falling and I don't think in general they're doing it to amuse anyone. I just think it's, it's pretty trivial. It happens so often and it's pretty trivial and when it's not trivial, you know, they probably die. But that's a tiny proportion of the time. [00:36:17] Speaker B: And never in your lab not, you. [00:36:19] Speaker A: Know, not going, you know, like when we, you know, originally, like years ago, I had parents keep a diary of every time their parents, their babies fell. And it was 1.2 times a month. You know, then we put video cameras on babies and just videotaped how often they fall. And you know, like a new Walker Falls 35 times an hour or something like that, you know. Yeah, babies fall 70 something times per hour in motion, you know, like, but, and then, you know, we've micro analyzed the falls and they truly are trivial. You know, a baby on average is back at play like, you know, from, from like fully torso on the floor, you know, to, to up and running around again in under two seconds. So, but think about it like this. In the be in the beginning, babies fall because their legs give out. They're just standing there and their legs collapse or they literally, they turn their head or lift an arm in it to tips them off balance and they fall. So would you want a robot or a baby to learn? Never turn your head, never take a step, never lift your arm? Of course not. So, so they, they ignore it, you know, because, because it doesn't matter. And so they ignore. [00:37:50] Speaker B: Wait, who ignores what babies ignore falling? [00:37:52] Speaker A: Yeah, yeah, yeah, yeah. And if you have a robot ignore it, so there's zero penalty for falling. It learns better, faster, deeper, better transfer, etc. So but think it's not, it's not just about our actions. Like you know, so if a baby learned that every time it said a word or mispronounced it, it would get a penalty. We would never learn to talk, you know, an algorithm, utterance. You would never learn to talk. You just shut down the system. So you know, for, for things that really need to be highly generative, highly creative, highly flexible, you know, you have to allow others. [00:38:42] Speaker B: So what do you, what is your prospect about AI in general? You know, you think if only they added what we learn in development that will solve it. Do you think it's something that is poorly defined in the first place? Because we don't, you know, we used the term intelligence differently when we're talking about organisms versus computers, for example. Or do you not think about it? Well, I know you don't have. [00:39:17] Speaker A: I try not to think about things like large language models. Why we've been testing, I guess is the right word. You're using testing automatic speech recognition tools like whisper for transcription because human transcription a really long time and I'm talking not about like newscaster transcription and against the background of quiet, but babies vocalizations and mothers infant directed speech or caregivers infant directed speech and noisy, you know, cluttery, noisily cluttery environments and also with code switching among languages, etc. And humans can do it beautifully. Highly accurate, you know, automatic speech recognition can sort of do it sort of so far. But you know, like if you run the same exact algorithm multiple times, you'll get different answers over different runs. A lot of the answers involve huge amounts of hallucinations where it just makes up stuff that never happened. Part of it is that the input to most of the AI models for speech, for behavior, for computer vision, whatever, are scraped off the web and they're not, you know, they're not. It's not good input. I don't know if you know about the databrave video library, but this is, this is the world's only large scale repository digital repository for research video. [00:41:08] Speaker B: This was started by your team, right? Yes. [00:41:14] Speaker A: And I don't know, I think there's 790 universities around the globe who are now authorized to access this. So it's a controlled access repository with open sharing within the users of the repository. Not for commercial purposes, not for, you know, some company to make money by learning your predilections and what the things are in your home. But for researchers, including AI researchers, to have, you know, really good data. A lot of it is human annotated or human transcribed. So there's a lot of, you know, gold standard data sets in Databury. I think things like that will help AI. I try not to think about the nefarious uses of AI and I try to think more about how. Well, like what I said to start, you know, how AI can help behavioral scientists, including developmental scientists, understand behavioral processes more deeply and more accurately. [00:42:30] Speaker B: How has the modern technology, machine learning techniques, tracking techniques changed or improved your research? I mean, you were just mentioning earlier that you used to ask people to just observe their children at home, right? And keep a little diary. And now you have people bring their babies and kids in the lab. You can track their every, you know, movement and really quantify things where they're looking. They can be wearing an EEG cap while they're doing some tasks. You can track their. One of the things that you've done is how babies at different stages of development will make different errors in picking up a hammer that has different Orientations, Wooden hammer. Not a, not a lethal hammer. So has that been like a, I hate to use the phrase a game changer, but you know, are you in the best place ever in terms of the tools that are available? Has it changed how you do your research? [00:43:32] Speaker A: I'm a great fan of technology, you know, but. And because I'm in movement science, people in movement science have been using recording technologies forever. Long before anyone in, you know, sort of mainstream psychology or cognitive science and neuroscience had the ability or even realized it would be important to do. And everyone trained as I was. You know, one of the golden rules is you, you, you use technology to answer the questions you have. You're not led around by the technology. And so I mean, one of the things I'm most proud of is my lab with Jason Babcock developed the first. So Jason Babcock is the owner of Positive Science, but we developed the first head mounted eye tracker for infants and children while they're fully mobile. So babies, children could run around, do whatever, you know, in their home, in the lab, walk over apparatuses near our bridges, whatever. All while wearing a head mounted eye tracker that actually tracks their point of gaze in the scene. [00:44:53] Speaker B: Yeah, I was going to say like it used to be a camera, but. [00:44:56] Speaker A: Like actual eye tracking. Amazing. But if I had to choose, would I rather have a head mounted eye tracker on a baby or a third person camera that gets the whole scene? I picked a third person camera in a nanosecond. Head mounted eye tracking tells you one kind of thing. Third person camera retails you another kind of thing. [00:45:16] Speaker B: But my sense is that you're using it all, you're using so many tools at once. [00:45:20] Speaker A: Yeah, I would, I, yeah, I do a lot of things in combination. But so for a lot of the behaviors that I'm interested in, it's not, it's prohibitive or at least it's not nice, not practical to put too much stuff on the baby. So we try to have everything like, you know, like instrument the floor rather than instrumenting the baby, or use computer vision to track the baby's joint angles rather than putting markers on the baby or wires on the baby, etc. So you know, like it's interesting. Like it's, there's so like for a human, absolutely trivial to know if a baby is taking steps. And I'm not talking about babies stepping as if they're walking on a treadmill. I'm talking about babies taking their natural steps as they approach an obstacle or during play. So the steps are like in place or backward or they kind of slither foot or you know, every direction, you know, windy paths, the whole thing. And if I showed you a video right now, we could both like in unison say step, step, not a step, step, step, step, not a step like that. Really hard for machine learning to do that kind of a thing, you know, not up to the task. On the other hand, computer vision can tell you the exact XYZ coordinates of the feet if the space is calibrated. A human can't do that. A human could say, you know, like if I were wearing a head mounted eye tracker, a human could say I'm looking at the wooden wall, but it can't say which pixels on the wall I'm actually looking at and which are out of my field of view. You know, a human can say you took a step, but it can't, you know, took a step closer to this target, but it can't say exactly what are the XYZ coordinates. So it's like a happy marriage where you find the right technologies and the right tools, some of which are humans and you can build tools to augment human abilities. So one of the Databerry suite of tools is dataview, which is a human computerized video annotation tool where user defined events, behaviors could be time locked to their location in the video. But you need stuff like that. Like if you don't have that tool and you're like writing it down with paper and pencil, it's just prohibitively expensive and horrible and not accurate enough to do. But you know, with a tool that gives you like perfect playback over the video and you can go forward or backward at different speeds with like fingertip control and you never have to touch your mouse. Like that's an amazing tool and that augments our human abilities. So, so that's how I feel about it. Technologies are amazing. And I've always, I'm very proud of the work that has come out of my lab where we push technologies forward. You know, instead of just measuring babies walking in a straight line, which you have to force them to do, using an instrumented floor to see what natural locomotion really looks like and what the gait parameters look like. There's, I mean, but I think it's very important that the neurosciences and behavioral sciences, developmental sciences are not getting led around by the promise or existence of certain tools. [00:49:17] Speaker B: Am I right to say that looking time studies in babies was kind of the gold standard for a long time? And what I want to ask you is, is that over? Does anyone still do, unfortunately. [00:49:30] Speaker A: And shamefully, it is long from over, not over. And I think, I think we need to be very clear. What's the looking time step? [00:49:39] Speaker B: Yeah, thank you. That's what I was going to ask. [00:49:41] Speaker A: So looking is a incredibly beautiful, amazing, important motor behavior. And coming back to where we started, James Gibson pointed out that looking is not a behavior that is limited to just the eyes. It's the eyes and a moving head and a moving body. And anyone who doubts it, you know, just watch your baby, your toddler on your, your toddler out of the car, out of the high chair, what, how he, how he looks at things. They, you know, they crutched it, you know, whatever, right? Looking time studies. All right, looking time studies started in the early 1970s with Robert Fance when people had limited ways of studying babies. So some of the ways that people were studying babies are still really good ways to study babies using operant and classical conditioning. Operant conditioning is an amazing way to study a baby or any other kind of animal. You know why? Because you can make inferences about a single participant. That's why Skinner and a lot of his famous studies, B.F. skinner, only needed one pigeon or one rat. Because you can say with 100% certainty that you have altered this animals behavior, you know, based on some reward structure. Right. But Fance realized that you could show babies displays and measure how long they look at the display. And that was originally used to ask whether infants could discriminate between different kinds of visual displays. And that's a totally good question. And the technologies they had for doing it were pretty limited. You know, so it started with just an observer peering down at a baby lying on its back and the baby's looking at displays or has the opportunity to look at displays. And the observer just pressed a button saying, did the baby look at the left display or the display on the right or display on the left like that. And we're essentially doing the same thing in looking time studies. Only now the questions aren't about discriminating visual displays. The questions are about cognition. So if an infant looks longer at a display, it means that they were surprised or the display was unexpected or whatever. And I think for normal people listening to this, when they hear that a baby's surprised, they think the baby's saying like, or something's happening. Besides, on average a group of babies looked like that. A little bit longer at one display than another, a few seconds longer, because that's what's, that's, that's the, that's, that's the data. Yeah, and typically it's not even in terms of the proportion of babies that look longer at the whatever display. And I've done those kind of studies. I mean, everyone in infancy research has done those kind of studies. And everyone knows that it all depends on what the display is if you want babies to look at it. And so huge file drawer problem. And I don't even know a way to get around it. Like you're trying to design your experiment, you have to play around with what the displays should look like. And then finally you get, you know, displays that it seems like the babies are interested in and you test them like that. This is not eye tracking. Eye tracking is a new technology that became widely available in the what, 2000. By 2010, any lab could be doing this. And that's where you actually track where the baby is looking on the display. Not did the baby look left or look right or look away, but where on the display. So was the baby actually looking at the thing that you, the adult researcher, thought was the surprise? Now you could actually know. So I don't think there's any excuse for using old fashioned looking time measures when if you want to get your outcome measure to be based on where babies look, eye track them and find out where they look, because you'll find really interesting things. There's a beautiful paper by Scott Johnson from, I think 2004. We showed that on average, babies look longer at, look longer at a display where an object appears, you know, to be moving behind an occluder like that. Turns out some of the babies are. When you actually use eye tracking and get the location of their looks, some of the babies are tracking the moving object. Some of the babies though are looking at the edges of the screen or they're looking at the occluder. They don't, they don't know how to guide their visual exploration. And the babies that have learned more about how to plan their visual exploration, those babies also perform better, you know, in the straight up habituation task, but they also perform better in a whole range of other tasks. So I just think like, why are we using a really weak tool method procedure when we have all kinds of other things available to us? And. [00:55:43] Speaker B: But it's still quite prevalent. [00:55:44] Speaker A: It's quite prevalent. I would say it's probably the most, what's still the most common way to study infant cognition and among the most widely. Looking time methods are among the most widely used methods in all of infancy research. And there's just no excuse for it. I mean, if you want to know if the baby's surprised. You could also be looking at their facial expressions and gestures and vocalizations and pupil, you know, dilation and lots of other things that are all there in the videos. People don't. Out of laziness and also because it often doesn't have anything to do with how long they're looking, which argues against it being a surprising event for the baby. [00:56:35] Speaker B: So I was telling you a little bit about my background. And what I didn't say is that I came. My PhD was in an eye movement slash neurophysiology lab and back in the old days. And this is non human primate. And it's still done this way frequently as well. But you would do eye tracking. But you would. You would want your experimental setup to be as controlled as possible. Right? So we would, you know, head fix the monkeys. They'd be sitting in a chair because we don't want their heads moving because we want just their eyes moving so we can track it and infer something about the. In my case, about the way that they were making decisions with their eyes. Right. These days is very different. I'm working with a data set now. So naturalistic behavior is all the rage now. And that's a lot. What you do, you do do controlled experiments, but you also just measure naturalistic baby child developmental behavior in the lab. And, and you set up the lab to make it, you know, to. [00:57:35] Speaker A: I want to come back to that, but I want to hear. I want to come back to a discussion about naturalistic, natural, whatever, unnatural setups. But I want to hear the rest of your story first. [00:57:48] Speaker B: Where was I? Oh, I'm doing now. Yeah. [00:57:51] Speaker A: Primate eye movements. [00:57:53] Speaker B: Right. Well, so I don't do that anymore. So that, that is. That contrasts with. With these days, naturalistic behavior has kind of exploded. But yeah, I mean, what you want. You want me to tell you my dissertation? Is that what you're. [00:58:07] Speaker A: No, I wanted to hear the. I didn't want to interrupt you and stop your. [00:58:12] Speaker B: No, that's okay. So these days I have a challenge. A problem. A challenge. Let's go optimistic. So I came to this lab where I'm doing another postdoc. The eric. Eric ITRI lab. And you know, we have these really high density neural recordings in freely behaving mice. They're just walking around the cage or a box in my case, in my data set. Case. Grooming, Locomoting. Yeah, Investigating. So it's, you know when you were talking about earlier how you can watch a child and say, oh, that's a locomotion. That's a locomotion. That's a step. That's a locomotion. Well, we have already, when I got to the lab, they had developed an unsupervised learning technique to basically use the kinematics data to categorize different behaviors. So I have like all these behavior labels, I have a bunch of neural data. And the question is how is, how is the brain enacting these behaviors? And it's kind of the wild rest west right now in neuroscience for naturalistic behavior because we have all this data now, we have all these machine learning techniques and the whole history of neuroscience, like my PhD experimental setup is just do the same thing over and over and over, exactly the same way, control it as much as possible. Then you average all of the same behaviors and compare them to some other behavioral condition and then you say, oh, there's a difference. That must be how the brain is doing it. Now we have a freely behaving mouse who sometimes grooms like this, sometimes like this. You know, sometimes it walks a slightly different way, slightly turns left. You know, there are all these like minutia of the behaviors and the neural activity is not nearly as clean, at least in the areas where I'm recording. And this is actually where I wanted to bring up a piece that you and I didn't know he was your partner wrote about developing motor cortex. One of my pet little. It's not a theory because it's not strong enough to be a theory, but thoughts is that the areas that we're recording in which are primary motor cortex and basal ganglia, probably aren't needed for these innate naturalistic type behaviors because there's no motivation, there's no reward, there's no attention. You know, there's no like there's nothing going on. And I'm not sure to interrupt you. [01:00:41] Speaker A: Because now you're getting annoying. So. [01:00:46] Speaker B: Awesome. [01:00:49] Speaker A: All right, so except for in the beginning of graduate school where I learned to do looking time studies with babies, you know, strapped in a car seat, I've only done studies where the baby, the child, whatever was free to spontaneously do something. Some kind of behaviors that were new to them, new to us, surprising to all. So you know, you mentioned our, our study of, you know, four or five year olds hammering a peg. You know, like we saw children do things in hammering the peg that no one would ever have imagined that a child would do. 100 whatever tiny little taps to hammer the peg down. Children using like their children chest and their bodies or whatever to turn the Hammer handle the right way when, if they grasped it in an awkward position, you know, coming like this to grasp it. So they would end up in a radial grip when all they had to do was just do an underhand grip to get it, to get it in the radial position. And in infants, you know, locomotive. I mean, even once when my child was. My child's grandpa, when my child was a child, and we did a science day for their sixth grade class and we just asked the sixth graders to crawl over, crawl in a straight line down a carpet. And we saw so many types of cross crawls. It was absolutely amazing. You know, bunny hops and children who could crawl with aerial flight phases with all four limbs in the air like a horse, you know, and they're only going a few yards. So. Okay, so that was a long winded way that you'll have to edit down to say that. I think it's very important that in any task the animal person has enough room that they can respond in ways with behaviors that are new to them and new to you. Otherwise, why even do it? Because you've sucked all the lifeblood and the fun out of the enterprise. [01:03:34] Speaker B: I could interrupt you there and tell you why the history of neuroscience has done it. [01:03:38] Speaker A: I know why they've done it, but. But really, you know, you don't need to do that. You don't need to do it to have sufficient control. And, you know, so I generally use psychophysics, which is, you know, psychophysics and operant conditioning are the two ways where you get the most strong data from a single participant and where. So where your inferences about what that participant as doing or knowing or perceiving or whatever are our most powerful. And you don't have to rely on group averages. Right. Naturalistic to me is like, oh, it's a weird, like, you know, a freely moving mouse in a cage or. Rob Frumpke's lab at NYU does these beautiful studies of mouse dams, you know, over consecutive births of their litters and caring for the pups and so on. [01:04:49] Speaker B: You said mouse. What's over the consecutive litters? [01:04:52] Speaker A: Litter, like giving birth to a litter of pups and then feeding the pups and managing the nest and you know, and then they have like their second. So it's 24, seven mouse recordings. They're in a cage. And I guess I would call that naturalistic. And I do similar kinds of things because we want to sometimes have studies where everybody has the same environment with the same objects and the same recording setup. Or it's a highly Calibrated space that's really conducive to computer vision, et cetera. I also do natural behavior where you just go out in the wild with your video camera and just, it's just people doing what they're, what they do. The end. And so it's not natural, it's not istic anything. It's just actual natural behavior. And so I think that whole continuum is extremely useful from a pretty controlled experiment where you want to do something like psychophysics or operant conditioning, but where the animal still has freedom to do its thing to truly natural behavior. And somewhere in the middle is the so called naturalistic, where it's a little bit like life, it's, you know, whatever. [01:06:11] Speaker B: What was the annoying part? Just the usage of the, of the term naturalistic. [01:06:17] Speaker A: Oh, you were talking, you were saying that naturalistic behaviors are innate behaviors. [01:06:23] Speaker B: Oh, no, I was, I didn't mean to conflate the two, but what, what I meant by innate. So naturalistic is just in, in this sense, just the mouse being able to just wander around. [01:06:33] Speaker A: Yeah. [01:06:33] Speaker B: On its own accord with its own timing. But the behaviors that it's enacting are innate behaviors. It like knows how to groom. It's not performing any task. Right. It's just, it's maybe it maybe, you know, innate. Maybe it learned how to groom. [01:06:48] Speaker A: So groom better. So you. It's not a, it's not a prescribed task that, you know, an externally directed task. It's just the mouse doing what it wants to do. But do mice learn to groom? Yeah, they do. [01:07:07] Speaker B: Yeah. [01:07:07] Speaker A: And there's a whole range of experiences that mice have to have for them to be effective at grooming when they need to trim themselves. [01:07:16] Speaker B: So we need a succinct way to a single term for naturalistic, non innate, some at some point learned, but non task related behavior, perhaps. [01:07:27] Speaker A: Yes, that would be good. That would be fine. [01:07:29] Speaker B: Okay. Yeah, we'll come up with something. Yeah. Anyway. Anyway, the data is not, you know, perfectly aligned to anything. So our unsupervised learning algorithm says, you know, at some time temporal scale, like when locomotion started. But this locomotion might have begun with left paw moving forward first, the other one with the right paw. And so we can't it, so we're doing a thing where we're aligning like traditional neuroscience, aligning the beginnings of a behavior to a specific time point, but the behaviors are enacted somewhat differently. Algorithm isn't perfect, et cetera. So we have all this data, it's fairly aligned, but there's a Lot of variation in it. And it's. It's been a challenge to relate the neural activity to the different. In this case, we have 16 different behavioral categories relate the neural activity to those behaviors. And one reason may be that cortex is maybe not the major player, maybe not needed, except when it is. And when I came across the opinion piece that you and your partner wrote or the perspective maybe about how early in development motor cortex is not motor, it's more sensory. And it takes, depending on the organism, it takes longer or shorter for motor cortex activity to be able to even directly behaviors, because the connections literally aren't there when it's younger. But it is getting like sensory information. So I just thought that was interesting. And I'm not sure if I'll present your perspective in my lab group because there's a lot of talk about like, well, motor cortex must be doing these. So how is it doing it? But maybe it's not doing it. So I wanted to go ahead. I mean, I'm getting. Am I getting annoying again? [01:09:26] Speaker A: No, well, no, no, not really. Adults, you know, across the lifespan, we do lots of movements that don't require cortex to control them. The argument in that and that perspective piece is essentially that there is no connections. So the things that, you know, the researcher is using to make inferences about what's happening in the baby's mind are not coming through cortex. So if you want to say that, you know, whatever month old, a seven month old or 15 month old has moral reasoning, you know, three month old has moral reasoning, but it's not the same moral reasoning as an adult would have. So it's like moral, you know, what is it? You know, and that's just sloppy. That's, that's just sloppy science. Let me get Mark. And he can. So he's asking about the motor cortex thing. And it's in the context of, in his own work, they're studying mouse behavior, freely moving mice behavior in a cage. [01:10:48] Speaker B: What I wanted to ask about is, so I have recordings in primary motor cortex and basal ganglia. And I had been thinking this for a while now that maybe for these less task, less motivated, less cortically necessarily driven behaviors, that it's, you know, mostly subcortical. Whereas the motor cortex, you know, you guys write about how before motor cortex develops the necessary connections to enact skeletal, skeletal motor behavior, it's actually getting sensory signals. Right. And so that would make sense if motor cortex and, or basal ganglia weren't really that important for these behaviors. It would kind of be sloshing around, sometimes delayed a little bit. But my thought is like, well, motor cortex would benefit by being in the right kind of range in case it needs to be called on, you know, but. But it might not necessarily need to just be like these beautiful tuning curves of neural activity aligned to the. To the behaviors. So I just wonder what you think about that. But also I just wanted, you know, you to explain a little bit more about that. And then you got some pushback on the perspective, right? [01:12:07] Speaker C: Yeah. [01:12:09] Speaker B: Published pushback. Anyway. [01:12:13] Speaker C: Yes, there's a lot to unpack here, obviously, because there has been relatively less work done on brainstem circuits relative to portable circuits and directional behavior. We don't understand fully the relationship between how those things, how they develop, how they are processed in real time, how they lead to the types of time sharing that you see in your behavioral repertoires, where they're going from one behavior to another, which is a really complex process. And because mice being mice, everything happens very quickly. So there are questions about how automatic a lot of these things actually are. So there's a developing literature in, in adults about how to learn behavior. You need motor cortex is more involved in adult behavior, that the motor cortex is involved in that process. So if you lesion the motor cortex prior to learning, you prevent learning, but after the motor. But after a certain task has been learned, now you destroy motor cortex, and there's no effect on the learned response, which leads to the suggestion which suggests that there's been an offloading of that behavior onto another structure not too different from what might think about, you know, systems consolidation of memory, where you pass. Right. And all of these things are controversial and they all have to be taken, of course, you know, with salt. But you also asked that about motor cortex being sensory early in development, both evolutionarily and developmentally. You know, the motor cortex, the motor cortex is a relatively late evolving structure. You see a lot of overlap in those circuits in certain types of marsupials, for example. And so the motor cortex is not distinct from sensory motor, sensory cortex in a lot of species. And then developmentally, what gives credence to that, I think, is when you look developmentally and you see the motor cortex is initially a purely sensory structure. And my work happens to be involved with looking at how sleep, sleep related motor activity reveals that sensory process. But it's also known that motor cortex in adults is receiving sensory input. There's a nice review by Nick Katsopolis in Neuron some years ago called sensing with motor cortex, which doesn't take any developmental perspective at all but still talks about the role that motor cortex plays in processing sensory input. So it's a lifelong process. And it makes sense because motor cortex needs to sort of integrate input and up from proprioceptors and other sorts of peripheral information and then use that to structure litigation. [01:14:50] Speaker A: Can you give the argument in the two papers and then like the basic argument, and then I don't know if we need to talk about the pushback. At least give the basic argument so Paul can edit that when he wants. [01:15:06] Speaker C: Yeah, so the basic argument is that I got to report. [01:15:16] Speaker A: That picture. Like so, you know, there's the behaviors, reasons, make inferences about cognition, you know, but. [01:15:23] Speaker C: Yeah, but let's start with the developmental picture is that, you know, motor cortex, like a lot of other structures like cerebellum, you know, it's a very late developing structure. And you can see this. There's some evidence from the human literature. For example, people with cerebral palsy, it's mostly caused by a perinatal stroke. That's one of the most predominant causes of cerebral palsy. But the symptoms don't show themselves until four or six or later months after birth. Why would that be if motor cortex is so. I mean, you know, we all know that if you have a stroke when you're an adult, you have immediate paralysis that doesn't show up in your face. But also work from cats, rats, developmentally, you see that motor cortex is not. All the evidence that we have suggest that motor cortex is not developing for weeks, in the case of rest, until after weaning. So that would mean that. And if you look at, you know, you work with mice, if you look at mouse behavior around the time of weaning, it's pretty complicated. You know, they're moving around the cage, they're drinking, they're eating, they're grooming, they're. The evidence would suggest that all that behavior is brainstem dominant. And I said cats. And your cats showing up. That's a dog. My dog to a dog stick. So the. [01:16:45] Speaker A: So where was I Bring it to the developmental. [01:16:49] Speaker C: No, I understand, I understand. But the point is that the. That in the. That in mice, you know, you're seeing complex behaviors when motor cortex is not involved. I think that ever has relevance to the work that you're doing. [01:17:03] Speaker B: So. [01:17:03] Speaker C: So the best evidence, or one of the best pieces of evidence that you have for demonstrating motor cortex involvement is sensory stimulus, electrical stimulation of the cortex. And you don't see movement of the limbs and rats until 25 days after birth, which is extraordinary. Right yeah. Now, that doesn't mean the circuits aren't. You cannot bring those circuits in to produce behavior. It's just that without a lot of manipulation of those circuits, you don't see evidence of movement. So that's pretty strong evidence. So if you take that, then you go back to the human, you know, the human literature and all of the things that are being put into the minds of babies, which, whether stated or unstated, you know, whether implicit or explicit, people think of as cortically required, you know, cognitive processes. Then the inference, the clear inference would be that a lot of these behaviors either are not being controlled by the. By the cortex, or some of these statements, some of these experiments are either not well done or they're based upon flimsy research or flimsy methods or what have you, and they're incorrect inferences being. [01:18:20] Speaker A: Drawn about what's happening either way. So in the best case, assuming the data are real and reliable, not motor cortex driving those looking times, not motor cortex driving the neonatal imitation, not motor cortex driving the behaviors. And that means that it's not the same underlying representations as in an adult. Yeah. [01:18:55] Speaker B: What does this say about. So at some point, motor cortex got the name motor cortex, and now it's. We can't. What does this say about how naming something affects the way that we assign function to it? [01:19:08] Speaker C: Well, yeah, naming things. I mean. I mean, science is filled with problems with that. Look at the. I mean, you could make the same arguments about Genesis genes. You know, you have a gene, you call it sleepless because a mutant mouse fly happens to not sleep, you know, but. But obviously these things are not doing just those things that that. So motor cortex has that name, but it also does a lot of sensory processing. So, yes, it's a problem for how we name. It's yet another example of how artilles get in the way. [01:19:44] Speaker B: Yeah, yeah, because you name something, it reifies it for sure. [01:19:47] Speaker C: Absolutely. [01:19:48] Speaker A: Yeah. [01:19:50] Speaker B: Let me just ask you, Karen, one more thing. So thanks for, I guess, Zoombombing our podcast. Oh, no, I was going to say thank you for solidifying my thought about why I'm studying something that doesn't exist. That's terrible. But I'll get there. I'll publish something. Because I have to publish something. So what I want to ask you about, Karen, is what you're excited about these days. I know you just submitted like 15 grants or something. I think it was four grants that you'll be submitting in August, I think you said. And so you're going to continue to be prolific and busy. Anything that is particularly exciting and, or challenging. [01:20:36] Speaker A: Well, I don't know if I can prioritize, so I just have to tell you the things that I think are super exciting. [01:20:44] Speaker B: Great. [01:20:45] Speaker A: So one thing that's really exciting is collecting corpora of what people's natural behaviors really are. And so we have done that for families in New York City. We are also doing it with the help of 73 PIs across the US and Canada for infants and their caregivers across 35 sites in the US and now we're taking it global. And so imagine you or, you know, any researcher being able to access Databury and see what life is like for babies and their caregivers anywhere in the world during not naturalistic, natural, you know, activities. We call it. The acronym is Babies B A B I E S and it stands for Babies activities and behaviors in Everyday Settings. [01:21:49] Speaker B: I don't know how hard you've worked on that acronym, but I know, yeah. [01:21:52] Speaker A: Good work, Karen, Good work. So that's something that I'm really excited about and something that would only be possible with a resource and a platform and a tool like Databury. I guess the honest answer though of what I'm really, really excited about are the same things that have excited me since I was a graduate student with Eleanor Gibson. I want to know how babies learn to guide their actions from eyes to toes in an adaptive and functional way to do the things that they need to do and the things they want to do. [01:22:34] Speaker B: How long is it going to take you to figure out? Five more years? [01:22:37] Speaker A: Well, I'm old, so I figure I'm giving myself 10 years to do it. And I think that. So currently one way we're doing that as same babies, we study them solving these problems on apparatuses, deciding how high of a drop off they can walk over, how steep of a slope they can walk up or walk down, etc. But then those same babies we also observe naturalistically in our laboratory playroom where we can see how much they move and how they move and whatever while they're playing alone and while they're playing with their caregiver. And the same babies during totally natural behavior in their homes for one hour of, you know, concerted activity between naps and meals, etc. And then also using video ecological momentary assessment where the Caregivers take little 10 second videos throughout the day over the course of a week. And then we have standard measures of walking skill, natural walking body dimensions, you know, the whole, the whole enchilada and so we'll study that, but we're also now going to turn that into something where we can ask it in terms of simulated robot learning curriculum. And there you can. You can sort of redo the baby's lives. So does it really matter if a baby steps on 179 different surfaces in the course of an hour? Maybe, but maybe it doesn't really matter for perceiving affordances. Does it really matter if a baby has experience on elevations for being able to cope with, navigate, perceive affordances of elevations, novel elevations in the lab? Maybe it doesn't. And we can address those things with simulated robots. Your question, does it matter that the body is changing, and does that aspect of embodiment really matter? Well, we've tracked babies longitudinally, you know, from two months to about 20 months in the same lab play room, where at first they just lie on their back and stare at the ceiling. And by the end, they're burning and touching everything in the room and using computer vision so that we know everything that was in their field of view and how that changed everything they touched with their hands and the exact location they touched in and the exact durations, everything everywhere their body went. So I think those kinds of approaches, it's big data, but it's also deep data, and I think that's the way to do it. And I think you got to put a lot of different minds on it. So if I were all alone, I'd be looking at my videos with the sound off. But luckily I have Catherine Thomas lamonda, the world's best collaborator, who's an expert in language and social interaction. I think if I were limited to what I could study solely on my own, you know, I'd be looking at babies in New York City living in apartments rather than, you know, out in West Lafayette, Indiana or whatever, with my collaborator, Laura Claxton. So, you know, and. And we'd all be studying babies that live in things we recognize as a home and not. And things that we go like, really? That's what a home looks like in another place in the world. So I guess that's a long way to say. Let me say you need to do the whole enchilada, Paul. You need to have the controlled experiments. They're highly enough controlled so that you can actually do something like psychophysics and say something definitive about each baby and each session, each participant in each session, just like visual psychophysics. Right. But where they're free to do things that'll surprise you and horrify you and delight you and Same with them. Right? And you need to know what is the actual learning input, which things matter, which things might not matter, and how that changes with development. And we actually, right now, we as a field, because if Karen Adolph can do it, anyone can do it. We have all the tools and technology to do the whole enchilada. And so what my colleagues are still doing babies in a car seat, looking at computer displays and taking group averages. Grow up, Get a life. You know, like, we have the tools and technology. Like the six Million Dollar man, we have the tools and technology. We just need to do it. [01:27:47] Speaker B: Will you do me a favor? When in 10 years you've figured this all out, will you title the. Oh, my gosh. I think. Is my microphone working? You can hear me. Will you title the paper the whole enchilada? [01:28:00] Speaker A: You got it, baby. [01:28:01] Speaker B: Yeah. It's a great phrase that doesn't get used enough anymore. Karen, thank you so much for your time. And tell Mark I said thank you for coming in. That was a surprise and it was fun. [01:28:12] Speaker A: All right. [01:28:18] Speaker B: 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 advanced research. Visit thetransmitter.org to explore the latest neuroscience news and perspectives written by 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're hearing is Little Wing, performed by Kyle Donovan. Thank you for your support. See you next time. [01:28:51] Speaker C: It.