Babies and toddlers have truly outstanding brains — they absorb information broadly, quickly, and indiscriminately as they learn about the world, with processing speeds that leave AI-powered robots in the dust. Alison Gopnik, professor of psychology and affiliate professor of Philosophy at U.C. Berkeley, has been studying baby brains for decades, and she joins us today to talk about how we could look to them to make computers smarter.
Phil Stieg: Hello, I’d like to welcome Allison Gopnik, a global authority in cognitive science and an expert in the study of children’s learning and development. She is a professor of psychology and affiliate professor of Philosophy at Berkeley. She writes the Mind and Matter science column for the Wall Street Journal and has written widely about cognitive science and psychology for The New York Times and The Atlantic, amongst others. Over the past two decades, her books, The Scientist in the Crib, the Philosophical Baby, and The Gardener and The Carpenter, have reshaped our understanding of the way that infant brains work. Allison, thank you for being with us today.
Alison Gopnik: Well, thank you for having me.
Phil Stieg: Children have an incredible capacity for learning and for exploration. Is the child’s less mature or is it just wired differently than your adult brain?
Alison Gopnik: In terms of brains, I think one of the things that we’ve discovered is that it’s not that brains in infants and young children are sort of defective compared to adult brains, which I think is a very natural way —
Phil Stieg: I didn’t mean to imply defective. I just meant like your muscular system.
Alison Gopnik: Or immature, right…
Phil Stieg: It’s a developmental process.
Alison Gopnik: Yeah. So one of the ways that people often have thought about development in general and in some ways a very natural way to think, is that you’ve got this adult system that has particular kinds of characteristics and then development means building up to that adult system. So you’ve got an adult brain that does various kinds of things. And the point of development is to turn this childhood brain into an adult brain.
But if you think about it from an evolutionary point of view, that doesn’t make a lot of sense. If it was so great to just have an adult brain, we could have an adult brain. And what I’ve argued is that there are features of the childhood brain that really make it much better at some tasks than the adult brain.
We get this long period when we don’t have to actually go out and get resources that’s done for us, by our caregivers. And that means that we can just explore, we can just look around the world, figure out as much as we can about the world, consider as many different possibilities about the world as we can. And when we look at brain development, we see this really interesting transition from a young brain that has many, many, many synaptic connections and is making more in the first several years of life, and then what happens is that those connections that are used that turn out to be useful get preserved, myelinated, and the ones that aren’t are pruned.
So you have an early young brain that’s very flexible, very plastic, as neuroscientists say, very good at learning, changing with new information, but not terribly good at doing anything effectively or efficiently.
Then you have this adult brain where the ineffective pathways have all been pruned and there are more long term pathways, there’s more of what’s called myelination. The pathways are more efficient. And that’s a brain that’s very good at doing things and doing things effectively and efficiently. But it’s a brain that’s much less good at changing, much less good at altering what it’s doing in the light of new information.
So it seems as if there’s this kind of trade-off between this young brain that’s really good at learning, not very good at acting and this old brain that’s really good at acting, not so good at learning.
Phil Stieg: It makes common sense, however, to think that a baby’s brain is what, 800 grams and the adult brain is 1500 grams, that over that process of development, as you said, there’s pruning. And the term you use was myelination, which is kind of an insulation of the nerve cells so that the nerve cells respond more rapidly to input, that there is going to be a change in processing information that occurs in a neonate, in an early born, versus the adult. Do you feel as strongly that the adult brain doesn’t have the adaptability that the child brain has?
Alison Gopnik: Well, we’ve discovered that the adult brain has more adaptability than we might have thought in the past. But I think it’s pretty clear, both in terms of behavior and in terms of brains, that typically the adult brain doesn’t have the same kind of flexibility. So the adult brain may be bigger, but the infant brain actually has many more neural connections. So there’s actually a sort of brain and a half in two or three year olds. So it depends how you measure. But if you look at things like language development or the development of the visual system across a wide range of different areas, children’s brains seem to be more flexible, more able to adapt to novelty than adult brains.
So an example that you’ll know as a physician is that you need to correct vision problems, for example, very early in development because the young brain can do something like fix amblyopia if you give it the right kind of information. Amblyopia is when the two eyes aren’t coordinating. You can fix that if you do it early. It’s much, much harder to do it if you try to fix it when you’re an adult.
The same thing’s true with language. It’s much easier to learn a new language, particularly learn the sounds of a new language when you’re young. My two year old grandson, who lives in Montreal already, much to my amazement, speaks French with a beautiful, perfect accent. And even though I lived in Montreal for many years, my French still sounds like it’s English speaker’s French. And that’s a pretty general phenomenon.
We’ve also done some work in our lab that shows that when it comes to sort of creativity, if you give adults and children a problem to solve, that has an unexpected, unlikely solution, children are actually better at getting to that unlikely solution than the adults are. So even though the adults are better than the children at getting to an obvious the more obvious solution, the children are better than the adults at finding the unlikely solution.
Phil Stieg: Can you explain to us in layman’s terms what the theory of the mind is?
Alison Gopnik: Yeah, this really goes back to work that I and some of my colleagues started in the 80s and the question is, if you look around a room with people in it, what you actually see is a bunch of bags of skin that are draped over chairs and covered with pieces of cloth.
Now, that’s crazy, right? Like, that’s not what you see when you see people. What you see are people with thoughts and beliefs and intentions and goals and things that they want to do and find out about. And it’s a very old philosophical problem about how is it that you ever understand all this about other people when all you see is these external bodies? And in particular, how do you understand that they have thoughts and beliefs and desires the same way that you do?
So, as I say, that’s a very old philosophical problem, but it’s one that we’ve only begun to think about in terms of children and babies relatively recently. For most of our understanding of development, people thought that babies and young children couldn’t understand what was going on in other people’s minds, that they were egocentric or solipsistic.
And what we’ve discovered since is that even the youngest babies, even arguably newborns, already have some ideas about how other people’s minds work. And as children develop, they learn more and more about other people’s minds. And that becomes one of the most important, arguably the most important kind of knowledge that we can have.
Phil Stieg: So do you separate the concepts of mind and consciousness? And do you believe that there are levels of consciousness as one gets older?
Alison Gopnik: One of the interesting sidelights of work in theory of mind is trying to figure out both what babies and young children think about consciousness, understand about consciousness, and also to try and say something about what their own consciousness might be like based on other things about what their minds are like. And one of the interesting things that we’ve discovered is that young children, for instance, don’t really understand consciousness the same way that we do as adults. And that may indicate something about their own consciousness. But I think there’s also evidence that children’s consciousness is actually different from adult consciousness in the sense that it’s more wide ranging, it covers more.
A metaphor that’s often used about adult consciousness is that it’s like a spotlight. It narrows in on a particular set of problems. But for children, that seems to be really different. Children literally seem to be taking in more information about the world around them at any moment than adults do.
So instead of thinking of it as a spotlight, you could think of it as a lantern that’s illuminating everything that’s going on around you.
When I go for a walk with my two year old grandson, it’s marvelous, but it takes twice as long. Of course, since this is in Montreal, we’re going around the corner to get a latte and what’s called a baby chino, which is like a late for babies. The espresso place is two blocks away and it would take 3 minutes if it were me walking. But of course, there’s leaves on the ground and there’s cars and there’s the most awesome thing of all, which is a garbage truck. So you have to stop and look at the garbage truck. And that’s kind of this experience of lantern consciousness.
One of the great things about being a caregiver is that as adults we don’t get to experience that very much outside of our meditation retreats and trips to Paris. But hanging out with a two year old, that’s what it’s like to be a two year old all the time. If you think that the main agenda of childhood is to try and learn as much as possible, having this kind of expanded consciousness is really a tremendous benefit.
Phil Stieg: Well, you talked a little bit also about nature versus nurture. The takeaway message for me when I was reading through that was more about as parents cutting yourself some slack and not being so, “my kid’s got to be like this.” It’s trying to understand that child and being reflective about it, not being so planning about it? Is that your sense?
Alison Gopnik: You know, part of the point behind the gardener and carpenter metaphor is the idea that if you think about what you do as a carpenter, what you’re doing is trying to shape a chair to come out a particular way. And that seems to especially to modern parents who have spent a lot of time going to school and working, that seems like a natural way of thinking about what you should do with children. But I think the evolutionary background and the scientific background suggests that’s not really what’s going on in caregiving. Instead, a better way of thinking about it is like a gardener.
A gardener can’t dictate exactly what is going to happen in the garden, but makes a kind of space where many, many different things can happen. Different flowers can bloom in different places at different times in different environments. And in fact, that kind of a system is more robust in the long run than it would be if you were just trying to make sure that this particular plant had grew up in a particular way. And I think that’s a better description of what we should do as adults. And I think also a liberating one for parents.
If they can kind of get out of the mode of thinking, here’s this thing that I’m completely responsible for, and instead be in the mode of thinking that they don’t have to make their children learn. All they have to do is let their children learn.
(Interstitial Theme Music)
Narrator: If you are fascinated by Dr. Gopnik’s story of how our brains experience the beginning of life – you may also be curious about how our brains experience the end of life.
Back in May of 2022 Dr. Stieg’s guest was Dr. Bruce Greyson, a researcher at the University of Virginia who has spent decades studying the phenomena of “NDE’s” or Near Death Experiences.
Here’s brief excerpt from that interview — Season Three, Episode Ten…Excerpt:
Phil Stiegr: How do you differentiate this from a hallucination? How do you know it’s different?
Bruce Greyson: That’s a great question. As a psychiatrist, I deal with hallucinations all the time. So, of course, that occurred to me, and there are lots of differences. The content of hallucinations are really different from person to person. You don’t see any consistency. And yet near death experiencers, they tell the same stories across the globe. And going back through centuries. We have accounts from ancient Greece and Rome that sounds just like the NDEs we hear today.
Furthermore, people have perceptions in a near death experience from an apparent out of body perspective that are accurate and can be corroborated by other people present at the time. And you don’t have that in hallucinations.
What’s most important to me as a psychiatrist is the aftereffects on people. Hallucinations usually make people more introverted, more terrified of interacting, more fearful, and less inclined to explore what this is all about. Whereas near death experiences make people more outgoing, more caring about other people and a greater desire to explore this phenomenon.
Narrator: Sound intriguing? You can find the entire episode (and many others) on our website, ThisIsYourBrain.com.Before you go there, let’s return to the second half of Dr. Stieg’s fascinating visit with Dr. Alison Gopnik as she describes how she brought her knowledge of baby brains into the world of Artificial Intelligence.
(Interstitial Theme Music)
Phil Stieg: You talk a lot about using childhood brain development, cognitive development, visual spatial understanding as a way to help us develop better artificial intelligence with computers. What have you learned from childhood brains that you are now directly applying to computer technology to facilitate artificial intelligence?
Alison Gopnik: Yeah, I think the biggest thing that we’ve learned is how far away even the most impressive AI systems are from doing things that every child does completely spontaneously. So a lot of the things that children do, like going out into the world and figuring out how cause and effect works very quickly. Something that every two year old learns how to use the remote on the TV but knows that there are some kinds of links that you should think of as causal and other ones that are just coincidences. It turns out that’s actually very hard for even very powerful AI systems to do.
AI systems are very good at taking a whole lot of data and increasingly more and more data and finding the statistical patterns in that data, but they’re not very good at taking that data and then the way a scientist would, figuring out what’s actually going on in the world, that is the basis for that data. And children are very good at doing that.
It’s really striking when I look at my colleagues in robotics for example, and they’re very proud of their robots, but those robots are so slow and so awkward compared to even my four month old grandson. And we’re only just beginning to learn how that’s possible.
One of the other important differences between the way that typical AI has worked and the way that children work that we’re working on is that the typical AI systems are what’s called supervised. In other words, someone is telling the system at every moment, here’s what you should do, here’s what you should be trying to. Here’s what the good response would be. Here’s what the bad response would be. Or here’s a dog, here’s a cat, here’s a dog, here’s a cat, here’s a dog, here’s a cat. And children have much less of that kind of supervision. Children are learning through their own spontaneous exploration of the world, through their curiosity, through going out and playing with things that are going on in the world.
One of the things that we’ve been working on is showing one of the things we’ve been working on is showing that AI systems do much better if you give them a chance to explore and play in the way that children are exploring and playing. That gives them representations that are much more robust, that can deal with change much better.
Phil Stieg: As a computer scientist, how do you teach a computer to play?
Alison Gopnik: So one idea, for example, is that instead of rewarding the computer for doing a particular thing, you could reward it for doing something new that it hasn’t actually done before. Instead of getting it to go to the place where there’s going to be a reward, you could get it to go to a place where something surprising has happened. And there’s been a number of really interesting attempts to do that. in the context of play.
One of the ideas we have there’s a nice idea in reinforcement learning, which is one of the big techniques for training computers of empowerment. And that means just getting the computer to be rewarded for doing things, for bringing about outcomes, for making things happen, whether or not those are things that you necessarily wanted the computer to do in the first place. And we think that that’s a lot of what’s going on in children’s play. But none of those techniques really seem to quite get at the subtlety of the way that children are playing. But that’s the sort of thing that you can do to try to get a computer to play.
Phil Stieg: I mean, you can’t turn on 60 Minutes on a week and not see another thing about artificial intelligence. In medical science, we’ve been exploring it in terms of answering patient questions and you go to Chat GPT and we’re finding now that the responses provided by Chat GPT are not only accurate and factual, they have a hallucination rate that they frequently talk about, but they’re also equally empathic. So there are changes that I assume are going on in the technology, that they are assimilating some of the childhood experiences that exploring sense, that makes it a broader response than just accumulating information and giving back a widget of information.
Alison Gopnik: Well, it’s interesting what those systems in particular like Chat GPT, what they’re very good at is summarizing what human beings already know. So if you think about most of what you have to do, say, when you go to medical school, is just learning a whole lot of information, getting it from the medical textbook into your head. And the new large language models turn out to be very, very good at doing that. And what I’ve argued is the best way of thinking about them is not as if they’re new individual intelligent agents who are smarter than we are, or dumber than we are, or malign or benign, but think of them as being like the invention of libraries or the invention of print. When we invented print, suddenly we individual humans could get access to all the information and thought and ideas that other humans had had in a much more efficient way than we could before print. And that really changed our intelligence, that really changed the way that we functioned in the world.
And I think these new systems are going to be very powerful ways of putting together information from lots of individual people. They’re not actually going out there and finding the information themselves the way that even a two year old does. Now, we know two year olds are also learning from other people in important ways, but a lot of what two year olds are doing comes from their own interventions. They’re actually being out there in the world doing the experiments that we think of as getting into everything and drawing conclusions from those experiments. And that’s important because the world keeps changing.
Even though an AI system may be very good at telling you “here’s what people have found out about the world so far”. When the world changes, people in AI talk about this as being out of distribution sometimes, then those systems don’t really know very well what to do. And that’s exactly where human children thrive.
So we’d have to have very different kinds of systems to be able to deal with a changing world and a changing environment. But that doesn’t mean that having a system that can take all the medical knowledge that’s there and give it back to you isn’t really important and useful isn’t a really important and useful thing.
Phil Stieg: Children learn by being curious. They explore, they touch, they feel, they taste, they see. And I would assume at some point we’re going to get better at teaching our computers to be curious. You give them a word, but then they start thinking. They go through the thesaurus and say, oh yeah, all of these words are related to it, and I want to gather information and then maybe give a broader, more diverse answer to the question.
Alison Gopnik: Yeah. And that’s exactly what we’re working on now, is trying to see if we can formalize what happens when we’re curious or what happens when children are curious. We’re even doing things like we can put a child in a virtual environment. We can put an AI agent in exactly the same environment. And now what we discover is if we do that with children, they start exploring right away. They want to find out how does this work? Where does this maze go? What happens if I put this key in the lock? And that kind of exploration enables them to master the environment.
The interesting thing is that when they do that, it’s this very interesting combination of sort of random noisy things just kind of trying this, but it’s not completely random. It’s also sort of intelligent and sensible. You’re exploring the right kinds of things and it’s actually been very, very challenging to get AI systems to do that. You can get AI systems to say, go and look at anything that’s new, but you don’t want to look at just anything that’s new. You want to look at the things that are relevant to the problem that you’re trying to solve that are new or the things that are new and interesting and informative. And that’s what we’ve shown that even babies can do and that’s very challenging for our current AI systems.
Phil Stieg: You talk about babies learning in kind of noisy environments and using their creative ability to create solutions to particular problems. What are your thoughts about the environment for a child and learning? Meaning that is it better to be in a quiet environment, exposed to multiple opportunities and then expecting the child to be creative or just putting them more of a real world? New York City, you’re walking down Broadway in Times Square exposed to a myriad of external stimuli and seeing whether they can somehow solve the problem. Which is better or are both equally good?
Alison Gopnik: You know, I think in general this gets back to the reason why we have this long childhood is so that we can deal with this wide range of different kinds of environments. And from that perspective, a lot of it depends on what you think the environment is that the child is going to have to deal with as an adult.
So if you’re growing up in big Eastern city I grew up in Philadelphia and Montreal you might learn very different kinds of things than if you’re growing up in a small town the way my husband did for example. You know, in a big city, one of the things you learn is don’t pay attention to a lot of the people around know, just pay attention to the people that you know. In a small town you should pay attention to everybody because you don’t know whether they’re going to be useful to you or not. So there’s differences in just the kinds of you have to learn how to cross the street.
What we’re skilled at – the kinds of skills that we develop depend on the environment that we’re in. If you think about children learning, for example, children love to be in the real environments that their parents are in. That’s their favorite thing to do. So if you’re cooking, children, even two year olds, they want to be in there cooking with you. If you’re talking on the phone, they want to be talking on the phone. If you’re going for a walk, they want to go for a walk. And they really want to master those skills that they see that are most important to the people around them.
And just integrating children into your daily life is a much better way of teaching them. I think a lot of contemporary parents think I have to be a teacher, so I have to do something that makes me be like the kind of teacher that you’d see in school. And in fact, just doing the things that you would do anyway. But integrating your child into those every every-day activities, I think, is much easier and more productive, both for you and for the children. Famously, it’s the big cardboard box that the toy comes in that the children really like to play with. And it’s the mixing bowls and the whisks that are the great toys that get used when the fancy learning toys get left behind.
Phil Stieg: One of the key things that I took away from your book was this concept of the difference between the gardener parent and the carpenter parent. And I would hope that personally, I would fall in more of the gardener parent. What’s the advantage to one over the other? And what can you take away from those titles?
Alison Gopnik: Yeah, I think it’s ironic. I liked the fact that I had a nice, vivid title that I think captured some of what I said. But of course, it’s sort of frustrating that when you say there isn’t such a thing as parenting, you shouldn’t be trying to get particular outcomes. The first thing that someone asks is, oh, okay, I get it. So I should be a gardener. So how should I do that again? Like, what’s the instruction book for how to be a gardener parent?
Phil Stieg: Well, they haven’t read Tiger Mom, obviously.
Alison Gopnik: Am I doing it all wrong because I’m trying to get my child to do in particular to accomplish particular things? And again, the point is, it’s not about doing it right or doing it wrong. It’s about having a child who’s growing up in your world, in your setting, with the kinds of values that you think are important. And the way that you pass on those values is often just in the course of your everyday conversations. So you might think it’s really important for children to be able to do certain kinds of things and really train them to do those things. And that’s something that often is really helpful for children. I mean, I think, for example, really training your children to take care of younger children is a really good thing to do.
So it really depends on what the circumstances are, how you’re going to grow up, how the children are growing up. I don’t think there are any prescriptions, and that includes prescriptions that say, oh, you should be a gardener parent.
Phil Stieg: Allison Gopnik, thank you so much for being with us today. I think your insights into the development of the child’s brain and also how we can use our understanding of the child’s brain in the development of artificial intelligence has been enlightening for all of us. I think also the key components that you bring to light is about the role of parents in having the time to actually spend with their children is extremely important, and I thank you for that.
Alison Gopnik: Well, thank you for having me.