How do babies learn to crawl, walk, and run? In this podcast, we talk about learning with Dr. Vanessa Simmering. She’s a Senior Research Scientist in Learning Solutions and studies learning and development.
Last year, she wrote about how babies learn to walk in a blog. In “Lessons from Learning to Walk,” Simmering shared three ideas about the foundations of learning and how they relate to “higher-order” skills like social interactions and language development.
For this podcast, we also discuss the Stroop Color and Word Test (try it here), crystallized versus fluid intelligence, and the Dynamic Systems Theory of learning.
Joining us on the podcast are Gunter Maris (listen to his interview on The Wiring of Intelligence), Michael Yudelson, Kristin Stoeffler, and Saad Khan. They share their views on learning and intelligence.
The views and opinions expressed in this podcast are those of the authors only and do not necessarily reflect the official policy or position of ACT, Inc.
[Vanessa Simmering] When you watch babies learning how to walk, they’re really bad at it at first and they do it very differently every time. But then as they do it more and more, they learn what works better and worse, what has the better outcomes, what makes them not fall down as much and so they’re able to reproduce that more often and that reproducibility is what we termed stability in the attractor state and that is the specific signature of learning in Dynamic Systems Theory. This increase in stability in an attractor state that occurs through experience.
[Adam Burke] That’s Vanessa Simmering talking about the Dynamic Systems Theory of learning. She’s a senior research scientist in learning solutions at ACTNext. Dr. Simmering studies learning and development and lives in Madison, Wisconsin.
We’ll talk more about the Dynamic Systems Theory but first I asked a few of Vanessa’s colleagues about learning and intelligence. Here’s Gunter Maris the director of advanced psychometrics at ACTNext.
[Gunter Maris] Learning is change usually for the better, that you’re not the same person tomorrow as you are today and it happens because you are instructed, you experience things. But it’s always change. Intelligence emerges out of the process of learning. It’s not something that you have but it’s something that literally emerges through your life experience and then it looks at the end as if you always have this intelligence. But if we would replay your life and we do it slightly different, you could either end up at a totally different intelligence level. So, intelligence is not something that exists. It’s something that comes into existence as a result of your life experience, biological maturation but the logic that learning and education and that goes the long way back to a pretty famous quote by Skinner.
So, Skinner wrote in a very nice article, “Once you have formed the noun ‘ability’ from the objective ‘able’ you are in trouble,” and that’s exactly what psychologists and educational researchers have been doing. We’ve been thinking about ability, something that you have whereas you should be thinking about things that you do. You don’t have an ability or abilities. Only the ability to learn but after the learning you are able or unable to do certain things and that’s what we can observe and that’s what we should deal with.
[AB] Now let’s hear from Michael Yudelson, like Vanessa he’s also a senior research scientist in learning solutions.
[Michael Yudelson] Learning is improving on something like a scalability. There are many names for that something as one is engaged in the focus to work toward learning that. Something so focused or could be reading, listening, looking at examples, single step, multi-step problems, constructing problems for others. Teaching is also learning.
[AB] Kristin Stoeffler is an assessment designer working on measuring and teaching creative thinking
[Kristin Stoeffler] Learning to me is improvement with a skill over time and I think that’s sort of a the traditional and foundational approach to what learning is for me it’s in my work it’s more about you know if I was to say that a learner was successful in something that I’m building or an environment that I’m creating for them to cultivate a skill or better understand a process that is I would define more as you know are they able to take this skill set and not just maybe acquire it but illustrate that they can use that and a variety of different contexts or a variety of different scenarios so that’s how when I’m building something I would say someone had successfully learned it because they’re able to take that process and show me that they can do it in a new a new task or a new environment not just did they improve on that from point A to point B but were they able to display that or demonstrate that across some different contexts or with some different content.
[AB] And now we have Saad Khan. He’s the director of AI and machine learning at ACTNext.
[Saad Khan] When we really think about human intelligence, many times we’re thinking about situations where people are able to adapt to new environments and problem-solve their way to achieving their goals and that very much is linked with the concept of learning. I believe when you take information as you would perceive in one context and distill it and save it into memory as knowledge which then can be applied in a different context, that is innately, I think, learning in action and is I think also a signature of intelligence or human intelligence.
[AB] I asked Dr. Simmering if there’s a difference between measuring learning versus measuring intelligence.
[VS] Some of the simplest definitions of learning are that any time you show a change in your behavior based on past experience that is learning and that kind of learning can be seen in all sorts of organisms. So very simple animals and lots of the early theories of learning were constructed from research with different species of animals and what kinds of things they could learn. But when you look at other types, other branches of psychology, people are interested in maybe higher forms of learning, you might say not just that your behavior shows some change from your experience but that that change is adaptive and that you’re better at something from having learned so that you can either find a faster solution or solve a new problem and that kind of learning looks more at what we might call generalization or transfer. You take an experience in one context and you bring that to bear on what you’re doing in a new context you have just learned a fact you’ve learned a way to process. Sometimes people talk about mental models of how things work and that as you learn a model for how something works so that would be kind of like a cognitive representation so that’s getting away from the behaviorist part. It’s not just that your behavior has changed but you actually have a change in the structure of your mind that reflects what you learned and so that type of learning we would expect to have much broader consequences then the simpler the behavioral learning we talk about different kinds of conditioning so that you can associate you know a smell with a taste or a song with the place that you were those are kinds of learning but they tend to be very narrow and specific and don’t lead to lots of new kinds of behavior but you might show a change in your behavior from having learned that.
I would say those are the more behavioral and cognitive ideas about learning and then those can be extended further in education when we talk about learning we are talking usually more about those cognitive ideas and those constructs and trying to think about really what the representation is in your mind and there is some neuroscience research that’s trying to pin down what is evidence of learning in the brain so you can see do you show different kinds of connectivity in the brain or activation in the brain when doing a task based on what you’ve learned before and so that might be something where you don’t necessarily see big differences in behavior maybe when reading for example.
When you’re learning a new language and reading a new land a new language you might be able to produce the word that you’re reading through different kinds of strategies and that so the behavior might look the same that you’re pretty in the word but if we were to look at how your brain was approaching that task we might see differences in what parts of the brain you’re recruiting to do that kind of task when you learn different things as you learn more about the language how you want to measure learning is you want to see often can you do something faster can you do it with fewer errors and then like I talked before about generalization. If you can take what you learned in one situation and apply it to something new but a lot of the tasks I know of that are supposed to be measures of intelligence are more like that last type of learning in like a fluid intelligence task you’re often asked to identify a pattern and how something is changing and so as you go through multiple patterns in a row there are little parts of it that repeat and if you pick up on those parts that repeat you learn that in the test and so then you get better and faster at the end because you’re picking up on what’s kind of cluing you in from one part of the test to another and so I think when we talk about fluid intelligence that ability to pick up on new information and use it for something later part of that fluid intelligence definition which i think is learning but may be learning may be so fluid intelligence is like the pinnacle of learning and the most sophisticated kind of learning but fluid intelligence is usually contrasted with crystallized intelligence which is just sort of your knowledge base crystallized intelligence is more like what I would say is a memory or like your vocabulary is your crystallized intelligence and so it’s unlikely that in the process of doing a test you’re going to gain a lot on that and so that’s why it’s thought that that’s more static and it’s just what you know and so you expect that while you’re testing that there’s no learning happening in the test versus in the fluid intelligence you’re actually looking at how learning is occurring in the test as part of what you’re measuring.
[AB] Maybe a better way to ask this question would be what’s the difference between measuring crystallized or fluid intelligence?
[VS] I think the expectation is that it doesn’t matter in a test of crystallized intelligence. We would be able to give it to you forward or back it wouldn’t make a difference but in the test of fluid intelligence that would make a difference because what you’re doing in the beginning of the test is supposed to lead to what you do in the end so it’s kind of like what you learned up until now.
Crystallized intelligence is trying to get like a snapshot of what you have learned and fluid intelligence is trying to get a picture of what your learning process is like, what you’re able to do, how you’re able to learn in that context of the test and then apply it later in the same test.
[AB] How does learning happen?
[VS] So I think this also depends on what kind of learning you’re interested in and also depends on I think what level of analysis you are interested in people who study brain mechanisms have pretty specific definitions of learning there so what you expect to see and how the connections get stronger through repeated use but that kind of learning doesn’t always lead to changes in behavior and like the kind of generalization we would be interested in.
So in my own theoretical training, I’m what’s called a Dynamic System Theorist and we think about learning in a particular way. It’s a little complicated to explain but we talk about behavior as being the product of soft assembly, which means that you take different components of a system and put it together in flexible ways. You can think about pieces of knowledge or you can think about parts of the body or any kind of system can be described this way. If we think about using pieces of knowledge and soft assembly in that way that you can use the same kind of language you know to write a newspaper article or write a fiction story or write a poem or a song and you’re using all the same kind of knowledge but you’re using it in different ways. When we talk about learning, a lot of times we’re talking about that soft assembly process and the learning to put things together. That’s the kind of learning that’s of most interest in Dynamic Systems Theory and using pieces of knowledge or parts of the system in a particular way is called an attractor state. It’s a specific relationship among those parts of the system and how they work and it doesn’t have to be exactly the same all the time but it’s mostly the same probably. An easier example for this is in different forms of locomotion so most adults can walk or run or crawl okay and those are different attractor states so your legs and your arms know what to do with each other when you want to walk versus run and walking and running for example seem very similar what your arms and legs are doing but there’s some qualitative difference some shift so if you’re standing on the tread when we’re walking on the treadmill and increase the speed there’s some point at which you’re going to stop walking and start running okay and that’s a transition between two attractor states and so how your legs and arms work together is the attractor and when we talk about learning and Dynamic Systems Theory getting practice in that attractor state makes it more stable so it makes it more similar to itself over time when you watch babies learning how to walk they’re really bad at first and they do it very differently every time but then as they do it more and more they learn what works better and worse what has the better outcomes what makes them not fall down as much and so they’re able to reproduce that more often and so that reproducibility is what we term stability in the attractor State and so that is the specific signature of learning in Dynamic Systems Theory is this increase in stability in an attractor state that occurs through experience.
[AB] Okay you wrote about this.
[VS] I did write about this on my blog.
[AB] okay we’ll put a link to that in the website. You talked about the stability and so it would be easy for me to say stability in terms of walking but you’re able to stand on two feet. What would be another example where we’d be more stable that’s not a locomotive or physical?
[VS] So this is a psychology task a lot of people are familiar with, it’s called the Stroop task. This is when there are words it’s different color words but they’re written in different colors than the word is so the color red is written in green and your task is to name the color and not read the color but word reading is a much more stable attractor state than color naming. You will have a really hard time naming the color even though you can do it if I just show you colored squares because there’s not competition between that more stable attractor stable but if you read very well or unless they put it in your non-native language so I know mostly my color words in French or Spanish. The Stroop task for me will be much easier in French and Spanish because reading those words is not nearly as stable as it is reading the English words that I’m more familiar with so that how easily you’re able to do something so in cognitive tasks we tend to talk about it as something’s very automatic or reflexive or you don’t have to think about it another example that’s a little bit of borderline between that’s where stable. If it’s more automatic so whatever it’s more automatic easier to do you kind of have to think about it less those will be the more stable things. It kind of blends between cognitive and motor tasks.
Driving a familiar route, many of us have had the experience where you’re supposed to stop and get something on the way home from work and you totally forget to do it because you have a very stable attractor for driving home from work because you’ve done it many times. You have to really try not to do something that’s very stable and this is actually one of the funny pitfalls of learning whereas if you practice something in the same way all the time then very small deviations from it are hard but if you practice something in multiple different ways then you have better flexibility in switching between attractors.
An example of this is in when kids learn how to solve equations and math equations, they tend to have a very familiar structure where you have like a number plus number equals something and if that’s the only way that kids ever encounter math problems then when you present a different math problem that technically relies on the same knowledge like a number plus blank equals a number it’s much harder for them to come up with what the answer is even though it’s using the same underlying knowledge. Because they have this very stable approach to how you solve a math problem and so spending too much time in only one kind of activity can sort of over stabilize your attractor and make it more difficult to get out of that state. It’s a fine line to walk where in early and learning you have unstable behavior where you’re kind of all over the place and you move through this period of stability where these attractors are forming and you can have some different sets of behaviors but the real pinnacle of learning is when you have multiple stable attractors that you can also flexibly switch between.
Back to the running and walking example, if you’re if you’re a person with really good balance you should be able to go back and forth between those two things without too much trouble but if you don’t have a lot of experience with one or the other so like when kids get to be good at walking they maybe don’t have experience and running yet and so it’s harder for them to switch between those attractors even though they’re what we would call a proficient walker so they’re stable in their walking but they’re not yet multi stable and having different stable states that they can switch between.
[AB] If you’re good at walking and that’s a stable state, does that make you good at running?
[VS] Quite possibly not, so there’s some really interesting yeah great about the same it kind of depends on how close it is so there’s usually an area around an attractor state where things kind of get sucked into it so if you’re trying to do something that’s very similar to what you’re good at you’ll end up just doing the thing that you’re good at and if you get far enough outside of that then if you’re still kind of putting the pieces together in a similar way you can get some benefit from it.
There’s a lot of interesting research by Karen Adolph, who I mentioned in the walking blog that I wrote, where she’s looked a lot at a lot of this kind of learning and she looks at it in babies who are transitioning from crawling to walking. She looks at their knowledge of what is a safe slope to this to descend. She’s got all these little ramps in her lab and what she finds is that when babies are at this transition point and they go to crawl down a ramp they can figure out very easily if it’s too steep but if you put them into a walking posture, they have no idea. It’s like they don’t transfer because knowledge of what a safe slope is defined in that attractor space so what you can crawl down and what you can walk down or not the same there’s some relationship between them but because it’s a different space you have to learn it all over again when you walking and so most people don’t go through an example of that in their adult life that they can remember but anybody who’s ever gone through a temporary mobility limitation. Let’s say you injure your foot or you know your crutches or something like that you tend to go through some of these stages where you get let’s say to a flight of stairs and now you have no idea if it’s gonna is it hard for me to go up these or down these stairs because your knowledge of walking and stairs are linked in that same kind of you have a practice walking on one foot exactly and so now you have to learn how to walk on one foot and there is some savings from knowing how to walk on two feet but it doesn’t work exactly the same way so you might learn to walk with one foot faster because you knew how to walk with two feet but it’s not an immediate transfer so you might see that you gain stability quickly but you have to start up unstable to begin with.
[AB] I asked some follow-up questions about dynamic systems and crystallized intelligence.
What’s the dynamic part?
[VS] What’s the dynamic part? So the attractor state itself is dynamic. The example I like for this is when you sign your name, you don’t sign it exactly the same every time. You don’t move your hand exactly the same way so if you did it exactly the same way it would actually not be dynamic, it would be static. It would be the identical. It would be like a stamp but that’s not how we do it. But we are consistent enough that we can recognize your signature versus a forgery and so it’s that ability to very similarly reproduce something that’s the dynamic part of it.
You never do exactly the same thing twice but you get in this sort of space that it’s very similar to what you’ve done before and so it’s that constant bringing together of the things so that every time you sign your name you have to create your signature in that moment and so it’s a dynamic process so there’s no module in your brain that comes out and says this is how you sign your name. You have to create it every single time. The harder analogy to make is then when I am producing these words right now, every time I produce the same word it’s actually different. To make that word this time it’s not exactly the same as when I say it next time.
Knowing what seven times eight is, I have to create that knowledge every time and that is very different from this cognitive idea of the structure that I have this knowledge in my brain. We tend to think of these like little diagrams of nodes and you know these things and they’re connected in this way and that’s very static. But you just have this thing in your head that 7 times 8 is 56 but you have to produce that but yeah so usually the cognitive idea is that that’s in your head and you just have to find how to get that information out to the world from a dynamic systems you have to create that knowledge when you’re asked what is 7 times 8 but the fact that you’ve created that knowledge many times before makes you really good at creating that knowledge which makes it appear that you always have it. It’s the appearance that that you know that. It kind of makes it sound like there is no knowledge.
[AB] And so why isn’t that crystallized?
[VS] The fact that we expect it to be the same all the time it’s why we call it crystallized. But if you look you can’t point to the part of your brain that is the 56 node or whatever.
My graduate adviser John Spencer does some neuroscience work now. He didn’t when I was in graduate school but he’s looking at some of these questions. It’s like how do you how does your brain actually represent the same information at different kinds of times and how similar does it actually look to itself when you’re making that same kind of memory representation or if you have to make a judgment do these two things are they the same as at what I saw before those kinds of things so looking at how the brain actually does it. He’s definitely been inspired by what we know about how the brain works and that there aren’t these little static things that just get reactivated but they’re actually changing all the time they’re just staying similar enough to give you the similar behavior when you know you’re asked on the test.
[AB] My next question is about motor skills and mental skills and we often want to divide those- the classic mind-body divide. What do you think? Is our motor skills learning motor skills different than learning cognitive skills?
[VS] I think there’s a lot of similarity. I think learning motor skills is a lot more transparent because we can see the parts of the body and how those parts are working together so thinking from a systems perspective when I talk about walking your legs have to do things your arms have to do things your trunk has to do things the floor has a big effect on you but I can see all those things and pretty easily understand that those are the components of the system that matters as an observer you can you see when your child is walking and how they’re doing right and so you can kind of get a sense of how the learning is going and whether maybe they need to take off their socks or something like that.
You can imagine what the parts of the system are, how they’re working together and what kind of intervention might make it go better. I think the similar concepts exist in cognitive learning but we don’t always know what those parts of the system are and we can’t necessarily see when somebody’s having a hard time.
Let’s say they can’t memorize their multiplication table. They’re having a hard time. I always had a hard time with the sevens and the eights and my multiplication tables and I don’t know why it’s something about those numbers. I still don’t know why. It’s hard to know what are the pieces of knowledge that go into learning that line of the multiplication tables. It’s different from the lines that are easier to learn and so I think that there still are those parts of the system that need to learn how to come together right but it’s much harder as an observer to note to identify what those parts are figure out which one is giving you the problem and then what’s gonna be the solution to make those parts work together better because it’s in a black box yeah and we don’t always we’re not necessarily good at articulating it ourselves especially when we’re working with young kids.
Being able to think about your thinking or your learning is called metacognition and that’s something that kids don’t get very good at until they’re coming along in the elementary school years and even then there are big individual differences. You can even see in a single person. We might have good metacognition about our reading but bad metacognition about our math. So there might be easier harder things for us to even understand in our own learning and so that ends up being the challenge is figuring out what are the parts of this system that are working together for this one particular outcome I’m interested in and is everybody using the same system get to this behavior that I want and if they are how do these pieces work together and how do I make sure that they have the foundation of each of those pieces and the opportunity to bring it together in the right way embodied knowledge is that and I know this is part of dynamic systems you embody knowledge.
[AB] And is embodied knowledge basically motor skill?
[VS] That’s the easiest way to think about it and talk about it. So, like the example I gave with the babies going down the ramps, that’s an example of embodied knowledge. I can understand a slope understanding a slope while crawling and understanding a slope while walking the visual information is very similar but it has a different meaning in those different postures and so that’s an embodied perception of a slope be that example you’re not looking in a diagram of a slope exactly. You’re saying well, that number is 30 degrees and I know that 30 is here and here it’s not like that. Making that judgment, even perceiving whether this slope is safe or what it looks like, is different for the position that you’re in, so those are the examples that I think are really easy to come up with.
[AB] How is cognition different than embodiment? How are those two knowledges different? One won’t become the other or are they totally separate or I just not knowing that you can’t see I can’t see that this child doesn’t knows his multiplication tables. I can see that they can’t walk or kick a ball.
[VS] Yeah, we classically have approached cognition as disembodied in psychology so we think of it as these little isolated pieces of knowledge that people just have but everything has to be learned through some modality you have to see it or hear it too for that information to get inside your brain and so we tend to neglect understanding how that knowledge is first built and we think that once you have the knowledge it’s just completely separate from how you learn to that how you gained that information so that’s where I think embodiment can still come in or I usually use the word contextualized learning more because I think it also gets to the fact it’s not just about your body but it’s about what were you doing the day. What other things were you learning at the same time? Was this something you learned in your normal teachers classroom or was this you went to the other teacher’s classroom for this day? What things do you bring to the situation when you’re learning are you learning with other kids next to you are you learning by yourself and those kinds of contextual pieces might change what knowledge you actually have from that situation and when we’re trying to assess knowledge, we usually just want to know that decontextualized piece do you know what 7 times 8 is but maybe you know you can come up with 7 times 8 in some particular context and not in another context.
An example I use to teach about cognitive development is Brazilian street children. These are kids who grow up without formal schooling in Brazil and they help their families sell food in the street market and so these kids are very proficient at calculating totals and making change. All of what you need to do to sell food. But if you give them a written math test that has the exact problems that they’re able to do in selling food, they typically don’t perform very well because they’re not used to that context of paper and pen. Thinking about the number 20 versus thinking about that this sells for 20 you know the number 5 times 20 is oh I want to buy 5 of these things that cost 20 each. If the equation isn’t contextualized in that way it’s harder for them because they’re really practiced at this context of, oh you need five of these things. They cost this much each. They might not know that’s called multiplication but they know how to do the process to figure out what you owe as the total and they might know the process of how to make change but really have a hard time with that abstract symbolic representation. That’s an example of that more contextualized knowledge, where if you’re asked to do reporting time as a number of minutes versus a portion of an hour and that the portion of an hour comes from the analog clock easily the quarter hour but most kids now are learning from digital clocks and so saying it’s quarter past two it’s 15 not 25 which calling it a quarter. Yes, it is a quarter hour but contextualizing that in an analog clock makes it a lot easier to understand versus it turns into just sort of a rote fact that you have to learn that 15 minutes is a quarter of an hour that you can do the math to figure that out but having contextualized it in the circle makes it the way that you know what a quarter is or thinking a quarter hour versus a quarter dollar. Those are the same word but very different contextual meanings so if I put you in a context where you’re thinking a lot about time or you’re thinking a lot about money and then you have to use the word quarter in some way that conflicts with what you’re thinking about is gonna be a lot harder to use that knowledge even though I wouldn’t say that you don’t know what a quarter is. It’s just that it has a slightly different meaning in those different contexts.
[AB] Okay thank you Vanessa and you’ve given me a lot to think about.
That’s our show. Thank you, Vanessa, Mikael, Kristin, Saad, and Gunter.
Did you catch our recent podcast with Gunther Maris? Look for The Wiring of Intelligence in episode 8 of the ACTNext Navigator podcast. Next month, we’ll post an interview with Saad Khan about artificial intelligence and machine learning. Thank you for listening.