Our brains are not just memory machines! A Conversation with Dr. Laura Desirèe Di Paolo

Welcome to the Future Learning Design Podcast.

It allows us to put together active inferences as a very simple and elegant principle and works for learning. It works for development. It works for knowledge.

Yeah.

And that is the key. The children develop because they learn things and so they know things.

Hi everyone, thanks for joining me for another episode of the podcast. My name is Tim Logan and the podcast is brought to you in partnership with Notosh. Over the last decade, there's been a surge of interest in the evidence and research backing that supports our choices as leaders and educators in schools. In an effort to prove what works, they reach for the cognitive and so-called learning sciences, Personally, I've always felt a bit uncomfortable about how narrow these discussions are, focusing only on memory, cognitive load, retrieval, and other terms that you'll be familiar with. As you'll hear in this conversation with Dr. Laura Desirée de Baolo, our brains are just not simply memory machines. In fact, our mind-brain-bodies are incredible prediction machines performing active inference constantly in order to make sense of and act in the world. Laura is a researcher and philosopher of cognitive science at the University of Sussex in the UK. And the Moritz Stern Institute at the University of Göttingen in Germany. Her research is focused on the material world in education, development, and evolution, and she's currently working in Andy Clark's team on the project Xscape: Material Mind, exploring the interactions between predictive brains, cultural artifacts, and embodied visual search. You can find links to much more of Laura's work in the show notes, but the thing that brought her brilliant work to my attention, which you'll also find there, is her paper Active Inference Goes to School. The Importance of Active Learning in the Age of Large Language Models. Hey Laura.

Hi.

How are you?

I'm fine.

Awesome. So this conversation has come out of primarily your recently published paper called Active Inference Goes to School. And as I say, I've been exploring this idea for a while, and that paper really was the first time I've seen anybody really making that explicit connection, and there's the AI connection, we'll come to that maybe briefly, but particularly the explicit connection between this active inference framing of cognition and learning and the way that schools are set up. So let's start because obviously we know that there have been other ways of thinking about learning and education and schooling historically. There's always been a progressive, marginal progressive stream of educational thought. Maria Montessori, for example, being a, a a very strong and important proponent of that. You quote her at the beginning of the paper, and I just, maybe that's a nice way in just to start thinking about it. 'Cause you, from The Absorbent Mind, which is 1949, you quote her a couple of quotes about movement. And I wondered, yeah, without movement, it's not possible to speak of an individual, which is Maria.

Yes, in Italian works, we had to struggle because the problem with that book, it's, I have to say, it's amazing. Okay. I have been reading and reading again because now I'm working on another paper also on the Montessori on the extended mind, particularly about extended mind in terms of learning. And I have been rereading that book over and over and she had lots of insights that today. So when, when I looked at some papers of the, the ones that you have just quoted, by the way, written together with Hutto, and I thought, okay, but these are exactly the same sentences that Montessori wrote. In 1910, in 1930, in 1940. The problem is that first, the translation of the book in English is honestly awful. We have just submitted a grant for retranslating the book because the English version that we have, that we all have at home, it's more like an interpretation.

Wow.

She talks literally about extended mind. The concepts are already there in terms of the four ideas of the active inference, right? So you have this cognitive system that is always coupled in some ways with the environment because otherwise it doesn't understand anything. And it has a mechanism of precision weighting and of error dynamics to update all the time what we think. So the beliefs or the set of information that it has in the model. They are already all in Montessori, the repetition. Like for instance, why the curricula in the Montessori system are not one year or a few months, but they are spanned on at least 3 years, from 3 to 6 years. The reason is exactly because she understands that for making the learning going deep, you need to repeat the same actions at different levels of understanding because our understanding of the world, it's very multisensory. It's very complex. You don't get on just a student that's sat on a chair and looks at the teacher, and the teachers can do whatever they want to attract their attention, right? Yeah. They can yell at them and say, oh, look at me, look at the board behind me. They can be as appealing as they want, but they cannot stop all the rest of the information that comes through. Even if a child is just sat in the classroom, he hears the birds singing. Thinking outside of it.

Yeah, yeah, of course.

He looks at what the classmates does, and all this information is constantly going in. If the lesson is even slightly boring, this information comes immediately way more up in your level of priorities, right?

Interesting. No, this, so I mean, because we're diving in deep already, so— Yes, sorry. No, no, this is good, but I'm really conscious of the fact that for some people listening, this will be the first time they've heard about active inference. So maybe we'll hopefully come back to Maria, but if we could just perhaps start just to distinguish the current kind of dominant or what has historically been dominant kind of view of the way that learning happens, maybe could you just speak a little bit about that? I don't even know what you would necessarily call it. I mean, cognitivist framing, but I don't know, like, but what is, what are the assumptions built into that? The way that we've just thought about learning.

In the way in which we thought about learning and particularly the way in which it's applied in schools is that everything happens inside the skull. That is the first assumption.

Yeah.

You know something because you are able to memorize a series of notions and information that a teacher has provided to you. And you are able to somehow reformulate this information in a way that you can give the answers to tests. This is more or less the way in which our schooling system works.

Yeah.

In all these approach, the parts that are called in the paper, but in general in cognitive science, embodied cognition. So the role of all our body, our sensory system, our motor system, our way to perceive reality is not taken into account. Another aspect is not taken into account what you use core learning. So the devices, the objects. Of course, you have in some respects, you have curriculum, right? In which you have a series of notions that are parcellized and put in a step of incremental difficulties, and you pretend children, young adults, to acquire, to go deep into this information that has been pre-divided for you. But your body, your understanding, everything that you are as a person, as as a student is not taken into account because first, the things that you are forced to learn, they might be completely not interesting for you. Second, the way in which we perceive these things, the way in which we read, you and me can read the same book and in the same book find completely different things, right?

Yeah.

Interpreting in a completely different way or highlighting in our mind completely different aspects of the same book because we are different people.

Yeah.

And these differences are somehow nullified. So we have an educational system and an idea of learning that is somehow standardized based on age. Classes are divided by age, assuming that all the children of roundabout 9, 10, they will behave roundabout in the same way, which is kind of not true. Depends. Depends very much. Then you have a context. So what I'm doing right now is more or less defining the 4E account of cognition. The context, which is called in the 4E account the embedding. So there are a bunch of studies that look at our classrooms, for instance, but in general, our physical surroundings for understanding how they resonate or not with our cognitive system. And you have kind of a Goldilocks. Spot, so like a sweet spot for which you need to have just the right amount of information. If the, the environment is too rich, you cannot focus your attention. If the environment is too poor, of course you don't have enough stimuli. Okay. And then there is this idea of extended mind. So the idea that we use devices and environments for performing cognitive processes, that in reality, our cognitive processes, most of them are done together with these embedding reality that we have.

Yeah.

The classic example would be for someone that writes. When I write a paper, I think while writing. I can like have some ideas before, but the reality of thinking happens to me in front of the screen with a keyboard under my hands.

Yeah.

Because in this way I can see my own thinking. This is the approach that we have right now. That's completely opposite to the old style of approach in which thinking, knowing, First are completely unrelated to one another. So you have knowing as a set of information that you need to retain in terms of memory. So I know what happened in the Second World War. I know a bunch of grammatical rules. So I know the subject, right? But most of the time is in fact matter of memory is not like really knowing because I don't feel what I know.

In your body. Yeah. Yeah.

I tried to explain this to my daughter. My daughter is 10 and I tried to explain to her at some point she had some difficulties in learning some piano and told her, well, look, the difficulties are normal when you go ahead in learning things, but there is one moment in which if what you are doing, you like it and you are putting effort in doing it, you have that moment the moment in which things make sense, right? We all, and I hope we all have experienced this feeling, but it's so seldom in schools, I think. I agree. Absolutely. That you have that aha moment because, well, some of the things that you're learning or you're memorizing, they might resonate with you, but most of the time that's not the case.

Somehow kind of a connectedness or fittedness or synchronicity or some, I don't know, the words escape me, but some, the way that those, all those layers that you're talking about all kind of align in a moment of embodied. And the other E being emotional, right? You say you like it, like there's some, there's emotional components to this as well, of course.

The emotional component, for instance, when you asking me like two different systems.

Yeah.

If you think about an idea of learning and knowing, first and foremost, divided. I've studied for a very long time social learning. Okay. That was my obsession, uh, in, uh, my previous work, how we transmit information across generation for forming our cultures and how this is different, for instance, from the cultural or traditional way of other animals. And one of the main thing is, for instance, all the studies on cultural learning, they focus a lot on the role of imitation.

Yeah.

Right? Tell me, what's the first thing that's not allowed in schools? You are not allowed to copy.

Yeah.

Copying from a classmate, it's immediately considered as something negative, which has at least two aspects. There is the aspect that you cannot do it, you are not allowed, but there is the emotional component in which you are socially stigmatized if you do. But the copying is one of our most natural ways to learn things.

Yeah, exactly. Amazing. Can I ask you, because I think this idea of separating is really important, like separating learning and knowing, as you're saying, but also I think it feels to me like there's also this, maybe it's kind of what you were saying about the embeddedness, but there's a deeper aspect of, like for example, Anil Seth talks about this kind of idea that we've historically had this sense that the world is out there and we, and all the sensory information is kind of coming in. So if you were to draw the diagram, all the arrows are coming in from this outside outside world, right? So we're not in that kind of embedded, connected way participating with it. We are almost like we are observing it sensorially through our different senses and we're taking in that information from the up and then we're processing that information. So there you've got a separation between perception and cognition, right?

Yes.

And I, I think, you know, for example, if you look at the diagrams about cognitive load theory, right?

Mm-hmm.

You've got all these arrows coming in. To the eyes and the ears and into the brain from the outside world. And then you hit these bottlenecks of information and then, you know, our cognitive architecture, our short-term memory gets overloaded and we can't cope with all this information coming in from the outside. And these ideas are really rife in education, right? And I'm not saying there is no limit to your short-term memory. And I'm not saying that, you know, the memory doesn't have certain architecture and all of that, but that sense of the information coming in from the outside world, and we are just, our brain is an information processing machine, and then we do the learning and then we end up with the knowing, is, as I understand active inference framework, is completely wrong, right? I mean, it's completely inaccurate to what active inference is saying about the way that perception and cognition are much more closely aligned.

Yes. And you have a third component that's development. That's also a very tricky one.

Okay. Okay.

Let's explain with something like hands-on.

Yeah.

If you are a 16-year-old and you are in a class and the bell rings and changes the subject, or you have to go to another class, right? And you move from mathematics to, I don't know, biology. Okay. You expect the teacher of biology to come in and to do your class. So you put on your desk, the biology book, very likely, or notebook or whatever you need. And you expect that a certain teacher is going to be there. That is a very clear case of predictive processing or active inference. If you know that there is the class of biology, but at some point a completely different teacher comes in and starts to do art, you would be very surprised because this is not our cognition processes world. We know that things happen with a certain regularity. And we expect them to happen with these regularities, but these regularities is something that we put in the world. And this putting into the world, it's projecting ideas based, of course, on a set of experiences of what is going to happen into the world, which also the emotional components that we were saying before, if you don't like biology and you know that you're going to have your class of biology is not just postponing to opening the book or just having that feeling of, okay, now I need to deal with this for 2 hours, right?

Yeah.

Because all our system, which is integrated and works all together, is already projecting what is going to happen, your own reactions to what is going to happen, and your own expectation. If at some point your teacher of biology has changed and it comes a very interesting person and makes your biology class super fun, right? You will be surprised. These two principles are the baseline of the active inference. So in one sense, you project what is going to happen, how it's going to happen, how you will perceive, how you will feel, because it's been proved till that point as a reliable key for understanding the environment, right? And the second aspect is the surprise. If something doesn't match with what you have expected to happen, you have this, oh wow, this wow moment that makes you rethink your own ideas of the world. And these are two basic principles of the active inference, which you have like a top-down, it's called, so from active way. To perceive the environment. You don't expect the information to come through, but you try to read it in advance. I am ready to understand it.

Yeah. And I think it's important to say that obviously your example of the lessons changing and that kind of temporal aspect, that's a good example. But also this thing is happening at tiny, tiny micro levels as well. So for example, in a visual illusion, you are expecting to see certain things and you either get an oh wow of You know, unexpected, like the example of the— Anil Seth uses this— of the squares on the checkerboard and the colors of the white squares, and one of them is in shade. And because we, our brains expect things that are in the shade of an object to be darker than they actually are, we predict it to be lighter than we actually see it. So there's, I don't know, I just wanted to make the point that this is happening really at this very minute level of—

I am explaining in terms of behavior because it's easier to get, but of course we are talking about like perceptual information, how you see reality. So literally where your eyes are looking at in a very specific moment in time, what you are hearing. The latest book of Andy, um, of Andy Clark started this example that it was a conference and he heard his phone ringing, but he was 100% sure that the phone was off. How is it possible? Because you expect it. It's normal for us to hear the phone ringing. And if nothing happens to your phone for like half of a day, you will be very surprised, right? But I can use a very old example from Jakob von Uexküll. So like we are talking about the first ecology studies of more or less the same period of Maria Montessori. He describes this event in which at some point he was on lunch or dinner with some friends. And he was looking for the bottle with the water inside and he couldn't see it. And his friend, he said, well, it's there in front of you. Since the bottle that had been on the table till that moment had been changed because it was broken, he literally couldn't see the bottle in front of him because it's not what he was expecting to see.

Right, exactly. Yeah. And the other, the other example is the x-rays that Dave Snowden uses a lot about the gorilla. There's experts of looking at x-rays and analyzing the x-rays and they can't see the x-ray of the gorilla because they're just clearly not expecting to see that in an array of, of other human x-rays. So yeah. So I'd really love to spend a bit of time if we can kind of breaking down some of the implications for learning, but I just, I just think it feels to me like this kind of information delivery mode that I, I mean, you referenced Ken Robinson at the beginning, right? That kind of industrial, the role of a teacher in this industrial system of delivering information. And that I think fundamentally has been based on this idea that the world is out there and we, we just need to find the right information to deliver into the cognition of these young people who are sitting in front of us as a very crude explanation of it, right? But, but what we are doing, I think, with active inference and with predictive processing is saying, that something fundamentally different is going on. And everybody's brain and body embodied self is making predictive models of what they're about to see and what they can expect to see. And then they're evaluating constantly the sensory information that they're obtaining and they're comparing it to their predictions and they're judging, like you say, the, the oh wow moments of whether it surprises you or doesn't surprise you based on your prediction. So there's something fundamentally different about the way that we think about this inhabiting of a kind of a learning environment, right?

So for how I see it, I hope I'm not going too far, that was exactly our aim in the paper, which I hope is going to be a first of a series. It changes completely how we think schools. And the reason is because the idea of how our educational system has been made and has been transmitted so far As you said, it's very much historically determined in, in point. It comes together with the industrialization, comes together with the necessity for people that were working in industry to be able to read and write information that were hanging inside the big factories, right? That was the beginning. So the information is not the information that an actual embodied system perceive from the world, but the information that someone else wants them to have. And this is the main problem with active inference. These children, young adults, these students, they live a world that is outside the school as well. So they, their cognitive system is already settled with certain level of expectancies. Then they go in school and everything is at the best, a wow moment, at the worst, it's just errors all the time. There is nothing of all the rest, how we make friends, how we do things, how we like things, that is actually important within the classroom. And this is a big problem. This explains a lot why people lose very often, um, motivation in studying and learning.

Yeah.

And I don't know you, but I mean, if you talk with most of people, you will get how much of the information that you got in school you retain now. For me, probably one class, the rest was, has been completely wiped out because my brain needs space for actual information.

Yeah. So my, my personal experience was not that. I mean, I got a lot out of school and I appreciated school, but because I was good, I understood the rules and I was good at the information processing game that was being played. Right., right? And I could play the game and I'm quite conscientious by nature, so I wanted to play by the rules. And you know, that's part of the point of the whole system, I think, is that it privileges certain people and certain personality traits or certain particular dispositions that most people are not like, right? It's a very small minority of people that get to kind of win at that weird game that gets played in these kind of industrial schools. But I think what a lot of what you're saying So far, people, I'm sure, will be kind of hearing resonances of, you know, Rousseau or Dewey, or, you know, all, all of people have had these kinds of insights all the way along. But the thing I get so excited about with the active inference framing is that they've always been trying to put those framings of what a nurturing learning environment would be like. They've been putting those framings on top of this very cognitivist information kind of processing model. So therefore, for example, you take, you know, a pedagogical approach like project-based learning or inquiry-based learning with these more kind of progressive approaches, you might say, they have to justify their own existence because they say they're better at helping young people learn the academic knowledge than the didactic method. Right. And so they're trying to compete at this same game of memory processing and academic learning and all of those things, but actually What is going on, if you accept the active inference framing, is something completely different that I think things like inquiry have really understood, that there is a fundamental coupling between the learner and the environment that needs to be taken into consideration when you think about learning as it actually happens through this predictive modeling and testing, rather than learning as it happens in this kind of purely information process.

So the main thing for like Putting on a lower level, as we said before, we have 3 aspects, right? One is learning, one is knowing, and one is developing. Okay. Most of our implementations in schools, they are based, for instance, on certain ideas of development in terms of stages, milestones, what a child is supposed to achieve at a certain age and so on. And you try to fill in knowledge, which is literally a bunch of pre-organized information in terms of linguistic or symbolic information most of the time.

Yeah.

And learning, which is the way in which you acquire and you deal with this information. And the problem with whatever implementation, even the best that you have, we have right now, for instance, the one that I like a lot, I don't know if you know the work of Elmore, the late Elmore, and it's called Learning by Design is the general.

Okay. Yeah.

But I'm referring particularly to the work that he has done with, I have it because otherwise I mispronounce the name surely, Prakash Nair.

Oh, okay. Yeah, I know that, but yeah, I've seen that one.

Together, working together with designers.

Yeah.

Okay. And also in that, which is a fantastic approach in my opinion, because It means exactly, okay, how do we improve learning? Let's start not from what the teacher must say, but let's start from where the kids are at every moment. Let's change the rules of how they behave because of the environment.

Which is what Montessori fundamentally understood, right? In the prepared environment.

Yes. Yes. So I don't think there is like, there is not the perfect educational Okay. I love a lot the constructivism and I love a lot deconstructionism. I love a lot the place-based education of David Sobel. All of them, they point to some aspect. Exactly.

Yeah.

The Montessori method is, at my knowledge, up to today, the most precise one. If you reread, that is also the reason why we have just submitted this grant hoping that it goes for retranslating The Absorbent Mind, is because in the Montessori method, all these aspects, the environment, the objects that you use, development as a push from the individual in choosing, doing, paying attention to things that resonate with them, and the connection between development, learning, and knowledge. It's already very well defined.

Yeah.

Okay.

Can we, would you mind if we spend just a little bit of time on each of those things? Because I think that idea of, so one of the phrases you use in the paper about learning is epistemic action. And I think there's something important as well. We've said it slightly with the embodied stuff, but I think it's worth dwelling on that learning is not just a brain-based process, right? It, and that's the other thing that this active inference seems to do is pull together action and perception together into the same, because this is back to your quotes from Maria Montessori about movement, right? There's something really fundamental about the way that we make sense in the world through moving objects just as much as we do about perceiving objects, you know, in that participatory way. So yeah, what's learning in that sense is epistemic action. So what does that phrase mean?

Okay, here it's kind of a complex— in its simplicity, is kind of a complex matter because there are different ways in which we can understand these epistemic actions. On the one hand, in a very simple and basic way, the way in which we act into the world, whatever movement, even action, even in terms of what we say, yeah, what we see, because that is the nice thing about the active inference, The moment in which I turn my eyes towards something because I heard a sound.

Yes.

Is an epistemic action. It's something that is supposed to give me some kind of information. Okay. All our activities and perceptions are epistemic. So they are a way to gather information into the world or to prove our own models of the world as right. Okay.

Yeah.

Again, let's think in a very— I try to put with very simple samples. In the moment in which you have in front of you a glass for beer or a glass of— for Merlot wine, you will grab the glass in a different way because your understanding of the reality of that glass in that particular moment is different given the specificity of glass, in the moment in which you turn your hands for grabbing the handle of a cup or grabbing a glass and you have a different grabbing way, it's already something that's intelligent in itself. It's already a way to say, okay, I know the world. I know that this glass, I can't grab a glass of beer from the top, right? And I won't grab a hot cocoa from the glass.

You're right, exactly. But if you take then a baby or, you know, a toddler into that scenario, they wouldn't necessarily have the experience or even the— I mean, in the, in the old frame, you'd call it kind of prior knowledge, right? But learning in relation to the way that they might grab it might be completely different because it's somehow— these might seem like new objects. They don't have the same learning associated with them.

So that's absolutely true. Indeed, the young, young infants, they tend explore.

Right.

Why you don't leave things that they can simply put inside their mouth is because they tend to explore and with whatever they have. There is a level for which they come into the world already knowing things. The work of Anna Chonika also that puts together, for instance, active inference and infancy and in utero life already gives us a quite strong case for the role of movement even before we are born, because somehow in utero, uh, fetus moves about. It has its own learning environment. But this allows us to see another aspect of epistemic actions. The work on, on active inference in this case, it's very useful because it makes you able to understand how the single individual understands its world and reacts to its world. Somehow inappropriate ways. On the other hand, the environment, particularly in humans, is already epistemically prepared.

Right.

It's already epistemically scaffolded. Right.

Yeah.

If you look at fetus in utero, the learning environment of the fetus is the uterus itself. What the mother eats, the sound of her voice, the music she listened to. We know that all this information, they have a strong impact after birth. Okay. We already know that. And again, this is helped by the active inference framework because in active inference framework, not differentiating between actions as something that you do with your whole body, but just perception, seeing, hearing, smelling, they are already activities. Okay. On the other hand. When you come into the world, when you are an infant and then a toddler and then so on, you have an environment that's already very much constructed for you in a strong sense. The, the, I don't know if you know the work of Barbara Rogoff. It was, uh, uh, Apprenticeship in Thinking.

Yeah.

It was very famous because she has these pictures with these young toddler, I think less than 2 years old, playing with a machete with the adults looking at him from the background, but not intervening. And this guy was not hurting himself. In our home, in our Western societies, homes, we don't allow kids to play with knives, right? We are very careful in avoiding everything. We put children blocks.

Yeah, yeah, exactly.

Yeah. But in reality, children learn from an environment that's very much prepared in a cognitive sense. And this goes back to our embedding. And this goes back to the idea that our schools as a bunch of chairs and desks and children sat for most of their time doesn't really fit with an idea in which their moving is no Yeah, exactly.

It's like somewhere in the paper, I, oh, epistemic foraging was the other phrase that I just thought was brilliant.

Yes.

Because you're like, you're foraging for information, you're exploring for information, constantly testing out your predictive models, right? You've created these generative models predicting what the world or what the information is gonna be, and then you're taking action to explore whether those predictions are correct. And, and I think that error The prediction error between the prediction you've made and what you then find is kind of where the learning happens, right?

Yes. Yes. The learning happens when something is reported as an error somehow. So the role of mistakes, of errors is very much interesting. Also, again, the active inference, I really like to work a little bit more on development because particularly the younger ages. We all know that children, young children, very good in learning a bunch of stuff. They learn languages without any difficulties. They learn grammar. They learn a bunch of things that if you think about it as an adult, it's very hard work. Yeah, right.

Exactly.

And the reason is because this epistemic foraging in that case, it's even more interesting because they are building kind of the basic models of their interaction with the world, which is the language that people around me speaks. I gotta learn this language because this is gonna be very important for me to be able to tell my mom that I want that particular toy.

Right.

So this is like the epistemic gathering. You gather all the information that you can, you explore the information, particularly as a child, as a, in a situation in which you feel safe. You don't have like the problem that will arrive a predator and is gonna eat you up.

Yeah. But it's, so again, it just, it's funny to me, but in a way, because there's, there's a lot of phrases and clichés around in education now about how, how we value failure and how you wanna fail well and how error is important, must be appreciated. And, but again, everything in the structural system doesn't send those signals to young people. Everything says the opposite in terms of exams and, you know, getting right answers, but people don't quite know why the failure is important. And again, I, I feel that this active inference framing just helps to really explain that it fundamentally drives learning, these error dynamics.

Uh, so there is, again, I should come back to Montessori because she, one of the 4 points that she shares with active inference is exactly error control. And active inference says that the errors are actually our way to do 2 things. Okay. In the moment in which you predict, so to say, how the world should look like to you. What you should perceive in the world and how you could act, and this prediction doesn't meet into the world, you have two fundamental ways to change these predictions. The first thing that you can do is updating your prediction. Okay, I made that 2 2 5. It's not 5. I change my belief. Okay, 2 2 4. That is the first way. The second way. Is to change the world, to make the world fit with my own predictions. So I create environments and structures for which, how I believe that I want to keep my temperature, that's 37 degrees, but outside is cold. I put on a coat, I invent the coat in the first place. So I do stuff that makes me able to say, okay, I want to feel this is my perfect temperature. I want to feel this temperature since I cannot change change the environment completely, I can grab a blanket and changing the situation surrounding me in a way that I am again in my prediction state that my temperature is 37 degrees.

Absolutely. But also the context you are in will allow you certain affordances or not to be able to do that, right? So if you're in a particular situation where it's not appropriate to go and get a coat, or you're in a school where you're told you're not allowed to, or you, you know, culturally there's some norm that's supporting or directing your behavior in a certain way. You're, again, that kind of embeddedness in context is all part of this, what's possible here in relation to my action and then my perception somehow.

Yes. Yes. I mean, if you are in the middle of nowhere, very likely you will try to build or to find a cave, or maybe you start to shiver. So you have mechanisms. For— they have to be— of course you can't— if the temperature drops suddenly to -30 and you are in the middle of nowhere, no blankets around, you will die. There is nothing that you can do about it.

No, but it's— yeah, I mean, it's just within the bounds of what's possible in that space. You are exploring and foraging, right? Yes. Or what is possible there. And I think that's what's really exciting about this whole thing.

Well, let's make it again about schools.

Yeah.

Right? I haven't studied for a particular test is copying or asking a classmate, right? Because this is my way to being sure that I don't have the stigma of the bad grade, but I can still get along with my environment and I find a solution. I copy.

Right, exactly. But also I think the idea that if you haven't studied for a test, the prediction error you're going to get, it's so big because you haven't got the understanding to then to engage with the learning in a way that is going to make, you know, so you're reading the questions, you've got reasonable expectations that you understand the question, you can provide an answer, and therefore there's this prediction error is minimized, whereas if it's so big, there's not really a lot of learning. I mean, a test is a bad example of that, but in any situation, if the prediction error you're getting is so significant, then there's not a lot of learning happening. But equally, if there's no prediction error, because you're anticipating everything.

It depends what you define with high prediction error. I mean, it can be high prediction error, I don't pass directly the whole year. That's, it's a high prediction error, but I learned a lot from him anyway. Maybe not what the school wanted to teach me.

Yeah.

But I learned a lot. If I am someone that tends to have all the time bad grades, the teachers treat me badly because I have some difficulties or I don't like to do the day-by-day homework. Okay. And despite I am a very smart guy, I keep having bad notes one after the other. I learn a lot. From that information, what I learn is that I am someone that doesn't fit here.

Right. You're still learning. Yeah. You always learn.

You're still new anyway learning.

I guess what I was trying to badly explain was the idea of the kind of that learning sweet spot, that there's a, if the prediction error is so high on a particular task, then you're not learning a great deal because it's perhaps too far off.

What you are thinking about is putting the Einstein mathematics in front of a 6-year-old.

Right, exactly. There's not a lot of learning that's possible.

There's very little that you can do there. Yes.

But equally, if there's no prediction error because you know everything, 'cause you're just being asked 2 2 what? Over and over again, there's complete certainty and you're not learning either. So it just, again, it gives a way to explain that kind of Goldilocks principle or the learning sweet spot. Yes. Or the zone of proximal development in a way that integrates back straight into the active inference. Like it gives an explanation for that as well.

Yes. Yes. We, we called it in the paper manageable goals. The idea is exactly that. You need to have something that's— Montessori referred to it way earlier than Vygotsky, unfortunately, as learning at the periphery. So she says learning is something that happens like an explosion and you need to put the child at the periphery of what they already know. They need to have every time something interesting enough. Okay. And for that enough. Hard part means that has to be challenging in a way that can be solved, right? If you have a puzzle like the Rubik's Cube and you are not someone that's completely into puzzles, and the Rubik's Cube is a particularly hard puzzle to solve, what you would do is either you go to a tutorial, you find a shortcut, or you let it go. Okay, that's not my style. I put it away, I trash it, I give it away because it's not my stuff. It's too complicated for me. Right. This, the, the good theory behind it for understanding, it's called the Apprenticeship Hypothesis from Kinsella. It's used on an evolutionary scale, but the idea is like that we learn better when we have like sort of personalized information, but in a, in a very embodied, embedded, and extended way. We deal with tasks from which there is just one piece, miss it every time.

Yeah.

If I give you something that's too complicated and they tell you, just give you the starting point, you might lose your interest and you might never achieve the result. If I give you something that's already done, it's already done. I don't have any reasons for paying attention to it a second time, right? So I need to find just the right amount, which is exactly depriving from a complete task, for instance, every time just a little bit more. So I give you something like already half made in a way that every time you have just to do one little thing.

Yeah. Which again, what you've just said there would feel very familiar to most educators, which is good. There is a familiarity there, but I think what's important, particularly the Montessori example or the active inference generally, is that that's not only a conceptual cognitive thinking act of of fitting a new piece of information into an incomplete picture of information. This is a much more embodied way of interacting and participating with the world as you explore.

For me, what surprises me when I try to, to read stuff in education is that, as I told you at the beginning, this complete gap between what we know about learning, what we know about cognition, and what is actually— do you know, are you familiar with the scaffolded education?

Yeah.

Right. So this is the idea of scaffolded education. When you read papers or implementations within schools means, okay, have some kids that have difficulties in a specific subject. How do I deal with that? And they do exactly what I just described it.

Of course. Yeah.

But this is something that we know in cultural learning that we have applied for understanding how we as humans learn. For a very long time. We know that the other way to see the epistemic actions is the scaffolded environment. You do epistemic actions in an environment that's already scaffolded. It's already prepared for certain kind of actions, activity, right? That goes from the learning inside the schools to everything that we do in our everyday life. You know how to behave on the table according to the specific culture in which you are, because everybody behaves in a certain way, because you have certain tools on the tables, because dinner is served at a certain time. This is all epistemic in some ways. Yeah, it's, it's knowledge, it's cultural content, it's cultural knowledge, and we deal with it all the time. Just, it's completely absurd to me that we have somehow to reimplement it as a new discovery inside schools. And instead of asking the question Why do these kids have problems in the first place for which I have to reimplement something that we do in our everyday life forever since ever?

It's true. Yeah. I mean, it's a bizarre thing, isn't it? That we've created these models.

It's a very bizarre thing.

Yes. That are so false. They're so unlike the way we actually live in our lives. And then we try and reimpose these radically new theories of learning, which we're actually just doing as soon as we leave school anyway. Yeah. Fascinating.

Yeah.

I think if I can, one of what you were saying about knowledge, I think is really important because we have privileged fixed knowledge and what I hear in the active inference, like even the movement and the action and the foraging and all these things are ongoing processes. They don't stop, right? We keep interacting with our environment. I'm doing it now. You are doing it now. It does, it never stops. And The idea of knowledge is somehow stationary, like we can pin it down. And there's something important there.

I think we have, we have done this thing to pin it down. We have done this.

Yeah, of course.

Tests. Yeah. And we have so far talked about 3 of the principles of the active inference. There is the 4th one that in this case is the most appropriate one, which is the precision weighting.

Right.

Understand the foraging to the environment, but we are every time we're more precise in doing so. So when you learn to do, to, if you paint, right, the first thing that you do is drafting the, on the canvas, like the main geometric figure or the proportions. Like here there is the body, here there is the sun, here there is the other. Then you go there a second time and you say, okay, now this is the base and then is the most important character in my, my painting. Then you go again and if you, if you have to do details, you will go there over and over and over. Again, let's think about Leonardo, over and over and over again till you don't have exactly because every time you see something more. In this case, it's not matter of errors, a little, but it's more matter about going deep and deep and deeper in what you know, in how you know, how you perform some certain tasks. The first time that you are in front of a piano, you do some stuff, right? The more you play, the better you are. And the better you are, the more your— all your multisensory systems will hear music, will think music, will be able to abstract music, seeing the music, right? So this is another principle which is— should be really much implemented in the very same concept of knowing. The more you like something, the more you go deep. Because the fastest you learn and the more you can make more precise your interaction with that particular aspect of the environment. And again, this is completely out in our educational systems because there is this idea that you, you need to get out from school that you know a little of everything, that in my opinion, it's very much of nothing because you didn't have time to develop the things that you actually liked to do. And you had to spend lots of time, to waste lots of time in learning a bunch of things that really are not useful.

Yeah. I mean, it's so dynamic, isn't it? There's so much moving because the environment is constantly moving and therefore our predictive models are constantly adjusting and updating. But I think what you're saying there is so important the way that, and you said it at the beginning, the way that the young people have no control often over the environment that they're in because it's being controlled by timetable and the teachers and the, the decision of the education system generally, those choices have been taken away. So they, when the bell rings, they move from maths to biology and then the bell rings again, they go somewhere else. But that idea of them being able to influence their learning environment as it changes, as you say, to go deeper and deeper into this particular thing is the same kind of process of updating predictive models and as you say, precision weighting and those things, but it's just, you're going deeper into a particular thing rather than skirting across superficially across a lot of areas.

You can meet the same thing. So the point is I'm not so good in math, let's be honest. Right. But if I have to read a paper in which there is some math, but I'm really interested in the paper because it's, it's, I'm writing a paper and I need to really understand particular is gonna be way less difficult.

Exactly.

That if I was put in front of the very same math out of the blue, right? So the perception of how something is difficult because is connected with an activity that you want to perform, something that you want to understand, is going to be difficult. We in the schools, in many, because of course schools are different, there are very good teachers out there, there are very good principles. There are very good schools even in the classical teacher-centered approach. We have to make it clear. But the point is that these aspects are not implemented in the idea of the system, for which you find things difficult simply because are completely unrelated to one another. They don't are related with something that you want to achieve, you want to build, you want to design, you want to, you want to, the want to, I think it's fundamental. Having learned about learning for a very long time, I can say without any doubt, you learn the most if you like what you learn. If you don't like it, why do we learn so fast from TikTok? Why do we learn so fast from, uh, because we want to do it. You read one thing that you are interested in one time and you already retain like 80% of the information. I read exactly the same thing because my teacher told me tool, and they will never retain a single word.

Absolutely.

3 times.

Yeah. And I think just to kind of wrap it up, I think my hope for this is that all of those things like intrinsic motivation or the, you know, emotional aspects of cognition, these are— there are lots of educators who are very aware of those things and they try to bring them in where they can, but fundamentally the system is built on a model of cognition, which is completely alien to the way that we actually explore our environments in embodied ways. Right? And so my hope for this is that the science is going to knock off its pedestal, this cognitivist way of thinking about learning so that we can actually start saying the science also supports a different way of approaching this. This is not some romantic fantasy about we want everybody to just be their best selves and explore their environment. This is based—

The good thing about the good thing about the active embed very much formalized.

Right.

It's really a formal way to look, this is scientific.

Exactly.

This is how organi works.

And in some ways it's unfortunate that we need, often we need the science. We've put so much value in the science in our decision-making. I say our, I mean in kind of, you know, Western Eurocentric modern approaches to institution building or whatever, but I think there is weird societies. Yeah, the weird societies, but there is an importance to that that will help the cause, as you said, of people like Maria Montessori who have been saying these things from before the grave and beyond the grave for over 100 years, right? It's, it's incredible that we have taken this long to get to it, but I do feel like active inference is a really important step forward.

I completely agree. And also because it, it allows us to put together active inferences as a very simple, if you wish, simple and elegant principle and works for learning. It works for development. It works for knowledge. And that is the key. The key is that we need to keep together the fact that children develop because they learn things and so they know things. So they develop and they learn things and so they know things. Instead, in the schools that we have inherited, they split these three phenomenon. And so you develop at home and then here you just know things and how you learn, who knows? That is more or less the— I have, I will start very soon on implementation in a school, in fact, in Montessori school. I don't have yet everything done, but the idea is to have a completely multisensory and multi-topic implementation. So the main topic is neurobiology, neuroscience. We'll make some kind of models of neurons. We will see how the neurons interact to one another. And then from there, we will go to machine learning on—

great.

All in English in a German school. So the idea is exactly that. I don't expect at the end, so now it's, this one is going to be a pilot because a study has never been done in this way in which I will do. So I call it an implementation and maybe a pilot for eventual studies. But the idea is exactly to supply some hours in which children are in a complete different way surrounded by objects. And languages and space. And they don't expect all the kids to know everything about neurobiology. I don't expect the kids to learn everything about neural networks. I don't expect all the kids to be able to speak English at the end of the implementation. What I do expect is that the kids that like the language, maybe they will speak better English than they did before. The kids that like to play on and to use glue, they will be much better in using hot glue and in making up stuff. And the kids that like science, maybe they will know something about it. And that's, is my, I don't expect everybody to learn everything. I expect to give an environment in which everybody can learn things.

Nice. And the, so the design of that whole space and curriculum is based around active inference.

It's about active inference and the material engagement theory. The material engagement theory is, yes, again, the Montessori method shares with the material engagement the particular attention to the hands.

Yes.

Because the hands are these, these hands and tool have a particular relationship. We know things by using our hands and our hands become somehow our own tools while we know things. And at the same time, they already filter the information in a very strong sense. It's not the case that they always have this example of rubbing, because they are very important, particularly because I'm Italian, so for us it's also very important.

Of course, always with the hands.

Yeah, it's true. A complete system of communication codified.

Yeah, but again, I mean, you quote Maria in the paper, and I wrote it down actually, that when man thinks— it's about the man, but when people think, when people think They think and act with their hands. I've changed the quote slightly to make it plural, but you take my point that she totally got it right. She totally got it.

She's totally— uh, there is a beautiful, uh, book that came out last year, I think, about architecture. And for instance, if you look at it, the old material respect exactly the same colors and the same material of Kandinsky. Yeah, the, the point the line, the surface. The idea was exactly because she knew a lot about perception. She not only had studied the work in perception done in Germany particularly, and afterwards the old Gestalt philosophy.

Yes, okay.

Ecology. So the idea is exactly that there is this progression in knowing things, and what she did with the material is trying to eliminate the overreach information. So all the material is made of wood, Yeah, it has primary colors, you know, and you add one element, one abstract element at a time. Of course, for instance, in my implementation, I work with stuff that is not in the Montessori material, is not because I work with neurons, I work with machine learning, I work with that. And I work with very young kids. I work with kids that are between 6 and 12 because the idea based also on very up-to-date research. When do we start to construct abstract knowledge? If you go in an idea of development like Piaget-wise, that you use widely in all educational things, you can't build development of mathematical things till you are 12, practically. Instead, we know that that's not the case. It's nonsense. It depends what the kids are playing with. They build their own abstraction. Oh, they know that they have to ask 3 bonbons, not 2, right? It's something that they have some very strong intelligence about the meaning, the understanding of things.

Yeah.

My idea is exactly, okay, then let's, let's try to make this understanding in a way that is perceived and on. We have already made a first trial because I wanted to understand what the kids like to do.

Amazing.

We made like a big two neurons touching each other with pom-poms and with the LED lights and stuff for understanding different transmission and so on. And for me, it was just like, okay, what you like to work with? I love the hot glue. Okay, we will have lots of hot glue.

Nice. Amazing. Thank you. I really, really appreciate the work you're doing and I, I want to try and use the platform I've created to get more of it out there to more educators because I think it's really important. So yeah, no, thank you. We didn't get onto the AI stuff, but maybe next time, maybe next time we can talk about large language models. But honestly, I think that's less interesting than, than this about the way the embodied learning that we just need to fundamentally shift in our schools.

So yeah, yes, there is a very good point that now because, and after Corona, we have a very good chance. We have the chance to change schools. Yeah. And if we don't do Now we will never have another chance. Now is exactly the right time.

Thank you so much. We hope you've enjoyed this episode. Please feel free to continue the dialogues with our guests, with us on our blog or on social media, or within your own networks.