Cognitive Load in the age of AI

Hello World, and welcome to the podcast for educators passionate about computing and digital making.

I'm James Robinson, Senior Learning Manager here at the Raspberry Pi Foundation, and this is the first episode to accompany the latest issue of the Hello World magazine, which explores critical thinking in the age of AI.

In today's episode, I'm talking to James Abela, director of digital learning and entrepreneurship at Garden International School in Kuala Lumpur, Malaysia.

He is also the founder of the Southeast Asian Computer Science Teachers Association and ReadySetComputing.com.

Welcome to the podcast, James. It's great to have you with us, how are you doing today?

Oh, really well, it's been a lovely day for me because it's the first day of my holiday. I fetched my son from the airport and had a pretty easy day, so it's been lovely.

Excellent. Before we get into our chat, do you want to just introduce yourself?

Yeah, sure. I'm James Abela. I've been teaching computing for 11 years now at the iGCSE and A level. Before that I spent two years teaching computing to primary students, and before that I was in the UK. Right now, and I have been for the last 11 years, I'm working in Malaysia for a school called Garden International School, and we like to think ourselves as one of the best in Southeast Asia for teaching computing. I have a team of six computing teachers and we love it.

Great. And it's really great to have you on the podcast.

So you recently authored a piece for the magazine which explored how pupils can manage cognitive load within deeper thinking activities.

Could you perhaps tell us a little bit about that article. One of the key things that I was talking about is how many people find it very difficult to get off the ground with a blank page, a blank piece of code, a blank page, you know, why hasn't everybody written their own books, etc., etc.?

And a lot of that is that that initial start takes effort. And the idea of scaffolding is that we give people the help that they need to get started, and we try to take on those challenges to get things running, basically, but also to provide those guides.

When we see students getting really frustrated. You know, coding is one of those things which if it's not done right, people give up in frustration.

And so we really want to be looking at managing the cognitive load. And I say the word managing not lightly. That means not too much, not too little, and really trying to get students into that state of flow with a hyper cognitive they're really, really going for it.

And they're learning at the pace that they best possibly can. And so scaffolding is a tool to reduce cognitive load. But there are things to make things more difficult as well. So I enjoy that kind of challenge and getting people to think about that and make sure the teaching is as good as possible.

And I think that's really interesting I know that's something that I've struggled with in the past, and I've seen it, you know, when I've been in the classroom, I see that frequently.

And I think different learners are going to struggle with different tasks in different ways. And so we need a multitude of different scaffolding approaches.

I think I really like that point. And we'll come on to it probably a bit more as we talk about that, that balancing, it's not about just reducing it, it's about managing. Perhaps we can start by saying a bit more about that.

You have talked about managing, not just reducing. Is there a sort of a sweet spot, and what are the risks of either going to far in one direction or the other, making it to structure too scaffolded and too supportive, or not having enough. Okay. Where do you see the balance?

So I'll bring the example of two easy in from maths not meaning to offend any maths teachers, but you know when you get those maths where you get like 20 of the same exercise and by the first five you've probably got the concept and yet you have to keep on doing this simpler thing.

And really badly designed computing works the same way that it's step by step by step. I've even seen courses where you have to go through the full prim model for every little step, and the kids are getting really bored at this point and turn off the subject.

And also because there's no mental challenge, they forget everything the minute they walk out of it. So they've done the test, they've forgotten about it, and they've, you know, it's disappeared. And the irony here is, is that students, if they're frustrated, will probably still tell you they're bored because they don't want to admit that they're frustrated and they can't do it.

That's not true. I will tell you straight away, I have never met a single student in the thousands that I've taught that cannot learn computer science and cannot learn basic programming enough for their everyday life to understand the concepts.

There isn't a student out there.

If they can talk, if they can solve a problem, they can learn the basics of computing.

But it has to be through really well managed tasks.

A really good point, if you've got too much scaffolding in, you can get what's called the expert reversal effect.

So the scaffolding that you've put in place as support, those novice learners, those that are sort of need that extra support, can actually become a barrier to those that have gained that experience, feel more confident, but they still have to use the same scaffolding to achieve the task.

And that can become it can slow them down, it can become frustrating.

And then at the other end of things that you want to reduce, you want to have enough scaffolding that you can make the difficult thinking, the hard thinking possible, but it still you still want there to be some hard thinking.

You want there to be a bit of friction, a bit of struggle.

Because, I mean, I believe that learning should be like there should be some challenge, there should be some challenge to offset against the reward you get.

So how do teachers know when they've struck that balance?

Well, obviously the productivity in the task, students start to enjoy it.

You can see it literally on their faces when they're in the zone.

And I'm afraid this is where experience does come in.

But if you have options in terms of scaffolding, you can choose when to do it.

And because we teach computing, we're very lucky that we can instantly deploy that scaffolding and only give it to certain students.

Even basics like Google Classroom, we can choose who we can give this to.

If you haven't got anything like that, you can have like prepared cards and things that you can just give out to the students.

And if we're looking at primary students, it's a really good example with scratch, right?

Scratch have made these amazing little cards that are designed to give scaffolding.

But the point is you don't just give them straight away, you give them when you can see the students struggling where they're not making progress.

And so that's the important point, is not just having the scaffolding and just handing it out. It's giving it to them when they really need it.

And you know, this is where I would argue that actually lamination still has a purpose. You know, those little bits of scaffolding that you're going to pass out just when the kid needs it and they're really struggling and it gives them a few hints and tips.

And I think that's really, really important.

Sometimes those hints and tips are just aid memoirs because the kid has forgotten something, and sometimes they are there just to provide that support in terms of how a concept works.

And we're getting to a point now where you can use web pages, you can use all sorts of things to really help you.

And one of the things that I have done is make my very own IDE.

And what that does is it does have a manual built in, but I don't tell them where it is exactly, but I can point them in that direction when the students need it.

And I think that's really important is.

But also, we're at a stage now where I can just literally flick a course at students and say, why do you have a go at this debugging course? Because that will help you with your projects and things like that, because we have so much available to us now, it's about making sure that we give the students what they want and give it to who we want, really personalising it.

With a bit of a bit of an anecdote, remember, I mean, it's been a long time since I've been in the classroom now, but when I was training, like, what was the current kind of thinking around supporting all students was you plan for your like your average student, and then you extend for the ones that are going to achieve, and then you provide a parallel activity for the ones that are maybe going to struggle more.

And I think our thinking has moved on so much more from that. But in that actually what we're probably doing is planning for everyone to achieve kind of pretty much the same goal. And then we have a multitude of different personalised approaches to scaffolding that we can deploy to help every learner get to that point, or the majority of

I think that's that's been a bit of a shift, I think, in certainly my thinking of teaching and learning, but also I think culturally within the teaching profession is much more about much more great, bespoke scaffolding.

Yeah, I think bespoke is good.

And the one thing I would add is the very, very best teachers will have students who perhaps have done some coding outside or done something outside of school.

So I would say for them we don't just extend them, we give them alternate project based tasks because we know they're capable, because what you shouldn't do is give the reward of more work for doing really well at the work.

So I would say, and I have something which I quite often deploy, which I call a jump task.

So it's like a mini test.

And if you pass that test you can A jump ahead and B you can choose to do a project your own because clearly you've learned what you need to for this.

And I don't want you to waste your time, but hopefully that project is more rewarding, more interesting, and naturally a lot more challenging.

And I've found particularly in the groups like in year nine, where we've got students who are competing at national level or they're competing in coding competitions and so on, or they've done extra work at home, read a book, or joined some kind of club.

Well, we don't want them wasting their time on loops for the 50th time. We give them a project.

And that's what I would say is not just extend, but think about different genuine differentiation at task. Not for the bottom, but for the top.

so I think we often I recently I've heard this term sort of crop up this cognitive offloading kind of idea.

And I think sometimes this idea, particularly with AI, that it's inherently bad, but we cognitively offload all the time in teaching and learning.

When we in maths, we use physical manipulatives to support the early conception of numbers.

Your scaffolds used physical manipulative, your diamond nine and diamond five and so on.

So I think we often using those kinds of tools to help take the thinking and the mental load out of our brains and facilitate that might be a useful framing before we go and start talking about AI.

But yeah, do you want to say a bit more about those kind of scaffolds that you're using and how they support the the cognitive load of the.

Yeah, I think the difference between just general AI and something like a diamond nine or a diamond five is that often it's the teacher that's using the AI to provide that greater personalisation.

And I think what we have here is a situation where we want them to learn at a certain speed.

And these are well proven, fairly old things, perhaps reused, reinvigorated and made much more quickly.

A good diamond nine when I was teaching way before AI could take me an hour to come up with, you know, because I want to think of the elements. I want to put it out, I want to chop, I want to paste it, whereas now I can do it in 1 to 2 minutes because my diamond fives and nines now have an AI to generate this for you, and because there's coding behind it, the students can instantly drag and play with it.

So instant deploy instant starters and really good to get the lesson off to a good start.

But those are well proven educational technique.

The problem we've got at the moment is that the AI's aren't at the moment like teachers. What they will often do is they will say, oh, I'm not supposed to give the answer. Therefore they give you a long explanation, but at the bottom they still give the answer.

And what does the kid do? Skip straight to the bottom, grab the code and do it.

And I think that's the big problem is if the students are given some of the support tools, what they're actually doing is perhaps getting far too much help. They'll say the AI was helpful, but actually what they've done is tried to grab the answer.

And that's where really skilled teacher is different. They will say to themselves, is this something I can give clues, hints to? Or do they genuinely not have the knowledge?

And that is the art of teaching is knowing whether they're likely to do it, how many hints to give it.

And we're not quite there with the AI yet.

I'm not saying AI won't be able to do this, and I've seen really big leaps in terms of study modes and things, but it still really wants to give you the answer, and I think that's a big challenge.

Yeah, I think as my personal use of AI, I think you have to spend quite a bit of time pushing back and showing restraint and saying, no, don't do that. Like and you have to kind of coach it in the way that you want to consume and learn that information, which I think for many young learners particularly, that isn't a skill or restraint or a kind of discipline that they've got to develop.

Perhaps we can just go back briefly to the diamond nine and five, because I think if teachers who are listening haven't come across those, what's the, the, the purpose or what are you trying to achieve through those tools that, yes, you might be generating through away? What's the kind of the underlying learning that you're trying to

Yeah, well, the fact is that this is one of those examples where we're trying to increase the cognitive load, because what we're doing is we're not asking for recall or just knowledge.

We're asking for prioritisation, which requires analysis and decision making.

So that's much harder to do.

But because we've given this kind of scaffold and we've given something to do it, we're creating discussion.

And good old Vygotsky talks about the power of discussion and in terms of working together.

So you're creating a situation where people could be working together, but also they have a framework on which to discuss things.

And I think that's really powerful because it increases the cognitive load from recall.

It gets you to analysis and decision making, but it doesn't do it from a blank page.

And that's why it's quite a well studied and and good method of doing things.

Yeah.

And I think there's there's lots of tools out there out there like it, but it's this kind of as you're saying, it's it's all kind of mentioned earlier on. It's making that hard thinking that we want to get to achieve possible accessible in a, in a, you know, limited sort of time perhaps.

I think there's another just on the AI point. I think there's a really interesting paper that I read recently. It's actually from some colleagues over at the Raspberry Pi Computing Education Research Center.

They analyzed lots of studies of how AI was being used in lessons. And a lot of it was it was fairly passive on the students parts, and it was largely substituting what, like, you know, existing teaching methods.

So it wasn't being used to enhance or to kind of really make a substantive difference to the teaching and learning. It was mostly for passive replacement type stuff rather than really, you know, using it to its fullest.

I'm not I'm not in the slightest surprised by that because the temptation is that if you're a non-specialist teacher or you've got too many kids in your class, you say, just use the AI because the AI will provide you with the answers.

But honestly, that's no different to looking at the back of the textbook for the answers. Is it now?

There are plenty of students, you know, when you're teaching that if you're teaching maths or whatever, there are those students who, if you do have the answers in the textbook, they will only go and look after they've had a really good go at it.

Right?

But there are plenty of teachers who say you have to rip those pages out of the textbook because we don't trust you to do it.

And that that, again, is teacher expertise at action, really, isn't it? Deciding how much you can allow the students to do it and how much you want to do it.

And I've seen all sorts of models to allow you to do it.

But you know, C3 before me is a classic model in computing. Have you heard of it?

Yeah.

Well, in this case, ask your friend, ask the AI, ask your teacher.

But it does require a lot of effort to get it right so that they do really try for themselves to have a decent go before they just try and find their friends.

And, you know, the friend often gives the answer.

And even in that situation, we're teaching the friends. Don't give them the answer, give them a hint, give them a clue.

And actually, kids get quite proud of. The fact is, I didn't tell them the answer. I gave them a hint.

You know, they're still struggling and you know, we have to go through that roll through.

And I think it sort of it sort of connects with the idea of a more knowledgeable other is, you know, is a well-documented thing in research, this idea that you're working with a partner, they might know more, but it's how they support you. That's the important thing, not just that they give you the answer.

How might AI in the future support our learners in terms of managing cognitive load, but also what are the risks that we might see? We might have

Yeah, I think I think the benefit is the amount of personalization it can provide a teacher.

But I think this is the main benefits I've seen in AI are in lesson preparation that you're able to give that level of support, scaffolding that you might not otherwise be able to do with your preparation time.

So you can do that.

I think that's the huge benefit.

I think that AI is getting better in terms of things like Notebook LM for revision.

So Notebook LM, if you don't know it is a revision tool. Now it's very tightly monitored against the materials it's given.

So it has a much lower level of imagination of making up things because it's much tighter in terms of its algorithm and it's given specific tasks.

So I think revision teacher preparation, it's really good at with things like the study mode, I don't think it's a great idea for unrestricted use.

And I also read in recent study that saying a lot of these AI's, that so-called tutors and stuff, they're not motivating the kids because they are basically interactive textbooks.

And that's certainly what I feel when I ask it to generate.

So if I ask it, oh, I don't know if I ask it to do a linked list, it will come up with the most mundane, boring examples from the textbook possible.

It's only because I tell it to localise it to make it about my local area, which is Mont Kiara, about the shopping center there, or about our school, or about something that's interesting.

Even things like the World Cup, will it actually make something interesting?

And it's a teacher being able to contextualise, make it interesting, make it about the students lives.

I mean, one of the funnest examples we had recently was all the students like to go to various restaurants nearby.

And so I made a random picker, but I didn't make any random picker.

I made a random picker for all of the restaurants that within a kilometer of our school, the students loved it and they loved engaging in that activity.

And that is not something from the textbook, but it only took me a few seconds to put those things together for an AI to make the exercise.

And I think that's really important, is using that teachers knowledge of the students together with the AI, to expand what is presented.

And I will say, even the last 2 or 3 months, both ChatGPT and Gemini are making a lot less spelling mistakes.

They're getting better and they're getting better at more esoteric languages.

Like three months ago, I would not have trusted it with Swift, but just last week I was able to make a whole code editor with Swift and using AI to make things much quicker.

And it makes things interesting and it makes things fun.

But it does need, you know, the teacher being involved and a certain amount of knowledge beforehand.

It's a lot like project based learning. You need to learn the knowledge first before we can give you that freedom.

think that the idea that teachers can build really interesting and engaging interactive kind of tools that are really bespoke to their context, their learners, their needs and can do so like you can vibe code.

I don't like that term, but you can, you know, you can make something pretty quickly that does that sort of interactive stuff.

I think that's really powerful. In your article, a lot of the examples you talk about are focused on the kind of ethical, moral, or kind of the really kind of sticky, meaty, like, you know, parts of the subject where learners are having there's no right answers, and learners are having to kind of way up different things and prioritize.

Is that for the examples you've given of the diamond now and the diamond five and so on, are is that the kind of the best fit for these?

Can you use the same kind of approaches within more technical subject matter areas of of computing?

I think it's one of those things where I'll just be honest.

There are other methods when you're teaching coding, and I think that's that's really important.

Instead of a diamond nine, we might use a Parsons problem, which is where you mix up the code.

And that really is scaffolding.

So it's designed to give you the code so that you have less syntax problem.

And that's one of the tools that I would use, say to teach coding to give them a bit of a sample in a mixed up order.

Now it's a little bit different to a diamond nine or some of the other tools, because those are tools to encourage higher level thinking.

This one here is a similar idea of mixed up things, but it's designed to reduce the thinking and to give you the code, or at least a part of it.

Now you can be really mean with things like Parsons problems and make it harder.

So for those really able kids, you might take out 3 or 4 of the actual pieces of the puzzle, and they hopefully will solve it in the same problem.

So even within those kind of things, it's a similar technique, but it's being used in a different way.

And I think the other thing is the other model I would bring in here, which I know we probably mention, every podcast is the print model, and I think that's really important.

And that was one of the few models where I felt it didn't just simplify thinking, it also allowed you to make things more difficult because you don't have to use the whole prim model each time.

You might say to your more able students, right, go straight to make, you can get on with it, have a go.

But for some of the other students, they will need every step of that model so that they have an understanding and they can make something useful in the end.

Yeah, I think it really nicely exposes the process of programing as opposed to focusing or emphasising the end result.

Another thing I've seen done in Parsons problems, which I think is quite mean, is red herrings, lines of don't feature in the program at all.

You've got to like discern what's the right bits and then sequence them.

I've really enjoyed this conversation So, James, we've talked a lot there about where we are right now in terms of how we support students.

Where do we see this going?

If we had this conversation in six months time, in 12 months time, or in five years time, Nothing can.

Education happens in less than a year.

There will be the few who are very keen, who will be motor speeding, who will be using the AI's, and their lessons are probably ahead of everybody else's anyway.

It takes academic years for things to change, and it takes even longer for exam boards to change.

So, you know, even basics like digital exams, they're going to take 3 to 5 years to happen.

And will they be based on very old premises.

Yeah.

I mean, there's a lot of stuff being talked about in terms of bitcoins and things that were flavour of the month 5 or 6 years ago, but have largely disappeared.

You know, this bit chain and things like that.

Blockchain, those things were interesting, but they didn't really go anywhere.

And so firstly, I would like to say the examples will probably get to where we are maybe 5 to 10 years.

But what I will say is I think the AI's, because they are so changing of everything that's going on in our lives, I think we will head towards a model where the AI's will provide that knowledge.

They will test you, they will be able to do these things, and the teachers will really be around to motivate, to focus on relationships, which is what we always say.

It's always about the relationships.

And I think that's going to be even more so.

I have said for a couple of years now that lecturers are infinitely replaceable, great teachers are absolutely not, and they will be the last to be replaced because it is absolutely all about the relationships.

It's all about the motivation.

It's all about getting the students to understand those basics and what's going on.

I also think there's a major pushback right now against AI, and I notice in the likes of Sweden, they're saying no way under 13.

So there will be a great need in 12 to 2 to three years to look at that, 13 to 16 to 18.

And how do we engage with AI?

I also think and it's not strictly computing, but I think it's really important.

The safeties, in terms of social networks, AI and that area, that kind of digital learning, I think will become more and more important.

And it won't be computer science, but it will be part of computing in terms of keeping people's online lives safe, in terms of not having electronic friends that give everything can take nothing.

So I think there's a lot happening in those five years, and honestly, a lot of this should be happening very quickly.

But I reckon unfortunately probably won't happen until 5 to 10 years.

We only have to look at social media and how slow education and governments have been with regard to that.

Is a real challenge is right now teachers are you know, they're living in a world where students are, you know, they're experimenting with AI, they're using AI outside of the classroom.

And it is kind of this Wild West, unregulated kind of space.

And I think, you know, that is a challenge.

And a lot of teachers will be kind of concerned about how they verify the the validity of the homework that's being done.

Has this work been done?

And I think for teachers, I think a lot of the sort of work around cognitive load and thinking about where to kind of place the emphasis and the learning on making sure the students are doing the thinking, and they're not easily able to hand that thinking off to an AI agent to kind of do it for them.

But how do we.

So I think that's a really interesting.

Yeah, the next few years can be really interesting, I Well, thank you so much for joining us today for the HelloWorld podcast. James.

Really enjoyed chatting to you today and hearing your thoughts on cognitive load and and how it impacts on students learning.

Thank you so much for having me.

It's been an absolute pleasure and it always is.

Talking to Hello World and the Raspberry Pi.

Cheerio. Thanks, James.

Well, that's it for today.

If you'd like to read the full article that inspired today's episode, you can find it in the latest issue of Hello World magazine.

And whilst you're at it, if you haven't already, head to HelloWorld to subscribe, explore our back issues and listen to more episodes.

What are you waiting for?

Until next time. Goodbye!