AI and the future of teaching and learning

Good morning, good afternoon, good evening. Thank you all for joining. I know we have thousands of people registered from all around the world. Thank you for taking your time. We are here to talk about AI and the future of teaching and learning and also technology and the role of technology in the education process. I am Rebecca Winthrop. I direct the Center for Universal Education at the Brookings Institution. We have been looking deeply, particularly at this question of AI, starting a couple years ago when we launched the Brookings Global Task Force on AI and Education. And we shared the findings of that task force earlier this year. And there has been a lot of conversation in the US and around the globe around what is the role of technology in classrooms, in schools, in the teaching and learning process. And I couldn't be more tickled to have wonderful experts and friends join us today to have an open dialogue and conversation about this topic, as well as taking your questions. And you can always continue to submit questions at events@brookings.edu. So I'm really pleased to welcome to the platform, please come online, Michael Horn, who's a lecturer at Harvard and co-founder of the Clay Christensen Institute, and Andreas Schleicher, who is the Director of Education and Skills at OECD. Michael will be leading us in a conversation, and hopefully Michael will be sharing your wisdom too. Over to you.

Well, I appreciate the opportunity, Rebecca, and it's always a pleasure to be with you both in this forum and to have this conversation. And I think it comes at a critical moment. In the world around technology in general, and certainly in our schools, as we will discuss. Just to frame the conversation before we dig in, if you will, it is an interesting moment, I think, because there are a lot of seeming paradoxes out there. There is a widespread global backlash around the concerns and challenges and harms from social media. Parents, in many cases, lawmakers are part of a backlash against technology in classrooms themselves. In the United States, for example, we are seeing a lot of legislation regulation, backing up these parents to limit screen time, to limit technology usage and the like. And then on the other hand, you have AI that has been growing in our consciousness, certainly through large language models over the last few years. And I would argue in many countries is actually ubiquitous, whether we know it or not, through a variety of platforms, intentionally or often unintentionally in the classroom and in school systems, certainly. Now, you also have this other weird thing where usage of AI will go up when students in schools are in session. And yet, if you look at higher education and the graduations at colleges around the country and around the world, in many cases, students don't seem very excited about the prospects of AI. So once again, you have AI up and prospects and hatred of it also there as well. And then finally, the last paradox I would just highlight is, even as you have parent parental and student concern over these AI and other technology models. You also have a group of families migrating to brand new school models created by educators that are in many cases powered in some sense by technology and AI. And so we have a lot of different currents. And so I think we've got the 2 right people to help us sort out what is going on and help us understand the signal, if you will, from the noise. And Rebecca, maybe let me start with you on this one. Just let's divide the conversation in 2 parts, if you will. First part, let's just talk about the technology and education in general, not AI specific, sort of more of a reflection of where we are now, but also the last really 10, 15 years. And then the second half, we'll get into AI where you both have done a lot of specific work. But starting with your Family Engagement Education Research Network, like I mentioned the parent backlash to technology, but dimensionalize it, help us understand what's really going on here a little bit more.

Thanks, Michael. Well, I, I— let's, let's start in the US, um, because you are seeing a big backlash around, um, technology in the classroom in the US, and then we can talk about the rest of the world, and I think Andreas, um, can weigh in too. But in the US, I just, I, I think we need to go back a few years, and part of what I think we need to look at is just how are parents doing. So in 2020, 2/3 of parents said, and this is, you know, at the onset of the pandemic, that parenting is harder today than it was 20 years ago. And yes, the pandemic was just beginning. It wasn't full-fledged. But, and a lot of parents said that is because of screens and technology and trying to manage this online world that is confusing and sometimes dangerous for their kids, but yet they have to use it for school and homework. And then in Fast forward in 2024, the U.S. Surgeon General Vivek Murthy under put out a advisory, a public health advisory, which is something that the U.S. Surgeon General does, saying this is an here is an urgent public health issue. And his advisory was saying that parental anxiety and stress was a public health issue because it had gotten so bad. They had done a fair bit of research, and they found that at least 40% of parents. in the US feel so stressed and overburdened that they can't fully function and be there for their kids. And they saw a big difference between parents and people, adults who did not have children. And so, you know, I just wanted to put in context that this backlash comes on top of a really big struggle that parents have been facing, certainly in the US, around raising their kids, navigating the online world. And then we see, you know, Jonathan Haidt's book, The Anxious Generation, came out, which really catalyzed sort of ongoing interests and discussion and awareness that, hey, maybe we shouldn't have phones in school. That, I think, is the most sort of impactful thing that has come from that, from that book. And again, in the US, this was rather a new concept. It was not a new concept in the world a decade ago. So if you look at the rest of the world a decade ago, you know, you had France who was saying, you know, let's limit screen time in schools. And they progressively changed. And Europe and many countries, you know, years and years ago were, were taking phones out of schools. You also had places like China who were saying, you know what, this social media stuff is not good for our kids. It's not healthy. They, you know, have a very different governing system, so could control things and said, you know, frankly, it would be every, you know, say what you will about geopolitics, but every parent's dream is to do what China did for their kids. Online regulation where they said you can only play video games for 3 hours on the weekends and you can't be, you know, scrolling social media between 10 and PM and 6 AM. And so, you know, in the rest of the world, we do see a growing sort of movement. Now increasingly it's against social media accounts. So having young people have a social media account, being able to access social media. Australia was the first one to do this for under-16s. It's catching on. But, but now that AI as a software is everywhere. It's in social media, it's in Google searches, it's in our edtech. You're seeing an increasing sort of backlash by parents. For example, in the US, the Fair Play Coalition with a lot of, you know, child psychologists said we should put a moratorium on student-facing AI use for 5 years in schools. And really at the core of this, you see this sort of wave, at the core of this is a loss of trust. Parents do not trust school systems to do what's right by their kids. And they realize that, you know, you can take, you know, they can take the phones away at home, but homework is still assigned online, and that a lot of kids are accessing social media on their school laptops.

Hmm. It's interesting. Your answer already maybe helps explain one of the paradoxes of lack of trust. But if I've chosen a different school, maybe I have trust in that because I made sort of a conscious movement. Andreas, let me bring you in here though, because I'd love you to give us, you know, some data around the use of technology in education globally so that we all have a shared understanding empirically of what's going on. Rebecca started to mention some of the country differences and the like. Just help us understand about usage in different geographies, impacts both at a macro level, but I also know you think about the micro level, the actual use cases in the classroom as well. So give us some of the data that you all have collected. Yeah,

I mean, first of all, our data show when we survey students that access is now pretty universal. That's, I think, post-COVID, something that you can take for granted even in, you know, less developed economies. In fact, what's interesting, you know, 5 years ago we would have said this is a kind of digital divide between students from wealthy families and less privileged ones. And now this is actually flipped. Yeah. Flipped. You know, we have just completed our Baby PISA assessment where we looked at 5-year-old children. And what you could see is that 5-year-olds from disadvantaged backgrounds have a lot more access to devices than 5-year-olds from privileged backgrounds. And privileged backgrounds, you know, parents are more likely to read to their child, which is a good thing, actually positively correlated to cognitive, social-emotional outcomes. In disadvantaged homes, kids are more likely to see a screen, which the same data show negatively correlated with cognitive, social, emotional outcomes. So I think that in a way, if there is a divide, it has flipped the other way around. But I would say access is pretty universal. Now, access isn't use. I think that's where the interesting story starts. And we see still quite big differences between countries, particularly when it comes to intentional use. By that I mean, you know, teacher-directed work where, you know, this is integrated into classroom practice. There are some countries where that, you know, we look at Korea where that is now, or, you know, China where this is now basically quite closely integrated into everyday teaching practice. And there are some very well-developed education systems. Japan would be an example which actually ends up at the low end of the spectrum. So it's nothing to do with GDP. It's really about kind of policy, the intensity of use. The big differentiator is also technology use for leisure. And that's much more related to the kind of regulatory policies that Rebecca just referred to. And you can see some countries that have become very restrictive. I think cell phone bans are becoming more the norm than the exception across countries, for good reasons, by the way. Whereas we do not see, you know, much of a correlation for teacher-directed work with technology. In some countries it does have a positive, you know, impact, in others not so much. And it's a kind of quite flat curve. For technology use for leisure, you can see a steeply negative curve in every country. I mean, that is pretty clear that basically And it could, you know, causation can always go both ways. You know, you have non-performing students resorting a lot of technology or technology actually, you know, we do not know that. But what is very clear is that that relationship is steeply negative. We also see that you have now in many countries 40-50% of 15-year-olds who say, I can no longer concentrate in my mathematics lessons because of my smartphone or somebody else's smartphone. So I think that technology used for leisure is something where we see huge differences across countries, much more so than with teacher-directed work. And largely that is affected by an outcome of regulatory policies that have been taken. But much of this actually predates AI. I think AI obviously puts a completely new dimension to this. AI also has put use of technology much more in the background. You know, the most powerful uses of AI technologies are not student-facing. You know, again, there you can question it. The most powerful uses of AI are, you know, learning analytics where, you know, students get that kind of information on how to learn better. Teachers get, you know, information how to teach better. Schools get information how to, you know, become more effective. That's, I think, you know, where AI has added a completely new dimension to this. But no, Overall, access is pretty universal. And then exposure is, when it comes to leisure, very much an outcome of regulatory policies.

It's interesting that you mentioned the leisure point because one of my impressions also of what's driving a lot of the backlash is parents, quite frankly, very upset when they find out that their kid was on YouTube for an hour of class. they can't figure out the educational value. It's certainly not active learning and things of that nature. Andreas, you also mentioned that tech use is quite heavy, though, in places like Korea and China that have been high achieving. So maybe help us understand in terms of effectiveness. So the leisure, if that is negatively correlated pretty consistently, where is it effective? Where is it not? What are you learning about sort of the next level question rather than just tech good, tech bad? Yeah,

you know, I think the big issue really is pedagogical intent. And I do think we can learn a lot from countries like Korea, China, Singapore on that, where you have, you know, very clear— first of all, I mean, they will still minimize screen time. Again, they use technology largely as a source of, you know, knowledge, you know, to figure out how different students learn differently and to, you know, personalize that learning. You know, I actually had my first experience exposure to AI in 2018 in Shanghai. It was a really interesting experience. I was watching a primary school, primary classroom learning calligraphy. That's one of the biggest headaches of Chinese teachers. We struggle with 30 characters and they have to learn 4,000. So it's just massive. And for them, it's not just a technique, it's an art. And you saw those students drawing the characters with ink and And on their tables, inside their tables, they had integrated scanners. And on that table, students had a tablet that was giving them real-time feedback on the quality of their drawing. Teachers got real-time feedback on where students had misconceptions. You know, I mean, it was just amazing how on the one hand learning was very traditional and at the same time technology was super empowering both learners and teachers. And I think that Korea would be a country where you can see that with the digital textbooks. It's all adaptive. I think that the big differentiator to social media is that these things are about creative uses of technology, not consumptive uses of technology. It's not, you know, putting stuff to students. It's really about enhancing cognitive activity. Really, I think that's the one thing I would always, you know, point to that, you know, you do not become, you know, fit by watching sports. You become fit by, by doing sports. And the same is with AI, you know, where you use AI to, you know, enhance cognitive activity, to get students engaged, to understand how they learn. You're going to see benefits where, you know, you try to, you know, and don't let the consumptive aspects dominate. You see drawbacks. And I think some countries have really figured it out. What's also very interesting for us is how, how much it the variation there is in actually teachers being on top of this. You have, you know, some countries where teachers are very creative, very active designers of innovative learning environments, and others where they are slaves of scripted algorithms and lesson plans. Then I think that's, I think, another dimension really, really important here. General-purpose AI tools, I haven't seen much, you know, good coming out of this. Most of the, you know, good applications are really designed and used for educational and pedagogical purposes.

Michael, can I jump in really quickly? Just on Andreas's earlier point about how most countries who are high-performing are either using AI in a very pedagogical integrated way. I love the calligraphy example. That's a great example. Well, I would say technology. Let's go back to technology, but it's not heavily one-to-one. Screens. So I had— this is just a story that illustrates this— had a lovely meeting with a delegation of folks from Estonia, Ministry of Education and others. And as you know, Michael and Andreas, Estonia is a very sort of tech-forward country across the board, including in education. And they were sort of shocked to learn that a lot of schools spend kids have 80, you know, at least above 80% of their time face-to-face with their, you know, educational experiences intermediated with a laptop, for example. And they said, that's crazy. We, we have 80% person-to-person, eyeball-to-eyeball. We teach handwriting, you know, when we read books, paper books. And of course, we want our kids to learn technology and have the benefits of technology. And, you know, maybe 20% of the time it's tech intermediated, but it's very intentional and it's not sort of a universal platform, which is really different. A lot of school districts I talk to, they're like, well, we're 80%, you know, our learning is intermediated by technology. So I think we need to completely flip that. I think we've lost our way.

Yeah, and that's a good— yeah, yeah, I think it's a good rule of thumb also for people thinking about it. The way I've thought about it is similar to what Andrea said, which is, you know, the person that does the work does the learning, right, at the end of the day. And so Look, if you're on 45 minutes of a reading program over the course of the entire day because you're getting very direct instruction, you're working to read and it's giving you feedback and so forth, fine. It's a purpose-built technology. Ideally, there's research around it. Ideally, I would argue there should be an outcomes-based contract when the school is working with that vendor so that they're actually very clear about these are the outcomes that we expect. This is the time that students will be on it. And we're both holding each other to that end of the bargain so that there's clarity on both sides. And then Rebecca, to your point, it's okay, so you built some basic knowledge, you built something, and now what are you doing with that, right? And now this is where conversations, small group projects, things like that start to come into play. And look, maybe technology is used in the context of a project. I would argue for older kids. I get suspicious when I see younger kids on the, Laptops making things. And, but again, it's a function. It's sort of like a small part of what you're doing, not the plot itself. And I think that's what we've often flipped where we see this go awry. I want to ask one more question on this before we jump into the AI-specific piece. Rebecca, you think a lot, you know, you wrote The Disengaged Teen. You think a lot About engagement, motivation, which in, in the learning sciences stands somewhat separate from the cognitive science and a lot of the memory work, right? But it points to what Andrea said around the importance of actually doing the work and act— like actively learning as well. How does that all fit with this, right? Like the student motivation piece and how they're maybe moving through their own experience in school.

Well, yes. I mean, my— yeah, you gave a shout out to my book with my co-author Jenny Anderson, The Disengaged Teen. And we found these kids arrive in these 4 kind of learner modes of how motivated they are, how engaged. Passenger mode, I'm coasting, doing the bare minimum. There we find that kids you know, might like school to see their friends, but they're really not psyched and excited and motivated about the learning process. And they're trying to get through as quickly as possible. And they will absolutely use technology to sort of just find the answers or, you know, do the work for them, copy and paste. And they see education as very transactional. Like, you know, you do these certain things, it's task-oriented, do these little certain things, add that up and boom, you, you know, pass. And so they're like, well, let me just get my tech over here to do the things and then I can go on to do what I want. Kids in achiever mode are really trying to get really good grades. They're focused on the outcome, not the learning journey. They are motivated by school. They want to do well. They think it's sort of a gateway to a better life, which is not bad. They're ambitious, they're organized, goal-oriented, and parents love them. So do teachers. But in some ways they end up acting like the achiever, the passenger mode kids with technology because they are actually not that focused on learning journey. They're focused on getting a good grade by whatever means necessary. And, you know, before ChatGPT, that was— there was a high, high incidence of cheating of, you know, from friends and others and passing the, you know, tests around. And there certainly is still a high incidence of cheating because they're just trying to optimize the outcome. And then you've got kids in, in resistor mode who are just like, I'm not having any of this. But kids in explorer mode, which is Really a small fraction we found with our colleagues at Transcend, less than 4% of kids in the US spend time in a— regularly said they spend time in explore mode in middle school and high school, which is where they are following their curiosity. They're interested in the learning journey. They're doing what Andrea said. They are becoming stronger learners by practicing learning. They are becoming stronger critical thinkers by practicing critical thinkers, which you need content knowledge for that in order to manipulate And it's because— and, and they're really invested. They end up getting really good grades. I always say this because you don't actually have to sacrifice performance for exploration, but they use technology as an unlocker. So for example, this is the, the kid who gets excited about tornadoes and then says, Mom, I learned about tornadoes. Do you know about tornadoes? And then goes on YouTube for following their curiosity about tornadoes. And so that is a way that tech can sort of unlock their learning. But, you know, it's really hard for schools to architect, you know, sort of in our current model to sort of architect that sort of explore mode all the time when they, when kids have access to the entire World Wide Web. So like a kid in passenger mode, for example, might be interested in something and be like, oh, that's interesting. You know, go on YouTube because it's assigned and a class might be interested, but then the infinite scroll pops up and it's on, you know, something that MrBeast did and they go on MrBeast, which might be interesting, and then they're just on a TikTok, on a YouTube rabbit hole far from, you know, what they might be curious about. So that's how we see it playing out.

So let's start to shift into the AI. It's interesting you just said that because one of my reflections on all of this also is that Andreas used the word pedagogical intent versus sort of general AI tools. I kind of wish we had single-purpose tech tools as well, as opposed to a computer with everything at its disposal, like the Speak and Spell when I was, you know, 6 or whatever. Might be better at this point. Let's jump into AI, though. You both have done pretty significant reports, sort of collecting, in your case, Rebecca, a pre-mortem, Andreas, a digital outlook around AI. I'd love to hear what both of you found as we get into this. And so, Andreas, maybe you start and Rebecca just jump in right after.

Yeah, you know, the one compelling finding, and this was actually already in our previous PISA assessment, was, you know, when we looked at digital literacy, now, which students are really good at, you know, the kind of critical thinking and, you know, skills. We didn't find any relationship with AI or technology. Actually, the best predictor of digital literacy was the share of students who had read books over 100 pages, paper books. And we ask ourselves, you know, how can this be? And then, but when you start to think about it, you know, you understand, well, if you read long books, you have to learn to build a mental representation of something that is really complicated. You have to entertain all these different characters and how they interact, and you have to sort of engage with the material in its complexity. And maybe that's a way that actually then builds those kinds of AI literacy skills that later on that we look at. So, but basically on the digital outlook, what we found is that there is a big distinction between general-purpose AI tools that are just, you know, put into the classroom and those where you have very clear pedagogical intent. I give you an example that I think is really striking. You take, for example, you give students AI to ChatGPT or whatever to write essays. And of course, you know, they produce better essays. You know, task performance will always improve. But then afterwards, when you ask them, hey, what did you actually write about? 80% of students in that group could not remember. It's a classical example where, you know, you improve task performance but not learning. Basically what AI has done, it has divorced task performance from learning. You know, in the pre-AI age, learning was always about task performance. You learn to learn a quadratic— solve a quadratic equation, you can do it, and then you have learned something. Now with AI, that is no longer the case. At the same time, you know, there, there was another, you know, experiment where they gave students AI to, you know, research a topic for an essay and, you know, understand, you know, different viewpoints and angles about it, and then write the essay in a classical way. And actually, in that case, you did get better products. Students did learn actually more. So it was really the use of the tool and the design of the learning environment. AI tools for collaborative learning, an amazing kind of impact, getting students to learn to collaborate, engage. And I mean, there are many examples, but the key really is you need to understand that task performance and learning are no longer the same, and therefore design your AI strategy through learning, not task performance. And that also, of course, it comes back to Rebecca's point about Achiever mode versus, you know, explorer mode. If you are in achiever mode, you are focused on task performance. You use AI simply to get a better outcome, which is very easy to do, but not necessarily with learning. Second point, you know, you need to learn to think before you learn to prompt. Now, that question of age appropriateness is really, really important. Now, if you're not ahead of the AI tools, if you do not understand an algorithm, you're going to be the slave of that algorithm. That is, I think, very, very clear. But actually, again, most of the users of AI in education do not account for that. If you confront students with a lot of AI tools that are well above their kind of what we call AI literacy kind of skills. So I think that is a really important part. Putting teachers at the heart of the design of those tools. What you can actually systematically see in high-performing systems and systems that do this well is that teachers are not users of AI tools. They're actually very actively involved in research and design of those kinds of tools. So they have a role in that. You know, Rebecca cited Estonia. You know, they work with OpenAI. That's their kind of counterpart. But they have actually teachers who tell OpenAI what kind of environments they actually want to see in Estonian classrooms. So I think there's another really, really important takeaway. Then, you know, questions around digital safety are, you know, very, very much on everybody's mind by now. And again, you know, there is a race between what you can do through consumer protection and what you can do through enabling and kind of strategies. And I think those are really major findings. So overall, you know, I must say the Digital Outlook has put up as many questions as, or as many doubts on the impact of AI technologies than showing really good examples of how to use them. But, you know, it just brings out the different kind of scenarios. What is so interesting for us here at the OECD is that you see really sharp differences in countries' approaches, which really shows us we're not very good at knowledge diffusion on what really good uses of AI are. Now, you have some systems that are actually very capable in doing this and others that, that really, really struggle or just, you know, focus on the easy consumptive path.

So we have actually very similar conclusions. Our, our report, and a shout out to Andreas's colleague because OECD was on our task force, was, was taking a little bit of a different lens. We were looking at students, students' lives in and out of school, their development in and out of school, not just within the school system. And we were asking, are we on the right track with generative AI tools and primary and secondary school students, K-12 students, and their learning and development? And our ultimate conclusion was no, we are not on the right track. We were headed in the wrong direction. direction because the potential risks, the current and potential future risks, are outweighing currently the benefits. And it's not because there are no benefits, like Andreas said. And, and if I had to boil it down, I would say we did find there's a lot of potential benefits from narrow AI use. The Center for Humane Technology talks about narrow versus wide AI use. So you do not need the entire power of the World Wide Web for most of the good uses in education. Like teachers do not need, you know, the whole, you know, everything that's digitized in the world to level reading materials to the reading levels of the kids in their classrooms, for example. So when AI is used very narrowly, it can have You know, a lot of potential benefits, both in terms of teachers' ability to do creative things and to meet kids where they're at. Neurodivergent kids and supporting assistive technology to be more personalized with AI infused is a real area of potential benefit that I think merits a lot of investigation. When it comes to direct student-facing technology, we— you have to be really careful. And we do see, again, some very narrow supports, Like there's some really good studies that show that a Gemini math tutor, it can be incredibly effective, highly rigorous study, helping tutor kids for short periods of time. This isn't— they're going to learn all their math from an LLM. It's a very narrowly designed math tutor with just good content knowledge in it. So those are good examples of narrow AI use. The problem was, and the reason we said that the risks overshadow the benefits, is that's not how most kids are interfacing with GenAI. GenAI, you know, colleagues say it's crashed the party. It has showed up. It's maybe the only edtech that schools didn't have to intentionally procure. It just showed up on every single digital device because it's software. It showed up on kids' phones and social media. And, you know, in the US, 85% of students say they use AI to help do their work. And some of the things that we're really quite worried about are really from wide AI use. This idea of kids having direct interface and long-form discussions with general-purpose large language models, chatbots, or AI companions that are tell-not-teach machines and possibly emotional manipulation machines. We're really worried about something I call— I'm now calling cognitive stunting. There's a very broad conversation around cognitive offloading, and there's many other terms used, but the general idea is, you know, AI, like a calculator or other— or, you know, Google Maps, offloads a key dimension of human sort of effort, cognitive effort. And it, you know, in theory, allows us to free up our our time for other things. I think that's true for a lot of technology. For general purpose chatbots and children, I think kids aren't even developing the skills in the first place to offload. It's more akin to sort of physical stunting for children, which is a problem where you don't get enough nutrients in order to grow physically. And we don't know how many sort of moments of effortful, deep learning do young people need to really wire their brain to be able to think independently and critically. And so that, along with the real risks around AI friends and companions that are very sycophantic, agreeing with whatever kids say or feel. And, you know, it's very difficult to see both for kids' mental health and their development how that will be helpful. But also, if young people are going to be socialized to always be agreed with, how are they going to take feedback from a teacher in the classroom? How are they going to have a give and take in a group project at school? So we have to worry about the in and outside of school use in terms of the implications for student learning and development. And that is sort of really why we say the risks overshadow the benefits, because they're really of a different nature than the benefits.

Makes total sense. I want to jump into questions in a moment. One of the things I'm consistently hearing though, also in both of your comments, is Like the false dichotomies, right? So it started, Rebecca, before AI, content versus skills. These are not false dichotomies, they're actually important to each other, right? And then it bleeds into sort of Andreas, your point about building mental models and representations so that you can use these tools effectively. And before we even get to— I prefer the term cognitive automation, but it's like the same idea, right, that you were talking about, Rebecca. Yeah. of like, I don't even build these knowledge and skills so that I can do the things. Because as we said in the very beginning, the person that does the work does the learning ultimately. And that's, I think, the fear, right? But also we see the opportunities in narrow uses. Last question before we go to the wider Q&A, which is, there are a lot of headlines that we see out of, you know, China's gone all AI on education, etc. Europe different stance, US different stance. Can you level set us on like how at that level China versus the US versus Europe is approaching AI in education? Are there big countrywide differences or is there more variability inside the countries? How are you thinking about that?

Yeah, there is variability between countries. I mean, some of these are really clear policy choices. If you look at system like China, you know, their slogan is, you know, make AI present but not visible. So basically it's not student-facing, it's not very much in interaction between students. It's really about learning analytics. They have very, very powerful tools. You know, they have AI-based classroom observation that they offer to teachers. You know, you as a teacher can get actually pretty good readout of the quality of your interactions with the students from which you can learn and become a better teacher. So that's the kind of approach. It's basically a tool to super empower learners, super empower teachers, but it's not very much in interaction directly with the students as a kind of consumptive tool. That's, I think, their approach. I think Korea is actually quite similar with their digital textbooks. They, the Koreans, have gone, you know, one step further in basically, you know, giving students a lot of, you know, ownership over the learning out of the classroom at home. This kind of flipped classroom idea, the Korean digital textbook. You know, you get your digital homework, you can do it on your mobile phone, it comes back into the system. That's something that I haven't seen in China nor in Europe. In Europe, I don't think you have much strategy. You know, you have basically, I mean, individual procurement decisions by schools matter a lot more than whatever is done through public policy. So I think apart from regulatory interventions, again, ban on AI or ban on smartphones or ban on social media, those are kind of system-level interventions. But when you actually go into school within those constraints, schools are pretty much on their own. The Estonian case is an exception where you have a government that is working very closely with the tech sector to design something. that works. I think a big difference is also educator capacity. To what extent are educators seeing themselves as delivering instruction or, you know, designing learning environments? And that makes a huge difference in this world, whether, you know, teachers are actively, you know, embracing this, using this, or simply just deploying tools on this. So yeah, you do see big differences across countries, but you see also a huge variety. I think in the Western world, most variation is between, you know, classrooms, between schools, whereas system policies are far behind the evolution of technology. And I think that's actually one of the, you know, challenges that I do see, that actually we are far behind the curve of what's technologically possible. And many teachers would say that, you know, when you ask teachers to what extent they feel you know, they are ahead of this, they're well prepared, well supported. On our teacher survey, a large— on the one hand, it was actually positive that teachers often say, you know, I think this is a positive development, I want to use it. At the same time, teachers also feel that, you know, this is overwhelming and they are not really feel that well prepared for that.

You unintentionally or intentionally answered a couple of questions also in your answer there that we've gotten from the audience, including the impact on teacher teacher education and how they get to know their students better. Let's jump into it. One of the questions— we got a few actually about this question of AI literacy and how we impart that to students methodically. Ryan Finley is one out of Colorado who asked that. I'm curious how you both think of this. There's a little bit of the sort of keyboarding, if you will, of 20, 30 years ago. And, but, but I'd love to hear your thoughts on how do we think about AI literacy in a, in a responsible way? Rebecca, you want to jump in first and then—

Sure, I'll sort of frame broadly and pass it to Andreas, who I know has a lot to say on this. So my perspective is that if we're going to make the same mistakes we did in the past where we did a lot of tool teaching, like, this is this tool, this is how to use this tool, that is not the way to go. A, because the tools are moving incredibly quickly. So if you're going to teach, you know, a 7th grader or an 8th grader now, like, this is how to, you know, prompt ChatGPT, these are the different functions of this tool, by the time they get to the workforce, It is going to look radically different. Instead, what we need to do is really think about how kids understand what AI is. And we— I know, at least again in the US, from spending lots of time talking to schools and parents and teachers and districts, that kids are hungry for this conversation. I've had talked to 5th grade classrooms. These are 10-year-olds. And I said, you know, are you excited about anything about AI? Are you worried about anything? And they just come out. They're like, what job am I going to have in the future? Like, what are the— like, what about artists? How are artists going to work? What, you know, they have concerns. These are 10-year-olds. So we need to begin having conversations. You can— AI literacy, I think, should start early and it does not need screens. What is AI? How does it work? What do we think about it? What are the ethical implications? And then I think progressively as young people get older, and I want to pass to Andreas because I— a huge shout out to the AI Lit Framework that the OECD, EU, and Code.org helped with. I think that's the right sort of framework to go through, so which really builds young people's sort of critical thinking competencies around using a new technology. But but doesn't stop at just consumption, really focuses on creation and how they can use it to solve the problems they care about. So anyways, Andreas, I'll hand it over to you rather than talking about your framework, which I think is one of the better ones out there.

Yeah, you know, I mean, actually my thoughts are very similar to what you just expressed, Rebecca. I mean, for a start, you know, AI literacy in our framework is actually not really about technology. It's very much about human skills. You know, can you think for yourself? Can you live with people who are different from you? Can you make ethical decisions? Can you navigate ambiguity, manage complexity? I mean, those kind of human skills is what we regard at the center of AI literacy. And then of course comes the knowledge of AI. If you do not understand the functioning of AI, then very likely you're going to be just the kind of consumer of those kinds of things. But really at the center of AI literacy is our capacity to sort of manage ourselves, manage others, or work with others and have our thinking process. In fact, I think the interesting part of AI is that it just pushes us to think much harder about what makes us human and to figure out those kinds of skills and support them. And I think where we do succeed, we will have students who become autonomous learners and who will use AI to augment their capacities. Where you do not do that, you'll end up with students who just replace their own capacities. Because I mean, the one thing that is really troubling in this, I mean, maybe the one design flaw that we have as humans is that our brain has been made to save energy. So we will always trade autonomy for convenience. And I think that if we do not have the human skills to sort of, you know, support ourselves in ways. That is, I think, something that, you know, I think it's at the heart of AI literacy, at least in the way we have framed it.

Yeah, and that's not a fear-based way of thinking about it, right? That's a, how do we take control? It's a tool of ours, not the master of us stance. Elizabeth from the Village School in Arlington, Virginia asks, AI exposes a gap that we've never taught students to know themselves, perhaps, and would love your thoughts against that framework, both of you.

Well, I think it's a very profound insight. I think in many ways what we're seeing is a transformational technology layered over a system that needs transforming. And AI has exposed what for many, many decades many of us in the education community have said are the weaknesses of the, of sort of the traditional schooling model that is very task completion oriented and sort of knowledge acquisition, but knowledge acquisition repetition, and then you can forget it and it doesn't, doesn't matter later on because you don't have to use the knowledge for anything. And it has exposed this idea that what we need education to do is not just knowledge acquisition. I do really worry that we're going to throw the baby out with the bathwater and not have kids learn content, which is so highly correlated with critical thinking and creative thinking. You need to think critically about something, but we need to make sure young people are using that knowledge to try to solve problems in their school, in their neighborhoods, in their life, use it in some way. And a piece of that is young people trying to figure out what problems they want to solve. Like what matters to them? What is their voice? And so I think this question gets to that and we need to do that a lot more. In many ways, we're going to need to raise kids and explorers who ask really good questions, not just achievers who know how to find the right answer. Andreas.

No, no, very much, much the point. You know, I think In the industrial age, our education systems divorced learning from assessment. And I think that is still very much what haunts us now. You know, if you go back far enough in history, we all learned through apprenticeship. We always learned from and with people through interaction. Learning and assessment were always two sides of the same coin. When we separated it, we created exactly what Rebecca describes. You know, we ask students to accumulate knowledge over 10, 12 years, and then one day we call them back and say, tell me everything that you've learned in a very contrived, constrained setting, and we call that an exam. And that has created what makes us now redundant in this world of AI. Now, that is sort of something that, in a way, I think it's a positive thing that AI pushes us to rethink that model, to bring us back to what is human agency, what is learning, and how can we foster those kinds of human capabilities. And I think if we do, AI is going to be an amazing kind of tool that will superempower almost anything that we can do. If we don't, I think again, our current systems have come to their limits. And in a way, AI does the wonderful job of exposing those limits.

I think that's exactly right. Rebecca probably knew at some point I would say this, that to me, the question is rarely the tech. It's like, do we have the right model of learning itself that prioritizes mastery and coherence and application and frankly, reflection opportunities to build agency and metacognition and a sense of what your own purpose will be as we develop. And we aren't, you know, most systems are not like that. So layered on top of it, it sort of exacerbates the current tendencies and resources and processes of the existing system. It ties directly, I think, into a question we got in a variety of forms norms around cheating in the age of AI. You certainly both mentioned it. I would love both of your takes. I'll give my quick plug, which I actually think cheating— I'll provocatively say I think cheating is the solution, not the problem at the moment. And it reveals that the challenges with the zero-sum education system, as opposed to one in which your teachers are actually trying to help you master these crafts, much more like the apprenticeship system, Andreas, that you said, where— It's in my interest to make sure every one of my apprentices, if you will, is in fact reaching master status, not to judge them. And therefore, it's in their interest to actually do the work of learning and not skirt it, because in that sense, they're cheating themselves. Whereas Rebecca, as you've written, in achievement mode right now, that's actually sort of— you get credit for that, if you will, in the current system. So I— but I'd love to hear both your reflections because you probably have something more profound. than that.

No, no, I, I, excuse me, I totally agree. I think I would say 3 things about the idea, which is a big conversation in education publicly around cheating and AI and kids learning. One, I do not think that we should put the burden of responsible AI use on students' shoulders. I think that we should have responsible products. That's what I think. I think we should have technology that is guardrailed so that kids access it in ways that are helpful. It is a big burden to say, here's all this crazy technology. Some of it may hurt you, some of it may cognitively stunt you, some of it may unleash your learning capacity. You, student, are the one who has to sort out how to do it. And by the way, you're a bad student and a bad person if you don't sort it out right. So I actually want to sort of abolish this idea of responsible AI use. So that's number one. Number 2, I think that if we talked— the task force and my co-authors and I talked about pedagogy that is AI aware, AI assisted, and AI resistant. That is sort of the move we need to go to. And when we say AI aware, it means do not, if you're an educator, assign an assignment that can be hacked by AI. AI assisted is figure out where AI can be incredibly helpful. I think feedback, real-time feedback, the calligraphy example is a great example. I hear from teachers all over the place that, you know, real-time feedback is really helping their kids learn. Often it's teacher-led. And AI resistant is we do need at this moment in time, young people to be spending face-to-face, eye-to-eyeball, shoulder-to-shoulder, in-person time with other people, because that friction of human interaction is how you learn, grow, develop social competence that leads to cognitive development. And in a world awash of technology, schools have a role to play to protect that and make it a good space. So that would be my— Second thing, and the third thing I would say is that the dialogue around cheating has transformed teachers into policemen or policewomen, and this is not what teachers want. And we did see in our report a huge problem with the fraying of student-teacher trust and relationships in the US. Half the teachers say they don't trust that their students' work is actually their work, and half of teach— and half of students say they don't trust that their teachers are actually preparing their lessons and grading them and teaching them. And so I think the entire conversation around cheating and banning and catching and AI plagiarism checkers is the wrong direction altogether.

Yeah, you know, I mean, I very much agree. You know, if students write their work with ChatGPT and teachers assess the work with ChatGPT, we haven't gained very much. And I think that So there's sort of several dimensions here. First of all, where learning is largely transactional and not relational, you're going to result in cheating. You know, I think that is basically an outcome. If all the emphasis is on, you know, you produce these results, of course everyone will take the shortest route to those results, and that shortest route is cheating. It's, by the way, not a new invention. If you want to actually learn something about cheating, you can go to the university— to the Museum of Exams in Nanjing, which was the ancient capital of China where they invented exams. And, you know, 1,000 years ago you had the same kind of phenomenon with different technologies. People didn't understand why they were doing this. They were asked to produce all of that, and what did they do? They cheated the system. And I think that's something I think we have to just take away from this. Cheating is a symptom of poor education, is not, you know, the cause of that. And related to this is simply, you know, we just need to find ways to make the learning process in itself interesting. Why do I want to do this? Where you focus on the what, you get compliance. People will produce what you ask them to produce. Where you engage people in the why, you're going to get that ownership in the learning process, you know, that sense of growth, that sense of, you know, even failure is an opportunity for growth. And I think that's really, really important in the learning environments. And where you do that, you do not see cheating. You know, I mean, there are certain learning environments where cheating is not an issue, not because we control for it, but because the students have an inherent interest actually for personal growth in there. And I think once again, you know, AI exposes a weakness in education, you know, that focus on answers rather than that process of iterating, to better questions, I think, and where we, you know, design our learning environments, you know, purposefully, I think we do this— this problem will not recur and occur. So, and then I think AI, you know, I must say, I mean, of course I work a lot, you know, on the assessment front and I am, you know, we're doing a lot of open-book exams, you know, where we actually ask students or give students whatever tools they like to use to find, you know, problems, and then our task is to design assessment tasks that are actually conducive to that. I'm still also in favor of, you know, closed-book exams. I do think there are certain things that you want students to learn and not to use shortcuts, you know, absolutely. But I think this is what we're asked to do in, in this century, to design learning environments where AI, you know, is, is, is a friend and not something that we have to constantly, you know, see as a, you know, as a source of cheating.

Yep. All right. 30 seconds each for both of you as we wrap up here. Some of the questions are asking for a little hope on the horizon. I actually think we've given it because, Andreas, you've repeatedly said it sort of exposes where, you know, the challenges with the existing and the breaking points, and that forces us to ask bigger questions. Maybe assessment specifically on AI opportunities where it actually could help us sort of leapfrog where we've been to look at this more relational deep mastery as opposed to transactional. I memorized it for the thing and then I'm done with it. I'd love to hear both of your quick reflections as we wrap up here. Andreas, why don't you go first and then Rebecca bring us home?

You know, I think the short point is that AI is not a magic power. It's an amazing accelerator and an incredible amplifier. It's going to amplify and accelerate good pedagogical practice in the same way it can amplify, you know, poor ideas and bad practice. And I think that in the end makes the role of educators even more important than it used to be.

Where I'm hopeful is thinking about a direction that sees a lot of small, narrow language models, a lot of small, narrow deployments. Your, your Speak and Spell example is a great example. I think this is the future. Little small models that are, are very tailored, very designed, designed by educators and students together to make learning better and make learning experiences more exciting and more engaging. I can't wait for a moment for the day when things like virtual reality headsets, costs come down and AI is, you know, seamlessly infused and you can, you know, learn chemistry and you can manipulate a, you know, a molecule for 10 minutes in your chemistry class and really bring learning to life in ways that it's never been able to do before. I think that has huge potential and is really exciting for the future of learning for kids.

Narrowly defined, though, once again, clear purpose pedagogical practice. Rebecca, Andreas, huge thanks for shepherding us through this incredibly consequential conversation. And for all of you, incredibly active sets of questions and curiosity, really appreciate it. And it's been a pleasure hosting this webinar with Brookings. Thank you so much.

Thank you.

Thank you.