Can Schools Teach AI Without Dependence?

A podcast for educators on Estonia’s national AI-in-schools strategy, exploring why it is training students to use generative AI, what guardrails matter, and what other systems might learn.

Transcript

SPEAKER_00 0:00

If this episode makes you think, please let us know in the comments and support us by subscribing and leaving a review. Thank you. Today we are exploring a political article by Eto Hernandez Morales called Why Estonia is gambling on AI in schools. It looks at Estonia's decision to put artificial intelligence at the centre of upper secondary education through a national partnership with OpenAI, and it asks a question many countries are still trying to avoid. Not how do we keep AI out of schools, but how do we use it in ways that actually strengthen learning rather than weaken it? And this is one of those stories that really matters because it moves the conversation out of theory and into system design. According to the article, Estonia's Education and Research Minister Christina Kallas is taken a very clear position. She argues that if you regulate AI out of school, you risk cognitive decline because students will be using it anyway. Now that is a strong claim, and it is worth sitting with for a moment because it cuts against the instinct many school systems have had over the last couple of years. The instinct has often been prohibition first. Block it, detect it, warn against it, treat it mainly as a cheating problem. But according to the article, Estonia thinks that is the wrong fight. Calla says the challenge is not how to keep AI out, but how to put it into the learning process so that it accelerates and enhances cognitive growth rather than replacing thinking. That distinction is everything, because there are really two very different futures schools could walk into here. One future is chaotic and underground. Students use AI constantly outside the school day. Teachers know it, leaders know it. But everybody pretends otherwise. So the official curriculum says one thing, student reality says another, and the whole system becomes increasingly performative. Lots of policy language, very little honesty, very little redesign. The other future is much harder but much more interesting. Schools openly acknowledge that AI is here, that students will use it, and that the job of education is to help them use it with judgment. Not just efficiently, not just cleverly, with judgment. And I think that is what makes this article so compelling. Estonia is not described here as simply buying a tool. It is trying to redesign the relationship between students, teachers, and AI at system level. According to the piece, around half of Estonia's 20,000 upper secondary students are already using this customized platform, with the rest expected to join over the summer, and vocational schools due to follow in the next academic year. That is not a pilot in one innovative school. That is not a little sandbox in one district. That is national scale movement. Now whether this works brilliantly, partially or badly remains to be seen. We should be careful here. An ambitious policy is not the same as proven impact. But the educational questions Estonia is willing to confront are exactly the ones many other systems are still dodging. The first is this what happens if students are already outsourcing traditional coursework to AI? The article says Callus redesigned her own university assignments after realizing students were doing exactly that, and that should sound familiar to just about every educator listening. Once students can generate passable essays, summaries, explanations, and project drafts in seconds, the old question of how do we stop them quickly becomes less useful than a different one. What are we actually assessing now? This is where the education lens gets really interesting, because Estonia's move is not just about adoption, it is about forcing a redesign of tasks. If AI can do the standard homework better than many students can, then standard homework is no longer giving you clean evidence of learning. So schools have a choice. Double down on surveillance and pretend the old tasks still work, or redesign the work so thinking remains visible. That takes us straight into something I come back to again and again. Outsource the doing, not the thinking. A good AI strategy in education does not mean students never use the tool. It means the tool does not get to do the part that matters most. The machine may help gather, organize, suggest, explain, draft, fine. But the student still has to judge, interpret, connect, challenge, and apply. The student still has to be visible in the work. And according to the article, Callis is actually careful on this point. She compares AI to a calculator but adds an important caveat. Its value depends on how and when it is used. That is a really important nuance because calculators are often used as an easy analogy for AI, but they only help if you look closely at timing and purpose. Give a calculator too early, and it can bypass number sense. Use it well and it can free up cognitive space for more advanced reasoning. The same principle may be true here. In fact, the article says Callis stresses that AI should not be introduced too early in childhood. She argues that students must first build foundational factual knowledge, literacy, numeracy, and social emotional skills before AI can become a productive aid. Her example is simple and powerful. Some things, like knowing that the Second World War began in 1939, just have to be known by heart. Now this is the bit that really got me thinking. Because too many AI conversations in education get trapped in a false binary. Either memorization is old fashioned and we should all move to pure skills, or knowledge is everything, and AI is a threat to proper schooling. But what the article points toward is a more intelligent middle ground. Foundational knowledge still matters a lot. You cannot critically think about nothing. You cannot evaluate AI output if you have no internal structure to compare it against. AI does not remove the need for knowledge. It may actually increase it, because once answers become abundant, judgment becomes the scarce resource. And judgment depends on knowledge. So I do think Estonia is onto something important. If its leaders are saying yes, use AI but not before foundations are secure. That is a much more mature position than either blind enthusiasm or blanket rejection. The article also says this initiative is doubling as a research project. Under the agreement with OpenAI, student diet entered into the education platform remains under Estonian control and cannot be used to train OpenAI's broader models. Researchers will analyze anonymized usage patterns to study how students engage with AI, including what they ask, how long they interact, and whether they use it for deep discussion or superficial fact checking with findings to be published later as part of a broader scientific study. Now that is fascinating, because one of the biggest frustrations in this space is that we have lots of opinions and not nearly enough high quality, context rich evidence. Schools are making decisions fast, companies are shipping even faster, and everyone is desperate for proof. But good evidence takes time. So the idea of embedding research into a national rollout is very sensible. It does not solve every problem, of course. It will still depend on what exactly is measured and how honestly results are interpreted, but at least it acknowledges that if you are going to do this at scale, you should learn from it systematically. And that idea matters for school leaders everywhere, even those nowhere near Estonia. You may not be able to run a national research program, but you can build local learning loops. What are students actually using AI for? Where does it deepen thought? Where does it encourage superficiality? Which tasks are being strengthened? Which are quietly being hollowed out? If schools do not ask those questions carefully, then innovation quickly becomes guesswork. There is also a very practical point in the article about screen time. Callis argues that policymakers are often focused on the wrong metric. According to her, it is not about the amount of screen time. It is about the purpose and the pedagogical strategy behind it. That is such a useful correction. Because education has a habit of debating the visible metric instead of the underlying design. How many minutes on device, how many hours online, how often using AI? Those questions matter a bit, yes, but they can be misleading. Forty minutes of lazy digital worksheet completion is not necessarily better than forty minutes of intense, guided, AI-supported feedback and reflection. Equally, lots of screen time in the name of innovation can be completely pointless if it is just old pedagogy on new hardware. And according to the article, Callis says earlier, digitalization efforts often failed because schools simply moved textbooks and worksheets onto tablets without rethinking teaching methods. Her preferred model is blended, handwriting and note-taking for memory formation, digital tools for testing, feedback, and guided AI assisted learning. Honestly, that sounds much wiser than either digital evangelism or digital panic. Because what it recognises is that different cognitive activities may need different mediums. Handwriting can still matter. Memory still matters. Note taking still matters. Discussion still matters. And digital tools can also matter when they are used with purpose. This is not about replacing one mode with another. It is about designing the right mix. In Dan's language, this is enhancement, not replacement. And that phrase matters here because the danger with a national AI strategy is that people assume the technology itself is the reform. It is not. The real reform is pedagogical. The technology only matters if it helps schools ask and answer better questions about learning. What should students know for themselves? When should they struggle alone? When should they use AI to stretch their thinking? What kinds of output should no longer count as enough? How should assessment shift? How do teachers model critical use rather than passive dependence? These are human questions, not product questions. The article also makes a broader political point. It says other European governments are watching Estonia closely, while many countries move in the opposite direction, by restricting smartphones in schools or debating online age checks for social media, and Estonia, according to the report, remains skeptical of broad technology bans, with Callus arguing such measures tend to provoke resistance rather than compliance. She even links that instinct to her own upbringing under Soviet rule, saying that when everything was banned, much of daily life became about figuring out how to avoid those bans. That is a striking line, and it opens a deeper education leadership issue. When schools ban things students find useful or inevitable or status laden, students often do not become more reflective. They become more secretive. So the question for leaders is not simply whether prohibition is morally satisfying, it is whether it actually produces the habits you want. If a ban drives AI use underground, all you may achieve is less visibility and less guidance. Now I am not saying schools should ban nothing. Of course, boundaries matter. Age matters, context matters. Some tools should absolutely be restricted in some settings, but blanket prohibition cannot be the whole strategy. Not if the tool is already shaping life outside school and increasingly inside it too. And that is why I think this Estonia story is so valuable. Not because it gives us a perfect model to copy, but because it forces us to think at the right level. Not should students touch AI, yes or no? But under what conditions does AI support cognitive growth, and under what conditions does it replace it? That is the real question. For classroom teachers I think the practical takeaway is this. Do not start with the tool, start with the learning. Where might AI help students think better, ask better questions, get more precise feedback, or see more examples? And where might attempt them into shallow completion? Build around that distinction. Purpose over technology. For school leaders the takeaway is slightly different. You do not need a grand national deal with open AI to learn from this story, but you do need a strategy that is more sophisticated than fear. You need clarity on age and stage, you need local guardrails on data and privacy, you need assessment redesign, you need professional development that helps teachers experiment safely. And above all you need to stop pretending that the main issue is catching students. The main issue is shaping the learning conditions in which AI becomes a cognitive support rather than a cognitive crutch. And for all of us, maybe the most important lesson is this. The systems that do best in the AI era may not be the ones that move fastest or ban hardest. They may be the ones willing to do the difficult work of redesign. To protect foundational knowledge while still embracing new capability, to hold human judgment at the center, to let the tool do some of the doing without ever handing over the thinking. Because that is the line we cannot afford to lose. Estonia, according to this article, is taking a gamble. Absolutely. But perhaps the bigger gamble now is pretending schools can avoid this altogether. The real risk may not be that students use AI. It may be that they use it everywhere except in the one place that is meant to teach them how to use powerful tools wisely. That's all for today. Thanks for listening.