Reflection
The problem of learning in the age of AI is overlooked only by those unconcerned with students’ information processing only working at a superficial level, or not at all. For everyone else, the elephant in the room – a focus on performance rather than mastery – has been stomping about for decades and has now become even more visible with the rise of AI. Yet it bears repeating a fundamental truth: contemporary learning science holds that the most effective learning happens through the learner’s own construction and co-construction of knowledge. In other words, genuine learning is still a slow, deep and argumentative or reasoning-based process – one that moves from practice to theory, where individuals, alone or with others, seek to construct explanations, definitions, formulas, conclusions, assumptions, solutions, insights, syntheses and applications through dialogue.31 This is precisely the process that AI must never do on our behalf, because thinking skills cannot develop in our minds without direct engagement (and deteriorate if we stop exercising them). Accordingly, in a well-designed educational framework, scientifically grounded AI learning tools – whose genuinely effective forms are still being developed at the time of writing – should mainly be used when no expert teacher or peers are present. In short, AI should serve learning only during a certain portion of the total study time, and time spent together with real people should not be wasted on AI.
AI should serve learning only during a certain portion of the total study time, and time spent together with real people should not be wasted on AI.
The human psyche is an integrated whole – and was so long before AI. To understand what AI can improve in education – or make worse – we must first look squarely at education’s own challenges. Perhaps the greatest of these is that many participants in the learning process still do not understand the nature of learning itself, which has wide-ranging consequences. Learners often judge the effectiveness of learning during the process, based on how easy or pleasant it feels, when in fact the opposite tends to be true: effortful learning is more effective.32 Supporting learning requires an understanding of how mental processes develop: the nature of cognition, the functioning and growth of executive functions, motivation, emotions, how these are interpreted and regulated, and much more. All these factors interact, and when they – or other factors such as social and emotional competence – create barriers to learning, AI-based support tools may help overcome them.
As the article notes, AI can act either as an accelerator of development or as a dangerous substitute for the learner’s own thinking – much like a physical education teacher running laps on behalf of their students. Ideally, AI should function as a scientifically grounded, step-by-step support system for learners struggling with specific learning difficulties – much like a teacher would, if they could always be there at the right moment. An AI-based solution should be used only up to the ideal point at which it becomes unnecessary – the moment when the learner has internalised the relevant learning skill. This is when learning strategies, attention control, reflection on learning, understanding of learning processes, or metacognition have become part of the learner’s own thinking – when the student has become an autonomous learner and no longer needs AI as a learning coach.
We do not yet know for how many types of learning difficulties AI could be useful, and optimism may be premature. One key question for the future is whether we can create an AI that acts as an attentive ‘learning coach’, informed by contemporary research and capable of interpreting a learner’s verbal input while considering the many interrelated factors that shape cognition – both universal and individual – and prioritising them appropriately. For example, it may be futile to teach learning strategies to a student at times when they are convinced the subject is simply ‘not for them’, or to offer deep discussion when the learner first needs step-by-step support in sustaining attention and realising that attention itself can be trained.
Developing research-based educational AI requires rigorous validation, testing and refinement, along with attentiveness to potential unintended or harmful side effects. I share the article’s concern that the greatest risk lies in delegating essential cognitive operations and deep information processing to AI due to an insufficient understanding of how learning works. Moreover, there are learners whose difficulties lie elsewhere and who may not even reach the stage of problematic AI use – for them, a specialist AI, which does not yet exist, could possibly become a powerful form of support.
The transformation of the entire learning process. AI in education resembles both the Covid-19 pandemic and the Trojan horse – it cannot be ignored and makes it impossible to continue as before, yet it also holds the potential to bring long-awaited changes to learning processes that have long been waiting to enter the education system. An education system that uses AI wisely could resemble an enhanced version of the flipped classroom, but one enhanced by all the insights of modern learning science. Everything students do at home should aim to deepen their understanding, not to produce something for submission. Home learning should involve preparing for upcoming lessons through reading, exploration and reflection, building prior knowledge that resides in the learner’s mind rather than in something tangible to be submitted, possibly for assessment. This does not mean that students should not be encouraged to represent their learning to themselves in various ways, such as through visualisations, questions, conclusions, idea summaries, notes identifying points of confusion, syntheses or derivations. In this process, AI could serve as a valuable discussion partner, helping the learner reach insights through experimentation; the goal is not for AI to think but for it to help the learner think more effectively and deeply. In school, learning would then take the form of co-construction with real human minds – applying knowledge collaboratively or individually to solve complex problems in new contexts. The aim would be to recognise that what was learned at home is not yet fully understood and that applying knowledge in different contexts exposes gaps in understanding – the point at which genuine learning begins. It is also important to remember that if home learning focuses on the individual construction of knowledge, such as through constructive reading, reflection or curiosity, the workload should not be excessive, as deep learning is both time- and energy-intensive.
Assessment’s influence on learning. Finally, we must ask: why do students so often focus on performance rather than learning? It is encouraging that assessment principles are currently under review. The grading system used in Estonia today originated in the Russian Empire in the first half of the 19th century.33 Its preservation is at odds with growing evidence that numerical grading does not promote deep learning, motivation, self-assessment skills, relationships, psychological well-being or a commitment to lifelong learning.34 In the current system, grades inevitably become ends in themselves rather than means. This is reflected in how AI is used today: when assessment practices signal that mistakes are punishable and grades matter more than understanding, and when assignments lack meaning from the learner’s perspective, it is hardly surprising that students turn to AI to produce flawless results rather than to deepen their knowledge. Students do what their learning environment directs them to do. Though we created this situation ourselves, AI now compels us to reconsider how students perceive what happens in school – the nature and purpose of learning itself.
