‘Studying doesn’t occur when the machine does the pondering for you.’ With generative AI reshaping the tutorial panorama, Instructor columnist and OECD Director of Training and Abilities Andreas Schleicher discusses the significance of age-appropriate guardrails, knowledge safety and human judgement to make sure GenAI turns into a scaffold for studying, not a crutch.
For lecturers, GenAI provides a robust set of helps. It may possibly assist put together classes, personalise curricula, design assignments and exams, draft suggestions, and even help with grading. It may possibly additionally help administrative work similar to documentation and routine reporting.
Within the best-case situation, GenAI buys lecturers time to construct relationships, give richer suggestions, and concentrate on the human judgement and care that no machine can replicate. OECD knowledge present that in some contexts, it might even help much less skilled tutors in actual time, elevating the general high quality of instruction.
College students, too, are already utilizing GenAI as a cognitive companion. Massive language fashions have turn out to be a sort of supercharged search engine. They clarify ideas, reply follow-up questions, brainstorm concepts, polish drafts, debug code and recommend options. When designed particularly for studying, academic GenAI instruments can scaffold understanding very similar to clever tutoring methods, guiding college students step-by-step somewhat than merely delivering completed solutions. On the system stage, GenAI may help develop assessments, develop automated analysis to oral duties, tag studying sources, and supply administrative help throughout establishments.
However right here’s the catch: Studying doesn’t occur when the machine does the pondering for you.
The danger of ability atrophy
Writing, problem-solving, and instructing usually are not simply duties; they’re cognitive exercises. Outsource them to GenAI, and chances are you’ll get a refined product, however you lose the educational that comes from battle.
The OECD’s 2026 Digital Training Outlook exhibits the chance of ability atrophy: when instruments designed to assist us suppose find yourself pondering for us as an alternative. The hazard is particularly acute with general-purpose GenAI, which is constructed to carry out duties effectively, to not help academic progress.
This raises thorny questions. When is it acceptable to make use of GenAI, and for what? Ought to younger youngsters use general-purpose GenAI for homework? Most likely not. However academic AI instruments are a special story. They are often purpose-built, secure and efficient – similar to tutoring methods for arithmetic or studying, which have lengthy been used efficiently in main training and at the moment are turning into extra conversational by way of GenAI interfaces. Nonetheless, to be taught with GenAI, college students want sufficient foundational information to guage its outputs critically. With out that, the instrument can result in hurt.
Elevating not changing human relationships and studying
There’s one other threat that issues simply as a lot: the erosion of human relationships. Training isn’t solely about buying information; it’s about studying work together with others – similar to receiving suggestions, incomes approval and navigating disagreement. If lecturers cease giving private suggestions, or if college students more and more work together with bots as an alternative of friends and adults, training dangers turning into environment friendly however empty.
The lesson is straightforward and pressing. GenAI ought to elevate human studying – not act as a alternative. It ought to strengthen human connection, judgement and understanding – not quietly take their place. If we use GenAI with intention and care, it could possibly turn out to be a instrument that reinforces what makes training profoundly human somewhat than eroding it.
However one factor is evident, we can not simply offload all accountability to the customers, college students and lecturers. We have to equally be certain that security, belief and accountability are positioned on the coronary heart of those methods. After all, it’s onerous to consider digital security in absolute phrases. If we do, the consequence is probably going paralysis, and we are going to find yourself doing little with GenAI in any respect. As an alternative, we have to take a look at GenAI security in relation to human capabilities and be certain that college students have discovered to suppose earlier than they immediate.
Understanding pupil preparedness and capabilities
That concentrate on functionality is why the OECD is at the moment concentrating the efforts of PISA (the Programme for Worldwide Scholar Evaluation) on framing AI literacy. By this, we imply the capability of younger folks to meaningfully have interaction with AI, to create with AI, to query it and to handle it with important pondering, self and social consciousness.
We now have constructed that idea of AI literacy additionally into our subsequent PISA evaluation, in order that we will monitor how pupil preparedness for AI is evolving, and the way that performs out throughout international locations, social background, gender and age. Understanding capabilities provides us a superb sense of the security dangers and the place we’d like training methods to behave.
AI literacy is one facet of the equation – the opposite facet is digital security. And it’s clear we’re enjoying catch-up on this. Generative AI didn’t knock politely on the schoolhouse door. It got here in by way of the wi-fi and the query each system now faces is: Will GenAI drift into school rooms by chance, or will we govern it by design?
Guardrails, knowledge safety and human judgement
Once we construct roads, we don’t give toddlers the identical guidelines as truck drivers. We construct pavements. We decrease velocity limits close to faculties. We add guardrails. Not as a result of youngsters are weak, however as a result of they’re nonetheless rising. GenAI calls for the identical sort of pondering.
In faculties, we don’t give each pupil unrestricted entry to each instrument. Scissors within the early grades are blunt. Chemistry labs are locked till college students are educated. Web entry is filtered by age. Not as a result of college students are untrustworthy, however as a result of highly effective instruments require context, steering and gradual launch. GenAI calls for a similar method.
A 6-year-old and a 16-year-old could faucet the identical display, however they don’t seem to be partaking with the identical actuality. Once we speak about age-appropriate design and safeguards, we aren’t speaking about censorship, we’re speaking about developmental match.
For youthful learners, GenAI should behave much less like an oracle and extra like a guardrail. Which means sturdy filters. No emotional manipulation. No pretending to be a buddy, a confidant or – worse – a alternative for social interplay. Kids are wired to belief authority. If AI speaks with confidence, they’ll imagine it.
As college students get older, the foundations can loosen, however they need to not disappear. Youngsters can deal with uncertainty however need assistance seeing the place it lives. GenAI ought to make any ambiguity clear and say: ‘Right here’s a suggestion, not the reality’, ‘Right here’s a sample, not a verdict’. That’s the way you train important pondering in an age the place machines can sound assured even when they’re improper.
Information safety should even be a central focus. Kids’s knowledge isn’t oil to be extracted; it’s DNA to be protected. There ought to be no emotional profiling. No long-term reminiscence. No recycling of classroom interactions into industrial coaching datasets. If we’d not permit a stranger to comply with a toddler house and take notes, we must always not permit an algorithm to take action silently.
And above all, we should be clear about authority. GenAI ought to help skilled judgement, not exchange it. The second we permit machines to make high-stakes academic choices on their very own, we don’t simply automate, we abdicate. Robust guardrails early. Rising autonomy over time. Human judgement at all times in cost.
If we get this proper, GenAI turns into a scaffold, not a crutch. A instrument for pondering, not an alternative choice to it. And training stays what it has at all times been at its greatest – a human endeavour, powered by judgement, relationships and belief. That’s the future price designing for.
References
OECD. (2026). OECD Digital Training Outlook 2026: Exploring Efficient Makes use of of Generative AI in Training. OECD Publishing. https://doi.org/10.1787/062a7394-en
