Last month, I was invited to give the closing keynote at UNESCO’s Digital Learning Week in Paris. My task was unusual: rather than arrive with a prepared argument about AI and education, I was asked to listen across the gathering and reflect back some of the tensions, contradictions and contestations I had heard. It was a remarkable vantage point. Across several days, governments, researchers, educators, technology organizations and civil-society actors wrestled with questions that are becoming increasingly difficult to avoid. How do we protect children while recognizing their evolving agency? How do educational institutions pursue technological sovereignty while becoming dependent on infrastructures they do not own or control? How do we demand explanation from systems whose operations may remain structurally opaque? How do we reconcile intense pressures to adopt and scale with the need to understand costs that are ecological, social, political and educational?
None of these tensions can be resolved by simply identifying the correct side and recruiting evidence accordingly1.
AI is arriving in education at precisely the moment when many institutions are under pressure to move faster than collective discernment can reasonably operate. Governments remain accountable for public education while much of the infrastructure shaping its future sits elsewhere. Hyper-individualized learning may support some students while weakening learning as a social process through which we encounter difference, negotiate meaning and become answerable to one another. Cognitive offloading may allow us to think further, or it may allow us to perform learning without undergoing it. These are not arguments for or against AI, but invitations to pay closer attention to the kinds of educational, relational and political arrangements that are being normalized as adoption accelerates.
The people whose futures are most directly implicated are not always those with the greatest authority in these conversations.
In the ministerial meeting, I heard only two students speak. Both raised ecological costs and the right to disengage from AI. I found myself wanting to amplify their intervention, not because students occupy some uniquely innocent position, but because they were asking a question institutional conversations can bypass surprisingly easily: will we retain a meaningful right not to participate? “Choice” becomes rather thin when refusing a technology carries academic, professional or social penalties, and meaningful alternatives matter precisely because infrastructural dependence often presents itself first as convenience. That intervention brought me back to something that happened more than thirty years ago.
In 1993, when the World Wide Web became publicly available, I was beginning my undergraduate degree in education. A group of professors at my university produced a manifesto opposing the internet.
In 1993, when the World Wide Web became publicly available, I was beginning my undergraduate degree in education. A group of professors at my university produced a manifesto opposing the internet. Many of their concerns were perceptive, and some of their warnings now look remarkably prescient. Their proposed response, however, was to prohibit it, ideally everywhere, world-wide, which was, let us say, moderately unrealistic. Students organized a strike. What they were asking for was neither technological enthusiasm nor dismissal of the professors’ concerns. They wanted educators willing to accompany them through something that was already changing their world. They wanted collective inquiry into what this technology might enable, erode and ask of us. Thirty-three years later, I suspect that request remains remarkably current.
The educational challenge before us is to cultivate the collective capacities to discern what AI is making possible, what it is making more difficult, what it is rendering invisible, and what we may need to protect, interrupt or redirect.
This requires learning to encounter profound technological change without surrendering either critique or curiosity, without mistaking complicity for inevitability, and without clinging to the comforting fiction that innocence is still available. It also requires a willingness to turn the questions we direct toward AI back upon education itself. Many of the tendencies we now find troubling in AI, including standardization, optimization, extraction, surveillance and the displacement of human judgment, were already well established in our educational systems long before generative AI entered the classroom2.
AI is not only disrupting education, but also holding up a rather unforgiving mirror to some of its most deeply normalized assumptions and practices.
We rightly ask what biases AI reproduces, but which biases has education already institutionalized as knowledge, intelligence and merit? We ask whether AI will homogenize languages and knowledge, but education has a long history of linguistic erasure and of marginalizing knowledge systems that do not fit dominant categories of evidence and expertise. We worry about AI hallucinations, and rightly so, but what about the civilizational hallucinations that modern education continues to reproduce as reality, including fantasies of limitless growth, human separability from the rest of nature and the possibility of knowledge without consequence? We ask whether students will become dependent on AI, yet education already organizes powerful dependencies around credentials, institutional authority, approval and standardized success. None of this makes dependence on AI irrelevant. It makes the question more demanding: which dependencies are reciprocal and enabling, which transfer agency elsewhere, and when does offloading help learners think further rather than help them avoid thinking?
The question of alignment is similarly revealing. We ask how AI can be aligned with “human values,” as though humanity had ever agreed on a single set of values, or as though the category of the human had historically included everyone equally. But we might also ask what education itself is already aligned with. Is it aligned with competition, productivity and economic growth? With democratic participation and collective wellbeing? With ecological continuity? With the reproduction of existing institutions? The alignment problem is therefore not only technical, but educational, political and ecological. Perhaps AI is doing something more consequential than introducing a new technology into education. Perhaps it is making some of education’s existing contradictions unusually difficult to ignore.
This was the point in the keynote where I shifted, explicitly, from rapporteur to provocateur. The provocation that has stayed with me most concerns scale. AI offers something historically unusual: extreme scalability and extreme risk. A system can interact individually with millions of people, continuously, adaptively and at extraordinary speed. Those interactions do not merely transmit information. Over time, they shape what people attend to, what they desire, what feels normal, how they understand themselves and others, and even what kinds of futures remain imaginable. So I asked the room a question that was deliberately provocative.
If you wanted to shift hearts and minds, at scale and at speed, what would you put in the digital waters?
The point of the metaphor is not that AI is natural or inevitable. Quite the opposite. The point is that repeated, personalized interaction at scale can become environmental in effect. The relevant questions are therefore not only about what AI tells us, but about what patterns of attention, desire, relation and imagination become more probable through repeated exposure. What is already being placed in that water? By whom? According to which assumptions about intelligence, learning, flourishing, success and the human? In whose interests? With what capacity for democratic scrutiny, refusal or reversal? If public institutions do not ask these questions, commercial, ideological and geopolitical actors certainly will.
These questions became particularly concrete during the gathering when a technology company presented its vision for educational AI: systems capable of knowing an individual learner more thoroughly than teachers, more thoroughly than parents, and eventually more thoroughly than the learner understands themselves. The reception was not uniform. While some appeared deeply unsettled by the proposition, others were intrigued by its possibilities. The presenters, for their part, seemed to regard this degree of intimacy as an unambiguous educational advantage. It was a revealing moment, precisely because the tensions were not difficult to understand. As a commercial proposition, the promise is seductive: more personalized support, earlier recognition of learning difficulties, better adaptation to individual needs, less administrative burden for teachers. Yet what appears as an educational breakthrough from one vantage point may look rather different from another. And therein lies the tension: the very capacities that make these systems appealing as educational companions are also those that demand much closer scrutiny.
What does it mean for a commercial system to “know” a child? What forms of inference are being produced from their conversations, errors, vulnerabilities, preferences, emotional patterns and learning histories? Where does that knowledge reside? Who controls it? Who can access it now, and who might access it later? What happens when intimate educational data becomes accessible to corporations or states operating under legal and political arrangements that offer little protection against surveillance, discrimination or political misuse? What happens if educational data gathered under one political regime survives into another, if corporations change ownership, or if databases are breached? What if information generated for personalized learning becomes useful for employment screening, insurance, policing, migration control, political persuasion or other purposes nobody imagined when the student first opened the application? And who gets to decide which risks are acceptable, particularly when those whose futures are most exposed have the least power to refuse?
Do we actually want a technological system to know a young person better than they know themselves?
Underneath those governance questions lies an educational one that I find even more difficult. Do we actually want a technological system to know a young person better than they know themselves? Or is part of education’s responsibility precisely to help a person become increasingly capable of knowing, questioning and authoring themselves in relationship with others and with the world? Prediction and personalization can be enormously useful, but they can also quietly narrow the space in which a person surprises themselves, changes direction, refuses a category or becomes someone not yet legible to the system. A civilization increasingly enchanted with prediction may need to recover a deeper respect for emergence.
One of the deepest educational questions raised by AI is not what machines can do, but what forms of human discernment education now needs to cultivate.
We need the capacity to engage complexity without rushing toward premature certainty, to face complicity without seeking innocence and without becoming immobilized, and to distinguish forms of assistance that extend human capacities from those that quietly reorganize them. We also need forms of redirection rooted in relationship with the Earth, including relationships with generations and beings whose interests rarely enter the room3.
Perhaps the task before us remains close to the one those students articulated in 1993: neither uncritical adoption nor unrealistic prohibition, but intergenerational accompaniment, collective inquiry and the cultivation of discernment. The question is not simply whether education can adapt to AI, but whether this encounter can help education remember and redirect its public purpose in service of the living planet we are part of and depend on for our existence.
This post emerges from the research project Meta-Relationality and AI: Discernment, Fields, and Relational Capacity in Generative AI Systems, based at the University of Victoria, Canada. The university also offers two online courses, open to the wider community, that draw on this research: Facing Human Wrongs: Redirecting Education and AI Towards Earth Alignment, taught by Professor Vanessa Andreotti, with its inaugural cohort beginning in January 2027 (registrations are open); and A Meta-Relational Approach to AI, designed by Professor Andreotti, which will be offered again in March 2027.
Further reading: