I advocate for AI in education loudly. I also believe that advocacy without a critical counterweight curdles into hype, and a new community-authored paper makes that case clear. Built from a UNESCO-hosted symposium and a survey of 185 researchers, educators, and policymakers across 56 countries, the paper sets out to drag Critical Studies of AI and Education from the margins to the centre of the conversation. The authors call it putting a stake in the ground. I’d call it overdue.
The argument that grabbed me is not the catalogue of harms, though the paper has plenty. It’s the reframing of AI literacy itself.
Critical AI Literacy as a Civic Capacity
Most AI literacy work treats literacy as a skill set. Awareness, usage, evaluation, ethics, all testable, all neatly scaled. The contributors here push hard against that. They define critical AI literacy as a capacity to reflect on AI’s social, ethical, and political dimensions, grounded in data literacy and aimed at democratic participation.
The authors then extend the idea into what they call the human dimension of AI. That covers AI’s effects on dignity, agency, and the digital divide, plus its history, its hidden labour, its role in surveillance, and its environmental cost. This is the part I keep returning to in my own writing. Literacy that stops at “how to write a good prompt” leaves students fluent and powerless at the same time.
The Refrigerator Problem
The authors argue that “asking individuals to critically evaluate AI is akin to expecting consumers to understand how their refrigerator works” (p. 18). The point is not that literacy is worthless. It’s that loading the entire burden of AI safety onto individual understanding lets the systemic problems off the hook.
The authors single out the OECD’s plan to fold AI literacy into PISA as a case in point, warning that it risks shrinking the concept to testable skills while the political and democratic dimensions drop out of view. That critique connects directly to the assessment-validity strand I’ve been tracking, where the act of measuring a construct narrows what the construct is allowed to mean.
This is where I’d add a qualification the paper makes well. The authors hold two ideas at once. Individual literacy can’t carry the weight, and yet in an unregulated market, critical literacy becomes a necessary layer of protection. They don’t resolve that tension cheaply, and neither should we.
The Cognitive Risks Are Mounting
The paper gathers the evidence on over-reliance into one place, and the list is sobering: cognitive offloading, metacognitive laziness, the illusion of learning, and a social penalty for visibly using AI. These echo concerns I’ve written about before, particularly the metacognitive laziness work from Fan and colleagues (2025) that this paper cites. The worry isn’t that students use AI. It’s that “personalised” adaptive systems can strip out dialogue and collaboration, turning learning into a solitary efficiency exercise.
This tracks with everything I argue about pedagogy. The tool doesn’t decide whether learning happens. The design around it does. A system optimised for frictionless answers is optimised against the productive struggle that learning requires.
The Part That Should Unsettle Us
Two threads in the second half of the paper deserve more attention.
The first is collective agency. The authors argue the field obsesses over individual agency and misses the deeper question, which is whether schools can remain democratic spaces at all. They point to the “sham agency” of AI career-readiness platforms that funnel students into predetermined paths while dressing it up as choice.
The second is their warning about “eugenics 2.0,” the pairing of AI with behavioural genomics, which the authors say risks new forms of educational stratification under the banner of optimisation. The authors put it without hedging: “education risks being subsumed into technological paradigms that may reflect eugenicist logics under new guises” (p. 21). That sentence is hard to shake.
One respondent voiced the dissent that runs underneath the whole survey. In a quote the authors include from an anonymous participant (a respondent’s words, not the authors’ own), the critique is total: “AI literacy presupposes that we need to develop a literacy for AI, this is something that certainly sells but in practice, it is a burden for the education sector” (p. 16). I don’t fully agree. But I want that voice in the room.
How I Read This in 2026
So where does this leave an AI-in-education advocate like me? Not in retreat. The authors aren’t anti-AI, and neither am I. Their call is to slow down, to ask whether we need what we think we need before we buy it. Education, they remind us, is deliberately slow and human, and it should not be forced to sprint to keep pace with a product cycle.
What I draw from this is a shift in priority. Build critical AI literacy as a civic capacity. Centre student voices, which this paper admits its own survey mostly missed. Demand transparency mechanisms with teeth and keep pedagogy in the driver’s seat.
References
- Fan, Y., Tang, L., Le, H., Shen, K., Tan, S., Zhao, Y., Shen, Y., Li, X., & Gašević, D. (2025). Beware of metacognitive laziness: Effects of generative artificial intelligence on learning motivation, processes, and performance. _British Journal of Educational Technology, 56 (2), 489–530. https://doi.org/10.1111/bjet.13544
- Holmes, W., Mouta, A., Hillman, V., Schiff, D., Laak, K.-J., Atenas, J., Bardone, E., Lochead, K., Gonsales, P., Havemann, L., Seon, J., Go, B., Schreurs, B., Zhgenti, S., Lee, K., Bali, M., Bialik, M., Medina-Gual, L., Knight, S., … Yeo, B. (2025). Critical studies of artificial intelligence and education: Putting a stake in the ground. SSRN. https://doi.org/10.2139/ssrn.5391793