I’ve argued for years that the smartest way to bring AI into education runs between two bad options: the hype that treats it as salvation and the panic that treats it as the end of learning. A new manifesto from Bozkurt and over forty co-authors (2024) plants its flag in exactly that middle ground, and does it with force.
The authors wrote it collectively, using a modified Delphi process to gather and reconcile the views of educators across the world until the themes settled. They are open that the result won’t generalize and that it holds contradictions on purpose, which suits a document built to provoke thought.
Their core claim is one central to my own work: GenAI is not a neutral instrument. As they put it, “no technology, including GenAI, is ideologically and culturally neutral. It reflects certain worldviews and ways of thinking that present both opportunities and challenges” (p. 506).
Generative AI in Education: A Map of Promises and Perils
The heart of the manifesto is two inventories. The authors lay out fifteen potential benefits of GenAI and twenty risks, and they pair every upside with a critical caveat. Time savings, personalized tutoring, support for lifelong learning, accessibility for students with disabilities or language barriers, assessment that rewards process over correct answers, each one arrives with a warning about what could go wrong.
That structure carries a message of its own. Almost no benefit stands clean. The same tool that personalizes learning can narrow it. The same automation that frees a teacher’s time can fill it with new busywork or thin out the human contact that makes teaching work. One recurring worry is over-reliance. The authors caution that when students lean on instant answers, the cognitive strain that builds real understanding fades.
I’ll name one reservation about the format. A list of fifteen goods and twenty bads, each hedged, can start to read as a refusal to commit. The authors know this risk and frame the tensions as deliberate, which is fair for a manifesto. Even so, a reader who wants a sharp position has to assemble it from the caveats.
The Deeper Worries
Where the manifesto sharpens is in the risks that run past the familiar cheating-and-hallucination talk. The authors argue that GenAI trained mostly on Western, English-heavy data narrows whose knowledge counts and flattens cultural difference, a concern that lines up with what Sourati and colleagues (2026) found about these models homogenizing human expression and thought. They also raise recursion, where models trained on AI output slide toward model collapse and a slow decay of knowledge.
If a model carries the worldviews and commercial priorities of the firms that build it, then every classroom that adopts it inherits those values unexamined. That is the same hidden curriculum Warr and Heath (2025) traced in generative AI, the lessons a tool teaches without anyone choosing them. The authors put the epistemic version plainly: “GenAI reuses knowledge rather than creating new knowledge, which is a key function of the academic enterprise” (p. 507). Their image for the training data, a “distorted mirror” of the internet, shows why uncritical use is risky.
A further thread concerns human agency. The manifesto warns that in a tight human-AI symbiosis, GenAI can become the de facto decision-maker while human judgment recedes. Bozkurt’s separate work on AI and human agency (2025) develops this same worry, and it reads as the philosophical spine beneath the whole document.
A Call to Inquiry, Not a Verdict
The conclusion refuses to be a conclusion. The authors describe the manifesto as “the initial steps of an inquiry” and a wake-up call, and they insist that educators stay the active shapers of AI’s role, not its passengers. Their closing line is the one I’d pin to the wall: “this is not a moment for passive acceptance but one for collective, conscious effort and action” (p. 508).
That stance fits mine, with one update the calendar forces. This is a 2024 document, written before agentic systems and AI-saturated tools became ordinary. Some of its framing already feels early, the talk of GenAI as “embryonic,” the assumption that detection might still be retooled. The core argument holds up better than the snapshot around it. Three commitments are not 2024 concerns: treating GenAI as an agent and not a mere tool, refusing both worship and dismissal, and demanding evidence-based decisions. They are sharper now than when the authors wrote them.
The manifesto closes by asking whether we follow the white rabbit into an AI-run world or move through it with eyes open. The authors choose the third path, refusing the binary, and so would I. The work is to keep deciding, together, what we let this technology become.
References
- Bozkurt, A. (2025). The three laws of artificial intelligence: Re-evaluating human-AI agency and interaction in a time of the generative and agentic AI ren[ai]ssance. Open Praxis, 17(3), 421-428. https://doi.org/10.55982/openpraxis.17.3.794
- Bozkurt, A., Xiao, J., Farrow, R., Bai, J. Y. H., Nerantzi, C., Moore, S., Dron, J., Stracke, C. M., Singh, L., Crompton, H., … Asino, T. I. (2024). The manifesto for teaching and learning in a time of generative AI: A critical collective stance to better navigate the future. Open Praxis, 16(4), 487–513. https://doi.org/10.55982/openpraxis.16.4.777
- Sourati, Z., Ziabari, A. S., & Dehghani, M. (2026). The homogenizing effect of large language models on human expression and thought. Trends in Cognitive Sciences. Advance online publication. https://doi.org/10.1016/j.tics.2026.01.003
- Warr, M., & Heath, M. K. (2025). Uncovering the hidden curriculum in generative AI: A reflective technology audit for teacher educators. Journal of Teacher Education, 76(3), 245-261.