The adoption of Generative Artificial Intelligence (GenAI), particularly Large Language Models (LLMs), in educational contexts presents several risks. While humans remain responsible for meaning-making in interactions with these systems, LLMs are not neutral technologies. Trained on vast datasets, LLMs reproduce and amplify imaginaries, narratives, values and biases embedded in their training data or instilled in fine-tuning processes (Bajohr 2024; Buttrick 2024). This inherent non-neutrality could potentially shape human perceptions, beliefs, and actions. Therefore, we argue that integrating LLMs into educational settings requires a critical perspective that goes beyond viewing them as neutral instruments to support teaching and learning processes. Educators and learners need to cultivate a critical awareness that leads them to question whose visions and imaginaries these models reproduce (Warr and Heath 2025). This requires developing competencies beyond effective prompting skills, such as proficiency in discourse analysis. In addition, educators and learners need to become aware of their own needs and visions for the future to strengthen their human agency over technology (Zuber and Gogoll 2024; Holmes and Miao 2023; Miao and Shiohira 2024). This can be facilitated by engaging learners and educators in activities that promote agency and subjectification processes, such as future workshops (Simon et al. 2025; Suoranta and Teräs 2023). To promote the integration of such activities and methods into educational practices and curricula, awareness must be raised of the inherent non-neutrality of LLMs and the narrative and imaginaries they reproduce, which may not align with the values and future visions of school stakeholders.

To illustrate the necessity of these methods, this contribution offers a comparative analysis of imaginaries generated by ChatGPT and visions articulated by teachers in two future workshops conducted in France. The analysis focused on technology- and education-related narratives and values. In particular, we analyzed which domains of educational purpose were emphasized (Biesta, 2015), as well as the presence of techno-solutionist narratives. Our findings reveal a striking divergence between the narratives and imaginaries produced by ChatGPT and the visions of the future articulated by teachers. We conclude by discussing implications for future-oriented educational practice.

References

Bajohr, H. (2024). Whoever controls language models controls politics. In I. Arms (Ed.). Training the archive (pp. 189–195). Walther König.

Biesta, G. (2015). What Is Education For? On Good Education, Teacher Judgement, and Educational Professionalism. European Journal of Education. 50 (1). 75–87. https://doi.org/10.1111/ejed.12109.

Buttrick, N. (2024). Studying Large Language Models as Compression Algorithms for Human Culture. Trends in Cognitive Sciences. 28 (3). 187–89. https://doi.org/10.1016/j.tics.2024.01.001.

Holmes, W., & Fengchun, M. (2023). Guidance for Generative AI in Education and Research. UNESCO Publishing.

Fengchun, M. & Shiohira, K. (2024). AI Competency Framework for Students. UNESCO Publishing.

Simon, S., Vassaux, C., Guigon, G., Čarapina, M., Consoli, T. (2025). The Postdigital Classroom: A Guide Toward the Imaginary. Postdigital science and education. 7(3). 916-938. https://doi.org/10.1007/s42438-025-00565-w.

Suoranta, J., & Teräs, M. (2023). Future workshops as postdigital research method. In Constructing postdigital research: Method and emancipation (pp. 317-331). Cham: Springer Nature Switzerland. https://doi.org/10.1007/978-3-031-35411-3_16.

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. https://doi.org/10.1177/00224871251325073.

Zuber, N., & Gogoll, J. (2024). Vox populi, vox ChatGPT: large language models, education and democracy. Philosophies, 9(1), 13. https://doi.org/10.3390/philosophies9010013.