AI at school and university: the end of independent thinking? Not necessarily

The company OpenAI recently presented the new "Study Mode" for ChatGPT to great international acclaim. By asking specific questions in a Socratic manner and providing step-by-step explanations, the AI aims to support learners in thinking independently instead of just presenting solutions. But is this really a new dimension or is it marketing gimmickry? Two scientists from Kassel have looked into this in a study. It has so far been published as a preprint, so has not yet undergone a peer review process.

As part of an interdisciplinary collaboration between members of the Center for Empirical Teaching and Learning Research (ZELL) and the International Center for Higher Education Research (INCHER) at the University of Kassel, Dr. Igor Asanov and Peer-Benedikt Degen conducted a study to investigate the effectiveness of an AI-based Socratic tutor. This AI tutor was designed, technically implemented and tested by Peer-Benedikt Degen as part of his doctorate. The study by the two Kassel scientists shows that the use of AI can support critical, independent and reflective thinking and not hinder it if the AI tool is specifically pedagogically instructed.

The results are particularly significant in light of a recent MIT study, which showed that students exhibited lower cognitive activity when using an uninstructed AI (in this case Chat-GPT 4o from OpenAI). A finding that underlines the importance of a focus on pedagogical framing in AI-supported learning.

The study by ZELL and INCHER was conducted with 65 student teachers in Germany and compared the students' interaction with a Socratic tutor with their interaction with an untrained AI chatbot. Students who used the Socratic tutor reported a significantly stronger promotion of critical, independent and reflective thinking, suggesting that dialogic AI can stimulate metacognitive engagement, contradicting recent narratives of skill loss due to the use of generative AI.

The preprint of the study has just been published on arXiv and can be accessed at arxiv.org/abs/2508.05116

Contact:

Peer-Benedikt Degen
University of Kassel
Email: Degen[at]uni-kassel[dot]de