Critical Thinking with AI, the EduGPT Approach

In recent months, I have worked with many schools, teachers, and educational organisations on one question that keeps coming up in conversations, training sessions, and pilots. It is not so much whether we should use AI in education, but rather how we ensure that students’ own thinking remains intact.

What you see happening in many places now is that critical thinking is added as a kind of extra step around AI, for example with checklists, signals or visual aids that help students to reflect afterwards on what a model has generated. And while that is certainly valuable, in practice you quickly notice that it often remains separate from the actual learning process, as if it is a correction mechanism that only comes into play after the thinking has already been handed over.

In the programmes we run with CivAI, now across more than 35 pilots and collaborations with schools and educational institutions, we see something different happen when you reverse the design and approach critical thinking not as an add-on, but as a starting point. This changes the interaction itself, challenging students during the conversation with AI to make choices, justify assumptions, and make their reasoning explicit, rather than only checking afterwards what has happened.

This also requires something different from the technology, but perhaps even more from the team developing it, because you don’t just need engineers but especially people from educational practice who understand how learning works, where friction is needed, and where guidance should be provided. That is precisely why we have involved teachers, educational advisers, and practical experts from the very beginning in designing EduGPT, so that it does not become a generic AI tool but a learning environment that truly matches what happens in classrooms and educational programmes.

What helps in this is that we did not come up with this all at once, but developed it step by step in practice, with feedback from lessons, training sessions, and implementations. This allows you to gradually see what works and what does not, and how important it is that an AI system sometimes slows down, asks further questions, or even deliberately does not give a direct answer in order to leave space for the student’s thinking process.

We notice that schools are becoming increasingly sensitive to this, because they see that the issue around AI is not only technical, but above all pedagogical and didactic. This means you also need a mature approach in how you develop and implement this, with a team that understands that complexity and has experience in education itself.

Ultimately, for me, it is less and less about the question of which tool you use, and more and more about the question of which learning process you design, and whether AI is an answer machine in that process or a true thinking partner that helps students to develop and make their own reasoning visible.

Want to pilot for free for 3 months? Get in touch via www.edugpt.nl.