There is growing interest in how artificial intelligence can work alongside humans, rather than simply replacing them. This is particularly relevant for education. We often hear about “human-AI partnership” “, augmentation”, or “hybrid intelligence”. These terms refer to the fact that, while early visions of AI tutors often imagined them as “teachers in a box” (potentially supplanting the teacher’s role), current research increasingly recognises the importance of teachers as expert collaborators within AI-powered learning environments.
Rather than viewing AI tutors as substitutes for teachers, the focus is shifting towards systems that actively involve teachers in the decision-making process. The aim is to leverage teachers’ expertise to guide, customise, and enhance the educational experience that AI systems provide. This approach acknowledges that teachers bring irreplaceable pedagogical insight, contextual understanding, and adaptability to the classroom.
Nevertheless, there is one fundamental challenge when discussing teachers’ involvement in the use of AI Tutors. By definition, AI Tutors are autonomous systems, which the user, i.e. the student, is meant to use autonomously. Therefore, the involvement of the teacher in the AI tutor’s functioning can occur before or after the AI tutor is used, but certainly not while it is being used.
One possible involvement is to adjust feedback and instructional content to suit the classroom context. Intuitively, teachers play a crucial role in crafting meaningful feedback and providing scaffolding and support for learning. In the work of Arends et al. (2017), Teachers play a crucial role in validating and adapting AI-generated feedback, ensuring that messages are contextually appropriate and directing students towards relevant resources.
Another possible involvement is to monitor student progress and intervene when necessary. In the field of AI in education, this involvement first began with Open Learner Models, which enabled interfaces which make the machine’s representation more understandable for learning. This could be used either by the teacher or the learner themselves. With the surge of Learning Analytics, the Teacher Dashboard has become the virtual place where relevant information about students is aggregated and visualised for teachers to monitor. Examples of teacher dashboards working with AI tutors have been developed by Xhakaj, Aleven & McLaren (2017), demonstrating how a teacher dashboard can affect teacher knowledge, decision-making and actions in the classroom.
A step beyond that is the concept of the Learning Analytics Cockpit by Karademir et al. (2024), which, in addition to reporting data about students, also provides the possibility for intervention. For example, to complement the AI Tutors with additional (human-generated) feedback or to refine the automated feedback to fit specific learning contexts or needs.
Whether through direct creation, validation, or refinement of feedback, or by informing system features with their professional expertise, teachers gain centrality in ensuring that AI-driven feedback is both effective and contextually appropriate for learners. But to what extent does this apply to Large Language Models and Generative AI? So far, we do not have clear insights. Being current LLMs based on text interaction and relying on large-scale representations, it seems impractical for teachers to oversee the internal representations of learners in the GenAI model, let alone to monitor conversations with the chatbot. However, the fields of GenAI and Explainable AI are moving rapidly; therefore, we can expect a breakthrough to occur soon.
References:
Arends, H., Keuning, H., Heeren, B., & Jeuring, J. (2017, November). An intelligent tutor to learn the evaluation of microcontroller I/O programming expressions. In Proceedings of the 17th Koli calling international conference on computing education research (pp. 2-9).
Xhakaj, F., Aleven, V., & McLaren, B. M. (2017, September). Effects of a teacher dashboard for an intelligent tutoring system on teacher knowledge, lesson planning, lessons and student learning. In European conference on technology enhanced learning (pp. 315-329). Cham: Springer International Publishing.
Karademir, O., Di Mitri, D., Schneider, J., Jivet, I., Allmang, J., Gombert, S., … & Drachsler, H. (2024). I don’t have time! But keep me in the loop: Co‐designing requirements for a learning analytics cockpit with teachers. Journal of Computer Assisted Learning, 40(6), 2681-2699.
Image credit: Viewsonic
