BeLearn-Socratic Tutor
How to learn better with LLM Tutors? Our project examines identifies patterns indicative of successful human learning when learners interact with Large Language Model (LLM)-based tutors.
Factsheet
- Schools involved School of Engineering and Computer Science
- Institute(s) Institute for Patient-centered Digital Health (PCDH)
- Research unit(s) PCDH / AI for Health
- Funding organisation BFH
- Duration (planned) 01.01.2026 - 31.12.2026
- Head of project Prof. Dr. Kerstin Denecke
-
Partner
BeLEARN
Universität Bern
PHBern
Situation
Recent advances in generative AI have enabled LLM-based chatbots to enhance personalized learning. One promising application is their use as Socratic tutors that stimulate active learning and critical thinking by posing thought-provoking questions rather than providing direct answers. However, the impact of LLMs on learning depends on how they are used. Active engagement and critical thinking, such as identifying and questioning plausible but incorrect information, can promote deep learning. In contrast, passive use may lead to cognitive offloading: tasks are completed correctly, but often with little understanding and minimal long-term learning. This underscores the importance of equipping learners with the skills needed to use LLMs in ways that genuinely support learning.
Course of action
The aim of this project is to identify features in interactions between students and an LLM-based tutor that are linked to effective learning, while also examining how individual characteristics like prior knowledge, cognitive ability, motivation, and technology acceptance influence interaction quality. The project builds on a learning sequence with an LLM-based tutor developed in a previous BeLearn project. Student–tutor interactions from a 2025 blended learning course will be analyzed to identify features linked to learning success. Methods will include natural language processing, machine learning, language models, and expert evaluations. To investigate the role of individual learner characteristics, students will complete a cognitive potential assessment, and we will assess their prior knowledge, motivation and technology acceptance. Based on each measure, they will be grouped into low, medium, and high levels, and differences in key interaction features will be analyzed across these groups.
Result
Following the implementations of the learning sequence and the training for teachers in various educational contexts within the project, the entire learning sequence will be released as an open-source solution on GitHub, ensuring broad public accessibility. The fully open-source workflow is designed with transferability in mind: educators across disciplines and educational levels will be able to adapt the materials to their specific teaching contexts, promoting efficient and reflective use of LLMs in learning.
