In a newly published pilot study, Isa Steinmann, Roar Bakken Stovner, Ove Hatlevik, Anne Kristine Øgreid, and Janne Herseth explored whether artificial intelligence (AI) large language models (LLMs) can reliably categorise master’s theses—an often overlooked but important part of teacher education—at scale. Using a coding instrument they developed, they asked:
- Can GPT-4 Turbo perform similarly to teacher educators when classifying thesis types?
- What are the common characteristics of 278 theses from a Norwegian university?
They found that GPT-4 Turbo showed promising potential for assisting in this kind of educational research and that most theses were small-scale, qualitative studies—often based on interviews or classroom observations.
These are promising results that provide emperical support for using LLM to catergorize master’s theses for the TEPS study master theses module.
The authors also reflect on what these results mean both for AI as a research tool and for strengthening research-based teacher education.