LLMath

Artificial Intelligence is rapidly changing how students learn and work within traditional university lecture formats. While AI tools are increasingly available to students, there is still limited understanding of how these systems are used in authentic long-term learning settings and how they affect learning processes and performance over time. It is usually suggested that writing high-quality prompts is enough to maximize the performance and accuracy of these tools, but little is still known about other factors such as specific help-seeking strategies, and how they evolve over time.

In particular, little is known about:

• How do students seek information with the Lytris tutor and regulate its use to make learning progress on a mathematical skill? Specifically, what domain-specific and domain-general learning strategies do they use and how do they evolve over the course of a semester?

• and which features of student-AI interactions can predict long-term learning progress.

  
   

The project follows a longitudinal research design and collects learning-related data across multiple mathematics lectures over the duration of a semester.

Students will use a digital learning platform that provides several forms of AI-supported interaction, including:

• open-ended chat interactions with a tutor-style AI assistant,
• short multiple-choice quizzes for self-assessment and knowledge checks,
• and support while working on mandatory exercise sheets accompanying the lectures.

We will use a combination of theory-driven and data-driven approaches to investigate the conversational data produced by students. We will leverage AI-supported methods such as epistemic network analysis to investigate how students transition between different types of help-seeking strategies (e.g. from executive answer-seeking to more instrumental one), their self-regulation and agency in using the system, as well as their domain-specific activity, following the relevant didactics frameworks. Finally, we will also investigate how different scaffolding strategies exhibited by the AI tutor can impact these interaction patterns, and how the latter can impact learning progress trajectories. 

The collected interaction and learning data will be analysed to better understand how students use AI tools in real learning situations, how their usage patterns evolve over time, and how these patterns relate to engagement and academic success.