RUME 28
2026
Alexandria, Virginia
MATH-AI: An Integrated Conceptual Framework for Student-AI (LLM) Interactions
Page: 846
With the rising popularity of generative artificial intelligence, particularly large language models (LLMs), it is imperative to understand how students are leveraging these rapidly evolving tools in their learning. Although researchers have begun to address issues of academic integrity, acceptance, and adoption, few conceptual frameworks exist for analyzing how students interact with LLMs in the context of mathematics problem solving. This paper introduces the MATH-AI framework, a synthesis of theories on metacognition, student agency, trust in automation, and human processing and regulation to examine how AI mediates math problem solving and potentially reconfigures the learning process. A case study of two students is provided to demonstrate the framework’s application and illustrate distinct patterns of trust, processing, and self-regulation during AI-mediated mathematics problem solving. These cases highlight the potential risks and opportunities associated with LLM use and provide a foundation for future empirical studies and development of subject-specific AI guidelines.