Collectives of humans-with-GenAI in university mathematical modelling: different forms of human-AI interaction in the production of mathematical knowledge

Abstract This study applied the humans-with media perspective to examine how collectives of students-with-ChatGPT produced mathematical knowledge during a university-level mathematical modelling task. We conceptualized generative artificial intelligence (GenAI) not as an external tool, but as a participant in a mathematical (modelling) activity. The qualitative study involved 94 engineering students enrolled in a Differential Equations course and focused on three modelling stages: problem framing, mathematical development, and interpretation/validation. Data were generated from modelling reports, oral presentations, questionnaires, and semi-structured group interviews. Findings indicate that interaction with GenAI varied in the organization of agency within each collective. ChatGPT participated in problem framing and interpretive synthesis but not in mathematically rigorous development, with several groups of students restricting or rejecting its use due to limitations in symbolic manipulation and numerical accuracy. Across groups, interaction configurations ranged from mediated participation to delegative incorporation, reflecting different distributions of agency and authorship. The study contributes evidence on how GenAI may reshape modelling practices and highlights emerging challenges related to mathematical knowledge production in humans-with-GenAI collectives.