Co-Constructing Mathematical Knowledge for Teaching: The Impact of Generative AI-Scaffolded Instruction on Preservice Elementary Teachers' Number Sense and Pedagogical Reasoning

Öz

The rapid proliferation of Generative Artificial Intelligence (GenAI) necessitates a paradigm shift in teacher education, particularly in foundational domains such as mathematics. This study investigates the effects of a "GenAI-Augmented Lesson Study" approach on Preservice Elementary Teachers' (PETs) Mathematical Knowledge for Teaching (MKT) specifically within the domain of "Number and Operations," as well as their self-efficacy in integrating AI into mathematics instruction. Utilizing an Explanatory Sequential Mixed-Methods Design , the research was conducted with 82 PETs (41 Experimental, 41 Control) at a state university during the 2020-2021 academic year. While the control group engaged in traditional collaborative problem-solving, the experimental group utilized Large Language Models (LLMs) as "cognitive partners" to diagnose student misconceptions, generate multiple representations of number concepts, and design remedial tasks. Quantitative data were collected using the "Mathematical Knowledge for Teaching Numbers (MKT-N) Instrument" and the "AI-Integration Self-Efficacy Scale." Data were analyzed using ANCOVA and MANOVA . Qualitative data, derived from screen-recorded planning sessions and reflective journals, were analyzed using Thematic Analysis . Quantitative findings revealed a statistically significant difference in favor of the experimental group, particularly in the sub-domains of "Specialized Content Knowledge" (SCK) and "Knowledge of Content and Students" (KCS) F(1, 79) = 14.23, p < .001, η_p^2 = .15. Qualitative results indicated that GenAI acted as a "virtual mentor," pushing PETs from procedural to conceptual understanding of numbers, although issues regarding "over-reliance" and "epistemological trust" emerged. The study proposes an updated AI-TPACK framework for elementary mathematics education.