As generative AI (GenAI) tools become more accessible, K–12 mathematics teachers are increasingly using them for instructional tasks. Survey instruments in the United States describe usage quantitatively including the purposes for which mathematics teachers use GenAI. Yet research into why teachers deploy GenAI as they do is sparse. This qualitative study funded by the U.S. Department of Education investigates how 15 STEM teachers in the Pacific Northwest of the US used GenAI for instructional tasks (lesson planning, teaching, assessment, professional learning, and administration). We conducted in-depth 90-minute interviews with 15 STEM teachers, 11 of them teaching mathematics, and 1-2 people at their schools to understand not only their GenAI use cases but the social and institutional context for their use. We analyzed 76 use cases, drawing on an adaptation of the SAMR model to code them as substitutive, amplified, or transformative based on whether they enabled new learning opportunities for students and whether other technologies or methods could reasonably be employed for the same purposes.

We found that teachers' GenAI use varied by three contextual factors: the institutional context, the social context, and teachers' pedagogical beliefs. On one end of a spectrum, teachers who did not use GenAI, some of them actively avoiding it, were in institutional contexts without clear policy and messaging about GenAI and technological innovation. They had limited resources and structures for collaborative professional learning. In these environments, the most advanced users deployed GenAI primarily as a substitute for other technologies in ways that did not alter students' mathematical learning opportunities. On the other end of the spectrum, teachers in schools with clear GenAI policies, endorsed tools, and resources and structures that supported collaborative professional learning and innovation had more transformative use. These teachers deployed GenAI in ways that provided new and expanded mathematical learning opportunities. We also found that teachers' GenAI use paralleled their pedagogical orientations. Teachers with a more procedural understanding of mathematics had a procedural understanding of GenAI that limited their experimental learning. Teachers who approached mathematical pedagogy as a social practice of exploration of problems and ideas experimented more with GenAI to learn how to use it for instructional tasks.

These findings have important implications for both mathematics education research and practice. Prior research on laptops in schools argue for an ecological understanding of teachers' use of technology for instructional tasks. GenAI, however, represents a new type of technology because it is generative. It is more like an instrument than a tool. A tool is designed for a specific purpose, but people partner with instruments to co-create outcomes. Each responds to the other in ways beyond a simple user-tool dynamic. This new technology has profound implications for mathematics teaching and learning. Our research helps to theorize the ecological considerations required to understand not just how mathematics teachers use GenAI but why they use it as they do. This has profound implications for mathematics teaching and the professional learning and organizational leadership that must be designed to support teachers' deploying GenAI to truly transform mathematics learning opportunities for students.

Keywords: Artificial intelligence, mathematics teaching.