Artificial intelligence in the teaching of mathematics in UNINBE's STEM Courses: students' perceptions and teachers' convictions

Abstract

This article aims to analyze students’ perceptions and teachers’ beliefs regarding the integration of Artificial Intelligence (AI) into Mathematics teaching in STEM programs at UNINBE, taking into account historical evolution, conceptual paradigms, didactical trends and classroom practices. The study assumes that there are significant discrepancies between students and teachers concerning theoretical grounding, historical awareness of AI and the identification of the most relevant didactical trends. The theoretical framework is structured into four dimensions: historical evolution of AI in mathematics education (from early intelligent tutoring systems to large language models), conceptualizations and paradigms (behaviourism, constructivism, connectivism, TPACK, SAMR), current didactical trends (personalized learning, gamification, adaptive assessment, generative AI) and practical uses in classroom and teachers’ work, including robotics, intelligent systems and language models. The review highlights both the potential of AI for adaptive feedback, personalized tutoring and lesson design, and the risks related to algorithmic bias, shallow reasoning, overreliance on systems and ethical issues such as privacy and academic integrity. Methodologically, a mixed methods design is employed through two parallel questionnaires (18 items each) administered to 271 students and 18 mathematics teachers, organized into four dimensions (history, theories/paradigms, didactical trends, practices) and combining five-point Likert items with open-ended questions. Instrument construction and data analysis are grounded in literature on AI in education and mixed methods research, with descriptive statistics for quantitative data and thematic analysis for qualitative answers. Findings show that students are highly aware of generative AI but have limited knowledge about earlier historical milestones, and tend to value behaviourist-adaptive and connectivist paradigms over constructivism and models such as TPACK. They positively assess personalization, gamification, automated assessment and generative AI explanations in terms of motivation and performance, although classroom integration remains sparse, mainly individual and not systematically oriented toward collaborative mathematical inquiry. Teachers report frequent use of digital technologies and AI tools but often with implicit or weak theoretical grounding, low explicit adherence to TPACK and limited translation of personalization rhetoric into systematic lesson planning. They recognize benefits such as reduced administrative workload and improved diagnosis, while simultaneously stressing the lack of adequate institutional training and clear policies for critical and ethical AI integration. The study concludes that there is a gap between discourse and practice, resulting in an “illusion of integration” characterized by fragmented, instrumental uses of AI rather than deep pedagogical innovation. It recommends formal professional development programmes on AI for mathematics teachers, didactical redesign of course units integrating exploration, adaptive practice and critical verification of AI outputs, and the establishment of an institutional AI policy grounded in didactical principles, ethics and equity.