Artificial intelligence in university physics education: a systematic review of empirical studies
TL;DR
The current evidence does not support a general conclusion that AI is uniformly effective in university physics education, and its educational value appears to depend on how AI is integrated into learning activities and on the extent to which students and instructors retain responsibility for reasoning, evaluation, and pedagogical decision-making.
Abstract
Artificial intelligence (AI) is increasingly being integrated into higher education, yet empirical evidence on its pedagogical use in university physics remains fragmented. This systematic review synthesized empirical research on the applications, reported learning and pedagogical outcomes, methodological quality, challenges, and educational implications of AI-supported university physics education. We used six search sources: SpringerLink, Scopus, Web of Science, ScienceDirect, Google Scholar, and ERIC. The searches in the five original sources were updated through September 2026, and ERIC was added as an additional search source. Eligibility criteria were defined using the PICOS framework, study selection followed PRISMA 2020, and methodological quality was appraised using the 2018 Mixed Methods Appraisal Tool (MMAT). Eleven empirical studies met the inclusion criteria and were synthesized narratively. Across the included studies, AI was used for tutoring and explanations, formative feedback and scaffolding, collaborative problem solving, simulation and modeling, instructional design, and developing AI literacy. Reported outcomes were mixed and context-dependent. Some studies identified favorable short-term or task-specific outcomes related to feedback, learning performance, simulation-based activities, and AI-related competencies, whereas others reported limitations in transfer to independent performance, the pedagogical quality of AI-generated materials, and students’ ability to recognize scientifically inaccurate AI-generated responses. Cross-study patterns suggested that instructional structure, alignment between AI-supported activities and learning tasks, students’ disciplinary knowledge, and opportunities to critically evaluate AI outputs shaped the educational use of AI. The current evidence does not support a general conclusion that AI is uniformly effective in university physics education. Rather, its educational value appears to depend on how AI is integrated into learning activities and on the extent to which students and instructors retain responsibility for reasoning, evaluation, and pedagogical decision-making. Given the small and heterogeneous evidence base, larger longitudinal and multi-institutional studies are needed to examine the durability and transferability of reported outcomes.