Artificial Intelligence in Physics Education: An Exploratory Systematic Review of Educational Applications, SDG 4 Alignment, and Sustainability Competencies

Abstract

This paper presents an exploratory systematic review of artificial intelligence (AI) in physics education and its alignment with Sustainable Development Goal 4 (SDG 4). Following Preferred Reporting Items for Systematic Reviews and Meta-Analyses 2020 guidelines (Page et al., 2021), a Web of Science search conducted on January 15, 2025, yielded 105 records; after independent dual-reviewer screening (Cohen’s κ = 0.82 on n = 105 title/abstract decisions), 12 empirical studies published between 2013 and 2025 were included and analyzed. Given the small corpus and emergent nature of the field, findings are explicitly characterized as exploratory evidence mapping rather than consolidated empirical conclusions. The corpus spans educational contexts from secondary school to teacher training and engages AI types including intelligent tutoring systems, machine learning, natural language processing, and generative models. Results indicate that AI is used in four pedagogical roles: tutor, evaluator, assistant, and analyst, with documented benefits in conceptual understanding, formative feedback, personalization, and self-regulated learning. Alignment with SDG-4-relevant sustainability competencies (critical thinking, problem solving, collaboration, and digital literacy) was identified through a secondary interpretive analysis by the review authors against Education for Sustainable Development competency frameworks (Rieckmann, 2011; UNESCO, 2017; 2024); this alignment is inferential and design-based rather than directly measured by the primary studies. Ongoing challenges include technical reliability, teacher preparation and acceptance, and ethical concerns regarding transparency, bias, equity, and data protection. The review contributes an initial, disciplinary-specific evidence map at the intersection of AI, physics education, and SDG 4, and identifies six evidence gaps for future empirical research. No protocol was preregistered prior to conducting this review; this absence is acknowledged as a transparency limitation.