This review synthesises emerging evidence (2022–2025) on the role of artificial intelligence (AI) in enhancing simulation and visualisation within STEM education. Traditional instructional approaches often face challenges in representing abstract concepts, ensuring equitable access to laboratory experiences, and fostering higher-order cognitive skills. AI-driven simulations address these gaps by enabling dynamic interaction, personalised feedback, and real-time adaptation of task complexity, grounded in constructivist and socio-constructivist learning theories. Drawing on recent empirical studies, this article examines how generative AI, multi-agent platforms, reinforcement learning controllers, and explainable AI frameworks transform conceptual understanding, engagement, self-regulated learning, and model-based reasoning. Findings highlight consistent gains in conceptual grasp, motivation, and misconception repair, particularly when AI systems are designed with inquiry cycles, cognitive load principles, and teacher oversight. However, persistent challenges remain in equity, scalability, data governance, and ethical safeguards, with gaps in early education contexts and longitudinal research. The paper proposes a structured research agenda that emphasises inclusive design, cross-disciplinary collaboration, and transparency. By integrating theoretical foundations with technological advancements, this synthesis contributes a framework for the effective, equitable, and accountable application of AI-driven simulation and visualisation in STEM teaching and learning.
Deckker et al. (Wed,) studied this question.