A PRISMA-Based Systematic Review of Generative AI in STEM Education: Taxonomy, Challenges, and Future Research Direction

Abstract

Generative Artificial Intelligence (GAI) has emerged as a transformative force in Science, Technology, Engineering, and Mathematics (STEM) education by enabling intelligent tutoring, automated content generation, real-time coding assistance, simulation-based learning, and adaptive assessment. Despite its rapid adoption, a structured understanding of how GAI is systematically applied across diverse STEM domains remains limited. To address this gap, this paper presents a PRISMA-based systematic review of generative AI applications in STEM education. The selected studies are classified into five major categories: concept explanation and intelligent tutoring, simulation and virtual laboratories, programming and coding support, design thinking and creativity, and assessment and feedback automation. A comparative analysis of widely used GAI tools, including ChatGPT, GitHub Copilot, and Bard, is conducted to highlight their educational roles and limitations. The findings indicate that GAI significantly improves learner engagement, conceptual understanding, and coding productivity. However, challenges such as limited personalization depth, weak multimodal integration, insufficient cognitive feedback, hallucination, and ethical reliability remain unresolved. Finally, key research gaps are identified and a future research agenda is outlined for building trustworthy and adaptive GAI-driven STEM learning systems.