ARTIFICIAL INTELLIGENCE TECHNOLOGIES IN STEM EDUCATION: A SYSTEMATIC LITERATURE REVIEW OF APPLICATIONS, OUTCOMES, AND IMPLEMENTATION CHALLENGES (2019-2026)

Abstract

Despite the rapid expansion of AI deployments, the literature remains fragmented across disciplines, educational levels, and AI technology types, hindering the synthesis of cumulative evidence. This systematic literature review (SLR) synthesizes empirical evidence from 54 included studies published between 2019 and 2026 to map AI technology types, educational purposes, learning outcomes, and implementation challenges across K-12 and higher education STEM contexts. Adhering to the PRISMA 2020 guidelines and the PICOC framework), a systematic search was conducted across seven databases. Out of 1,267 identified records, 54 studies met the inclusion criteria after a two-stage screening process (Cohen's kappa = 0.82). Bibliometric analysis, in conjunction with thematic synthesis (Thomas & Harden, 2008), was employed to elucidate publication trends, outlet distribution, and conceptual clusters. Four distinct AI application categories were identified: (1) adaptive learning and personalized tutoring (38%); (2) intelligent assessment and management (26%); (3) profiling and prediction (20%); and (4) emerging products, including educational robotics and generative AI (16%). Experimental methods were the predominant approach (42.6%), while higher education received the most attention (46.3%). Notably, positive effects on STEM achievement were consistently moderated by teacher readiness and pedagogical alignment. Artificial intelligence (AI) can significantly enhance STEM learning outcomes when integrated into coherent pedagogical frameworks. However, technology-driven adoption without adequate teacher preparation, theoretical foundation, and ethical governance consistently falls short of expectations. Key research gaps include longitudinal studies, qualitative inquiry, and equity-disaggregated analyses conducted in underrepresented contexts.