STEAM education for AI literacy: a systematic literature review

Abstract As K-12 education prepare learners for an AI-driven society, the interdisciplinary STEAM approach is widely presented as a pathway to AI literacy, yet evidence on where, how, and how well it advances distinct literacy areas remains fragmented. Following PRISMA procedures, we systematically reviewed 39 studies (2016–2025) from four databases that addressed AI as content within STEAM contexts. We conducted (1) descriptive mapping of publication year, region, educational level, research design, and instructional modality; and (2) deductive coding with the TIECD framework (Technology, Impact, Ethics, Collaboration, Design), followed by inductive refinement into ten AI Literacy Elements (AILEs). Co-occurrence analyses examined emphases and gaps, and mappings linked AILEs to school subjects and to STEAM clusters. Results show a sharp increase in publications from 2021 onward, with most work in middle and high schools. The corpus is geographically concentrated; methods are mainly mixed or qualitative; and instruction is predominantly technology-enhanced. Across AILEs, technical foundations, specifically Fundamental AI Concepts, Computational Thinking, and Data Literacy dominate, while Ethical Awareness, Creative Imagination, Creating with AI, Managing AI, and Designing AI are comparatively underrepresented. Contributions are led by Technology disciplines (Computer Science and Data Science), with thinner coverage across Science, Mathematics, Engineering and Arts disciplines, and Integrated STEAM. Overall, current STEAM implementations chiefly develop technical literacies while offering limited, uneven opportunities for ethical reasoning, creative futures thinking, collaborative management of AI, and design of AI systems. We propose a revised, evidence-informed AI literacy framework aligning TIECD with ten elements and recommend broadening early-years provision, diversifying disciplinary pathways beyond Technology, and designing tasks that jointly evidence technical fluency, ethics, collaboration with AI, and iterative design.