AI Literacy in STEM Classrooms: What a Decade of Research Reveals About the Frontier

Artificial intelligence has moved from the research laboratory into the everyday lives of schoolchildren and university students, and educators are now scrambling to work out what it actually means to be literate in AI. A new editorial published in the International Journal of STEM Education takes stock of this fast-growing field and delivers a sobering verdict: despite an explosion of studies, the conceptual foundations of AI literacy remain shaky, the evidence of its benefits for STEM learning is inconclusive, and the most important questions are only now being asked. The article, authored by Jonas Hallström of Linköping University together with Thomas K. F. Chiu, Meixia Ding, Milo D. Koretsky, Yeping Li and colleagues, synthesizes the most highly cited research on AI literacy and STEM education published between 2022 and 2026, and maps out where the field must go next.

The notion of AI literacy borrows deliberately from older literacies. Just as digital literacy, media literacy and data literacy expanded the original idea of reading and writing into broader competencies, AI literacy describes the knowledge, skills and attitudes that non-experts need to engage with artificial intelligence in daily life. The most widely cited definition, from Long and Magerko, frames it as a set of competencies that enables individuals to critically evaluate AI technologies, communicate and collaborate effectively with AI, and use AI as a tool online, at home and in the workplace. That framing matters, because it shifts the educational focus away from hands-on technical skills alone and toward an understanding of, and critical reflection on, the role of AI in society. In recent years the research conversation has also migrated from university computer science departments, where AI literacy was first studied, down into K-12 classrooms, a shift driven largely by the sudden availability of generative AI applications such as ChatGPT.

To establish the state of the art, the authors conducted a qualitative evidence synthesis, searching the Scopus database for the most cited international articles on AI literacy in STEM education. After screening titles, abstracts and full texts, they settled on the ten most highly cited papers published between 2022 and 2026, restricted to prestigious, high-impact journals. The approach has limitations the authors acknowledge candidly: highly cited papers attract further citations, newer and more peripheral studies may be obscured, and a small sample cannot represent an entire field. Still, by examining the most visible scholarly output, the review captures what the field currently considers important, and where its blind spots lie. The selected studies span early childhood through higher education, confirming that AI literacy is now seen as relevant at every level of schooling.

Three broad dimensions emerged from the analysis. The first concerns instructional focus. In K-12 settings, researchers have tested AI literacy programs with preschoolers, designed summer workshops for middle schoolers, and piloted AI-integrated science lessons in lower secondary classrooms. One study examined whether upper primary students could develop AI literacy through digital story writing, with children imagining robots that recycle waste, care for injured animals, or help elderly people with household chores. In higher education, researchers developed and validated AI literacy tests, explored ChatGPT as a virtual tutor in university physics, and modeled how undergraduates’ technological readiness, self-efficacy and attitudes predict their use of AI tools. Notably, almost all of the top-cited studies focused on students rather than teachers, a gap the authors flag as urgent.

The second dimension, how AI literacy is characterized, reveals striking conceptual variation. Some frameworks emphasize technical concepts and processes, others ethical and societal implications, and still others future careers. In early childhood contexts, AI literacy has been described as learning about AI as content, using it as a tool in hands-on activities, and being aware of its limitations and incorrect output. One framework developed specifically for technology education divides AI literacy into conceptual, procedural and contextual technological knowledge, treating it as part of a broader technological multiliteracy. Another widely used structure includes four components: knowing and understanding AI, using and applying it, evaluating and creating with it, and considering ethics. Despite the differences, most conceptions now combine technical understanding with critical and ethical awareness, echoing the influential definition that treats critical evaluation as central.

The third dimension, STEM disciplinary content, exposes an uneven landscape. Interdisciplinary projects with environmental and sustainability themes dominate, particularly in early childhood and primary education, where children have learned basic AI inference through tools like Teachable Machine and connected robotics activities to topics such as ocean pollution and recycling. Biology and physics appear most often as disciplinary anchors, mathematics less so. Teachers piloting AI-integrated science lessons saw complementary roles: science provides the context for a problem, such as determining the proper conditions for solar panels on a Mars Rover, while AI contributes a predictive model that helps solve it. Yet technology as a disciplinary subject in its own right was almost entirely absent from the reviewed research, mentioned only in the framework developed for technology education. In higher education, meanwhile, STEM content often receded into the background, with studies focused on AI tools themselves rather than on how AI reshapes disciplinary understanding.

The evidence on outcomes is mixed, and this is where the review becomes most provocative. Some interventions report genuine gains, including improved conceptual and epistemic understanding of AI among upper elementary students. Others are more cautionary. In one study, primary students who used ChatGPT to support STEM programming gained confidence and programming knowledge but showed no improvement in computational-thinking skills or independent thought. A mixed-methods study of ChatGPT as a physics tutor found that students must learn to recognize AI-generated content and develop prompt engineering skills, and identified a persistent misconception that AI is super-human and infallible. The authors of that study concluded that AI literacy education is first and foremost about improving students’ critical thinking in assessing information, a conclusion that resonates throughout the review. Anthropomorphism, the humanlike quality of chatbot communication, may foster unwarranted trust in AI responses, compounding the risk.

Underlying all of this is a deeper tension the authors articulate with unusual clarity: AI enters STEM education in two distinct roles, as a domain of knowledge to be taught and as a tool for learning, and these roles can interact or even conflict. Students should understand how neural networks recognize patterns and how algorithms process data, but they should also use intelligent tutoring systems, automated feedback and generative AI to support inquiry. The danger is that AI used as a crutch may undermine the learning of foundational problem-solving skills, precisely the competencies that deep disciplinary understanding requires. Whether AI literacy can genuinely transform STEM learning, rather than simply accompany it, remains an open question that few studies have directly investigated, though some point to increased authenticity when AI literacy is woven into real-life STEM problems.

The ethical stakes receive pointed attention. The authors call for urgent research on bias, data privacy, cybersecurity, the energy footprint of AI systems, data ownership and rights, and the responsiveness of education systems to rapid technological change. They also note that prominent AI competency frameworks from UNESCO, the European Commission and the OECD often lack connection to empirical research or sufficient theoretical underpinning, and few are specific enough to be usable in classrooms. Cultural and contextual factors add another layer of complexity: given differences in AI governance between, for example, the United States and the European Union, AI literacy is likely conceived differently across national contexts, yet comparative scholarship is scarce. Gender differences in AI literacy development also warrant closer study.

The editorial closes with a call to action that doubles as a research agenda. Future work must move beyond one-size-fits-all frameworks toward context-sensitive approaches that empower educators and learners to engage critically and responsibly with AI while preserving core STEM learning goals. Teachers, the authors argue, need to be AI literate themselves before they can deliver effective instruction, yet research on promoting AI literacy among teachers is strikingly thin. Assessment presents its own frontier: generative AI threatens traditional assessment formats, inviting multimodal and interactive alternatives, while AI-generated feedback demands a new form of teacher expertise the authors call AI feedback literacy, the capacity to judge machine inferences and adapt instruction accordingly. As AI becomes simultaneously a subject of study and a learning partner in STEM classrooms, the field’s central challenge is to ensure that technical competence and critical reflection grow together, and that the promise of AI does not come at the cost of the deep disciplinary understanding that STEM education exists to cultivate.

Subject of Research: AI literacy and its role in STEM education research

Article Title: AI literacy and STEM education: emerging frontiers and research priorities

Article References: AI literacy and STEM education: emerging frontiers and research priorities. (n.d.). https://doi.org/10.1186/s40594-026-00653-8

Image Credits: AI Generated

DOI: 10.1186/s40594-026-00653-8

Keywords: AI literacy, STEM education, generative AI, ChatGPT, critical thinking, K-12 education, higher education, AI ethics, educational assessment, teacher training, technological literacy, research review