Generative AI in STEAM education can contribute explanations, design ideas, code, and media to classroom projects. Whether those contributions support learning depends on how teachers organise the activity and what students remain responsible for doing.
Dahal, Hasan, and Islam’s (2026) review brings together research and guidance on these decisions. Its practical contribution is a way to connect AI applications with pedagogy, ethical responsibilities, and disciplinary purposes; it does not establish the effectiveness of every proposed classroom strategy.
I read this paper with a planning question in mind: what should students understand and be able to do by the end of the project? That question helps determine which assistance is useful. A tool that produces an impressive artefact may still bypass the reasoning the activity was intended to develop.
Generative AI in STEAM education needs disciplinary purpose
STEAM connects science, technology, engineering, the arts, and mathematics through inquiry, design, and problems that draw on more than one discipline. The authors caution that these connections need substance. Adding a generated image to a science task does not, by itself, create meaningful interdisciplinary learning.
Their discussion of the arts is particularly useful. Arts integration can involve interpretation, narrative, aesthetic judgement, and sensitivity to human experience. A visual component needs a purpose within the project. Students should be able to explain how its design communicates an idea or responds to the people for whom it is intended.
The review also distinguishes generative systems from other forms of educational AI. A system that generates explanations raises questions about reliability and authorship. An adaptive tutoring system raises questions about the sequence of tasks and the support learners receive. These differences matter when choosing resources for a particular outcome.
This is the kind of contextual judgement that belongs within AI and teaching practice. Teachers need to understand both the subject and the learning process they are asking a tool to support.
Summary of the Study
Niroj Dahal, Md. Kamrul Hasan, and Md. Tarikul Islam (2026) conducted a structured narrative review of literature on AI, generative AI, and STEAM education. They searched scholarly databases and UNESCO and OECD repositories, supplemented by citation searching.
Fifteen sources formed the reviewed corpus, while additional theoretical and contextual literature supported the discussion. The review combined different source types, including research syntheses, conceptual work, and policy guidance, rather than calculating a pooled estimate of learning effects.
The authors organised their discussion around pedagogical foundations, categories of tools, classroom integration, assessment, ethics, and institutional support. They identified potential applications in lesson planning, inquiry, project work, feedback, coding, visualisation, and interdisciplinary learning.
Their proposed framework links these applications to pedagogical principles and ethical considerations, with teachers responsible for designing learning and students responsible for judgement, reflection, and checking.
The breadth makes the article useful for planning and professional discussion, but it also limits what we can infer. This is a synthesis and a proposed framework, not a classroom trial of the recommended activities.
Evidence about established tutoring systems cannot automatically establish the benefits of generative tools. Nor does a plausible teaching application demonstrate improved understanding. I would use the review to identify decisions and possible approaches, then evaluate those approaches against the needs of a specific class.
Practical insights: decide what students must investigate and explain
The authors recommend connecting AI use to inquiry, testing, reflection, and human decision-making. A practical implication is to define the intellectual work before introducing assistance.
If the outcome involves evaluating a design, students need evidence and criteria for that evaluation. If it involves understanding code, they need opportunities to explain its logic and investigate its behaviour.
My suggested adaptation would be an engineering design activity in which students first identify the problem and its constraints. They could then examine an AI-proposed design, test its assumptions, and justify changes.
A brief record of a failed test or rejected suggestion could help the teacher see their reasoning. This example applies the review’s principles; it is not an intervention tested by the authors.
The same logic can guide coding activities. Depending on the outcome, generated code might provide material for inspection and debugging. Students could explain why a section behaves as it does, test a case where it fails, and account for their correction. The relevant evidence would include their explanation and testing decisions alongside the working program.
Teacher preparation also needs this connection between content, pedagogy, and technology. The paper draws on TPACK, and the issue connects with professional knowledge for integrating AI in education. Knowing how to operate a tool is part of preparation. Teachers also need to judge whether its contribution fits the concept, the learners, and the evidence of learning they require.
Equity and ethics should influence those choices from the start. The review discusses access, privacy, bias, and contextual responsiveness. Teachers can consider whether a project requires resources students lack, what information enters the system, and how generated representations might overlook local knowledge. A successful activity should give students a workable route to the intended learning and leave them able to explain the decisions behind what they produce.
Reference
Dahal, N., Hasan, M. K., & Islam, M. T. (2026). GenAI and AI tools for STEAM educators: Evidence-informed strategies and resources for the future of teaching. Frontiers in Education, 11, Article 1948064. https://doi.org/10.3389/feduc.2026.1948064
