Large Language Models and Agentic AI in STEM Teacher Education: Opportunities, Risks, and Responsible Implementation

 

Guest Editors

Alkinoos Ioannis Zourmpakis¹ and Stamatios Papadakis²

¹ Department of Special Education, University of Thessaly, Volos, Greece
² Department of Preschool Education, University of Crete, Rethymno, Greece

Problem statement

Large Language Models are increasingly used by STEM teachers and teacher educators to draft activities, generate feedback, and support inquiry in STEM classrooms. That potential comes with risks, including factual errors, content that is not age appropriate, and references that turn out not to exist, especially when relying on a single LLM. Agentic AI, in which distinct roles plan, retrieve information, and check each other's output, is one useful response, especially when different models get involved. Also, another interesting aspect is when LLMs are grounded in retrieval-augmented generation (RAG) and a curated knowledge base is utilized rather than left to rely on memory alone. Locally hosted, open-weight models add a further layer of data protection, and this matters especially when involving young learners. None of this works, however, without teachers who are prepared to implement it in class. The teacher remains central to this process: judging scientific accuracy, questioning sources, and deciding when to reject a draft are decisions no system can make on the teacher's behalf. Consequently, teacher education is just as central to this Special Issue as the technology itself, since responsible use depends on how well teachers develop this judgement alongside their pedagogical and content knowledge.

This Special Issue looks at how LLMs and agentic AI are shaping STEM teacher preparation and practice at every level, from early childhood through higher education. Early childhood and primary contexts are of particular interest, since the research there is still fairly limited, but contributions addressing any stage of STEM teacher education are welcome.

Focus Areas

We welcome submissions from diverse perspectives, including but not limited to:

  • Design and evaluation of LLM-based or agentic AI tools for pre-service and in-service STEM teacher education
  • Multi-agent architectures, such as generator and reviewer workflows or cross-model verification, for producing and checking STEM/STEAM instructional content
  • Retrieval-augmented generation and curated knowledge bases for grounding AI output and reducing hallucination
  • Locally hosted or open-weight LLM deployment for data privacy and institutional data sovereignty in schools
  • Adaptive and personalized STEM learning environments driven by generative or agentic AI
  • AI literacy and critical evaluation skills for STEM educators, particularly at early childhood and primary level
  • Prompt design, human-AI verification workflows, and safeguards against inaccurate or age-inappropriate AI output
  • Ethics, safety, and policy considerations relevant to AI-supported STEM education, including the EU AI Act and GDPR by design
  • Equity, inclusion, and accessibility in AI-supported STEM teacher preparation
  • Systematic reviews and meta-analyses on LLMs or agentic AI in STEM teacher education

Aims and Scope

By engaging with these themes, this special issue aims to advance our understanding of how LLMs and agentic AI can be integrated responsibly into STEM teacher education, enriching both teacher preparation and classroom practice. We encourage submissions that offer new theoretical insights, empirical findings, instructional designs, or practical applications that contribute to the ongoing discourse on AI-supported STEM education, with particular interest in early childhood and primary contexts where the evidence base remains limited.

Submission Guidelines

All submissions undergo a rigorous double-blind peer-review process.

Submission Process

  1. Visit https://www.editorialpark.com/ejmste and log in (or create an EditorialPark account if you do not have one).
  2. Click the “Submit New Manuscript” button.
  3. Under the “Issue Type” section, select “Special Issue”.
  4. Select “Large Language Models and Agentic AI in STEM Teacher Education: Opportunities, Risks, and Responsible Implementation” from the special issue list and submit your manuscript.

Key Dates

Submission Deadline: July 31, 2027

Expected Publication: December 2027

Contact

If you have any questions, please feel free to reach out to us at ejmste@ejmste.com.