Research Summary: A framework for integrating large language models in secondary physics education: practical design, opportunities, risks, and pedagogical principles

A framework has been proposed for integrating Large Language Models (LLMs) into secondary physics education, addressing the lack of structured guidance for their responsible and pedagogically sound deployment. This framework is rooted in established educational theories and identifies three core application modalities: teacher-facing instructional design tools, guarded student-facing inquiry tools, and assessment-augmentation tools. The analysis highlights both domain-specific opportunities, such as addressing misconceptions in mechanics and electromagnetism, and inherent risks including cognitive offloading, epistemic opacity, and algorithmic bias.

Why it matters

The responsible and effective integration of advanced AI technologies like LLMs into educational systems represents a critical strategic imperative for maintaining pedagogical relevance and fostering future-ready competencies. Understanding the opportunities and risks associated with these technologies is essential for developing robust, ethical, and effective educational strategies that leverage innovation while mitigating potential harms to learning outcomes and equity.

Key insights

• The rapid advancement of LLMs presents significant potential for transforming secondary physics education.
• Existing theoretical and practical frameworks for responsible LLM integration in education are currently underdeveloped.
• A new framework, grounded in constructivism, cognitive load theory, and the TPACK model, is proposed for LLM integration.
• Three primary LLM application modalities are identified: teacher support, student inquiry, and assessment enhancement.
• Opportunities include targeting specific student misconceptions in physics domains like mechanics and electromagnetism.
• Key risks associated with LLM integration are cognitive offloading, epistemic opacity (lack of transparency in knowledge source), and algorithmic bias.

Source

Frontiers in Education — https://www.frontiersin.org/articles/10.3389/feduc.2026.1874510