Development of an Adaptive AI Physics Tutor Based on a Large Language Model to Improve Scientific Literacy and Remediate High School Students’ Misconceptions
TL;DR
A theoretical framework for an Adaptive AI Physics Tutor, a new-generation Intelligent Tutoring System integrating LLM architecture, Retrieval-Augmented Generation, a Misconception Detector, and a Student Model into one personalized, adaptive learning ecosystem is formulates.
Abstract
The shift toward Society 5.0 and generative AI (Artificial Intelligence) has opened new opportunities for personalizing physics learning, yet Indonesian students’ scientific literacy in PISA (Programme for International Student Assessment) 2022 remains far below the OECD (Organization for Economic Co-operation and Development) average, and misconceptions involving force, motion, energy, electricity, and waves remain resistant to conventional instruction. LLM (Large Language Model) chatbots such as ChatGPT, Gemini, and Claude are used ad hoc by students, without being designed to diagnose misconceptions, grounded in conceptual change theory, or equipped with PISA-aligned scientific literacy assessment. This article formulates a theoretical framework for an Adaptive AI Physics Tutor, a new-generation Intelligent Tutoring System integrating LLM architecture, Retrieval-Augmented Generation, a Misconception Detector, and a Student Model into one personalized, adaptive learning ecosystem. Using narrative review, critical review, concept analysis, and theory synthesis on 2021-2026 literature, it combines Conceptual Change theory, Interactive Engagement, Modeling Instruction, Vygotsky-Piaget constructivism, and TPACK (Technological Pedagogical Content Knowledge) into a conceptual model mapping student input, the AI engine, and learning output. The resulting model positions the AI Physics Tutor as a mediator of misconception remediation explicitly aligned with three PISA competencies, namely explaining phenomena scientifically, evaluating and designing scientific inquiry, and interpreting data and evidence, and offers a staged 2026-2045 development roadmap with a Design-Based Research agenda for prototype validation, along with theoretical and practical implications for curriculum developers, physics teachers, universities, and policymakers in Indonesia.
