False understanding in AI-assisted physics problem solving: a theoretical framework

Abstract The rapid adoption of generative artificial intelligence (AI) in education has renewed longstanding concerns in physics education regarding the separation of answer correctness from conceptual understanding. While AI systems can efficiently generate correct solutions and fluent explanations, their use may unintentionally enable a form of learning in which students appear successful without developing robust, transferable understanding. This paper introduces a theoretical framework to conceptualize false understanding in AI-assisted physics problem solving. False understanding is defined as a state in which learners produce correct or acceptable solutions while lacking the conceptual structures required to independently generate, evaluate, or transfer knowledge. Drawing on research in physics education, cognitive science, and epistemic cognition, the framework identifies key mechanisms; answer–reasoning decoupling, epistemic outsourcing, cognitive short-circuiting, and the illusion of explanatory depth; that explain how AI can alter core problem-solving processes. The framework distinguishes productive pathways, in which AI functions as a scaffold for sense-making, from risk pathways, in which AI substitutes for learners’ cognitive and epistemic engagement. Instructional, assessment, and research implications are discussed, emphasizing the need for AI-resilient task design, process-oriented assessment, and future empirical validation. The framework is positioned as a foundation for understanding and mitigating risks to meaningful physics learning in AI-rich environments, not as a critique of AI use.