Conference Symposium2-2

Generative AI, especially large language models (LLMs), is rapidly accelerating peripheral research tasks such as document drafting, code templating, summarization, and figure preparation. However, hallucinations, weak extrapolation, black-box dependence, privacy risks, tool-chain security, and energy costs remain serious concerns in medical, educational, and research settings. This educational article explains these benefits and limitations for readers in physical fitness science and introduces AI to Learn (AI2L), a practical framework that uses AI for learning support while removing AI-originated black-box elements from final deliverables. AI2L is organized around transparency, accountability, data protection, and Green AI. From the perspective of what should remain as the final product, the article presents examples including local pulse-wave peak detection based on dummy-data prototyping, video-based extraction of hand/body/mouth motion, visualization-assisted classification AI for learning tacit skills, symbolic regression for generating interpretable candidate laws, local reusable scripts for seating-plan preparation, and educational infrastructure that separates teacher-side AI use from learner-side evidence. The aim is to help non-AI specialists begin using AI without sacrificing learning, explainability, or responsibility.