Modularizing Educational LLM-Agency for Fostering Responsible Learning Assistance

LLMs alone don't follow good teaching practices: While large language models have access to educational research, they naturally tend to ignore pedagogical principles, which can harm students' ability to transfer knowledge, think critically, and be creative—highlighting why simply deploying standard chatbots in classrooms is risky.

Modular design enables responsible AI in education: Rather than using generic, "out-of-the-box" AI chatbots, the researchers propose breaking down the educational assistant into separate modules for different stages of problem-solving, allowing educators to embed specific teaching strategies at each step and maintain better control over how students are guided.

Transparency and oversight matter for learning: The modular architecture makes the AI's decision-making process more transparent and "overseeable," meaning teachers and students can better understand why the system is providing certain guidance, rather than receiving opaque responses from a black-box AI system.

Pedagogical principles should be deliberately designed in: The paper emphasizes that responsible educational AI requires intentional incorporation of evidence-based teaching methods into the system's structure, rather than hoping an LLM will naturally adopt good practices—suggesting that responsible AI deployment requires active human expertise in instructional design.