Scaffolding students-AI dialogue for safe educational interactions
Adolescents are the largest and fastest-growing age group adopting large language models, tools not designed for their needs. The paper proposes SCAFFOLD, an architecture that surrounds a model’s output with external verification, targeted repair and safe fallback while preserving data privacy. It keeps the effort that learning requires. A first prototype was piloted in a classroom with 12 to 16-year-old students and a social robot.
































