Mansi, Newton, Moore, Alemdar & Riedl (2026) — cs.HC 📄 Full text (arXiv)
Mansi et al. (2026) introduce Concept Catalyst, a system designed around 'scrutable interfaces' — interfaces that make AI reasoning visible and editable by users. Working with K-12 teachers, the study shows that when teachers can inspect and modify how a generative-ai tool processes their inputs, they report higher trust, greater sense of control, and better alignment with their pedagogical goals. This directly addresses a critical gap in edtech-platform design: most teacher-facing AI tools operate as black boxes, undermining the teacher-role as a professional decision-maker. The Concept Catalyst approach empowers teachers to become co-designers of AI-assisted lesson planning, not just consumers. The work contributes to ai-literacy by demonstrating that 'scrutability' as a design principle can bridge the gap between AI capability and classroom reality. The study also has implications for k-12 education policy, suggesting that AI tools adopted in schools should meet scrutability standards.