Yuming Feng, Yuan Tian, Erica Zhao (2026) 📄 Full text (arXiv)
The rapid integration of llms into intelligent-tutoring threatens to reduce mathematical learning to mere answer generation. This paper presents a design framework for AI tutors that act as reasoning facilitators rather than answer generators, specifically targeting high-stakes exam preparation environments. Through usability studies, the authors demonstrate that scaffolding approaches — where the AI guides students through reasoning steps without providing final answers — yield superior learning outcomes. The framework provides concrete guidelines for designing student-ai-interaction patterns that prioritize deep understanding over superficial completion in higher-ed mathematics.
Related Pages
- automated-grading — Automated assessment approaches in computing education
- llm — Large language models and their applications
- k-12 — K-12 education context
- ai-literacy — AI literacy frameworks and assessment
- student-ai-interaction — How students interact with AI systems
- scaffolding — Instructional scaffolding techniques
- ai-generated-content — AI-produced educational materials
- learning-analytics — Data-driven analysis of learning behaviors
- formative-assessment — Formative assessment in AI-enhanced education
- active-learning — Active learning approaches
Citation
APA: Yuming Feng, Yuan Tian, Erica Zhao (2026). From Answer Generators to Reasoning Facilitators: Designing AI Tutors for Mathematical Reasoning in High-Stakes Environments. arXiv:2607.01692.