Simulated Teaching and Learning at Scale
Developing frameworks to evaluate AI-generated educational dialogues along two critical dimensions: simulation fidelity and interaction effectiveness.
Overview
Developing frameworks to evaluate AI-generated educational dialogues along two critical dimensions: simulation fidelity (how realistically LLMs mimic student behavior) and interaction effectiveness (whether these exchanges produce meaningful learning). This research helps distinguish between LLMs that merely "sound like" students versus those that facilitate genuine educational progress.
Project Details
Started
2024-10
Status
In Progress
Topics
Research Team
- Kevyn Collins-ThompsonPrincipal InvestigatorUniversity of Michigan
- Michael IonResearch Lead and Postdoctoral FellowUniversity of Michigan
- Sumit AsthanaResearch Assistant (PhD Student)University of Michigan
- Tianyi WangResearch Assistant (Undergraduate)University of Michigan
- Fengquan JiaoResearch Assistant (Undergraduate)University of Michigan
Related Scholarship
Papers
Simulated Teaching and Learning at Scale: Balancing Fidelity and Effectiveness in Tutoring Interactions
Michael Ion, Kevyn Collins-Thompson, S. Asthana
Bayesian Hierarchical Modeling of Large-Scale Math Tutoring Dialogues
Michael Ion, Kevyn Collins-Thompson
Joint Statistical Meetings (2025) - Under review
Talks & Presentations
Adaptive Knowledge Assessment in Simulated Coding Interviews
Michael Ion, S. Asthana, Fengquan Jiao, Tianyi Wang, Kevyn Collins-Thompson
iRAISE Workshop at AAAI Conference (2025)