Where should an AI tutor’s knowledge of a student’s interests come from?
My dissertation compares asking students with reading the records they already produce, and measures what each gives an AI tutor to teach from.
Research
An AI system that adapts to a person must first know something about them, by asking or by reading records the person already produced. My research asks which to rely on, how accurate and well grounded the result is, and what it costs in privacy.
An AI system that adapts to a person first needs to know something about them
- By askingwhat the person tells it
- By reading recordswhat the person already produced
My research asks
- Which to rely on?
- How accurate and well grounded is the result?
- What does it cost in privacy?
The work falls under three contexts: personal (what systems learn about a student or user), embodied (learning through hands, puppets, avatars and devices), and civic (crowd-sourced urban digital twins).
Personal
Grounding: checking what models assert about people
A unit-level check of every claim a model makes about a student — is it grounded in the student’s own material, or inferred?
Teacher support technologies
Teacher-facing work on bringing students’ personally relevant information into instruction.
Also: explanation and recommendation
- Explanation types and user motivations in recommender systems (AAAI HCOMP 2020, co-author).
- Everyday-inspired movie recommendation from personal social media (IFIP INTERACT 2023, co-author).
Both are listed on the publications page.
Embodied
Embodied and creative learning
Hand enactment, puppets, avatars and AR glasses in storytelling and everyday learning, and the systems I built or helped build along the way.
Civic
Civic AI: crowd-sourced urban digital twins
Our design framework for gamified, LLM-assisted crowd-sourced urban digital twins, and a flood prototype for Jacksonville that confirms an event when three channels agree.