Dissertation research
My dissertation research asks how we can understand and support students’ learning when they use generative AI for help. I’m especially interested in the reasoning that happens between receiving an answer and deciding whether to use it.
This matters for assignment design. A student can submit a plausible solution without showing how they evaluated the chatbot’s contribution. I use perturbed discrete mathematics questions to make some of those decisions observable, then examine the conversations students have around them.
How the work connects
The online, on-demand coursework study examined why undergraduates seek computing instruction outside their degree programs and why they stop. Some participants reported turning to LLMs. That raised a further question for me: what kinds of help do students seek from these systems, and what do they do with it?
The AIED case study examined eight students’ chatbot conversations during a perturbed discrete mathematics assignment. It identified patterns of cognitive scaffolding and GenAI dependence associated with assignment grades in that sample. Those observations motivate instruction in checking a chatbot’s reasoning; they do not by themselves show that the instruction will improve learning.
Current work
An ongoing classroom study places instruction between two perturbed assignments. Its preprint describes three procedures for checking a proof, strengthening a prompt, and working through a problem with a chatbot. Analysis was still in progress when the preprint was written.
A manuscript under review examines mathematical performance on models available through free access tiers. This work considers the capability evidence students might use when deciding how much to rely on a tool.
I’m also developing a comparison study to examine whether the instruction transfers to other problems. That is a study design, rather than a reported result.
Status
This page describes the research direction. A committee-approved dissertation title and abstract are not posted here. The linked records distinguish published work from preprints and work under review.