Description
- Abstract:
- College is a critical transitional period, and students spend much of it learning how to balance academic, personal, and financial responsibilities while adapting to daily stressors that can meaningfully shape mental health. As GenAI – particularly large language models (LLMs) – become increasingly embedded in students’ lives, their intersection with psychology raises the question of whether LLM-based tools can responsibly detect and distinguish complex, human traits. In this thesis, we examined 1) the relationship between interest curiosity (IC), deprivation curiosity (DC), and generalized anxiety disorder (GAD) in artificial college-aged students, 2) whether LLMs can be used as a monitoring tool for these two primary responses to the unknown, and 3) the extent to which LLM-based assessments can reliably score these traits across repeated sessions and different model backends. Overall, our research indicated that LLM-based psychometric scoring is highly sensitive to how personas are represented and that an intermediate prompt structure (Dataset 2) provides the most reliable extraction of anxiety-curiosity relationships across both ChatGPT‑4o and ChatGPT‑4.1. While synthetic participants can support generalizability checks and bias surfacing, they cannot substitute for human-subject testing of emotional and ethical risks. Finally, our study shows that LLM-based psychometric scoring is highly sensitive to prompt structure, and that an intermediate prompt condition produced the most reliable scoring pattern across both model backends. As AI and LLMs become more integrated into students’ help-seeking behavior, building tools that are reliable, interpretable, fair, and clinically responsible is essential. Our research takes a step in that direction by showing that LLM-based psychometric scoring can be coherent and meaningful, but only when the narrative/prompt has been engineered with the same care as the psychological instruments it seeks to administer.
- Notes:
- Thesis (Sc. M.)--Brown University, 2026
Citation
Gandhi, Saachi,
"Can LLMs Reliably Detect Complex Human Emotions such as Curiosity and Anxiety"
(2026).
Biology and Medicine Theses and Dissertations, Biotechnology.
Brown Digital Repository. Brown University Library.
https://repository.library.brown.edu/studio/item/bdr:s5gqg74f/
Relations
Collections:
-
Biology and Medicine Theses and Dissertations
Theses and Dissertations for the Biology and Medicine department....