Date of Award

Spring 2026

Document Type

Thesis (Master's)

Department or Program

Computer Science

First Advisor

Soroush Vosoughi

Second Advisor

Paul Barr

Third Advisor

Nikhil Singh

Abstract

Artificial intelligence has increasingly been adopted in healthcare, largely for specialized tasks and under significant human oversight. The use of large black-box systems raises important concerns about transparency in high-stakes environments such as clinical decision-making. Clinical communication is fundamentally human-centered, and failures in judgment can have serious consequences for patient care. Overestimating the reasoning abilities of large language models may lead to undue trust in fabricated or “hallucinated” outputs, while rejecting AI-assisted tools altogether may preserve inefficient workflows and contribute to missed or delayed diagnoses. These concerns reflect a broader tradeoff between accuracy and interpretability: although more complex models may achieve stronger performance, their opacity can limit accountability, safety, and trust. We posit that interpretability should be considered a core requirement for the responsible use of AI in healthcare. This is especially true in domains such as electronic health records and clinical speech recognition, where AI may reduce administrative burden and improve efficiency only if it can be understood, validated, and appropriately overseen.

In this work, we discuss the limitations of EHR systems that may be safely addressed using AI or ML methods. We specifically explore the task of automatic medical annotation by evaluating three methods to extract medical insights from transcriptions of primary care visits. We implement a zero‑shot prompting of Meta‑Llama‑3.1‑8B‑Instruct, low‑cost parameter‑efficient fine‑tuning (PEFT) of the same LLM using LoRA, and span‑candidate generation with a multinomial logistic‑regression classifier. We aim to identify instances of discussion regarding medications, symptoms, medical conditions, and test/imaging results. We compare both the performance and interpretability of these methods.

Download

Share

COinS