[Colloquium] TODAY: Karen Zhou Candidacy Exam/Dec 10, 2025

[via Colloquium]

This is an announcement of Karen Zhou's Candidacy Exam.
===============================================
Candidate: Karen Zhou

Date: Wednesday, December 10, 2025

Time:  4 pm CST

Remote Location: https://uchicago.zoom.us/j/99812609634?pwd=OCu8lhnrpiap2TGzuIGU0EgGw0BCYR.1 Meeting ID: 998 1260 9634 Passcode: 488370

Location: DSI 105

Title: Structured Frameworks for Human-Centered AI

Abstract: As AI is increasingly used in high-stakes domains such as politics and medicine, its application and evaluation must be reliable and aligned with human reasoning. This work presents domain-grounded frameworks for linguistic analysis, language model evaluation, and assessment of AI systems in these domains. The first suite of metrics quantifies rhetorical distinctiveness and divisiveness in political discourse by combining large language models, lexical resources, and graph-based analyses to characterize how U.S. presidents differ in linguistic style and use of antagonistic language. The second framework introduces a systematic approach for distilling large-scale clinician feedback into structured, LLM-enforceable checklists for evaluating AI-generated clinical notes. These checklists yield interpretable and auditable criteria that improve the alignment between automated evaluations and expert assessments, outperforming a baseline approach in coverage, diversity, and predictive power. Building on these insights, subsequent experiments will explore checklist-guided data selection and augmentation as a means to enhance model specialization and fine-tuning efficiency. We also leverage structured criteria to guide the design of interactive AI systems that can facilitate informed civic decision-making, evaluating how AI can support voters' comprehension, confidence, and reasoning about ballot measures. Together, these studies propose structured, human-centered methods for analyzing language, evaluating models, and guiding AI system design in applied domains.


Advisors: Chenhao Tan

Committee Members: Chenhao Tan, Mina Lee, Lexing Xie