Eunkyu (Eunice) Park
박은규
Ph.D. Candidate, Seoul National University
Visiting Researcher, CMU HCII
Hi! I am a Ph.D. Candidate in the Interdisciplinary Program in Artificial Intelligence at Seoul National University. I work at the intersection of NLP, computer vision, and HCI, studying how multimodal AI systems reason and how people come to trust them. My research spans three threads: evaluating the trustworthiness of vision-language models (faithfulness of reasoning, localizing perception-to-understanding failures); studying how people perceive, rely on, and are misled by AI explanations; and building benchmarks and alignment methods for multimodal moral and social reasoning. I design human-subject studies to evaluate how users perceive and calibrate trust in these systems, and translate those findings into better learning signals for model alignment.
I am advised by Professor Gunhee Kim at SNU Vision & Learning Lab. At CMU, I am visiting Professor Motahhare Eslami's lab at HCII and work closely with Professor Maarten Sap at LTI.
News
- Sep 2026MM-SCALE is accepted to NeurIPS 2026 (Evaluations & Datasets Track).
- Aug 2026Critical or Compliant? is accepted to EMNLP 2026.
- Jul 2026CoCoT is accepted to COLM 2026.
- Feb 2026New preprint MM-SCALE is on arXiv.
- Feb 2026Extending my position at CMU HCII as a Visiting Scholar through the end of 2026.
- Jul 2025CoCoT is out, written during my first month at CMU.
- Jun 2025HalLoc V1 dataset released on HuggingFace.
- May 2025Joining CMU HCII as a Visiting Scholar through Fall 2025.
Research
How can we design multimodal AI systems whose reasoning is transparent, trustworthy, and aligned with human values, and how do humans perceive, interact with, and sometimes misplace trust in these systems?
-
Human Perception of AI Reasoning
How people evaluate, trust, and are misled by AI reasoning chains. When chain-of-thought explanations support critical thinking versus create false confidence, and what interaction designs encourage careful oversight.
-
Multimodal Moral & Social Reasoning
Benchmarks and preference learning frameworks that capture the continuous, pluralistic nature of human moral judgment across text and image, moving beyond binary labels toward human-aligned supervision.
-
Trustworthiness & Interpretability of VLMs
Detecting and localizing failures in vision-language models, examining how model biases lead to unfaithful outputs, and whether model reasoning faithfully reflects human values rather than mimicking them.
Education
- 2021 –MS/PhD in Artificial Intelligence, Seoul National University
- 2017 – 20B.S. in Computer Science, Columbia University
- 2014 – 17B.A. in Mathematics, Bard College