Responsible AI

Societal AI

Project 1 Human-centered Approach for Measuring Societal Biases in Vision Language Models

This project develops methods to detect hallucinations and societal biases in vision-language models. Through tools like HalLoc and Behavior-SD, the team uncovers how these models deviate from human norms. Current work involves building moral judgment datasets and personalization benchmarks, enabling AI alignment with complex social expectations.

Publications & Presentations

• Park, E., Kim, M., Kim, G. (2025). HalLoc: Token-level Localization of Hallucinations for Vision Language Models. (Learn more)
• Song, S., Lee, T., Ahn, J., Sung, J.H., Kim, G. (2025). Is a Peeled Apple Still Red? Evaluating LLMs' Ability for Conceptual Combination with Property Type. (Learn more)
• Lee, S., Kim, K., Kim, G. (2025). Behavior-SD: Behaviorally Aware Spoken Dialogue Generation with Large Language Models. (Learn more)

---

Project 2 Understanding the Impact of Personalized AI on Self-understanding and Communication

This project creates personalized LLM-based agents—digital doppelgängers—that reflect users’ inner voices. The research shows how these agents support career reflection, identity development, and psychological clarity. The team further explores how agents can simulate multi-party social interactions, contributing to emotionally intelligent AI design.

Publications & Presentations

• Lee, K., Kim, S. H., Lee, S., Eun, J., Ko, Y., Jeon, H., ... Kim, E.M.,& Lim, H. (2025). SPeCtrum: A Grounded Framework for Multidimensional Identity Representation in LLM-Based Agent. (Learn more)
• Jeon, H., Yoon, S., Lee, K., Kim, S. H., Kim, E. H., Cho, S., ... Kim, E.M.,& Lim, H. (2025). Letters from Future Self: Augmenting the Letter-Exchange Exercise with LLM-based Agents to Enhance Young Adults' Career Exploration. (Learn more)

---

Project 3 Research on Simulating Problematic Nurse-Patient Communication and Building LLM Training Agents

Aiming to improve nursing communication, this project simulates emotionally charged patient interactions using adaptive LLM-based virtual agents. It also develops training tools with reflective debriefing cycles and collects comparative dialogue corpora to inform curriculum design. The work provides an evidence-based foundation for training novice nurses in real-world communication.

Publications & Presentations

• Kim, D., Lim, H., Kim, E.M., Suh, E.E. (2025) Navigating Challenging Conversations: Communication Strategies from Experienced Nurses in South Korea
• Lee, K., Lee, S., Kim, E. H., Ko, Y., Eun, J., Kim, D., ... & Lim, H. (2025). Adaptive-VP: A Framework for LLM-Based Virtual Patients that Adapts to Trainees' Dialogue to Facilitate Nurse Communication Training. (Learn more)

Participating Researchers