Towards Safe, Robust, Interpretable Dialogue Agents for Democratized Medical Care Reverse Engineering the Invariances of the Primate Visual System

PI: Julia Hirschberg
Co-PI: Sarah Ita Levitan, CUNY; Tatiana Emmanouil, CUNY

Julia Hirschberg
Columbia

Sarah Ita Levitan
CUNY

Tatiana Aloi Emmanouil
CUNY

Abstract

With the growing capabilities of Large Language Models (LLMs) in enhancing communication, AI-powered dialogue agents are increasingly used in clinical psychology, promising to augment human therapists. However, deploying these technologies in medical settings involves significant risks, including the lack of safety regulation and robust evaluation methods. Our project aims to develop reliable LLMs that provide immediate mental health support by establishing comprehensive safety guidelines and creating interpretable, effective medical AI/NLP systems tailored for psychological counseling. A cornerstone of this effort is our continuous interdisciplinary collaboration with medical professionals to ensure our technological advancements are safely grounded in clinical reality.

Picture1

Publications

  • Chen, R., Liang, W., Gong, Z., Ai, L., & Hirschberg, J. (2026). Detecting Mental Manipulation in Speech via Synthetic Multi-Speaker Dialogue. arXiv. https://doi.org/10.48550/ARXIV.2601.08342
  • Bojic, I., Ong, Q. C., Ma, S. H. X., Ai, L., Liu, Z., Gong, Z., Hirschberg, J., Ho, A. H. Y., & Khong, A. W. H. (2025). SMARTMiner: Extracting and Evaluating SMART Goals from Low-Resource Health Coaching Notes. Findings of the Association for Computational Linguistics: EMNLP 2025, 16288–16305. https://doi.org/10.18653/v1/2025.findings-emnlp.885
  • Gong, Z., Shi, P., Donbekci, K., Ai, L., Chen, R., Sasu, D., Wu, Z., & Hirschberg, J. (2025). Learning More with Less: Self-Supervised Approaches for Low-Resource Speech Emotion Recognition. arXiv. https://doi.org/10.48550/ARXIV.2506.02059
  • Cai, Y., Wu, C., Ma, B., Chen, B., Xue, Y., Hirschberg, J., & Gong, Z. (2026). SURE: Synergistic Uncertainty-aware Reasoning for Multimodal Emotion Recognition in Conversations. arXiv. https://doi.org/10.48550/ARXIV.2604.01916 
  • Ma, B., Li, Y., Zhou, W., Gong, Z., Liu, Y. J., Jasinskaja, K., Friedrich, A., Hirschberg, J., Kreuter, F., & Plank, B. (2025). Pragmatics in the Era of Large Language Models: A Survey on Datasets, Evaluation, Opportunities and Challenges. arXiv. https://doi.org/10.48550/ARXIV.2502.12378
  • Gong, Z., Ai, L., Deshpande, H., Johnson, A., Phung, E., Wu, Z., Emami, A., & Hirschberg, J. (2024). CREAM: Comparison-Based Reference-Free ELO-Ranked Automatic Evaluation for Meeting Summarization. arXiv. https://doi.org/10.48550/ARXIV.2409.10883 
  • Liu, Z., Gong, Z., Ai, L., Hui, Z., Chen, R., Leach, C. W., Greene, M. R., & Hirschberg, J. (2025). A Review of Incorporating Psychological Theories in LLMs. arXiv. https://doi.org/10.48550/ARXIV.2505.00003 
  • Liu, Z., Gong, Z., Ai, L., Hui, Z., Chen, R., Leach, C. W., Greene, M. R., & Hirschberg, J. (2025). A Review of Incorporating Psychological Theories in LLMs. arXiv. EACL 2026. https://doi.org/10.48550/ARXIV.2505.00003
  • Gong, Z., Cai, D., Ma, B., Sharma, M., On, T. Q., Morparia, M., Enos, B., Yu, Y., Resnik, P., Levitan, S., & Hirshburg, J. (2026). MCML - A Survey on Mental Health Datasets and Resources. https://mcml.ai/publications/gdm+26/

Resources