Integrating Physiological Data with Large Language Models for Empathic Human-AI Interaction
Project Overview
The document explores the integration of generative AI, specifically through the use of Large Language Models (LLMs), in the educational context, focusing on their application for mental health support. It introduces an Empathic LLM (EmLLM) chatbot, which is designed to monitor and manage stress among students. A pilot study is detailed, showcasing how this innovative approach can provide personalized mental health assistance, thus enhancing empathic human-AI interactions. The findings reveal both the effectiveness of the EmLLM in facilitating better emotional support for students and the challenges encountered in deploying such technology within educational and therapeutic frameworks. Overall, the document emphasizes the potential of generative AI to improve mental health outcomes in educational settings while acknowledging the complexities involved in its implementation.
Key Applications
Empathic LLM (EmLLM) chatbot
Context: Mental health support for students, specifically Ph.D. students experiencing work-related stress.
Implementation: The EmLLM chatbot was developed by integrating physiological data from a smartwatch with a fine-tuned LLM (Falcon-7B), utilizing deep learning to predict user stress and customizing the chatbot's responses based on these predictions.
Outcomes: The pilot study showed improved accuracy in predicting user stress and a positive therapeutic alliance between users and the chatbot, with participants reporting satisfaction with the empathetic nature of the interaction.
Challenges: Challenges included ensuring the reliability of stress predictions, managing privacy concerns related to physiological data, and the limitations of current LLMs in generating human-like empathetic responses.
Implementation Barriers
Privacy and Security
Concerns regarding the management of sensitive physiological and conversational data could lead to breaches of confidentiality.
Proposed Solutions: Implementing robust data protection measures and ethical guidelines to ensure user trust and safeguard sensitive information.
Reliability of Outputs
Concerns about the ability of LLMs to produce reliable and consistent outputs in the context of mental health support.
Proposed Solutions: Fine-tuning LLMs on specialized datasets and employing prompt engineering techniques to enhance response quality.
Project Team
Poorvesh Dongre
Researcher
Majid Behravan
Researcher
Kunal Gupta
Researcher
Mark Billinghurst
Researcher
Denis Gračanin
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Poorvesh Dongre, Majid Behravan, Kunal Gupta, Mark Billinghurst, Denis Gračanin
Source Publication: View Original PaperLink opens in a new window
Project Contact: Dr. Jianhua Yang
LLM Model Version: gpt-4o-mini-2024-07-18
Analysis Provider: Openai