LLM-Powered AI Tutors with Personas for d/Deaf and Hard-of-Hearing Online Learners
Project Overview
The document examines the integration of generative AI, specifically large language model (LLM)-powered intelligent tutoring systems (ITS), in the education of d/Deaf and Hard-of-Hearing (DHH) learners. It highlights how the inclusion of culturally relevant personas in AI tutors can enhance the educational experience by fostering relatability and trust. A user study conducted with DHH participants demonstrates that AI tutors who reflect an understanding of DHH community culture are viewed as more human-like and credible. Key findings emphasize the unique needs of DHH learners, such as the necessity for visual sign language support and the significance of transparency concerning the AI tutors' backgrounds. Overall, the document illustrates that by tailoring AI educational tools to better align with the experiences and requirements of DHH students, educators can improve engagement and learning outcomes in this demographic.
Key Applications
LLM-powered AI tutors with personas for DHH learners
Context: Online learning for d/Deaf and Hard-of-Hearing students
Implementation: Developed a prototype interface allowing DHH learners to interact with ChatGPT and three AI tutors with different backgrounds in DHH education while watching an educational video.
Outcomes: DHH learners found tutors with DHH education experiences to be more trustworthy and human-like. Participants suggested that these tutors elicited more engagement and provided a better learning experience.
Challenges: LLMs sometimes lack understanding of specific needs for DHH learners, such as the multimodality of sign language and the need for transparency in background information.
Implementation Barriers
Technical barrier
Current LLMs do not adequately support sign language, provide visual demonstrations, and lack effective visual representation capabilities.
Proposed Solutions: Future LLMs should incorporate sign language gloss and improve visual representation capabilities.
Trust barrier
Participants expressed the need for more transparency regarding the AI tutors' backgrounds, their ties to the DHH community, and the need for credibility.
Proposed Solutions: Provide detailed backgrounds and clarifications regarding the AI tutors' experiences and credibility.
User experience barrier
Responses from AI tutors varied in length and detail, leading to mixed preferences among users.
Proposed Solutions: Allow users to customize the length and style of responses from AI tutors.
Project Team
Haocong Cheng
Researcher
Si Chen
Researcher
Christopher Perdriau
Researcher
Yun Huang
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Haocong Cheng, Si Chen, Christopher Perdriau, Yun Huang
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