KAIST announced that it has successfully implemented an artificial intelligence-powered virtual teaching assistant (VTA) in a large-scale undergraduate course. (Image courtesy of KAIST)
DAEJEON, June 6 (Korea Bizwire) — The Korea Advanced Institute of Science and Technology (KAIST) announced on Thursday that it has successfully implemented an artificial intelligence-powered virtual teaching assistant (VTA) in a large-scale undergraduate course, marking a significant milestone in AI-driven education.
The AI assistant, developed by Professors Yoonjae Choi of the Kim Jaechul Graduate School of AI and Hwajeong Hong of the Department of Industrial Design, was used in the fall 2024 semester for the course Programming for Artificial Intelligence, which had 477 enrolled students — the largest deployment of its kind in a real-world academic setting.
Unlike general-purpose generative AIs like ChatGPT, the KAIST VTA was specifically tailored for educational use. It utilizes retrieval-augmented generation (RAG) technology, enabling it to vectorize and index lecture slides, coding exercises, and video content to provide contextually accurate and material-grounded responses to student questions.
When a student asks a question, the system searches through course materials in real time to deliver a precise and trustworthy answer — enhancing reliability compared to standard large language models.
Unlike general-purpose generative AIs like ChatGPT, the KAIST VTA was specifically tailored for educational use. (Image supported by ChatGPT)
According to Kwon Soon-joon, the doctoral candidate who served as head teaching assistant and first author of the study, the VTA significantly reduced the burden on human TAs. “The AI assistant handled routine questions about basic concepts and lecture content, allowing human TAs to focus on more complex student needs,” he said. The number of questions requiring human responses dropped by about 40 percent compared to the previous year.
Over the 14-week semester, more than half of the students used the system, resulting in 3,869 recorded interactions. Usage was especially high among non-computer science majors and students with limited prior knowledge of AI. Many students preferred asking theoretical questions to the VTA, appreciating the nonjudgmental environment it provided.
Students who had previously hesitated to approach human teaching assistants reported greater satisfaction with the AI-driven interaction, the research team noted.
Professor Choi, who led the course and the research, said the project demonstrated the real-world potential of AI in academic settings. “We’ve confirmed that this technology can benefit both students and instructors. We look forward to expanding its use across a wider range of courses,” he said.
To promote broader adoption, the research team has released the VTA’s source code on the open-source platform GitHub, inviting other institutions and researchers to build on their work.
Kevin Lee (kevinlee@koreabizwire.com)
