Brief AI Pre-Lecture Chats Match Human Tutors on Neural Engagement — HKUST's GPT-4 Instructor Study

Researchers at the Hong Kong University of Science and Technology (HKUST), led by Prof. LI Ping, Dean of the School of Humanities and Social Science, tested whether a brief AI-led conversation before a lecture could match the learning benefits of a human-led one.

Fifty-seven university students were randomly assigned to three groups: no pre-lecture interaction, an 8-10 minute face-to-face chat with a human instructor, or an 8-10 minute conversation with a GPT-4-powered AI instructor combining speech recognition, content generation, text-to-speech synthesis and real-time talking-head animation.

Both interaction groups outperformed the no-interaction group on recall, comprehension and knowledge-transfer tests, especially on harder comprehension questions, and showed stronger synchronized neural activity in brain regions tied to information processing and cognitive resource allocation. No significant difference was found between the human and AI groups on any learning measure. The AI group did show significantly lower gaze alignment and reported lower perceived social closeness than the human group, despite matching learning outcomes.

The study, published in Neuron in May 2026, is among the first to bring neuroscientific measures to bear on human-versus-AI instruction, but its findings rest on a single small sample (n=57) at one institution and a narrow 8-10 minute intervention, so scalability and generalisation to full courses or younger learners remain untested.

Read the full analysis: https://phys.org/news/2026-05-ai-human-teachers-pre-chat.html