Chatting with AI: enhancing the supervision safety of near-L3 automated driving by engaging in non-driving–related tasks

Current near-L3 automated driving systems still require continuous supervision, but prolonged supervision easily induces passive fatigue. This study examines whether non-driving-related tasks (NDRTs), especially chatting with AI, can improve drivers' cognitive state (cognitive load, sleepiness) and hazard perception during 15-minute supervision in daytime, suburban and low-traffic environments. The experiment compared a baseline condition with three NDRTs: listening to music, chatting with human, and chatting with AI. Results showed that all NDRTs increased cognitive load and reduced sleepiness. For novices, human chatting imposed excessive cognitive load and failed to improve hazard perception. Both music listening and AI chatting enhanced it, with AI chatting being more effective in reducing sleepiness. Eye-tracking data confirmed larger pupil sizes and more balanced fixations during AI chatting. In summary, NDRTs alleviated passive fatigue during supervision, and AI chatting improved novices' sleepiness and hazard perception with a moderate cognitive burden, emerging as the relatively optimal solution.