Unlike conventional social AI agents, AI streamers are multi-modal artificial intelligence systems that engage in autonomous, real-time social interactions with audiences in a dynamic and public online space. Through a qualitative thematic analysis of 1891 comments on a YouTube channel of Neuro-sama, an exemplar of popular AI streamers, we reveal that AI streamers enhance viewers’ experiences through their unique personality development, behavioral autonomy, and nuanced AI-creator relationships; yet they also raise concerns about emotional damage, problematic training data, and heightened moderation challenges in real-time streaming environments. We contribute to HCI at the unique intersection of AI for social needs and live streaming research by highlighting how AI streamers reshape live streaming practices through innovating its creative content creation, novel streamer identity practices, and rich streamer-audience interaction mechanisms. We also propose three design principles for strengthening AI streamers’ social and creative affordances while mitigating identified risks, which inform broader AI agent designs in public online social spaces.
ACM CHI Conference on Human Factors in Computing Systems