Tell Me More, Tell Me More: AI-Generated Question Suggestions for the Creation of Interactive Video Recordings

Time-Offset Interaction Applications (TOIAs) are narrative-sharing systems that use databases of previously recorded videos of real people to mimic conversations with them. These video databases comprise large (the larger, the better) collections of videos of answers paired with specific questions. This paper focuses on a solution to the challenge of creating such databases without exhausting their creators’ creativity, energy, and interest. We describe the design and development process of Question Suggester (QS)-an intelligent GPT-3-based service that generates suggested questions following up a conversation based on the history of recorded questions and answers. We conduct a user study to empirically evaluate the value of QS for reducing the effort to create a video database while creating an interaction that is enjoyable. The users’ average experience rating for QS is 4.6 compared to 4.0 when QS is not used (on a 1-5 scale, p-value<0.05). The experience with interactions so created is more enjoyable, too (3.7vs. 3.3, p-value<0.05). The usage metrics and qualitative feedback confirm that QS is essential for interactive video-recording systems and for increasing their adoption.