Engineering Socially-Interactive AI

In recent years, considerable efforts have been made to improve the expressive behaviors of artificial interactive agents, ranging from animated virtual characters and anthropomorphic robots to digitally enhanced everyday objects. Experience shows, however, that realistic appearance and animation alone do not suffice to maintain long-term engagement beyond an initial novelty effect. Motivated by the goal of enabling smooth, natural interactions, this talk discusses the challenges of engineering socially interactive AI. Particular emphasis is placed on temporal alignment, including the concurrent generation of appropriate system responses, such as backchannels, during ongoing interaction with the human interlocutor. The talk highlights the need for tightly integrated real-time systems that coordinate Multimodal Social Behavior Analysis and Generation under dynamically evolving interaction conditions.