This paper explores the development of proactive conversational agents (PCAs) from a human-centered perspective, emphasizing the importance of balancing technological advancement with human-centered design principles. The authors propose a new taxonomy with three key dimensions: Intelligence, Adaptivity, and Civility. They re-examine existing literature on PCAs based on this taxonomy and outline a five-stage process for PCA system construction: Task Formulation, Data Preparation, Model Learning, Evaluation, and System Deployment. The paper discusses the challenges and future research directions in each stage, focusing on how to integrate human-centered principles into these stages. It also introduces a multidimensional evaluation framework to assess the abilities of PCAs in terms of Intelligence, Adaptivity, and Civility. Finally, the paper emphasizes the importance of human-centered system deployment, including user interface design and fostering appropriate trust and reliance between users and PCAs. The authors conclude by highlighting the need for human-centered approaches in the development of superintelligent AI systems to ensure they serve human interests effectively.This paper explores the development of proactive conversational agents (PCAs) from a human-centered perspective, emphasizing the importance of balancing technological advancement with human-centered design principles. The authors propose a new taxonomy with three key dimensions: Intelligence, Adaptivity, and Civility. They re-examine existing literature on PCAs based on this taxonomy and outline a five-stage process for PCA system construction: Task Formulation, Data Preparation, Model Learning, Evaluation, and System Deployment. The paper discusses the challenges and future research directions in each stage, focusing on how to integrate human-centered principles into these stages. It also introduces a multidimensional evaluation framework to assess the abilities of PCAs in terms of Intelligence, Adaptivity, and Civility. Finally, the paper emphasizes the importance of human-centered system deployment, including user interface design and fostering appropriate trust and reliance between users and PCAs. The authors conclude by highlighting the need for human-centered approaches in the development of superintelligent AI systems to ensure they serve human interests effectively.