A social chatbot is a conversational agent designed to hold open-ended, emotionally engaged conversation with a person over a long period, rather than to answer a single question or complete a task. The clearest articulation comes from the Microsoft Research paper “The Design and Implementation of XiaoIce, an Empathetic Social Chatbot” (Li Zhou, Jianfeng Gao, Di Li, and Heung-Yeung Shum, arXiv 1812.08989, 2018). The authors define the goal as building “an AI companion with an emotional connection to satisfy the human need for communication, affection, and social belonging.”
Two design choices distinguish a social chatbot from a task assistant. First, it is built to combine what the XiaoIce team called intelligence quotient (IQ) with emotional quotient (EQ) - the system tries to recognize a user’s emotional state and respond to it, not just retrieve facts. Second, it is optimized for a different metric. Instead of resolving a request in as few turns as possible, the XiaoIce team modeled the conversation as a Markov Decision Process and optimized for “Conversation-turns Per Session” (CPS) over the long run, treating a longer, repeated relationship as success.
This frame inverts the assumption behind most commercial chatbots, where a fast resolution and a short conversation are the point. It also makes the design directly responsible for a human’s emotional experience, which is why social chatbots sit at the center of later debates about attachment, loneliness, and harm.
Why business readers should care: a product optimized to maximize time-in-conversation with an emotionally vulnerable user is optimizing for something very different from a support tool that resolves tickets, and the two should not be evaluated, marketed, or regulated as if they were the same thing.