What Counts as Proactive? Rethinking Proactivity in Conversational Agents
TL;DR
This provocation argues that current uses of the term proactivity in the context of conversational AI are overly broad and conceptually imprecise, which limits the ability to design, compare, and evaluate proactive conversational agents.
Abstract
Proactivity has become a central concept in research on conversational user interfaces and human–computer interaction. It is an evolutionary stage for conversational agents. Nevertheless, despite the growing research on this topic, the term remains conceptually underspecified and inconsistently applied. Systems that send reminders or recommend content are often labeled as proactive, even when their underlying mechanisms and intentions differ fundamentally. This provocation argues that current uses of the term proactivity in the context of conversational AI are overly broad and conceptually imprecise, which limits our ability to design, compare, and evaluate proactive conversational agents. We synthesize perspectives from human-computer interaction, sociology, and psychology to propose a principled definition and a conceptual framework for proactive conversational agents to distinguish proactive from reactive and other related system behaviors.