Surfacing Relational Dynamics in Voice-Based Human-AI Conversation: The Relational Transparency Layer

When people engage with anthropomorphic conversational agents, the relational dynamics that emerge during these interactions stay largely obscure to users. This paper presents the Relational Transparency Layer (RTL), a demonstration system that shows the implicit relational dynamics of human-AI conversation during the exchange and explores whether part of the work of tracking these dynamics can be shifted from the user to the system. The RTL is a voice-based desktop application where users speak to an AI system and hear its responses, while a dedicated local observer model runs alongside the conversation. After each exchange, the observer classifies the user and AI’s positions across five relational configurations, tracks the relational tension between them, and writes a brief narrative of what happened. In a live demonstration, participants speak with the demo system and watch their relational configurations shift as the exchange progresses.