Beyond one-on-one: Authoring, simulating, and testing dynamic human-AI group conversations

Google Research · 5 months ago

Researchers introduced DialogLab, an open-source framework for authoring, simulating, and testing multi-party conversations between humans and AI agents, presented at ACM UIST 2025. The framework was evaluated with 14 participants who preferred the human-control mode for testing conversations, rating it as more engaging and realistic than fully autonomous or purely reactive agent behaviors. The tool enables applications in education, game design, and social science research by allowing designers to balance scripted structure with improvisational AI responses in group dialogue scenarios.