Researchers' Virtual Roundtable Uses Multi-Agent AI to Simulate Human Brainstorming

In a new arXiv preprint, researchers Tim Dorn, Saara A. Khan, and Julie Mumford unveil the Virtual Roundtable, a multi-agent system designed to overcome human brainstorming limitations like groupthink and echo chambers. The architecture operates in two phases: divergent thinking to generate a wide variety of ideas, and convergent thinking to evaluate and rank the most promising ones. It employs diverse AI personas that engage in roundtable discussions, each maintaining private thoughts while commenting publicly. An agentic facilitator steers the conversation toward productive outcomes, while per-persona quotas on idea submissions and votes enforce balanced participation and produce natural rankings. Throughout the session, the system tracks each idea's lineage, capturing how concepts originate, evolve, and cross-pollinate over time.

Demonstrated through a case study generating consumer ideas for AI smart glasses, the Virtual Roundtable produced diverse, relevant ideas with clear insights into their evolution. The researchers observed that the cumulative exchange of perspectives across personas cultivated a shared context that progressively deepened the quality of discussion and the ideas produced. This approach suggests that multi-agent AI brainstorming could become a practical tool for product development teams, offering a scalable, diverse alternative to traditional human-only sessions. By simulating the dynamics of human creativity—without the social biases—the system could help organizations rapidly explore innovative concepts before committing resources to development.