Diverse Evidence, Better Forecasts: Multi-Agent Deliberation Under Information Asymmetry

Researchers have developed a framework for multi-agent AI systems that optimizes how information is distributed among individual models during collaborative forecasting. By adjusting the specific data provided to each participant, this method improves the accuracy and calibration of the group's collective predictions. This approach addresses common inefficiencies in multi-agent deliberation where agents otherwise lack the necessary information to contribute effectively to complex problem-solving tasks.