Multi-Agent Systems (MAS)

Multi-Agent Systems (MAS) are increasingly deployed in critical environments such as healthcare, autonomous transportation, and industrial automation. However, the complexity and dynamic nature of MAS make failure attribution challenging, especially when relying solely on correlation-based methods. This article proposes a novel framework based on causal inference to automate failure attribution through three sequential stages: Abduct, Act, and Predict. By combining causal discovery, intervention mechanisms, and predictive modeling, our approach enhances explainability, robustness, and reliability in decision-making for MAS. We illustrate the framework with simulated scenarios in healthcare robotics and autonomous logistics, demonstrating improved accuracy in failure attribution and prevention compared to traditional methods.