Cloud-Based Large Language Models for Enhancing BDIx Agents' Decision-Making in Network Control Emergencies

This paper presents a novel framework for integrating Large Language Models (LLMs) into cloud environments to enhance the decision-making capabilities of Belief-Desire-Intention eXtended (BDIx) agents. By leveraging RESTful services, BDIx agents can evaluate cases using LLMs beyond their predefined plan library, thereby enhancing their belief systems. Our methodology includes the deployment of LLMs on cloud platforms, the development of RESTful APIs, and the modification of BDIx agent frameworks to enable seamless integration. Additionally, transitioning from traditional fuzzy logic to the Adaptive Neuro-Fuzzy Inference System (ANFIS) in the plan library improves the system's adaptability. The effectiveness of this approach is demonstrated through a feedback loop mechanism. This mechanism continuously refines agent behaviour by suggesting priority adjustments to the plan library for cases not previously evaluated based on LLM responses. As a result, the system ensures adaptability and enhances decisionmaking in complex environments.