Self-Adaptive Large Language Model (LLM)-Based Multiagent Systems

Abstract—In autonomic computing, self-adaptation has been proposed as a fundamental paradigm to manage the complexity of multiagent systems (MASs). This achieved by extending a system with support to monitor and adapt itself to achieve specific concerns of interest. Communication in these systems is key given that in scenarios involving agent interaction, it enhances cooperation and reduces coordination challenges by enabling direct, clear information exchange. However, improving the expressiveness of the interaction communication with MASs is not without challenges. In this sense, the interplay between self-adaptive systems and effective communication is crucial for future MAS advancements. In this paper, we propose the integration of large language models (LLMs) such as GPTbased technologies into multiagent systems. We anchor our methodology on the MAPE-K model, which is renowned for its robust support in monitoring, analyzing, planning, and executing system adaptations in response to dynamic environments. We also present a practical illustration of the proposed approach, in which we implement and assess a basic MAS-based application.

Introduction. In autonomic computing, the development of self-adaptive multiagent systems (MASs) is known to be a complex task [1]. Self-adaptation is a well-known approach used to manage the complexity of these systems as it extends a system with support to monitor and adapt itself to achieve a concern of interest [2]. For example, by adjusting to changing scenarios, these systems can optimize resource allocation or become fault tolerant by expressing high-level goals as utility functions. Communication is key in this regard. Even with basic communication constructs, simple agents can develop robust collective behaviors [3] [4]. Conversely, complex tasks often trigger the emergence of adaptive behaviour, leading to selforganized, collaborative agents. In advanced scenarios involving agent interaction, these communication systems enhance cooperation and reduce coordination challenges by enabling direct, clear information exchange [5]. The interplay of selfadaptive systems and effective communication is crucial for future autonomic MAS advancements.

Discussion / Conclusion. Integrating Large Language Models (LLMs) like GPT-3 or GPT-4 into multiagent systems is a novel and emerging field. The application of such models in this area could potentially revolutionize how agents understand, learn from, and interact with their environment and other agents. The potential of using natural language processing capabilities of LLMs could lead to more sophisticated communication between agents, improved adaptability in dynamic environments, and more robust problem-solving capabilities. Furthermore, LLMs can serve as a common platform for diverse agents to interact, facilitating heterogeneous multi-agent systems. However, this integration also brings up significant challenges, such as the computational overhead of LLMs, the interpretability of their decisions, and ethical considerations. Our approach presents the integration of Large Language Models (LLMs) within multi-agent systems (MASs) to develop self-adaptive agents.