Personalization in multi-agent systems

In this paper we study the aspects of the human-MAS interaction found in organizational applications based on MAS. The increasing number of MAS applications inside organizations introduces a basic problem: how should the user interact with such systems? The problem involves many questions that influence the proper architecture of the MAS and must not only be considered as a fine detail, but should be taken into account from the beginning in the MAS development. Among the most important questions we can stress the interaction modes (multimodal, based on natural language, voice-based, point-and-click interfaces) and the personalization or adaptability of the interaction and the services. From such questions and using our experience in the development of MAS applications for organizations, we present in this article a MAS architecture based on the concept of digital butler. In this architecture, the user interacts with a single assistant agent that can in turn interact with other agents in the system. We use an adaptive assistant agent able to personalize the interaction with the user as well as a set of other adaptive agents able to personalize the execution of the services. Beyond the previous questions, such a distributed architecture offers solutions for the computational aspects found in a MAS, like the complexity and the use of resources.