Intelligent agents are an emergent research area of particular interest in multimedia systems. Recent proposals --- e.g., [101,288] --- regard the development of agent architectures for the distributed control of complex tasks, in the real-world and in real-time. Operating in the real world means to cope with unexpected events at several levels of abstraction both in time and space. This is one of the main requirements of intelligent agent architectures, and is also the typical scenario for a multimodal interaction in multimedia systems. It is therefore of crucial importance to develop and experiment integrated agent architectures in the framework of multimedia systems.
For an overview on intelligent agents, we suggest to read the paper by Wooldridge and Jennings . It covers theoretical aspects of agents as well as agent architectures and programming languages which have been specifically designed to be used with agents. Another interesting survey of existing agent architectures is , and a first attempt of an agent architecture developed in the project is described in . As stated in , ``the intention is to provide an implementation of a software control architecture which is competent, functionally rich, behaviourally diverse, and which encourages and readily facilitates extensive experimental evaluation.'' experiments involve a set of autonomous agents, each delegated to a precise skill (e.g., input modality), characterized by flexible and modular integration in the different experimental scenarios.
For a multimodal system, several aspects of intelligent agents are especially interesting, e.g. intentional behavior, believe, autonomy, negotiating capabilities, etc. All of them will significantly contribute to an intelligent user interface which is able to interpret the user's actions, decide for itself the best way to present information to the user, and cope with one of several input modalities selected by the user without the need to explicitly tell the computer.
Relations between metaphors and diagrammatic or pictorial representations are also other interesting recent developments [239,124], which can be useful in the design of multimedia systems.
In cognitive musicology, mental models and analogical representations based on metaphors are widespread. They are mostly related to the problem of music imagery. For example, Todd  argues that musical phrasing has its origin in the kinematics and the self-stimulation of (virtual) self-movement. This is grounded on the psycho-physical structure of the human auditory system. Another example comes from robot navigation in 3D space. In this task domain, a bipolar force field is a useful metaphor: the moving robot corresponds to an electric charge, and a target to be reached corresponds to a charge of opposite sign. Obstacles correspond to charges of the same sign. Metaphorical reasoning implies the use of multiple representational levels and environment-based representations.