Multimodal embodied agents

Among the goals of research on Autonomous Agents one important aim is to build Believable Interactive Embodied Agents, ones apt to be applied to friendly interfaces in ecommerce, tourist and service query systems, entertainment and as pedagogical Agents. A Believable Agent is one able to show (even, may be, to feel?) emotions and one who has a definite personality. An Interactive Agent has to take into account the particular User and the particular Context where the interaction takes place, and therefore has to make up its own model of the User and the Context, and interact with it by following the rules of face-to face interaction, like turn-taking, back-channel and so forth. An Embodied Agent must be able to interact with the User not only through written text, but in all the modalities a human Agent may use: through words, voice, gesture, gaze, facial expression, body movements, body posture (sometimes, maybe, even through touch?). But it also must be able to conceive, represent and convey all the possible meanings that natural language and multimodal interaction may convey in Humans. The list of capacities required by a Believable Interactive Embodied Agent allows us to sketch the outline of the steps to move in this field of research; some of them have already been moved in recent work, and are fairly represented among papers in this workshop. Research must include three phases. First, a phase of empirical research aimed at finding out the regularities in the mind and behavior of Human Agents, and at constructing models of them. Second, a phase of modelling of Believable Interactive Embodied Agents, where the rules found out are formalized, represented and implemented in the construction of Agents. Third, a phase of evaluation of the implemented Agents, aimed at testing how they fit the User’s needs and how similar they look to a real Human Agent.