We created the characters Agneta & Frida with the intent to strengthen and encourage exploration of information spaces. In a follow-up study we tried to capture whether users found the characters believable, whether they raised affective responses in users, and whether they created a richer more narratively oriented experience of the space. In order to do so, we had to develop new criteria and methods of understanding users ’ conceptions and affective responses. We discuss the study in detail, as well as the general implications for how to perform user studies and design of character-enhanced systems. 1

ally be sure enough about to take on its own. Otherwise, the continual approval of such actions will decrease the utility of the agent. The agent will eventually have to take autonomous action. To allow an agent to do this, the user must have a certain amount of trust in the agent. The user must also be aware of the agents state so that uncertainty does not undermine this trust. Researchers have just begun to explore methods to represent the agent's state to the user. One can imagine that any representation, ranging through all of the senses, could be used. As with the representation of anything else in the system, the designer should strive for the maximum intuitiveness and the minimum cognitive load. The representation or the representational metaphor will probably be used across the entire system not only to represent state but the agent itself. A very important aspect of this representation is how the user initially appraises the representation. This initial appraisal may shade the

In this paper, we describe user experiences with a system equipped with cognitive vision that interacts with the user in the context of personal assistance in the office. A cognitive vision computer can see the user and user responses and react to situations that happen in the environment, crossing the boundary between the virtual and the physical world. How should such a seeing computer interact with its users? Three different interface styles – a traditional GUI, a cartoon-like embodied agent and a realistic embodied agent – are tested in two tasks where users are actively observed by a (simulated) cognitive vision system. The system assists them in problem solving. Both the non-embodied and the embodied interaction styles offer the user certain advantages and the pros and cons based on the experiment results are discussed in terms of performance, intelligence, trust, comfort, and social presence. Author Keywords Cognitive vision, embodied agent, personal assistant,