Using Cognitive Artifacts to Bridge the Tool-Agent Divide

A fundamental hurdle in building intelligent systems is in acquiring enough information about a user and her context, in a form that the system can use, without impairing the usability of the system. A variety of interaction paradigms have been explored in this regard – agents, spoken language understanding, the incorporation of biometrics and other interaction modalities – that aim to make interaction more natural for the user. In contrast to these efforts, we have studied how structure can be added to a system to improve the user’s performance in the domain task. In this paper, we demonstrate how this structure can then be leveraged to provide useful intelligent support. Our domain of investigation is a groupware system, and we show how structured support for co-referencing activity can be used by the system to infer user intent. We also provide empirical results demonstrating the effectiveness of an adaptive component that is built on the intent inferencing procedure.