User Models and User Interfaces

Abstract

In the past several years, it has become increasingly apparent that there is strong disagreement between researchers within and outside of AI on how to build systems for man-machine communication. Within AI, and specifically within natural language, researchers have noted that human speakers and hearers draw on their knowledge about each other when communicating. This knowledge is used both in understanding, and responding to, a speaker’s utterances. User models are a means for representing various types of information about speakers and hearers so that systems are able to reason about their users when interpreting input and producing responses. While there is disagreement in the AI community about the form of user model that should be used, there is an implicit assumption that some form of knowledge about users is essential for successful man-machine communication. In contrast, in the user interface and information retrieval community there has been a reaction against user models. Researchers in this community have argued that with a properly designed interface, users are able to get the information that they need. Modelling a user requires access to and representation of all sorts of ill-defined aspects of human cognition (e.g., how do we know when a user believes that a specific fact is true?). Rather than being so presumptuous as to assume that we can carry out such a task, why not rely on the fact that users, as humans, are smart enough to be able to use interfaces as intended to satisfy their needs? The purpose of the panel is to bring together these disparate opinions in an open forum. Are we really as far apart as we seem ? Are there opinions in the opposing viewpoints that ought be adopted by the other group?