Intelligent Interface Agents

This paper deenes my depth area for the AI qualifying exam. In the next paragraphs I will give an intensional deenition. Following that will be a roughly-grouped list of papers which form a corresponding extensional deenition for the area. The goal of building an intelligent agent is almost synonymous with the goal of building an artiicial intelligence. This depth area scopes that to managable size by surveying systems which attempt to incorporate agents directly in service of a user's needs. This excludes much of the distributed AI (DAI) literature, where agents are used to make a problem more manageable by breaking it up into pieces. This area focuses on whatever AI components of complete systems which comprise agents, rather than focusing on speciic AI technologies and all their possible extensions. For example, machine learning is an important component in many of the papers listed below, but never is PAC learning mentioned. Finally, the emphasis on user interface has brought in some work from outside of the traditional realm of AI (e.g. from ACM TOIS). It is important to examine this work for two reasons. First, it is often work done by people from outside the AI community, with diierent (typically more pragmatic) ideas about what agents are and can be { we can learn from them. Second, it is sometimes the case that people make claims about AI outside of AI that would not be acceptable in AI journals (e.g. we expect a full speech recognition interface for this in the next release) { we should correct them.