Interactive Intelligent Assistants

Dialogue-based communication: We believe that the most effective human-computer interaction is based on principles of natural language dialogue between people. Why should this be? First, it is clear that this is how people interact. If we want the computer to accomodate the person rather than the person learning how to accomodate the computer, this is is where we have to start. Second however, it turns out that thinking of the interaction as a dialogue provides many benefits that are missed in simpler forms of human-computer interaction. Dialogue provides a shared context for interpreting what the other participant is saying or doing. This permits much more efficient communication, since the shared context can be used to fill in unmentioned details. The dialogue also provides the structure for such conversational activities as asking questions, revisiting a topic or decision, deciding whether to say or do something, and so on.

Collaborative Problem-Solving: We have been concentrating on a particular type of intelligent, interactive assistant, namely one that helps the manager solve some kind of problem. This is not very restrictive, since most things that you can think of needing a computer's help with involve a problem that needs to be solved. Finding information, assessing a situation, developing a course of action, achieving a goal, comparing alternatives, and the like are all examples of ``problem-solving'' tasks in this sense.

Robustness

Mixed-Initiative Interaction

A Note About ``Agents'': The term ``agent'' has a certain currency, and one might ask why this project is not about Intelligent Interactive Agents. In fact, it is, in one sense. We would define an assistant as a cooperative agent. However, note that many things billed as ``agents'' are simply what would have simply been called ``programs'' not so long ago. They are told to do something and they do it, sometimes more intelligently than other times, sometimes doing exciting things like remote data access and distributed processing. But to us, the defining characteristic of an agent as opposed to a program is that agents have non-trivial goals and intentions and, typically, they need to reason about these in determining what to do. The systems we are building in the Intelligent Interactive Assistants Project most definitely have goals and intentions, and in fact also represent the goals and intentions of their managers as well, and reason about all of these in deciding what to do.