Brad A. Myers
Human Computer Interaction Institute
School of Computer Science
Carnegie Mellon University
Pittsburgh, PA 15213-3891
To be presented at:
AAAI 2007 Spring Symposium on Interaction Challenges for Intelligent Assistants
All intelligent assistants, human or computerized, will make mistakes. Therefore tasks performed by an assistant will generally proceed with some level of oversight and participation by the user. How can we design intelligent assistants with which people will be happy? How can we avoid people turning off our intelligent assistants and just doing the tasks by hand?
This talk will present an informal equation for determining when a user will be happy with an intelligent assistant. This equation includes factors such as the utility of the intelligent assistant, the success rate at performing the user's desired tasks, the difficulty of monitoring and detecting mistakes, and the ease of corrections. One intuition is that this equation is not a simple linear combination of these factors -- instead there seem to be discontinuities, where small changes in the value of these factors have dramatic changes in the user's attitudes. This talk will discuss the factors and present many examples of intelligent assistants to illustrate different values. Some of the examples will come from our work at Carnegie Mellon University on demonstrational interfaces and on the Radar project. Other examples will come from other research and commercial systems.