Agent software is a topic of growing interest to users and developers in the computer industry. Already, agents and wizards help users automate tasks such as editing and searching for information. But just as we expect human assistants to learn as we work with them, we will also come to expect our computer agents to learn from us. This paper explores the idea of an instructible agent that can learn both from examples and from advice. To understand design issues and languages for human-agent communication, we first describe an experiment that simulates the behavior of such an agent. Then we describe some implemented and ongoing instructible agent projects in text and graphic editing, World Wide Web browsing, and virtual reality. Finally, we analyze the trade-offs involved in agent software and argue that instructible agents represent a "sweet spot" in the trade-off between convenience and control.
Note: The Institute of Electrical and Electronics Engineers, Incorporated is distributing this Article with permission of the International Business Machines Corporation (IBM) who is the exclusive owner. The recipient of this Article may not assign, sublicense, lease, rent or otherwise transfer, reproduce, prepare derivative works, publicly display or perform, or distribute the Article.