Task Learning by Instruction: Benefits and Challenges for Intelligent Interactive Systems

Jim Blythe, Prateek Tandon, Mandar Tillu

The CALO desktop assistant aims to provide assistance through many AI technologies, including several techniques for learning to perform tasks. Based on our experiences implementing Tailor, a tool for task learning by instruction (TLI) in Calo, we explore the requirements for integrating TLI more closely into the assistant. The benefits of integration include a more coherent user experience and more powerful combined approaches for learning and interacting with a task-based assistant. We consider the TLI component both as one of the components for communicating with the user about tasks, and as a component for learning procedure knowledge. Components in both these groups need to share significant capabilities for task generation and recognition and for processing information from the user about tasks to be learned. We discuss strategies for invoking tasks by name that can also be used in task learning, enabling an integration of task learning by instruction and by demonstration.

Subjects: 2. Architectures; 10. Knowledge Acquisition

Submitted: Jan 29, 2007

This page is copyrighted by AAAI. All rights reserved. Your use of this site constitutes acceptance of all of AAAI's terms and conditions and privacy policy.