Goal Reconstruction: How Teton Blends Situated Action and Planned Action
Abstract
This chapter discusses an extra capability that few cognitive architectures have, even though it is both useful from a programming point of view and arguably a good approximation to a human capabilities. People can reconstruct goal structures and other aspects of their internal state that have been forgotten. For instance, suppose one is interrupted in the middle of solving a difficult problem by a long involved phone call. When the phone call is over, one can eventually pick up the problem solving where one left off. This capability is called goal reconstruction. Because goal reconstruction requires no special training to acquire it and it does not have to be acquired separately for each new problem solving procedure one learns, goal reconstruction is arguably a fundamental, task-general capability of human problem solvers. Goal reconstruction is also a useful capability even for an artificial problem solver. It permits recovery from interruptions of the problem solving by processes that modify the body of procedural knowledge, such as an inferential learning process or a programmer debugging the procedural knowledge. In short, goal reconstruction is both a fundamental human capability and a useful capability for AI architectures. Keywords: Cognitive modelling; Reactive planning; Problem solving; Artificial intelligence.
Document Details
- Document Type
- Technical Report
- Publication Date
- Nov 03, 1989
- Accession Number
- ADA222362
Entities
People
- Kurt VanLehn
- William Ball
Organizations
- Carnegie Mellon University