Towards the Knowledge Level in SOAR: The Role of the Architecture in the Use of Knowledge
Abstract
Soar has been described as an architecture for a system that is to be capable of general intelligence. One way to specify what this might mean is to define general intelligence as the ability to approximate an ideal knowledge level system across a sufficiently broad set of goals and knowledge. In this chapter we use this definition as the basis for evaluating the degree to which Soar achieves general intelligence. A complete evaluation is beyond the scope of this chapter, so we focus more narrowly on how the Soar architecture supports and constrains the representation, storage, retrieval, use and acquisition of three pervasive forms of knowledge: procedural, episodic, and declarative knowledge. The analysis reveals that Soar adequately supports procedural knowledge - to some extent it was designed for this - but that there are still significant questions about episodic and declarative knowledge. These questions arise primarily because of consequences of the principle source of constraint in Soar, the fact that all learning occurs via chunking. New results are also presented on the acquisition of declarative knowledge.
Document Details
- Document Type
- Technical Report
- Publication Date
- Aug 07, 1989
- Accession Number
- ADA218926
Entities
People
- Allen Newell
- John E. Laird
- Paul Simon Rosenbloom
Organizations
- Carnegie Mellon University