Engineering for Artificial Intelligence Software
Abstract
Rule based systems are being applied to tasks of increasing responsibility. This report focuses on techniques for the verification and validation of these systems. Conventional software quality assurance depends on the availability of requirements and specification documents. For rule systems, there are generally none because the capabilities of these systems evolve through a development process that is partly experimental in nature. Conventional testing techniques are considered; however, such techniques do not carry over absence of errors. Methods for proving the consistency of rule systems are examined. These methods require that the rules be viewed declaratively, which may be too much of a simplification. A semantics for rule systems based on term rewriting is developed. Standard tests for confluence of term rewriting systems cannot be converted to rule systems, however, because the firing of rules can depend on the absence, as well as the presence, of elements in working memory. Finally, we consider deductive methods for the validation of rule systems.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 01, 1990
- Accession Number
- ADA234793
Entities
People
- John Rushby
- Mark E. Stickel
- Richard J. Waldinger
Organizations
- SRI International