Formalization of Tradeoff Rules and Other Techniques for Comprehending Complex Rule-Based Models
Abstract
The authors and colleagues have recently developed a large and complex knowledge-based simulation that includes political-military decision models, each with thousands of qualitative rules. As part of our knowledge architecture, we emphasized rule hierarchies which require recursive combining rules that evaluate a given qualitative variable as a function of several lower- level variables -- often in ways that involve value-laden tradeoffs. Recently, we began to formalize procedures for defining the qualitative variables and characterizing the tradeoff relationships in algebra-like terms. This have proven valuable in speeding model development and communicating results. It has also improved the quality of rule-writing and made it easier to differentiate among and deal intuitively with different types of combination rules -- many of which are quite different from the standard weighted-sum approach. This paper begins with examples, notes that the problems involved are generic rather than domain specific, and then illustrates an approach for dealing with them. In the longer run, there should be implications for rule algebras in formal modelling and new syntaxes in programming languages. In essence, the objective should be to state combining rules at a high level of abstraction. (edc)
Document Details
- Document Type
- Technical Report
- Publication Date
- Nov 01, 1987
- Accession Number
- ADA216994
Entities
People
- Paul K. Davis
- Robert Weissler
- Steven C. Bankes
Organizations
- RAND Corporation