How Probability Theory Can Help Us Design Rule-Based Systems
Abstract
This paper is concerned with finding improved methods of reasoning for use in rule-based systems that perform diagnosis in general and situation assessment is particular. It is argued that, because the rules used in rule-based systems typically have exceptions, the rules must be interpreted probabilistically. Thus, if a rule If A then B has exceptions, then what the rule really means is that the conditional probability of B given A is close to one. A rule-inference criterion that makes use of second-order probability concepts is advocated. Interestingly, this inference criterion is equivalent to some non-probabilistic inference criteria. The paper expounds a scheme for constructing situation-assessment systems that could be used as either decision aids or as reasoning components of computer generated forces.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 01, 1998
- Accession Number
- ADA364392
Entities
People
- Donald Bamber