Introduction
Building expert systems often encounters a significant hurdle: experts frequently disagree, often without realizing it. Knowledge is rarely centralized in a single individual; instead, it is distributed across a group of specialists. This distribution introduces "tacit disagreement," where experts use different words for the same concept or the same word for entirely different concepts. If a knowledge engineer aggregates these views without a formal framework, the resulting system becomes a muddled collection of contradictions. This paper introduces a computer-supported methodology using repertory grids and fuzzy logic to map these disagreements, transforming conceptual friction into a structured negotiation for better system design.
Methodology: Repertory Grids and Fuzzy Logic
The authors base their approach on the Repertory Grid Technique, a psychological method where experts identify "entities" (objects in the domain) and "attributes" (the distinctions used to describe them). The methodology follows a three-phase process:
- Domain Discussion: Experts agree on a common set of entities to ensure they are referencing the same objects.
- Conceptualization: Each expert independently creates a grid, rating the entities against their personal attributes.
- Exchange and Compare: Experts exchange grids, rating the common entities using their colleague's attributes.
To analyze these grids, the authors employ Fuzzy Set Theory. Attributes are treated as fuzzy predicates, and similarity scores are calculated using Lukasiewicz’s equivalence operator and Zadeh’s sigma count. This algorithmic approach allows a computer to act as a neutral mediator, identifying conceptual relationships without bias.
The Four-Quadrant Model of Conceptual Dissent
Based on the similarity scores between terminology and distinctions, the relationship between any two experts' concepts is categorized into a four-quadrant model:
- Consensus: Same terminology and same distinctions.
- Correspondence: Different terminology but same distinctions (identifying synonyms).
- Conflict: Same terminology but different distinctions (the most dangerous form of dissent).
- Contrast: Different terminology and different distinctions (highlighting unique areas of expertise).
Case Study and Results
The methodology was demonstrated using a dataset from two experts in geographic mapping techniques. Entities included methods like Kriging and Bicubic splines, while attributes involved distinctions such as "Linear vs. Non-linear interpolation."
The analysis successfully flagged conceptual issues using quantitative thresholds (e.g., >80% for consensus/correspondence and <70% for conflict/contrast). For instance, the term "linear interpolation" was identified as a major point of conflict, whereas "low-level data" and "nominal data" were found to be in correspondence—different labels for the same underlying concept.
Conclusion and Future Directions
By making tacit disagreements overt, this framework shifts knowledge acquisition from simple data extraction to a collaborative negotiation process. While the method scales linearly with the number of experts, it relies heavily on the initial agreement on a common set of entities. Future work involves exploring the cognitive load on experts during the exchange phase and integrating these multi-expert logic findings directly into production environments, such as tagging rules with their source expert to handle dissenting opinions gracefully.
Summary updated