Introduction
In the mid-1980s, the configuration of complex hardware systems emerged as a significant challenge for artificial intelligence. While early expert systems like Digital Equipment Corporation’s XCON demonstrated the feasibility of automated configuration, they often suffered from maintenance difficulties due to their reliance on vast, unstructured rule bases. This paper, originally presented at the 1986 AAAI proceedings, proposes a hybrid knowledge representation technique that marries the structural advantages of frame-based systems with the procedural flexibility of production rules.
The Configuration Problem
The primary difficulty in hardware configuration lies in balancing structural representation with judgmental reasoning. Purely rule-based systems lack inherent structure; a change in a single component often necessitates updating numerous scattered rules, leading to a "spaghetti code" effect. Conversely, frame-based systems excel at representing hierarchical relationships and inheritance but struggle with the "fuzzy logic" required to guide users through complex decision-making processes. The author identifies this as the "Expert Gap," where a system must not only enforce technical constraints but also act as a consultant for non-expert users.
Proposed Methodology: A Hybrid Approach
The proposed solution utilizes a representation language based on the Frame Representation Language (FRL), augmented with production rules that function as "demons."
Frames for Structural Knowledge
In this system, every hardware component is represented as a frame. Slots within these frames define attributes (such as price or performance) and relationships (such as "IS-A" links). This hierarchical structure enables inheritance, allowing specific hardware models to automatically inherit properties from broader categories.
Rules as Demons
Rather than maintaining a global rule pool, production rules are attached directly to specific slots within frames. These "demons" are triggered by specific events:
- IF-NEEDED demons: Employed via backward chaining when the system requires a value that is not currently present (e.g., determining which PC model to recommend based on user needs).
- IF-ADDED/REMOVED demons: Employed via forward chaining to manage side effects and maintain consistency when data is modified.
To handle the inherent ambiguity of user requirements, the system incorporates certainty factors, a technique borrowed from the MYCIN medical expert system.
Implementation and Knowledge Base
The methodology was demonstrated using a manually curated knowledge base focused on the IBM Personal Computer family. This domain provides a clear example of how hierarchical frames can organize hardware specifications while rules manage the constraints of specific configurations.
As shown in Figure 1, the "PC System" serves as a parent frame for specific models like the IBM PC, XT, and AT. The diamond-shaped node (RB1) represents a rule block or demon associated with these components, illustrating how procedural logic is scoped to specific structural elements.
Results and Discussion
The hybrid architecture offers several qualitative advantages over traditional approaches:
- Centralization and Maintainability: Partitioning knowledge into logical sections makes the system easier to update as hardware specifications evolve. All information about a specific component is located in one place.
- Hybrid Reasoning: The integration of declarative knowledge (frames) and procedural knowledge (rules) allows for more sophisticated reasoning that mirrors human expertise.
- Consultative Logic: By scoping rules to specific attributes, the system can provide more targeted guidance to users, effectively bridging the "Expert Gap."
At the time of publication, prototype systems were being developed at ROLM Corporation to validate these concepts. Preliminary findings suggested that the hybrid approach would be significantly easier to develop and maintain than pure rule-based commercial shells.
Conclusion and Future Work
This 1986 study provided a foundational framework for modern object-oriented expert systems. While the author acknowledged potential scaling issues as knowledge bases reach extreme sizes, the modularity provided by the frame-rule hybrid remains a powerful tool for managing complex configuration tasks. Future work was suggested to explore the applicability of this technique in domains beyond hardware configuration, though the immediate focus remained on refining the prototype implementations.
Summary updated