Introduction
The development of expert systems is frequently hindered by the "knowledge acquisition bottleneck"—the difficulty of extracting specialized knowledge from human experts and encoding it into machine-executable formats. Historically, developers faced a binary choice: use general-purpose tools that lacked domain specificity or invest significant resources into building custom tools from scratch.
This paper introduces DOTS (Domain-Oriented Tool Support), a metatool designed to bridge this gap. Unlike previous systems like PROTEGE, which were "method-oriented" and tied to specific problem-solving strategies, DOTS adopts an "architecture-oriented" approach. By defining a high-level architectural blueprint, developers can rapidly generate custom-tailored knowledge-acquisition tools without manual coding.
The Abstract-Architecture View
The core innovation of DOTS is the Abstract-Architecture View, which decomposes any knowledge-acquisition tool into four fundamental building blocks:
- Knowledge Editors: These provide the user interface (e.g., forms, graphs, or menus) through which the expert interacts with the system.
- Knowledge Modules: These serve as internal data structures that store the acquired information in a neutral format.
- Update Rules: These act as the mapping mechanism, ensuring that data entered into an editor is correctly routed to the appropriate slot within a knowledge module.
- Transformation Rules: These instructions convert the stored knowledge from the internal modules into the final representation required by the target AI system, such as production rules or frame-based instances.
By decoupling the interface (Editors) from the final output (Transformation Rules), DOTS allows for a high degree of flexibility. A developer can modify the target AI system's format without needing to redesign the expert-facing interface.
Implementation and Evaluation
DOTS allows developers to visually arrange these architectural components to generate operational software. To evaluate its effectiveness, the author applied DOTS to the Sisyphus VT task, a benchmark involving the configuration of elevator systems. This task requires capturing complex constraints and "fixes" for design violations.
The results were significant: a functional tool for the VT task was generated in just nine hours. This tool performed comparably to SALT, a well-known hand-crafted knowledge-acquisition tool specifically built for the VT domain. The speed of development highlights the efficiency of the metatool approach.
Beyond the VT task, the principles behind DOTS were used to create ALFA, a real-world tool for troubleshooting DNA sequencing machines, further proving the framework's versatility across different domains.
Key Insights
The shift from method-oriented to architecture-oriented metatools provides several advantages:
- Method Independence: DOTS is not restricted to specific problem-solving strategies like "propose-and-revise." It can support any strategy that can be modeled using its four architectural components.
- Separation of Concerns: The clear division between UI, storage, and transformation simplifies maintenance and allows for the reuse of components across different projects.
- Rapid Prototyping: The ability to generate a complex tool in hours rather than weeks or months drastically reduces the cost of building custom expert systems.
Limitations and Future Work
While DOTS is highly effective for graphical and form-based tools, it has limitations. It currently does not support interview-style interfaces that rely on natural language processing or textual dialogs. Furthermore, it is constrained to tools that fit its specific architectural paradigm.
Future research aims to develop DASH, a successor metatool. DASH is intended to allow developers to design knowledge-acquisition tools directly from domain ontologies (class definitions), further automating the link between domain modeling and tool generation.
Summary updated