Introduction: The Knowledge Engineering Bottleneck
In the late 1980s, the development of expert systems faced a significant hurdle known as the "knowledge engineering bottleneck." This process required human engineers to manually extract heuristics from domain experts and encode them into software—a labor-intensive and error-prone task. Early automation efforts typically fell into two categories: technique-driven systems, which offered generality but lacked depth, and knowledge-driven systems, which provided deep expertise but were restricted to narrow domains.
Jitu Patel identifies a clear gap in this landscape: the need for a system that is both general enough to work across different domains and powerful enough to elicit deep, expert-level heuristics.
The ASKE Hybrid Approach
To address this trade-off between generality and depth, Patel proposed the Automatic System for Knowledge Engineering (ASKE). ASKE employs a two-stage hybrid methodology to bridge the gap:
- Stage 1 (Technique-driven): The system uses general question-asking strategies to help the expert define an initial "task model."
- Stage 2 (Knowledge-driven): Once the model exists, ASKE uses it to drive a deeper, system-led dialogue to extract specific problem-solving rules.
The core of ASKE's architecture is its Template Hierarchy. Knowledge is categorized into four distinct template types, each serving a specific role in the acquisition process. These templates are linked via isa relations, allowing the system to inherit structures from one domain to help build another.
Evaluation and Results
As a symbolic AI system, ASKE was evaluated through practical application in three distinct domains: Archaeology (interpreting settlement and burial sites), Motor Mechanics (diagnosing engine problems), and Nursing (selecting patient care plans).
The system successfully guided experts through dialogues to identify critical data categories (e.g., "pottery," "fabric," "decoration") and automatically generated complex IF-THEN rules. A key innovation was the use of Reference Templates (RTEMPs). If the system already possessed a structure for interpreting an "archaeological burial site," it could use that as a guided blueprint to help an expert describe a "settlement site," significantly accelerating the knowledge elicitation process.
Conclusion and Limitations
ASKE represents a significant step toward automated knowledge engineering by successfully balancing generality and depth. However, it operates under the assumption that most expertise can be represented as classification hierarchies (following Clancey’s 1985 paradigm). This may limit its applicability in domains requiring complex causal reasoning or temporal logic. Furthermore, while rule generation is automated, human experts are still required to review and edit the output, indicating that the path to fully autonomous knowledge engineering remains a work in progress.
Summary updated