Introduction
In university laboratory settings, "Practical Works" (P.W.) sessions are often limited by the availability of the instructor. Students frequently encounter hurdles such as theoretical misunderstandings, assembly errors, or equipment failures. While AI has long been applied to troubleshooting, many existing tutoring systems suffer from two major flaws: the "Magical Expert" problem, where the system provides answers without explaining the underlying methodology, and "Maintenance Rigidity," where the system is difficult for non-AI experts to update or adapt to new domains.
The SIAM system was developed to address these issues, serving as a "computerized complement" to the teacher. It is designed to be both pedagogically effective—focusing on the cognitive process of diagnosis—and easily maintainable by instructors without deep AI expertise.
The SIAM Architecture
The core innovation of SIAM is the strict separation of descriptive knowledge (the "what") from expert diagnostic logic (the "how"). This modularity allows the system to remain generic and adaptable across various scientific fields, such as electronics, optics, and electrotechnology.
As illustrated in the architecture, the system is organized into several key components:
- Models and Descriptives: These include "descriptive sheets" for materials, circuits, and domains. Teachers can edit these frames to define attributes like Directions-for-Use and Physical-description.
- Knowledge and Metaknowledge: This layer contains generic production rules and metarules. These rules do not reference specific devices like "oscilloscopes"; instead, they reason about any device possessing the attributes defined in the descriptive sheets.
- Inference Engine and Interfaces: The system uses an inference engine to apply diagnostic logic to the current "Context of Work," communicating with students via a natural language interface or a menu-based system.
Knowledge Representation and Modeling
To facilitate troubleshooting, SIAM models the experimental setup as a set of interconnected components. For example, in an electronics lab, a relaxation oscillator is represented not just as a schematic, but as a structured set of nodes and links that the system can analyze.
By using this representation, the expert modules—such as Problem Analysis, Signal Control, and Circuit Checking—can systematically drill down into technical checks. The goal is to demystify the "magical" intuition of a human teacher by showing the student the explicit cognitive steps required to identify a fault.
Evaluation and Results
The SIAM system was evaluated over the 1988-1989 academic year at the Institut Supérieur d’Electronique de Paris. The authors developed a "Bibliotheca of descriptives" covering three scientific domains, 50 material descriptions, and 40 specific lab scenarios.
The evaluation highlighted several key successes:
- Pedagogical Impact: Students were able to grasp the expert's cognitive process, helping them approach future problems more independently.
- System Genericity: The modular architecture proved successful in handling real-world failures across distinct scientific departments.
- Authoring Efficiency: Instructors were able to describe new tutorial subjects in approximately one hour, provided the basic materials were already in the library.
Conclusion and Future Work
SIAM demonstrates that a generic, modular architecture can effectively bridge the gap between expert diagnostic logic and classroom instruction. While the current system relies on manual student input and faces some interface challenges, it provides a robust framework for teaching troubleshooting skills. Future developments may include direct hardware sensing and more sophisticated student modeling to further personalize the learning experience.
Summary updated