Q13Principles of Artificial Intelligence
Question
Explain the process of Knowledge Acquisition in Expert Systems.
Answer
A definitive guide to Knowledge Acquisition in Expert Systems. Details the horrific architectural bottleneck of manually extracting subconscious heuristic logic from human domain experts and violently translating it into strict mathematical IF-THEN rules.
An Expert System (like MYCIN for blood infections) is entirely dependent on its Knowledge Base. The physical hardware is useless without the data. Knowledge Acquisition is the highly aggressive, incredibly difficult architectural process of extracting absolute domain knowledge from a human expert and mathematically encoding it into the machine.
The Feigenbaum Bottleneck
This process is universally recognized as the absolute greatest bottleneck in AI architecture. Human experts (like senior doctors) utilize "Subconscious Heuristics." They look at a patient and "just know" the diagnosis. They mathematically struggle to explain the exact rigid IF-THEN rules they used. The AI engineer (the Knowledge Engineer) must violently force the human to break down their intuition into strict, programmable logic.
The Algorithmic Steps
- 1. Elicitation: The Knowledge Engineer executes highly structured interviews, case-study analyses, and protocol analysis (forcing the expert to think aloud while solving a problem).
- 2. Interpretation: The Engineer mathematically analyzes the interview transcripts to identify the absolute core concepts, objects, and relationships.
- 3. Formalization: The extracted concepts are violently forced into a rigid mathematical Knowledge Representation architecture (e.g., translating "If the patient has a high fever, they might have an infection" into strictly weighted PROLOG rules:
infection(X) :- fever(X, high), CF 0.8). - 4. Verification & Validation: The Expert System is fed test cases. The output is aggressively compared against the human expert's output. If it fails, the rules are violently mathematically tweaked until perfection is achieved.