Q11Principles of Artificial Intelligence

A Knowledge Representation (KR) system (like First-Order Logic or Semantic Networks) is the absolute foundational architecture that allows an AI to mathematically store and manipulate facts about the universe. To be viable, a KR system must violently enforce four strict mathematical properties.

1. Representational Adequacy

The absolute most critical property. The architecture must possess the mathematical vocabulary to represent EVERY SINGLE fact, relationship, and constraint that exists in the specific problem domain. If the AI is diagnosing cancer, the KR must be able to mathematically represent complex biological probabilities and time-series mutations. Propositional Logic often fails here because it cannot mathematically represent "All humans are mortal" without explicitly listing every human.

2. Inferential Adequacy

Storing data is useless if the AI cannot generate new knowledge. The system must possess strict algorithmic inference rules (like Modus Ponens or Resolution). It must be able to mathematically violently combine existing facts in the Knowledge Base to deduce absolute, previously unknown truths.

3. Inferential Efficiency

The inference algorithms cannot run until the heat death of the universe. The KR must be structured to allow highly aggressive, lightning-fast heuristic searches. This is why Horn Clauses are used in PROLOG; they severely restrict Representational Adequacy but provide catastrophic, mathematically explosive Inferential Efficiency.

4. Acquisitional Efficiency

The system must easily accept new data. If adding a single new fact requires an engineer to manually rewrite 5,000 lines of existing logical constraints, the architecture is a failure. It must support automated mathematical updates (like Machine Learning weights or automated ontological merging).