History

The evolution of AI knowledge representation reveals a fascinating shift from rigid, rule-based systems to dynamic, context-aware frameworks. In the 1960s, early systems like GPS (General Problem Solver) relied on heuristic search and problem decomposition, treating knowledge as procedural rules. By the 1970s, Minsky’s frames introduced a structured way to represent stereotyped situations, enabling agents to handle common scenarios with predefined slots and default values. For example, a medical diagnosis agent using frames might have slots for symptoms, treatments, and contraindications, allowing it to infer missing information from context. 

The 1980s and 1990s saw the rise of hybrid systems, combining rules (e.g., expert systems like MYCIN) with frames to create more adaptable agents. MYCIN, for instance, used backward-chaining rules to diagnose infections while leveraging frames to organize patient data hierarchically. However, these systems struggled with scalability and maintenance, as rule bases became unwieldy. 

The Semantic Web era (2000s–present) introduced ontologies (e.g., OWL, RDF) and classifiers (e.g., support vector machines, neural networks) to represent knowledge as machine-readable, interoperable graphs. Modern agents like IBM Watson or autonomous vehicles use ontologies to integrate diverse data sources (e.g., medical records, traffic patterns) and classifiers to infer context. For example, an autonomous car’s agent might use an ontology to link road signs, weather conditions, and traffic laws, while a classifier predicts pedestrian intent. The trade-off here is between interpretability (rules/frames) and flexibility (ontologies/classifiers), with modern systems often combining both for robustness.