Semantic Memory

Semantic memory represents generalized knowledge that remains useful independently of a particular previous experience.

It can provide persistent facts, concepts, relationships, preferences and learned propositions for future reasoning and behavior.

Episodic memory represents particular experiences or events. Semantic memory represents generalized knowledge.

An episode can provide evidence from which semantic knowledge is eventually derived, but the resulting knowledge no longer requires retrieval of that particular episode to be useful.

Semantic memory can represent multiple forms of durable knowledge depending on the agent's domain.

A vector database provides storage and similarity retrieval infrastructure.

It can participate in semantic-memory implementations, but the architecture still needs to define what qualifies as durable knowledge and how that knowledge is governed.

RAG retrieves information into model context. The retrieved information can come from semantic memory, episodic memory or external sources.

Semantic memory additionally concerns the representation and lifecycle of durable generalized knowledge.

Durable knowledge can preserve information about its origin and reliability.

This allows the agent to distinguish explicit assertions, external observations, model inferences and generalizations derived from previous experiences.

Semantic memory should be durable but revisable.

Knowledge can become obsolete, contradictory or better supported by newer evidence, requiring revision, replacement, reduced confidence or retirement.

Patterns across episodic memories can support generalized semantic knowledge.

Consolidation allows repeated experiences to influence future behavior without requiring the agent to reconstruct the same conclusion from every individual episode.

Semantic memory is one subsystem within the broader architecture of agent memory.

It provides generalized knowledge while episodic memory preserves particular experiences and working memory maintains information actively relevant to current cognition.

Semantic memory allows learned knowledge to survive individual context windows and execution boundaries.

This contributes to persistent agents whose future behavior can incorporate information learned in previous interactions.

Semantic memory contributes to cognitive continuity when knowledge acquired previously remains available and appropriately influences later cognition.

The knowledge must also remain revisable so continuity does not become persistence of obsolete beliefs.

Semantic memory originates as a concept in cognitive psychology and neuroscience.

In AI-agent architectures, the term is used functionally for persistent generalized knowledge and does not imply human-like understanding or consciousness.