Definition
Episodic memory is a storage mechanism in AI agents that captures and retrieves discrete past interactions, observations, and state transitions to inform current decision-making. It enables agents to maintain continuity and perform few-shot learning from historical experiences, though it introduces trade-offs between increased personalization and higher retrieval latency or context window saturation.
Focuses on 'what happened' in specific sessions rather than general facts (Semantic Memory) or hardcoded skills (Procedural Memory).
"A flight data recorder (Black Box) that the pilot can query mid-flight to review and learn from previous maneuvers."
- Semantic Memory(Contrast (General knowledge vs. event-specific knowledge))
- Context Window(Prerequisite (The finite space where retrieved episodes are processed))
- Reflection / Self-Criticism(Component (The mechanism used to distill episodes into insights))
- Vector Database(Component (The physical storage layer for episodic embeddings))
Conceptual Overview
Episodic memory is a storage mechanism in AI agents that captures and retrieves discrete past interactions, observations, and state transitions to inform current decision-making. It enables agents to maintain continuity and perform few-shot learning from historical experiences, though it introduces trade-offs between increased personalization and higher retrieval latency or context window saturation.
Disambiguation
Focuses on 'what happened' in specific sessions rather than general facts (Semantic Memory) or hardcoded skills (Procedural Memory).
Visual Analog
A flight data recorder (Black Box) that the pilot can query mid-flight to review and learn from previous maneuvers.