Learned Skills
Learned Skills represent procedural knowledge acquired through practice, distinct from declarative knowledge (facts) or short-term working memory. In cognitive architectures and AI frameworks, this corresponds to the ability to perform tasks automatically without conscious deliberation.
Key Characteristics
- Procedural Nature: Focuses on how to do something rather than what is known.
- Automation: Repeated execution reduces cognitive load, moving tasks from controlled to automatic processing.
- Persistence: Unlike short-term memory, learned skills are retained long-term, forming part of an agent’s or individual’s core competency.
AI Context: CoALA Framework
In the CoALA framework for AI Agent memory, “Learned Skills” are one of the four critical memory types, analogous to human procedural memory.
- Distinction: Separated from semantic memory (facts) and episodic memory (experiences) to optimize inference speed and task execution.
- Source Analysis: As detailed in AI Agent Memory Types: CoALA Framework Overview, Martin Keen (IBM Technology) categorizes this memory type as essential for agents to execute complex workflows without re-computing basic operations.
- Implementation: Often realized through fine-tuned model weights, reinforcement learning policies, or tool-use proficiency that persists across sessions.
Related Concepts
- Procedural Memory
- short-term-memory
- Semantic Memory
- Episodic Memory
- ai-agent-architecture