Short-Term Memory
Short-term memory (STM) is a cognitive system capable of holding a limited amount of information in an active, readily available state for a brief period. Often used interchangeably with working memory, though the latter implies active manipulation of information, STM serves as the immediate buffer for sensory input before it is either discarded or encoded into Long-Term Memory.
Key Characteristics
- Capacity: Typically limited to 7 ± 2 items (Miller’s Law) or 4 chunks (Cowan’s Model).
- Duration: Seconds to minutes without rehearsal.
- Function: Acts as a gateway between Sensory Memory and Long-Term Memory, enabling immediate cognition, decision-making, and conscious processing.
AI Agent Analogy: CoALA Framework
In the context of artificial intelligence, specifically the coala-framework (Context, Operational, Agentic, Long-term), STM parallels the immediate context window or active reasoning state of an agent.
- Integration Point: As detailed in AI Agent Memory Types: CoALA Framework Overview, IBM’s Martin Keen outlines four critical memory types for AI agents, explicitly drawing analogies to human cognitive structures.
- Short-Term Equivalent: In this framework, the immediate context or “scratchpad” used by an LLM during inference functions as the AI’s short-term memory, holding transient data necessary for current task execution before being summarized or stored.
- Distinction: Unlike human STM which decays rapidly, AI short-term memory is bounded by context window limits rather than biological decay, requiring strategic summarization or retrieval mechanisms to extend effective capacity.
Related Concepts
- Working Memory
- Long-Term Memory
- Sensory Memory
- Chunking