Engram
In neuroscience and psychology, an engram refers to the hypothetical physical or chemical trace in the brain that represents a specific memory. The concept was introduced by German biologist Richard Semon in 1904 to describe the neural substrate of memory. Semon proposed that sensory experiences leave persistent marks on neural tissue, which could later be reactivated to produce memory recall. Although his specific mechanistic proposals were not fully validated, the term established the foundational idea that memories are stored as tangible biological changes rather than abstract entities.
Modern research has sought to identify the specific cellular and molecular mechanisms underlying engrams. Scientists generally agree that engrams involve long-term potentiation, synaptic plasticity, and the formation of new neural connections. Techniques such as optogenetics have allowed researchers to label and manipulate specific engram cells, providing evidence that distinct populations of neurons encode specific memories. This shift from a purely theoretical construct to a biologically identifiable phenomenon has significantly advanced the understanding of how learning and memory function at the neural level.
In the context of artificial intelligence, the term “engram” has been adopted by DeepSeek to describe its context-aware knowledge retrieval system for language models. This usage draws an analogy to the biological concept, suggesting a mechanism for storing and retrieving information efficiently within digital architectures. While the computational implementation differs fundamentally from neural biology, the terminology highlights the system’s role in maintaining persistent, accessible states of knowledge that inform subsequent processing and response generation.
Source Notes
- 2026-04-14: “But OpenClaw is expensive…”
- 2026-04-07: DeepSeek Engram Solving LLM Inefficiency Through Context Aware · ▶ source