Definition
An active symbol, in Douglas Hofstadter's Gödel, Escher, Bach, is a self-sustaining pattern of brain activity that represents something and can in turn trigger other such patterns. Unlike a static label or a single neuron, an active symbol is a dynamic distributed pattern — a region of brain activity that becomes self-reinforcing when activated and that primes related symbols to fire next.
Your "kitchen" symbol is the pattern that wakes up when you think kitchen-thoughts, hear the word, walk in, smell coffee. It includes connections to "fridge," "breakfast," and a long tail of associations. When active, it primes its neighbors.
Why it matters
How it works
Active symbols compete for activation in the brain's symbol-level economy. Each receives input from perception, action, and other symbols; each integrates these inputs and either becomes active (if the integration exceeds threshold) or stays dormant. Active symbols modulate which other symbols are likely to become active next, producing trains of thought that flow along contextually weighted paths.
Hofstadter's later research program — the FARG group at Indiana University, the Copycat and Tabletop systems — attempted to build computational architectures based on active symbols. The systems used parallel codelets competing to build representations in a workspace, with concept neighborhoods (slipnets) that allowed symbols to mutate into neighbors under contextual pressure. The work was intellectually rich but eclipsed commercially by the deep-learning approach, which arguably achieves something like active symbols through different machinery.