Working Memory
The small pool of information the mind actively holds and manipulates, and the classic multicomponent model of how it works.
Working memory is the system that holds and manipulates a small amount of information for ongoing tasks: keeping a phone number in mind while dialing, tracking the thread of a conversation, or carrying intermediate results while solving a problem. It is distinct from long-term memory, which stores vast amounts of information for years, and from the brief sensory registers that hold raw input for fractions of a second. Working memory is severely limited in capacity, and that limitation shapes much of human cognition.
The scientific study of its limits began with George Miller's 1956 paper "The Magical Number Seven, Plus or Minus Two," which showed that people can hold roughly seven items in immediate memory — though the items could be chunks that package several elements, such as a familiar acronym or a chess position. The modern landmark is the multicomponent model of Alan Baddeley and Graham Hitch (1974), which replaced the older idea of a single short-term store. It proposes a phonological loop for verbal material (an inner voice plus an inner ear, evidenced by the word-length effect and articulatory suppression), a visuospatial sketchpad for visual and spatial material, a central executive that allocates attention and coordinates the subsystems, and, added in 2000, an episodic buffer that binds information from the subsystems with long-term memory into integrated episodes.
The capacity question was refined by Nelson Cowan, who argued that the true limit is about four chunks when rehearsal and chunking are controlled. Individual differences in working memory capacity are substantial and predict performance on reasoning and fluid-intelligence tests, making capacity one of the most studied individual differences in cognitive psychology. Capacity is typically measured with digit span, operation span (recalling items while solving arithmetic), and the N-back task, in which participants must detect items that repeat N positions back.
Neuroscientifically, working memory engages a frontoparietal network, with the dorsolateral prefrontal cortex central to maintenance and manipulation; the classic finding of persistent activity during the delay period of delayed-response tasks was recorded in monkeys by Fuster and by Goldman-Rakic. Neuroimaging confirms the same regions in humans.
Working memory has direct applications: instructional design and cognitive load theory use it to explain why worked examples and split-attention formats aid learning, and deficits are implicated in conditions such as ADHD. As a concept it remains a working hypothesis — the exact structure of the system is still debated — but the constraints it describes are among the most replicated in psychology.