What working memory is
Working memory is the system that keeps information active and manipulable while you use it: holding a phone number while you find a pen, tracking who said what in an argument, keeping a subtotal in mind while adding the next column.
Baddeley's influential model splits it into a phonological loop for verbal material, a visuospatial sketchpad for images and locations, an episodic buffer that binds them, and a central executive that allocates attention. Later accounts emphasise controlled attention over storage, but every model agrees the capacity is small and interference-prone.
The real capacity limit
The famous 'seven plus or minus two' figure came from studies with rehearsal-friendly material. Careful experiments that block rehearsal put the number closer to three or four chunks.
A chunk is a unit of meaning, not a character. 'FBICIANASA' is ten letters and three chunks if you recognise the agencies, ten chunks if you do not. This is why knowledge is such a powerful multiplier on apparent memory ability.
Why it predicts reasoning scores
Complex reasoning requires holding intermediate results while generating and testing hypotheses. On a matrix item, you must keep the candidate rule active while checking it against another row and simultaneously suppressing the rule you already rejected.
Individual differences in that suppression — resisting interference from the discarded hypothesis — track fluid reasoning scores closely. In several factor-analytic studies the correlation between working-memory capacity and fluid intelligence is high enough that researchers argue about whether they are separable at all.
Cognitive load and how to lower it
Instructional research distinguishes load intrinsic to the material, load added by poor presentation, and the productive effort of actually building understanding. Only the middle category is wasted, and it is usually the easiest to remove.
Practical moves: put a diagram next to the text that explains it rather than on the next page; name intermediate results so you can refer to them instead of recomputing; write the sub-goal down. Every element you externalise frees a slot for reasoning.
What degrades it
Sleep deprivation, anxiety, pain, alcohol, task-switching, and background speech all consume capacity. Anxiety is a particularly cruel case: worrying about performance occupies the very resource the task needs, which is one mechanism behind test anxiety.
Interruptions are expensive beyond their duration. Reconstructing the state you were holding takes far longer than the ten seconds someone spent asking their question.
Building around the limit
Since capacity itself resists training, the productive strategies are structural: chunk deliberately by learning patterns until they become single units, offload aggressively to notes and diagrams, and sequence work so only one hard thing is in flight at a time.
Experts do not have bigger workspaces. They have better-organised knowledge, which means each thing they hold carries far more content.