What is working memory? A plain-language explanation

A workspace, not a warehouse

Working memory is the small amount of information you can hold and manipulate at the same time: a phone number you are about to dial, the first half of a sentence while you read the second, the running total while you add prices in a shop. It is distinct from long-term memory, which stores knowledge for years and has no practical capacity limit. The useful image is a workbench rather than a warehouse. What matters is not how much you own but how much you can have out in front of you, and what you can do to it while it is there.

Where the term comes from

The term was popularised by Alan Baddeley and Graham Hitch in 1974, who proposed a model with a central attention-control system and separate short-term stores for speech-like and visual-spatial information. Before that, psychology mostly spoke of short-term memory as passive storage. The shift mattered because it moved the emphasis from holding to using: the interesting part is not that you can keep a few items alive for a few seconds, but that you can work on them while doing so.

The components in the classic model

Baddeley's model describes a phonological loop for speech-like material, a visuo-spatial sketchpad for images and locations, and a central executive that allocates attention between them. A fourth component, the episodic buffer, was added later to explain how information from different sources is bound together. You do not need the model to be literally true to find it useful, and Baddeley's own 2012 review is explicit that the architecture remains debated. What the model does well is explain why holding a picture and a spoken phrase at once is easier than holding two spoken phrases.

Why capacity feels so small

Research summarised by Nelson Cowan in 2010 suggests that adults reliably hold about three to five separate items in the focus of attention when they cannot rehearse or group them. That figure is smaller than the famous number seven proposed by George Miller in 1956, and the difference is mostly about method: Miller's estimate came from tasks where grouping was possible, and Cowan's from tasks designed to prevent it. Both describe something real, and the gap between them is one of the most instructive arguments in the field.

Chunking, and why it changes everything

People manage far more than five items in daily life by chunking: a familiar phone number is stored as a few groups rather than ten digits, a chess position as a handful of patterns rather than thirty-two pieces. The limit is on separate, unfamiliar pieces, not on the amount of meaning you can carry. This is why expertise looks like memory from the outside. A specialist is not holding more items than you, they are holding items that are bigger because their long-term knowledge lets them be.

Why it fades so quickly

Items in working memory decay within seconds unless something maintains them, and they are also vulnerable to interference from new material. Both mechanisms are debated as explanations, but their practical consequence is not: a distraction that lasts a few seconds is often enough to lose what you were holding. This is why you can walk into a room and forget why, and why an interruption mid-calculation costs the whole calculation rather than a part of it.

How it is measured

Common laboratory tasks include digit span, in which you repeat a growing list of numbers; complex span tasks, which alternate remembering with a processing step such as judging a sentence; and n-back tasks, in which each new item must be compared with the one presented N steps earlier. Each loads working memory differently. Scores from one are not interchangeable with scores from another, and none of them is an intelligence test, though some correlate with reasoning measures more strongly than others.

The link with reasoning

Complex span measures correlate with tests of reasoning more strongly than simple span does, and Conway and colleagues reviewed that relationship in 2003. A correlation is not a mechanism, and it certainly is not a promise that raising one raises the other. It does explain why working memory attracts so much attention: it is one of the few laboratory measures that reliably relates to how people perform on tasks nobody would call a memory task.

What a dual n-back task asks of it

Dual n-back keeps two streams in play at once, a position and a sound, and demands continuous updating: each turn you must add the newest item, drop the oldest and compare the current one with the item from N turns ago. Nothing is stored for you and nothing stays still. This is why the task feels demanding at N=2 and above in a way that memorising a list does not. It loads maintenance, updating and attention control at the same time.

Storage tasks versus updating tasks

The difference matters more than most explanations admit. Digit span asks you to hold a list that has already finished. N-back asks you to hold a list that keeps being rewritten while you answer questions about it. The first is closer to remembering a shopping list; the second is closer to following a conversation in which the topic keeps shifting. They are related but they are not the same thing, and improving one does not automatically improve the other.

What working memory is not

It is not intelligence, though the two correlate. It is not long-term memory, and being good at one says little about the other. It is not concentration, though attention control is part of it. And it is not a fixed quantity you can read off a single test, because every measure of it is also measuring the particular task used. Treating a score from any one task as your working memory capacity is the most common overreach in this area.

What changes it, and what does not

Capacity varies with age, following a rise through childhood and a slow decline in later adulthood. It varies within a person from hour to hour with sleep, stress, illness and distraction, and those swings are large enough to swamp most other effects. What has not been reliably shown is that training a working-memory task raises capacity in general. The training literature finds clear improvement on the trained task and little that travels, which the companion guide on the evidence covers in detail.

Why the limit is probably useful

A workspace that held everything would be a workspace in which nothing stood out. Part of what attention does is decide what deserves to be in front of you right now, and a small capacity forces that decision constantly. Seen that way, the limit is not a design flaw to be trained away but the mechanism by which anything gets prioritised at all. That framing also explains why the most reliable way to handle a demanding task is to reduce what it asks you to hold, rather than to try to hold more.

What this means in practice

If you want to work around the limit rather than fight it, three things help more than any exercise. Write down intermediate results instead of carrying them. Group information into meaningful units before you need to hold it. And protect the seconds that matter from interruption, because a few seconds of distraction costs the whole contents rather than a fraction. None of that is glamorous, and all of it works reliably.

What this means for training

Practising an n-back task reliably improves performance on that task and on very similar ones. Whether it changes working memory in general, or reasoning, is an open question in the research literature, and the honest answer today is that well-controlled studies have mostly failed to find it. Dual N-Back shows your accuracy per session so that you can follow your own progress on the exercise itself, which is the part the evidence actually supports.

Further reading

• Baddeley & Hitch (1974), Working Memory, Psychology of Learning and Motivation
• Miller (1956), The magical number seven, plus or minus two, Psychological Review
• Baddeley (1992), Working Memory, Science
• Conway et al. (2003), Working memory capacity and its relation to general intelligence, Trends in Cognitive Sciences
• Cowan (2010), The Magical Mystery Four: How is working memory capacity limited, and why?, Current Directions in Psychological Science
• Baddeley (2012), Working Memory: Theories, Models, and Controversies, Annual Review of Psychology