Working memory: how much you can hold and how to make room

Holding information and using it compete for room

Imagine following a recipe while someone changes the number of people coming to dinner. You need to retain the original quantities, adjust them and remember which ingredients have already gone into the bowl. Knowing how to multiply will help you here. Keeping the changing situation straight places another demand on your attention, which is why a familiar calculation can become awkward midway through a busy task.

Alan Baddeley and Graham Hitch's 1974 chapter, Working Memory, argued for a system that supports ongoing thought as well as temporary storage. Their model distinguished a coordinating component from stores supporting verbal and visual or spatial information. The account has developed since its publication. Its useful lesson for ordinary work is that remembering a sequence and reasoning about that sequence draw on resources that have limits.

A written recipe helps because its quantities remain available when your attention moves elsewhere. The same principle applies to a train connection, a spreadsheet calculation or an argument whose conclusion depends on several earlier claims. When the task becomes tangled, identify what you are trying to retain alongside what you are trying to work out, because that distinction suggests where a note or diagram could hold quantities steady while you decide how to change them.

Capacity depends on what counts as an item

George Miller's 1956 paper reviewed limits in immediate memory and the judgement of sensory information. Its memorable title has often become a supposed rule for everything from menus to presentations. Miller discussed how recoding information into chunks changes what can be recalled, and he warned against treating different experimental limits as one underlying process. A universal allowance of seven thoughts is unwarranted.

Nelson Cowan's 2001 review argued for a central storage limit of about four chunks when rehearsal and other ways of extending recall were controlled. Four chunks is an approximate proposal drawn from several kinds of experiment, rather than the average from a single participant sample, so it should be treated as a model of limited storage under particular conditions. Experimental conditions and individual differences still matter. A familiar word may function as a chunk, whereas an unfamiliar sequence of letters may require several separately maintained pieces.

Try reading the digits in the figure as separate characters, then reading the grouped version as familiar hundreds. Both rows contain the same twelve digits. Grouping changes the units you are trying to keep available, although this particular sequence also follows a predictable counting pattern that would help recall in its own right. The drawing illustrates a possible representation, without measuring how many items any reader will remember.

Schematic, not data: the same twelve digits appear as twelve separate items or four familiar chunks containing three digits each.

Illustrative representations of the same digit sequence
Representation	Sequence	Units represented	Digits per unit	Total digits
Separate digits	1, 0, 0, 2, 0, 0, 3, 0, 0, 4, 0, 0	12 items	1 digit	12 digits
Familiar number groups	100, 200, 300, 400	4 chunks	3 digits	12 digits

Chunks grow from knowledge you already have

In William Chase and Herbert Simon's 1973 study, Perception in Chess, three players ranging from beginner to master were asked to reconstruct chess positions from memory. When the positions came from real games, recall rose with playing strength, whereas for randomly arranged pieces that advantage largely disappeared. With only three players the study is better read as a demonstration of the idea than as an estimate of how large the effect is. It supports the view that familiar configurations let a knowledgeable player organise pieces into meaningful units.

This makes chunking a consequence of learning a subject, as well as a deliberate memory technique. A cook can recognise a sauce base where a beginner sees disconnected ingredients, just as a reader recognises a phrase without consciously maintaining every letter. Arbitrary labels have less value without familiar structure behind them. When learning a procedure, connect each group to its purpose and practise recognising where that group begins and ends.

For a small experiment, take instructions you often forget and group their steps by the outcome each group produces, such as preparing the workspace or checking the finished result. Keep the order visible while you learn. On a later attempt, hide the instructions and describe the groups before completing the procedure, then check whether any necessary step disappeared inside a label that was too vague to retrieve it.

Written states and worked examples reduce unnecessary load

John Sweller's 1988 paper, Cognitive Load During Problem Solving: Effects on Learning, proposed that searching for a solution can consume capacity that might otherwise support learning its structure. This became a foundation for cognitive load theory. A novice may spend considerable effort deciding what to try next while also retaining the problem's conditions. Struggle alone therefore cannot guarantee useful learning. The practical response is to reduce avoidable demands while keeping the relationships that the learner needs to understand.

Writing down intermediate states offers a direct way to do that in everyday work. During a budget calculation, put the current total beside the adjustment that produced it, including the currency and what that total covers. An isolated number leaves its meaning open to confusion. A labelled state lets you inspect the operation, recover after an interruption and check whether a later change used the intended starting point, especially when you need to distinguish money already spent from money merely set aside.

A worked example provides a related form of support: the solution steps are visible, leaving room to examine why they follow. Sweller and Graham Cooper's 1985 experiments on learning algebra found that students who studied worked examples tended to solve later problems of the same structure faster and with fewer errors than students who had practised on conventional problems alone. This worked example effect matters most when the method is unfamiliar. Explain each step in your own words, then cover part of the example and complete a similar problem; as the procedure becomes familiar, remove more of that support.

Practice teaches a task without promising broad transfer

Getting better at a memory exercise does not establish that your capacity has increased across daily life. You may have learned its materials, discovered a useful grouping or become more efficient at answering. Daniel Simons and colleagues' 2016 review, Do “Brain-Training” Programs Work?, found substantial evidence for improvements on trained tasks, with weaker evidence as the tested outcomes became more distant. Many studies also had design or analysis shortcomings.

Monica Melby-Lervåg, Thomas Redick and Charles Hulme's 2016 meta-analysis found no convincing evidence that working memory training improves intelligence or wider academic skills when trained groups were compared with groups doing an active alternative activity, although trained groups did improve on closely related memory tests. Learning a specific technique can still be worthwhile. Its value should be checked on the activity you want to perform, rather than inferred from a rising exercise score.

Funga Wega's “track the state” exercise asks you to read a starting state and some steps, hide them and answer about the resulting state; no AI generates or grades it. Try keeping the current state distinct from the operation that changes it. After checking an answer, locate the point where your representation went wrong and compare approaches on later attempts. For an everyday task that keeps going astray, practise the same procedure with a written state first, then see which parts you can retain reliably without the note.

Frequently asked questions about working memory

Is four chunks a limit for everyone?

Cowan's proposal describes an approximate capacity under conditions intended to limit rehearsal and supplementary memory support. The task, existing knowledge and individual differences affect performance, so it cannot determine how many instructions a particular person can follow.

Does writing things down prevent memory practice?

Writing supports a task by keeping information available while you work with it. When remembering is your learning goal, you can later hide the note, attempt recall and check what was missing.

Can working memory exercises make me better at studying?

Broad improvements from practising unrelated memory exercises have not been established convincingly by the reviews discussed here. Practising your study material with clearer grouping and fewer unnecessary demands addresses the activity whose improvement you can observe directly.