What Is Cognitive Load? Why the Mind Can Hold Only So Much at Once

Human memory has two stores that work very differently. Long-term memory is vast and durable, holding everything from the layout of a childhood home to the grammar of a language, organised into schemas, patterns that let an expert see a chessboard as a few positions where a novice sees thirty pieces. Working memory is tiny and brief: George Miller's seven plus or minus two of 1956 has been revised down to about four chunks, and a chunk is whatever long-term memory can supply as a unit, so that a phone number is seven chunks to a stranger and one to its owner. Learning is the transfer of patterns from the small store to the large one, and it can only happen through the small one, which is why the amount of information arriving at once matters so much.

Sweller's theory divides the demand on working memory into parts:

• •Intrinsic load: the difficulty inherent in the material, set by how many elements must be held and related at once; adding two numbers is low, solving a simultaneous equation high, and it can be reduced only by breaking the task into pieces or by the learner's growing expertise
• •Extraneous load: demand that comes from how the material is presented rather than from the material, a diagram with its labels in a separate list, a lecturer talking over a dense slide, a manual that makes the reader hunt for the relevant step; it teaches nothing and can be designed away
• •Germane load: the effort of building and automating the schemas that are the point of learning, which is what the freed capacity should be spent on

The theory grew from findings that upset common sense. Students learning algebra from worked examples, in which the solution was shown, outperformed students who solved the same problems themselves, because solving from scratch loaded working memory with search and left nothing for learning the pattern; the effect reverses once the learner is expert, when worked examples become redundant. Splitting attention between a diagram and text placed elsewhere hurt learning, and integrating the labels into the diagram helped. Presenting the same information twice, in speech and on a slide, hurt, because processing both was extra load; presenting a diagram with spoken rather than written explanation helped, because the visual and auditory channels are partly separate. Each finding is a case of the same rule: whatever working memory spends on things other than the pattern to be learned is lost.

The applications are wherever people must think under pressure or learn something hard. In teaching: worked examples before problems, one new element at a time, diagrams with labels in place, and the removal of decorative content, since interest that adds load is not free. In interfaces: the cockpit, the operating theatre and the control room, where a checklist exists to move a sequence out of working memory onto paper, and where an alarm that requires interpretation during an emergency is a design failure. In everyday work: the finding that interruptions cost far more than their duration, since the interrupted task's contents are lost and must be rebuilt, and that multitasking is switching, at a price paid each time. The theory's limits are argued, since the three loads are hard to measure separately and the germane category has been redefined more than once, but the central constraint, four things at a time, has not moved.

Expertise is the escape from the limit. A chess master holds a position in working memory as a few chunks that long-term memory supplies, a fluent reader takes in words rather than letters, and a surgeon runs a procedure as a sequence of automated steps that leave capacity free for the unexpected; in each case the load of the task has been moved from the small store to the large one by practice, which is what practice is for. The novice looking at the same task sees every element separately and is overloaded by what the expert does not notice, which is why experts make poor teachers of beginners unless they remember, and why the most useful thing a teacher can do is decide what to leave out.

Cognitive load cannot be observed directly, so researchers infer it, and each method has its own blind spots. The simplest is to ask: after a task, learners rate how much mental effort it took on a short numbered scale, and a nine-point version introduced in the early 1990s is still widely used. It is quick and tracks performance reasonably well, but it relies on people judging their own effort, which they do imperfectly, and a single rating given at the end cannot show which moment of a lesson was the hard one.

The other methods are indirect measures of behaviour and the body. In a dual-task design, a learner does the main task while responding to a simple secondary signal, such as a tone, and the slower the response to the tone, the less spare capacity the main task has left. The pupil of the eye also widens slightly with mental effort, a finding from the 1960s, so eye-tracking cameras can follow load moment by moment, although the pupil reacts to light and emotion too. You can notice the dual-task effect in everyday life: conversation in a car often stops while the driver works through a difficult junction.

One of Sweller's earliest findings is also one of the easiest to try. A typical geometry exercise asks for one specific thing, such as the size of a particular angle in a diagram. To solve it, a novice tends to work backwards, comparing where they are with where they need to be and hunting for any step that narrows the gap, a strategy called means-ends analysis. It often succeeds, but it fills working memory with the goal, the current state and the difference between them, so a student can reach the right answer while learning very little about the relationships that got them there.

Sweller's alternative was the goal-free problem. Instead of asking for angle x, the question says: find the size of as many angles as you can. With no single target there is no gap to keep track of, so the learner simply works forwards, applying whichever rule fits the information in front of them, and each step becomes a small, clear use of one principle. In experiments, students given goal-free versions tended to learn the underlying rules better than those given the conventional version. The approach suits topics where working forwards is possible, and it matters less once learners have the expertise to handle goals comfortably.

A study published in 2017 by Adrian Ward and colleagues reported that students performed worse on tests of working memory and problem solving when their smartphone was on the desk beside them, even turned face down and silent, than when it was in another room. The authors suggested that resisting the pull of the phone used up some of the mind's limited capacity.

Later attempts to repeat the finding have produced mixed results, and some found little or no effect, so the size of the effect is uncertain. The advice drawn from it is still sensible and costs nothing: when you need to concentrate on something demanding, putting the phone out of sight and reach removes one more thing competing for a limited supply of attention.

Cognitive load is the demand a task places on working memory, which holds only about four chunks at once and is the only route into long-term memory, so that learning fails when the load exceeds it. The load divides into the intrinsic difficulty of the material, the extraneous burden of poor presentation and the germane effort of building patterns; worked examples, integrated diagrams, one element at a time and short checklists reduce it, and expertise escapes it by packing what was many things into one.