Cognitive Load

What it is

Everything a person learns passes through working memory first, and working memory is small. George Miller's 1956 paper put its capacity at about seven items and that figure became one of the most repeated numbers in popular psychology. Current estimates are lower, around three to four, following Nelson Cowan's 2001 work. Whichever number is used, the shape of the problem is the same: an effectively unbounded long-term store, reached only through a very narrow opening. Nicholas Carr's image for it is a thimble filling a bathtub. Information arriving faster than the thimble can carry it does not get stored more quickly. It gets spilled.

John Sweller's contribution, in a 1988 paper in Cognitive Science, was to notice that the traffic through that opening is not all of one kind, and that only some of it is doing useful work. Cognitive load theory divides the demand into three.

Intrinsic load is the difficulty of the material itself. Simultaneous equations are harder than addition because of what they are, and no amount of good design makes them easy. Intrinsic load is a property of the subject and the learner's existing knowledge, not of the presentation.

Extraneous load is the demand added by how the material is presented. A diagram whose labels sit in a key on the opposite page forces the reader to hold arrow positions in mind while they look up what the arrows mean, and that holding costs capacity that could have gone to the content. Extraneous load contributes nothing. It is pure overhead, and it is the part a writer, teacher or designer controls.

Germane load is the productive effort of actually building the understanding, of assembling the pieces into a structure that can be recalled and reused. This is the effort that learning consists of, and the reason the theory is not simply an argument for making everything as easy as possible.

The practical consequence follows directly from the three sitting in one fixed container. Because intrinsic load cannot be lowered without changing the subject, and germane load is the thing you want more of, extraneous load is the only term worth attacking. Every unit of overhead removed is a unit freed for the work. This is why identical content, presented two ways, produces different learning, and why the finding that presentation matters is not a claim about motivation or effort at all.

In effect

The Mali presentation, with and without the extras

The clearest demonstration in the material this site draws on is a comparison of two versions of the same lesson. Participants who watched a text-and-image presentation about Mali scored 7.04 on a ten-question quiz. Participants who watched a version of the same presentation with added audio and video scored 5.98. The addition was not distraction in any obvious sense. It was more of the subject, delivered in more channels, and it made performance worse. The second half of the result is the part worth keeping: the multimedia group rated the presentation more interesting and believed they had learned more. The gap between how well people think they understood and how well they did is the practically useful part of load theory, and it recurs throughout this literature.

Every link is a decision

Erping Zhu's 1999 study varied only the number of links in otherwise identical passages. Comprehension fell as link count rose, and so did readers' confidence in their own understanding. Load theory supplies the explanation. Deciding whether to click is a decision about navigation, and the capacity it consumes is capacity not spent on the argument the sentence was making. The link may be relevant, well chosen and useful later. Within the passage it is still overhead. Hypertext is the applied case, and screen inferiority is where that literature settled.

The recipe that hides the quantities

A recipe that lists ingredients with quantities at the top, then writes the method as "add the flour, then the butter", forces the cook to hold six numbers in mind or to look back up six times. A recipe that repeats the quantity inside the step, "add the 220g of flour", does not. The dish is identical and its intrinsic difficulty is identical. One version spends the cook's working memory on retrieval and the other does not. This is an illustration rather than an experiment, but it is the shape of the extraneous-load claim in a form anyone can check against their own kitchen.

Onboarding, checkout, and the design case for asking less

The commercial version of load theory is the observation that every field, option and decision in an interface is a demand on the same narrow channel, and that most of them are extraneous to what the user came to do. Progressive disclosure, sensible defaults, splitting a long form into steps that each hold one idea, and removing optional fields from a checkout are all the same move: reducing the overhead so the capacity goes to the decision that matters. Elements of simplicity is the design-side vocabulary for this, and the translation is direct, because reducing friction is reducing load. It is worth being clear about the asymmetry, though. In teaching, load is cut so the learner can do the harder work. In commerce, load is usually cut so the user does not stop, and the two motives happen to recommend the same interface.

What it does not say

Cognitive load theory is one of the better-replicated frameworks in educational psychology, which is why this page is tagged robust. That status attaches to the framework, not to every result derived from it. Specific effects within the literature vary in size and in how reliably they reproduce, and a page citing the theory should not treat any individual demonstration as settled.

It does not say that easier is better. Germane load is effort, and removing it removes the learning. A presentation that requires nothing of the reader can be perfectly comfortable and leave nothing behind, which is the trap the Mali result describes from the other direction: the version that felt richer was the version that taught less. Difficulty in the right place is the mechanism, not the obstacle.

The numbers on this page should be handled with care. Miller's seven is out of date by decades and the working figure is closer to three or four, on Cowan's revision. The Mali comparison reaches us through Carr's The Shallows, and our notes record the two quiz scores without the sample size, the design or the original citation, so the scores are reported here as they were reported to us rather than verified at source. Zhu's link-density result is a single 1999 study and we hold no independent replication of it.

Finally, load theory says nothing about brains being rewired, and it is not evidence for the popular argument that the internet is damaging cognition. It is a theory about the capacity limit that has always been there, and about the fact that presentation can waste that capacity or spend it well. That claim is old, narrow and well supported. The larger claims that borrow its authority are neither.

Who developed it

John Sweller proposed cognitive load theory in 1988, in Cognitive Science, working on problem solving and instructional design. The capacity limit the theory is built on comes from George Miller's 1956 work on working memory, revised downwards by Nelson Cowan in 2001. Nicholas Carr is the writer who brought load theory into the argument about reading on screens, in The Shallows.

---