Cognitive load
Learning is bottlenecked by working memory, which can hold only a few things at once, while long-term memory is effectively limitless. The idea splits the mental effort a lesson demands into three kinds — the difficulty built into the material, the difficulty added by how it is presented, and the useful effort of forming durable mental structures — and argues that good teaching strips away the second so the limited capacity that remains can do the third. Content can be correct and still fail to be learned if its presentation overloads the learner.
John Sweller's theoretical framework holding that learning is constrained by the limited capacity of working memory, and that instructional design either respects that constraint or fails despite the content being correct. The framework distinguishes three components of the load any learning episode imposes on working memory: intrinsic load, set by the number of interacting elements in the material itself; extraneous load, imposed by the format of the presentation; and germane load, the cognitive effort that goes into building the schema. Effective teaching, in this account, is the work of suppressing extraneous load so that the surviving capacity can do the schema-building.
Etymology§
The phrase cognitive load in this technical sense is Sweller's in the 1988 Cognitive Science paper Cognitive load during problem solving: Effects on learning. The underlying working-memory model is George Miller's seven plus or minus two (1956), refined through Alan Baddeley's working-memory architecture in the 1970s and 1980s. The three-component decomposition (intrinsic / extraneous / germane) was developed across Sweller's collaborations with Paul Chandler, Fred Paas, and Jeroen van Merriënboer through the 1990s and 2000s. Cognitive Load Theory (Springer, 2011, second edition 2023) is the field-summary volume.
The argument's anchor is the asymmetry between working memory and long-term memory. Working memory is small and time-limited; it holds roughly four to seven elements at once and loses them within seconds unless they are being actively rehearsed or transformed. Long-term memory is, for practical purposes, unbounded. Learning is the process of building structures in long-term memory — schemas — that allow multiple related elements to be retrieved and processed as a single unit when they are needed again. Once a schema is built, a problem that previously consumed several working-memory slots can be handled with one; the bottleneck releases.
The three-load decomposition follows from this account. Intrinsic load is set by the material itself — by how many elements interact in the problem and have to be held together to make sense of it. Solving a single-step arithmetic problem has low intrinsic load; balancing a chemical equation has higher intrinsic load; debugging a multi-file program has higher still. Intrinsic load is a property of the content and the learner's current schemas, not of the instruction; it can be sequenced (simpler problems before harder ones) but not erased.
Extraneous load is imposed by the form of the presentation rather than the content. A diagram on one page and its caption on another forces the learner to split attention and reconstruct the mapping in working memory; an integrated diagram-and-caption removes the split-attention demand. A worked example with the solution steps and the diagram in the same visual frame demands less than the same content in separated panels. The split- attention effect, the modality effect (mixing visual and auditory channels), and the redundancy effect (redundant information actively increasing load) are the empirical signatures of extraneous load and the levers instructional design uses to suppress it.
Germane load is the cognitive work that actually goes into schema-building — the effort the learner spends linking new material to existing knowledge, generalising across worked examples, and anticipating the next case. Germane load is the component that should not be suppressed; the design problem is to clear extraneous load out of the way so the learner has working memory left over for the germane work.
The empirical core of the framework is the worked-example effect. A novice studying complete worked examples builds the underlying schema faster than a novice solving equivalent problems, because problem-solving from scratch consumes working memory on search behaviour (trying possible moves, evaluating, backtracking) that does not contribute to schema acquisition. The finding has held across forty years of replication in mathematics, programming, statistics, and physics. The expertise reversal effect is its companion: the same worked- example design that helps novices begins to hinder experts, who already have the schema and for whom the worked example becomes redundant load. The instructional implication the framework gives is worked examples first, problem-solving as the schema develops.
Cognitive load theory's relationship to deliberate practice is one of the recurring tensions in the skill-acquisition literature. Sweller's account makes the careful study of worked examples primary in the acquisition phase; Ericsson's account makes effortful problem-solving at the edge of present ability primary throughout. The two can be reconciled — worked examples for novices, deliberate practice once the schema is in place — but the emphases differ, and the literatures cite each other less than the overlap warrants.