Cognitive Load

John Sweller's Cognitive Load Theory: why working memory has hard limits, the difference between intrinsic, extraneous, and germane load, and what this means for how you design flashcards.

Cognitive Load Theory, developed by Australian educational psychologist John Sweller in the late 1980s, holds that learning is constrained by working memory's hard capacity limits. Working memory can hold only about four "chunks" of new information at once (Cowan, 2001), for about 20 seconds. Instructional design — and flashcard design — that respects these limits produces better learning than design that overloads them.

Key Takeaways

Working memory is the bottleneck: it processes 2–4 chunks of new information at a time, for ~20 seconds - Cognitive load comes in three forms — intrinsic (inherent to the material), extraneous (caused by poor design), and germane (effort that builds long-term schemas) - Reducing extraneous load and managing intrinsic load lets germane load do the actual learning work - Flashcards with multiple facts on one card, ambiguous prompts, or visually cluttered formatting impose extraneous load that has nothing to do with the material - Schema-building — chunking knowledge into reusable patterns — is how learners move beyond working-memory limits

What working memory actually is

Working memory is the mental workspace where you hold and manipulate information consciously, right now. It's distinct from short-term memory in that it includes active processing, not just storage. When you're trying to solve a problem in your head, working memory is what holds the partial steps.

George Miller's famous 1956 paper ("The Magical Number Seven, Plus or Minus Two") suggested working memory holds about 7 items. Later research, particularly Nelson Cowan's 2001 work, revised this downward: when you remove chunking strategies and rehearsal, the true limit for novel information is closer to four items. For unfamiliar material — exactly what you face when studying something new — this lower number applies.

The duration is also short: unrehearsed information in working memory decays within about 20 seconds.

These limits are stable and well-replicated across studies. They're a hardware constraint, not a skill to overcome.

How Sweller built the theory

John Sweller, an Australian educational psychologist, developed Cognitive Load Theory (CLT) from problem-solving research in the late 1980s. His 1988 paper "Cognitive load during problem solving: Effects on learning" laid out the framework. The theory has been substantially extended since — Sweller, Paul Chandler, Fred Paas, and others have published hundreds of empirical studies refining it.

The central claim: learning fails when working memory is overloaded. Instructional design should explicitly account for working memory limits, not just topic content.

The three types of load

Sweller's model distinguishes three sources of cognitive load:

Intrinsic load

Caused by the inherent difficulty of the material. Learning to add two single-digit numbers has low intrinsic load. Learning to solve simultaneous differential equations has high intrinsic load. Intrinsic load depends on:

The number of interacting elements in the material
The learner's existing knowledge (more prior knowledge = lower intrinsic load for the same content)

You can manage intrinsic load by sequencing material from simple to complex, breaking topics into smaller units, and building prerequisite knowledge before advanced topics. You can't directly reduce intrinsic load below what the material requires — but you can stage when learners encounter it.

Extraneous load

Caused by how material is presented, not the material itself. Poor formatting, irrelevant decoration, split attention between text and diagram, unclear instructions, ambiguous wording — these consume working memory without contributing to learning.

This is the type of load instructional design has the most leverage over. A well-designed lesson can have the same content as a poorly designed lesson but produce much better learning, simply because less working memory is wasted on irrelevant processing.

Germane load

The effortful processing that builds long-term schemas. When working memory has room left over after handling intrinsic load, it can do the work of integrating new material with prior knowledge, looking for patterns, and constructing mental models. This is where learning actually happens.

The goal of good instruction: minimize extraneous load, manage intrinsic load through sequencing, free up working memory for germane load.

Schemas: how learners escape working memory limits

If working memory holds only 4 chunks, how do experts handle complex material with apparent ease?

Through schemas. A schema is a knowledge structure that combines many elements into a single chunk in working memory. A chess grandmaster sees "Sicilian Defense, Najdorf variation" as one chunk; a beginner sees twenty separate piece positions. The grandmaster's working memory is no larger than the beginner's — but the grandmaster's chunks are denser.

This is why expertise feels qualitatively different from novicehood. Experts haven't expanded their working memory; they've built schemas that pack more information into each chunk.

Spaced repetition, retrieval practice, and worked-example study all serve schema-building. The point of repeated review isn't just to "remember the fact" but to integrate the fact with related facts until they merge into a single retrievable unit.

What this means for flashcard design

Sweller's framework gives concrete guidance for designing study materials:

One fact per card. A card asking "What are the three types of cognitive load and an example of each?" overloads working memory during retrieval. Three separate cards, one per type, fit. The difference isn't pedantic — multi-fact cards produce noisier ratings (you remembered two of three; how do you grade?) and weaker memory traces (the brain doesn't form one strong association but three weak ones competing for attention).

Eliminate decoration. Background images, decorative borders, and unnecessary formatting on flashcards add extraneous load. The card should be visually as simple as possible: question, answer, nothing else. The "split-attention effect" specifically warns against forcing learners to integrate spatially separated elements (text on one side of a diagram, labels on the other) — keep related elements physically close.

Make the question unambiguous. A card asking "When did this happen?" without context forces the learner to first reconstruct what the card is about before they can answer. Each card should be self-contained: "When did Hermann Ebbinghaus publish Über das Gedächtnis?" instead of "When did the book come out?"

Build schemas through related cards. Cards about a single domain (the Krebs cycle, French regular verbs, JavaScript array methods) build a denser schema than cards scattered across unrelated topics. This doesn't mean batch-studying only one topic — interleaving still helps — but related cards should be in the same deck so they're encoded together.

Pre-build vocabulary. Cards using technical terms the learner doesn't yet know force a "look up the term in my head" step that consumes working memory. If a card references "stability" in an FSRS context, the learner should already have a card establishing what stability means in FSRS.

In Neurako

Neurako's AI card generation explicitly targets one-fact-per-card output, but it's worth reviewing AI-generated cards for unintended multi-fact prompts. When you see a card asking "What are the X and Y of Z?", split it into two cards. The marginal time cost is small; the recall benefit compounds over months of review.

Worked examples vs. problem-solving

One of Sweller's most cited findings: worked examples often beat unguided problem-solving for novices. A learner shown a fully worked solution to a problem, then asked to do a similar problem, typically outperforms a learner asked to solve the problem from scratch. The worked example reduces extraneous load (you don't have to invent the approach) and frees working memory for schema construction.

This effect reverses for experts — once schemas are in place, doing problems is more productive than studying solutions. The "expertise reversal effect" (Kalyuga, Ayres, Chandler, & Sweller, 2003) describes this transition.

For studying: when learning a new topic, don't immediately throw yourself at hard problems. Study several worked examples first. Once you've absorbed the pattern, switch to problem-solving practice. Flashcards naturally support this: early cards can include worked-example fronts ("How do you compute X? Here's a worked example. What's the next step?") and graduate to bare-problem fronts as the skill matures.

Implications for session design

Cognitive Load Theory also suggests practical session structure:

Long sessions accumulate fatigue. As working memory fatigues, extraneous load that was tolerable becomes overwhelming. This is why a card that was easy in session 1 sometimes feels harder in session 5. Take real breaks.

Don't mix unrelated dense topics in one session. Switching between two topics that both make heavy working-memory demands (e.g., organic chemistry mechanisms and a foreign-language grammar) leaves both poorly encoded. Interleaving works best within a domain, not across domains with no shared schema.

Match difficulty to recovery. If a card consistently feels overwhelming, the problem may not be that you "don't know it" — the card may be poorly designed and imposing extraneous load. Rewrite it.

Dual Coding

Paivio's theory of verbal + visual processing channels — and Mayer's multimedia learning research extending it.

Spaced Repetition

How distributed review builds the schemas that overcome working-memory limits.

Active Recall

Why retrieval practice is the main way working memory builds long-term schemas.

Sources

Sweller, J. (1988). Cognitive load during problem solving: Effects on learning. Cognitive Science, 12(2), 257–285. https://doi.org/10.1207/s15516709cog1202_4
Cowan, N. (2001). The magical number 4 in short-term memory: A reconsideration of mental storage capacity. Behavioral and Brain Sciences, 24(1), 87–114. https://pubmed.ncbi.nlm.nih.gov/11515286/
Miller, G. A. (1956). The magical number seven, plus or minus two: Some limits on our capacity for processing information. Psychological Review, 63(2), 81–97. https://pubmed.ncbi.nlm.nih.gov/13310704/
Sweller, J., van Merriënboer, J. J. G., & Paas, F. (2019). Cognitive architecture and instructional design: 20 years later. Educational Psychology Review, 31, 261–292. https://link.springer.com/article/10.1007/s10648-019-09465-5
Kalyuga, S., Ayres, P., Chandler, P., & Sweller, J. (2003). The expertise reversal effect. Educational Psychologist, 38(1), 23–31. https://doi.org/10.1207/S15326985EP3801_4

Ready to turn this into a study ritual?

Start studying with Neurako