Mark ở vũ trụ khác · Another Universe Papers That Changed How We Learn #5 · Learning · Working memory

Cognitive load

John Sweller · "Cognitive load during problem solving: Effects on learning" · Cognitive Science 12(2), 257–285 · 1988

Who, whereeducational psychologist John Sweller, University of New South Wales (Sydney, Australia)
Core ideasolving problems by aiming straight at the answer crowds out learning
Lessonnovices should study worked examples before solving on their own
LabelCaution: the core idea is useful, but "load" is hard to measure

Source: Sweller (1988), Cognitive Science 12(2), 257–285; Sweller's profile on the University of New South Wales (UNSW) website. The working environment blocked the journal site, so the original was checked through excerpts in search results, not the full text.

One-minute recap

Essence

Solved is not learnedPeople reached the answer, but very few noticed the rule behind it.
A crowded deskAiming at the answer forces the mind to hold the goal, the current state, the gap and subgoals. No room left to learn.
Worked examplesFollowing ready-made steps leaves room to see the structure. A 2023 meta-analysis in maths: a medium effect.
Not for everyoneExperienced learners do worse with examples than with solving: the expertise reversal effect.
Hard to measure"Load" is usually self-rated. The types of load overlap, and researchers still argue.

Source: Sweller recounts his early-1980s experiments in Sweller (2023), Educational Psychology Review 35:95; abstract of Sweller (1988); Barbieri et al. (2023), Educational Psychology Review 35 (g = 0.48); Kalyuga, Ayres, Chandler and Sweller (2003), Educational Psychologist 38(1); de Jong (2010), Instructional Science 38(2).

Part 1

The paper's question

Why doesn't solving lots of problems always make us better at them? Sweller's answer: while we push toward the answer, the mind is so busy that there is no room left to learn how to solve.

Source: abstract of Sweller (1988). The question is Mark's restatement of the paper's core idea.

early 1980sA number puzzleSolvers reach the goal, but hardly anyone sees the rule.
→
1988The paperGoal-directed solving crowds out learning "schemas" (ways of solving).
→
todayWorked examplesA standard teaching tip, with a limit: experts are different.
Part 2

The discovery

Who, where, how, and what people thought about practice problems at the time.

The number puzzle Sweller describes: get from a start number to a goal number using only multiply by 3 or subtract 69. The numbers were chosen so only one solution existed: alternate the two moves. Solvers reached the goal, but very few noticed the alternating rule. the number puzzle (as Sweller tells it) start goal × 3 − 69 × 3 − 69 hidden rule: alternate the two moves. It is the only solution. result: solvers reached the goal, but very few noticed the alternating rule illustrative diagram; no specific start or goal numbers, as none appear in the sources checked
Solved without learning. Sweller says the numbers were chosen so only one solution existed, and "very few" found the rule of alternating the two moves. His conclusion: goal-directed problem solving is for reaching the goal, not for learning.Source: Sweller (2016), "Story of a Research Program", and Sweller (2023), Educational Psychology Review 35:95, recounting the original study Sweller, Mawer and Howe (1982), American Journal of Psychology 95, 455–483.
1969–1972 · AdelaideJohn Sweller takes his BA and then his PhD in psychology at the University of Adelaide, and later works at the University of New South Wales (UNSW), Sydney.
1972 · Means-ends analysisNewell and Simon describe how people solve problems: look at the gap between where you are and the goal, find a move that shrinks it.
1980 · Experts differ from novicesStudies of physics problem solving: novices work backward from the goal and set subgoals; experts work forward using "schemas" they already hold.
Early 1980s · The number puzzleOnly multiply by 3 and subtract 69. Goal reached, rule rarely seen.
1985 · Worked examplesSweller and Cooper give algebra learners examples to study instead of problems to solve: faster learning, and later problems of the same type solved faster with fewer errors.
1988 · The paperPulls it all into one explanation: goal-directed solving fills working memory.

Source: Sweller's UNSW profile; degree years from his ATD profile (secondary source); Newell and Simon (1972), Human Problem Solving; Larkin, McDermott, Simon and Simon (1980), Science 208; Sweller (2023); Sweller and Cooper (1985), Cognition and Instruction 2(1), 59–89, per the paper's abstract.

Part 3

What the paper says, in 3 sentences

  1. When we solve a problem by aiming at the goal, we must hold at once the goal, where we are, the gap between them, and any subgoals.
  2. That uses up most of working memory, leaving little room to learn "schemas" (recognising the problem type and how to solve it).
  3. So teaching that reduces goal-chasing, such as problems without a specific goal or worked examples, can help novices learn faster.
"Conventional problem solving in the form of means-ends analysis requires a relatively large amount of cognitive processing capacity which is consequently unavailable for schema acquisition."Sweller (1988), abstract.
Two desks of the same size. Left, goal-directed solving: full of blocks for the goal, current state, gap, subgoal, another subgoal; the block "learn the method" has fallen off. Right, worked example or goal-free problem: only the step being read, with room for "learn the method". goal-directed solving worked example / goal-free problem goal where am I the gap subgoal another subgoal learn the method step being read learn the method illustrative diagram of the paper's idea, no numbers; the desk = limited working memory
Same desk, two ways to use it. The 1988 paper used a computer simulation to compare two ways of solving the same problem: the goal-directed route needed more information in working memory and a far more complex set of rules. Its experiments showed that people given a specific goal still solved the problems but learned less of their structure.Source: Sweller (1988), simulation and experiment sections, checked through excerpts in search results. No participant counts or error numbers are given here because they could not be checked against the original tables.
Part 4

How this paper is rated

Every line names its source and lookup date.

PublishedCognitive Science, vol. 12 no. 2, pp. 257–285, 1988. The peer-reviewed journal of the Cognitive Science Society.
CitationsMore than 8,500 according to Semantic Scholar (looked up 15/10/2026 via the Semantic Scholar page shown in search results). Google Scholar: could not be looked up (blocked in the working environment). Databases count differently.
Replicated?The idea that worked examples help novices has been replicated many times. A 2023 meta-analysis in maths (43 articles, 55 studies): a medium effect, g = 0.48. The "goal-free problem" idea has been tested less.
Main criticismTon de Jong (2010): concepts unclear, hard to measure, easy to explain after the fact. A 2022 meta-analysis of load questionnaires: reliable, but correlations between load types partly conflict with the theory. Manu Kapur and colleagues: letting learners try first, then teaching, can deepen conceptual understanding.
Place in educationThe New South Wales Department of Education (Australia) published a 2017 review, "research that teachers really need to understand", on this theory. Sweller is a Fellow of the Academy of the Social Sciences in Australia.
VerdictCAUTION The core holds for novices; "load" is hard to measure and the effect reverses for experienced learners.

Source: the paper's Semantic Scholar page (looked up 15/10/2026); Barbieri, Miller-Cotto, Clerjuste and Chawla (2023), Educational Psychology Review 35; de Jong (2010), Instructional Science 38(2), 105–134; Krieglstein et al. (2022), Educational Psychology Review; Sinha and Kapur (2021), Review of Educational Research; CESE, NSW Department of Education (2017); Sweller's UNSW profile. The label is Mark's assessment.

Part 5

Fact check: what holds and what is myth

12 things often said about the paper and cognitive load theory. Click a line to see why.

SOLIDNovices learn better from worked examples than from solving from scratch

Sweller and Cooper (1985) in algebra; Barbieri et al. (2023) meta-analysis: 55 maths studies, g = 0.48, medium. Correct examples alone worked better than incorrect ones or a mix.

SOLIDSolving a problem does not mean you learned how to solve it

The multiply-by-3, subtract-69 puzzle: goals reached, but very few saw the alternating rule (Sweller recounting in 2016 and 2023 the work of Sweller, Mawer and Howe 1982). The 1988 paper: solvers with a specific goal learned less of the problem structure.

SOLIDThe 1988 paper says goal-directed solving crowds out learning

That is what the abstract says: capacity used for means-ends analysis is "unavailable for schema acquisition".

SOLIDThe theory has entered official guidance for teachers

NSW Department of Education, Centre for Education Statistics and Evaluation (CESE), 2017: "Cognitive load theory: Research that teachers really need to understand".

CAUTION"The 1988 paper proved it with large classroom experiments"

The paper's evidence is a computer simulation and controlled problem-solving experiments. Evidence from many real classrooms came later, through meta-analyses.

CAUTION"Cognitive load can be measured precisely"

It is usually measured with self-rated effort scales. de Jong (2010) criticised how hard it is to measure and to separate load types. A 2022 meta-analysis: the questionnaires are reliable, but correlations between load types partly conflict with the theory.

CAUTION"Letting students struggle first is a waste of time"

Sinha and Kapur (2021), 53 studies: trying first and being taught afterwards can help conceptual understanding when well designed; for procedural skills the two orders were equal. Ashman, Kalyuga and Sweller (2020): the more interacting elements in a task, the better explicit instruction first. It depends on the task and the learner.

CAUTION"Working memory holds 7 items"

Miller's 7 (1956) has been revisited: Cowan (2001) argued for about 4 chunks. See Papers That Changed How We Learn #2. Load theory only needs "working memory is limited".

CAUTION"The 1988 paper is the whole of cognitive load theory"

The 1988 paper was the seed. Three load types and "germane load" came in 1998 (Sweller, van Merriënboer and Paas), the expertise reversal effect in 2003, a major update in 2019. Sweller (2023) writes that failed replications forced the theory to become more detailed.

CAUTION"It is the single most important thing for teachers to know"

Dylan Wiliam wrote this on Twitter on 26 Jan 2017 (secondary source: blogs embedding the post). He did say it, but it is an opinion. An article on the Chartered College of Teaching site calls it a bold claim.

FALSE"Worked examples are always better, for everyone"

Kalyuga, Ayres, Chandler and Sweller (2003): techniques that help novices a lot can lose their effect, or even backfire, with experienced learners. Experts should move on to solving.

FALSE"The easier learning feels, the better"

The theory says cut extraneous load (messy presentation, irrelevant tasks), not all effort. Since 1998 it has included "germane load". Self-testing (paper #6) is effortful and still helps memory last.

Source: as noted in each line; full list in the Sources section.

Part 6

What the latest research says

2019 · "20 years later"Sweller, van Merriënboer and Paas review two decades of the theory and set new directions (Educational Psychology Review 31, 261–292).
2021 · Try first, learn afterSinha and Kapur pool 53 studies, 166 comparisons: trying a problem before instruction helps conceptual understanding when well designed (Review of Educational Research).
2022 · Measuring load by questionnaireA meta-analysis of 4 common questionnaires: reliable, but the relations between load types partly do not fit the theory (Educational Psychology Review).
2023 · Worked examples in mathsBarbieri et al. pool 43 articles, 55 studies, 181 effect sizes: a medium effect, g = 0.48 (Educational Psychology Review 35).
2023 · Sweller looks backFailed replications did not topple the theory; they forced it to say more clearly when it applies and when it does not (Educational Psychology Review 35:95).

Source: the papers named in each line. Study counts and g values come from the papers' own abstracts, checked through search results on 15/10/2026.

Part 7

Assessment of the paper

Scores are Mark's assessment based on the sources listed, not data from a study.

Novelty at the time
5/5
Evidence in the original
3/5
How well it holds today
4/5
Ease of measurement
2/5
Ease of use
5/5
Strengths
  • Neatly explains something every student meets
  • Produced usable advice: worked examples
  • Replicated across many meta-analyses
Weaknesses
  • "Load" is hard to measure, load types overlap
  • Does not hold for experienced learners
  • Original evidence is a model and small experiments

The paper is right where it matters most: novices need to see how a problem is solved before they wrestle with it alone. Just don't turn it into "the easier, the better".

Part 8

Apply it: study a worked example before solving alone

Usable from tomorrow in any subject with exercises: maths, physics, programming, accounting, grammar.

Four steps: read a worked example and explain each step to yourself; solve a twin problem of the same type; use examples with steps gradually blanked out; once fluent, switch to solving on your own. 1 · read an exampleexplain it to yourselfwhy is this step here 2 · solve a twinsame typeright after the example 3 · fade it outfill in missing stepsone more step missing each time 4 · solve aloneonce fluentnow the example gets in the way Novices: see first, do later. Experienced learners: do first.
From example to independent solving. Steps 1 and 2 rest on the worked-example effect; step 4 on the expertise reversal effect. Step 3 (fading) bridges them and is discussed in reviews of learning from examples.Source: Sweller and Cooper (1985); Barbieri et al. (2023); Kalyuga et al. (2003); Atkinson, Derry, Renkl and Wortham (2000), Review of Educational Research 70(2). The four-step order is Mark's suggestion.
Each new type: one exampleDon't dive into solving a problem type you have never seen. Find a correct solution and read it slowly.
Ask "why this step?"Write one line of explanation per step. If you can't, you don't understand it yet.
A twin problem right afterClose the example, solve the same type with new numbers. Stuck? Look at just that step.
Keep it in one placeMute notifications while studying, keep diagrams next to their explanations. Cut extra load, not thinking.

The first three rules rest on the worked-example effect (sources in the figure above). "Keep it in one place" is Mark's suggestion based on the theory's idea of cutting extraneous load.

Self-check: am I ready to drop the examples?

1. Close the example. Can I rewrite the solution and explain each step?

2. Same type, new numbers: can I solve it in about the time it took to read the example?

3. A slightly different type: can I spot which part resembles the example?

4. If all three are "yes", drop the example and solve alone. If question 1 is "no", go back to step 1.

Goal-directed problem solving serves to reach the goal, not to learn.Sweller's conclusion from the number puzzle, as he recounts it in Sweller (2023); paraphrased by Mark.
Part 9

Further reading

Sweller (2023), "The Development of Cognitive Load Theory: Replication Crises and Incorporation of Other Theories Can Lead to Theory Expansion"Educational Psychology Review 35:95. The author himself on how the theory grew, mistakes included. Free to read.
CESE, NSW Department of Education (2017), "Cognitive load theory: Research that teachers really need to understand"A summary for teachers with many classroom examples.
Papers That Changed How We Learn #2: Miller's number 7How much working memory holds. On markhocai.com/vu-tru-khac.
Papers That Changed How We Learn #6: Testing beats rereading (Roediger and Karpicke, 2006)An effortful habit that makes memory last. Next in the series.
Part 10

Sources

The original paper first, then peer-reviewed reviews and meta-analyses; secondary sources are labelled. The working environment blocked journal sites and citation databases, so sources were checked through excerpts in search results on 15/10/2026, not the full texts.

  1. Sweller, J. (1988). Cognitive load during problem solving: Effects on learning. Cognitive Science 12(2), 257–285. doi.org/10.1207/s15516709cog1202_4
  2. Sweller, J., Mawer, R. F., & Howe, W. (1982). Consequences of history-cued and means-end strategies in problem solving. American Journal of Psychology 95, 455–483.
  3. Sweller, J. (2016). Story of a research program. Education Review 23. Sweller, J. (2023). The development of cognitive load theory. Educational Psychology Review 35:95. link.springer.com
  4. Sweller, J., & Cooper, G. A. (1985). The use of worked examples as a substitute for problem solving in learning algebra. Cognition and Instruction 2(1), 59–89.
  5. Newell, A., & Simon, H. A. (1972). Human Problem Solving. Prentice-Hall. Larkin, J., McDermott, J., Simon, D. P., & Simon, H. A. (1980). Expert and novice performance in solving physics problems. Science 208, 1335–1342.
  6. Sweller, J., van Merriënboer, J. J. G., & Paas, F. (1998; 2019). Cognitive architecture and instructional design. Educational Psychology Review 10(3); "20 years later", 31, 261–292. link.springer.com
  7. Kalyuga, S., Ayres, P., Chandler, P., & Sweller, J. (2003). The expertise reversal effect. Educational Psychologist 38(1), 23–31.
  8. Barbieri, C. A., Miller-Cotto, D., Clerjuste, S. N., & Chawla, K. (2023). A meta-analysis of the worked examples effect on mathematics performance. Educational Psychology Review 35. doi.org/10.1007/s10648-023-09745-1
  9. de Jong, T. (2010). Cognitive load theory, educational research, and instructional design: some food for thought. Instructional Science 38(2), 105–134. link.springer.com
  10. Krieglstein, F., et al. (2022). Meta-analysis of the reliability and validity of cognitive load questionnaires. Educational Psychology Review. link.springer.com
  11. Sinha, T., & Kapur, M. (2021). When problem solving followed by instruction works: Evidence for productive failure. Review of Educational Research. doi.org/10.3102/00346543211019105 Ashman, G., Kalyuga, S., & Sweller, J. (2020). Problem-solving or explicit instruction: Which should go first when element interactivity is high? Educational Psychology Review 32.
  12. Atkinson, R. K., Derry, S. J., Renkl, A., & Wortham, D. (2000). Learning from examples. Review of Educational Research 70(2), 181–214.
  13. CESE, NSW Department of Education (2017). Cognitive load theory: Research that teachers really need to understand. education.nsw.gov.au
  14. Cowan, N. (2001). The magical number 4 in short-term memory. Behavioral and Brain Sciences 24(1), 87–114.
  15. Citation count: the 1988 paper's Semantic Scholar page (looked up 15/10/2026). Sweller profile: UNSW website.
  16. Secondary sources: ATD profile (Sweller's degree years); blogs embedding Dylan Wiliam's Twitter post of 26 Jan 2017 and an article on the Chartered College of Teaching site; World Economic Forum, swissinfo and the University of Texas reporting Sinha and Kapur (2021).