What is cognitive apprenticeship?

Cognitive apprenticeship is a model of instruction in which teachers make thinking visible: they model expert reasoning aloud, coach students as they attempt the work, scaffold the parts students cannot yet manage, and fade support as competence grows. Allan Collins, John Seely Brown and Ann Holum introduced the model to a wide audience in 1991.

Who developed the cognitive apprenticeship model?

Allan Collins, John Seely Brown and Susan Newman first described the model in a 1989 research chapter. The 1991 American Educator essay by Collins, Brown and Ann Holum, restored on this page, is the version most educators cite and assign.

What are the four dimensions of a cognitive apprenticeship learning environment?

Content (domain knowledge, heuristic strategies, control strategies, learning strategies), method (modeling, coaching, scaffolding, articulation, reflection, exploration), sequencing (increasing complexity, increasing diversity, global before local skills), and sociology (situated learning, community of practice, intrinsic motivation, cooperation).

How is cognitive apprenticeship different from traditional apprenticeship?

Traditional apprenticeship teaches physical work that is naturally visible, in the place where the work is used. Cognitive apprenticeship adapts the same arc to thinking, which is invisible: reasoning must be deliberately externalized, school tasks must be situated in meaningful contexts, and practice must be varied so skills transfer beyond a single setting.

What do modeling, coaching, scaffolding and fading mean?

Modeling: the expert performs the task and voices the reasoning while the learner observes. Coaching: the learner performs while the expert watches and guides. Scaffolding: the expert supports only the parts of the task still beyond the learner. Fading: support is progressively withdrawn until the learner works alone.

The Homework Question, Asked Honestly

Homework and AI meet at an awkward truth: the research case for homework was always weaker than practice assumed, and machine assistance has now removed the completion signal that made it administrable. What survives honest scrutiny is homework built on retrieval, reading, and preparation, kept short, and never graded as unsupervised proof of learning.

What the evidence showed before machines could do it

The homework debate did not begin with this technology, and pretending otherwise flatters the present. The most cited synthesis, by Harris Cooper, Jorgianne Robinson, and Erika Patall in 2006, reviewed the research through 2003 and found a positive association between homework and achievement that was moderate for secondary students and close to zero for elementary students, with the authors themselves cautioning that most of the underlying studies were correlational and could not establish that homework caused the gains. That same year two popular books pressed the case against, Alfie Kohn's The Homework Myth and Bennett and Kalish's The Case Against Homework, and the Initiative's archive holds a restored dual review of both that remains the best starting point on this site. The honest pre-machine summary was already uncomfortable: benefits concentrated in older students and diminished with excess, costs in family friction and inequity were real, and much assigned homework was busywork defended by tradition rather than evidence.

Two details from that synthesis deserve more attention than they usually receive. First, the correlational caveat is not a technicality. Students who do more homework differ from students who do less in motivation, home support, and prior achievement, and no synthesis of observational studies can cleanly separate what homework causes from what it merely accompanies. Second, the association was not linear. Cooper's work supported the rule of thumb he had long endorsed, roughly ten minutes per grade level per night, with returns flattening and then reversing beyond it, which means the marginal hour of homework defended in most staffroom arguments was already, on the evidence, the least defensible hour.

The same literature, read closely, showed which homework earned its keep. Practice that forces retrieval draws on the best documented effect in learning science: Roediger and Karpicke's 2006 experiments showed that recalling material strengthens retention more than restudying it, and spaced practice compounds the effect, one of Robert Bjork's desirable difficulties. Sustained reading, and preparation that lets the next lesson start further ahead, also survived scrutiny. What never had good evidence behind it was volume: hours of routine completion work, graded on submission.

What machine assistance actually changes

Large language models did not make homework useless. They made a specific management model of homework untenable: assign product, collect product, grade product, infer learning. An answer engine completes a worksheet, a reading response, or a standard essay in seconds, so completion no longer signals anything except that the student, or someone, submitted. This is the domestic case of the general collapse examined in Proof of Work: unsupervised product has stopped being evidence of process, and homework is the most unsupervised product a school assigns.

This is less novel than it feels. Homework's accounting function has been quietly compromised for generations by tutors, older siblings, helpful parents, and copied answers in homeroom; teachers have always known that a completed worksheet certified a household more than a mind. The machine did not introduce the corruption, it universalized it, and in doing so performed a service of a kind: a signal that is now visibly worthless can no longer be politely overvalued. Schools are being forced to decide what homework is for, a decision the evidence has been recommending since before the machines arrived.

Notice what is and is not lost. The learning function of good homework is untouched, because retrieval practice, reading, and thinking about the next day's material still work exactly as before, for the student who does them. What is lost is the accounting function, homework as a graded deliverable certifying that learning occurred at the kitchen table. Machine assistance also sharpens an old equity observation from the Kohn side of the debate: homework has always measured the home as much as the student, and a home's stance toward machine help is now one more thing it measures.

The teacher's evening is part of the ledger as well. Hours spent marking homework that certifies nothing are hours withdrawn from the planning and feedback that the effectiveness research actually rewards, and a school that keeps collecting worthless signal is spending its scarcest resource, teacher attention, on the least defensible line of the budget. Honest accounting runs in both directions: if homework no longer proves what students did, marking it no longer achieves what teachers hope.

There is a subtler cost, too, for adolescents especially. The evening hours homework colonizes are hours the research on adolescence has long argued are needed for sleep, autonomy, and the unstructured social learning that secondary schooling tends to crowd out. The Initiative's own tradition pressed this point for decades: adolescents are built to practise independence, and an evening regimen of compliance tasks practises the opposite. Homework that no longer even certifies effort has a weaker claim on those hours than it had in 2003, and the claim was under argument then.

What honest homework looks like now

Asked honestly, the question is not whether to abolish homework but what homework can still be for, and the answer is: for learning, not for accounting. The shape varies by subject but the grammar is constant. In mathematics, a nightly set of six mixed retrieval problems, answers printed at the bottom of the page, does more for retention than twenty graded exercises, and the printed answers remove the incentive to outsource since there is nothing to submit but one's own preparedness for tomorrow's similar problems in class. In languages, spaced vocabulary and structure practice is the single best documented use of out-of-class time, and its verification is built in: the student either produces the language in the next lesson or does not. In science and humanities, the strongest home tasks are inputs to class rather than outputs of it, reading with a question to answer, an example to find, a position to stake, because a discussion consumes preparation in a way no machine substitute survives. Across all of them, sustained reading survives on its own merits, verified by what students do with it in class rather than by summaries a machine writes better. And where a longer product must be built at home, the process expectations described in our assessment essay travel with it: drafts, disclosure of assistance, and a few minutes of talking through the work. A teacher deciding where machine help belongs in any of this will find the fuller framework in AI Literacy for Educators.

One objection deserves its own paragraph: does not all this simply concede the evenings to the machine? It does the opposite. A student assigned six retrieval questions with answers attached has no reason to consult a machine, because the task's whole value is private and the machine's contribution would be nothing. A student assigned a collectible essay has every reason. The concession to the machine is built into the design of outsourceable homework, not into the decision to stop assigning it. Schools do not defend the evening by policing it harder; they defend it by assigning work whose worth cannot be delegated, and by being willing to say aloud that the rest was never worth the argument.

Implications for practice

Stop grading homework completion as evidence of learning; the signal is gone and was never strong. Assign less and design more: ten minutes of retrieval practice with answers available beats an hour of collectible product. Move certification into the classroom, where process is visible, and let home study be what the evidence always said it was at its best, rehearsal and preparation. Tell students plainly why unassisted recall is worth their effort when a machine could answer instantly: because the effort is the mechanism, and what the machine does for them, their memory does not do at all. A school that makes these moves loses nothing but paperwork, and recovers the one claim homework could always honestly make, that practice, done by the learner, works.