A polished answer can leave the room and still be the old habit with a cleaner surface. The person agrees with the new method. On the next real task they do what they already trusted. The quiz was fine. The work was not.
A National Research Council committee put the mechanism in the opening chapter of How People Learn, the expanded edition. People build a new understanding out of what they already know and believe. If that starting belief is never engaged, they may fail to grasp the new idea, or they may hold it long enough for a test and go back to the old one once the room is gone. The chapter organises the evidence as three findings. Learners arrive with preconceptions. Competence needs facts held in a framework, not a pile of items. People also have to learn to watch their own understanding, because that watching does not appear on its own.
The belief that rewrites the lesson
The chapter uses a children's book to show the reshape. A fish hears a frog describe birds, cows and people. It pictures birds as fish with wings, cows as fish with udders and people as fish who walk on their tailfins. The report is new. The model is not. Told the earth is round, children picture a pancake, because they can still imagine standing on it. Told that round means a sphere, they put the pancake inside the sphere or on top of it. The old model explained how a person stays upright, so the new fact was edited until it did too.
The persistence is not limited to children. The chapter describes a study in which physics students from elite, technologically oriented colleges played a game. Steer a simulated object into a target, and do it with as little speed as possible at the hit. They had a hands-on trial first, tapping a tennis ball with a small mallet. Elementary schoolchildren played the same game. Both groups failed. Success needed an understanding of Newton's laws. The college students, despite their training, aimed the moving object straight at the target and left momentum out. One student who was looked at more closely knew the formulas and the properties. In the game she used the conception she had before the course.
A common misreading, which the chapter corrects, is that a teacher should never tell anyone anything. That confuses a theory of knowing with a theory of teaching. People construct as they listen, including during a lecture. There are times, usually after they have grappled with the problem, when telling works well. The teacher still has to watch what they made of it. The picture of the learner as an empty vessel to be filled has to go. If the initial belief is ignored, the understanding they leave with can differ from the one the session intended, and a test will not catch the swap.
Why a list of prompts is not a skill
The chapter is plain that facts matter. Experts in chess, history, science and mathematics are not faster minds with nothing behind them. Their skill sits on a rich body of knowledge. Usable knowledge is not a list of disconnected facts. It is organised around important concepts, and it is tied to the situations where it applies. Experts do not necessarily have better memories overall. The framework lets them see a pattern, a relationship or a discrepancy a novice misses, and it lets them pick out what is worth keeping. Their attention is not used up on the whole scene.
A map makes the difference concrete. A student can memorise states, cities and countries and fill a labelled map accurately. Take the borders away and the task collapses, because no concept is holding the names. Someone who understands that borders often follow a mountain or a river, and that large cities often grew where trade could move, can still reason when the lines are gone. A workshop that hands over prompt patterns with no idea they belong to is that labelled map. Remove the slide and the next task has nowhere to sit.
The teaching implication is depth on fewer topics, not a light pass over all of them. A tour of every feature is coverage. It does not build the framework. For a working team the concept is a task they already owe: what the page is for, who will act on it, what must be true, what the tool cannot know and what would be expensive if it were wrong. Several passes on that concept beat one pass across the feature list.
The check they have to run alone
The third finding is about watching your own understanding. Experts who were asked to talk through their work monitored it as they went. They noticed when they needed more information, whether the new piece fitted what they already knew and which analogy would move them on. It is easy to treat that inner conversation as something adults already possess. The chapter says many of these strategies are learned, and that they reflect methods of inquiry a person has been shown. They can be taught. Predict what will happen. Explain the step so the explanation itself improves the understanding. Notice that you did not understand. Bring the relevant background forward. Plan the next move.
The sequence in the chapter matters. In the reading method it describes, the model is provided first by the teacher. Learners then practise the strategies and discuss them. Later they prompt themselves and monitor their own comprehension without the teacher there. The strategies are not generic across subjects. Teaching them as a free-floating skill can fail to carry into the work. A history student asks who wrote the document and what that does to the reading. A physics student asks whether the principle in play is the one they think it is. Those are different watches. A slide that says to be more critical is the generic version. The check has to live inside the task the team actually does.
There is no single best method. The chapter compares the question to asking whether a hammer is better than a screwdriver. A lecture can carry new information. It is a poor way to draw a preconception into the open, and a poor way to practise the monitoring. Hands-on work can ground a new idea. It does not, on its own, produce the concepts that let a person use the idea on a different task. A quiz of definitions tests recall. It does not test the belief that will run the next task.
What the agenda gives away
Read the proposed running order before you book. If it opens on the product and never asks what people already believe the tool is for, you are filling an empty vessel. The old model will still be there at the end, wearing the new vocabulary. If the practice tasks are invented topics, with no framework for the work this team sends, you have the labelled map. People will perform while the example is on the wall. If nobody runs the check with the facilitator silent, you stopped when the teacher was still supplying the model. The chapter's order is the other way. The belief is said. The concept is taught on the real task. The person takes the watch.
What a later task should show
The chapter reports a comparison that should change how you read a group who have already had a course. Sixth grade students taught physics through an inquiry approach, testing what they believed at the start and monitoring where they were as they went, did better on conceptual problems than eleventh and twelfth grade students in the same school system who were taught by conventional methods. A further comparison, of younger urban students taught that way against older suburban students taught conventionally, again favoured the inquiry group on the fundamental principles. Prior schooling was not the advantage. The way the belief was handled was.
That is the standard I would take into a proposal. Do not treat an earlier course as evidence the belief has moved. The college students knew the formulas and still aimed straight at the target. Score the next real task, not the smile sheet. Was the prior belief spoken, in the person's own words, before the method arrived. Can they state the framework for their own task without the slide. Did they run the check with the facilitator quiet, including on a pass where the fluent answer was wrong or thin. If they can recite a pattern and still send the first clean page, the old conception kept the job.
Be willing to cut the catalogue so one task is taught with enough cases to grasp it. The argument against covering everything is practical. A little of each feature leaves the defining concept ungrasped, and the person reverts when the slide is gone. Name the task. Name the belief you expect to hear about it. Name the check and the moment the facilitator stops prompting. If the proposal cannot drop the rest of the tour, it will cover features, and the feature tour is the labelled map with the borders still printed on it.
What the room has to be allowed to hear
The awkward part is the belief itself. People will say the tool is a shortcut, a search box or something they must not be caught using. Each of those will absorb a new method and keep its shape, the way the pancake absorbed the sphere. Correcting the sentence and moving on does not engage it. The belief has to be worked with: kept where the task shows it is sound, and dropped only when the person can see it fail. A room that is not allowed to say the unfashionable version will perform the new one and revert on the way out.
Collective Campus is building high-performance culture and teams for the AI world. The AI training workshop is for a team that can bring a live task and is willing to say what they already believe the tool will do with it. What changes when the practice includes a task the tool gets wrong is in how to judge the hands-on day. What one day should leave a named person able to run is in what a corporate AI training day should produce. Why a mixed room slides back to a generic demo is in role-based AI training.
What to bring
Bring the task the team will actually send next, the belief they already hold about the tool and the name of the person who reviews that work. If you cannot say the belief, the day will teach a method the old one can swallow.
