My son is seven. The kids do their own breakfast most mornings, and this morning there was milk on the bench when I came through. He had poured his cereal into a cup instead of a bowl. Going to fill his sister’s, he knocked the cup over. He picked it up, started again, knocked it again, got milk on the bench and down his shirt, wiped the bench with his hand, then wiped his hand on his top.
I got frustrated with him. He got defensive, produced a run of reasons why none of it was really his doing, and ended up in his room upset. I went in after a bit and told him I should not have talked to him like that, and that he was allowed to be angry about it. Then I asked him why he had tried to wipe the milk with his hand and hide it, because milk goes off and his shirt was going to stink either way.
He did not have an answer for that.
What he did have was one last reason, and it was the one holding all the others up.
There were no bowls.
There were bowls. They were in the sink and they needed washing.
As soon as I said that, everything he had built went quiet. Not because I had won the argument, but because the thing it was resting on turned out not to be there.
Two ways of going backwards
Later the same morning he could not find his shoes. I asked him where they were when he last had them. He said he could not remember, that they were lost, that he had looked everywhere. Then he stopped, thought about it, and remembered he had put them in his school bag.
That is where they were.
Both episodes involved going backwards from the situation he was in now toward whatever had put him there. The difference was that one route was allowed to discover something and the other was trying to close the question.
There were no bowls was an explanation that arrived finished. It worked for as long as nobody checked the sink. Where were your shoes when you last had them? stayed open long enough for the answer to change.
That seems like a small distinction, but I think a great deal sits inside it. One way of going backwards is a retrace: each step remains answerable to what actually happened. The other produces an account that makes the present position make sense. An account can be perfectly coherent and still rest on something that is not true.
Children do not need to be taught how to do the second one. They work it out. If being right feels safe and being wrong does not, then producing a reason that protects the position you are already in is a useful skill. That is not some unusual fault in a seven-year-old. It is a sensible response to the conditions, and adults have a great deal to do with setting those conditions.
The useful thing is not to make being wrong comfortable by pretending it does not matter. It is to make being wrong cheap enough that there is no need to defend it.
The cup was not the mistake
The cup is worth looking at again, because the obvious reading is that he made a silly choice and then made a mess.
But the cup was already a solution.
The bowls needed washing and he did not want to wash one, so he found another container. That solved the immediate problem and let breakfast continue. Then the cup tipped, which created the next problem. He picked it up and tried again. It tipped again. There was milk on the bench, so he wiped it with his hand. Then there was milk on his hand, so he wiped that on his shirt. Then there was a mess to hide and an explanation to produce.
Every step made some sense in relation to the thing immediately before it.
Not one of them touched the sink.
That shape turns up everywhere once you start noticing it. Something underneath has not been done, built or carried, but the next thing proceeds anyway. A workaround gets it moving. The workaround produces consequences, and those consequences are handled with another local fix. Before long a great deal of intelligence and effort can be going into managing effects several steps removed from the original absence.
From inside the sequence, every decision can be reasonable.
The sequence can still be wrong.
Below the words
A while before that I was throwing a Fruit Burst at my kids in the lounge because there was no ball to hand. I told them to keep their eye on it.
On one throw I faked the release. Their eyes shut immediately.
On another I moved the sweet sideways and their eyes stayed on my face rather than following the thing I had thrown.
They understood the instruction perfectly. They could repeat it back. They were trying to follow it. It failed anyway, twice, and for two different reasons.
The first failure was sitting underneath the instruction. Something coming toward their face made their eyes close. There was no better wording that was going to talk them out of that reflex.
The second was different. They were watching me because I was the person doing the throwing. They had not yet separated the object that mattered from the person who had set it moving. Keep your eye on the ball assumes that distinction has already been made. It sounds like one instruction, but it rests on several capacities that the instruction itself cannot provide.
That is true of a lot of teaching. We name the visible action because that is the part we can see: catch the ball, cut along the line, hold the pencil properly, listen carefully, check your work. But the visible action may depend on things underneath it that are still unreliable.
Those things do not necessarily form a neat ladder. Children develop messily. Capacities overlap, support one another and sometimes arrive together. But some are load-bearing enough that when they are missing, the same absence appears through several different tasks.
A child struggling with balance, visual tracking or timing may look as though they are bad at catching. A child whose small hand movements are not yet well calibrated to what their eyes are seeing may struggle with scissors, threading, handwriting, rulers, buttons and drawing. On the surface those can be treated as separate problems. Underneath, they may be several expressions of the same thing not yet holding.
There is a signature to that kind of failure. The same fault keeps returning. Repetition stops paying. The correction seems to be understood but does not take. You keep working on the thing you can see and it keeps failing for reasons that instruction at that level cannot reach.
That is when it becomes worth asking what the task is resting on.
When the gap is found late
My wife teaches intermediate. Some of her twelve-year-olds cannot thread a needle or use a pair of scissors confidently.
By twelve, that is no longer only a question of whether they can learn the movement. There is now a social cost to learning it. A five-year-old can be visibly bad at scissors without much consequence. A twelve-year-old being asked to practise something a much younger child can already do is being asked to display that they are behind.
Refusal makes sense from inside that position. Declining costs very little today. The cost of not learning is distant, invisible and attached to future tasks that nobody will necessarily connect back to the scissors.
The teacher is under a different constraint. There is a class to run and a lesson to finish. There may be no room in the timetable to go back several years and build something that was assumed to be there already. So the practical response is accommodation: do part of it for the student, change the task, provide a tool that removes the requirement.
Sometimes that is exactly the right thing to do.
But accommodation and construction are not the same operation.
The task gets completed while the capacity underneath it remains unchanged. The next task that needs the same capacity arrives, the gap appears again, and another accommodation is made. Every step is locally sensible. The result can be that a capacity nobody has time to go back and build becomes a capacity nobody is responsible for building.
Eventually the workaround stops looking like a workaround. It simply becomes how the system operates.
What absence costs
That gives another way to find what is missing.
Look at what is being spent to manage its absence.
Extra supervision. Repeated explanations. Special processes. Rework. Workarounds. Someone else quietly doing the difficult part. A tool introduced to remove a judgement the person never got the chance to build. Meetings about recurring failures that are treated as separate events because the thing connecting them sits underneath all of them.
The absence itself can be hard to see. Its maintenance cost usually is not.
That morning, the cost of not washing one bowl was not one dirty bowl. It was the cup, the first spill, the second spill, the milk on the bench, the milk on the shirt, the attempt to hide it, the argument and the reasons afterwards.
The original cost had not disappeared.
It had travelled.
The same thing at larger scales
The same pattern becomes harder to see as the system gets larger because the distance between the missing condition and the visible consequence grows.
A campaign line like fix the basics and build the future names an outcome while leaving almost everything that has to carry it unstated. That is part of why such lines work. There is very little in them that can be found wrong.
The difficulty begins when the slogan meets particular people and places.
Stop an industry and the cost does not land on an abstraction. It lands on crews, contractors, families and towns that can be named and counted. That does not settle whether the industry should continue. It means that ending one thing and carrying people through its ending are different operations. If the second part is absent, the policy has proceeded as though a condition underneath it had already been satisfied.
The specification is often the cheap part.
The carry is where the cost becomes real, because naming what has to carry an outcome usually means naming who carries it, for how long, and at whose expense.
AI and judgement
AI makes this question especially interesting because it can remove difficulty from exactly the places where judgement used to be formed.
I have worked in architectural design for about ten years, but the longer thread for me is not really professional experience. I have spent most of my life using systems and tools that respond to what I do.
Video games are an obvious example. You make a move and something happens. You miss the jump, lose the fight, build the wrong thing, take the wrong route, mistime an action. The system does not need to explain that your response was suboptimal. Your response meets a consequence, and you learn from the difference.
A great deal of judgement is built that way. You act, the world answers, you adjust. Over time you become sensitive to differences you could not have been taught as rules because you had to encounter them before they became visible.
AI is different in one important respect.
It talks back.
It can produce an answer and then produce an explanation for the answer. If you object, it can produce another explanation. It can agree with your correction, improve the wording, reorganise the reasoning and make the next version sound more complete than the last. The response does not simply meet a consequence outside itself. The system can participate in telling you what the consequence means.
That is enormously useful, but it changes the feedback loop.
For the last ten months or so I have used AI intensively enough that a working language has developed through the interaction. I know more now about how to question it, when to distrust fluency, when an answer has actually moved and when it has merely been rewritten, when to push further and when to go back to whatever the claim is resting on. Some of that judgement has been formed through using the tool itself.
But that only works if something in the interaction remains answerable to more than the conversation.
Otherwise there is a danger that a coherent response is mistaken for a checked one.
AI can generate there were no bowls extremely well.
Not because it is trying to deceive anyone, but because producing a plausible account and retracing something until it meets what is actually there are not the same procedure. A person using the tool needs to know when to keep going backwards, when to check the sink, and when a better explanation is not the thing that is needed.
The question is not whether AI produces good work. It often does.
The question is what capacities we are allowing it to replace before those capacities have been built, and whether the person receiving a good answer can recognise the occasions when it is wrong.
Output and judgement are not the same thing.
Takes
This is where I think the word mistake causes trouble.
On a film set, a take can fail without the failure having to transfer onto the person. The take did not work. Run it again.
Nothing about that requires lowering the standard. A failed take is still rejected. Nobody needs to pretend it was secretly good because mistakes are part of learning. The work can be wrong without the person having to become wrong with it.
That distinction matters because learning needs both correction and enough safety to remain exposed to correction.
If being wrong carries too much weight, the sensible response is defence. Explain it. Hide it. Avoid the task. Produce an account that closes the question.
If the standard disappears, nothing gets built either. Mistakes are learning becomes a reason not to say that something did not work, and the learner loses the information they needed.
Keep both.
This one does not work.
Go again.
The reason I got to five throws with the Fruit Burst was that we were laughing. Nobody was defending anything. A missed catch had almost no cost, so another attempt could start immediately.
Then I stopped, because five was enough and a sixth was beginning to become a different thing.
Repetition is not automatically learning. There comes a point where another attempt stops adding information and starts adding weight. Knowing where that point is cannot always be reduced to a method either. It is another judgement built by having done it, by having got it wrong, and often by remembering what it felt like when somebody else went one round too long.
Three questions
What does this rest on.
Was that thing ever actually built.
If it was not, what is now being spent to manage the absence.
Run those questions on a classroom, a policy, a project programme, a piece of technology or a seven-year-old with a cup of milk, and the answer comes back in the same shape often enough that it stops feeling accidental.
Something underneath was missing or unreliable. The next thing proceeded anyway. A workaround carried it for a while. Then more effort was needed to carry the workaround, and more again to manage what followed. Eventually the activity above the gap became so normal that the gap itself was difficult to see.
And the thing missing underneath is often one of the least impressive things to build.
Repeated attempts. Failure that stays visible. Correction. Another attempt. Someone nearby who can see what actually failed. Enough room to be wrong without needing to turn the failure into a defence of yourself.
It is slow, awkward work. It produces things that are visibly not good enough. It is easy to remove in the name of efficiency, kindness, technology or simply getting on with the next thing.
But some capacities are built almost entirely there.
Once they have been skipped, explaining them later is not the same as having built them.
There were bowls.


