AI-Proof: What Machines Can Never Be — Volume I: Awakening, Part III: The First Awakening, Chapter 10: Building Deep Human Capital

Part III looks at the person doing the work. Chapter 9 asked how to decide what the work is for. This tenth chapter asks what the person doing it is made of, once a machine can give anyone the knowledge.

← Previous: Chapter 9: The Architect Mindset · Series hub · Next: Chapter 11: Judgement: The Skill That Compounds →

A machine now explains science to Kavitha’s class more patiently than she can. When anyone can have the knowledge, what is depth? An old Indian teaching answers: hearing is never enough. Depth is what knowing does to you.

In Brief: For readers who wonder what their expertise is worth when a machine can explain their field to anyone. Chapter 10 asks what depth means once a machine can give anyone the knowledge. A science teacher in Sagar watches a platform teach her class, and a toolmaker in Nagoya, a planner in Medellín and a sensor engineer in Shenzhen each find their depth tested by a machine that can do part of their work. Yājñavalkya’s three steps of knowing (hearing, reflecting, living with), Confucius on knowing what you do not know, Aristotle’s first principles, the Nyāya logicians’ hidden condition and the strange story of N-rays help answer it, without promising that depth will pay or that it brings the right to decide.

A machine can hand anyone the knowledge; depth is what the knowledge does to you when you hear it, test it and live with it, and no one, human or machine, can do that part for you.

Opening: The Platform at the Back of the Room

By every measure Kavitha had, the platform taught the chapter on tissues better than she did.

It arrived in August 2025, part of a district pilot sent down from the education office. Twice a week her forty-three Class 9 students shared tablets in pairs and worked through its lessons on plant and animal tissues. Kavitha sat at the back of her own classroom and watched. The platform explained what xylem does, and then explained it again in different words to the students who had not understood. It asked questions, read the answers and chose the next question to fit. It did not get tired on the fourth repetition. It did not lose its patience with the boy who kept tapping the wrong picture. Two boys who had failed the July unit test were now passing its checks.

Kavitha Sharma teaches science at a government school in Sagar, in Madhya Pradesh. This was her twelfth year of teaching. She was not afraid of machines. For more than a year she had used AI tools to draft lesson outlines and the notes she sends to parents. Chapter 1 told how she made a rule about them: nothing would leave with her name on it that she had not read and could defend. Chapter 8 told how that rule caught a study the machine had invented about the neem tree in the school courtyard. Those tools had helped her do her work. This one did the part of her work she had always thought was the centre of it. It explained. And nobody had asked her whether it should come, or what her periods were now for. That was decided in the district office, as the retirement of Priya’s module in Chapter 9 was decided in a meeting she was not invited to.

That evening she marked notebooks at the kitchen table while her mother watched the news and her husband read the paper. She found she was asking a question she had never needed to ask. Her students could now get the knowledge from a screen, as patiently and as often as they wanted. If knowledge was what she had been giving them, the machine gave it better. And if it was not knowledge, what had it been?

The question is not only a teacher’s. As of 2026, a machine will explain law to a junior lawyer, chemistry to a nurse and accounting to a shopkeeper, at any hour, at the level each one needs. Explanation used to be scarce, and part of what experts were paid for was explaining. Explanation is becoming cheap. But receiving an explanation is not the same as understanding it, knowing when to trust it, or seeing when it does not apply. So this chapter asks: When a machine can give anyone the knowledge, what is depth, and how is a person to build it?

The first answer most people reach for is: know more. Go deeper than the machine. The print edition of this book said something close to that. It said that deep human knowledge lives in the body, takes years, and is something “no platform, no tool, and no dataset could replicate”. Chapter 2 has already shown why that cannot be the answer. Knowledge that people could not put into words once seemed safe from machines. That safety has gone. Machines learn from examples, and they have learned many things no one could write down. If depth means holding knowledge the machine lacks, depth is a shrinking island. The chapter has to look for it somewhere else.

10.1 Domain Mastery: What Practice Builds

In a toolmaking shop in Nagoya, Ishikawa Kenji runs his thumb along the inside curve of a die. A die is a hardened steel form. This one will press steel brackets for a supplier to a car maker, many thousands of times a day, for years. Kenji has been making dies for twenty-six years. This one has taken him three days.

The shop’s measuring machine has already passed it. That machine touches the surface with a fine probe at hundreds of chosen points and compares each point with the drawing. Every point is within the limit. Kenji’s thumb stops near the third curve. There is a faint ridge where the curve meets a flat face, too small to see. It lies between the points the probe touched.

The print edition told this story as a victory of the thumb over the instrument. This edition tells it differently. In 2025 the shop bought an optical scanner that measures the whole surface, not chosen points. When Kenji asked for a scan, the scanner found his ridge. It also found a second ridge, smaller, on the other side of the die, which his thumb had missed. On the job of detecting a ridge, the thumb is no longer the better instrument. It was the better instrument for a short time only, while the measuring machine sampled points.

So what does Kenji know that the scanner does not? The scanner reports the ridge and says it is within the limit. Kenji knows something about what a ridge like that becomes. Twice in twenty-six years he has seen one turn into a crack, after a few hundred thousand strokes in a customer’s press. The die fails, the press stops, and someone in another company’s factory loses a day. Being within the limit is a fact about the drawing. Whether the ridge matters is a judgement about what will happen under load, in a press he will never see.

Be honest about where that knowledge lives. As of 2025, nothing in his shop links a ridge on a new die to a crack two years later in another firm’s press. The shop records what it ships. It rarely hears what happens afterwards, and when it does, nobody writes the reason next to the die’s number. Kenji’s knowledge of consequence lives in his memory partly because the institution never kept the record. If the records were kept, a model could learn what Kenji learned, and from more cases than one man will ever see. Part of what looks like his depth is a missing record. That part is not safe.

Say the rest plainly. A record nobody keeps does two things. It makes Kenji look like the only person who could know, and it keeps the next toolmaker from learning what he learned. If the shop wrote the reason next to each die’s number, a junior could check a ridge against more cases than Kenji has seen, without waiting twenty-six years. A kept record would be one way to rebuild the missing rung of Chapter 4: the beginner’s work, now done by machines, on which the next person used to learn. And even with the record, someone would still have to decide how much risk is acceptable in a customer’s press. No measurement settles that.

He has also been wrong. Early in his career he reworked a surface that did not need it. It cost his employer a day of production, and his supervisor said so. That mistake did not make him less careful. It taught him to test his own alarm. Some feelings come from the steel. Some come from his own nerves on a bad day. Telling them apart took years, and he still does not always manage it. That lesson is his. The unrecorded crack is the shop’s.

An Indian text describes three steps of knowing that fit Kenji’s years closely. In the Bṛhadāraṇyaka Upaniṣad, one of the oldest Upaniṣads, the sage Yājñavalkya is about to leave home and renounce the world. He wants to settle his property between his two wives. One of them, Maitreyī, asks whether wealth could make her free from death. He says it could not. Then he teaches her. Ātmā vā are draṣṭavyaḥ śrotavyo mantavyo nididhyāsitavyo maitreyi. In Patrick Olivelle’s translation: “it is one’s self (ātman) which one should see and hear, and on which one should reflect and concentrate.”¹

The philosopher Śaṅkara, writing about twelve centuries ago, turned those verbs into an order. In Swami Madhavananda’s translation of his commentary, the self “should first be heard of from a teacher and from the scriptures, then reflected on through reasoning, and then steadfastly meditated upon.” Only when the three are combined, he adds, does true knowledge come, “not otherwise, by hearing alone”.² The tradition names the three steps. Śravaṇa is hearing: receiving the teaching. Manana is reflecting: testing what you heard against reasons and doubts until the doubts are answered. Nididhyāsana is living with it: holding the knowledge steadily until it stops being something you know about and becomes the way you see.

I am using this teaching far from its home, and the reader should know how far. The Upaniṣad is about the self, the one thing it says cannot be known as information. Its tradition would put die-making in what Chapter 2 called the lower knowledge, everything that can be written down and taught. Śaṅkara’s third step is steadfast meditation on the self. Twenty-six years at a die are not that. So I borrow the order of the steps, not their object, and I do not claim that the Upaniṣad offers a theory of how people learn a trade. Borrowed this way, the order fits Kenji. He heard: from senior toolmakers, from manuals, from the properties of each grade of steel. He reflected: he tested what he heard against what happened, and the day he wasted taught him to argue with his own feeling. And he lived with it for twenty-six years, until the knowledge was not something he consulted. It was in how his thumb read the steel. What the borrowing earns most clearly is Śaṅkara’s warning, and it is a warning for the age of the platform: not by hearing alone.

Confucius drew a similar line. In James Legge’s translation of the Analects: “Learning without thought is labour lost; thought without learning is perilous.”³ The two passages agree on the first two steps. Learning without testing is wasted, and testing without learning is dangerous. Read side by side, they lean different ways about what comes after. The Upaniṣad’s third step turns inward, until the knowledge has changed how the knower sees. A later passage of the Analects, used in 10.4, turns outward, to honesty about the edge of what you know. That is my reading of a few passages, not a summary of two traditions, each of which says much more. For a toolmaker’s shop, the second test is the more useful one.

Two ideas about mastery need correcting: that AI cannot shortcut it, and that it takes ten thousand hours. A kept record is one shortcut, as Kenji’s shop shows. The idea that mastery takes ten thousand hours was made popular by Malcolm Gladwell, drawing on the work of the psychologist Anders Ericsson. Ericsson himself rejected the number. “It may not take exactly 10,000 hours,” he said, “but it will take a lot.”⁴ What he studied was deliberate practice: practice designed to improve one thing, usually by a teacher, with quick feedback on every attempt. Hours of doing the same thing do not count.

In 2014 three psychologists, Brooke Macnamara, David Hambrick and Frederick Oswald, combined the studies of practice and performance. Practice explained about a quarter of the differences between people in games, about a fifth in music and slightly less in sports. In education it explained 4 per cent. In the professions it explained less than 1 per cent.⁵ Ericsson objected to this study, and to a later one by two of the same authors.⁶ They had counted hours of any kind of practice, he said, not deliberate practice designed with a teacher and feedback.⁷ The dispute is not settled. Two things survive it. Hours alone do not make depth. And in most professions, the clean, quick feedback that deliberate practice needs is rare. That second point will matter when the chapter returns to Kavitha. There is also a limit to what the figures can say. As I read them, they compare people with one another. They do not measure how much practice changes the one who does it, and that change is what this chapter calls depth.

Now put the machine beside each of the three steps.

Śravaṇa, hearing, is what the machine does best. It is the most patient explainer of facts ever built. It explains at any level, as many times as needed, in many languages. The platform in Kavitha’s classroom is śravaṇa at a scale no school has had.

Manana, reflecting, is where the machine can help or harm. Chapter 4 told of two studies. Students given a machine that supplied answers did better in practice and worse in the exam taken without it. Students given one that offered only hints did better in practice and no worse in the exam. Among junior programmers, those who used the machine to ask questions about the code seemed to keep more than those who handed it their work, though that comparison was not a controlled test. The machine can be the sharpest partner you will ever argue with. It can also do the arguing for you and leave you with nothing.

Nididhyāsana, living with it, cannot be done for you. This is not a claim about what machines cannot do, which would need a date. It is a claim about what the word means. To live with knowledge until it changes how you see is a change in you. Another person can no more do it for you than they can sleep for you. A machine may have something like depth of its own. That would be the machine’s, not yours.

Why should Kenji’s depth matter, if the scanner sees better? Not because it makes him hard to replace. It may not. It matters because it lets him meet the conditions Chapter 7 set for a full position. He can check the scanner, because he knows what the steel does; Chapter 8 said to believe a machine only as far as you can check it. He can ask the question the scanner was never asked: does this ridge matter under load? And when he signs the job card, he can read the evidence behind his signature. A signature from someone who cannot read the evidence is a name on a form.

The converse is not a job. Depth can fill a position. It does not grant one. The shop decides whether Kenji keeps the right to stop a die, and whether he has time to look. It decides whether a day lost on a ridge that proves harmless is held against him. Sanjay, in Chapter 6, could read the image and was moved to a room he did not choose. Priya, in Chapter 9, could read the plan and was not in the meeting. Depth is a capacity in a person. The chance to use it is a condition that an institution gives or withholds.

10.2 Cross-Disciplinary Thinking: Where Originality Lives

Kenji’s knowledge is deep and narrow. For most people, depth in one field is not enough. The broad middle of a profession, the competence that takes five to ten years, is exactly where the machine arrives first. The print edition’s advice was to go sideways. That advice needs testing too.

Camila Restrepo is an urban planner in Medellín, Colombia, with fifteen years of work behind her. Some of it was in Bogotá and Buenos Aires, and one year was with a development charity in Nairobi. In 2025 her firm was asked to propose a way for people to reach the city from a hillside neighbourhood on Medellín’s edge. On the firm’s maps the neighbourhood was almost blank. It had grown without permission, house by house, over decades. No road on it was wide enough for a bus. People went up and down by footpaths, concrete stairs and shortcuts between houses.

Medellín knows this problem well. In 2004 the city opened a cable car line that carries people from the valley floor to hillside districts.⁸ In December 2011 it opened a chain of outdoor escalators in Comuna 13, one of its poorest districts. They run 384 metres up the hill, a climb that officials compared to a building of twenty-eight floors. A walk of thirty-five minutes became a ride of six.⁹ Camila’s engineering colleagues proposed extending a cable car line. The firm’s AI planning tool, given the slope and where people lived, proposed an escalator line like Comuna 13’s, along the steepest straight path. Both proposals were sound.

Camila spent three weeks walking the paths. She sat in the small shops where people stop on the way up. She talked to women carrying shopping and to men coming home from work. She listened to the long, tangled things people say when nobody is filling in a form. She learned that the paths were not only a way to get somewhere. An old woman sat on her step at one bend every morning and watched the children going down to school. A vendor sold arepas at another bend where everyone passed. At night the stairs were safer because people were always on them. A straight line of escalators would carry people past all of it.

She proposed something less dramatic. Keep the path network. Improve it where it is worst, with lighting, handrails, places to rest and drains for the rain. Add mechanical help only on the two steepest stretches. Join the network to the cable car below. And let the neighbourhood’s own elected board, the junta de acción comunal that Colombian neighbourhoods have, choose among the options. Camila did not invent escalators or cable cars. Medellín had built both years earlier. What she added was knowledge of this hillside.

Her plan had costs too. Paths with two short stretches of escalator leave most of the climb to people’s legs, which matters for the old and the sick. A cable car costs money to run for decades. Each option shares out benefits and burdens differently. Her proposal was better informed. That does not make it certainly better.

Here is the part the print edition left out. Camila typed her field notes into the planning tool: counts of people on each path at each hour, short summaries of her conversations, the places where people stopped. In a few minutes the tool proposed something close to her design. It also noticed something she had missed. One stretch of path she wanted to keep crossed a gully that, by the city’s drainage maps, flooded in heavy rain. She had never walked it in a storm. She asked the residents. It flooded. Given her material, the machine combined planning, engineering and the residents’ habits quite well. The combining was not only in her head.

So where was her contribution? In the material. The residents told her what they would never tell a survey, because she came, sat, and came back. That is standing, not capability. And her years in two kinds of work, planning and the patient watching of how people live, told her what was worth writing down. The decision itself was not hers either. It belonged to the people who live on the hill. The Prologue called them the third chair: the people whose lives a decision is about. The residents were not one more dataset to feed the model. Their knowledge helped show what each design would do. Their standing came from something else: they would live with the result. Choosing was theirs. If the path they chose flooded, the firm that drew it and the people who built it would still answer for that.

Charles Darwin is the usual example of what happens between fields, and the story is true. In October 1838, he wrote later, he “happened to read for amusement ‘Malthus on Population’”, the economist’s argument that populations grow faster than food. He saw at once that in such a struggle useful variations would be kept and harmful ones lost. “Here then I had at last got a theory by which to work.”¹⁰ But notice the other half of the story. Twenty years later, Alfred Russel Wallace lay ill with fever in the Moluccas, the islands of what is now eastern Indonesia. He too remembered Malthus, and reached the same theory on his own.¹¹ The combination was there for anyone who held both fields. That should make us careful about calling such combinations rare and human. Machines combine fields all the time.

The best evidence on where originality comes from is less romantic. In 2013 Brian Uzzi and his colleagues studied 17.9 million scientific papers. The papers with the most influence were mostly built on very ordinary combinations of earlier work, with a small intrusion of unusual ones. Papers with that mix were about twice as likely to be among the most cited. Papers that were unusual without the ordinary grounding did no better than average.¹² The study measured papers, not people. I read it as a hint about people too: deep, ordinary grounding may help a person tell which unusual connection matters. It is a hint, and only a hint.

There is an Indian parable about this, and it is less flattering than it looks. In the Udāna, a collection of the Buddha’s sayings, a king of Sāvatthī has the people of the city who were blind from birth brought to an elephant. Each touches one part. The one who touched the head says an elephant is like a jar. The one who touched the ear says a winnowing basket; the one who touched the tuft of the tail, a broom. They begin to argue, and then to fight. The Buddha compares them to teachers who cling to their own views: “people seeing one side”.¹³ The parable is told about religious teachers, not professions, so applying it here is my extension. It is often retold as if the lesson were: be the one who sees the whole. Read it again. The king who sees the whole elephant does not help the blind men. He watches them fight, and the text says he was pleased. Seeing more than one side is not the same as seeing the whole. Partial views can correct one another, as the drainage map corrected Camila and the residents confirmed it. Putting them together does not give anyone the right to decide. Camila had touched more parts of the elephant than her colleagues. She still had to ask the people on the hill what they had touched.

There is a cost to working across fields, and it is paid in the currency of careers. Camila has no qualification in the study of how people live together. She read the methods at night in Buenos Aires, and practised them in Nairobi because the charity’s director insisted. In the systems that decide hiring and promotion, she is a planner who dabbles. Depth in one field shows on paper. Depth across two is harder to show, and it is often seen only after the work is done.

10.3 First-Principles Reasoning: Thinking from the Ground Up

Zhang Rui runs a firm of twelve people in Shenzhen, in an office above a shop in the electronics district. The firm makes low-cost dust sensors for factories. Inside each one, a small laser shines across a stream of air. When dust passes through the beam, it scatters the light, and a detector counts the flashes. From the flashes, the sensor estimates how many micrograms of fine dust are in each cubic metre of air. The sensors cost a small part of what the reference instruments cost, and hundreds of factories in the Pearl River Delta use them.

In the spring of 2025 the complaints began. A factory manager in Dongguan, a city an hour away, sent a message that Zhang Rui later printed and pinned above his desk. On the worst days of the huí nán tiān, she wrote, his sensors reported the dirtiest air of the year. Huí nán tiān, “the return of the south”, is what people in southern China call certain days in early spring. Warm, wet air from the sea meets walls still cold from winter, and the walls sweat. Floors are slippery with water. On those days, she said, her workshop was cleaner than usual, because two of the furnaces were off.

For six weeks the firm did what firms do. Its engineers adjusted the calibration for each site. They borrowed a reference instrument and trained a correction on its readings. AI tools helped them fit better curves. The fit improved at one factory and failed at the next. Someone proposed a heater on the air inlet to dry the air before it reached the laser, as the reference instruments do. It would have added power, cost and a part that could fail.

Zhang Rui went back to what the sensor actually measures. It does not measure dust. It measures light scattered by whatever crosses the beam. The firm turns that light into a weight of dust by assuming particles of a typical size and density. That assumption holds when the air is dry. In very humid air, many fine particles take up water and swell. Swollen particles scatter more light. In fog, droplets of water scatter light too. The sensor sees more light and reports more dust. Nothing was wrong with the laser or the detector. The weakness was in the inference from light to dust.

Scientists who test such sensors had found exactly this, and published it. One study in 2018 found that a low-cost counter, at humidity above about 85 per cent, read 2.5 to 3.9 times the true amount. A correction based on how much water the particles take up brought its readings within a third of the reference.¹⁴ Zhang Rui’s firm added a cheap humidity sensor to each unit. It applied a correction of that kind. It marked readings taken in fog as unreliable instead of reporting them as dust. And it began telling customers, in plain words, what the number on the screen means and when not to believe it. The correction is not perfect. How much water particles take up depends on what they are made of, so it has to be tuned to the air of each place.

Aristotle opens his Physics with the idea behind the phrase “first principles”. In R. P. Hardie and R. K. Gaye’s translation: “we do not think that we know a thing until we are acquainted with its primary conditions or first principles, and have carried our analysis as far as its simplest elements.”¹⁵ On this view you understand something when you can build it up from what it rests on.

The Indian logicians of the Nyāya school asked a different question, and it fits Zhang Rui’s case even better. Their textbooks teach a sound inference: the hill has fire, because it has smoke. Then they turn it round: “The hill has smoke, because it has fire.” It sounds just as good. It is wrong. The logicians’ own counter-example is a red-hot iron ball, which has fire and no smoke. In their account, smoke goes with fire only where the fire is in contact with wet fuel. That contact is a hidden condition the inference silently needs. The Nyāya name for such a condition is upādhi.¹⁶ An inference holds only when no such condition is hiding behind it. In the textbooks’ standard example, the hidden condition is water. In Zhang Rui’s sensor, it was water again.

The two lamps disagree about where to start. Aristotle goes down to the simplest elements and builds up. The Nyāya logician does not start from nothing. He starts from the inference you already make every day and asks what condition it quietly depends on. Zhang Rui did the second. “First principles” in business talk often sounds like starting from nothing, as if a clever person could rebuild a field in an afternoon. What he did was smaller and harder. He found the condition the whole design had silently assumed, and asked what happened when it failed. His engineers had been adjusting the curve. He changed what the curve was taken to mean. That needed a working picture of the instrument as well as a willingness to doubt.

Now the machine’s side. The answer was published years before the complaints. If anyone at the firm had asked a model in 2025, “Why do optical dust sensors read high in humid air?”, it would have answered correctly and offered references. Some of those might have been invented, like the study about the neem tree that Kavitha caught, and each would have needed checking. The explanation itself is the one the published studies give. The print edition said that what the machine did “not yet” do was feel “the frustration that generates the question”. That misses the point. The machine was not missing the knowledge. The firm was missing the question. Its tools had been asked how to fit the numbers, not what the numbers meant. The question came from a factory manager in Dongguan who knew her own workshop. It became a question, and not a nuisance, because Zhang Rui knew what the instrument actually does. Without her message, the question might never have come. Without his depth, her message would have been one more complaint about calibration.

This too carries a date. A system that read every complaint from every customer might have found the pattern between the bad readings and the weather by itself. As of 2026, systems that watch for such patterns are improving quickly. The question will come from the machine more often. Someone will still have to know what the instrument measures, to tell a real pattern from a lucky one. And knowing is not deciding. If the firm’s owners had chosen to keep shipping the uncorrected sensors to meet an order, Zhang Rui’s understanding would have changed nothing on the factory floor in Dongguan.

10.4 Pattern Recognition: The Silent Teacher

Kavitha had always read faces. After twelve years she could look across a room of forty-three fourteen-year-olds and feel, she believed, who had understood and who had not. She had never been able to test it.

Chapter 3 described when such a feeling can be trusted. The psychologists Daniel Kahneman and Gary Klein agreed that skilled intuition needs two things: a world that gives clear signals, and enough chances to learn them with feedback. A classroom is poor in the second. A teacher reads faces all day and almost never learns whether she read them right. The unit test comes weeks later and measures many things at once. A face she misread in August is forgotten by October.

The platform changed this, and Kavitha noticed it in early September. At the end of each lesson, the platform gave every student a short check and sent her the results that evening. So she began a practice. Before each check, she wrote in a notebook the names of eight or ten students and, beside each, one word: understood or not yet. In the evening she compared her words with the platform’s results.

Chapter 6 told of a radiologist who reads ten images each morning before he sees the machine’s flags, so that his eye does not fade. Kavitha’s purpose was different. She was not trying to keep her eye. She was trying to find out where it could be trusted.

After six weeks she had not proved anything about her eye. About a hundred guesses, checked against short quizzes, cannot do that. But she had found where it seemed useful and where it kept misleading her. It misled her in a pattern. She read silence as confusion. A girl in the second row named Pooja almost never spoke and rarely looked up. Three times Kavitha wrote not yet beside her name. Three times the platform showed that Pooja had understood. Meanwhile a boy named Aman nodded hard, answered first and smiled at the right moments. Kavitha wrote understood. Twice the platform showed he had not. For twelve years she had been reading the performance of understanding. Nodding is a performance. Silence is not ignorance.

This call went the machine’s way, and the machine was right. Kavitha’s face-reading was trained by twelve years of faces. The platform’s checks taught her where that training was wrong, in a few weeks.

Confucius said what she had found. In Legge’s translation, he says to his student Zilu: “Yu, shall I teach you what knowledge is? When you know a thing, to hold that you know it; and when you do not know a thing, to allow that you do not know it; this is knowledge.”³ This is the outward test. Depth in pattern recognition is not a feeling that is always right. It is knowing which of your feelings to trust. Engineers calibrate an instrument by checking its readings against a known standard. For a person, calibration is the same check made on yourself: comparing how sure you felt with what actually happened. Kavitha’s eye seemed reliable for some students and not for others, and now she had a better idea which.

But the platform’s check is not the truth either. It is a measure, and Chapter 9 showed what happens to measures. A short quiz tests whether a student can pick the right answer from four. A student can do that without understanding. Aman, in October, passed the platform’s check on how water moves through a plant. At lunch, Kavitha pointed at the neem tree in the courtyard and asked him where the water was going up. He laughed and did not know where to look. Each check tested something different. The quiz tested whether he could recognise the answer. The question at the tree tested whether he could use it. Neither the teacher’s eye nor the platform’s check was the truth. They were two witnesses. Where they disagreed, Kavitha now looked hardest.

Chapter 8 said: believe the machine as far as you can check it. In Kavitha’s classroom the checking ran both ways. She checked the platform’s results against what she saw at lunch. The platform checked her eye against what the students could do. This is the machine as a teacher of the teacher. It gave her the one thing a classroom had never given her: quick feedback on her own perception. Ericsson’s deliberate practice needs exactly that, and the study in 10.1 suggested that most professions lack it. A machine that gives it can make a person deeper, if the person keeps score and keeps the score honest.

10.5 Narrative Intelligence: Making Meaning Out of Complexity

On a Thursday at the end of October 2025, the platform’s lesson ended with twenty minutes left before the bell. For weeks Kavitha had been filling such gaps with stories and questions she could not have defended to an inspector with a checklist.

That afternoon she told them about Jagadish Chandra Bose.

She had spent two evenings reading so that she could tell it accurately. People had known for centuries that the leaves of the touch-me-not, chhui-mui, fold when touched. In 1880 Charles Darwin and his son Francis published a book on how plants move.¹⁷ In 1873 the physiologist John Burdon-Sanderson had already found that the leaf of the Venus flytrap produces an electrical change when it is touched, much as an animal’s muscle does.¹⁸ So Bose did not discover that plants respond. He was a physicist in Calcutta who had built instruments to study radio waves. Around 1900 he turned very sensitive instruments of his own design on plants. He recorded their electrical and mechanical responses with a precision nobody had managed. And he made a bold claim: that living animal tissue, plants, and even metals respond to a blow, to heat or to poison in ways that are alike.

Some of what he claimed has held. Plants do send electrical signals. Some did not persuade. Few scientists accepted that metals respond as living tissue does. British physiologists, Burdon-Sanderson among them, objected to his methods and his claims. One long paper he sent to the Royal Society was never published.¹⁹ In May 1901 he gave a lecture at the Royal Institution in London. He ended it with a message that he said his ancestors had proclaimed on the banks of the Ganges thirty centuries earlier: “They who see but one, in all the changing manifoldness of this universe, unto them belongs Eternal Truth, unto none else, unto none else!”²⁰ I hear in it a verse of the Kaṭha Upaniṣad, which ends teṣāṃ śāntiḥ śāśvatī netareṣām, “theirs is eternal peace, not others’”.²¹ The verse promised peace. Bose, speaking to scientists, promised truth.

Kavitha did not tell this as a life story. She told it as a puzzle. Everyone around Bose had instruments. What made him point them at a leaf?

The students went still, even the restless ones. Then Pooja, from the second row, who almost never spoke, asked a question. “Miss, can what you believe about the world make you see things that are actually there? Things that other people miss because they believe something different?”

Kavitha opened her mouth and closed it again. She did not have an answer good enough for the question. She said she would think about it and answer the next day.

That evening she did what she now did with hard questions. She asked the machine, the assistant she used at home, not the platform. It gave her a good answer. Philosophers of science, it said, have long argued that what we see depends partly on what we already know. In 1958 the philosopher Norwood Russell Hanson imagined the astronomers Tycho Brahe and Johannes Kepler watching the sun rise. Tycho believed the sun circled the earth; Kepler believed the opposite. Did they see the same thing? “There is more to seeing than meets the eyeball,” Hanson wrote.²² Then the machine gave her something she had not known. In 1903 a respected French physicist, René Blondlot, announced a new kind of ray, which he called N-rays. Within months scientists, most of them French, reported seeing them, and hundreds of papers were published. In 1904 an American physicist, Robert Wood, visited Blondlot’s laboratory in Nancy. In the darkened room, during a demonstration, Wood quietly removed the prism the experiment depended on. Blondlot went on seeing the rays.²³

The machine gave Kavitha the second half of her answer. Before she used it, she did what her rule required. She found accounts of Wood’s visit, and his own short report in the journal Nature of September 1904. The story was true.

The next day she told the class both stories side by side. Bose believed that living and non-living things were one, and his belief made him look where others had not, and something real was there. Blondlot believed in his rays, and his belief made him see what was not there. So the answer to Pooja’s question was yes, and also yes in the other direction. Belief can open the eyes. It can also fill them. What tells the two apart is checking: an instrument that is allowed to say no, a second scientist who is allowed to remove the prism. She added one more thing, because the neat version is too neat. Bose’s belief helped him choose his question. His measurements of plants mostly held. Some of his conclusions, about metals, did not. A belief can help you choose where to look, and still mislead you about what you found. Then Kavitha gave the question back. “So how would you know which one is happening to you?”

Chapter 3 met narrative intelligence, the quality of finding the true shape in a set of facts and telling it so that others can see it. That chapter also warned that a good story is not the same as a true one, and that machines tell stories too. So what did Kavitha add here?

The print edition answered with a phrase: “conviction of origin”. A story from a person, it said, is generated “from the inside”, while a machine’s is “assembled from the outside”, and readers can often feel the difference. That answer is withdrawn. Chapter 3 reported a study in which readers could not tell machine poems from human ones better than chance. And the machine in this story supplied half of the best answer. The platform could have told the Bose story too, perhaps more smoothly than she did.

What Kavitha added was three things, and none of them was information. She chose this story, for these students, on that afternoon. She knew from her notebook that Pooja’s silence was attention, and she was waiting for it to speak. She checked what she told them, because it would leave the room with her name on it. And she made a promise: “I will answer tomorrow.” Chapter 4 drew on Hannah Arendt’s idea that promising is one of the ways a person appears among others. Kavitha was the one who could be asked again the next week, and the one who would have to correct the story if it turned out to be wrong. A platform could be built to return to a question too. Her promise matters because of where she stands in that room, whatever a machine can or cannot do.

There is a fourth thing, and it brings back the three steps. The platform gave the class śravaṇa, the hearing, very well. Kavitha’s story turned the hearing into a question that belonged to a student. Pooja’s question was the beginning of manana: knowledge being tested by someone who now wanted to know. A teacher with depth in her subject and in her students can start that step. The machine can supply the hearing to anyone. The question that starts a student thinking came, that afternoon, from a person who knew the student.

Name the scale, though. Most of what the platform teaches needs hearing and some testing, and no more. A fourteen-year-old does not need to live with the parts of a leaf until they change how she sees. The platform’s patience on the fourth repetition is not a failure of depth. It is hearing, done better than Kavitha could do it. The third step, in the borrowed sense of 10.1, is real here in a smaller place: a question carried home, an answer written down and then crossed out. That is nididhyāsana at the size this book can honestly claim. It does not rescue every lesson the machine teaches better.

Kavitha later began to call such stories her teaching stories.

10.6 From T-Shaped to Two Depths: Rebuilding the Missing Rung

For many years, people who hire have praised the T-shaped person. The idea is decades old, and Tim Brown of the design firm IDEO helped make it popular.²⁴ The vertical stroke of the T is depth in one field. The horizontal stroke is a working familiarity with many others, enough to talk to the people in them. The horizontal stroke is what a machine now supplies on request. Ask a model to explain the basics of marketing, data science or contract law, and it will, at the level you need. So the print edition proposed a second vertical stroke: real depth in two or three fields, held in one person.

The idea is worth keeping. The reason the print edition gave for it is not. It said: “The distance creates friction. The friction produces originality. The originality is the moat.” A moat is a defence against competitors, and this chapter has found no defence. A machine has more verticals than any person will ever have. If two depths are meant to keep you ahead of the machine, they will not.

What two depths give you is something else, and the second need not be a second profession. Zhang Rui’s was the workshops of his customers, and the question arrived in the gap between them and the physics. Camila’s was the slow watching of how people live, which told her what to feed her planning. Kavitha’s is these children. A second depth gives you a second place to stand: a way to see your first field from outside, and to know where your own knowledge stops and someone else’s begins. Uzzi’s study of papers, read as 10.2 read it, points the same way.

How is such depth built, when the machine now does the work beginners used to learn on? Chapter 4 called this the missing rung, and the three steps give it a shape.

Let the machine give the hearing. It is the best śravaṇa there has ever been, and refusing it out of pride is foolish.

Insist on the reflecting. Use the machine as a partner to argue with, not a source of finished answers. Ask it to attack what you believe. Write your own answer before you read its answer. Hints are not always better than answers; a beginner often needs a worked example first. The narrower rule is this: use the machine to make learning possible, not to hide whether it has happened. This is where the two studies in Chapter 4 point, and where a teacher or a senior colleague still matters most: in making the junior do the testing. Not every beginner’s task deserves saving. A junior engineer need not do every calculation by hand. She does need to learn to recognise an answer that cannot be right, and to say what it assumed.

The other side

The strongest objection to this section comes from researchers who study how people learn from instruction. Asking a beginner to try first, they would say, often wastes the attempt. Working memory is small, and a beginner’s search for a solution can fill it, leaving little room to learn the method.

They have evidence. In 2001 Slava Kalyuga and three colleagues taught apprentices in a mechanical trade, either with worked examples to study or with problems to solve. On later tests, the beginners did better with the examples, and rated the mental effort lower. With more experience in the trade, the examples became redundant, and solving problems worked better. In 2003 Kalyuga and colleagues reviewed such findings and named the pattern the “expertise reversal effect”. Methods that help inexperienced learners can lose their value, and even do harm, with experienced ones.

The objection partly wins. The advice to write first suits someone who knows enough to have a view of her own, as the Practice below assumes. For a true beginner, a worked case from the machine may be the better first step. On my reading, what survives is the purpose of the order: the help should shrink as the learner grows, until the first attempt is hers. Someone has to notice when the help has begun to get in the way, and that is the institution’s work as much as the learner’s.

Protect the living with. Time with the real material, without the machine’s answer in front of you, on real cases, with feedback on whether you were right. Kenji’s thousands of dies. Kavitha’s notebook of predictions. That time is the expensive part, and someone must pay for it. Chapter 9 called Kamala’s skill, gathered in one person’s hands with nobody learning it, a debt. Depth that an organisation uses but does not pay to build or pass on is the same kind of debt.

None of this can rest on the learner alone. A junior who is judged by how many tasks she finishes will hand the tasks to the machine, and by the measure she is given she will be right to. Chapter 9 showed what measures do. An organisation that wants depth would ask for a first attempt before the machine’s answer on some chosen tasks. It would ask juniors to explain the work, not only to deliver it. It would widen their responsibility in steps, from checking the machine’s work to taking the hard cases alone. And it would build feedback into the work, as Kavitha’s notebook built it into her week. A school could give every teacher on purpose what the platform gave her by accident. Where an institution does none of this, a person can still do a small part alone, as in the Practice below, on time the measure will not count.

Put the stories together and the word depth has carried several things. Knowing why: Kenji’s sense of what a ridge becomes, Zhang Rui’s of what his sensor measures. Knowing how far to trust yourself: Kavitha’s notebook. Helping another person begin to test: the Bose lesson. None of these is the right to decide what happens, and none is the power to act on what you see. An engineer may know why a sensor is wrong and still be told to ship it. A teacher may know a lesson failed and still be told to follow the platform’s schedule. Those powers come from the people and institutions around the knower. When they overrule what the knower saw, they answer for what follows. The knower’s part is to have said it, plainly and in time.

And be honest about what depth does not promise. In the professions, Macnamara and her colleagues found, practice explained less than 1 per cent of the differences between people. Their study did not say what explains the rest. Ability, luck, health, money and family are among the likely causes. A person can do everything in this chapter and still lose the job. This book does not promise employment. Depth is not a ticket.

The chapter’s title uses an economist’s word. Capital is what a capability adds to production. It does not say what the capability is worth to the person who has it.

Run it through two futures, as earlier chapters did. If machines grow quickly, depth will be worth less in many markets as a paid skill, because the machine will do more of what the deep person did. There will be more machine work to check. But mattering for checking is not the same as being paid for it. Chapter 5 named three things that stay scarce when thinking is cheap: choosing ends, checking and answering. Checking is the one that carries a date. And the seats where people are asked to check and to answer may be fewer than the people able to fill them. If machines grow slowly, depth stays a paid skill for longer. In both futures one thing survives that no market settles. Chapter 6 said that some work is valuable partly because it forms the person who does it. Depth is what that forming leaves behind. It does not employ anyone. It makes the knowledge yours.

Practice

Write first, then ask. For one month, whenever you are about to ask a machine a question in your own field, stop. Write your own answer first, in two lines. Beside it, write how sure you are (sure, fairly sure or guessing) and one thing your answer assumes. Then ask the machine. Where its answer can be checked against the world, check it, as Chapter 8 asked. Then mark your own answer: right, wrong, or still unknown. At the end of the month, look for patterns, not single misses; anyone can be right by luck or wrong for good reasons. Read two lists. The first is where you were sure and wrong. It is your own Analects 2.17: the places where you thought you knew. The second is where you were guessing and right: the places where you know more than you trust. Choose one item from the first list. Write down what would have told you sooner that you were wrong, and use it on a real case, without the machine. This builds depth only where answers can be checked. Some answers can be measured, some need a person who knows, and some nobody can settle yet; note which kind each one was. And none of this promises a wage. It tells you where you are deep. It does not tell anyone to pay you for it.

Ending: The Page with One Line Through It

Six weeks after the Bose lesson, Kavitha’s notebook was half full. Most pages were columns of names with understood and not yet beside them, and small ticks and crosses added in the evenings. She no longer wrote not yet for silence. Next to Pooja’s name, on three pages, there were three crosses.

One page was different. At the top she had copied Pooja’s question in Pooja’s words. Under it was her first answer, written that Thursday evening before she asked anyone: Yes. A single line ran through it. Below that were two names, Bose and Blondlot, and below them the question she had given back to the class. She had not found a better answer since. Neither had Pooja. Through the window, the neem tree stood in the courtyard where it had always stood, taking water up through wood that nobody in the room could see.

Words to keep

• ātman: the self.

• calibration: knowing how far to trust your own judgement, by checking it against what actually happened.

• deliberate practice: practice designed to improve one thing, usually by a teacher, with quick feedback on every attempt (Ericsson).

• depth: in this chapter, what knowledge does to the one who knows it: heard, tested and lived with until it changes how they see. It is not the same as holding more information than a machine, and it is a capacity, not a position or a right to decide.

• first principles: the most basic conditions a thing rests on (Aristotle, Physics I.1).

• manana: reflection; testing what you have heard against reasons and doubts until the doubts are answered.

• nididhyāsana: steady contemplation; living with knowledge until it becomes the way you see.

• śravaṇa: hearing; receiving a teaching from a teacher or a text.

• T-shaped: a person with depth in one field and working familiarity with many others.

• upādhi: in Nyāya logic, a hidden condition that an inference silently depends on, such as damp fuel for “where there is fire, there is smoke”.

Sources

1. Bṛhadāraṇyaka Upaniṣad 2.4.5, trans. Patrick Olivelle, The Early Upaniṣads: Annotated Text and Translation (New York: Oxford University Press, 1998), p. 69. The same teaching is repeated at 4.5.6.

2. Śaṅkara, Bṛhadāraṇyaka Upaniṣad Bhāṣya on 2.4.5: śrotavyaḥ pūrvam ācāryata āgamataś ca, paścān mantavyas tarkataḥ, tato nididhyāsitavyo niścayena dhyātavyaḥ … na anyathā śravaṇamātreṇa; trans. Swami Madhavananda, The Bṛhadāraṇyaka Upaniṣad with the Commentary of Śaṅkarācārya (Mayavati, Almora: Advaita Ashrama, 1934); also as quoted in “‘nididhyAsana’ as Shankara explains, 1/2,” Advaita Vision. The translator’s dash before “by hearing alone” is printed here as a comma.

3. Confucius, Analects 2.15 and 2.17, trans. James Legge, The Chinese Classics, vol. 1 (1861). In 2.17 Legge’s semicolon and dash before “this is knowledge” are printed here as a semicolon.

4. Joseph Lee, “Tes talks to… Anders Ericsson,” Tes, 18 November 2016.

5. Brooke N. Macnamara, David Z. Hambrick and Frederick L. Oswald, “Deliberate Practice and Performance in Music, Games, Sports, Education, and Professions: A Meta-Analysis,” Psychological Science 25, no. 8 (2014): 1608–1618; doi:10.1177/0956797614535810. Variance explained: 26 per cent (games), 21 (music), 18 (sports), 4 (education), less than 1 (professions).

6. K. Anders Ericsson, reply to Macnamara, Hambrick and Oswald (2014), posted on his Florida State University web page, September 2014; summarised in “Macnamara et al (2014) meta-analysis on deliberate practice not convincing,” Progress-Focused Approach, November 2014.

7. K. Anders Ericsson, “Summing Up Hours of Any Type of Practice Versus Identifying Optimal Practice Activities: Commentary on Macnamara, Moreau, & Hambrick (2016),” Perspectives on Psychological Science 11, no. 3 (2016): 351–354; doi:10.1177/1745691616635600. The 2016 study is by Brooke Macnamara, David Moreau and David Hambrick.

8. “Metrocable (Medellín),” Wikipedia: Line K began operating in 2004, linking the river valley at Acevedo with comunas 1 and 2.

9. Associated Press, “Colombian city gets giant outdoor escalator,” 27 December 2011 (via Salon): 384 metres in six sections; officials compared the climb to a 28-storey building and the time to six minutes from thirty-five.

10. Charles Darwin, The Autobiography of Charles Darwin (written 1876; ed. Nora Barlow, 1958), “My Several Publications”.

11. Alfred Russel Wallace, My Life: A Record of Events and Opinions (London: Chapman and Hall, 1905), vol. 1, chapter XXII, on the fever of 1858 and the recollection of Malthus. Wallace remembered the place as Ternate; his notebooks suggest nearby Halmahera (Gilolo).

12. Brian Uzzi, Satyam Mukherjee, Michael Stringer and Ben Jones, “Atypical Combinations and Scientific Impact,” Science 342, no. 6157 (25 October 2013): 468–472. Hit rates: 9.11 per 100 papers high in both conventionality and novelty; 5.33 for papers high in novelty only, against a background of about 5. The reading of the result as a hint about people is mine.

13. Udāna 6.4, Tittha Sutta, trans. Ṭhānissaro Bhikkhu (Access to Insight).

14. Leigh R. Crilley et al., “Evaluation of a low-cost optical particle counter (Alphasense OPC-N2) for ambient air monitoring,” Atmospheric Measurement Techniques 11 (2018): 709–720: a positive error above about 85 per cent relative humidity, 2.5 to 3.9 times the reference before correction; within 33 per cent after a correction based on κ-Köhler theory.

15. Aristotle, Physics I.1, 184a10–16, trans. R. P. Hardie and R. K. Gaye (Oxford, 1930; Internet Classics Archive).

16. Annambhaṭṭa, Tarkasaṃgraha §56 (seventeenth century): sādhyavyāpakatve sati sādhanāvyāpaka upādhiḥ … parvato dhūmavān vahnimattvād ity atrārdrendhanasaṃyoga upādhiḥ (“the hill has smoke, because it has fire”: here contact with wet fuel is the hidden condition); the red-hot iron ball (ayogolaka) is from Annambhaṭṭa’s own commentary, the Dīpikā. GRETIL text via Zenodo record 6466333.

17. Charles Darwin, assisted by Francis Darwin, The Power of Movement in Plants (London: John Murray, 1880).

18. Darwin to J. S. Burdon Sanderson, 13 September [1873], Darwin Correspondence Project, DCP-LETT-9055 (the editors quote his diary for 12 September 1873: “Work at Electrical phenomena of Dionaea”); and Darwin to A. G. Dew-Smith, 19 October [1873], DCP-LETT-9101, with the editors’ notes on the results: when the leaf was irritated, its current was disturbed, as in an animal’s muscle.

19. J. C. Bose, “On mechanical and electrical response in plants,” unpublished paper received 22 December 1903, read 25 February 1904, Royal Society Archives AP/76/1, with referees’ reports RR/16/10–20 (Burdon Sanderson’s, RR/16/10, calls the methods crude).

20. J. C. Bose, Friday Evening Discourse, Royal Institution, 10 May 1901 (date as given in his Response in the Living and Non-Living, London: Longmans, Green, 1902). The closing sentence as printed in Biswanath Banerjee, “The Scientist and the Poet: Acharya Jagadish Chandra Bose and Rabindranath Tagore,” Rupkatha Journal 2, no. 4, which calls the lecture a discourse to the Royal Society; other printed accounts give “they who behold the One”. The original dash before “unto none else” is printed here as a comma.

21. Kaṭha Upaniṣad 2.2.13 (5.13 in single numbering). The reading of Bose’s sentence as an echo of this verse is mine.

22. Norwood Russell Hanson, Patterns of Discovery (Cambridge: Cambridge University Press, 1958), chapter 1.

23. “September 1904: Robert Wood debunks N-rays,” APS News, August 2007; R. W. Wood, “The n-Rays,” Nature 70, no. 1822 (29 September 1904): 530–531.

24. “T-shaped skills,” Wikipedia (earlier uses of the term; Tim Brown of IDEO’s endorsement of it for building teams). On worked examples (for “The other side”, 10.6): Slava Kalyuga, Paul Chandler, Juhani Tuovinen and John Sweller, “When problem solving is superior to studying worked examples,” Journal of Educational Psychology 93, no. 3 (2001): 579–588 (abstract: “inexperienced trainees benefited most from the worked examples condition”; “With more experience in the domain, worked examples became redundant and problem solving proved superior”); Slava Kalyuga, Paul Ayres, Paul Chandler and John Sweller, “The expertise reversal effect,” Educational Psychologist 38, no. 1 (2003): 23–31 (abstract: techniques “highly effective with inexperienced learners can lose their effectiveness and even have negative consequences when used with more experienced learners”). Checked 31 October 2026.

Named people in this chapter are composites, and their conversations and circumstances are constructed; the studies, laws and figures cited are real and sourced.

Revised from the print edition (2026)

• What the print edition claimed: that the depth of Kavitha, a toolmaker, a planner, an engineer and an investor is something “no platform, no tool, and no dataset could replicate”; that mastery takes “Ten Thousand Hours That AI Cannot Shortcut”, the first section’s title; that the toolmaker’s thumb was “more informative than the instruments beside him”, and that “AI has no equivalent to this internal negotiation, because AI does not experience doubt about its own outputs”; that the planner’s walkway and escalator system was her own design, and “the only person in her firm who could have designed” it; that “Originality lives in the space between domains”, which AI “was not trained on”; that the engineer’s breakthrough was to use an LED as “a coherent light source”; that AI “does not yet do” the frustration “that generates the question”; that an investor’s losses “deposited something in her nervous system”, lessons “no training dataset contains”; that Bose “demonstrated that plants respond to stimuli” when “the dominant scientific view held plants to be passive”; that a person’s story carries a “conviction of origin” a machine’s lacks, and “that ‘not always’ is itself an argument for consciousness”; that “The originality is the moat”, in a section titled “From T-Shaped to Polymathic”; and that “AI has shallow knowledge of everything”, against “a single conscious mind”. It also gave the planner fifteen and then eighteen years, named the toolmaker Hayashi Kenji and the engineer Zhang Wei, set Kavitha awake under a ceiling fan, and ended with “The next chapter asks”.

• What I now hold: when a machine can give anyone the knowledge, depth is not holding more of it. It is what knowledge does to the knower: heard (śravaṇa), tested (manana) and lived with (nididhyāsana) until it changes how they see; not by hearing alone. The machine is the best hearing there has been, can help or harm the testing, and cannot do the living with for you. That last claim is about meaning, not a dated ability. Part of what looks like depth is only a missing record, and a machine can learn it once the record is kept. Depth matters because it lets a person check the machine, ask the question it was not asked, and answer for what they sign. Combinations across fields are found by machines too; originality comes mostly from deep grounding with one unusual neighbour. First principles, in practice, often means finding the hidden condition (upādhi) an inference depends on. A machine can teach a person’s intuition by giving it feedback, and depth in pattern recognition is knowing which of your feelings to trust (Analects 2.17). Depth is not a moat and promises no wage. It can fill a position; it does not grant one. The right to act on what a person sees, the time to look and protection for a good refusal are conditions an institution gives or withholds, and building depth when the machine does the beginner’s work is the institution’s task as much as the learner’s (a first attempt before the machine’s answer, explanation as well as delivery, responsibility widened in steps, feedback built into the work). A kept record can be a new rung. The three steps are borrowed for their order, not their object, and the third step is claimed only at a small scale: a question lived with, not every lesson redeemed. The people a plan is about may choose among its options, while those who drew and built it still answer for its safety; those who overrule what a knower saw answer for what follows.

• What changed my mind: the Prologue’s rule against defining people by what machines cannot yet do; Chapter 2’s finding that tacit knowledge is no longer safe; scanners that measure whole surfaces; the published research on humidity and low-cost sensors; Ericsson’s own rejection of the ten-thousand-hour rule and the 2014 meta-analysis; Uzzi’s study of 17.9 million papers; Wallace reaching Darwin’s theory from the same book; the history of plant electrophysiology before Bose; Medellín’s escalators of 2011.

• What survives: Kavitha watching the platform from the back of her classroom; the toolmaker’s thumb, now as judgement of consequence rather than detection; his early false alarm; the broad middle of a profession, where the machine arrives first; Camila’s three weeks on the paths and the cost to her career’s legibility; the engineer’s return to what the instrument measures; the term “narrative intelligence”; the Bose lesson and the girl’s question, “Miss, can what you believe about the world make you see things that are actually there?”; the notebook; the second vertical stroke, without the moat.

• Section titles in the print edition: 10.1 was “Ten Thousand Hours That AI Cannot Shortcut”, and 10.6 was “From T-Shaped to Polymathic”.

Next in the series: Chapter 11: Judgement: The Skill That Compounds.