AI-Proof: What Machines Can Never Be — Volume I: Awakening, Part I: The Treasure You Already Carry, Chapter 3: The Qualities of the Conscious Self

Part I looks at the person. Chapter 1 drew the distinctions, and Chapter 2 heard eight traditions. This third chapter asks what makes a quality yours when a machine can show it too.

← Previous: Chapter 2: The Ancient Knowing · Series hub · Next: Chapter 4: Why This Matters Now →

A machine can now show judgement, taste, patience, even empathy. So what makes a quality yours? Aristotle and Patañjali give one old answer: a path is made by walking it. Answering for where it leads is one more step.

In Brief: For readers who wonder whether their judgement, taste and patience still count when a machine can show them too. Chapter 3 names twelve qualities that matter more as intelligence gets cheaper, and asks what makes any of them a person’s own. A machine is given its abilities. A quality, in the old sense of Aristotle and Patañjali, is worn into a person by their own repeated choices; answering for what it leads them to do is a further choice. The chapter shows how leaning on machines can let these qualities grow over, and meets the strongest objections, from Gilbert Ryle and from the Gītā itself.

A machine is given its abilities. Your qualities are worn into you by your own choices, and answering for them is one more choice.

Opening: The Cold Room at Seven

The heart-team meeting starts at seven in the morning, before the first operation of the day. The room is small and too cold. There is a long table, a screen on the wall, and six people with coffee. Renata Oliveira has been a cardiac surgeon in São Paulo for twenty-three years. She sits near the end of the table and says nothing for the first ten minutes.

On the screen are the arteries of a sixty-one-year-old bus mechanic. Three of the vessels that feed his heart are narrowed. In two of them the narrowing is long, and the walls are heavy with calcium, hard as old pipe.

A heart team exists for decisions like this one. There are two main ways to treat such a heart. A cardiologist can open the narrowed arteries from inside, with small metal tubes called stents, pushed in through a blood vessel in the wrist. Or a surgeon can open the chest and build new routes around the blockages, with pieces of vein or artery taken from elsewhere in the body. This is a bypass. A stent is quicker and gentler. The patient goes home in a day or two. A bypass is a major operation, with weeks of recovery, but in some hearts it lasts longer. Medical guidelines ask that a team, not one doctor, weigh the two, and that the patient make the choice with them.¹

The hospital now has a decision-support system in the room. It has read his scans, his blood results, his medicines and the outcomes of hundreds of thousands of patients. It has ranked the options. Stents come first, with a high confidence score. Below the ranking it has flagged two things as minor: the pattern of calcium, and a gap in his history.

Renata has seen this pattern before. Not often. Three times in twenty-three years: long, chalky narrowings in more than one vessel, of this same shape. In all three, the stents narrowed again within a year, and the men came back to her for the operation they had tried to avoid. The system’s data treat such cases as rare, and they are rare. Three patients do not make a statistic. They make a memory.

She asks to see him before the meeting ends. In the ward he tells her he cannot afford to be off work for long. His son is studying engineering. He wants the stents.

Back in the meeting she argues for the bypass. The cardiologist argues for the stents, and he has the ranking on his side. Renata does not say the ranking is wrong. “It may be the better bet,” she says. “I can’t show you that it isn’t. I can only tell you what I have seen.” The team does not agree. In the end that is not its job. Its job is to put the choice to the patient honestly.

So that afternoon the two of them go to him together. The cardiologist explains the stents and the ranking. Renata explains the bypass and the three men. She tells him that three cases are not proof, that the computer has looked at far more hearts than she ever will, and that she might be wrong. His daughter, on the phone, asks why anyone would choose the bigger operation when the hospital’s own computer says otherwise. It is a fair question. Nobody in the room can answer it with certainty.

He takes a day to think. Then he chooses the bypass. He says later that the three men were not what decided him. What decided him was that she had told him she might be wrong.

Renata operates two days later.

Six months later he is back at work. He went back after eight weeks, not the two he wanted, and the family borrowed money to cover the gap. His recovery proves nothing. Most patients like him do well after either treatment. Renata will never know whether the stents would have done as well. Nobody will. A decision that is taken removes the world in which the other decision was taken. That is a fact about every choice, and it matters for this chapter.

Chapter 1 told the other side of Renata. Two years before this meeting she overrode a system’s recommendation on a feeling, and the feeling was wrong. Her patient had a week of needless pain. She carries both cases. They weigh against each other every time she sits at that table.

So this chapter asks one question. If a machine can show these qualities in what it does, what makes a quality yours?

From one light to many paths

The print edition of this book used an image for this chapter. It said that consciousness is like light passing through a prism: one source, many colours. The colours were twelve qualities: judgement, taste, trust, presence, empathy, patience, moral imagination, intuition, discernment, creative courage, narrative intelligence and distribution. The twelve are not a full list of what is good in a person. Honesty and fairness are not on it, and they matter as much. The twelve are the ones that seem to matter more as intelligence gets cheaper. A machine can now be made to show most of them in what it does.

I keep the twelve. I give up the prism. A prism makes colours without effort. Shine light into it and the colours come out, every time, from anyone’s light. If the qualities came out of consciousness that way, every conscious person would have all of them, equally. They do not. Some people have good judgement and no patience. Some have deep empathy and no discernment. And a person can have a quality at forty that they did not have at twenty. Something else is happening.

Here is a better image. Think of a field of grass between a village and a well. Nobody plans a path across it. But people walk to the well every day, and they tend to walk the same way. After a season a path appears. The grass is worn down. The earth is pressed hard. Now the path is easy to follow, so more people follow it, and it grows clearer still. If people stop walking it, the grass comes back in a year or two, and the path is gone.

The Chinese philosopher Zhuangzi, writing more than two thousand years ago, put it in one line: “A road is made by people walking on it.”² He meant something about how names come to fit things, but the line works for this chapter. A quality is a path worn in a person by their own walking. It is made by repeated choices. It makes the next choice easier. And it disappears if you stop.

Two old accounts of character reach a similar point from very different worlds.

In Greece, Aristotle asked how a person becomes good. His answer is plain. We are not born good, and we are not born unable to be good. “Neither by nature, then, nor contrary to nature do the virtues arise in us; rather we are adapted by nature to receive them, and are made perfect by habit.” And then: “men become builders by building and lyreplayers by playing the lyre; so too we become just by doing just acts, temperate by doing temperate acts, brave by doing brave acts.”³ Aristotle’s word for a quality of this kind is hexis, often translated “state of character”. It means something you have, in the way you have a skill: something your past acts have given you, which now shapes your future acts.

His definition adds two points that matter for machines. A virtue is “a state of character concerned with choice, lying in a mean”. And the mean is “determined by a rational principle, and by that principle by which the man of practical wisdom would determine it.”⁴ So for Aristotle a quality is not only a habit. It is a habit of choosing, and the right amount of it is fixed by a person’s practical wisdom, not by a rule. Courage is the mean between cowardice and rashness. Where exactly the mean lies depends on the person and the situation. That is why the quality cannot be written down as a procedure.

In India, the Yoga Sūtra of Patañjali says something close about practice, which it calls abhyāsa. Practice is one of the two means by which the restless movements of the mind are brought to rest. But practice only holds under conditions. The sūtra says: sa tu dīrghakāla-nairantarya-satkāra-āsevito dṛḍhabhūmiḥ. In James Haughton Woods’ translation: “But this [practice] becomes confirmed when it has been cultivated for a long time and uninterruptedly and with earnest attention.”⁵ Three conditions: a long time, no breaks, and care. Take away any one of them and the practice does not become firm ground.

The commentary on the Yoga Sūtra adds a mechanism. Every movement of the mind leaves an impression, a saṃskāra, and impressions in turn produce new movements. “In this wise,” Woods translates, “the wheel of fluctuations and subliminal-impressions ceaselessly rolls on.”⁶ This is the footpath in the language of the mind. Each act presses the grass a little. The pressed grass guides the next act.

The two are not saying the same thing, and the difference matters. Aristotle wants a good life lived among other citizens. His virtues are made for the city. Patañjali wants the restless mind to fall still, so that the self can stand free of it. His practice is made for release. One builds a character for living with others; the other trains the mind to let go. But on one point they meet, and it is the point this chapter needs: what you do again and again, you become, and it takes time. Using Patañjali’s conditions for a surgeon’s judgement or a planter’s eye is my application, not his subject.

Indian thought also made lists of qualities, as this chapter does. The sixteenth chapter of the Bhagavad Gītā opens with more than twenty, beginning with “Freedom from fear, purity of heart”, and including “gentleness, modesty”, “forgiveness, courage”.⁷ It calls them the endowment of a person “born to godlike endowments”. Nobody reading that list thinks it is a checklist. It is a portrait of a character, built over a life.

So here is the definition this chapter uses. A quality is a settled way of choosing, worn into you by your own repeated choices, which shapes the choices you make next. Three things follow.

First, a quality is yours, though you did not make it alone. Teachers, patients, colleagues, family and luck put the field in front of you and showed you where others had walked. Chapter 2 learned from the Ubuntu tradition that a person is formed among other people, and that is true of every quality here. But the walking was yours. That is why you can be praised for it, and why it is fair to hold you to it. Aristotle notes that praise and blame belong to what is voluntary.⁸ The field was not given equally. A person who was never given a field is not to blame for a path they could not walk. And a quality is something to praise. It is not the measure of what a person is worth.

Second, a quality can be lost. It needs walking. This is the danger section 3.8 describes, and it is the most practical point in the chapter.

Third, having a quality and answering for what it leads you to do are two different facts. Priya, in Chapter 1, put her name on a release without the path that would have let her check it. That is answering without the quality. A surgeon can walk the path for twenty years and then refuse to come to the meeting where she would have to explain herself. That is the quality without the answering. A quality is the path. Answering is what you do with it in a particular case: you explain the decision, undo it where it can be undone, and put right what can be put right. It is one more choice, and it can be refused. You can even answer for a path you let grow over, which is what section 3.8 will ask of you.

Now the machine. A machine trained on millions of human decisions can be built to act as if it has many of these qualities. It can weigh evidence, rank options, write warmly, wait, and find patterns no person has seen. These are real abilities, and in some of them the machine is better than any of us, as of 2026. But the machine did not walk to them. Its makers built them into it, by training it, and they can change them tomorrow. When it acts, the question “whose quality was that?” has an honest answer. It was the makers’, and the people who chose to use it. This is the second of the three “nevers” from the Prologue: a machine’s makers and users will never be free of what it does. A machine can give one part of an answer. It can explain what it did, often well. Putting right what went wrong falls on the people who built it and chose to use it. They are the ones a patient or a court can hold.

There is an obvious reply. Machines learn too. A model is trained on examples, corrected by feedback, and changed by its mistakes. Is that not a path worn by walking?

Look at who walks. Today the makers choose the training: what the model learns from, what counts as success, when it is changed. When a model makes a mistake, people decide whether and how it will be corrected. The mistake becomes a line in the makers’ records. It does not become part of a life the model carries forward and must answer from. Renata’s three men are different. They are not data she stores. They changed who she is, and she brings them to every meeting whether she wants to or not. What is the difference between a system storing a mistake and a person being changed by one? That question runs under this whole chapter. A performance happens in a moment. A character has a history.

I do not say this line will hold forever. Picture a system that chooses some of its own goals, keeps its own record of its failures, cares what it will become, and changes itself over years in ways its makers did not decide. If such a system ever exists, the argument of this chapter will have to be made again. Section 3.9 comes back to that question and leaves it open. The claim here is narrower. The words “quality” and “character” have always meant a history of one’s own choosing. That is the test any machine would have to meet.

So the spine of the chapter is three words. Abilities can be built. Qualities are grown by the one who has them. Answering is a further act, case by case. Keep the three apart: ability, quality, answering. The rest of the chapter takes the twelve qualities one at a time, with the people who showed them to me. Of each one it asks two questions. What can the machine do here, as of 2026? And what remains yours?

3.1 Judgement: Deciding When the Data Runs Out

Go back to the heart-team meeting.

The system did something very useful. It weighed hundreds of thousands of cases. It ranked the options. It said how confident it was. Renata did something different. She weighed three cases, with her own mistake in the other direction in mind, and judged that this patient might be more like the three than like the hundreds of thousands. Then she said so, in her own name, and said she might be wrong.

Was that judgement or stubbornness? From outside, you cannot always tell. The difference lies in how the decision was made and what the person was ready to do afterwards.

Judgement is the quality of deciding well when the evidence does not settle the question. It is not the same as being right. Renata has been wrong. It is not the same as overriding the machine. Most days she agrees with it. Judgement is the practised ability to see what kind of case this is, to weigh what matters against what is merely measured, and to choose.

Aristotle had a word for this: phronēsis, practical wisdom. He defines it as “a true and reasoned state of capacity to act with regard to the things that are good or bad for man.”⁹ Notice the word state again. Practical wisdom is a hexis, a worn path. You get it the way you get the other qualities: by judging, being wrong, and remembering. Renata’s three patients and her one needless override are the path. A first-year surgeon reading the same scan has the same information and less judgement.

Here is the case for the machine, and it is strong. Since the 1950s the American psychologist Paul Meehl, and many who followed him, have compared expert judgement with simple statistical rules: in medicine, in psychology, in decisions about parole, and in many other fields. In most comparisons the rules did as well as the experts, or better.¹⁰ A confidence score is honest. It says: in cases like this, this is how often each choice turned out well. A human “feeling” may be memory, or it may be pride. Renata’s own history shows both. So why should a family trust three memories over a hundred thousand records?

Meehl allowed one kind of exception. He imagined a professor who usually goes to the cinema on a Friday night. From his habits, a formula puts the chance that he will go this Friday at ninety per cent. But you know something the formula does not: he has recently broken his leg. Here you should overrule the formula.¹¹ The “broken leg” is a fact about this case that the formula has not seen, or cannot weigh. The trouble, as Meehl and his colleagues later showed, is that experts see broken legs everywhere. Given the freedom to overrule, they find too many exceptions, and they do worse than the rule.¹²

So is Renata’s calcium a broken leg? It might be. The pattern is rare, the system has seen few cases like it, and her three men are exactly the kind of fact a formula can underweigh. It might also be three coincidences, kept in memory because they were vivid. She cannot tell which. Nobody can. That is why the honest thing she did in the meeting was not to argue that the ranking was wrong. It was to say that it might be right, and to let the patient choose with that uncertainty in front of him. Judgement, here, is not the override. It is knowing what your own memory can and cannot support.

The same week, a woman came before the same team with a heart much like the mechanic’s. Renata felt the same pull toward the bypass. Then she looked again. The narrowings were short, and the calcium lay differently. Her three men did not fit. The ranking said stents, and this time Renata said so too. Nobody remembers that meeting. It is most of what judgement is.

Now notice what kind of thing the score is. A score is uncertainty with the person removed. Judgement is uncertainty with someone attached. A family can want both: the score, and a named person who will still be there if the score turns out wrong. That combination is what this book is for. The score cannot sit with the daughter on the phone. It cannot say “I advised this, and here is why, and I will be here in six months.” Renata can. And if she is wrong, she must explain it, try to put it right, and carry it into the next meeting. That is what “answer for” means in this book.

There is one more thing the meeting showed. Renata may have known best. She still did not decide. The mechanic did. The Prologue’s distinction applies to doctors as much as to machines: knowing what is best is not the same as having the right to decide it. A heart team that overrules its patient with a surgeon’s memory is no better than one that overrules him with a score. Had he been unconscious, unable to choose, someone would have had to speak for him, as the Prologue said of everyone in the third chair who cannot ask. The team would have owed that person the same honesty.

The print edition said that deciding “requires someone who has something to lose.” I keep the sentence and make it exact. Machines in tests have acted to avoid being switched off or replaced, as the Prologue and Chapter 1 reported. So it is not quite true that a machine has nothing to lose. What it has to lose was put there by its makers, and it can be copied and restored. A machine can run a thousand versions of the future. The mechanic will live only one, and so will Renata. Their losses cannot be restored from a copy. Judgement belongs to people whose losses are real in that way.

3.2 Taste: The Courage to Prefer

Taste sounds like a luxury. Now that machines can produce a thousand adequate options in a minute, it is one of the hardest qualities to keep.

Taste is the quality of telling what is good from what is merely acceptable, for reasons that a count of other people’s preferences cannot settle, and acting on the difference. A machine can predict what people will like. Taste asks what is worth liking. A product tests well and feels dead in the hand. A building meets every specification and moves nobody who enters it. In each case the numbers say yes. A person with taste says no, and can give some reasons, though never all of them.

Is taste only a private feeling? The Scottish philosopher David Hume asked this in 1757, in an essay called “Of the Standard of Taste”. He knew the saying that there is no arguing about taste. He also knew that nobody really believes it. Everyone accepts that some judges of wine or poetry are better than others. What makes them better? His answer: “Strong sense, united to delicate sentiment, improved by practice, perfected by comparison, and cleared of all prejudice, can alone entitle critics to this valuable character.”¹³

Look at the middle of that sentence: “improved by practice, perfected by comparison”. For Hume, taste is a path too. You get it by looking at many things, good and bad, and comparing them.

Seven hundred years earlier, in Kashmir, the philosopher Abhinavagupta described the ideal reader of poetry in almost the same terms. He called such a person a sahṛdaya, literally “one with heart”. He wrote a commentary on Ānandavardhana’s Dhvanyāloka, a famous work on how poetry means. In it he says that the sahṛdayas are those in whom the mirror of the mind has become clear through long practice of poetry. They can enter fully into what a poem describes.¹⁴ Again: practice. A polished mirror is not born polished.

The two did not mean the same thing. Hume wanted a standard: the joint verdict of good judges, which could settle arguments about beauty. Abhinavagupta wanted something more inward: a reader who can enter the feeling of a poem until, for a while, it becomes their own. But on one point the Scottish lamp and the Kashmiri lamp agree, and it is the point this chapter needs. Neither kind of taste is born. Both are polished by practice.

What does the machine do here? In 2024 two researchers at the University of Pittsburgh gave more than sixteen hundred ordinary readers a set of poems. Some were by famous poets, from Chaucer and Shakespeare to Sylvia Plath. Others were written by an AI in the same poets’ styles. The readers could not tell which were which. They did slightly worse than chance. And they rated the AI poems higher for qualities such as rhythm and beauty. The researchers suggest why. The AI poems were simpler and easier to understand. And readers took the difficulty of the human poems for the confusion they expected from a machine.¹⁵

That study shows two things at once. A machine can satisfy the average reader better than Shakespeare does. And the average reader’s preference is not taste. A sahṛdaya, Abhinavagupta’s practised reader, is exactly the person who can stay with a difficult poem long enough to see why it is difficult. Machines can also predict, from data, what most people will like. That is a real ability, and it is often useful. But it is a prediction of preference, not a preference. When the machine “chooses” one design over another, it is following a target its makers set: what users will rate highly, or what the person asking seems to want. If it chose otherwise, nobody would say the machine had bad taste. They would say the system needed adjusting.

A person with taste is in a different position. Their preference is theirs. It was formed by years of looking. They can be wrong, and be shown wrong, and change. And they can be held to it. An editor who rejects a book that later becomes a classic carries that rejection.

There is an honest case against taste, and it should be heard. Taste can be snobbery. It can be the preference of one class or one culture dressed up as a standard. Hume knew this too, which is why his list ends with “cleared of all prejudice”. The person with taste must keep asking whether their “no” is a judgement about the work or a judgement about who made it. Taste without that question is just habit with good manners.

Taste does not make you right. It makes the preference yours, so that you can be right or wrong in your own name.

3.3 Trust: Built Slowly, Destroyed Instantly, Impossible to Automate

In the banking district of Nairobi, a relationship manager named Njeri Mwangi sits across a table from a client she has known for eight years. His business exports avocados and French beans. This year it is in trouble. Prices have dropped. A shipment was held at the port for two weeks. His numbers do not justify the short loan he needs to get through the season. The bank’s credit-scoring system has marked his application as high risk.

Njeri knows things the model does not hold. She has watched this man get through three earlier bad years. She has visited his packing sheds near Mombasa and seen how his workers talk to him. She knows his numbers are bad and his character has not changed.

She also knows that trust is not a prophecy.

Four years ago she lent to a woman in Kisumu who exported fish. The woman was rebuilding after three bad seasons, and her determination impressed Njeri. Njeri trusted her. The woman did not cheat. The market for her fish collapsed, and no determination could save the business. The loan was not repaid. Njeri had to explain the loss to her own managers. She still thinks about it.

She approves this loan. She does not yet know how it ends. Neither do I.

What is she doing? The philosopher Annette Baier gave the best short answer I know. In a 1986 essay called “Trust and Antitrust”, she wrote: “Trust, then, on this first approximation, is accepted vulnerability to another’s possible but not expected ill will (or lack of good will) toward one.”¹⁶ Read it slowly. To trust is to accept that the other person could harm you, while not expecting them to. Trust is a chosen exposure. It is a wager.

That is why the credit model does not trust anyone, however well it predicts. It estimates risk. It exposes nobody. If the loan fails, the model will be retrained on the result. Njeri will be asked why. The money at risk is the bank’s, not hers. What she exposes is her name. Her managers will ask her why, and the answer will follow her, as Kisumu has.

And it is why trust is something a person grows. Njeri’s trust is not a feeling she has about the client. It is a practised way of reading people, worn into her by eight years of meetings, three good wagers and one that failed. Each wager taught her something. The failed one taught her the most.

Confucius saw trust as what holds human arrangements together. In the Analects he asks how a cart could move without the piece that joins it to the animals that pull it: “I do not know how a man without truthfulness is to get on. How can a large carriage be made to go without the crossbar for yoking the oxen to, or a small carriage without the arrangement for yoking the horses?”¹⁷ The Chinese word is xìn, often translated “trustworthiness”. It is a carriage image, which this book cannot resist. Without it, the cart and the oxen are both there, and nothing moves.

Notice that Confucius and Baier are speaking of two different things. Confucius means being worthy of trust: being the crossbar that others can rely on. Baier means the act of giving trust: choosing to be exposed. Njeri needs both. She gives her trust to the client. And her bank, and her client, rely on her being worthy of theirs. The first is a wager, made case by case. The second is earned slowly, by being the kind of person whose wagers are honest, including the one in Kisumu that failed.

Trust of this kind also has a blind side. It can reach only people Njeri has had years to read. A newcomer with the same numbers has no such history, and the score at least treats him and the old client alike. So her trust is a reason to look harder at this client. It is no reason to turn away the stranger.

Can people trust a machine? They do, every day, and often they are right to. But look at what is happening when they do. They are trusting that the makers built it well, that the bank tested it, that someone will answer if it fails. Trust in a machine is trust in the people behind it, whether we notice them or not. When nobody stands behind it, what remains is not trust. It is reliance on something that cannot be asked why.

The print edition said that trust “cannot be automated because it requires two conscious beings who recognise each other as selves.” I would now put it more carefully. A machine can be reliable, and reliability matters. Trust in Baier’s sense needs someone who chose to be exposed and will answer for the choice. That someone can work with a machine. It cannot be one.

3.4 The Qualities That Cost the Giver

The print edition said that three qualities, presence, empathy and patience, “cannot be faked”. That was wrong. People fake all three, and machines now perform two of them well, empathy and patience. In a 2023 study, health questions that patients had posted online were answered separately by doctors and by a chatbot. Judges rated the chatbot’s answers as empathetic or very empathetic far more often than the doctors’.¹⁸ Chapter 2 discussed what that study does and does not show.

So the question is not whether these qualities can be imitated. It is what they are when they are real. My answer has two parts, and they should be kept apart.

The first part is a description. When these qualities are real, they are spent from a finite life. That is what makes them gifts from someone. But it is not a test. A cost does not prove that a person is present, and a machine built to attend to one person at a time would pay a cost of its own. The second part is the one the argument rests on. When these qualities fail, someone has to answer for the failure.

Presence. On the island of Flores, in eastern Indonesia, a midwife named Ibu Mariana kneels beside a woman in labour. There is no diagnostic system in this room. There is a lamp, a clean cloth, and Mariana’s hands, which have delivered more than a thousand babies in three decades. What she gives is undivided attention. She is not checking a phone. She is not thinking about the next house. She is here, for as long as it takes, and the night is long.

The French philosopher Simone Weil wrote in a letter in 1942: “Attention is the rarest and purest form of generosity.”¹⁹ Notice the word generosity. A gift is generous when it costs the giver something. Mariana’s attention is generous because her night is hers, and she cannot spend it twice. While she is here, she is not anywhere else. That is what makes it presence. A wage does not change this. A midwife may be paid for the night, and it is still a night of her life, spent once.

A machine’s attention is spent differently. It can attend to a million people at once, and its attention to you takes nothing from its attention to anyone else. This is a real strength. A chatbot that answers a frightened question at three in the morning does real good. But what it gives is not taken from anyone’s life. Presence, in Weil’s sense, is the offering of a finite existence, in a world the two people share: the same night, the same room, the same risk. The chatbot shares none of these. It may still help, and help is not nothing. It is not presence.

Empathy. In a resettlement office in Amman, a caseworker named Hana Bishara listens to a Syrian civil engineer. He describes qualifications earned at a university that no longer exists, recorded in documents he lost when he crossed the border. He speaks with careful dignity. Hana notices what he does not say: the stillness when he mentions his family, the change in how he sits when she asks about plans. She builds, inside herself, a rough picture of what it is like to be this man in this room. She speaks more slowly. She offers tea. It is not in the procedure. She can see that a small, ordinary kindness matters more right now than a fast interview.

Adam Smith, the Scottish philosopher, described this act in 1759. “As we have no immediate experience of what other men feel, we can form no idea of the manner in which they are affected, but by conceiving what we ourselves should feel in the like situation.”²⁰ Empathy is imagination, used on another person. But imagining is only the first step. The second is what the imagining does to you: the other person’s state changes what you believe you owe them. Hana did not only picture the engineer’s loss. Because of it, she treated him differently.

There is a strong case against empathy too. The psychologist Paul Bloom argues that “Empathy is a spotlight focusing on certain people in the here and now,” and that “Empathy is biased.”²¹ It makes us care about the person in front of us and forget the thousands we cannot see. Hana knows this. She has eighty cases. The man with the tea is one of them, and the ones who arrived on a bad day got less of her.

And here the machine has an honest advantage. Hana’s empathy runs out. Some mornings she arrives already empty, and her listening becomes routine, and the people across the table can feel it. A system does not get tired on the eightieth case. If what we need is a kind and accurate response to every case, the machine may serve the eightieth person better than an exhausted Hana.

So what is hers? Her empathy is limited because it is real: it draws on her own feeling, and feeling can be spent. When she notices that she is empty, she has to decide what to do: rest, ask for help, or carry on and do her duty without it. That decision is hers. And when she gets a case wrong, she cannot say “my empathy module was miscalibrated.” She has to say “I did not listen well that morning.” The cost and the answering go together.

Patience. In a tea garden in upper Assam, an estate manager named Bora has watched the second flush come in for twenty-seven years. The second flush is the crop of late May and June. The best Assam tea comes from it: strong, malty, with golden tips on the leaf. Bora can tell from the colour of the young shoots, the weight of the morning air and the timing of the first heavy rain whether this year’s second flush will be ordinary or fine.

Some years ago the estate installed soil and weather sensors and a model that tells him, almost to the day, when to pluck each section. At first he followed it everywhere, for two seasons. The tea was consistent. It was adequate. It was never fine. The sensors measure what can be measured: temperature, humidity, the water in the soil. They do not record what Bora has learned to see: how the shoots look after a dry April, or the particular softness of the second leaf when the bush has been stressed just enough. Since then he has divided the estate. In the low sections near the river the model’s timing is better than his, and he follows it there. On the higher ground he walks the bushes every morning and decides himself.

His patience is tested now in a different place. Last year the estate gave him an assistant, a young graduate named Rupjyoti, who trusts the model and is good with it. Bora makes him walk the higher sections every morning for a full season before he may decide anything there. Rupjyoti thinks this is a waste of time. The model is right most of the time, and he has a better offer from a company in Guwahati. He may leave before the season ends.

Bora is not sure he is right to insist. Perhaps the young man will never need the eye. Perhaps the model will improve every year, and walking the bushes will become a hobby for old men. But Bora knows how he learned what he knows. Patience, as the Yoga Sūtra says of practice, takes “a long time” and no breaks. Bora’s knowledge is twenty-seven layers deep, one for each season, and each season was different. He could not have learned it faster. He cannot give Rupjyoti those seasons. He can only refuse to let him skip the first one. Some understanding cannot be hurried, because it is made of time. Patience is the quality of giving it that time, which means accepting slowness, and sometimes paying for it. Here the price may be a young man who walks away.

These three qualities share one thing. Each is spent from a finite life: a night, a feeling, a season. Machines can supply the effects of all three, and sometimes better. What they do not supply is a life the gift was taken from, or a person who answers when the gift fails. The first is a description. The second is the argument.

3.5 The Qualities That See What Isn’t There

Three qualities look ahead of events. They notice what has not happened yet, or what has not yet been named.

Moral imagination. In a product development lab in Shenzhen, an engineer named Lin Wei reviews a camera for the home that recognises faces. It works. It meets its specifications. The automated test suite passes it. Lin Wei is not celebrating. She is picturing who will use it, and on whom.

She thinks of a woman whose former partner still has the password, and can see from the camera log when she comes and goes. This is not a new idea. Researchers had documented exactly this kind of abuse years before. A 2018 study of survivors of abuse in New York found that abusers mostly used simple, everyday technology: shared accounts, known passwords, devices in the home.²² The harm was known. What was missing was anyone in this room who connected that knowledge to this product.

The British philosopher Iris Murdoch wrote: “I can only choose within the world I can see, in the moral sense of ‘see’ which implies that clear vision is a result of moral imagination and moral effort.”²³ Moral imagination is the effort to see the world your choice will land in, including the people who will never be in the meeting.

Here is the machine’s case. Companies now use automated “red-teaming”: systems that attack a product in thousands of ways to find failures and misuses before customers do. They are thorough, and they find harms that humans miss. A well-built red-teaming system, given the right instructions, might well have flagged the camera log. Lin Wei’s imagination is not the only way to see this harm.

But someone has to decide what the red team looks for, and someone has to care when it finds something. Lin Wei raised her concern. It was dismissed. The product met its specifications. The lawyers had approved it. The market was waiting. So she wrote a memo that told short stories instead of citing rules: this is what could happen to this kind of person in this kind of home. Some colleagues began to see what she saw. The company added a feature that tells every user when a new device is watching the log. It did not do the other three things she asked for. She won one argument out of four. That is often what moral imagination wins, and it is still worth having. The three she lost are not hers to answer for. Her part was to see the harm and to put it in writing. Those who decided against her cannot now say that nobody told them.

Intuition. In Istanbul, a financial crimes detective named Elif Kaya reviews a fraud case. The accounts balance. Each transfer, alone, is ordinary. The compliance system has raised several alerts, and each one cleared when someone checked it. The case looks clean. Elif feels that something is wrong.

She cannot yet say what. Seventeen years of financial investigation have trained her to notice patterns before she can name them. The timing of the transfers. The route the money takes, through places that each make sense, but together form a shape she has seen before. Documents that are a little too tidy, as if they were made to survive exactly the checks a machine runs.

The Indian grammarian and philosopher Bhartṛhari, writing in the fifth century, gave this kind of knowing a name: pratibhā, a flash of understanding that arrives whole. He saw it everywhere in nature. “Who goads beasts and birds on to actions like eating, loving, hating, swimming etc.?” he asks.²⁴ A bird builds a nest without being taught. For Bhartṛhari, pratibhā can come from nature, but also from practice. Elif’s is the second kind.

Science has studied when such flashes can be trusted. In 2009 the psychologists Daniel Kahneman, who distrusted intuition, and Gary Klein, who studied experts who rely on it, wrote a paper together and found that they agreed more than they expected. “Two conditions must be satisfied for skilled intuition to develop: an environment of sufficiently high validity and adequate opportunity to practice the skill.”²⁵ Intuition is reliable where the world gives clear signals and quick feedback, and the expert has had years to learn them. Firefighters and chess players have it. People who forecast politics usually do not.

By that test, Elif’s world is a hard one. Fraud is designed by people who are trying not to be seen, and who study the checks. Feedback is slow: a case can take years to close, and many never do. Kahneman and Klein would not trust her hunches very far. Neither does she.

Here the machine is often ahead. Machines learn patterns in exactly the kind of environment Kahneman and Klein describe, from more examples than any human sees. In 2016 the Go-playing system AlphaGo played the champion Lee Sedol. In the second game it made a move that, by its own data, a human would play about one time in ten thousand.²⁶ Experts called it a mistake. It won the game. If intuition means recognising patterns below the level of words, machines have it now, and in games and scans they have more of it than we do.

So what is Elif’s? Not the flash. Her quality is what she has learned to do with it. Two years ago she followed a feeling for three months, found nothing, and a small company spent a year under a suspicion it did not deserve. She remembers the owner’s face. Since then she has kept one rule. A hunch is a reason to look, never a reason to accuse. She looks quietly, for a fixed time, and nobody is named until there is evidence that someone else could check.

This time she gives herself two weeks. In the second week she finds a payment that should not exist. She reopens the case. Whether it ends in a conviction is not known as I write.

The flash may be the same kind of thing in her and in a machine. The rule is the path her mistake wore in her, and it is hers.

Discernment. The third looking-ahead quality is the plainest. An editor uses it each morning, when a hundred stories arrive and only five can lead. A doctor uses it when a patient describes twelve symptoms and only one of them matters.

Discernment is the quality of seeing what is significant. In 1971 the economist and computer scientist Herbert Simon saw where this would lead. “A wealth of information creates a poverty of attention,” he wrote, “and a need to allocate that attention efficiently among the overabundance of information sources that might consume it.”²⁷ When information is cheap, the scarce thing is knowing what to look at.

Indian philosophy calls the power to separate the lasting from the passing viveka, discrimination in its old and good sense. It was always a practised quality, not a gift.

Machines rank and filter well. They decide much of what you see every day. But ranking is not the same as significance. A ranking answers the question “what will this person probably click?” Discernment answers “what matters here, for this purpose, now?” That second question has an owner. When an editor decides which story leads, or a doctor decides which symptom to chase, they can be asked why. When a feed decides, the answer is “the system ranked it”, and that is the end of the answer. As the Prologue put it, that is abdication.

3.6 The Qualities That Create

Two of the twelve qualities bring something new into the world.

Creative courage. In a studio in Mexico City, a filmmaker named Valentina Reyes is arguing with her producers. The audience-testing tool they paid for is clear. Test viewers prefer a straight timeline, a familiar score, and an ending where the hero succeeds. Valentina wants a broken timeline, a score built from recordings of rain and street markets, and an ending in which the hero finds something she was not looking for.

She has read the data. She knows what the test audience says it wants. She also has a specific reason to think the data is incomplete. The test measures what people say after seeing a film once, in a hired room, on a weekday evening. She is making a film to be seen twice. In twenty years of making films she has watched the scenes that tested worst become the ones people quoted back to her years later.

The American psychologist Rollo May described this in 1975: “Whereas moral courage is the righting of wrongs, creative courage, in contrast, is the discovering of new forms, new symbols, new patterns on which a new society can be built.”²⁸ Creative courage is the quality of acting beyond what the existing evidence can establish, when you have good reason to think the evidence is incomplete, and standing behind the result. It is not the refusal of evidence. Ignoring the data without such a reason is usually just a mistake.

She makes the film her way. It does not earn back its cost in its first month. It wins no prize. A year later people are still talking about it. Two years later students are studying it.

I have to be honest about this story, because it is the kind of story that flatters people like me. For every Valentina whose film is remembered, there are many filmmakers who ignored the data and were simply forgotten. We hear about the survivors. Creative courage does not promise success. What it promises is smaller and more certain: the work is yours. If it fails, it fails as your choice, and you can learn from it. If you follow the testing tool and the film fails, you learn only that the tool was wrong. And a film is made with other people’s money. Courage spent with it owes them the truth. That means telling them what the test said and why you are going against it, so that they can see the risk they carry.

As of 2026, machines generate new images, melodies and stories in seconds, and some of them are surprising. A 2024 study found that writers who used AI ideas wrote stories that judges rated as more creative, but the stories also became more alike.²⁹ The machine raised the floor and narrowed the range. That is what a tool trained on what most people have already liked will tend to do: it pulls new work toward the middle of what already exists. Creative courage, now, is partly the refusal of that pull: the willingness to stand outside the range of what is likely, and to answer for standing there.

Narrative intelligence. In a courtroom in Lagos, a lawyer named Chidinma Okafor makes her closing argument in a patent case. A large foreign company says her client, a small Lagos firm, copied its invention. The other side has filled the court with charts produced by an analysis system: each one impressive, all of them together impossible to follow.

Chidinma does not add another chart. She tells a story, with dates. Two years before the foreign company filed its patent, her client’s engineers showed the same idea in public, at a technology fair in Lagos, and a newspaper photographed it. A patent can only protect an idea that was new when it was filed. If the idea was already public, the patent may fall. She does not argue that her client was clever. She argues that the order of events matters, and she makes the judge see the order.

The judge asks for the newspaper.

The psychologist Jerome Bruner said there are two basic ways of thinking: the logical argument and the story. “A good story and a well-formed argument are different natural kinds,” he wrote. “Both can be used as means for convincing another.”³⁰ Narrative intelligence is the quality of finding the true shape in a set of facts, the order and connection that makes them mean something, and telling it so that others can see it.

What does it add that judgement does not? Judgement decides. Narrative intelligence shows. The other side had every fact, including the newspaper photograph. It was in their own exhibits, on page forty. What they did not have was the shape: the one fact that put all the others in order. Finding that shape is a quality, worn by years of building cases. And the same quality, used to give false facts a convincing shape, is called propaganda. A good story is not the same as a true one.

Machines tell stories too. As of 2026 they write fluent, well-built narratives, and some find the shape in a set of facts very well. So the quality itself may not stay ours alone. What a court case in New York shows is the other half of this chapter’s definition: not the quality, but the answering.

In 2023, in the case of Mata v. Avianca in New York, two lawyers filed a brief that cited court decisions which did not exist. A chatbot had invented them. The judge fined the lawyers and their firm $5,000. He wrote: “Technological advances are commonplace and there is nothing inherently improper about using a reliable artificial intelligence tool for assistance. But existing rules impose a gatekeeping role on attorneys to ensure the accuracy of their filings.”³¹ The court did not fine the chatbot. It could not. Nor were its makers before the court. Whatever they owe for a tool that invents cases, a filing answers first to the person who signs it. The lawyers had signed. A signature on a court filing is not only a mark of who wrote it. It is a promise to answer for it. Chapter 1 left Priya asking what her name on a release was for. This is one answer.

Chidinma’s story is hers in both senses. She found its shape, through years of practice. And she stands behind it: she checked every date, she will defend it if the other side attacks it, and she can be punished if she is wrong. A story in a court of law is a promise that the story is true, made by someone who can be held to it.

3.7 Distribution: The Quality That Sits Apart

The twelfth item on the list needs a different kind of honesty, because it is not a quality at all.

Distribution means putting your work, your ideas and your judgement where the people who need them can find them. It is a behaviour in the world, not a state of character. The first eleven are paths worn inside a person. Distribution is the road between that person and everyone else.

It is on the list for a practical reason. The print edition used a shorthand, which it called the AI-Proof Equation: judgement, taste, trust and distribution. The first three are qualities. The fourth was there because the book was written partly for people earning a living, and in a market, being found matters as much as being good. Judgement nobody knows about helps nobody but you. “Distribution” is a market word, and I keep it on purpose. A book for working people should not pretend that being found does not matter.

A musician in Buenos Aires spends three years building an audience: small venues, short essays about how she works, releases placed where her particular listeners gather. She is practising distribution. Every choice in it uses her judgement, her taste and the trust she has built. Distribution is where the inner qualities reach the outer world. It is a channel, and only a channel.

And because it is a channel, it can carry harm as easily as good. What is found can be found by anyone. A journalist who distributes a true story may expose a person who did nothing wrong. A teacher who posts her students’ best work may make a child searchable for the rest of her life. Machines now make distribution almost free and almost instant, which multiplies both kinds of result. So distribution, too, needs someone who answers for it: who decided this should reach these people, and who will deal with what follows.

Keeping distribution on the list, while saying plainly that it is different, keeps the book honest. Removing it would give you a rich inner life and no way to reach anyone. Calling it a quality of consciousness would be the prism again.

3.8 The Danger You Did Not Expect

You may have read this far expecting one warning: that machines will replace you if you do not grow these qualities. That warning is real. There is a deeper one.

Remember the footpath. It exists because people walk it. If they stop, the grass returns. Now imagine a moving walkway built across the same field, smooth and fast. Nobody walks the old path any more. In two years it is gone. Then one day the walkway breaks.

Machines do not only lack our qualities, or have their own. By doing the work that used to wear the path, they can let the path grow over. The danger is not that the machine takes the quality away. It is that we stop walking.

This was noticed long before the current machines. In 1983 the psychologist Lisanne Bainbridge wrote a short paper called “Ironies of Automation”, about factories and power stations. “Unfortunately, physical skills deteriorate when they are not used,” she wrote. “This means that a formerly experienced operator who has been monitoring an automated process may now be an inexperienced one.”³² And the irony: the human is kept for the rare moments when the automation fails, which are exactly the moments that need the most skill.

There is now direct evidence that the same thing happens with AI. In 2025 a study in four endoscopy centres in Poland looked at doctors who examine the bowel for small growths that can turn into cancer. Once the centres began using an AI system that points out such growths, the doctors also worked some days without it. On those days, after months of AI help, they found growths in about 22 out of every 100 patients. Before the AI arrived, they had found them in about 28.³³ The study was observational, and one study is not the last word. But the direction is exactly what Bainbridge predicted. The eye that is always helped gets weaker.

A 2025 survey of 319 office workers found something similar about thinking. The more confidence people had in the AI tool, the less critical thinking they reported putting into the work. The more confidence they had in themselves, the more they thought critically.³⁴

Put plainly: performance can rise while practice falls. With the tool, today’s work is better. Without it, tomorrow, the worker may be worse. And, as the survey suggests, confidence can rise at the same time.

Now take the twelve qualities in turn.

Judgement grows through hard decisions, made under uncertainty, some right and some wrong. When a machine handles the easy decisions and presents the hard ones already analysed and ranked, the person gets fewer hard decisions of their own. Aristotle’s sentence applies directly: we become judges by judging. You do not become a better judge by approving recommendations.

Taste grows through exposure to the whole range: good, bad, ordinary and extraordinary. When a machine selects what you see, choosing what it predicts you will like, the range narrows. The study of AI-assisted stories found the same narrowing in what people make.²⁹

Patience suffers first in a world that answers every question in seconds. Intuition needs years of slow, confused, immersive experience. When the machine offers the pattern at once, the temptation is to take it and skip the confusion. Each time you skip it, the path is not walked.

The Yoga Sūtra named the three conditions: a long time, no interruption, and care. A machine that does the work for us breaks all three. The time is shortened. The practice is interrupted. And care moves from the work to the tool.

It would be convenient to blame all this on laziness. Mostly it is something else. Organisations measure output, speed and cost, and the path shows up in none of them. To a hospital manager, Hana’s tiredness is a problem the machine solves. Bora’s season of walking the bushes with Rupjyoti is a season of lower output. A team that makes its own judgement before it checks the machine looks slower than a team that does not. So walking the path is not only a private discipline. Inside most institutions it looks like waste, and someone has to defend it. Later chapters, on organisations and roles, ask who that someone should be. The point here is only this: the walkway is measured, and the walking is not.

None of this means giving up the machine. Bora still reads his sensors. Renata still reads the ranking. It means keeping some of the walking. There are two kinds of reliance. You can use a tool for the work, as you use a calculator for a long sum. Or you can let it take over the practice that keeps you able to judge its work. The first is usually wise. The second should sometimes be refused. The reason is not that the machine is worse. It is that doing the work yourself is how you stay able to check it. That refusal is the power Chapter 1 gave the lord of the chariot, the power to say no, used on yourself. Not every lost skill matters. Few people still do long sums by hand, and nobody is harmed, because nobody is kept at work to catch a calculator’s mistakes. The skills that matter are the ones you are kept for: the hard case, the check, and the day the walkway breaks. Those are the paths to keep walking.

In practice, that means sometimes choosing the harder road on purpose: making your own judgement first and comparing it with the machine’s afterwards, reading the unfiltered pile now and then, sitting with a problem before you ask. The same tool, used in two ways, leaves two different people after ten years.

What would prove this wrong?

My claim in this section is that leaning on a machine can wear away the very skill a person is kept to use when it fails. Emerging. The direct evidence is thin. The Polish study looked back at the records of 1,443 patients examined without the machine, 795 before it arrived and 648 after. A study that looks back cannot rule out other changes over the same months, and the authors say only that steady exposure “might reduce” detection.³³ The survey of office workers measured what people said about their own effort, not what they did.³⁴

Here is what would overturn the claim. Suppose people are assigned at random to work with a machine or without it for some months, and then everyone is tested without it. If the ones who had the machine do as well on their own as the others, or better, then I am wrong about that kind of work. Bainbridge’s irony would then be a risk to manage, not a rule.

The test has to be of unaided work, after long help. A score taken with the machine switched on settles nothing here, because it measures the pair, not the person.

3.9 The Map

Twelve qualities, but not a ladder. Judgement is not higher than patience. Empathy is not more important than discernment. And they fight.

Judgement without empathy becomes cold calculation. Empathy without discernment becomes being overwhelmed. Creative courage without patience becomes recklessness. Patience without distribution becomes obscurity. Renata’s judgement and the mechanic’s wish to be back at work in two weeks pulled in opposite directions. Njeri’s trust and her duty to her bank pulled apart.

Each quality also casts a shadow. Judgement can become pride, as it did for Renata two years ago. Trust can lose the bank’s money, as Njeri’s did in Kisumu. Empathy can become favouritism. Patience can become delay, and may cost Bora his assistant. Taste can become snobbery. A hunch can cost an innocent company a year. Narrative can become propaganda. Distribution can make a child searchable. A list of qualities is not a list of guarantees. Having them does not save anyone from tragedy. It makes a person able to choose inside it.

The philosopher Isaiah Berlin argued that this conflict cannot be removed. “If, as I believe, the ends of men are many, and not all of them are in principle compatible with each other, then the possibility of conflict, and of tragedy, can never wholly be eliminated from human life, either personal or social. The necessity of choosing between absolute claims is then an inescapable characteristic of the human condition.”³⁵

If the qualities conflict, and there is no ranking, then something must decide between them each time. What is it?

No thirteenth quality settles it. Aristotle’s answer was practical wisdom: the mean “relative to us”, fixed by the person of practical wisdom in the situation. The Prologue’s answer was the old Indian phrase deśa, kāla, pātra: place, time and person. They come from different worlds, and they agree on one point. The balance is struck by a particular person, in a particular case, and it cannot be struck in advance by a rule. And whoever strikes it answers for it. That is where the map ends: with a person who has to choose.

Now the case against this whole chapter, in full. There are two versions, one from Oxford and one from India, and both are serious.

The first comes from the philosopher Gilbert Ryle. In 1949, in The Concept of Mind, he attacked the idea that inside each person there is a hidden inner self that does the real thinking. He called it, “with deliberate abusiveness, ‘the dogma of the Ghost in the Machine’.”³⁶ For Ryle, to have a quality is simply to be disposed to act in certain ways. “To possess a dispositional property is not to be in a particular state, or to undergo a particular change; it is to be bound or liable to be in a particular state, or to undergo a particular change, when a particular condition is realised.”³⁷ Being brittle means breaking when struck. Being patient means waiting when waiting is called for. Nothing more is needed, and nothing hidden.

Apply that to machines. If a quality is a disposition to behave, and a machine reliably behaves that way, then the machine has the quality. A system that reliably answers frightened people kindly is kind. A system that reliably flags misuse has moral imagination. To deny it, Ryle would say, is to look for a ghost.

The second objection comes from inside the Indian tradition, and it goes deeper. The Sanskrit word for “quality” is guṇa. In the Sāṃkhya philosophy, which the Bhagavad Gītā draws on, the guṇas belong to prakṛti, nature, the whole material world, including the body, the senses, the mind and even the intellect. The true self, puruṣa, is a witness. It does not act. The Gītā says it directly: “He whose mind is deluded by egoism thinks himself the doer of the actions, which, in every way, are done by the qualities of nature.”³⁸ Prakṛteḥ kriyamāṇāni guṇaiḥ karmāṇi sarvaśaḥ: all actions are done by the qualities of nature.

Put a machine into that picture. A machine is prakṛti. It is matter arranged to process. If qualities belong to nature and not to the self, there is no reason a machine cannot have them. Chapter 1 already accepted that the intellect, buddhi, is the charioteer, and Sāṃkhya counts buddhi as part of nature. And there is worse for this book. If the self is not the doer, then my whole language of “answering for” a decision looks like the very egoism the Gītā calls delusion: ahaṃkāra, the “I-maker”, saying “I did this.”

Both objections deserve more than a quick reply. Here is mine, in three parts.

First, against Ryle. I accept most of what he says. This chapter has not defended a ghost. It has defined a quality as a settled way of choosing, made by one’s own repeated choices. That is close to Ryle’s disposition. But notice what the definition includes and his leaves out: how the disposition was made, and by whom. Two people may be equally patient today. One became patient by twenty-seven seasons of waiting. The other took a pill this morning. We admire the first and not the second, though their behaviour is the same. This is no judgement on medicine. A person whom treatment helps to stay calm still chooses, day after day, what to do with the calm, and that walking is theirs. The point is only that praise follows the walking. Aristotle drew the line in exactly that place: praise and blame belong to what is voluntary, and a virtue is a state that our own acts produced. A machine’s dispositions were produced by its makers’ choices. So the honest Rylean conclusion is not “the machine is kind.” It is “the makers built something that behaves kindly, and the makers answer for it.” Ryle removes the ghost. He does not remove the authors.

Second, against the Sāṃkhya objection about machines. I accept it too. On this view the machine may well have guṇas, as a river has force and a fire has heat. The book never needed to deny that. It needs only the distinction this chapter has drawn: between abilities, which nature and engineering can produce, and the question of who answers. Sāṃkhya does not say that nobody answers. It says the witness is not the doer.

Third, the hardest part: the Gītā’s attack on “I am the doer”. Here the Gītā helps more than it harms. In its last chapter it lists five causes of every action: “The substratum, the agent likewise, the various sorts of organs, and the various and distinct movements, and with these the deities, too, as the fifth.”³⁹ The body, the agent, the instruments, the efforts, and powers beyond our control. It then says that the person who sees the self alone as the agent “sees not (rightly).”⁴⁰

That is exactly the point the Prologue made with the political philosopher Dennis Thompson’s “problem of many hands”. Every decision in this book has many causes: the machine, the makers, the data, the institution, the patient, luck. To own an act is not to claim that you caused everything that produced it. It is to answer for the part that was yours. Renata did not cause the mechanic’s disease, or build the ranking system, or set the hospital’s rules, or make his choice. She answers for her part, which was her advice, and then the operation. The Gītā loosens the claim “I alone did this.” It does not loosen the duty to act and to own what one did. Arjuna is not told to walk away. He is told to act without the delusion that he controls everything.

There is a danger here, and I should name it plainly. The next verse of the Gītā says that the one who has no feeling of egoism, “even though he kills (all) these people, kills not, is not fettered.”⁴¹ Read carelessly, that is the most dangerous sentence a manager could find. “It was not I. It was the system. It was the qualities of nature.” Chapter 1 borrowed the researcher Madeleine Clare Elish’s name for a person who carries blame without control: a moral crumple zone. This is the reverse: a person who has control and takes no blame. The verse describes someone who has truly let go of the self, not someone looking for an excuse. But it can be misused, and in an age of machines it will be. The book’s Epilogue returns to the Gītā and to the choice it leaves Arjuna. Until then, the rule is this: the doctrine that loosens “I alone did this” must never become “nobody did this.”

So the case against this chapter changes it, and is not defeated. Machines may have abilities that look like every quality on the list, and may in time have more of them. Whether any machine experiences anything when it acts remains, as Chapter 1 said, an open question. Nothing here depends on it. What the chapter claims is narrower. A quality in the full sense is grown by the one who has it, through their own choices. Answering for what it leads them to do is a further choice, and it is the one the world can hold them to. A machine’s abilities were grown by others, and those others answer for them.

That leaves things I cannot settle, and I will not pretend to. Suppose a machine learned for years from its own choices. Suppose it chose some of its own goals, kept its own record of its failures, and was changed by them in ways its makers did not decide. Would it be growing qualities in Aristotle’s sense? Would its makers still answer for what it became, or would it begin to answer for itself? And when machines help to train the machines that follow them, where does “made by its makers” end? I do not know. Later chapters return to these questions. This chapter has only tried to say what the words would have to mean before the answer could be yes.

---

A week after the operation, Renata visits the mechanic in the ward. He is sitting up, grey and sore, angry about the eight weeks. He asks her straight out whether she has ever been wrong about one of these.

She tells him about the patient two years ago, the one for whom she overruled a recommendation that was right. He did not expect that. He is quiet for a while.

“So you might be wrong about me too,” he says.

“I might,” she says. “If I am, you will know where to find me.”

He laughs, which hurts him, and tells her to go away.

Practice

Walk first, then ride. Choose one kind of decision you now hand to a machine: a draft, a ranking, a diagnosis, a forecast. Once a day for a month, make your own call first, in one written line, before you look at the machine’s answer. Then compare. Do not try to beat the machine. Keep a simple tally: where you agreed, where you differed, and, when you later learn the outcome, who was right. At the end of the month you will know two things you do not know now: where the machine is better than you, and whether your own path in that field is still walked. The point is to use the machine without letting the grass grow back.

Ending: The Ranking on the Screen

The ranking is still on the screen in the cold room every morning. Most days Renata agrees with it.

On the days she does not, she says so in her own name, and she says what she cannot prove. Then she writes the case down. The list is longer now. Some of the cases are saves. Some are not.

Words to keep

• ability: something a person or a machine can do. Abilities can be built.

• answer for: to be the one who must explain a decision, undo it where it can be undone, and put right what can be put right. Answering is a choice made case by case, and it can be refused.

• broken leg: Paul Meehl’s name for a fact about one case that a formula has not seen. A good reason to overrule the formula, and a rarer one than experts think.

• creative courage: acting beyond what the existing evidence can establish, when you have good reason to think the evidence is incomplete, and standing behind the result.

• discernment: seeing what is significant, for this purpose, now.

• distribution: putting your work where the people who need it can find it. A channel, not a quality.

• empathy: imagining what another person feels, by imagining yourself in their place.

• heart team: a group of heart specialists who decide together, with the patient, how to treat a complex case.

• intuition: a judgement that arrives before you can explain it, built from long experience.

• judgement: deciding well when the evidence does not settle the question.

• moral imagination: the effort to see the world your choice will land in, including people who are not in the room.

• narrative intelligence: finding the true shape in a set of facts, and telling it so others can see it.

• patience: giving understanding the time it needs.

• presence: undivided attention, given from a finite life, in a world the two people share.

• quality: a settled way of choosing, worn into you by your own repeated choices, which shapes the choices you make next.

• taste: telling what is good from what is merely acceptable, for reasons that a count of other people’s preferences cannot settle.

• trust: in Baier’s words, “accepted vulnerability to another’s possible but not expected ill will.”

• abhyāsa: practice; in the Yoga Sūtra, effort held steady over a long time.

• ahaṃkāra: the “I-maker”; the sense “I am the doer”.

• buddhi: the intellect; in the Kaṭha Upaniṣad, the charioteer.

• deśa, kāla, pātra: place, time and person.

• guṇa: quality; in Sāṃkhya and the Gītā, the qualities of nature that do all acting.

• hexis: Greek, a state of character, made by habit (Aristotle).

• phronēsis: Greek, practical wisdom.

• prakṛti: nature; the whole material world, including mind and intellect.

• pratibhā: a flash of understanding that arrives whole (Bhartṛhari).

• puruṣa: in Sāṃkhya, the self as witness, which does not act.

• sahṛdaya: “one with heart”; the cultivated reader of poetry (Abhinavagupta).

• saṃskāra: an impression left by an act, which shapes the next act.

• viveka: discrimination; the power to separate the lasting from the passing.

• xìn: Chinese, trustworthiness.

Sources

1. Franz-Josef Neumann, Miguel Sousa-Uva et al., “2018 ESC/EACTS Guidelines on Myocardial Revascularization,” European Heart Journal 40, no. 2 (2019): 87–165, on the Heart Team and shared decision-making; the advice on chronic coronary disease was updated in the 2024 ESC Guidelines for the Management of Chronic Coronary Syndromes (European Heart Journal, 2024, doi:10.1093/eurheartj/ehae177), which keep shared decision-making with the patient at the centre.

2. Zhuangzi, chapter 2, “Discussion on Making All Things Equal” (道行之而成), trans. Burton Watson, The Complete Works of Chuang Tzu (New York: Columbia University Press, 1968), p. 40.

3. Aristotle, Nicomachean Ethics II.1, 1103a23–26 and 1103a33–b2, trans. W. D. Ross.

4. Aristotle, Nicomachean Ethics II.6, 1106b36–1107a2, trans. W. D. Ross.

5. Patañjali, Yoga Sūtra 1.14 (with 1.12 on practice and dispassion), trans. James Haughton Woods, The Yoga-System of Patañjali, Harvard Oriental Series 17 (Cambridge, MA: Harvard University Press, 1914), p. 35.

6. Vyāsa, Yoga Bhāṣya on Yoga Sūtra 1.5, trans. Woods (1914), p. 17.

7. Bhagavad Gītā 16.1–3, trans. Kashinath Trimbak Telang, The Bhagavadgîtâ, Sacred Books of the East 8 (Oxford: Clarendon Press, 1882), p. 114.

8. Aristotle, Nicomachean Ethics III.1, 1109b30–32, trans. W. D. Ross: “on voluntary passions and actions praise and blame are bestowed.”

9. Aristotle, Nicomachean Ethics VI.5, 1140b4–6, trans. W. D. Ross.

10. Paul E. Meehl, Clinical versus Statistical Prediction: A Theoretical Analysis and a Review of the Evidence (Minneapolis: University of Minnesota Press, 1954); William M. Grove et al., “Clinical versus Mechanical Prediction: A Meta-Analysis,” Psychological Assessment 12, no. 1 (2000): 19–30.

11. Meehl (1954), chapter 4, “The Special Powers of the Clinician”, p. 24, the “broken leg” case, paraphrased.

12. Robyn M. Dawes, David Faust and Paul E. Meehl, “Clinical versus Actuarial Judgment,” Science 243, no. 4899 (1989): 1668–1674, on clinicians who, given the freedom to overrule, identify too many exceptions, at 1671.

13. David Hume, “Of the Standard of Taste” (1757), in Essays, Moral, Political, and Literary, ed. Eugene F. Miller (Indianapolis: Liberty Fund, 1987), p. 241.

14. Abhinavagupta, Locana on Ānandavardhana’s Dhvanyāloka 1.1: yeṣāṃ kāvyānuśīlanābhyāsavaśād viśadībhūte manomukure varṇanīyatanmayībhavanayogyatā te svahṛdayasaṃvādabhājaḥ sahṛdayāḥ, paraphrased; see Daniel H. H. Ingalls, Jeffrey Moussaieff Masson and M. V. Patwardhan (trans.), The Dhvanyāloka of Ānandavardhana with the Locana of Abhinavagupta, Harvard Oriental Series 49 (Cambridge, MA: Harvard University Press, 1990), commentary on 1.1; the page of the translation not seen.

15. Brian Porter and Edouard Machery, “AI-Generated Poetry Is Indistinguishable from Human-Written Poetry and Is Rated More Favorably,” Scientific Reports 14, 26133 (14 November 2024): 1,634 participants in the discrimination study identified the source with 46.6 per cent accuracy; poets included Chaucer, Shakespeare, Whitman, Dickinson, Eliot and Plath.

16. Annette Baier, “Trust and Antitrust,” Ethics 96, no. 2 (1986): 231–260, at 235.

17. Confucius, Analects 2.22, trans. James Legge, The Chinese Classics, vol. 1 (2nd ed., Oxford: Clarendon Press, 1893).

18. John W. Ayers et al., “Comparing Physician and Artificial Intelligence Chatbot Responses to Patient Questions Posted to a Public Social Media Forum,” JAMA Internal Medicine 183, no. 6 (2023): 589–596: chatbot responses rated empathetic or very empathetic in 45.1 per cent of evaluations, physicians’ in 4.6 per cent.

19. Simone Weil, letter to Joë Bousquet, 13 April 1942: “L’attention est la forme la plus rare et la plus pure de la générosité.” Published in Simone Weil and Joë Bousquet, Correspondance (Lausanne: L’Âge d’Homme, 1982); first printed in Cahiers du Sud (1950); cited by letter and date, the page not seen.

20. Adam Smith, The Theory of Moral Sentiments (1759), Part I, Section I, Chapter I, paragraph 2.

21. Paul Bloom, Against Empathy: The Case for Rational Compassion (New York: Ecco, 2016), prologue, p. 9.

22. Diana Freed, Jackeline Palmer, Diana Minchala, Karen Levy, Thomas Ristenpart and Nicola Dell, “‘A Stalker’s Paradise’: How Intimate Partner Abusers Exploit Technology,” Proceedings of CHI 2018, paper 667.

23. Iris Murdoch, “The Idea of Perfection,” in The Sovereignty of Good (London: Routledge and Kegan Paul, 1970), p. 37.

24. Bhartṛhari, Vākyapadīya 2.150 (the passage on pratibhā runs 2.143–152), trans. K. A. Subramania Iyer, The Vākyapadīya of Bhartṛhari, Kāṇḍa II (Delhi: Motilal Banarsidass, 1977). In 2.152 pratibhā arises from nature, practice, yoga and other sources; verse numbers as in Iyer.

25. Daniel Kahneman and Gary Klein, “Conditions for Intuitive Expertise: A Failure to Disagree,” American Psychologist 64, no. 6 (2009): 515–526, at 520.

26. Demis Hassabis, “What we learned in Seoul with AlphaGo,” Google blog, 16 March 2016: “AlphaGo’s bold move 37 in Game 2 had a 1 in 10,000 chance of being played by a human.” Game 2 was played on 10 March 2016.

27. Herbert A. Simon, “Designing Organizations for an Information-Rich World,” in Martin Greenberger (ed.), Computers, Communications, and the Public Interest (Baltimore: Johns Hopkins Press, 1971), pp. 40–41.

28. Rollo May, The Courage to Create (New York: W. W. Norton, 1975), chapter 1, section 6, “Creative Courage”, p. 22.

29. Anil R. Doshi and Oliver P. Hauser, “Generative AI Enhances Individual Creativity but Reduces the Collective Diversity of Novel Content,” Science Advances 10, eadn5290 (12 July 2024).

30. Jerome Bruner, Actual Minds, Possible Worlds (Cambridge, MA: Harvard University Press, 1986), p. 11.

31. Mata v. Avianca, Inc., No. 1:22-cv-01461-PKC (S.D.N.Y. 22 June 2023), Opinion and Order on Sanctions (Judge P. Kevin Castel), slip op. at 1; penalty of $5,000 at 34.

32. Lisanne Bainbridge, “Ironies of Automation,” Automatica 19, no. 6 (1983): 775–779, at 776.

33. Krzysztof Budzyń et al., “Endoscopist Deskilling Risk after Exposure to Artificial Intelligence in Colonoscopy: A Multicentre, Observational Study,” Lancet Gastroenterology & Hepatology 10 (2025): 896–903: adenoma detection rate in standard colonoscopy 28.4 per cent before and 22.4 per cent after AI was introduced, four centres in Poland. For the test in 3.8 (abstract): a retrospective, observational study at centres taking part in the ACCEPT trial; 1,443 patients had colonoscopy without AI between 8 September 2021 and 9 March 2022, 795 before and 648 after AI was introduced; odds ratio 0.69 (95% CI 0.53–0.89); “Continuous exposure to AI might reduce the ADR of standard non-AI assisted colonoscopy.”

34. Hao-Ping (Hank) Lee et al., “The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers,” Proceedings of CHI 2025: 319 knowledge workers.

35. Isaiah Berlin, “Two Concepts of Liberty” (1958), in Four Essays on Liberty (Oxford: Oxford University Press, 1969), pp. 168–169. The original uses dashes around “and of tragedy”; they are given here as commas.

36. Gilbert Ryle, The Concept of Mind (London: Hutchinson, 1949), chapter 1, “Descartes’ Myth”, section 2, pp. 15–16, as given in the book’s own index; the printed page itself not seen.

37. Ryle, The Concept of Mind, chapter 2, p. 43.

38. Bhagavad Gītā 3.27, trans. Telang (1882), p. 55. Sanskrit: prakṛteḥ kriyamāṇāni guṇaiḥ karmāṇi sarvaśaḥ | ahaṃkāravimūḍhātmā kartāham iti manyate. See also 13.29 (13.30 in some editions): “He sees (truly), who sees (all) actions (to be) in every way done by nature alone, and likewise the self (to be) not the doer” (Telang, p. 106).

39. Bhagavad Gītā 18.14, trans. Telang (1882), p. 123; the five causes are announced in 18.13 and applied to every action in 18.15.

40. Bhagavad Gītā 18.16, trans. Telang (1882), p. 123: “the undiscerning man, who being of an unrefined understanding, sees the agent in the immaculate self, sees not (rightly).”

41. Bhagavad Gītā 18.17, trans. Telang (1882), p. 123.

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 consciousness is a single source from which twelve qualities flow, pictured as “light passing through a prism. One source, many colours.”; that the machine “cannot decide” and “has nothing to lose”; that the surgeon’s override was the measure of her judgement; that presence, empathy and patience “cannot be faked”; that a cardiac surgeon chose between a stent and a bypass with the patient on the table; that Assam’s second flush has a “muscatel” note; that home-camera abuse was a harm “no one has yet linked”; that a Lagos patent case was won on simultaneous invention; that as of 2026 “no artificial system exhibits” any of these qualities.

• What I now hold: abilities can be built; qualities are grown by the one who has them, through their own repeated choices, among other people; answering for what a quality leads you to do is a further choice, made case by case. Machines may show abilities that look like every quality on the list, and whether they experience anything is open. Their abilities were made by their makers, who answer for them; whether a machine that formed itself over years would change this is left open. Presence, empathy and patience can be imitated; when real, they are spent from a finite life, which describes them but does not test for them. Judgement is not the override: an expert’s memory is a “broken leg” only rarely, and the patient, not the surgeon, decides. Distribution is a channel, not a quality. Trust grown from long acquaintance cannot reach the newcomer; a score at least treats the newcomer like anyone else. A lost skill matters when you are kept to check the machine or to catch its failures. The decision is made in the heart-team meeting, with the patient; the tea is malty; the abuse was documented in 2018; the patent case turns on prior public disclosure. The Gītā’s teaching that the qualities of nature do all acting does not remove answering; it removes the claim to be the only cause.

• What changed my mind: the audit’s finding that “consciousness” had been stretched to cover whatever the human got right; Aristotle’s account of hexis and Patañjali’s of abhyāsa; Meehl on clinical and statistical prediction; Ryle’s dispositions; the Gītā 3.27 and 18.14–17, read together; the 2023 empathy study; the 2024 poetry study; the 2025 colonoscopy study.

• What survives: the twelve, and their people; Renata’s three saves and her one needless override; Njeri’s failed Kisumu wager; Bora’s two seasons of adequate tea; 3.8, now with evidence; “someone who has something to lose”, now made exact.

Next in the series: Chapter 4: Why This Matters Now.