# AI-Proof: What Machines Can Never Be — Volume II: Application, Part V: Conscious Strategy, Chapter 28: AI-Proofing Sectors

By Nithin Palal · 21 November 2026 · 45 min read
Section: Artificial Intelligence · Series: AI-Proof: What Machines Can Never Be
Source: https://nithinpalal.com/ai/ai-proof/chapter-28-ai-proofing-sectors

> A teacher’s stories are asked to become a state’s template. A sensor sees a widow’s sugar rise and not why. No sector is proof against machines. Each must decide who may make the exception its rules did not foresee, and who pays.
*Part V looks at strategy. After vision, intermediaries, verification, speed, regulation, business ideas and investment, this last chapter of Part V, and of Volume II, asks what every sector must keep a person for, on purpose.*

← Previous: [Chapter 27: The Investor’s Playbook](https://nithinpalal.com/ai/ai-proof/chapter-27-the-investors-playbook) · Series hub · Next: [Volume III: Transformation (overview)](https://nithinpalal.com/ai/ai-proof/volume-3-transformation) →

**A teacher’s stories are asked to become a state’s template. A sensor sees a widow’s sugar rise and not why. No sector is proof against machines. Each must decide who may make the exception its rules did not foresee, and who pays.**

**In Brief: For readers who wonder what, in their own line of work, should stay with a person. Chapter 28 asks whether anything stays a person’s in every sector when machines can do much of the work. A teacher in Sagar takes her teaching stories to a committee in Bhopal; a widow in Secunderabad wears a sensor that sees her sugar rise; a food-ration reform in Jharkhand cuts theft and shuts people out; a designer in Mumbai loses to a tested design; an adviser in Pune promises to be there next month; an exporter in Coimbatore finds that his loyalty was habit; a farmer near Moga weighs his field against his phone; a plantain seller in Accra loses part of her evening crowd to a chain. Aristotle on equity and the leaden rule, and the Buddha’s advice to Sigāla on duties that run both ways, help show who may make the exception a rule did not foresee, and what each side owes.**

**No sector is proof against machines; what each must keep, on purpose, is someone who may make the exception its rules did not foresee and who explains and repairs it, and what that person owes differs from sector to sector.**

## Opening: An Invitation from Bhopal

Kavitha Sharma was marking Class 9 notebooks in the staffroom when the invitation reached her.

It was July 2026, early in her thirteenth year of teaching science at a government school in Sagar, in Madhya Pradesh. The district education officer, Suresh Tiwari, had forwarded it from Bhopal, the state capital. A committee reviewing the use of AI in the state’s government schools wanted her to present her method in September. It might go into the state’s guidance for teachers.

Her method had a plain name. She called them her teaching stories. For a year a learning platform had taught her students the syllabus twice a week, patiently and well, on tablets shared in pairs. What she added were stories about how someone came to know something, chosen for the moment a particular class had stopped understanding. In May she had sent four of them to the state programme. In March she had sent the district office a one-page note about the rubric of its certification course. The rubric gave marks for using AI at every stage of a lesson plan. Nobody had replied to the note. The stories had brought an invitation.

Her first thought was pride. Her second, arriving before the pride had settled, was dread.

She did not have a framework. She had a practice. It had grown from the neem tree in the school courtyard and from the fourteen-year-olds in front of her. She had learned to read their silences slowly, with a notebook. A committee in Bhopal would want something every science teacher in the state could use. She suspected that her practice would not survive the journey from a staffroom in Sagar to a committee room in Bhopal without becoming something else.

Every sector in this chapter has the same trouble in its own clothes. A rule is written for everyone. A machine now applies many such rules faster and more evenly than people do, and in some places better. Then a case arrives that the rule did not foresee. It may be a widow whose sugar is rising for a reason no sensor records, or a ration card that fails a fingerprint. It may be a lender whose customers must trust it for years, or a field whose clay lies nearer the surface than any map shows. The question is the same in every sector. **When machines can do much of the work in every sector, is anything a person’s in all of them, and who decides what it is in each?**

The print edition answered with one sentence for eight sectors: “AI makes intelligence cheap and consciousness valuable.” It said that “The twelve qualities apply everywhere.” This edition cannot say either. No quality, and no consciousness, does the same work in a classroom, a clinic and a ration shop. What the sectors share is smaller and more useful. In each, someone must hold a position. It belongs to the one who may make the exception a rule did not foresee, and who must explain the exception and put right what it costs. Who that is differs by sector, and the sector’s law, money and history decide it. The eight sectors below are examples and do not cover every kind of work. The claim rests on something every sector has, a general rule meeting a case it did not foresee. So the chapter’s title can mean only one thing. To “AI-proof” a sector is to decide, before the case arrives, who holds that seat and who pays for it. It keeps nobody’s job. What it can protect a sector against is a failure that nobody owns.

Naming the seat is not enough, and it is easy to do badly. An institution can give someone the right to sign without the evidence to examine the case, the time to understand it or the power to change it. The signature then becomes a place to leave the blame while the decision stays elsewhere. A rule that asks only for “a human in the loop” can be met in this way. Four things are needed: the right to depart from the rule, the knowledge to see when departure is justified, the duty to explain, and the means to repair. They need not sit in one person. Where they are split, each part needs a name. Without them, a seat is furniture.

The exception is also not the only place where people decide. Before any rule meets a case, someone chose what the rule is for: what the machine is asked to achieve, which trade-offs it may make and whose interests count. Chapter 9 called this the frame. After a case has gone wrong, someone must be able to review it, reverse it and change the rule. The exception sits between these two seats, one at the design of a rule and one at its review. This chapter follows the middle seat because that is where the sectors differ most. A sector that fills only the middle one leaves the other two to whoever happens to hold them. And a seat need not be one person. It can be a team, a review board or a tribunal, as long as its powers and its duties have names.

An old text makes the move this chapter needs. The *Sigālovāda Sutta* is the Buddha’s advice to a young householder named Sigāla, in the long discourses of the Pāli canon. In Nārada Thera’s translation¹, the Buddha finds him early one morning “with wet clothes and wet hair”, and he “worshipped with joined hands the various quarters”. His dying father had told him: “The six quarters, dear son, you shall worship.” The Buddha does not tell him to stop. “It is not thus, young householder, the six quarters should be worshipped in the discipline of the noble.” He keeps the six directions and fills each with people. “The mother and father are the East”; “The Teachers are the South”; “Servants and employees are the Nadir”, the direction below one’s feet. Each relation then carries five duties in each direction. The pupil owes the teacher, and the teacher owes the pupil. The worker owes the employer, and the employer owes the worker. The sutta is advice to a layman about his household, not a theory of economic sectors, and I borrow only its move. A sector is not a direction to bow to. It is a set of relations, and their duties run both ways. The sutta gives the grammar of those duties. It does not give the means to enforce them, and a relation with duties on both sides can still be unequal. Enforcing them is the work of laws, appeals and bodies that speak for many people at once.

## 28.1 Education: Teaching Students How to Know

In Bhopal, in September, Kavitha told the committee about one lesson. The previous October she had told her class about Jagadish Chandra Bose, a physicist in Calcutta. Around 1900 he turned instruments of his own design on plants and recorded their electrical responses.² A quiet girl in the second row asked whether what you believe can make you see things that are really there. The next day Kavitha told a second story beside the first. In 1903 a respected French physicist, René Blondlot, announced rays that did not exist, and his belief made him see them. He went on seeing them after another scientist quietly removed part of his apparatus.³ Belief can open the eyes, she told her class. It can also fill them. What tells the two apart is checking.

The committee listened. Then it asked what committees must ask. Could this be written down so that every science teacher in the state could do it? Could it be scored? Its members were not wrong to ask. A state cannot rely on every teacher inventing stories alone. Most teachers, on most days, need good material more than they need freedom. That is the true half of a template.

The other half is old. Aristotle, in the fourth century BCE, wrote about the gap between a law and a case. In W. D. Ross’s translation⁴, “all law is universal but about some things it is not possible to make a universal statement which shall be correct”. He did not blame the lawgivers; the fault lay in the subject. So he defined *equity* (*epieikeia*) as “a correction of law where it is defective owing to its universality”. When a case arises that the general rule does not cover, someone must say what the lawgiver “himself would have said had he been present”. Then he took an image from the builders of Lesbos, who cut curved mouldings in stone. “For when the thing is indefinite the rule also is indefinite, like the leaden rule used in making the Lesbian moulding; the rule adapts itself to the shape of the stone and is not rigid”. A measuring rule made of lead bends to the stone it measures.

Aristotle wrote about laws and judges, not schools, and I borrow only his point: a general rule needs someone at the particular case. His test is also a limit. Equity corrects a rule in the direction of the rule’s own purpose, saying what the lawgiver would have said. An exception that serves the person who makes it, or that is given to one person and refused to the next in the same case, is favour. Telling equity from favour needs a written reason for each exception, set beside the next case like it. Kavitha’s teaching stories are a leaden rule. The Bose lesson was chosen for one silence, in one class, on one afternoon. A module written in Bhopal is a rigid rule, and that is what a module is for. The framework must bend to the sector, not the other way around.

Here the machine has a part to play. As of 2026, a learning platform bends better than a module. Kavitha’s chooses the next question to fit each student, and it could have told the Bose story too, perhaps more smoothly than she did. What it did not hold was the standing to depart from the plan. Nor did it hold the duty to say why to a parent, a head teacher or a district office. Aristotle’s question is who may bend the rule, and who must answer for the bend. The platform already bends within its own design, each time it changes the next question for one student. It could be allowed to bend further, even to leave the plan. Then the seat would belong to whoever allowed it, and that person would answer for what the bends cost. Letting a machine bend is a decision, and someone makes it.

The version that travels will not be her practice. The tension between the two is permanent. It is also a decision. Someone in Bhopal will choose what travels. It could be the stories themselves, the habit of setting a success beside a failure, or a score for “using stories”. Choosing is the committee’s work, and so is standing behind the choice. Her part is to say plainly what the travelling version will lose.

Assessment is where the change costs most. As of 2026, a machine can write a good answer to almost any question a student takes home. The obvious proposals are oral examinations and reasoning done in front of a teacher. A student who explains her answer aloud shows what she can do outside the conditions a machine controls. That matters because doing well with a machine beside you is not the same as having learned. In a trial in Türkiye reported in Chapter 37, students who practised mathematics with an ordinary chat assistant did better in practice and worse in the exam taken without it. A version that gave only hints did no such harm. Checks in front of a teacher are also expensive, and so are the others: drafts that show their revisions, problems unlike the ones practised, and work done in the room. Ten minutes with each of forty students is most of a day of a teacher’s time, and somebody has to give that day. If the office that sets the examination does not give it, the teacher gives it from her evenings. Checking a machine’s work against the world costs time in the same way. In August a boy brought her a machine’s report on the water from the school’s hand pump, written from what is usually true of water in the region. She took a sample to the district laboratory herself, paid the fee, and waited a fortnight to learn that, this time, the pump agreed with the report.

The sutta’s list for teachers fits better than talk of cultivating consciousness. In Nārada’s translation, teachers “see that they grasp their lessons well” and “provide for their safety in every quarter”. It gives the teacher duties to her pupils. It says nothing about what an office owes its teachers, and that is where Kavitha’s year had been decided. The platform came in August 2025 because the district office chose it. Nobody had asked her what her periods were now for. On 1 April 2026 the Ministry of Education announced a curriculum on AI and computational thinking for Classes 3 to 8.⁵ The Central Board of Secondary Education would carry it out from the 2026–27 session. It does not reach her Class 9, and what her own state adopts is the state’s decision. The hours she spent preparing for Bhopal were her own evenings. Evenings given by one teacher at a time are a virtue that an office has learned to depend on. Protecting them is the work of a rule: a timetable, a budget line, an agreement made with teachers as a body.

## 28.2 Healthcare: The Numbers and the Life

Sarojini is seventy-two. She lives alone in a small flat in Secunderabad, with a cane chair by a window that looks towards Hussain Sagar, the lake between the twin cities. She has type 2 diabetes and mildly high blood pressure. She has never agreed that either needs more attention than a household did. For forty years her friend Leela telephoned her at eight every morning to talk about the news, the neighbours and their grandchildren. In early May 2026 Leela died of a stroke in the night.

Her daughter Meera visited on Sundays from Banjara Hills, across the city. Her son, in Chennai, had bought her a glucose sensor in March. It was a small disc worn on the arm, reading her sugar every few minutes. Its app also took her steps, her sleep and her heart rate from a watch he had sent with it. It sent a weekly summary to her doctor’s clinic in Begumpet. She had agreed to it the way a mother agrees to a son who has already paid. Nobody had asked her what she wanted her doctor to see. Agreeing to wear the disc was one permission. Letting the clinic see her readings was another, and letting a program interpret them, or her children read them, would have been others. Nobody had separated them.

After Leela died, Sarojini sat in the cane chair. She cooked less and slept in pieces. She stopped walking to the temple in the evenings. On the telephone she told her children that she was fine.

The app saw what none of them saw. Over three weeks her fasting sugar had drifted up. No single morning’s reading showed it; the trend did. The app flagged it and suggested that her doctor review the dose of one of her two tablets. This call goes to the machine first. The trend was real, and the app found it weeks before any visit would have.

Her two tablets are not alike. On its own, metformin is “unlikely to cause hypoglycemia”, the dangerous fall in blood sugar.⁶ Gliclazide is different. It belongs to a family of drugs called sulphonylureas. It “may cause low blood sugar”, and people who take it are advised against “skipping or delaying meals”.⁷ The app had suggested raising that one, for a woman who had begun to skip meals.

Her doctor had treated her for eleven years, and he knew about Leela. Sarojini had told him at her last visit, a week after the funeral. The fact the app lacked was in a conversation, not in its data. As of 2026, a system that reads someone’s steps, sleep and heart rate can be built to ask whether something has changed in her life. It would still not know what had changed, unless someone told it.

He did not change the dose. He telephoned her. He asked about her sleep, and her food, and the temple. Somewhere in the conversation she said Leela’s name and stopped speaking for a long time, and he waited. Then he told her what he could not prove. Grief can keep blood sugar high. Stress hormones make insulin work less well, and “If stress doesn’t go away, it can keep your blood sugar levels high”.⁸ A stronger dose of a tablet that lowers sugar whether or not she had eaten could take her too low. He gave her the choice. She could change nothing for two weeks and have her readings looked at again, or start the higher dose now. She chose to wait. She asked him not to tell her children that she had cried. That evening he wrote his reason in her notes: her friend’s death, the meals she no longer cooked, and her choice to wait.

Neither of them was simply right. By the middle of June her fasting readings had come down a little, though not to where they had been. In July he raised her metformin by one small step. The app had seen a real problem and proposed a risky remedy for it. The doctor had seen what came first, and he still needed the trend. Neither proves the other wrong. The exception was the doctor’s to make. He could set aside the app’s suggestion, and he would have answered for what followed. The choice between the two courses he offered was Sarojini’s, because it was her life. The risk could have run the other way. A tired doctor can accept a machine’s suggestion because it looks precise, and a proud one can wave away a true alert. In one study, radiologists rated mammograms after seeing a suggested rating said to come from AI. When the suggestion was wrong, even the most experienced rated fewer than half correctly.⁹ What protects a patient is a written reason, whether the doctor follows the machine or not, kept where someone can check it.

The gap between what a machine finds and what a patient knows is the sector’s, not only Sarojini’s. In a trial published in February 2026, researchers at Oxford gave members of the public medical scenarios to work through with language models. The models on their own identified the relevant conditions in 94.9 per cent of cases. The people using them did so in fewer than 34.5 per cent, no better than people without them.⁹ The trial did not test doctors, so it does not say whether the models knew more than a doctor would. What was in the machine did not become what the patient knew. Healthcare happens in that gap, between the numbers and the life they describe.

Suppose the dose had gone up on the app’s suggestion, and Sarojini had fallen in her kitchen. The doctor who signed the prescription would have had to explain it and put right what he could. So, through its people and its money, would the company that built the suggestion. What no company had paid for was the call. His clinic is paid by the visit, and a quarter of an hour on the telephone with a grieving patient was his own time. A system that pays for visits and not for calls depends on doctors who make the calls anyway.

In the sutta’s map, Leela had been the North. Friends, in Nārada’s words, “do not forsake him in his troubles”. A doctor is not a friend, and should not have to be. What he owed was smaller and harder. He had to look at the whole case before he signed a dose, and then leave the choice with the woman whose life it was. He did both.

## 28.3 Government and Public Service: The Accountability Layer

A customer can go elsewhere. A citizen whose benefits application is denied by an algorithm cannot switch governments. That is why accountability is the definition of public office, not a constraint on it: the obligation to explain, justify and correct decisions that affect people’s lives. The difficulty is the word *correct*.

The strongest evidence in this sector is Indian, and it cuts both ways. India’s public distribution system sells subsidised grain through ration shops. It is the country’s largest social protection programme. From August 2016 the state of Jharkhand began to require people collecting grain to prove who they were with a fingerprint, matched to Aadhaar, India’s biometric identity number. In July 2017 it tied the grain it sent each shop to the transactions the fingerprint machines recorded, and paused that rule four months later. Three economists, Karthik Muralidharan, Paul Niehaus and Sandip Sukhtankar, studied the reforms with randomised and natural experiments.¹⁰ “Corruption fell but with substantial costs to legitimate beneficiaries,” they wrote. Between 1.5 and 2 million people lost access to their grain at some point during the reforms.

On one count the evidence is on the machine’s side. Tying the grain to recorded transactions sharply cut the value of grain sent to the shops, and most of the cut was grain that had been leaking away.¹¹ Dealers who had stolen could steal less.

It was also a rule that did not foresee many cases. Among households that had not linked an Aadhaar number to their ration card, nearly a quarter of the total, the grain they received fell by about a tenth. Their chance of receiving none at all rose by 10 percentage points. People made more trips to the shop to collect the same grain. The researchers placed the cause where it belonged: “Adverse effects appear to have been driven primarily by decisions about the way the transition was managed”. Those were decisions, and people made them. Whether less theft was worth that much exclusion, managed as it was, is not something a study can settle. It is a choice about who bears which loss, and the state that made it answers for both. In Telangana, reporters found in January 2024 that the state had cancelled ration cards on a welfare algorithm’s flags, “several thousands” of them wrongly.¹² A citizen refused by a rule has nowhere else to go, which is why the refusal must be explicable, and reversible, by someone.

The two old texts meet hardest here. Aristotle gives the exception to someone who judges, saying what the lawgiver would have said. In a ration shop the person on the spot is the dealer, and the dealer’s discretion was where the grain had leaked. A rule that bends can also bend towards a cousin. The sutta does not leave the weaker side to a superior’s goodwill. It writes each side’s duties into the relation before anyone needs them. Read together, and applied to a public office (the application is mine), they ask for three things written down before the first refusal. The exception needs a named holder who is not the person with the motive to cheat. The exception must be recorded. And the person refused is owed a reason and a way back. The researchers end on the same point in their own words: the effects of a new technology “can be highly sensitive to the protocols governing their use”.

Where nobody holds the exception, a person may still sit beside the machine. An officer may sign the list of the excluded without the evidence, the time or the authority to change it. That is a name on a decision. It is not a person who answers for it.

A named holder is also not the only door. The holder may be absent, mistaken or under pressure from above, and a reason is not yet a remedy. A person can be told why she was refused and still be unable to correct the record or get back what she lost. So a public system owes three different things: a reason, a review by someone who is not looking at the same score, and repair when the refusal was wrong. India’s food law of 2013 already names such a review for grain: a grievance officer for every district, with an appeal to a state food commission.¹² Where one rule has wrongly refused thousands, as in Telangana, letting each person appeal is not enough. The office must find the pattern and stop the rule. It must also go back to the people it reached, as Chapter 31 asks of any builder whose system has wronged people. The state here is the machine’s user, not only its regulator, and it owes what it would ask of a company. The person refused by a public system must not depend on the goodwill of the one who refused her.

The district office in Sagar is a public office too. It decided that the platform would come to Kavitha’s classroom. In December 2025 it required every government schoolteacher in the district to finish sixty hours of certification in AI by the end of March. Anyone who did not would have it noted in the annual appraisal. It never replied to her note on the rubric. Its accountability ran upward, to the state’s targets, more readily than downward, to the teacher it governed. That is common, and it is what this sector has to change if the definition is to mean anything. Explaining is the easy part. Correcting needs a seat.

## 28.4 The Creative Economy: When Taste Meets the Market

Arjun Mehta designs brands. His studio is two rooms above a printing shop in Parel, in Mumbai. In September 2025 a start-up that lends money to small shopkeepers through an app asked him for a complete identity. He spent a week on it. By his own judgement it was the best work he had done in a year. The type was restrained. The colours remembered the red cloth ledgers of Indian traders without imitating them. The design said *we take your money seriously*, not *we are exciting*.

The client liked it. Then it tested it. In an A/B test, two versions of a page are shown to different visitors, and the version more of them act on is kept. The client set Arjun’s design against several that a machine had produced in an afternoon: louder, brighter and more like other lending apps. A machine-made design won on every measure the client counted, from clicks to time on the page to sign-ups. Arjun’s design was better. The other design worked better.

On the client’s terms, the machine won. The test measured honestly what it measured. Arjun believed it measured the wrong thing. A lender asks people to trust it with their shops for years, and a first click is a month’s evidence. He could not prove it. Arjun’s “better” was a judgement about years. The test’s “worked better” was a count of first clicks and sign-ups. A longer test could have counted whether shopkeepers stayed and repaid. It would have taken months, and not spending them was also the client’s choice. He had a habit, learned after a hotel chain in Goa rejected a website of his. He wrote three arguments against his own work in the voices of the people who would live with it. Two of the three went against him.

He declined to redesign towards the louder version, and said why in writing. In November 2025 the client chose the machine’s design, and Arjun lost the project and a fee he had counted on. He may have been wrong. Taste is a judgement, and judgement can be wrong.

What does not survive is the print edition’s claim that “Taste is the irreducible human quality in creative work.” As of 2026, machines make competent images, type and layouts in seconds, and buyers use them. Researchers studied a large online marketplace for freelance work after ChatGPT was released.¹³ Freelancers in the most exposed occupations got about 2 per cent fewer jobs a month, and earned about 5.2 per cent less. Choosing among outputs is not beyond machines either; in this story the test chose. What was the client’s was the choice of measure, and the client was entitled to it. Someone always chooses what counts as success. Being entitled to choose the measure is not the same as choosing it well. A count of first clicks stood in for years of trust. A measure that stands in for something can be met while the thing itself is lost, which Chapter 9 called Goodhart’s law. A designer can also hide a weak design behind a long-term value nobody can measure. The test could not tell which of the two had happened here. The exception the test could not foresee, a lender whose customers need trust before excitement, was the client’s to make. Arjun’s part was to say it in time and in writing, and to accept being refused. Working for himself, he carried the loss alone.

Finding clients who can tell what tests well from what lasts is part of the work. That is true, and it promises nothing. The start-up’s app launched with the machine’s design. Whether that design will age badly is not known. If it does, that will not prove him right.

## 28.5 Legal and Financial Services: When the Client Already Knows

A lawyer’s metaphor goes like this. Being a lawyer used to be like being a librarian, paid to find information. Now it is like being a priest. The image catches the collapse of an old advantage. Many clients now arrive having read a machine’s analysis of their case. The image gets the remedy wrong. A priest is believed on faith. A client should believe a lawyer as far as she can check what she is told, or as far as the lawyer stands behind it.

These clients arrive not ignorant but partially informed, and partially informed in ways that may be subtly wrong. Picture a lawyer meeting such a client. He says the machine told him to sign. She knows his business, the other party and what the contract leaves unsaid, and she replies: “The terms are standard. But given what you told me about your plans for next year, I’d want to negotiate clause seven.” The machine read the contract. She read the client. A lawyer and an investment adviser work under different duties and different regulators. What they share is advice that a client acts on, and a person who must answer for it.

Bhoomi Agarwal is an investment adviser in Pune, registered with SEBI, the Securities and Exchange Board of India, which regulates the country’s securities markets. In April 2026 two of her clients received the same recommendation: the same rebalancing, produced by the same software. One accepted it by email. The other, a man saving for his daughter’s university fees, came in with a machine’s analysis of his own, which proposed almost exactly the same changes. He asked her why he should pay her.

In both offices the advice was mostly the machine’s, and it was good.

What she did in the hour was to say what the numbers meant for one date: the month his daughter’s fees fall due, in July 2027. A fall in the markets in the year before that date would leave him no time to recover. So she moved part of the fund into steadier holdings sooner than the software proposed, and told him that this would probably cost him some growth. Then she said one sentence that is worth keeping word for word: “Here is what I would do, and here is why, and I will be here next month to answer for it.”

The sentence needs one correction to stay honest. To answer, in this book’s sense, is to explain and to put right. Bhoomi can explain. She can put right some things: a recommendation she got wrong, a change she forgot to make, a fee she should not have charged. She cannot put right a market fall, and she told him so. A promise is honest only about what is in the promiser’s power.

The print edition said that “The AI cannot make a promise.” That is no longer the point. As of 2026, companies make written promises about what their machines do, and back them with money.¹⁴ India’s securities regulator has also decided where the burden of a machine’s advice falls. Since February 2025, its rules have said that any person it regulates who uses artificial intelligence “shall be solely responsible” for “the output arising from the usage of such tools and techniques it relies upon or deals with”.¹⁵ The software’s advice is, in law, Bhoomi’s advice. As I read it, the rule settles whom the regulator will hold to account. It does not settle what Bhoomi could recover from the software’s maker, which depends on their contract. Nor does it settle what a client could recover from her, which depends on negligence and on what she promised. A rule like this can also pass a burden down to someone who cannot see inside the tool. Bhoomi chose her software, so the burden fits her. An adviser in a large firm who must use the firm’s tool and sign for it has a name on the decision and less of the seat. The firm that chose the tool holds the rest of the seat, and should be named as holding it. The client’s own analysis had nobody behind it. If it had been wrong, no one would have had to explain it to him or put it right. What her client bought, beyond an answer, was a person’s judgement about his circumstances, her duty of care, and her promise to be there next month. By law and by her own words, she has agreed to stand behind what she used.

Her fee was the same for both clients. Her hours were not. The second client took an afternoon; the first took a minute. The difference came out of her own week.

## 28.6 Manufacturing: What the Data Revealed About the Man

Rajan Subramanian makes cloth in Coimbatore. The business, in Tamil Nadu, has been his own since 1997. About 140 people weave and hand-finish cloth on his floor. His yarn is dyed for him by dye houses in Tiruppur, and he sells the cloth to buyers in fourteen countries.

In early 2025 his nephew Karthik ran ten years of Rajan’s records through a tool that scored every supplier by late deliveries, rejected lots and complaints. Near the bottom was a dye house in Tiruppur whose owner Rajan had liked for more than a decade. It charged more than two others of the same quality, and its lots came late more often. Rajan looked at that line, and kept the dye house for another year.

In April 2026 he opened the list again, with fifteen more months in it. Nothing had changed, except that there was more of it. Two of the dye house’s lots had needed dyeing again. The record was right about the dye house. What it showed was something about Rajan. His loyalty to that owner had not been judgement. It had been habit dressed as judgement.

The print edition said of such a tool that “The AI could not make that distinction. Only Rajan could.” Its own story said otherwise. In the print, the tool’s advice had been to keep the old supplier for special orders and move the rest, which is the distinction, made by a machine. Weighing price against reliability was a sum, and the tool did it well. Loyalty can buy things a record does not count, such as being served first in a shortage. Loyalty that can name what it buys is a judgement, and its reason can be written beside the record. Loyalty that cannot name it is habit. What was Rajan’s was the decision, and its cost. Here the usual roles were reversed. The rigid rule was Rajan’s, kept for years without a reason, and the machine was the one that bent it. His part was the other half of the seat: to decide, and to pay the debt the decision left.

Between April and June 2026 he moved that work to two other dye houses in Tiruppur. He did not send an email. He went to the dye house, sat with its owner, showed him the record and asked which entries he disputed. Two of the late lots had followed Rajan’s own late change of shade, and Rajan struck them out. The rest stood. He gave the owner three months’ notice instead of none. That September, in Singapore, Rajan would ask a platform to give each mill he introduced the same chance to see its record and reply. In Tiruppur, answering for a decision meant explaining it to the man it hurt and giving him time.

It cost Rajan something, and it cost other people more. The owner was polite, and has not telephoned since. Rajan’s orders had been a large share of the dye house’s work, and by July two of its dyers had lost shifts. They had no part in the decision and paid for it without being asked. A buyer owes a supplier no permanent business. But when one buyer’s orders are a large share of a small firm’s work, his decision reaches people his contract never named. The time he gives them is part of what he owes. The saving was Rajan’s. Whether it goes to his own floor’s wages, to his prices or to his margin is a choice he has not yet made.

The old sutta has a list for the floor, not for the supplier. The employer, in Nārada’s translation, owes his workers five things, among them “assigning them work according to their ability”, “supplying them with food and with wages”, “tending them in sickness” and “granting them leave at times”. The same passage says what the workers owe in return, and part of it should be refused outright: “they rise before him”, “they go to sleep after him”, “they take only what is given”. A duty that runs both ways can still run uphill. The sutta does not reach a dye house’s workers in another town, and neither does any law Rajan works under. Only his decision reached them.

## 28.7 Agriculture: One Field and the Forecast

Before dawn on a January morning in 2026, Kesar Singh walks his wheat outside Moga, in Punjab. The fog lies flat on the fields. The wheat was sown in November, after the paddy harvest, and stands as high as his shin. Farmers there are told to sow by the middle of November, and a late paddy crop pushes the wheat back.¹⁶ He carries a stick that was his father’s, the right length for parting the crop to look at the soil.

His eastern field holds water differently from his western one. The clay lies nearer the surface there. No satellite has mapped it, and the advisory app on his phone does not know it. He knows because his father once pulled a handful from a drainage trench and showed him its colour. Part of what he knows is a record that nobody kept. A soil sensor in the eastern field would give the app what his father gave him.

The app reads the satellite’s view of his crop and the weather forecast, and it tells him when to water. In most weeks its timing has been better than his own. Some mornings the system is right and he is wrong. He has learned this. Some mornings he is right and the system is wrong. He has learned this too. Each can be wrong in a way the other cannot see. His memory is of a climate that is changing, and the app learned from years that may not come again.

At a larger scale the machines’ case is stronger. In 2025 India’s agriculture ministry sent forecasts of the monsoon’s arrival to about 38 million farmers in 13 states of central and north-eastern India.¹⁷ The forecasts were made with AI weather models and went out by text message, up to four weeks ahead. They saw a stall in the monsoon’s advance that other available forecasts missed, and the monsoon did stall. Kesar’s wheat is sown after the rains, so an onset forecast does not touch this decision. But nearly half of India’s workers, 46.1 per cent in 2023–24, work in agriculture.¹⁸ A forecast that reaches 38 million of them is not a small thing.

That January morning the app advised watering both fields before a dry spell. He watered the western field and left the eastern one, because its clay would hold water for days. The app’s rule had not foreseen that, and the exception was his to make. A calculation optimises. A decision commits. Here *commits* means something plain: if he is wrong, the eastern wheat is his family’s loss, and no one else’s. If he follows the app and the app is wrong, the loss is still his family’s. The app advises, and it does not share the harvest.

Much of what decides his year is not his. On 1 October 2025 the central government set the minimum support price for the wheat he will sell in 2026, at ₹2,585 a quintal, a hundred kilograms.¹⁹ His grain is sold at the *maṇḍī*, the market yard. When the paddy may be transplanted, what the *maṇḍī* pays and how much water lies under the fields for his sons are decided mostly by people he will never meet.²⁰ An app can also be right for every field and wrong for all of them together. If it tells every farmer in a district to water on the same dry morning, each field gains and the water they all draw on falls. A farmer can hold the exception in his own field. The sector’s larger exceptions belong to others. They owe him a reason too.

### The glitch

The law that sets Kesar’s transplanting date was written to save groundwater, and it seems to. Punjab passed it in 2009. A study published in 2019 reports that simulations suggest such laws have significantly reduced the fall in groundwater.²⁰

The same study found a cost that nobody had planned. The laws forbid transplanting before a date in June. Paddy planted later is harvested later, about eight days later on average, and the wheat must still be sown on time. Burning the rice straw had become the quick way to clear a field for wheat. After the laws, the burning crowded into a narrower window later in the autumn, with a peak 39 per cent more intense. In New Delhi, the average level of fine particles in the air in November was 29 per cent higher. The authors offer these as first inferences, to be confirmed by more monitoring and modelling.²⁰

If they are right, a rule that was good for the water was bad for the air, and many of those who breathed it lived far from any paddy field. No farmer could make that exception in his own field. A water rule is usually judged by the water it saves. The seat that reviews it has to look past its own sector.

## 28.8 The Informal Economy: AI-Proofing Without a Corporate Structure

Ama Nkansah sells roasted plantain and groundnuts. Her place is a junction in Osu, a neighbourhood of Accra. The traffic slows there at dusk, and people going home from offices are hungry enough to stop. She has been at this junction for nine years. Her spot, between a tailor’s shop and a kiosk that sells phone credit, is hers because everyone in the neighbourhood knows it is hers. No lease says so.

Every day she makes the decisions a firm makes. She chooses which plantains to buy and from whom, how many to roast and when. She decides how much to keep for stock and how much to send to her mother in Kumasi. Each is a judgement made with her own money at risk.

In February 2026 a chain opened at the junction’s other corner. It sells fried plantain chips in printed packets, about a third cheaper than Ama, delivered by vans whose routes are planned by software. The chips are always the same. In Ama’s judgement they are too thin and too dry. For two months her evening crowd thinned. By June some came back, among them the woman in the blue car who wants extra groundnuts and the man from the insurance office who buys two portions on Fridays. Not all of them came back.

But this book owes Ama honesty, not romance. The customers who did not come back are not wrong. They chose a lower price and a packet they could predict, and that was a reasonable choice. The chain’s side has a case too. It can learn its regulars. As of 2026, an app remembers a favourite order better than most sellers do. If the chain cuts its price again, or opens a second shop, Ama’s evenings will get harder, and nothing in her nine years protects them.

Most of the world’s jobs are informal. In modelled estimates published in 2025 by the ILO, the United Nations’ labour agency, informal employment was 57.8 per cent of all employment worldwide.²¹ For such workers the nearer risk, as of 2026, is not displacement. It is exclusion. Tools, credit and protection reach the formal economy first, and the gap grows. Informal work is not one condition, either. A street vendor, a domestic worker, a waste picker and a delivery rider whose pay an app sets meet different machines. For some of them the risk is displacement after all.

What Ama lacks is the structure the other sections take for granted. Nobody owes her notice before a chain opens beside her. No office can make an exception for her, because no rule covers her. Her relations carry the duties that a firm spreads across contracts, insurers and laws. The sutta’s friends do that work. In its words, friends “become a refuge when he is in danger” and “do not forsake him in his troubles”. The tailor watches her stall when she goes to buy plantains. The wholesaler gives her a week’s credit in a bad month. That is a real structure, and a thin one.

One proposal is tools in local languages, for the phones people actually use, that bring machines’ benefits to informal workers without making them register as firms. It is a good proposal, with three questions attached. Who builds such tools? Who pays for them? And could Ama see what a tool says about her, and dispute it, before a lender or a city office acts on it? Registering as a firm brings credit and protection, and also fees, inspections and taxes. The question is how to give informal workers the first without making survival dearer. As I read the record, where informal workers have won a seat at all, they have mostly won it together. India’s street vending law of 2014 gives vendors, elected by vendors, at least two-fifths of the seats on each town’s vending committee. In 2021 a parliamentary committee found that not every eligible town had one yet.²²

Across the eight sectors the same structure recurs. A general rule or model meets a case it did not foresee, and someone must be able to depart from it, explain the departure and repair what it costs. In the classroom, the clinic and the adviser’s room a person held that seat. In the dye house the roles were reversed: the machine bent a person’s rigid rule, and the person’s part was to pay for the change. At the junction there was no rule to bend and no seat at all. And those who held the seat paid for it themselves, in Kavitha’s evenings, the doctor’s quarter of an hour and Bhoomi’s afternoon. A seat that rests on unpaid virtue is held only as long as the virtue lasts.

Three more things decide whether such a seat lasts. The first is competence. Everyone in this chapter who made a good exception had spent years on the routine work that machines now do. Kavitha had more than twelve years of teaching, the doctor eleven years with one patient, and Kesar a field learned from his father. If the routine goes to the machine and nobody keeps a place to learn it, the next holder of the seat will never have worked the stone by hand. Chapter 4 called this the missing rung. The second is sight. Not every case the rule did not foresee arrives as a flag. Some failures look like success for a while. A student passes without learning; a supplier gets fewer orders without knowing why; advice works for a season and costs the soil. They are found only by an institution that goes back to see what happened to people. The third is a way to stop. A school, a hospital or a ration office must be able to suspend a system that is failing. It must also be able to keep working while the system is off, and it cannot once nobody remembers how the work was done without it. Rules, data and money also cross from one sector to another, so no sector can build its seats alone. A ration card in Jharkhand could fail at the shop because of an identity number that no ration office had designed.

This is the last chapter of Part V, and the end of Volume Two, the volume of application. Each chapter of Part V found something that has to be settled before anyone needs it. A decade is a commitment that someone keeps, answers for and is ready to end. Machines move into the middle rather than empty it, and the trust every middle stands on is kept or spent, never owned. A check against the world is paid for by someone, or it is not done, and an honest signature says what it did not check. Slowness belongs where a choice cannot be undone, and it should be a check, not a rest. A law can make an institution name who answers, but it cannot do the answering in the room. Work that lives in one person’s hands becomes a business only by becoming a different promise. An investor can judge how people act when they are wrong, and he answers for the trust he lends. This chapter adds the exception, the seat between a rule’s design and its review, which every sector must keep someone for. Together they change what Part IV meant. Part IV found, in products, organisations and roles, that what a machine does not take over is someone who answers. Part V shows that this someone is not found at the end, like a remainder. The seat is designed beforehand and paid for by someone. The person in it may change; the institution’s duty to keep it filled, paid and able does not. No institution should automate its decisions faster than it can keep the capacity to understand, challenge and repair them. That is where Volume Two ends: not with a human residue, but with seats that someone must decide to fill.

## Ending: Two Lines She Could Not Defend

The committee’s draft guidance reached Sagar in the second week of October, through the district office. Her teaching stories had become a module called “Teaching Through Stories”. It had a rubric: five points for a story’s completeness, five for its link to the syllabus, five for its use of the platform. At the foot of the first page a line said that the method had been developed by Smt. Kavitha Sharma of Sagar.

She read all of it. Before the platform came, she had noticed that notes drafted by a machine were leaving under her name unread. She had made a rule then: nothing would leave with her name on it that she had not read and could defend. She could defend most of the module. She could not defend two lines. One gave full marks only for a complete story, when her best lessons had ended on a question. The other told teachers to use the same stories with every class.

She wrote to the committee through the district office. She said which parts she could stand behind and why. She named the two lines she could not, asked that they change or that her name come off them, and offered to explain at the next meeting. Suresh Tiwari read the letter in his office and told her, kindly, what she already knew. A module with a teacher’s name on it travels further than one without. The committee might keep the rubric whatever she wrote.

She sent the letter the next morning. If the lines stayed, her name would go. With it would go the one sentence that might have sent a teacher in another district to ask her what the stories were for.

### Practice

*The case the rule did not fit.* Take the most common decision in your work that a machine now makes, drafts or checks. Write the rule it follows in one sentence, and who chose what it is asked to achieve. Then find one real case from the last year that the rule did not fit. Write down who had the right to make the exception, and whether they knew they had it. Write who explained the exception to the person it affected, and who paid for it, in hours or in money. Where the answer is nobody, write who could hold the seat, and whether anyone has told them. Write who could stop the system, and how the work would go on if they did. Then write two short lists, in the manner of the sutta. One lists what you owe the people your work touches. The other lists what the institution you work for owes you, in time, safety and leave. This takes an hour; if the hour is your employer’s, note who gave it. It will not keep your job. It shows where your sector still needs someone, and whether anyone has been given the seat.

## Words to keep

- **A/B test:** showing two versions of a page to different visitors and keeping the one more of them act on.
- <strong>equity (</strong><strong><em>epieikeia</em></strong><strong>):</strong> Aristotle’s word for correcting a general law where it fails a particular case, by saying what the lawgiver would have said.
- **exception:** a decision that departs from a general rule for a case the rule did not foresee; in this chapter, it needs a named holder, a record, and a reason and a way back for the person it affects.
- **informal economy:** work done without a written contract, a registered business or social protection.
- **leaden rule:** the bending measure of the builders of Lesbos, which “adapts itself to the shape of the stone”; Aristotle’s image for equity.
- <strong><em>maṇḍī</em></strong><strong>:</strong> Hindi, a market yard where farmers’ crops are sold.
- **minimum support price:** the price at which India’s government offers to buy certain crops, announced before each season.
- <strong><em>Sigālovāda Sutta</em></strong><strong>:</strong> Pāli, the Buddha’s advice to Sigāla (*Dīgha Nikāya* 31), which turns the six directions into six relations with duties both ways.
- **sulphonylurea:** a family of diabetes tablets, gliclazide among them, that lower blood sugar whether or not a person has eaten, and so can take it too low.
- **teaching stories:** Kavitha’s name for stories about how someone came to know something, chosen for the moment a particular class has stopped understanding.

## Sources

1. *Sigālovāda Sutta* (*Dīgha Nikāya* 31; also called the *Siṅgāla Sutta*), trans. Nārada Thera, in *Everyman’s Ethics: Four Discourses by the Buddha*, The Wheel Publication No. 14 (Kandy: Buddhist Publication Society, 1985), Access to Insight.
2. J. C. Bose, *Response in the Living and Non-Living* (London: Longmans, Green, 1902); see Chapter 10.
3. “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; see Chapter 10.
4. Aristotle, *Nicomachean Ethics* V.10 (1137b), trans. W. D. Ross (Oxford: Clarendon Press, 1908), MIT Internet Classics Archive. The archive’s text reads “the error is in the law nor in the legislator”, evidently a slip for “not in the law”; the paraphrase above follows that sense.
5. Press Information Bureau, Government of India, release of 1 April 2026 on the launch of the curriculum on AI and computational thinking for Classes 3 to 8, “implemented from classes 3rd to 8th, in the session 2026-27”, by CBSE under the Department of School Education and Literacy (Ministry of Education, PIB2247963).
6. Calette Corcoran and Tibb F. Jacobs, “Metformin,” *StatPearls* (Treasure Island, FL: StatPearls Publishing; updated 17 August 2023), NCBI Bookshelf NBK518983, “Mechanism of Action”: “the drug is unlikely to cause hypoglycemia”.
7. NHS, “Side effects of gliclazide” (nhs.uk, accessed 23 October 2026): gliclazide “may cause low blood sugar”; “not skipping or delaying meals”.
8. Diabetes UK, “Stress and diabetes” (diabetes.org.uk, accessed 23 October 2026): stress hormones “make it harder for insulin to work properly”; “If stress doesn’t go away, it can keep your blood sugar levels high”; “coping with the loss of someone close” among the causes of stress listed.
9. Andrew M. Bean et al., “Reliability of LLMs as medical assistants for the general public: a randomized preregistered study,” *Nature Medicine* 32 (2026): 609–615, published 9 February 2026: models alone identified conditions in 94.9 per cent of cases; participants using them in fewer than 34.5 per cent, no better than the control group; the models tested were released in 2024; the trial tested 1,298 members of the public in the UK, not doctors, and doctors helped create its ten scenarios (Oxford Internet Institute, “Reliability of LLMs as medical assistants for the general public”, Synthetic Society project page). See Chapter 8. On suggested ratings: Thomas Dratsch et al., “Automation Bias in Mammography: The Impact of Artificial Intelligence BI-RADS Suggestions on Reader Performance,” *Radiology* 307, no. 4 (2023): e222176, published 2 May 2023: 27 radiologists rated 50 mammograms; where the category “purportedly suggested by AI” was wrong, very experienced readers rated 45.5 per cent correctly; the authors conclude that readers are “prone to automation bias” (Radiological Society of North America press release, 2 May 2023).
10. Karthik Muralidharan, Paul Niehaus and Sandip Sukhtankar, “Identity Verification Standards in Welfare Programs: Experimental Evidence from India,” *Review of Economics and Statistics* 107, no. 2 (2025): 372–392; NBER Working Paper 26744 (February 2020; revised September 2021), abstract; dates from the NBER conference version (8 August 2020), sections 2.1 and 4.4.
11. Karthik Muralidharan, Paul Niehaus and Sandip Sukhtankar, “Balancing corruption and exclusion: Incorporating India’s Aadhaar into public food distribution,” VoxDev, 17 April 2020 (the value of grain disbursed fell 18 per cent in control and 36 per cent in treatment areas, 78 and 66 per cent of the drop being reduced leakage; beneficiaries paid the costs of “having to make multiple trips to obtain grains”, and their costs of collecting it rose 17 per cent, or ₹7). For the 23 per cent of households without an Aadhaar number linked to their ration card: the value received fell by ₹49 (10.6 per cent), and the probability of receiving any benefit fell by 10 percentage points (NBER Working Paper 26744, introduction and section 4.3).
12. Tapasya, Kumar Sambhav and Divij Joshi, “How an algorithm denied food to thousands of poor in India’s Telangana,” Al Jazeera, 24 January 2024 (Samagra Vedika; the state cancelled more than 1.86 million existing food security cards, 2014 to 2019; “several thousands of these exclusions were done wrongfully”); the National Food Security Act, 2013 (Act No. 20 of 2013), section 15(1) (a District Grievance Redressal Officer “for each district”, for grievances “in matters relating to distribution of entitled foodgrains or meals”), section 15(6) (an appeal “before the State Commission”) and section 16(1) (legislative.gov.in); see Chapter 25.
13. Xiang Hui, Oren Reshef and Luofeng Zhou, “The Short-Term Effects of Generative Artificial Intelligence on Employment: Evidence from an Online Labor Market,” *Organization Science* 35, no. 6 (2024): 1977–1989: on Upwork, a 2 per cent fall in monthly jobs and a 5.2 per cent fall in monthly earnings for freelancers in the most affected occupations; see Chapter 6.
14. Microsoft, “Microsoft announces new Copilot Copyright Commitment for customers,” 7 September 2023: if a commercial customer is “challenged on copyright grounds, we will assume responsibility for the potential legal risks involved”; see Chapter 20.
15. Securities and Exchange Board of India (Intermediaries) (Amendment) Regulations, 2025, F. No. SEBI/LAD-NRO/GN/2025/226, dated 6 February 2025, *Gazette of India*, Extraordinary, No. 92, 10 February 2025 (in force on publication), inserting Chapter IIIB, “Usage of Artificial Intelligence”, regulation 16C(1)(b); under regulation 16C(2) the Board may, for a violation, “take such action as it may deem fit” (sebi.gov.in). Investment advisers are registered with SEBI under the SEBI (Investment Advisers) Regulations, 2013.
16. “Sow wheat by Nov 15 to prevent grain shrivelling: PAU to farmers,” *The Tribune*, 6 November 2022 (Punjab Agricultural University’s advice; a late-maturing paddy variety delays wheat sowing).
17. Robert Sanders, “AI revolutionizes weather prediction to help farmers in India,” Berkeley News, 16 September 2025 (forecasts by India’s Ministry of Agriculture and Farmers’ Welfare to “about 38 million farmers across 13 states in central and northeastern India”, by SMS, up to four weeks ahead; models from Google (NeuralGCM) and ECMWF (AIFS) blended with India Meteorological Department data; the stall of 2025 foreseen).
18. Ministry of Statistics and Programme Implementation, *Periodic Labour Force Survey Annual Report, July 2023–June 2024*: 46.1 per cent of workers (usual status) in agriculture and allied sectors, as stated in the Ministry of Labour and Employment’s answer to Rajya Sabha Unstarred Question No. 3760, 3 April 2025; see Chapter 5.
19. “Govt raises wheat MSP by 6.59% to Rs 2,585 per quintal for 2026-27,” *The Tribune*, 1 October 2025 (Cabinet decision on minimum support prices for the rabi marketing season 2026–27).
20. The Punjab Preservation of Subsoil Water Act, 2009: “No farmer shall sow nursery of paddy before 10th Day of May” (or another notified date) and “No farmer shall transplant paddy before such date” (UNEP LEAP, national legislation summary). For The glitch in 28.7 (added in this edition): Balwinder-Singh, Andrew J. McDonald, Amit K. Srivastava and Bruno Gerard, “Tradeoffs between Groundwater Conservation and Air Pollution from Agricultural Fires in Northwest India,” *Nature Sustainability* 2 (2019): 580–583. The “groundwater acts” it studies, Punjab’s among them, forbid transplanting before 10 June, a date later moved to 20 June. “Simulation studies suggest that groundwater depletion is significantly reduced through the groundwater acts”; burning “became the method of choice for accelerating the turn-around time between crops”; “an 8-d harvest delay”; burning concentrated “into a narrower window” with “a peak intensity that is 39% higher”; in New Delhi, “average daily PM2.5 concentrations in November were 29% higher after the groundwater acts”. The authors present these as first inferences, needing more air-pollution monitoring and modelling.
21. International Labour Organization, *World: Informality, Regional Statistical Profile* (Geneva: ILO, 2025; web page dated 21 February 2025): informal employment 57.8 per cent of total employment, world (ILO modelled estimates, series 2004–2024; the profile does not state the figure’s year).
22. The Street Vendors (Protection of Livelihood and Regulation of Street Vending) Act, 2014 (India; Act No. 7 of 2014), section 22(2)(d) and its first proviso: at least forty per cent of the members of each Town Vending Committee are to be representatives of street vendors, “elected by the street vendors themselves”, a third of them women (*Gazette of India*, Extraordinary, Part II, Section 1, No. 8, 5 March 2014; legislative.gov.in). Standing Committee on Urban Development (2020–21), Eighth Report, *Implementation of Street Vendors (Protection of Livelihood and Regulation of Street Vending) Act, 2014*, presented 6 August 2021: such committees “are yet to be constituted in all eligible towns and cities” (eparlib.nic.in).

*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 “AI makes intelligence cheap and consciousness valuable” in every sector, that “The twelve qualities apply everywhere”, and that the professionals “who thrive” will be those developing their consciousness; that education’s value has shifted to “consciousness development” and that its qualities “cannot be templated”; that Kavitha’s value lay in an “irreplaceable” narrative intelligence and her method was “verification-based pedagogy”, refined for “two years”; that a platform “could not understand why” a widow’s sugar rose, that “The fundamental limitation is meaning”, that “Trust is a consciousness phenomenon”, and that raising metformin in a woman eating less “risked pushing her glucose dangerously low”; that “The algorithm cannot exercise discretion”; that “Taste is the irreducible human quality in creative work”, that markets “can only ask for what they already know”, and that Distribution is “the twelfth quality”; that “The AI cannot make a promise” and that trust is “the stratum of value that AI cannot reach”; that judgement becomes “the entire competitive advantage”, that a dearer supplier was “an insurance policy, paid in loyalty, denominated in trust”, that “The AI could not make that distinction. Only Rajan could”, and that the choice was “the most conscious act of his career”; that a senior dyer on Rajan’s floor was “teaching apprentices”; that farming knowledge is “distributed consciousness” and “the empirical register of a consciousness calibrated by decades”, and that a farmer might “plant a week early because the monsoon feels different”; that Ama’s consciousness “is what every working human being exercises”, and that she “has known what she is worth”. It also sold “grilled” plantain chips, counted “roughly two billion” informal workers without a date, called the Goa rejection a “disaster”, gave Arjun “twelve years” at about thirty, said that Rajan “had resisted the implementation for months”, timed two conversations at “forty minutes”, ended on a summary of the Part and a bridge to the next chapter, and named Meera’s mother’s friend Padma, a doctor Venkatesh, a farmer Hardeep Singh, a seller Ama Boateng and a lawyer Carolina Martins.
- *What I now hold:* no sector keeps a human remainder that machines cannot reach, and no sector is proof against them. Each sector is a set of relations with duties both ways (the *Sigālovāda Sutta*’s move), and what recurs is a position: the one who may make the exception a rule did not foresee, and who must explain it and put right what it costs (Aristotle’s equity, with his question of who corrects the rule). Who holds it, and what it owes, is set by each sector’s law, money and history. Discretion is the leaden rule and also where favours live, so the exception needs a named holder without the motive to cheat, a record, and a reason and a way back for the person refused, written before the first refusal. A platform bends better than a module; what it lacks is the standing to depart from the plan and the duty to say why. Letting a machine bend leaves the seat with whoever allowed it. Metformin alone rarely causes dangerous lows; a sulphonylurea can. A test that chooses honestly by its measure leaves the measure as someone’s choice. In law, since February 2025, India’s securities regulator makes the regulated person “solely responsible” for an AI tool’s output; a promise is honest only about what is in the promiser’s power. Loyalty kept against the record is habit, and correcting it is a decision whose cost falls partly on people outside it. The farmer’s exception is his own; the sector’s larger ones belong to others. For informal workers the nearer risk, as of 2026, is exclusion. To “AI-proof” a sector means deciding, before the case arrives, who holds the seat and who pays for it; it keeps nobody’s job. The seat needs four things, which need not sit in one person: the right to depart, the knowledge to see when, the duty to explain and the means to repair. It sits between the seats at a rule’s design and at its review, and it lasts only while the competence to hold it is kept. A public system owes a reason, an independent review and repair. No institution should automate its decisions faster than it can keep the capacity to understand, challenge and repair them.
- *What changed my mind:* the Prologue’s rule against defining people by what machines cannot yet do; Aristotle’s own account of equity, which places the problem in the nature of the case, not in a lack of awareness; the Jharkhand experiments, which showed a machine right about theft and a rule wrong about people, and placed the cause in decisions about the transition; SEBI’s 2025 rule; the medical sources on metformin and gliclazide; the 2025 monsoon forecasts; the audit’s corrections (the crop, the plantain, the drug, the names); the print’s own story of the supplier, in which the machine made the distinction it was said to lack; and three outside readings of an earlier version of this chapter, in October 2026, which pointed out that the seat compressed four things into one person, that not every decision is an exception, that the holder must stay able to judge, that a reason is not a remedy, and that a duty written on paper can be a burden passed downward.
- *What survives:* the eight sectors, each with its subject; Kavitha’s invitation, “She did not have a framework. She had a practice.”, her pride and her dread, and the journey from Sagar to Bhopal; “The framework must bend to the sector, not the other way around”; the widow’s trend line and the doctor who waited; “A citizen whose benefits application is denied by an algorithm cannot switch governments” and accountability as “the obligation to explain, justify, and correct”; “Arjun’s design was better. The other design worked better.” and “He may have been wrong.”; “They arrive not ignorant but partially informed” and “partially informed in ways that may be subtly wrong”; “Here is what I would do, and here is why, and I will be here next month to answer for it.”; “habit dressed as judgement”; “Some mornings the system is right and he is wrong ...” and “A calculation optimises. A decision commits.”; “But this book owes Ama honesty, not romance.”; the customers who left “are not wrong”; exclusion as the risk.
- *Section titles in the print edition:* three kept withdrawn claims: 28.1 was “Education: From Content Delivery to Consciousness Cultivation”, 28.2 was “Healthcare: The Conscious Clinician”, and 28.7 was “Agriculture: Consciousness Rooted in the Earth”.

*Next in the series:* [*Volume III: Transformation (overview)*](https://nithinpalal.com/ai/ai-proof/volume-3-transformation)*.*