AI-Proof: Prologue. Two Minds in a Room
Next: Chapter 1. The Observer Behind Your Eyes
To the awakened, the awakening, and those who don’t yet know they already are.
यः पश्यति निजां दृष्टिंन यन्त्रेण स हीयते । बुद्धिः सुलभतां याताबोधो नित्यं सुदुर्लभः ॥
yaḥ paśyati nijāṃ dṛṣṭiṃna yantreṇa sa hīyate | buddhiḥ sulabhatāṃ yātābodho nityaṃ sudurlabhaḥ ||
The one who sees their own seeingis not diminished by the machine.Intelligence has become cheap.Awareness remains forever rare.
The machine may drive. The journey is still yours, and so is the choice of driver.
Imagine a room. There are two chairs, with a table between them. You sit in one. In the other sits a machine. As I write this, in 2026, it can solve olympiad mathematics at the level of a gold medal. It can turn a paragraph of instructions into working software. It can summarise a two-hundred-page contract before you have found your glasses.¹
Ask it what it is, and it will answer well. Ask whether it knows it is in the room, and it will give you a careful answer. Nobody alive can tell you for certain whether that answer describes anything.²
You know you are in the room. Tomorrow you will also know what was decided in it, and you will still be carrying it.
That second fact is the subject of this book.
When thinking became cheap
For most of history, if you wanted something thought through, you needed a trained human mind. A legal brief needed a lawyer. A diagnosis needed a doctor. A plan needed someone who had made plans before. Making that person took twenty years, a family’s savings and an institution willing to train them. Thinking was expensive because thinkers were rare. Within a single working life, that has changed, and the machine in the other chair is the reason.
When you are thirsty, water seems like the most important thing in the world. When water is everywhere, two things happen. First, you remember that you also needed food, shelter, company and a reason to get up in the morning. Second, you learn that plenty brings its own trouble: floods, dirty water, the need for a filter. Cheap intelligence does both. It shows what else we always needed. And it fills the world with answers that someone has to check. Someone has to say which water is safe to drink, and that someone can be asked. In the end, the filter is a person.
So this book asks one question. If a machine can do much of what you spent years learning, what is left that is yours?
The case against this book
Many classical Indian philosophers kept a rule for honest argument. Before you give your own view, you give the opposing view in full, in its own best words. They called it the pūrvapakṣa, the prior view. I will keep that rule in every chapter, so I begin with the case against the book itself. It has three parts.
The first part is a record of failure. In 1950 the mathematician Alan Turing collected the claims people made in his day about what machines would never do. His list begins: “Be kind, resourceful, beautiful, friendly, have initiative, have a sense of humour, tell right from wrong, make mistakes, fall in love, enjoy strawberries and cream …”³ Turing answered that such lists came from the machines people had already seen. For more than seventy-five years people have kept making these lists, and machines have kept crossing items off. In 2022, two groups of careful forecasters were asked whether a machine would reach gold-medal standard at the International Mathematical Olympiad by 2025. One group was made of experts, the other of trained forecasters. Both gave it less than a one-in-ten chance. It happened in July 2025.⁴
The first edition of this book made such a list too. It said a machine could not stop in the middle of an argument and think: wait, this is wrong. Machines released in 2024 and 2025 do exactly that, in so many words, and then go back and correct themselves.⁵ That claim was false before the book was printed.
There is a name for this mistake. Some religious writers once found God in whatever science could not yet explain. As science explained more, that God grew smaller. In 1894 the Scottish evangelist Henry Drummond warned believers against this habit: searching science for gaps, “gaps which they will fill up with God.”⁶ The habit later got a name, the “God of the gaps.” A book that finds the human in the gaps of the machine makes the same mistake. Each year the gaps close, and each year the human it describes grows smaller. So this book keeps one rule: it never defines a human being by what machines cannot yet do.
The second part of the case is older and stronger. It comes from Plato. In the Republic, he asks us to picture a ship. The sailors fight over who should take the helm, though none of them has studied navigation. They flatter the owner, take control of the ship, and feast on its supplies. The one person who has studied the seasons, the stars and the winds is mocked as a useless stargazer. Plato’s point is simple. A ship should be steered by the one who knows how to steer, and a city should be ruled by those who know what is good.⁷
Now put the machine on that ship. If it becomes the best navigator there has ever been, Plato’s argument says it should steer. That means the ship, and also the loan, the diagnosis, the prison sentence and the plan. If knowing best gives the right to decide, the machine will soon have that right. This is the strongest case against everything this book says. I will not pretend it is weak.
The third part concerns the machine’s inner life. Some serious philosophers think a machine could one day be conscious, and nobody can yet prove that none will be.² A book that rests on the claim that machines have no inner life rests on something nobody knows.
The rider
My reply begins with a chariot, and with the difference between two of them.
Plato described the soul as a chariot too. In the Phaedrus, it is a charioteer driving two winged horses, one noble and one unruly.⁸ Readers have long understood the charioteer as reason. In Plato’s picture the best part of you is the driver, and there is no one else in the chariot. That is the picture the machine threatens. If your worth is your driving, a better driver replaces you.
An Indian text more than two thousand years old, the Kaṭha Upaniṣad, draws the same chariot with one more figure in it.
“Know the self as a rider in a chariot, and the body, as simply the chariot. Know the intellect as the charioteer, and the mind, as simply the reins.”⁹
Look at who drives. The Upaniṣad gives the driving to the intellect, buddhi, long before anyone built a machine. The self is not the driver. It rides. The next verse names the one who makes the journey: the self joined to body, senses and mind is the bhoktṛ, a word from a verb meaning to eat, to enjoy and to undergo.⁹ The great commentator Śaṅkara insists on the difference: “the pure atman cannot be the enjoyer.”¹⁰ It is the self in the journey, with a body, a history and a mind, that goes through it. That is the figure this book is about: the person to whom the journey happens.
Plato’s chariot has no rider. Reason drives, and reason is the highest thing in it, so a better driver would be a better you. The Kaṭha’s chariot has a rider, and the rider was never the driver. The machine may become the finest charioteer there has ever been. It will still not be the one the journey is for.
But the Upaniṣad does not let the rider relax. The verses that follow are a warning. “He who has no understanding and whose mind (the reins) is never firmly held, his senses (horses) are unmanageable, like vicious horses of a charioteer.” Such a person never reaches the end of the road. “But he who has understanding for his charioteer, and who holds the reins of the mind, he reaches the end of his journey.”¹¹ Arrival depends on the driver. The rider is not safe because someone else is driving. The rider arrives, or does not.
In the Upaniṣad, the charioteer is your own understanding. A machine is a charioteer from outside. That makes the warning sharper. Your own understanding must still choose the driver, watch the road, and take back the reins when the chariot goes wrong.
Knowing best and having the right
Here someone will object. Grant that the machine is not the rider. Why should the rider not simply hand it the reins? If it drives better, let it drive.
Often, the rider should. A doctor relies on a scan that a machine has read. A pilot lets the autopilot fly for hours. Handing over the reins is delegation, and a sensible person delegates all day. What a rider cannot do is hand over the journey. That is abdication: letting “the machine decided” become the end of the answer. To delegate, you choose the machine, you watch it, and you keep the power to take the reins back. To abdicate is to keep none of these.
Look again at what Plato asked of his rulers. They had to know what is good, not only how to steer. A machine may become the best at routes: the fastest, the cheapest, the most likely to work. Choosing where a life should go is a different kind of question. It asks what counts as better, and the people who live the life have a voice in that answer which no finder of routes can replace.
Plato’s own student, Aristotle, adds a second point. In the Politics he notes that some things are judged best by the people who use them, not the people who make them. The one who lives in a house judges it better than the builder. “The guest will judge better of a feast than the cook.”¹² Aristotle does not settle the whole matter. The guest does not cook the feast, and he cannot be blamed for it. But he has standing: the feast is for him, and he may judge it. The navigator may steer best. The voyage still belongs to the people on board, and the navigator answers to them.
So the first distinction of this book is simple, and it carries everything else. Knowing what is best is not the same as having the right to decide it.
The third chair
So put a third chair in the room. In it sits the person the decision is about: a patient, a client, a citizen, a child.
Suppose a machine approves a loan, and the loan goes bad. A family loses its house. The family does not go to the machine. They come to a desk, and a person sits behind it.
What does the family need from that person? Not only a reason. “The score was 0.73, and the policy said approve” is an explanation. It is not yet an answer. To answer for a decision is to be the one who must explain it, undo it where it can still be undone, and put right what can be put right. An explanation next month does not give a family back its house. This is what has become scarce. Intelligence is getting cheaper every month. What stays scarce is a person who answers for a decision in this full sense. I will call that person answerable.
The desk can also disappear. Behind it stands an institution: a bank, a hospital, a ministry. When a decision passes through many hands and many machines, it can end with no name on it at all. The political philosopher Dennis Thompson called this “the problem of many hands.”¹³ A person who answers is scarce partly because institutions can be designed without one. Someone has to refuse that design.
Some people in the third chair cannot ask at all. A child cannot. A patient who is unconscious cannot. The people not yet born cannot. For them, someone must answer on their behalf, the way a parent or a trustee does. Answerability is not only for adults signing their own loans.
Can a machine answer for a decision? It can explain one, often very well. But what does it carry? What a machine can lose was placed there by the people who built it. In tests, some systems have even acted to avoid being switched off or replaced.¹⁴ Yet whatever a machine can lose can be copied, restored or changed by its makers. Nobody can restore your life from a copy. That is a fact about today, and it is also part of what we mean by a life.
The danger was never only the machine. In The Philosophy of Everything I described intelligence as something that “acts and reacts algorithmically to distinct parameters without being conscious.” In the same book I named the habit that makes such intelligence dangerous: “Believing that someone remotely controls human destinies makes us outsource our crucial decisions.”¹⁵ I was writing about fate and gods. The sentence now has a new subject. The machine does not have to take our decisions from us. We are ready to give them away. The real risk is that we mistake intelligence for authority, and delegation for abdication.
Consciousness, carefully
I will use the word consciousness in this book, and carefully. I mean the awareness that you are aware: experience, and knowing it as yours. I do not mean the power to check your own thinking. Machines now do that, as the case against this book showed, and it can be built. Consciousness, in this book, is how a person notices, before handing a decision to anyone, that the decision is still theirs.
The verse at the head of this book speaks of “the one who sees their own seeing.” That is the witness. The Kaṭha’s bhoktṛis the one who goes through the journey. They are two sides of one person. The witness notices that the decision is yours. The bhoktṛ is the one who will live with it, and so the one who answers.
Whether a machine is, or could become, conscious is an open question, and serious people disagree.² Nothing I argue depends on the answer. This book does not need to prove that machines are unconscious. Even if a future machine has an inner life, it will not be you. It will not bring your history to the decision. It will not wake tomorrow to the results of what you chose.
Consciousness also does not make people good at deciding. Conscious people make lazy decisions. They protect their position, believe the number that suits them, and grow careless late in the day. A good part of this book is about the ways people fail the machine, and not only the ways the machine fails people.
What this book claims
A good watch does not promise that water can never reach it. It tells you a depth: water-resistant to fifty metres. The promise is honest because it is exact. This book tries to make its promises the same way.
AI-proof does not mean out of the machine’s reach. It means the part of your work, and of your life, that stays yours when the machine can do the rest: the decisions you sign, the answers you owe, and your voice in what counts as better.
Four questions run through every chapter, and they should be kept apart.
What will work? That is a question for intelligence, and the machine is often best at it.
What should we do? That is judgement. It needs a view of what counts as better.
Who may decide? That is authority. Knowing best does not settle it.
Who answers? That is the person who must explain, undo and repair, and who carries what follows.
I say “never” about three things only.
A machine will never be you. It will not live your life after the decision: your body, your name, your people, your future.
A machine’s makers and users will never be free of what it does. Children are also made by others, but a child grows into someone who answers for themselves, and the parents step back. This book holds that a machine’s makers and users never step back. However many machines stand between, the people who make it answer for what it is, and the people who use it answer for what they use it to do.
Being intelligent will never give a machine the right to decide for people. Knowing best is not the same as having the right.
The first is true by what the words mean. The second and third are commitments, not predictions. A world could abandon them; this book asks you not to. The commitments do not change, but the way they are kept will. Indian ethics has a phrase for this: deśa, kāla, pātra, place, time and person.¹⁶ What it takes to keep a decision answerable in a village bank in 2026 will not be what it takes in a city hospital in 2040.
Everything else I say about machines is “for now.” When I say what a machine can or cannot do, I give the year, because the year will change the answer. When I say what a machine cannot be, I give no year, because no year will change it.
How this book is built
The book has three volumes. The first, Awakening, asks what you are and what the machine is. The second, Application, carries those answers into products, organisations, strategy and whole sectors of work. The third, Transformation, is about living through the change: fear, identity, practice, and what we hand to the people who come after us.
Each chapter is built the way many classical Indian thinkers built an argument. They called a topic examined in this way an adhikaraṇa. It has five steps: the subject, the doubt, the opposing view, the reply, and the conclusion.¹⁷ Each chapter opens with one sentence that holds its conclusion. In the old tradition a short sentence that carries a whole teaching is called a sūtra, a thread. Then comes a story, which is the subject. Then the doubt, stated plainly. Then the case against me, as strong as I can make it. Often it is the machine’s case, and sometimes the machine wins it. Then my reply, my conclusion, and what I still cannot settle.
The people in the stories are composites: each is made from several real lives. They are there to show how the machine works and how people work with it. The cases and figures they meet are real, and each one is sourced.
This is the second edition of the book, published as a series of articles. Where I have changed my mind since the first edition, the chapter says so at its end: what I claimed, what I now hold, and what changed my mind.
Refusing this technology would be as useful as refusing electricity, and this book does not ask you to. It asks how to use the most capable tool anyone has built without giving it what does not belong to it: your decisions about your own life, and the answers you owe to other people.
The room again
Now the room again. Two chairs, and the third.
The machine is ready. Ask it what you have that it does not, and it will give you a thoughtful answer. It can ask the question as well as you can.
Only one of you has to live with the answer. And when the person in the third chair asks why, they will be asking you.
Words to keep
abdication: handing over a decision and with it your answer, so that “the machine decided” ends the matter.
AI-proof: the part of your work, and of your life, that stays yours when the machine can do the rest.
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.
answerable: bound to answer for a decision in that full sense.
authority: the right to decide. Knowing best does not give it.
consciousness: the awareness that you are aware: experience, and knowing it as yours.
delegation: letting the machine do the work while you stay answerable: you choose it, watch it, and can take the reins back.
intelligence: the power to solve problems and produce answers. A machine can have a great deal of it.
witness: the one who sees their own seeing; how you notice that a decision is still yours.
adhikaraṇa: a topic examined in five steps: subject, doubt, opposing view, reply, conclusion.
bhoktṛ: the one who goes through the journey; in the Kaṭha Upaniṣad, the self joined to body, senses and mind.
buddhi: the intellect; in the Kaṭha Upaniṣad, the charioteer.
deśa, kāla, pātra: place, time and person; what changes in how a commitment is kept.
pūrvapakṣa: the prior view; the opposing case, stated in full before it is answered.
sūtra: a thread; a short sentence that carries a whole teaching.
Sources
Google DeepMind, “Advanced version of Gemini with Deep Think officially achieves gold-medal standard at the International Mathematical Olympiad,” 21 July 2025. OpenAI announced “gold medal-level performance” from an experimental model on 19 July 2025, graded by former medallists it engaged rather than certified by the IMO.
Patrick Butlin, Robert Long et al., “Consciousness in Artificial Intelligence: Insights from the Science of Consciousness,” arXiv:2308.08708, 2023; David J. Chalmers, “Could a Large Language Model Be Conscious?”, arXiv:2303.07103, 2023, also in Boston Review, 9 August 2023.
A. M. Turing, “Computing Machinery and Intelligence,” Mind 59, no. 236 (1950): 433–460, §6.5, “Arguments from Various Disabilities.” Turing’s page cross-references inside the list are omitted; the list continues. His reply: such claims “are mostly founded on the principle of scientific induction.”
Forecasting Research Institute, report on the accuracy of near-term forecasts from the Existential Risk Persuasion Tournament, 2 September 2025: superforecasters gave 2.3 per cent and domain experts 8.6 per cent to an IMO gold by 2025.
OpenAI, “Learning to Reason with LLMs,” 12 September 2024; DeepSeek-AI, “DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning,” arXiv:2501.12948, January 2025, published as “DeepSeek-R1 incentivizes reasoning in LLMs through reinforcement learning,” Nature 645 (2025): 633–638, 17 September 2025.
Henry Drummond, The Lowell Lectures on the Ascent of Man (London: Hodder and Stoughton, 1894), p. 426: “There are reverent minds who ceaselessly scan the fields of Nature and the books of Science in search of gaps … gaps which they will fill up with God.” The phrase “God of the gaps” is later; Drummond does not use it.
Plato, Republic VI, 488a–489a (the ship), and V, 473c–d (philosophers as rulers), paraphrased.
Plato, Phaedrus 246a–b and 253c–254e (the charioteer and the two horses), paraphrased.
Kaṭha Upaniṣad 3.3–4 (1.3.3–4 in the older numbering), trans. Patrick Olivelle, Upaniṣads (Oxford: Oxford University Press, 1996), pp. 238–239. Verse 3.3 is quoted; verse 3.4 is paraphrased.
Śaṅkara, commentary on Kaṭha Upaniṣad 1.3.4, trans. S. Sītārāma Śāstrī: “for, the pure atman cannot be the enjoyer.” [verify edition and year]
Kaṭha Upaniṣad 3.5, 3.7 and 3.9 (1.3.5, 1.3.7 and 1.3.9), trans. F. Max Müller, The Upanishads, Part II, Sacred Books of the East 15 (Oxford: Clarendon Press, 1884).
Aristotle, Politics III.11, 1282a, trans. Benjamin Jowett (1885).
Dennis F. Thompson, “Moral Responsibility of Public Officials: The Problem of Many Hands,” American Political Science Review 74, no. 4 (1980): 905–916.
Anthropic, “Agentic Misalignment: How LLMs Could Be Insider Threats,” 20 June 2025; Jeremy Schlatter, Benjamin Weinstein-Raun and Jeffrey Ladish (Palisade Research), “Shutdown Resistance in Large Language Models,” arXiv:2509.14260v1, 13 September 2025 (retitled in v2, January 2026). Both were invented test settings.
Nithin Palal, The Philosophy of Everything (2022), pp. 114 and 212.
The triad appears in the Bhagavad Gītā 17.20, where a gift is rightly given deśe kāle ca pātre ca, “at the right place, at the right time, and to the right person”; later ethical writing uses deśa-kāla-pātra for what must be weighed when a rule is applied.
A traditional verse lists the five members of an adhikaraṇa: viṣayo viśayaś caiva pūrvapakṣas tathottaram | nirṇayaś ceti pañcāṅgaṃ śāstre ’dhikaraṇaṃ smṛtam (“subject, doubt, prior view, reply and conclusion: these five limbs are known in the śāstra as an adhikaraṇa”; translation the author’s). Its source is uncertain. Mīmāṃsā counts saṅgati, the connection to what comes before, as the fifth member; Vedānta counts nirṇaya, the conclusion.
Revised from the print edition (2026)
What the print edition claimed: that the machine lacks judgement, taste and trust “not because a future version will finally crack the code” but because it is not conscious; that “every quality that makes a human irreplaceable… flows from this single source”; that the machine “cannot pause mid-sentence and think: wait”; that it “is all engine and no driver.”
What I now hold: whether machines are or could be conscious is open, and the argument does not depend on it. “Never” is said of three things only: a machine will never be you; its makers and users will never be free of what it does; intelligence never gives it the right to decide for people. Everything else is “for now,” and dated. In the Kaṭha’s image the intellect was always the charioteer; the self is the rider, and the rider answers for the driver it allows. Knowing best is not the same as having the right to decide. Delegation is allowed; abdication is not.
What changed my mind: the print edition’s own later sections (1.3, 3.9, 40.3) conceded the question; reasoning systems released in 2024 and 2025 stop, check and correct themselves; the Kaṭha, read to the end of its warning, set beside Plato’s chariot.
What survives: the room, the water, and the difference between producing an answer and answering for it.
Comments