AI-Proof: What Machines Can Never Be — Epilogue
Where we are: the end. After Chapter 40, the book returns to the room from the Prologue, and to the choice that knowing more never by itself gives anyone the right to make for you.
Late in the Gītā, the one who knows most says: ponder it thoroughly, then act as you like. A machine can now set out the reasons on every side. Saying “I answer for this” then is not pride. It is a reply, for your part, to those owed one.
In Brief For readers who wonder whether a choice is still theirs once a machine can weigh every reason for them. The Epilogue returns to the room where the book began. A designer in Mumbai finds that a machine has copied the silences of his own old work, and sends his client both versions with the case against his own. An operations head in Hyderabad tells her managing director that a check can be automated and should not be. A verse near the end of the Bhagavad Gītā, which hands the choice back to Arjuna, and Sartre’s pupil, who was told to choose, help show what stays yours when every reason can be weighed for you. Choosing, the right to decide, being affected, explaining and repairing can sit with different people, and “I answer for this” need not be pride.
Knowing more never by itself gives anyone the right to make your choice; “I answer for this” means replying for your part, not claiming you acted alone.
Opening: Two Minds in a Room
Two minds in a room.
You have been here before. The Prologue set two chairs at a table, a machine in one and you in the other. Then it added a third chair, for the person the decision is about: a patient, a client, a citizen, a child. The room has not moved. The machine in the other chair has.
Since the print edition of this book came out, earlier in 2026, it has changed again. In July 2026 the Forecasting Research Institute reported that several machines had “likely reached parity” with the best human forecasters on its test1. Good Judgment, the company that grew out of the research that first identified superforecasters, disputed the comparison eight days later2. It said that improving machines had been measured against a human score frozen in 2024. METR, a group that measures how long a task machines can finish on their own, counts it in a skilled person’s hours. In May 2026 it warned that its measurements above sixteen hours were unreliable3. And the machine no longer waits in its chair. It carries out long tasks while you do something else. In 2025 one began to complete purchases for people inside a conversation, in a limited launch in the United States that its maker moved and scaled back the next March4.
The print edition ended by telling you that you had changed. A book cannot know that. It can only bring you back to the room.
So this ending asks one question. If a machine can weigh every reason better than you can, in what sense is the choice still yours, and is it only pride to say “I answer for this”?
The “if” is a supposition, not a finding. No test has shown that machines weigh every reason better than people, and the forecasting result above is one test, and disputed. This ending grants the supposition anyway. An argument that holds only while the machine is worse would not last.
The print edition asked a question of its own instead: “now that the machine handles intelligence, what will you do with your consciousness?” It closed: “You are aware. That is the difference. That has always been the difference.” This edition cannot end there. Whether a machine is aware of anything is an open question, and nothing in this book depends on the answer. The difference the book can defend is narrower. It is the first of the three things the Prologue said “never” about: “A machine will never be you. It will not live your life after the decision: your body, your name, your people, your future.” The choice is handed to the one whose life it is, once the reasons are in. One line of the old ending stays as it was: “The choice is not between them.”
The title can sound like one choice, made once, between you and the machine. The choice in these pages is smaller, and it comes more often: the one handed back at the end of counsel, once the reasons are set out. Where a decision touches other lives as well, who may decide it is a further question, and whoever decides answers to the people it touches.
Making a choice, having the right to make it, being touched by it, owing an explanation for it and being able to repair it can belong to different people. The three stories that follow show them apart. A designer’s client runs a hotel chain, a business with owners, staff, guests and a freelancer whose work depends on it. A managing director’s budget is bounded by the law on medicines, by a quality unit that must stay independent, and by patients he will never meet. A warrior’s choice touches his kin, his teachers and the soldiers on both sides. Sometimes the right to decide is given to another, as to a doctor for an unconscious patient or a parent for a small child. That right comes from the gift, not from knowing more. Deciding is still not consenting. The person whose life a decision touches must not be erased from it, and the decider’s responsibility must not be mistaken for that person’s consent.
The Leaving-Out
Arjun Mehta is a brand and packaging designer. He works from two rooms above a printing shop in Parel, in Mumbai, and he uses machines every working day. Since May a freelancer, Neha, has done his smaller work for a hotel chain in Goa, his client since 2023. Since June she has been paid a monthly fee, and she has a desk in the second room.
The chain has four small hotels and plans more. In October 2024 its operations director, Vandana, rejected the first website he designed for it. “This is beautiful,” she said. “It will not work for us.” She renewed him for a year in April 2025, and in April 2026 for six months only, after asking whose name would go beside hers. The six months end this month. Early in October 2026 he is designing the identity for the chain’s next hotel.
He gives a machine his brief, the booking figures Vandana sends him each quarter, and the chain’s own identity, which he designed in 2023. She has agreed in writing that her figures may go to the machine he uses. In a few seconds it returns a complete identity for the new hotel. It is good. It has used the figures: its sample pages put the room sizes and the distance to the beach first, where families look for them. And it extends his own work faithfully, because his brief asked it to keep the brand: soft letters, pale colours, much left unsaid. Beneath the design the machine has written a short note. It calls the design “elegant”.
That evening he writes a sentence in his notebook: “The machine left out what it did not have, and called the leaving-out elegance.”
Half of the sentence is literal. The note did call it elegant. The other half needs care. Subtracting has become cheap: ask a machine for a design with one image and one colour, and it will make one, and make it well. Restraint is something else. It is the decision about what a brand will not say, taken by someone who will be blamed if the silence fails. The question is what the machine “did not have”. If the sentence means that machines lack something by their nature, it claims what nobody can prove, and nothing here needs it. If it means what nobody had given the machine, it is true, and it turns back on him.
The silences the machine copied were his. In 2023 he had designed the chain for couples from Europe looking for quiet. In 2024 Vandana told him that the guests had changed: more Indian families, more bookings through the sites that compare prices. He did not hear it then. The machine had learned the chain’s restraint from his own work and kept it without its reason, because the reason had been his and he had never written it down. Part of what it left out, he had left out first.
The machine was faithful to the old identity, and the new figures changed only what its pages put first. Fidelity to what a system is given is not relevance to what has changed. A system can reproduce an old assumption faithfully, and its fluency makes the old assumption look like a fresh decision. Every brief, every set of data and every identity leaves things out. What matters is who decided what to leave out, for what purpose, and who notices when the purpose has changed.
He spends three days on a version of his own. It keeps less of his old restraint. The letters are heavier, so that a grandmother can read the hotel’s name on a phone in the sun, and there is a second, warmer colour. The difference from the machine’s version is small. Most people would not see it.
The print edition made a virtue of that. It said the difference would be invisible to anyone who had not spent years learning to see it, and “The difference is everything.” That is how taste becomes gatekeeping. A difference that only the designer can see is his preference, however good, and a client should not pay for it on faith. The test is whether he can say what the difference does for the people who will live with it. He can say part of it: the heavier letters are for older relatives booking on their phones. He cannot fully say why the second colour is right. Not everything he cannot yet say is mere preference. Some of what a designer knows comes before the words for it. But it reaches a client honestly only when he sorts it: what he knows, what he suspects, and what he cannot yet show. He knows what the heavier letters are for. He suspects the colour is right. He cannot yet show that either will bring in one more family. The fee for the new identity was agreed as one sum, and the three days are not in it. They are his.
Since his second design for Vandana he has written the case against his own work, in the voices of the people who will live with it. This time he asks the machine to write it too. It gives three objections, and they are good. The third he cannot answer. Guests who know the chain’s four hotels will recognise the machine’s version at once, and his looks like the start of another chain. Families who book on price may not notice either. A year earlier a machine-made design beat his on a client’s own measures, and he lost that project. Being sure did not make him right then.
The objections came from his brief and his question, so they are not independent of him. Neha had no part in the 2023 identity; she has worked for him only since May. She knows how he works, but she made neither version. He shows her both designs without saying which is his, and asks her to read them cold. She says something neither he nor the machine had said. Most families will first meet the new hotel as a small picture on a site that compares prices, beside thirty others. At that size she cannot tell the two designs apart. He has no reply for her yet.
Ponder Over It Thoroughly
An old book ends at the same kind of table.
The Bhagavad Gītā, the Song of the Lord, is set in a chariot between two armies. The warrior Arjuna rides in it. His charioteer is Kṛṣṇa, who in the poem is God. In Kashinath Trimbak Telang’s translation5, Arjuna gives the order himself: “station my chariot between the two armies”. He looks at the teachers and kinsmen he would have to kill. In Telang’s words he said “I shall not engage in battle”, and “verily remained silent”. An older chariot, in the Kaṭha Upaniṣad, gave this book its picture of a person. There the intellect is the charioteer, the sārathi, and the self is the rathin, the lord of the chariot. This book read the rathin as the one who can refuse the journey. Arjuna has just used that power.
The rest of the poem is teaching. Near the end, Kṛṣṇa makes the strongest case against Arjuna’s freedom: “If entertaining egotism, you think that you may not fight, vain, indeed, is that resolution of yours. Nature will constrain you.” The Sanskrit throws Arjuna’s own words back at him (na yotsye, the words of his refusal) and names their source, ahaṃkāra: the “I-maker”, the sense that “I am the doer”. Telang has “egotism”, and so does a modern scholarly translation by Swami Gambhirananda. Two classical commentators read the word differently. Śaṅkara, explaining egotism earlier in the chapter, calls it the notion “I am the agent”. Rāmānuja, in Swami Adidevananda’s translation, reads it here as “self-conceit”, “a false sense of independence that you know what is good for you and what is not”. Telang’s note gives Śaṅkara’s reading of “nature” as the nature of a kṣatriya, a warrior by birth. That reading ties a person’s duty to the caste he was born into, and this book refuses it. Two verses later the Lord is “turning round all beings (as though) mounted on a machine”. The word is yantra. Telang says “machine”; other translators say “wheel”. The verse predicts nothing about the machines in this book; its picture is of the Lord and all beings.
Then comes the verse this book has kept for its last pages:
“Thus have I declared to you the knowledge more mysterious than any mystery. Ponder over it thoroughly, and then act as you like.”
iti te jñānam ākhyātaṃ guhyād guhyataraṃ mayā | vimṛśyaitad aśeṣeṇa yathecchasi tathā kuru || (18.63)
The one in the poem who knows most has just said that every being is moved as if on a machine. Still he does not choose for Arjuna. He tells him to think it all through, and he hands the choice back. Its verbs of wishing and doing are spoken to “you”: icchasi, you wish; kuru, do. Swami Gambhirananda’s translation keeps the second person: “Pondering over this as a whole, do as you like.”
The Gītā is not a modern book about freedom, and this verse is not its last word. Kṛṣṇa speaks again at once, and three verses later he says: “Forsaking all duties, come to me as (your) sole refuge.” And Arjuna’s last words take the counsel: “Destroyed is my delusion; by your favour, O undegraded one! I (now) recollect myself. I stand freed from doubts. I will do your bidding.” The choice that was handed back is used to accept the advice. That is still a choice. Advice he could have refused, and took knowing what it was, becomes his own act. The same holds when the advice is a machine’s. Being influenced does not abolish a choice, but influence comes in kinds that are not alike. Being persuaded by reasons differs from being steered by how the options are shown. Both differ from being pressed by someone with power over you, or forced by circumstance. Handing a decision over on terms you agreed differs again from giving up your judgement. The poem is God persuading a man to fight a war against his own family. I borrow neither the war nor its theology, and nothing here likens a machine to Kṛṣṇa. Nor does this book make Kṛṣṇa a champion of modern freedom. I borrow only the order of the ending: the counsel given in full, the choice handed back, and a man who says “I will”. What the order shows is one feature: after the whole teaching, Arjuna is still spoken to as someone who must think it through and act.
That order has become ordinary. In my experience, a machine in 2026 can set out the reasons on every side of a decision when asked, more fully than most advisers have time to. “Ponder over it thoroughly” is the part it now does well. It will also go on adding reasons for as long as you keep asking. The time to decide is set by the people who need the decision. Then it is made on the reasons already in. When a machine acts for you, as some now do, the choice came back to you earlier: when you let it act, and on what terms.
A choice is not handed back just because a screen offers a button marked “reject”. If refusing would cost a person her livelihood, or there is nothing else to choose, the choice came back in form only. Chapter 36 looked at what makes speaking up safe in practice, and refusing needs the same.
A French philosopher saw the other side of the same order. In a lecture in Paris in October 1945, Jean-Paul Sartre told the story of a pupil6. The young man’s elder brother “had been killed in the German offensive of 1940”. He had “the choice between going to England to join the Free French Forces or of staying near his mother and helping her to live.” He came to Sartre for advice. “Similarly, in coming to me, he knew what advice I should give him, and I had but one reply to make. You are free, therefore choose, that is to say, invent.” Choosing whom to ask, Sartre said, is already part of the choice: “to choose an adviser is nevertheless to commit oneself by that choice.”
The two teachers face the same kind of choice, a man deciding whether to go to war, and they part twice. Kṛṣṇa gives the whole teaching before he hands back the choice. Sartre thinks no teaching could settle it: “No rule of general morality can show you what you ought to do”. And where the Gītā says that nature will constrain you, Sartre says: “We are left alone, without excuse.”
The machine, as I read it, sharpens both. It can give the whole teaching, more patiently than any adviser most people could afford. But you choose which machine to ask, what to tell it and what to leave out, how to put the question, and whether to ask it to disagree. Each of these is already a commitment, in Sartre’s sense. And machines tend to agree with the person asking7, a pull that researchers call sycophancy. Sartre’s pupil knew what his teacher would say. Often, so do we. Arjun’s habit is one partial remedy. He writes the case against his own work, and this time he asks the machine for it too: the side he would rather not hear.
Neither teacher wins, and they are not simple opposites. Read alone, Sartre’s chooser carries everything, without excuse, which is the “I alone did this” that this book has refused. But Sartre did not think a person chooses from nowhere. In the same lecture he names limits that do not change: “What never vary are the necessities of being in the world, of having to labor and to die there.” Elsewhere he calls the given part of a situation its facticity. What he refused was its use as an excuse. The Gītā’s machine, read by itself, becomes such an excuse: it was nature, it was the system. Where the two meet can be said once. We cannot claim to be the authors of every condition we act in, and we cannot use those conditions to erase the part that was ours. In the online rewrite of my book, The Philosophy of Everything: On the Pursuit of Happiness, published as essays on nithinpalal.com in October 2026, I described that place like this: “You are not the author of the match. You are not a passenger either. You are in it, with a set of controls that are smaller than you wish and larger than you fear.”8
You Can. You Should Not.
Meera Reddy is vice-president of operations at the Hyderabad site of NovaChem Pharmaceuticals, a drug company with its head office in Basel. She has worked for the company for fifteen years. The site is in Genome Valley. A system there reviews the batch records, which show how each batch of medicine was made. Early in October Meera named its operator: Madhavi Krishnamurthy, the quality manager who signs for those batches. She also asked that one record in twenty be read in full by hand before anyone opens the system’s reading. The hours come from the time the system saves.
In the middle of October, at the site’s budget meeting for the next year, the managing director for India asks whether the hand reading can be automated. Last year the hand re-checks matched the system every time.
“You can,” she says. “You should not.”
Her answer keeps apart two questions the meeting had joined. Whether the reading can be automated is a question about what will work. She says plainly that it can: on the evidence so far, the system applies its rules as well as the people do. Whether it should be is another question, and her reason is about the person, not the records. The operator’s name is on every batch, and her right to stop one is only as good as her eye. A signature can outlast the practice that made someone able to give it. The hand reading keeps the person who signs able to sign.
She says what she cannot prove: that any batch will ever need it. She also says what the reading cannot do. When a hand and a system agree, both may still be wrong in the same way. One record in twenty will seldom meet a failure that comes once in a thousand batches. And a reading done every month can turn into a ritual, done to agree. That same week she had written to the site’s governance team that she would sign fewer of its reviews, and read in full every one she signed.
Her strongest claim is about the company, not about reading. It must not keep a person’s signature while it removes the skill, the independence and the authority that give the signature its meaning. The hand reading could one day give way to another reliable check that does not lean on the system. It should not give way to nothing. She is not defending people against machines, or praising work by hand, which is slow and tiring and on most records finds nothing the system missed. She is defending the reading behind a signature. And she says whose decision it is. The budget is his, with Basel’s approval. If the reading stays, its hours are hers to explain, month by month, in the report that goes to Basel.
He says he will think about it.
Under the hood
For each product the site keeps a master record, which says what each step should be, and specifications, which set the limits the results must stay within. The system reads each batch record against both, and flags what is missing, what falls outside a limit and every departure from the written procedure. It does not release a batch.
The operator’s signature comes after the system’s reading. Under the site’s approved procedure, it says that she has reviewed the record and the system’s flags, and that every deviation has been explained and closed. It is a step before release, not the release itself, which another named person signs. If she will not sign, the batch waits.
The hand reading has a second use besides keeping the signer able to sign. Because the two readings are compared each month, it checks the system against one that did not begin from the system’s flags.
I Will
The question this ending began with has a second half. Is it pride to say “I answer for this”?
The Gītā names the pride. The one deluded by ahaṃkāra, it says, thinks “I am the doer” of what nature does. This book met half of that charge earlier, with the same poem’s five causes of every act. The poem lists them as the seat of action, the agent, the organs, the many movements and, fifth, the deities. They deny that the one who says “I did this” is the act’s only author. They do not remove the agent, who is one of the five. To read them as a map of how responsibility is shared in a company is an analogy, and I mark it as one. To answer for a decision is to answer for your part in it, not to claim you caused it all. Three things are easy to run together here. Doership is the belief that I alone caused an act. Agency is the power to form an intention and act on it. Answerability is the duty to explain my part and repair what I can. The five causes loosen the first and keep the second; the third is this ending’s question. The other half lies in two sentences of Arjuna’s: his refusal and his last reply.
Both are in the first person. “I shall not engage in battle.” “I will do your bidding.” What follows is my reading, and the tradition has others, Rāmānuja’s among them. Kṛṣṇa calls the first egotism. On my reading, it is the resolve of an “I” that thinks it can stand outside everything that moves it, as if a decision could be made from nowhere. The second is said by the same man after the whole teaching, with his doubts gone, to the one who handed him the choice. The poem does not remove the “I”. It moves it, from a place outside the causes to a place among them, where it can reply.
“I answer for this” belongs with the second sentence when three things are true. Someone is owed it: the verse that hands back the choice is spoken to “you”, so the reply begins with another person, not in self-assertion. It is about your part, said by one cause among several. And it comes after a real effort to understand the reasons, the doubts and what may follow. Said that way, it is a reply, not a boast.
Being asked does not create the debt; being owed does. A company that finds a dangerous fault in what it sells should report it before anyone asks, and a doctor who has made an error should say so unasked. The patients who take the medicines made at Meera’s site will almost never ask who read the batch records. The explanation is owed to them all the same, and someone must be ready to give it. Nor is answerability blame. A person can act reasonably and still owe an explanation. What each person owes depends on her role, what she knew and the power she had.
In three other cases the same words rebuild the ego. They do so when they claim to stand outside the causes, as the only doer, and when they are said in order to be admired. They do so, too, when a leader takes all the blame and so closes the question of the other hands. A chief who says “the fault is mine” can end an inquiry that would have found a bad process, a wrong incentive or a manager nobody asked. Taking all the blame can be a way of keeping the record short.
Nor is it innocence. Arjuna’s “I will” led into the war. The epic that holds the Gītā has a whole book, the Strī Parva, in which the women walk the field afterwards and weep beside the dead9. Answering for a choice does not make it a good one. It makes it yours to explain, and to put right where it can still be put right. An explanation is not a remedy, and an apology is not repair. When a harm cannot be undone, answering for it may mean paying for it, correcting a record, opening a way to appeal or changing the institution that let it happen.
Go back, last, to the room. Suppose the machine in its chair weighs the reasons better than you. That does not give it, or whoever owns it, the right to decide. Being helped does not take your part away: Arjuna was persuaded, and the act was still his. Causing part of an outcome and owing an explanation for it are different questions, and neither settles the other. A decision is not made legitimate because someone in your chair accepted what the machine advised. And the third chair cannot be made invisible: the grandmother booking on her phone, the patient, the soldiers across the field. You need not be the best intelligence in the room. You must not become only the one who lives with a decision nobody will answer for.
Ending: Both Versions, Sent
At the end of the week Arjun sends Vandana both versions, his and the machine’s.
With them goes the page he now writes at the end of every project. It says what the machine produced, and what he chose and changed. This time it also carries the machine’s three objections to his own version, and Neha’s point about the small picture. The machine’s third is marked as the one he cannot answer.
He does not tell her which to choose. The choice is hers to make for the chain, and its guests will live with it. She may also choose neither, ask for a third, or try both on guests first. The six months end this month, and whether to renew is hers to decide as well.
Her reply has not come.
Words to keep
- ahaṃkāra: the “I-maker”; the sense “I am the doer”, which in the Gītā deludes a person into thinking he alone acts; translators render it “egotism” or “self-conceit”, and commentators read it in different ways.
- answer for: to be the one who must explain a decision, undo it where it can still be undone, and put right what can be put right.
- batch record: the document that shows how one batch of a medicine was made and checked.
- doership: the belief that I alone caused an act; not the same as agency, the power to intend and act, or answerability, the duty to explain my part and repair what I can.
- facticity: Sartre’s word for the given part of a situation, which a person did not choose and must act within.
- master record: the record of how each step of making a medicine should be done; each batch record is read against it.
- rathin: the lord of the chariot; the one whose journey it is, who can refuse it.
- restraint: the decision about what a design or a brand will not say, taken by someone who will be blamed if the silence fails.
- sārathi: the charioteer.
- yantra: a machine or a wheel; in Gītā 18.61, what all beings are turned on, “(as though) mounted”.
- yathecchasi tathā kuru: “act as you like” (Telang); the last words of Gītā 18.63, spoken to Arjuna after the whole teaching.
Sources
- Forecasting Research Institute, “AI models have likely reached parity with superforecasters on ForecastBench,” Substack, 16 July 2026 (several systems “statistically indistinguishable from superforecaster-level accuracy”; the results “more consistent with superforecaster parity than with outperformance”).
- Good Judgment, “Not So Fast,” Good Judgment News and Insights (Substack), 24 July 2026: the comparison sets “continuously improving AI systems against a frozen human baseline from a single 2024 engagement”.
- METR, “Task-Completion Time Horizons of Frontier AI Models” (metr.org/time-horizons, consulted 23 October 2026), update log: “May 8th, 2026: Added Claude Mythos Preview (early) and notice that ‘Measurements above 16 hrs are unreliable with our current task suite.’”; “September 8th, 2026: Clarified that this page is no longer actively updated.” The measure is the length of task, in a skilled person’s hours, that a system finishes on its own about half the time; the tasks are mostly software tasks.
- OpenAI, “Buy it in ChatGPT: Instant Checkout and the Agentic Commerce Protocol,” 29 September 2025 (openai.com), launched for users in the United States buying from US sellers on Etsy. In March 2026 an OpenAI spokesperson said that “Instant Checkout is moving to Apps”, retailers’ own apps inside ChatGPT (Brian Warmoth, Digital Commerce 360, 6 March 2026), and OpenAI scaled the feature back, “deprioritizing the feature as a standalone offering” (Tim Hardwick, MacRumors, 25 March 2026; Modern Retail, 27 March 2026).
- Bhagavad Gītā, trans. Kashinath Trimbak Telang, The Bhagavadgîtâ with the Sanatsugâtîya and the Anugîtâ, Sacred Books of the East 8 (Oxford: Clarendon Press, 1882): 1.21, p. 39; 2.9, p. 43; 18.59, p. 128, with Telang’s note on “Nature”: “The nature of a Kshatriya, Sankara”; 18.61, 18.63 and 18.66, p. 129; 18.73, p. 130 (sacred-texts.com). 3.27 (p. 55) is paraphrased: Telang has “He whose mind is deluded by egoism thinks himself the doer”, and the Sanskrit kartāham iti manyate, “thinks ‘I am the doer’”. Sanskrit of 2.9, 18.59, 18.61 and 18.63 from the IIT Kanpur Gita Supersite (heritage.iitk.ac.in/SBG). In 18.59, yad ahaṃkāram āśritya na yotsya iti manyase repeats the na yotsya iti (na yotsye, “I shall not fight”) of Arjuna’s refusal in 2.9; in 18.61, yantrārūḍhāni, mounted on a yantra, which some translators render “wheel”. 18.13–14, p. 123 (page confirmed in the sacred-texts.com copy, 4 November 2026): “these five causes of the completion of all actions” are “The substratum, the agent likewise, the various sorts of organs, and the various and distinct movements, and with these the deities, too, as the fifth.” Modern translation and commentaries compared (added 4 November 2026): Swami Gambhirananda, trans., Bhagavad Gītā with the Commentary of Śaṅkarācārya (Kolkata: Advaita Ashrama), and Rāmānuja’s commentary in Swami Adidevananda’s translation, both as given on the IIT Kanpur Gita Supersite and read through its data copy (vedicscriptures.github.io, 4 November 2026); the printed books were not seen. Gambhirananda: 18.14 “The locus as also the agent, the different kinds of organs, the many and distinct activities, and, the divine is here the fifth.”; 18.59 “That you think ‘I shall not fight’, by relying on egotism” and “vain is this determination of yours” (quoted in two parts because the text sets a dash between them); 18.63 “Pondering over this as a whole, do as you like.”; 18.73 “I stand with my doubt removed; I shall follow Your instruction.” Śaṅkara on 18.17, in Gambhirananda’s translation: “the feeling of egoism, in whom does not occur the notion in the form, ‘I am the agent’”; on 18.59, “your own nature of a Ksatriya”. Rāmānuja on 18.59: “in your ‘self-conceit,’ i.e., under a false sense of independence that you know what is good for you and what is not”. Telang and Gambhirananda both keep 18.63 in the second person and render ahaṃkāram “egotism”; Adidevananda has “self-conceit”.
- Jean-Paul Sartre, “Existentialism Is a Humanism,” a lecture given in Paris on 29 October 1945, published as L’existentialisme est un humanisme (Paris: Nagel, 1946); trans. Philip Mairet as Existentialism and Humanism (London: Methuen, 1948), reprinted as “Existentialism Is a Humanism” in Walter Kaufmann (ed.), Existentialism from Dostoevsky to Sartre (Meridian, 1956; 1989 printing at marxists.org). The punctuation of “You are free, therefore choose, that is to say, invent.” follows the reprint, as does the spelling “labor” in “What never vary are the necessities of being in the world, of having to labor and to die there.” Facticity (added 31 October 2026): Sartre’s term (facticité) in L’Être et le néant (Paris: Gallimard, 1943) for the given side of a situation; see Pierre-Jean Renaudie, “Jean-Paul Sartre,” Stanford Encyclopedia of Philosophy (revised 10 July 2026).
- Myra Cheng, Cinoo Lee, Pranav Khadpe, Sunny Yu, Dyllan Han and Dan Jurafsky, “Sycophantic AI decreases prosocial intentions and promotes dependence,” Science 391, no. 6792 (March 2026): eleven models affirmed users’ actions about 49 per cent more often than humans did. DOI 10.1126/science.aec8352.
- Nithin Palal, The Philosophy of Everything: On the Pursuit of Happiness, Part II, Ch 8, “Free Will or Determinism?” (nithinpalal.com, 16 October 2026). The Philosophy of Everything (print, 2022) is not quoted here.
- The Mahabharata, Book 11, Stri Parva, trans. Kisari Mohan Ganguli (1883–1896), sacred-texts.com, Sections XVI and XVII: Gandhārī, given sight of the field, sees the women among the dead.
Named people in this chapter are composites, and their conversations and circumstances are constructed; the studies, laws and figures cited are real and sourced.
Revised from the print edition (2026)
- What the print edition claimed: that the machine does its work “without knowing it is doing anything at all”; “The machine processes. You experience the processing. The machine produces output. You produce meaning.”; that “One of them processes the world. The other experiences it”, the one “fast, tireless, and continuously improving”, the other “slow, fragile, and conscious”; that the age of AI asks “now that the machine handles intelligence, what will you do with your consciousness?”; that the machine “is very good at waiting”, with “no experience of time”; that “You have changed”; that “knowing without being able to prove that you know is the beginning of everything this book has tried to say”; that teaching is “the cultivation of another person’s awareness”, and said of her platform, “It will not use her”; that Arjun rejects a good design “Because it is not his”, that the difference “will be invisible to anyone who has not spent twelve years learning to see it” and that “The difference is everything”; that “the old Meera” answered at once and “This Meera” pauses; and that “You are aware. That is the difference. That has always been the difference.” It also gave Rajan “thirty years”, Kavitha “thirty-four” students and “eleven years”, Meera a “governance framework” with “four colours of marker”, ended on a roll-call of the five, and pointed to an orphan (“the final was honest about the limit”).
- What I now hold: whether machines are aware is open, and nothing here turns on it. As of October 2026 machines forecast about as well as the best human forecasters on one test (a disputed result), work alone on long tasks and act for people; they do not wait. The choice is handed to the one whose life it is, after the reasons are in, and knowing more does not by itself give anyone the right to make it for her, though a right given in other ways can (a doctor for an unconscious patient, a parent). Choosing, the right to decide, being affected, explaining and repairing can sit with different people, and a decider’s responsibility is not the affected person’s consent. Choosing whom to ask, and how, is part of the choice; a refusal that costs a livelihood returns the choice in form only. “I answer for this” does not rebuild the ego the Gītā loosens when it is a reply to those owed one, about one’s own part, after a real effort to understand; it does when it claims to stand outside the causes, closes the question of other hands, or is said to be admired. Answering is not innocence. Being asked does not create the debt; being owed does. Doership, agency and answerability differ; answerability is not blame, and an explanation is not a remedy. Advice one could have refused, and took knowing what it was, is still a choice; letting a machine act is a choice made earlier. Restraint is the decision about what will not be said, by someone who will be blamed; a machine can copy its silences without its reason, and part of what it left out may be the designer’s own; fidelity to what a system was given is not relevance to what has changed. A difference only the designer can see is not paid for on faith: he says what it does for the people who live with it, and sorts the rest into what he suspects and what he cannot yet show. “You can” and “You should not” answer different questions, and whose decision it is is a third. Agreement between a hand and a system is not proof, and a signature must not outlive the competence, independence and authority that give it meaning.
- What changed my mind: the Gītā read to its end, from Arjuna’s refusal (2.9) through 18.59 to 18.73; Sartre’s pupil and his adviser; The Philosophy of Everything: On the Pursuit of Happiness (essays on nithinpalal.com, October 2026); the forecasting and task-length results of 2026; the Prologue’s rule that the book never defines a person by what machines cannot yet do; the audit’s corrections of the five vignettes; one outside reading of an earlier version of this chapter, in October 2026.
- What survives: the room (“Two minds in a room.”; “You have been here before.”); “The machine left out what it did not have, and called the leaving-out elegance.”; “You can,” she says. “You should not.”; “The choice is not between them.”; the question handed back to the reader, now through the verse kept for it.
- Section titles in the print edition: none; the print Epilogue ran without sections, and its four headings are this edition’s.