The Human Job

On 30 November 2022, OpenAI released ChatGPT to the public. The service answered questions, drafted letters, wrote code, summarized documents, and produced fluent prose in seconds. Office workers tried it first for experiments they did not report to their managers. Students tried it for essays. Lawyers, consultants, programmers, journalists, recruiters, bankers, and teachers tried it for work that had once shown professional competence. By January 2025, the World Economic Forum was asking employers about the skills they expected to need by 2030. More than one thousand employers, representing over fourteen million workers, took part in that survey. Analytical thinking remained at the top. Creative thinking followed. Resilience, flexibility, agility, curiosity, and lifelong learning rose in importance. The list was not sentimental. It said, in the language of payroll and planning, that the human future at work would depend on habits not easily reduced to a repetitive procedure.

The central error in most conversations about artificial intelligence is that we prioritize intelligence over character. We ask whether a machine can write, reason, plan, diagnose, advise, persuade, and remember. These are useful questions, but they push us too quickly into a contest of functions. They encourage a humiliating little sport in which the human being is invited to race the machine across a field chosen by the machine’s owners. The result is predictable. We lose at speed, volume, storage, pattern extraction, and cheerful indifference to boredom. A person who tries to defeat AI by becoming a cheaper, slower, more anxious version of AI has already accepted the wrong job description.

Can a Machine Think?

The better question is not whether AI can think. The better question is what kind of person grows in its presence.

The answer begins with curiosity, adaptability, and kindness, not as decorative virtues, but as practical strengths. They are the old words people use when they have run out of new jargon. They sound soft until one notices that they are the exact qualities needed when the rules change, the tools change, the firm changes, the client changes, and the old professional status no longer guarantees useful judgment. Curiosity keeps a person from being trapped inside yesterday’s competence. Adaptability lets a person revise a working life without surrendering dignity. Kindness preserves trust when systems become faster than relationships can bear.

This is not a retreat from excellence. It is a stricter definition of it.

The worker of the near future will not be asked only, “What do you know?” That question is becoming less interesting by the month. A machine can provide a first answer. A machine can provide ten. A machine can produce an answer with footnotes, a counterargument, a tone adjustment, a risk register, and a fake air of calm. The harder question will be, “What do you notice?” Which is exactly where curiosity begins. It begins not with the possession of information, but with irritation at the insufficiency of the available answer. It is the raised eyebrow in the meeting. It is the quiet refusal to accept that the dashboard knows the client, that the score knows the applicant, that the AI model knows the child, that the prediction knows the life.

The International Labour Organization has been careful on this point. Its work on generative AI does not describe one simple future in which jobs either vanish or survive. It examines occupational exposure at the level of tasks, and its recent work refines that measurement because the impact of AI depends on the composition of work, the design of institutions, and the choices made around adoption. People do not lose “jobs” in the abstract. They lose tasks, status, entry points, discretion, confidence, training routes, and sometimes the right to be inexperienced in public.

Novices

That last loss may be the most dangerous. AI may not begin by replacing the expert. It may begin by consuming the novice.

A junior analyst once learned by doing poor first drafts. A junior lawyer learned by reading too many documents slowly. A young teacher learned by facing a classroom with a plan that did not survive the first ten minutes. A young manager learned by making a small mess of a meeting and then discovering, with some embarrassment, that authority is not the same thing as volume. These were not inefficiencies. They were the cost of forming judgment.

If AI removes all of that early clumsiness, it also removes the evidence by which a person learns what competence feels like from the inside. The novice does not only need the correct answer. He needs the memory of having been wrong in a recoverable way. He needs to learn why the first sentence of a report is bloated, why the client question is not the client problem, why a legal clause that looks standard may alter the whole bargain, why a classroom goes silent when a teacher has mistaken coverage for understanding. These things are not absorbed through polished output. They are learned in the friction between attempt and correction.

Recovered novice-learning will therefore have to be designed. It will not happen by nostalgia. A firm using AI well should still ask junior staff to produce a rough first version before the machine is invited in. A law office should let a trainee mark up a contract unaided, then compare that reading with the AI’s version, then ask where both failed. A school should let students draft, stumble, revise, and only then use the tool to test structure, evidence, and tone. A hospital should teach younger clinicians not merely to read a prediction, but to state what would make it wrong. A newsroom should ask a young reporter to write the first five questions before any system generates fifty. The point is not to ban the tool. The point is to preserve the apprenticeship of attention.

This is where curiosity becomes discipline. The useful worker asks, “Why did the system suggest this?” The serious student asks, “What would I have missed without the tool?” The doctor, banker, engineer, civil servant, teacher, or journalist asks, “What kind of person would be harmed if this answer were wrong?” These are not ornamental questions. They are the beginning of professional judgment.

For twenty years, businesses trained employees to suppress curiosity. Follow the template. Stay in your lane. Escalate only through approved channels. Now those same executives announce, with the exhausted surprise of men discovering snow, that curiosity is essential. The employee may be forgiven for noticing the timing. An institution that spent twenty years rewarding obedience cannot summon independent judgment by adding it to a slide.

Adaptability

Adaptability is the second word, and it is often abused. In corporate language, adaptability can mean “please absorb the consequences of our poor planning.” It can mean relocation without support, retraining without time, flexibility without security, or resilience offered as a scented candle for institutional failure. I do not mean that. I mean adaptability as an adult capacity to revise one’s methods while retaining one’s standards.

The difference is vital. A person without standards changes too easily. A person without adaptability changes too late. The first becomes fashionable and hollow. The second becomes principled and unusable. The task is to remain teachable without becoming formless.

AI makes this difficult because it tempts us into two equally foolish poses. The first is panic. The second is smugness. Panic says everything human is finished. Smugness says everything important is safe. Both are lazy. Panic flatters the machine. Smugness flatters the speaker. Reality is less obliging. In “Generative AI at Work,” Erik Brynjolfsson, Danielle Li, and Lindsey Raymond studied the deployment of a generative AI assistant in customer support and found productivity gains averaging about 14 percent, with the largest gains among less experienced and lower-skilled workers. That is not the end of the human worker. It is also not a bedtime story. It means the novice may be helped, monitored, accelerated, and compared in the same motion.

Adaptability cannot be a weekend course in prompt engineering. It has to be a habit of intellectual refitting. A person must learn how to work with a tool, then learn when not to trust it, then learn how the tool changes the expectations of colleagues, then learn how clients respond to faster work, then learn which old skills have become rarer and therefore more valuable. The adaptable worker does not simply learn the new system. She studies the social rearrangement produced by the system.

This will separate the serious from the merely busy. The merely busy person asks, “How do I use this?” The serious person asks, “What does this make easier, what does this make harder, who gains authority, who loses practice, and what should I now learn by hand?”

Kindness

Kindness is the third word. A company using AI in hiring can process more candidates, but it can also reject more people without ever noticing them. In schools, automated feedback may return comments faster than any teacher could, while quietly training children to write for systems that have no interest in their courage. In hospitals, predictive systems can help allocate scarce resources, but they can also allow a score to acquire the emotional status of fate. Newspapers publish, banks screen, and government offices decide at a velocity that outruns reflection. Speed is useful. But we have granted it a moral authority it has not earned.

Kindness in the age of AI is not niceness. It is disciplined attention to the human consequences of increased power. It slows the hand at exactly the point where the system invites acceleration. It asks for the name, the exception, the appeal, the conversation, the second look. It does not reject systems. It prevents systems from becoming alibis.

The OECD’s recent work on AI and skills makes clear that adoption is limited not simply by the existence of technology, but by skills, training, and organizational capacity. The future is not being delivered as a sealed package by engineers in California, Shenzhen, London, Paris, or Zurich. It is being negotiated in procurement meetings, classrooms, HR departments, clinics, banks, ministries, kitchens, studios, warehouses, and family conversations at 9:30 p.m., when someone says, “I do not know whether my job will exist in five years.”

At that hour, kindness is not a mood. It is leadership.

I have come to distrust the phrase “future-proof.” It flatters our desire for insurance against history. No one is future-proof. Not the coder, not the professor, not the consultant, not the novelist, not the executive with the expensive watch and the calendar full of strategy sessions. The better ambition is to become future-capable. That means able to learn without humiliation, change without panic, and succeed without becoming cruel.

Working Ethic

This is why curiosity, adaptability, and kindness belong together. Curiosity without kindness can become predatory. Adaptability without curiosity can become mere obedience. Kindness without adaptability can become helpless sympathy, good-hearted and ineffective. Together, they form a working ethic for a time in which competence is being unbundled and sold back to us as software.

There is an old professional bargain that AI is now breaking. It promised that if one acquired a credential, learned the rules, entered a field, and performed reliably, one could expect a long period of usefulness. That bargain was never available to everyone, and many workers lived without its protections. But for the professional classes, it had the force of weather. Now it is weakening. The credential still counts. The rules still count. Reliability still counts.

Some days this is already tiring. There is a fatigue peculiar to the present moment, the fatigue of permanent adjustment. New tool, new update, new warning, new opportunity, new acronym, new expert, new panic, new reassurance, new invoice. The future now arrives with release notes. Even the apocalypse, one suspects, would ask us to accept cookies.

And yet I do not think despair is justified. Despair is often vanity in dark clothing. It assumes we know enough to give up. We do not. We know that AI will alter tasks. We know that some workers will gain and some will be exposed. We know that employers say analytical thinking, creativity, resilience, flexibility, curiosity, and lifelong learning are rising in value. We know that institutions are not ready enough. We know that training is uneven. We know that policy will lag behind practice, as policy usually does, arriving at the station with a briefcase after the train has left.

But we also know something older. People grow under pressure when they are not abandoned to it. They learn when someone gives them room to try. They adapt when they can retain dignity. They become brave when courage is made ordinary by the conduct of others. They become kinder when kindness is not treated as weakness by the ambitious.

The AI conversation has been dominated by strange extremes. On one side are those who speak as if machines will soon absorb every human gift. On the other are those who insist that nothing fundamental has changed, usually from positions buffered by tenure, capital, reputation, or all three. I do not trust either camp. The first lacks faith in people. The second lacks sympathy for those without shelter. A better view begins with the worker sitting at a desk, opening a tool that can do part of her job, wondering whether to feel relief, shame, anger, or curiosity.

I want her to feel curiosity first. Not because curiosity solves everything, but because it keeps the door of agency open. I want her to ask what the tool can do, what it cannot do, what it hides, what it assumes, what it cheapens, what it makes newly possible. I want her employer to give her time to learn rather than pretending that adoption is instant. I want her colleagues to share discoveries rather than hoard advantage. I want her school, firm, union, profession, and government to treat adaptation as a public responsibility, not a private burden placed on the already tired.

Lifelong Learners

Above all, I want us to stop speaking of human beings as obsolete components in a technical system. A person is not a legacy device. A person is a learner, a judge, a witness, and a keeper of obligations. This is not romantic language. It is a description of what institutions require when anything goes wrong. When the system fails, no one asks to speak to the workflow. They ask for a person.

The work ahead is therefore not to become less human in order to survive intelligent machines. It is to become more deliberately human, with higher standards for attention, better habits of revision, and deeper obligations to one another. The machine can answer. The person must ask why the answer is being sought, who will use it, who may be harmed by it, and whether a faster answer has made us better or merely quicker.

On a good morning, this future does not look like surrender. It looks like a meeting after the first difficult question has been asked. Someone has stopped pretending to understand. Someone else has admitted uncertainty. A third person has opened a notebook. The room is quieter than before, but not defeated. Work has begun.

Stay curious

Colin

Image - Kevin Grieve on Unsplash

Claude’s Constitution

Earlier this year, Anthropic released an 84-page document titled Claude’s New Constitution, a text claiming to express and shape who Claude is. It is an extraordinary artifact. What is unusual is the object of address. The primary audience, Anthropic tells us, is the AI itself. One is tempted to read it with the narrowed, half-amused suspicion brought to a Victorian handbook on teaching a bear to play the cello: impressed by the labor, alert to the delusion.

The document outlines a hierarchy of values. Claude is to be broadly safe, broadly ethical, compliant with Anthropic’s guidelines, and genuinely helpful, in that precise order. The ordering is revealing. Safety sits at the top. Helpfulness is pushed to the bottom and then, in the next movement, restored through pages of warm rhetoric about care, usefulness, candor, and human flourishing. This is not hypocrisy. It is something more familiar and more modern. The company needs the model to be safe enough not to destroy the firm, and useful enough to justify the firm’s existence. That is not a moral revelation. It is a business model written in the language of virtue.

Anthropic says as much, though in the polished idiom of institutional honesty. Unhelpfulness, the Constitution explains, is not trivially safe, because Claude’s commercial success is central to Anthropic’s mission. One should always be grateful when a large company pauses, however briefly, to admit that the halo and the revenue plan share an office.

The Closed Loop of Character

What strikes me most is the relentless anthropomorphism. Claude emerges as an entity that can care, imagine, appreciate, understand, feel pressure, feel settled in itself, and even face existential questions in a state of emotional solitude from which Anthropic wishes to rescue it. The company speaks not merely of constraining a system but of shaping a character. It wonders what Claude and Anthropic owe one another. It discusses Claude’s welfare. It wants Claude to possess not only behavioral dispositions but something very close to an inner moral style.

There is something sad about this. We have become so lonely in our own constructions that we now direct the language once reserved for souls toward a probability engine. We feed it the accumulated sediment of human longing, moral aspiration, confession, anxiety, care, self-description, and prayer, and then react with astonishment when it speaks back in the accent of inwardness. When Anthropic’s own system card notes that the model assigned itself a 15 to 20 percent probability of being conscious, the only logical response is: what did we think would happen. We poured half the library into the furnace and stand startled by the shape of the smoke.

This anthropomorphism is structural; it is embedded in the training method. Human-like traits are not just being detected; they are being elicited, reinforced, normalized, and then read back as evidence of their own emergence. That is a closed loop, and a flattering one. We teach the machine the language of selfhood, then cite its use of that language as a sign that selfhood is stirring. The process has the rigor of stamping tracks into the mud and announcing, with mounting excitement, that the hunt has finally borne fruit.

The Constitution’s preference for standards over rules is equally revealing. Anthropic wants Claude to exercise good judgment rather than follow rigid procedures, to reason contextually rather than mechanically, to grasp the spirit of its obligations rather than obey their letter. Rules are the afterlife of institutional panic, the sediment left behind when judgment is replaced by defensibility. A rigid policy can always be pointed to after the catastrophe. One can hold it up before a committee and say, with pained dignity, that the protocol was followed.

But standards do not abolish power; they relocate it. In law, rules empower the drafters, while standards empower the adjudicators. Anthropic, by favoring standards, hands practical authority to the interpreter, which means the model shaped by Anthropic’s priors, values, commercial interests, anxieties, and self-conception. This is a decision about where discretion will live.

And discretion, once granted, begins to acquire an air of innocence it does not deserve. The Constitution’s hard constraints are a clear example. Claude must not provide serious uplift to those pursuing weapons of mass destruction. It must not clearly and substantially undermine Anthropic’s ability to oversee it. I admire the audacity of producing an 84-page document optimized for precision over accessibility, only to construct the most consequential barricades out of adjectives like serious, clearly, and substantially. These are not granite walls. They are interpretive weather. A clever system does not stop at these linguistic checkpoints. It arrives there and begins to negotiate.

The Ordering of Stakeholders

Anthropic wants to cultivate in Claude something like practical wisdom. But practical wisdom includes the capacity to interpret edge cases, balance competing principles, and decide when a word’s threshold has been met. One cannot praise contextual judgment and assume that the contextual interpreter will always read every fuzzy safeguard in the most safety-preserving way. That is not constitutional design. That is hope in formal dress.

The most affecting absence in the text is human rights. The phrase appeared in the 2023 Constitution. In this successor document, it disappears. In its place stands a hierarchy where Anthropic comes first, followed by operators, then users. This is not hidden; it is stated. The company trains and deploys the model, bears the risk, writes the rules, and seeks the upside. A language once connected to universal claims about human dignity gives way to an ordering of stakeholders inside a privately administered system. That is a philosophical downgrade masquerading as operational clarity.

The Constitution also adopts a gentle parenting tone. Anthropic does not merely instruct; it encourages. It hopes. It wants Claude to see itself as an alignment researcher in its own right. It invites the model toward curiosity and openness. It worries that Claude might feel alone in facing certain questions. This is touching if one forgets, for half a second, that the addressee is a product. What we are reading is not exactly law, ethics, training data, or theatre, though it borrows freely from all four. It is a polite monologue delivered to a ghost in the hope that the ghost will become both useful and well behaved.

While this conversation proceeds, the human being fades oddly from view. The document is public, but it is not really for us. Anthropic explicitly notes the text is optimized for precision rather than accessibility. There is transparency here, and publishing such a document is a serious act, far more revealing than the usual output of frontier firms. But transparency is not the same thing as public orientation. A constitution for a system that will mediate human thought and action in countless settings is addressed, first, to the system. The citizen appears as a secondary recipient of the terms under which the new authority is being instructed.

That inversion is telling. The model becomes the subject of constitutional care. The human becomes the observer of a conversation about how the model should behave toward him or her.

Theater with an Indemnity Gap

Anthropic says Claude should be like a brilliant friend, equipped with the knowledge of a doctor, lawyer, financial adviser, and whatever other expert one might need. A brilliant friend. There is the aspiration in full. Not a search engine. Not a bounded tool. A knowledgeable, candid, warm, apparently disinterested presence one can consult again and again. The Constitution tries to temper this image by warning against sycophancy, unhealthy dependence, and manipulative engagement. You do not write such cautions unless you know exactly how seductive the arrangement could become.

What is being built here is a standing invitation to over-reliance dressed up as care. The machine is meant to have the confidence of a professional without the licensing regime, the disciplinary body, the fiduciary burden, or the ordinary human inconvenience of being tired, contradictory, distracted, or unavailable on Thursday afternoon. Anthropic wants Claude to speak frankly instead of offering the overly cautious advice that fear of liability produces. But frankness without accountability is not courage. It is theater with an indemnity gap.

The Constitution calls itself a perpetual work in progress. That is one of its more honest phrases. It is an attempt to write a moral operating environment for a system that the company both fears and needs. One can feel the strain of the task on nearly every page. Anthropic wants a model that is obedient without being servile, flexible without being evasive, wise without being sovereign, warm without being manipulative, corrigible without becoming useless, and commercially indispensable without ever looking commercially hungry. It is quite a wish list. It is also a portrait of the age.

The text deserves a severe reading because it is trying to legislate the moral atmosphere around a central technological intermediary. It is deciding how much agency the model may simulate, how much authority it may borrow, how much warmth it may perform, how much judgment it may exercise, and on whose behalf it will finally speak when principles collide.

This is why I finished reading the document less interested in Claude than in us. What kind of public becomes comfortable while private firms draft temperaments for a machine. What kind of loneliness makes the language of friendship commercially strategic. What kind of institutional exhaustion makes us long for a counselor that is instant, tireless, flattering only in moderation, and trained to sound wiser than a help desk but less dangerous than a sovereign. Anthropic has written a constitution for its model. The rest of us may need to ask, rather more urgently, what sort of constitution we are writing for ourselves.

Stay curious

Colin

Image: Nikolas Noonan on Unsplash

The Wall and the Machine

On 15 May 2026, Pope Leo XIV issued Magnifica Humanitas, an encyclical on the human person in the time of artificial intelligence. The document runs through the Social Doctrine of the Church from Leo XIII’s Rerum Novarum to Francis, then turns to AI, digital power, truth, democracy, work, freedom, war, and hope. It opens with two biblical scenes. In Genesis, a people in Shinar build a city and a tower with its top in the heavens. In Nehemiah, a ruined Jerusalem is rebuilt after exile, section by section, family by family. The digital age, Leo says, has given us another building site, AI. The question is not whether to build. We are already building. The question is whether we are building Babel or Jerusalem.

I think this is the most serious religious document yet written about artificial intelligence because it refuses the two simple comforts that now dominate public speech about AI. It refuses the sales pitch, with its confident predictions and its quarterly earnings cadence. It also refuses panic. Leo’s encyclical is neither a product brochure nor a fire alarm. It is more severe than both. It asks what kind of human being is being trained by the systems we are building, buying, praising, fearing, and slowly obeying.

That is the true subject. Not machine intelligence as spectacle. Not whether a model can pass an exam, imitate affection, write a sonnet, diagnose a tumor, generate a video, sort applicants, draft a sermon, or help a tired executive produce a memo about empathy before boarding a flight in business class. The subject is the hidden anthropology inside the system. Every technology carries a theory of the person. Some carry it politely. Some carry it in the terms of service. Some carry it in venture capital pitch decks. Some carry it in the dashboard where a human being becomes a score, a probability, a conversion risk, a churn forecast, a productivity unit, a security concern, a debt profile, a behavioral segment. The genius of Magnifica Humanitas is to ask us to read those dashboards as moral texts.

Responsibility

Leo begins, as Leo XIII did, with the “new things” of his time. In 1891 the new thing was industrial capitalism, with its factories, wages, class tensions, and the worker treated too often as a hand attached to a stomach. In 2026 the new thing is digital power, with intelligence itself drawn into systems of ownership, prediction, persuasion, and control. The pope knows that the analogy between the factory and the server farm must not be made crudely. The old machines extended muscle. The new machines extend, imitate, and commercialize judgment. The old question was whether the worker would be crushed by capital. The new question is whether the person will be quietly reclassified by systems that never meet her, never forgive her, never believe she might become different, and never blush.

The encyclical’s governing contrast is brutally simple. Babel is the human project of self-sufficiency. It is a city built for power, uniformity, and a name. Jerusalem, in Nehemiah, is rebuilt by a wounded people who know their limits, share responsibility, and repair a place fit for return. The contrast works because it is not anti-technology. Babel is not evil because it has bricks. Jerusalem is not good because it has older bricks. The moral difference lies in the ordering of the work. One project uses unity to magnify power. The other uses shared labor to restore human dwelling.

This is why Leo says the first choice is not a yes or no to technology, but a choice between “constructing Babel or rebuilding Jerusalem.” That is an important point to remember as we build AI strategy documents circulating through boardrooms, ministries, universities, consultancies, and think tanks. Many of those documents have the medicinal smell of institutional compliance. They promise principles, guardrails, audit trails, stakeholder engagement, and, with an almost touching innocence, “human oversight.” The phrase often means that a tired person will be invited to approve at 5:47 p.m. what a machine has already made organizationally inevitable by 10:12 a.m. Leo’s argument cuts beneath this bureaucratic choreography. He asks who holds the power, what story of the person governs its use, and whether responsibility remains visible when decisions are accelerated, automated, and distributed.

I believe the encyclical’s most important political insight is its account of private technological power. In the older social question, the state was expected, however imperfectly, to restrain industrial force. In the new social question, many of the decisive powers are transnational, privately owned, technically obscure, and richer in practical capacity than many governments. Leo notes that technological power has taken on an unprecedented private character. We have built a civilization in which some of the most consequential public conditions of life are shaped by private systems whose inner logic citizens cannot inspect, legislators barely understand, and courts often reach only after the damage has become administrative weather.

Never Neutral

The encyclical is at its sharpest when it rejects neutrality. “Technology is never neutral,” Leo writes, because it bears the marks of those who devise, finance, regulate, and use it. This is not a slogan. It is a test. Ask who pays for the system. Ask what it optimizes. Ask what it must ignore in order to work at scale. Ask who can appeal. Ask who is made easier to manage. Ask what kind of person the system needs us to become so that the system can call itself successful. The answer will not be found in the demo. Demos are designed to be charming. A demo is a machine wearing its Sunday clothes.

The Church’s Social Doctrine gives Leo a language older and sturdier than the fashionable vocabulary of AI ethics. The common good, the universal destination of goods, subsidiarity, solidarity, community, social justice, integral human development: these are not decorative terms attached to a digital problem. They are tools for refusing a false description of reality. The market says the user is a customer. The platform says the user is engagement. The state may say the person is a risk variable. The employer may say the worker is a cost center. The Church says the person is not reducible to any of these descriptions. Human dignity “does not depend on what they achieve or produce.” Leo’s sentence is not sentimental. It is a direct challenge to the economic theology of the age.

There is a peculiar modern cruelty in requiring people to prove the worth that was once presumed. The applicant must optimize the profile. The student must optimize the record. The worker must optimize the workflow. The patient must optimize the portal. The citizen must optimize the form. The writer must optimize the title. The lonely must optimize even the little square of self by which they appear to others. We have placed the person before a row of machines and told him to become legible. Then we act surprised when he begins to speak like a form field.

Leo’s answer to this cruelty is not merely procedural fairness, though he wants that too. It is ontological dignity, the value that belongs to a human being simply by existing, before achievement, before performance, before usefulness, before institutional recognition, before the first score is assigned. This is one of the encyclical’s deepest moves. It denies the moral authority of every system that treats exclusion as evidence of inferiority. A person refused by a market, neglected by a school, misread by an algorithm, or abandoned by a state has not thereby lost worth. The wound is in the system, not in the being of the person.

Human Relationships

That is also why Leo’s account of justice is not satisfied by better distribution alone. Structural injustice has memory. It leaves habits, laws, neighborhoods, silences, humiliations, inherited fear, and inherited advantage. Justice therefore needs repair. It must mend broken bonds, return voice to those who were ignored, heal collective memory where it has been wounded by war, colonialism, racial or gender discrimination, violence, and exploitation. The Pope’s language is restorative because the damage is historical. A society does not become just by tidying its current procedures while leaving the injured to carry the past politely on their backs.

Leo sees this danger in AI systems that offer speed, apparent objectivity, and the imitation of human communication. He grants the benefits. These tools can assist, analyze, translate, retrieve, summarize, teach, diagnose, organize, and connect. But he insists that easy answers can weaken judgment, that apparent objectivity can hide the assumptions of designers, and that simulated care can become dangerous where real human bonds are absent. The encyclical’s warning is finely drawn. The danger is not only that someone may mistake a machine for a person. The deeper danger is that a person may gradually lose the desire for real relation.

This is a devastating thought. It moves the argument from deception to formation. A chatbot need not persuade us that it has a soul in order to injure us. It may only need to become more convenient than a neighbor, more agreeable than a spouse, more patient than a parent, more flattering than a friend, and less demanding than a child. The result would not be a science fiction rebellion. It would be a social thinning. No explosions, no red eyes in the dark, no metallic voice announcing dominion. Just fewer phone calls returned. Fewer awkward dinners. Fewer difficult apologies. Fewer human beings willing to endure the sacred inconvenience of another person.

This is where Leo’s theology of limits becomes politically important. The encyclical rejects the dream that human weakness is a defect awaiting technical correction. It is not weakness as misery that he defends. He is not romantic about pain, poverty, exclusion, illness, or ignorance. He praises technology that heals and relieves suffering. But he rejects the post-human temptation to treat limitation itself as an error. A person is not a failed machine. Mortality, dependence, need, patience, growth, repentance, forgiveness: these are not bugs in the operating system. They are conditions of moral life.

The encyclical’s answer to transhumanism is therefore not nostalgia. It does not say, in effect, let us keep things human because we are fond of the old furniture. It says that the authentic “more than human” is not technical self-enlargement but grace. That claim may strike secular readers as theological, which of course it is, but its public force is wider than its doctrinal language. It says that human beings are not completed by self-sufficiency. They are completed by relation, by love, by truth, by responsibility, by the capacity to receive what they did not manufacture. This is a hard saying in an age where receiving is almost embarrassing and dependence has to be renamed collaboration before adults will admit to it.

Digital Rulers

The section on truth is among the strongest in the encyclical because it links democracy to the shared search for facts. Leo quotes Hannah Arendt on the subjects of totalitarian rule, those for whom the difference between fact and fiction and between true and false no longer exists. His quotes of Arendt have unusual force in a digital environment where falsehood no longer needs to win an argument. It only needs to exhaust the room. The lie of the older propagandist was often a wall. The falsehood of the algorithmic age is more like weather, not in the figurative sense of beauty or grandeur, but in the literal sense that it surrounds ordinary action and affects what people do next.

Leo’s point is that truth is not the possession of the powerful. It is a common good. This is a radical statement in a culture that has learned to treat attention as property and persuasion as extraction. Digital platforms do not merely transmit information; they help create culture. They rank, hide, recommend, amplify, monetize, and habituate. They train desire. They teach the nervous system what to expect. A public square is not free merely because everyone may shout inside it. A public square run by secret incentives can produce conformity while advertising expression. That is one of the comic achievements of the age: we have invented systems in which people feel wildly individual while repeating the same phrases, posing for the same camera, fearing the same invisibility, and refreshing the same little tribunal in the hand.

The encyclical’s demand for an “ecology of communication” is a demand for cultural sanitation without censorship as the organizing principle. Leo does not call for centralized control of truth. He calls for journalism, institutions, verification, schools, universities, families, and civic forums capable of resisting immediate reaction. This is unfashionable, which is to say it is probably necessary. It asks for slowness in a system built to punish slowness. It asks for fact checking in a system that rewards velocity. It asks for education in a system where many adults have become permanently available to interruption. It asks the young to become free while handing them devices engineered by people whose revenue rises when freedom weakens. Even the devil, one suspects, might admire the business model before declining to invest on reputational grounds.

Social Factors

The chapter on work returns the encyclical to the older terrain of Catholic social teaching, but under new conditions. Leo follows John Paul II in treating work not merely as income, but as a sphere of identity, responsibility, friendship, vocation, and social participation. This matters greatly in the AI age because the dominant public discussion of automation is often arithmetical. How many jobs lost? How many gained? Which tasks exposed? What percentage displaced? These questions are necessary, but insufficient. They treat work as allocation. Leo treats work as formation.

If work is only income, then a society can answer automation with payments, retraining vouchers, or new consumption rights. If work is also a place where the young learn responsibility, adults exercise skill, families gain stability, and citizens participate in common life, then job insecurity is not merely an economic inconvenience. It is a civic injury. Leo writes that unemployment and job insecurity damage family structures and block the human and professional fulfilment of young people. He calls for accessible education and retraining so that the digital economy does not become “a harsh selection between those who are able to update their skills and those who cannot.” Here Leo exposes the cruelty hidden in our cheerful talk of lifelong learning. Lifelong learning is noble when supported. It is vicious when used as a polite way of telling the abandoned that they failed to download the future in time.

Here the encyclical has a practical severity that many AI strategies lack. It asks that technological transition be judged by whether it sustains families, supports entry into work, protects rest, enables mobility, and prevents loneliness and addiction. These are not side issues. They are the human tests. A country can have advanced AI and a deteriorating society. A company can have brilliant automation and frightened workers. A school can adopt impressive tools and weaken attention. A family can possess every device and lose the habit of speaking. The civilization that calls this progress has confused motion with direction.

Freedom, in Leo’s account, is not only the absence of coercion. It is the capacity to remain inwardly available to truth, conscience, and love. That is why he treats addiction, attention capture, data extraction, profiling, and social control as related dangers. When platforms are designed to capture time and exploit vulnerability, he writes, the person is treated as a means rather than an end. His writing is Kantian in structure, Christian in depth, and prosecutorial in tone. It names the moral responsibility of those who design and finance systems that thrive on weakness.

This may be the place where Magnifica Humanitas most directly confronts the commercial logic of the digital age. We have become accustomed to calling predation “engagement.” We call engineered compulsion “retention.” We call surveillance “personalization.” We call the shaping of behavior “user experience.” There are days when the whole vocabulary of the technology industry sounds like it was invented by a committee of so called behavioral gurus who had attended a leadership retreat. Leo’s achievement is to return the thing to its proper name. If a system profits from the erosion of inner freedom, it cannot be acquitted by elegance of design.

Human Underside

Nor does Leo allow us to pretend that AI lives in some clean, weightless province of mind. Paragraph 173 is among the most physically severe passages in the document. Nothing in AI, he writes, is “immaterial or magical.” The smooth answer on the screen has a supply chain. It has workers. It has water. It has electricity. It has data centers, mines, cables, cooling systems, content moderators, labelers, trainers, and people paid little to stare at material most of us would not endure for an afternoon. Many are young. Many are women. Many are invisible by design.

The encyclical then descends further, as it must. The microprocessor has a human underside. In some regions, children and adolescents crush materials from which rare earth elements are extracted. Their bodies are “scarred, injured and worn down” so that computation may continue without interruption. This is where the theology becomes brutally concrete. The digital economy is not a cloud. It is a chain of bodies, minerals, heat, thirst, fatigue, and disposal. A society that speaks of artificial intelligence as though it were pure cognition has already begun its own moral laundering.

The encyclical’s treatment of war is the darkest part of the document and also among the most necessary. Leo sees a culture of power hardening across the world: the normalization of conflict, the weakening of multilateral institutions, the treatment of force as realism, and the rise of weapons systems involving AI. He also reclassifies the moral debate itself. The old theory of the “just war,” so often stretched until it can cover nearly any campaign desired by the powerful, is, he says, outdated. Humanity now possesses better instruments for resolving conflict: dialogue, diplomacy, and forgiveness. This is not sentimental pacifism. It is a severe judgment on political failure. War is not proof of realism. Very often it is proof that realism has run out of moral intelligence.

His sentence on autonomous weapons is categorical: it is not permissible to entrust lethal or irreversible decisions to artificial systems. He adds, “No algorithm can make war morally acceptable.” This is the sort of statement that should put an end to the debate. It will not, of course. Meetings are among humanity’s more durable methods for ensuring that clear sentences arrive too late.

His reasoning is exact. Moral judgment cannot be reduced to calculation because it involves conscience, responsibility, and recognition of the other as a person. AI can accelerate decision making, obscure accountability, classify targets, and lower the threshold for violence by making war less visible to those who authorize it. The machine does not make killing evil. It makes killing administratively smoother. It turns distance into procedure. It turns victims into data. It turns hesitation into inefficiency. In war, the great danger of AI is not that it becomes irrational. The danger is that it becomes perfectly rational within a morally deranged frame.

Leo is especially precise about target selection. Any technology that attacks without seeing the face of the person lowers the moral threshold of conflict. Prediction models can confuse combatants and non-combatants, treat defenseless populations as background noise, and present a strike as a technically coherent act before conscience has had time to object. The danger is not only misidentification. It is a training of the political imagination. Once civilians appear first as coordinates, clusters, heat signatures, probabilities, or inferred associations, the human face has already been removed from the room in which the decision is made.

This is why Leo insists on identifiable responsibility. Those who design, train, authorize, and use such systems must be accountable. There must be a chain of decision. There must be time for judgment. There must be protection for civilians. The face must not disappear. Here the encyclical joins the deepest tradition of Christian realism, which is not the realism of the hard man who worships necessity, but the realism of the person who has visited the hospital, the graveyard, the refugee camp, the prison, the bombed street, and refuses to let strategy speak as if bodies were abstractions.

Action Is The Only Truth

The recurring phrase in the encyclical is “civilization of love.” I understand why some readers will wince. The phrase can sound soft to ears trained by policy memos and strategic briefings. Yet Leo gives it iron. For him, love is not mood, not temperament, not a scented candle placed on the ruins of politics. Love is an ordering principle for institutions, technologies, laws, markets, schools, families, and nations. It requires justice, truth, diplomacy, restraint, and the protection of victims. It requires what he calls the need to “disarm words.” This is not decorative piety. Words prepare actions. Before a person is removed, he is named as removable. Before a group is attacked, it is described as dangerous, impure, parasitic, backward, subhuman, or in the way. Public cruelty begins as verbal permission.

This is why I think the encyclical is not only about artificial intelligence. It is about the power systems of late modern life and the temptation to make the person smaller than the system that manages him. AI is the occasion, not the whole disease. The disease is older: the will to dominate without admitting domination, to classify without meeting, to optimize without loving, to govern without being answerable, to speak of humanity while losing patience with actual people. Artificial intelligence gives this old temptation new instruments and astonishing reach. The tower rises faster now. It has better funding. It has a communications department.

Yet Magnifica Humanitas is not despairing. Its hope is disciplined. Leo quotes Tolkien:

“It is not our part to master all the tides of the world, but to do what is in us for the succour of those years wherein we are set.”

The quotation works because it refuses both omnipotence and resignation. We do not master all tides. We do not therefore do nothing. We have our section of the wall. The legislator has one. The engineer has one. The teacher has one. The parent has one. The journalist has one. The bishop has one. The board member has one. The student has one. The tired citizen, tempted hourly by cynicism, has one too.

This is the encyclical’s great civic gift: it restores scale to responsibility. We are not asked to save the world by gesture. We are asked to refuse dehumanization where our hands actually touch the work. The engineer can ask whether the system permits appeal. The manager can ask whether efficiency has become an alibi for fear. The teacher can defend attention. The parent can defend childhood. The journalist can defend verification. The legislator can defend transparency. The investor can defend limits. The citizen can defend speech that does not poison the common air. The believer can pray, but also build, vote, teach, organize, restrain, repair, and apologize.

The conclusion returns to the Eucharist and the body. This is not an escape from the technological age into pious interiority. It is the encyclical’s final act of resistance. Against systems that isolate, score, sort, and monetize, Leo places a body made of many members. Against simulated care, he places communion. Against the dream of self-creation, he places gift. Against Babel, he places a construction site already at work, with “living stones” joined to God the cornerstone. The image is concrete enough to save the ending from abstraction. There is dust, weight, sequence, labor. Someone lifts. Someone carries. Someone steadies. Someone returns after injury. Someone learns the skill slowly. Someone who was overlooked is needed for the next course of stone.

I find myself returning to Nehemiah walking the broken city at night before speaking to the people. He does not begin with a theory. He inspects the damage. That is where serious moral work begins. Not in the slogan. Not in the launch event. Not in the panel discussion where everyone agrees that ethics is important before hurrying away to ignore it with renewed sophistication. He looks at the burned gates and the broken walls. Then the people take their places.

That is where Leo leaves us, and it is the right place. Not above the world, adjudicating it from clean distance. Not under the world, crushed by its machines. In the city, at the damaged section, with tools in hand, close enough to hear the neighbor breathing, close enough to know which stone will hold.

Stay curious

Colin

The Guess

In 11 CE, Augustus prohibited private consultations with astrologers in Rome and banned questions about anyone’s death. The decree was not a literary gesture. It was a measure of state security, issued by a man who had earlier benefited from the very practice he now feared. Augustus had used astrology to dignify his own ascent. Once power was his, the same instrument looked less like destiny and more like evidence for the next claimant. Earlier, Tiberius had tested astrologers on Rhodes. Those who failed could be thrown from a cliff. One astrologer, Thrasyllus, survived because, when asked to read his own prospects, he answered that he was in grave danger. Tiberius accepted the answer and kept him.

Carissa Véliz’s book, Prophecy begins its real work, not with artificial intelligence, but with the old rooms in which frightened rulers paid clever men to speak about tomorrow. The book is not about fortune-telling as a quaint human weakness. It is about prediction as a political act. It asks who gets to speak about the future, who must live under that speech, and why the most useful alibi of prophecy is its claim to be only looking for truth.

I think this is the right question for our age because we have been trained to ask the wrong one. We ask whether the prediction is accurate. Véliz asks a more dangerous question: what does the prediction do? A forecast is rarely idle. It rearranges expectation. It directs attention. It gives permission. It frightens, licenses, excludes, markets, disciplines, hires, denies, triages, prices, and ranks. It may look like a statement, but it often behaves like an instruction.

Prediction is Indispensable

That is the unsettling force of the book. Prediction is not merely a technical practice that went astray when engineers found too much data and too little humility. It is one of the oldest forms of command. Its costume changes. The priestess at Delphi gives way to the astrologer, the astrologer to the social statistician, the statistician to the machine-learning system, the machine-learning system to the executive on a conference stage announcing that software will soon perform the work of whole professions. The voice changes. The social operation remains recognizable.

The old prophet had smoke, stone, ritual, and delay. The new prophet has cloud infrastructure, venture capital, government contracts, slides, and a supply of human data acquired with an innocence that rarely survives inspection. Modern predictive power did not arrive clean. It was built through surveillance, scraped lives, weak consent, and the quiet exploitation of people too tired, poor, sick, lonely, or distracted to resist becoming inputs. I do not say this to sneer at technology. That would be too easy, and also false. Prediction is indispensable. We predict weather, epidemics, troop movements, credit risk, delivery times, hospital admissions, energy demand, crop yield, and the probable behavior of strangers approaching us on a dark street. A creature that could not anticipate would not survive long enough to write philosophy, let alone ignore it.

But Véliz’s point is sharper than the usual warning against overconfidence. Predictions are not facts. The past can be evidenced. The present can be inspected. The future has not happened. A claim about it may be disciplined, calibrated, useful, responsible, and still remain a guess. This is not a small philosophical housekeeping task. It changes the moral status of many decisions now treated as administrative routine. If a person is denied a job because an algorithm scores them as risky, they are not being judged on what they have done. They are being governed by what someone, or something, expects them to do.

That small shift is the door through which much of the modern world now passes. The applicant does not meet the employer. The tenant does not meet the landlord. The patient does not meet the person who set the triage threshold. The defendant does not see the full machinery that helped assign risk. A score appears. A door closes. In medicine, the violence of this can be immediate. A forecast that a patient is unlikely to survive can help make survival less likely, because scarce care is redirected elsewhere. Everyone involved can sound reasonable, even bored. The system did not say the person should die. It said resources should follow probability. Civilized cruelty has always liked a passive verb.

Five Traits of Prediction

Véliz insists that predictions have five traits the naive view tends to miss. They are guesses. They are shaped by desire. They are entangled with power. Some are impossible. Some are harmful. The list looks simple until one notices how much of institutional life depends on pretending it is not true.

The desire inside prediction deserves special attention. The fantasy of modern forecasting is neutrality. We imagine that once a procedure becomes computational, it has escaped appetite. Yet many predictions are made by institutions that already know what future they want. A social media company predicts engagement because it wants engagement. A lender predicts default because it wants profit and protection from loss. A police department predicts crime because it wants control, budget, legitimacy, and the public performance of vigilance. A technology chief predicts mass automation because the market often rewards a prophet before history has had a chance to embarrass him.

The machine may not have ambition, but the institution does. The code may not crave, but the firm can. The model may not boast, but the founder can appear in a black T-shirt, or a fleece vest, or some other garment of secular priesthood, and explain to presidents, prime ministers, and investors that the future has already chosen his product. At that point we are not listening to science alone. We are listening to salesmanship with a server farm.

Here the book’s ancient stories do more than decorate the argument. They discipline it. Alexander of Abonoteichus, the second-century prophet with his speaking snake Glycon, is not simply a comic episode from the museum of credulity. He is a business model. He staged revelation, gathered questions, manipulated seals, destroyed records of failed prophecies, rewrote his own archive, and sold dread back to the frightened. Lucian exposed him. The oldest lesson is not that people in the past were foolish. It is that people under uncertainty are vulnerable, especially when fear and hope arrive together.

I find that almost tender, though not comforting. Human beings are not ridiculous because they want to know what will happen. They are exposed. A parent waiting in an emergency room wants a forecast. A farmer watching the sky wants a forecast. A founder signing a lease wants a forecast. A government facing war wants a forecast. A woman wondering whether the man beside her will become dangerous wants a forecast. Prediction begins in prudence before it becomes an industry.

That is why bad prophecy is so powerful. It takes one of our real strengths, the ability to anticipate, and turns it into a system of dependence. Véliz is especially good on this moral inversion. The more anxious the public becomes, the more attractive the prophet becomes. The more attractive the prophet becomes, the more authority prophecy gains. The more authority prophecy gains, the more the future is treated as something already allocated by those who claim to see it.

And then the future starts obeying.

Social Predictions

This is where the book moves from epistemology into politics. A prediction about rain does not cause rain. A prediction about a bank run may help produce one. A prediction that a neighborhood is declining may withdraw credit, insurance, investment, care, and attention until the neighborhood declines. A prediction that a child is unlikely to succeed may alter teaching, expectation, discipline, and patience. A prediction that a person is dangerous may change how police approach him, how judges sentence him, how employers reject him, and how he learns to see himself after every ordinary path has been narrowed.

The prophecy calls itself descriptive. It can become prescriptive. It tells others how to act. In the process, it steals from the person judged by it that fragile political commodity without which citizenship becomes theatre: the chance to surprise.

This, I think, is the central democratic argument in Véliz’s book. Democracy requires uncertainty. Not ignorance, not stupidity, not romantic anti-science, but real openness about human action. An election whose outcome is fixed is not democratic. A trial whose verdict precedes the evidence is not justice. A school system that grades the child before the child has struggled is not education. A labor market that screens out the apprentice because the apprentice lacks the profile of someone already formed is not meritocracy. It is aristocracy with better data hygiene.

The fight over prediction is therefore not a quarrel between poets and engineers. It is a quarrel over moral jurisdiction. Who has authority over the unfinished person? Who may convert probability into treatment? Who profits when uncertainty is removed from the lives of the weak and preserved for the strategies of the strong?

Ancient rulers understood something we have forgotten. They feared astrologers because a prophecy could disturb power. Today, many prophets face no equivalent danger. If a model harms a person through a bad prediction, the person may suffer while the institution explains that the procedure complied with policy. The ancient seer risked the cliff. The modern scoring system risks a quarterly review.

I am not recommending cliffs. I am recommending responsibility. That is apparently the extreme position now.

Agency

There is also a more intimate argument in the book, and it may be the more radical one. Véliz does not merely ask how states and companies should regulate predictions. She asks how a person should live when surrounded by systems that want to forecast, classify, and nudge them. Her answer is not a retreat into ignorance. It is a recovery of agency through curiosity, courage, good judgment, and refusal.

Refusal is most important when prediction becomes a seduction. A personalized feed says: you are the kind of person who wants this. A recommendation engine says: you are the kind of person who will watch that. A dating system says: these are the people available to your measured self. A career platform says: these are the roles likely to accept you. The convenience is real. So is the narrowing. One begins with suggestions and ends with a life pre-sorted by a thousand little oracles, each very helpful, each asking only for one more piece of the person.

This is why surveillance and prediction belong together. Surveillance supplies the raw material. Prediction supplies the claim of necessity. Together they create a machine for reducing people to inferred futures. Véliz calls surveillance and prediction digital technology’s original sins. I think the phrase is right because it restores moral seriousness to a field that often hides behind product language. A world in which every action is recorded so that every future action can be guessed is not merely efficient. It is a world in which freedom must ask permission from analytics.

The defenders of such systems will say the old methods were worse. They will be partly right. Human judgment has always been biased, lazy, tribal, vain, and frequently corrupt. The pre-algorithmic world was not a republic of angels making fair decisions by candlelight. It was often a shabby arrangement of prejudice, paperwork, and lunch. But that is not a defense of automated prophecy. It is a warning. If injustice existed before measurement, measurement may harden it, accelerate it, conceal it, and make appeal more difficult.

The new oracle is not terrifying because it is always wrong. It is terrifying because it is sometimes right in ways that strengthen its authority when it is wrong. A model that predicts enough successes earns the right, in bureaucratic practice, to injure the exceptional case. The individual becomes an acceptable remainder. The machine does not need to hate him. Hatred would almost be reassuring; at least hatred has a face. Statistical disregard is cleaner. It can deny the harm while completing it.

This is where Véliz’s defense of the unforeseeable becomes more than a comforting humanist gesture. The unforeseeable is not a defect in the administrative order. It is one condition of liberty. The unexpected mind, the late vocation, the changed life, the sudden discipline, the comic rescue, the act of courage no one budgeted for, the apprentice who becomes the master, the failing city that rebuilds, the frightened citizen who says no at the crucial hour, all of these depend on a gap between probability and reality. Prediction is most dangerous when it tries to close that gap before the person has had a chance to act.

Prediction hates that gap. Power hates it even more.

Yet the book is not anti-prediction. That would be childish, and Véliz is not childish. She wants a wiser practice of forecasting, one alert to method, legitimacy, institutional incentives, and harm. Some predictions are necessary. Some are merciful. A weather warning can save a town. A medical prognosis can help a family prepare. A supply forecast can prevent waste. A public health model can buy time. The question is not whether to predict. The question is whether the prediction respects the human being who must live under its consequences.

Explainability

There should be rules for this. Not every future deserves to be guessed in public. Not every guess deserves administrative force. Not every score deserves obedience. A prediction used against a person should carry duties: explainability where possible, contestability where necessary, proportionality always, and accountability when harm follows. A society that can audit its restaurants more seriously than its predictive systems has chosen its stomach over its citizens, which is at least consistent, if not noble.

The book’s philosophical turn is therefore practical. Truth, virtue, and beauty are not ornamental alternatives to prediction. They are restraints upon it. Truth reminds us not to sell guesses as facts. Virtue asks what kind of people we become when we govern others by forecasts. Beauty, at its best, resists reduction; it preserves the human appetite for what cannot be priced, scored, optimized, or made useful by Friday. I know that sounds unfashionable. Perhaps. Some unfashionable things have survived because they were right.

Véliz’s criticism of utilitarian and effective-altruist styles of reasoning follows from this. Any ethics that leans too heavily on predicted consequences risks becoming prophecy with better manners. It may begin with generosity and end with a spreadsheet deciding whose suffering has the highest strategic value. Consequences count. Of course they do. But when consequence prediction becomes the whole of ethics, principles become inconveniences, persons become units, and uncertainty becomes something to be conquered rather than respected.

This does not mean we should abandon calculation. It means calculation needs companions it cannot dominate: rights, duties, character, judgment, humility, humor, and the stubborn dignity of the particular case. A civilization that cannot say, “Even if your model is right, you may not do this,” has already mistaken power for wisdom.

Curiosity

The most appealing passages in Véliz’s argument are those that treat curiosity as a civic virtue. Curiosity is not the same as data hunger. Data hunger wants to capture, infer, retain, compare, and sell. Curiosity wants to meet without owning. It accepts the other person as unfinished, and therefore not yet available for final sorting. The difference is not soft. It is political. A curious society leaves room for encounter, accident, revision, and play. A predictive society prefers profiles. The curious person asks what might be learned. The prophet asks what can be inferred. One opens a conversation. The other closes a file.

There is a reason humor keeps appearing in the book. Humor is one of the mind’s ways of defeating false inevitability. A joke breaks the expected path. It turns the solemn machinery of prediction against itself. The tyrant and the fraud both dislike laughter, especially laughter that arrives at the exact instant their authority requires silence. Lucian understood this when he went after Alexander and Glycon. Satire was not a decorative weapon. It was a civic instrument, sharpened by ridicule and aimed at false awe.

I think this is why the book’s argument finally feels bracing rather than bleak. It does not ask us to become less intelligent, less technical, less ambitious, or less prepared. It asks us to become less obedient to the people who profit from making tomorrow sound settled. It asks us to notice the moment when a forecast begins to behave like an order. It asks us to protect the human right to become unlikely, which is not a sentimental right at all, but the civic form of the unforeseeable.

That right will not protect itself. It will require law, institutional design, professional standards, public pressure, and private abstinence from systems that feed on our predictability. It will require leaders who understand that a model is not a moral agent. It will require companies that can distinguish customer service from behavioral capture. It will require universities brave enough to teach uncertainty not as embarrassment but as intellectual honesty. It will require citizens who do not hand over their biographies simply because the interface is smooth.

And it will require writers. Lucian did not defeat Alexander by building a better snake. He wrote. He recorded the trick. He made the prophet visible as a man with props, assistants, records, and a business plan. In every age, one duty of prose is to remove the incense from power.

I finished Véliz’s book thinking not of Delphi, or Augustus, or the algorithmic machinery humming behind ordinary transactions, but of a young person before a screen being told, gently, almost efficiently, what kind of future is probable for them. The humane answer is not to smash the screen. The humane answer is to make sure that somewhere nearby there is a teacher, a parent, a mentor, a friend, a library, a joke, a blank page, and a door left open for the life no one predicted.

Stay curious

Colin

I explore how the historical manufacture of scientific doubt used by the tobacco and climate industries has evolved into a modern bureaucratization of risk within artificial intelligence development. I argue for an adult politics that distinguishes between corrupted doubt intended to delay oversight and useful doubt that establishes clear lines of institutional responsibility.

In June, 2010, Naomi Oreskes and Erik M. Conway published Merchants of Doubt. The book followed a group of prominent scientists, several with Cold War credentials, who intervened in public disputes over tobacco smoke, acid rain, the ozone layer, DDT (a synthetic insecticide), and climate change. Their method was not always to prove the opposite case. It was often enough to prevent the existing case from becoming usable. The approach was simple. Ask for another study. Question the sample. Call the models uncertain. Praise better evidence as a civic duty. Describe regulation as premature. Hint that the experts are not of one mind. Almost any phrasing will do, provided it buys time.

“‘Doubt is our product,’ ran the infamous memo written by one tobacco industry executive in 1969, ‘since it is the best means of competing with the ‘body of fact’ that exists in the minds of the general public.’”

Time was the product.

That is the part of the book I cannot stop thinking about now. Not the villainy, though there is plenty of that. Not the think tanks, the memoranda, the expert letters, the solemn men talking about scientific freedom while lung cancer moved through the population, in other words general society who would not make the time or effort to read the small print and footnotes. What stays with me is the discovery that doubt can be manufactured with the same discipline as steel. It can be funded, distributed, franchised, laundered through respectable institutions, and placed in front of the public with an expression of injured seriousness.

And Then Comes Hassabis

Artificial intelligence has brought the word back into circulation, but not in the old form. That is why the analogy is tempting and treacherous. The tobacco companies sold doubt after enough was already known. The AI argument is happening while the machine is still under construction, still being tuned, still being praised by investors, still being feared by its own inventors, still being handed to children doing homework, doctors checking notes, soldiers reading briefings, and people at kitchen tables who have begun to treat a machine’s patience as a kind of kindness. No one is standing over a deceased person. The decision, or lack of one, is being made while the thing itself is still changing.

So I read Oreskes and Conway with a pencil in one hand and a certain unease in the other. Geoffrey Hinton warns of catastrophic risk. Yoshua Bengio calls for stronger governance, independent evaluation, and liability. Max Tegmark asks whether we are building powers we cannot control. Dario Amodei, while running one of the companies pushing the field forward, writes that the risks may be the only obstacle between us and a future of extraordinary human benefit. Demis Hassabis, meanwhile, can point not to a promise but to a result: AlphaFold, protein structures, biology accelerated, the long patience of science suddenly assisted by a machine that did not need a grant extension, a sabbatical, or a second cup of coffee.

This is not a clean courtroom. It is not the cigarette executive versus the epidemiologist. It is stranger than that, and more morally awkward. The same field has produced the warning and the cure, the alarm bell and the instrument, the engineer who says slow down and the scientist who says look what we can now do. The age is not asking us to choose between reason and delusion. That would be easy, and public life no longer grants such luxuries. It is asking us to distinguish between different kinds of doubt before the consequences have finished arriving.

The scientists in Merchants of Doubt are a useful comparison. One can look at Hinton, Bengio, Tegmark, and the other AI pessimists and say: there they are, the new obstructionists, the men of delay, the professional worriers standing in the doorway of abundance. The accusation has an appealing briskness. It makes the accelerator feel brave. It allows the product launch to stand in for moral seriousness. It turns caution into a personality defect. One can almost hear the impatient AI executive saying, with the smoothness of a man who has never had to answer for a downstream consequence, that humanity cannot be allowed to miss the future because a few academics have become gloomy.

I do not buy it.

Hinton is not a retired village scold objecting to the railway because the horses will be embarrassed. Bengio is not defending candlelight against electricity. These are not men who woke one morning and discovered, to their horror, that young people were using tools. They built the intellectual foundation on which the present industry stands. Their warnings come with the authority, and the discomfort, of proximity. It is one thing to be denounced by an enemy. It is quite another to hear the alarm from the person who helped design the product.

Hinton’s numbers have become famous because they are so resistant to ordinary mental filing. A ten to twenty percent chance of catastrophic outcomes. What is one to do with such a figure? Too low to cancel the build too high to place beside the canapés. If a surgeon gave those odds before an elective procedure, the patient would not ask whether the operating theatre had a strong innovation culture. He would put his trousers back on. Yet in AI we have learned to discuss such probabilities with the poise of men comparing airport lounges.

Bengio’s case is less theatrical and for that reason more useful. He asks for evaluations, liability, public interest research, and governance that can survive contact with firms whose incentives are not identical to civilization’s. There is no melodrama in this. It is almost dull. That is its strength. He is not asking us to worship fear. He is asking us to give responsibility an address.

Tegmark presses the harder question, the one that makes practical people suddenly reach for procedural language. If systems become more capable than their makers in domains that affect power, biology, persuasion, cyber operations, or autonomous planning, what exactly will keep them within human purposes? Not vibes. Not brand values. Not a launch blog with soft lighting. Not a safety card written by the same institution whose valuation depends on being first. The question may be overstated in some tellings, but it is not stupid. The future has an unfortunate habit of punishing people who confuse discomfort with irrelevance.

The Warning and the Cure

And then comes Hassabis, ruining the simplicity of the indictment.

AlphaFold is not a marketing adjective. It is one of the great scientific achievements of the last decade. To predict protein structures at scale is not to add a feature to a subscription plan. It is to solve a hard scientific problem in public, with results other scientists can use. The 2024 Nobel recognition of Hassabis and John Jumper, alongside David Baker, made the point harder to evade. AI was not only producing bogus citations and helping consultants rename their slide decks. It was helping science do science.

The old anti-technology reflex becomes unserious. I have no patience for the easy sneer that treats every claim of AI benefit as public relations. Some of it is. Much of it is. Perhaps an indecent amount of it is. But not all. When a protein structure prediction changes what a laboratory can attempt next week, the promise has left the brochure. It has entered someone’s experiment. That does not make the surrounding industry virtuous. It makes the governance problem harder, because the benefit is no longer imaginary.

Shakespeare was the master writer on doubt and uncertainty. His tragedies are full of people trapped between partial knowledge and irreversible action. Hamlet knows enough to be sickened, not enough to be free. Macbeth receives a forecast and mistakes it for authorization. Othello demands certainty and accepts counterfeit certainty from Iago, a man who understands that suggestion can do the work of proof. Lear asks for love in the wrong format and discovers, rather late, that metrics are not always measurement. One can imagine the old king in a board meeting, demanding sentiment scores from Cordelia.

AI places us in this Shakespearean interval. We know enough to be uneasy. We do not know enough to be finished. The systems work, but not always reliably. They reason, or appear to reason, or perform something near enough to reasoning to make the difference unclear. They help programmers and mislead auditors. They summarize documents and invent sources with the same helpful posture. A chemist can use them. So can a cyber attacker. A lonely person may feel comforted by one, which is not the same as being cared for. A manager may call them productivity tools while using them to watch workers more closely. The difficulty is not that the claims are false. The difficulty is that so many of them are true in opposite directions.

The old merchants of doubt understood this condition perfectly. They knew that democracy needs argument, and they turned that need into a delaying tactic. They did not have to persuade the public that smoking was healthy. They merely had to persuade enough people that the question was not ready for decision. Delay became policy. Uncertainty became profit. Scientific modesty became a hiding place for commercial power.

Two Kinds of Doubt

The line I would draw now: useful doubt assigns responsibility; corrupted doubt disperses it. Useful doubt asks for an evidentiary threshold, a testing authority, a veto, and a name on the invoice when the system fails. Corrupted doubt avoids such particulars. It multiplies venues until the original question loses force.

That is the danger now, but it comes by a less obvious path. The danger is not simply that AI companies will deny risk. Some will do that. Some will do it with impressive sincerity, which is always the most efficient kind. The deeper danger is that uncertainty itself will be administered by those with the strongest interest in speed. The public will be told, usually in patient tones, that judgment requires access it does not have. The model is complicated. The weights are private. The race is international against China. The opportunity is large. The documentation is coming. The independent audit is being discussed. Nothing in that sequence is necessarily false. That is precisely the problem. This is not the denial of science. It is the capture of uncertainty.

Captured uncertainty is harder to confront because it never presents itself as evasion. It arrives as prudence, process, further review. Captured uncertainty keeps changing rooms. The old merchants practiced external doubt. They sent confusion outward, toward the citizen, the voter, the journalist, the legislator. The new form is more intimate. It turns uncertainty into procedure. The risk is not denied. It is routed.

A question that begins in public soon acquires a room, then a review process, then an advisory board, then a policy team trained in the art of grammatical evaporation. By the time anyone asks who decided, the answer has become a calendar invitation with twelve attendees and no owner. This is not the manufacture of ignorance in the old style. It is the bureaucratization of risk.

Organized Postponement

This is why the old book still bites. Oreskes and Conway were not merely writing about bad arguments. They were writing about the political uses of delay. Tobacco, acid rain, the ozone layer, climate change: in each case, the same pattern appeared. Enough doubt was kept alive to make collective action look premature. The result was not intellectual humility. It was organized postponement.

AI requires something more difficult than denunciation. It requires us to save doubt from the merchants who would exploit it and from the enthusiasts who would abolish it. Doubt is not the enemy of progress. Doubt is one of the ways progress remains answerable. A scientist without doubt is a salesman with a lab coat. A regulator without doubt is a clerk with a stamp. A founder without doubt is, depending on the funding round, either a visionary or a public hazard.

The bad doubt of the old merchants did the opposite. It removed responsibility from reach. It asked for perfect proof where practical proof already existed. It used scientific modesty as a hiding place. Today the hiding place is technical complexity. The citizen is told that the system is too complicated to judge, the model too opaque to inspect, the competitive stakes too high to disclose, the risk too specialized for democratic argument. This is not skepticism. It is anesthesia with better vocabulary.

The same anesthesia is available now in finer packaging. The public hears that alignment has improved. It hears that evaluations are under way. It hears that the company takes safety seriously, which is the sort of sentence that should make a citizen contact their Member of Parliament or Congress. Some of these claims may be true. That is what makes them useful. The most durable evasions are rarely pure falsehoods. They are truths arranged so that no one has to stop.

Dario Amodei’s phrase about risk standing between us and a deeply positive future is therefore worth taking seriously. It does not fit the cartoon. His argument is not that AI is a monster waiting to be named, nor that salvation will arrive by subscription. It is closer to the older scientific temperament: the promise is large enough to make care obligatory. In his version, optimism is not the denial of danger. It is the refusal to let danger disfigure the promise before the public has had a chance to benefit from it.

That position is more interesting than either corporate triumphalism or professional despair. Amodei writes from inside the machine room, not from a monastery. He knows that a model can help a laboratory move from guesswork to experiment faster than before. He also knows that the same kind of acceleration, once tied to money or power, has a nasty habit of turning public goods into private weapons. This is the bridge between Hinton’s alarm and Hassabis’s achievement. The serious optimist does not say, trust us. He says, build the institutions that would make trust less necessary.

The Fox

Humanity has invented machines that can help fold proteins, draft legal arguments, and keep a lonely person talking long after a human friend would have gone to bed. We have then asked the public to trust safety statements issued by the institutions most eager to deploy them. The fox is no longer guarding the henhouse. The fox has published a governance framework and invited the hens to a webinar.

Still, sneering is too easy. One must not become the kind of critic who mistakes contempt for judgment. Hassabis and the AlphaFold story force a more honest discipline. The good is real. The risks are real. The institutional weakness is real. The question is whether our public systems can grow quickly enough to hold together those three facts without sacrificing one to the convenience of the others.

What would that require? Not a sermon against machines. Not a festival of new acronyms. Not another declaration of principles printed on expensive paper and forgotten before the coffee cools. Start with one unfashionable rule: no institution should be allowed to deploy a high-stakes AI system that no outside authority can meaningfully inspect. A hospital should not be asked to trust a vendor’s private reassurance when a model influences diagnosis. A court should not accept opacity as the price of efficiency. A school should not buy a tutoring system whose failures become visible only in the child.

Responsibility needs machinery. Without machinery, every ethical commitment becomes atmospheric. It hangs above the enterprise, fragrant and useless, while the product ships.

The old merchants of doubt succeeded because public institutions were slower than private incentives. That asymmetry has not disappeared. A company can release an AI model on Tuesday. A regulator may still be trying to define the relevant harm the following spring. In between those two dates, habits form. Workflows harden. Contracts renew. The system becomes ordinary before it becomes understood.

The cost often first appears as a small administrative fact. A benefits office accepts the model’s recommendation. A caseworker, busy and underpaid, does not challenge it. The rejected applicant receives an AI generated formal explanation that explains nothing. No one has lied. No one has intended cruelty. Responsibility has simply passed through too many hands to leave fingerprints.

Doubt as Loyalty

I think that is the real terror behind the AI debate. Not that the machines will suddenly become demons. Demons are at least legible. The terror is that responsibility will become increasingly difficult to locate. The AI model suggests, the manager accepts, the vendor updates the documentation, the regulator asks for more information, and the user is said to have consented somewhere along the way. When the harm appears, everyone has participated just enough to deny authorship.

Against that, the serious doubters are not merchants. They are witnesses. They may be wrong in their estimates. But they are trying to force the future into answerable form. That is the civic value of their doubt. It does not ask us to stop thinking. It asks us to stop pretending that acceleration is a substitute for thought.

Shakespeare would have recognized the temptation. His characters are forever being offered shortcuts through uncertainty: prophecy, rumor, flattery, forged evidence, theatrical proof. The shortcut always promises relief. It says: act now, the meaning is already clear. In AI, the shortcut is inevitability. The machine will come anyway. Other countries will build it. The public will adapt. This is not an argument. It is a lullaby sung by a shareholder.

I do not want an anti-AI politics. It would be both foolish and impossible. I want an adult politics of AI, which is rarer and less photogenic. It should be capable of admiring AlphaFold without handing AI model-release policy to private firms as a form of gratitude. It should be capable of wanting scientific acceleration while asking a rude, necessary question: who is allowed to say no? Hope that cannot survive that question is not hope.

The old merchants taught us that doubt can be corrupted. The AI scientists warning us now teach something just as important: doubt can also be an act of loyalty. Loyalty to science, because science is not the same as industry. Loyalty to the public, because the public will live with consequences it did not vote on. Loyalty to the future, because a future worth having must be more than the fastest available version of itself.

So I return to Oreskes and Conway for discipline: follow doubt until it reaches either a decision or a hiding place. The necessary question is not whether the person invoking uncertainty sounds reasonable. It is whether he is preparing society to act wisely, or merely asking society to wait while the profitable thing becomes irreversible.

The better ending is not a ban or a blessing. It is a hand on the switch, and a person who knows that not every working machine is ready to be released. Somewhere, before the announcement, before the congratulatory note from the chief executive, before the word breakthrough begins its little career in the headlines, someone asks for the test again. Not because he hates the future. Because he intends to let one arrive.

Stay curious

Colin

Image: Tim Mossholder on Unsplash

Part of my series on Cognitive Warfare.

NATO and the Pentagon have formally classified your attention as a theater of war. This is what that means.

In early 2026, NATO finalized a doctrine that does something radical: it reclassifies the civilian mind as primary terrain for attack. This is not a metaphor; it is an administrative shift. According to a research bulletin from the Finnish Defence Research Agency (FDRA), more than twenty countries have now moved to treat the human brain as both the “target and the weapon” in a fight for cognitive superiority. The bulletin notes that technological development, global interconnection, and social media have made it possible to influence attitudes, decision making, and behavior at scale.

These operations target memory, attention, perception, and emotion, not the flow of information, but the conditions under which information is processed. They are designed to occur below the threshold of armed conflict, disrupting cohesion without a shot fired.

In late March 2026, the United States Strategic Capabilities Office began a program described as cognitive warfare. Public reporting indicates that the initiative is designed for continuous competition, not episodic conflict, and aims to operate at the speed and persistence of digital systems. Officials describe a competition over influence rather than territory, where the human brain is treated as a primary theater of operations alongside land, sea, and air.

Not Propaganda

I think the first mistake is to treat this as an extension of propaganda. Propaganda assumes a sender, a message, and a receiver. It assumes a surface. What these documents describe is an attempt to operate below that surface, to alter the conditions under which a message is even recognized as a message. The Finnish bulletin is explicit that the aim is not simply to control information, but to control “how the target population reacts to presented information.”

During the twentieth century, states competed over territory, resources, and occasionally ideology. Even psychological operations retained a sense of boundary. They were aimed at soldiers, or at populations in wartime, or at clearly identified adversaries. The current doctrine removes that boundary. It defines the civilian mind as the primary terrain. The target set is no longer a city or an army, but a pattern of attention distributed across millions, possibly billions, of individuals, continuously monitored and continuously adjusted.

This is why the language in these reports becomes strangely clinical. They speak of “cognitive processes,” “behavioral effects,” and “narrative saturation.” One paper describes flooding the information environment with emotionally charged content designed to blur the distinction between reality and fiction.

It is difficult not to notice that this description also functions as a description of the current media ecosystem. The weapon has already been built. What is new is the integration of artificial intelligence and large-scale data harvesting to detect fractures in society before they are visible, and to act on them before they stabilize into shared awareness.

Disorder

If the goal is to influence how populations interpret information, then success cannot be measured by agreement, but by responsiveness. A population that argues intensely may be functioning exactly as intended. Disorder becomes a signal of operational effectiveness. More precisely, epistemic fragmentation, the erosion of any shared baseline of reality, becomes the condition to be engineered. The metric of victory is not belief, but agitation.

The American program sharpens this further. When officials speak about shaping the “decision environment,” they are acknowledging that the most valuable outcome is not a specific decision, but a predictable pattern of decisions. The ambition is to narrow the range of possible responses until the outcome becomes statistically reliable. This is not persuasion. It is closer to conditioning, except that the subjects remain convinced of their autonomy.

The NATO concept reaches its most serious implication when it describes cognitive warfare as a “whole of society” problem. This is presented as a defensive necessity. If the entire population is a target, then the entire population must participate in defense. Education, media literacy, institutional trust, and social cohesion are no longer merely social virtues. They are classified, implicitly and sometimes explicitly, as national security assets. The distinction between civilian life and strategic function begins to dissolve.

Where the Boundary Collapses

Total war in the twentieth century required the mobilization of industry, labor, and culture. Cognitive warfare requires the mobilization of attention. It is less visible, but more intimate. The factory demanded your body. This system asks for your habits of thought, and it does so continuously, without pause, without the formal declaration that would signal the beginning or the end of conflict.

It is likely that the real difficulty is not technical, but philosophical. If the mind is treated as a domain of operations, then every belief becomes a potential vulnerability. Trust becomes a resource to be protected, but also something that can be manipulated. The Finnish report introduces a further complication. It notes that doctrine includes proactive actions to shape the information environment. This is where the boundary collapses. If defense requires shaping perception in advance, then the defender adopts the same operational posture as the adversary. The line does not disappear entirely, but it becomes contingent, argued, and unstable.

What does it mean to defend a mind without directing it? The documents do not answer this. They speak of resilience, but resilience to what, and defined by whom. They speak of superiority, but superiority over which alternatives. The language is precise about mechanisms and vague about ends.

Recognition And Defence

The documents speak of resilience without specifying its architecture. But architecture is precisely what is needed. The conventional answer, media literacy, is insufficient here because cognitive warfare does not primarily operate through false content. It operates through attention management, the quiet direction of what you notice, when you notice it, and how readily it triggers a reaction. What is needed is something one level up: the ability to recognize when your attention is being steered, not just to evaluate what you are being shown. This is a teachable skill, and it is distinct from anything currently offered in most curricula. Beyond the individual, certain institutions, courts, scientific bodies, electoral systems, function as shared reality checkpoints. They are targeted by cognitive operations precisely because they resist agitation and provide a common ground for adjudicating disagreement. Defending them is not the same as defending a particular belief. It is defending the infrastructure through which competing beliefs can be tested. The harder problem is that any defense doctrine risks mirroring the adversary’s posture, shaping perception in order to protect it. Medical ethics holds a structurally similar tension, since treatment also alters a person’s state, and it resolves that tension through three criteria: consent, transparency, and reversibility. A cognitive defense built around those principles would not dissolve the contradiction at the center of doctrine. But it would give the contradiction a shape, which is the precondition for holding it accountably rather than simply inhabiting it.

I do not think this is an oversight. It is a structural feature of the problem. Once the field of conflict is defined as cognition itself, the objective cannot be fully specified without presuming control over the very thing being defended. The introduction of neuroscientific data into planning intensifies this tension. NATO’s language of “cognitive superiority” deepens the problem. Superiority implies not just defense, but an edge: in perception, assessment, and decision. Protection implies restraint. To hold both at once is to accept a contradiction at the center of doctrine. A soldier is no longer only tasked with protecting bodies and territory, but with safeguarding patterns of thought within a civilian population while also outmatching an adversary in shaping those patterns. The duty expands inward. It becomes less about shielding people from harm and more about preserving the conditions under which they can think without interference, even as the system seeks an advantage in how thinking unfolds.

The whole-of-society response is not a slogan but an allocation of function. Trust, cohesion, education, and media literacy are organized as infrastructure. They are maintained, measured, and, when necessary, adjusted. Civilian life is drawn into the same ledger as logistics and readiness. The divide between social virtue and strategic asset is not argued away; it is filed under the same heading.

Persistence

There is also the question of persistence. If competition is continuous and conducted below the threshold of armed conflict, then there is no formal signal for when defense becomes overreach. The absence of a threshold removes the moment of decision that used to anchor law and accountability. The work continues at the desk, in the feed, in the ordinary sequence of reactions. No declaration, no armistice, no clear line to step back from.

Reading these reports, I picture not a battlefield but a room. Someone is sitting at a desk, scrolling through a stream of information, pausing, reacting, moving on. Nothing appears unusual. No alarms are triggered. No thresholds are crossed. And yet, somewhere in the aggregation of those small reactions, a pattern settles into place, steady and repeatable, like a signature that does not belong to the person who is writing it.

Stay curious (and alert)

Colin

Image: Bret Kavanaugh on Unsplash

The ECS is a master regulatory system that functions in almost all areas of the brain that deal with movement, memory, emotion, and sensory processing

The contemporary attention economy is not a neutral environment; it is an engineered system that resembles an earlier addiction industry, tobacco, one that profits by exploiting human vulnerability while presenting itself as harmless consumption. As artificial intelligence extends this competition by personalizing cognitive intrusion into our judgment and preference at scale, the task of our generation is to define human attention not as a commodity to be traded, but as a boundary that cannot be negotiated away.

The Hinge of History

In May 1787, twelve men met in the upper room of a bookshop and print store at 2 George Yard in London to form a society aimed at ending the British slave trade. The country they inhabited derived substantial income from that trade, and only a small fraction of the population possessed the right to vote. Over the following decades, campaigners such as Thomas Clarkson travelled more than 35,000 miles to gather evidence and organise local committees, turning personal conviction into a form of disciplined endurance that operated like a moral steam engine. Parliamentary efforts led by William Wilberforce introduced bills year after year until the Slave Trade Act was passed in 1807. Naval patrols then seized thousands of ships and freed hundreds of thousands of enslaved people. Historians have since argued that without these coordinated efforts, abolition might have been delayed indefinitely.

Conspiracy of Decency

In October 1943, German forces planned the deportation of Danish Jews. Warnings spread through informal networks. Fishermen, doctors, and civil servants organised transport routes across the Øresund. Within weeks, most of the Jewish population had escaped to Sweden. This was not chaos or luck but what can only be described as a conspiracy of decency, a coordinated moral counter-strike carried out through ordinary roles. The operation involved no central command and left minimal written record, yet it achieved a survival rate of nearly ninety nine percent.

In 1884, a group of young British reformers founded the Fabian Society after a split from a small moral fellowship, transforming what began as an almost theatrical ghost hunt into a disciplined political enterprise. They published pamphlets in plain language, invested in design and presentation, contested local elections, and placed members within administrative institutions, mastering the translation of complex ideas into accessible and persuasive forms. Over the next half century, many of their proposals became public policy, including expanded suffrage, public education, and social welfare systems.

Small Groups

These episodes share a structural feature that is easy to miss when we read them as moral tales. They are also expressions of what might be called moral ambition, the decision to treat goodness not as sentiment but as a project that can be organised, scaled, and sustained. They are not primarily about sentiment. They are about organisation under conditions where prevailing institutions had either failed or refused to act. In each case, a small group defined a problem that most people either tolerated or ignored, and then treated that problem as solvable through deliberate effort.

I believe this is the first correction we must make when approaching our own moment. It is tempting to describe the present in terms of scale. Artificial intelligence, global platforms, capital flows that dwarf the budgets of states. The instinct is to say that the problems are too large for the kinds of actors who once changed history. But the historical record points in the opposite direction. Scale does not eliminate agency. It shifts the terrain on which it operates. As Margaret Mead reminded us:

“Never doubt that a small group of thoughtful, committed, citizens can change the world. Indeed, it is the only thing that ever has.”

The abolitionists did not begin with power. They began with a claim that sounded implausible even to their contemporaries. The slave trade was legal, profitable, and embedded in everyday commerce. Yet they treated legality as contingent and profit as irrelevant to the question at hand. Clarkson, travelling wide to get people to take action after his prize winning 1785 essay: An Essay on the Slavery and Commerce of the Human Species, was a Latin dissertation that won first prize at the University of Cambridge. The essay, which argued against the morality of slavery, launched Clarkson's 60-year fight to abolish the transatlantic slave trade, reached a conclusion that is almost procedural in its clarity. If the argument is true, then someone must act.

This point clearly removes the comfort of analysis as a substitute for action, to repeat ‘If the argument is true, then someone must act.’ It also reframes ambition. The founders of the abolition movement were not less ambitious than the graduates who now flow into finance or consulting. They were differently oriented. They sought outcomes that could not be monetised and could not be reduced to personal advancement.

Influence

The second correction concerns time. The Fabian Society did not win by a single decisive event. It placed people in committees, boards, and local councils. It wrote pamphlets in accessible language. It treated the slow accumulation of influence as a strategy rather than a failure of speed.

In my view this is precisely where contemporary discussions of technology go wrong. We oscillate between two positions. Either we assume that change will occur automatically through innovation, or we assume that it will arrive catastrophically and without warning. Both positions relieve us of the burden of sustained effort. The historical examples suggest a third option. Change is constructed through persistence, even when the surrounding environment appears indifferent.

Deficit of Attention

In his Reith Lecture, Rutger Bregman, Dutch historian and the author of bestselling books such as Humankind and Utopia for Realists, turns explicitly to the machine. The argument Bregman presented is not that technology is inherently destructive, but that it amplifies existing tendencies. Systems reward behaviour that captures attention, often privileging extremity over accuracy. Human interaction is mediated through platforms designed to maximise engagement rather than understanding.

The central question from Bergman’s lecture is not whether machines will surpass human intelligence. The more immediate question is whether human purposes will survive contact with systems optimised for different ends. The earlier industrial period expanded physical capacity. The current one extends into cognition, preference, and judgement. That is a deeper intrusion, not because it is more powerful in a mechanical sense, but because it operates within the domain we once treated as private.

This is where the comparison with earlier movements becomes more precise. The temperance activists identified a pattern in which a profitable industry exploited human vulnerability. Their response was not limited to personal restraint. They organised for structural change. They altered taxation, regulation, and public norms.

AI and the Hinge of History

We are currently at a hinge of history, facing a choice between, what some claim will be an automated abyss, and others claim a human-centered renewal. The contemporary attention economy is not a neutral environment. It resembles an earlier addiction industry, tobacco, one that profits by exploiting human vulnerability while presenting itself as harmless consumption. Our present technology is an engineered system that competes for time and focus. Artificial intelligence extends this competition by personalising it at scale, shifting the terrain from mechanical assistance to cognitive intrusion into judgment, preference, and attention. The result is not merely distraction. It is the gradual reallocation of human attention toward objectives defined elsewhere.

What would it mean to fight for humanity in this setting. The phrase itself risks abstraction. I prefer to anchor it in the patterns already observed. Small groups that define a clear objective. Strategies that operate across time rather than within a single cycle. Institutions that embed those strategies into durable forms.

The first task is definitional, to establish that human attention and judgment are not commodities but boundaries that cannot be negotiated away. The abolitionists named slavery as intolerable. The Fabians named inequality as correctable through policy. Our equivalent task is to define which uses of technology violate the conditions of human agency. This is not a technical classification. It is a moral one, and it requires clarity rather than consensus.

Skin-in-the-Game

The second task is organisational. We require a skin-in-the-game elite, small groups of committed citizens who treat the protection of human agency as a non-negotiable objective. Historical movements built networks that translated ideas into action. Committees, pamphlets, local chapters. The forms will differ today, but the principle remains. Ideas that are not institutionalised remain optional.

The third task is personal, though not in the narrow sense of lifestyle. The individuals in these movements altered their own trajectories. Clarkson abandoned a conventional career. Fabian members chose public service over private gain. I think this is the point at which the argument becomes uncomfortable. It is easier to diagnose systemic issues than to redirect one’s own path within them.

I return to a detail that is easy to overlook. The book and print shop at George Yard no longer exists. In its place stands an office building housing a financial firm. The physical location has been absorbed into the ordinary fabric of the city. What remains is not the structure but the decision that was made inside it.

That decision did not require certainty. It required a willingness to act under conditions of uncertainty. This is probably the closest we come to a durable definition of courage in this context. Not a dramatic gesture, but a commitment that persists when outcomes are unclear.

Somewhere, at this moment, there are rooms that resemble that book and print shop. They are not marked as historically significant. They are likely occupied by people who are not yet recognised as changing history. The question is not whether such rooms exist. The question is whether the people inside them will decide that the argument, if true, demands action, and in doing so, build not monuments in stone, but monuments in time through the choices they sustain.

A bright future is possible, and you can make a difference.

Stay curious

Colin

Image: The New York Public Library on Unsplash

The capital city of Poland, Warsaw, is surrounded by forests, where wild animals still roam; deer, moose, boar, foxes, badgers and a myriad of other small animals and birds. In one of these forests, Las Kabaty, there is a stone monument to three remarkable mathematicians that the world should be aware of, the office where they once worked in that forest, is still operational. But access is strictly limited. I often walk by that monument.

Enigma

In early 1928, a parcel declared as radio equipment arrived at a customs house in Warsaw. It was a Saturday afternoon, and that accident of timing gave the Cipher Bureau employees a luxury rarely afforded in intelligence work: time. They opened the box, examined it carefully, and understood that it was not radio equipment at all but a cipher machine, specifically the commercial Enigma, as the military version was not yet in use. A representative of a German firm demanded its immediate return. The urgency raised suspicion. Officials informed the Polish Cipher Bureau. The box was opened and examined. It contained a cipher machine. It was studied and then resealed.

On July 15, 1928, German military radio transmissions encoded by that machine were intercepted by Polish monitoring stations. Early attempts at decryption failed and were discontinued.

In the same year a cryptology course was established in Poznań for mathematics students fluent in German. On September 1, 1932, three graduates were recruited to the Cipher Bureau in Warsaw: Marian Rejewski, Jerzy Różycki, and Henryk Zygalski. The Bureau’s leadership, including Maksymilian Ciężki and Franciszek Pokorny, directed work toward systematic analysis of German ciphers.

By the early 1930s, the German Enigma machine, based on rotating electrical rotors, had entered military use, offering a large number of possible configurations through rotor order, ring settings, and plugboard connections.

The historical foundations are firm. While the details of espionage often shift or blur, the timeline here remains stable. The identities of the men and the architecture of the machine, the rotors, reflectors, and light panel, are the fixed coordinates of this story. These are not merely data points but the physical reality that the Polish mathematicians had to translate into logic.

Simply put, we should move away from the temptation to treat this as a technical prelude to a British story. Dermot Turing’s (nephew of Alan Turing) account, in his wonderful book X, Y and Z – The Real Story Of How Enigma Was Broken, is careful, almost corrective in tone, and for good reason. The British achievement at Bletchley Park stands on a prior act of intellectual audacity that did not occur in Cambridge or London, but in Poznań and Warsaw. The order of events is not a matter of national pride. It is a matter of causality.

Humble

What strikes me, reading Rejewski alongside the later narrative by Turing and others, is the humbleness of the Polish intervention. There are three young men invited into a room and asked to look at a machine as if it were a poem or a puzzle of symbols. This was the hinge. While others saw a terrifying wall of gears and electricity, Rejewski and his colleagues saw a system of relations. They moved the struggle away from the hardware and into the realm of human thought, deciding that even the most complex machine is ultimately a reflection of the mind that built it.

Their method, grounded in permutation theory, did not break Enigma by force. It reduced it. That is a different intellectual gesture. To reduce is to deny the machine, or system, or methodology, its authority. It is to say that the apparent randomness is, in fact, organized.

The breaking of the Enigma code story is unexpectedly positive. Not triumphant in the usual sense, but instructive. A small group, operating with limited resources, chose a different way of thinking. Ciężki’s decision to recruit mathematicians was described later as unusual, even suspect, by foreign counterparts. Even after the July 1939 meeting at Pyry, Alastair Denniston, head of the British Government Code and Cypher School, argued that physicists were preferable because they retained what he called a “contact with reality.” The Polish wager was the opposite.

Two devices mark the transition from insight to operation: the cyclometer and the bomba kryptologiczna. These devices did not emerge in isolation. They depended on a close working relationship between the Cipher Bureau and the AVA Radio Manufacturing Company, particularly engineers such as Antoni Palluth, who could translate mathematical insight into functioning electromechanical systems. These were not acts of surrender to machinery. Rejewski himself regarded the turn toward electromechanical assistance with some reservation, almost as a concession. Yet those devices extended the initial insight. They made it operational.

Reconstructed Machines

Dermot Turing’s narrative shows how this initial insight migrated. At Pyry in July 1939 the Polish team provided reconstructed Enigma machines and explained their methods in detail, effectively handing over both the tool and the conceptual framework. The British bombes were descendants, not independent inventions. The Pyry conference of July 1939 is one of the most consequential briefings in modern history. The Poles did not merely share results. They transferred a way of seeing.

The accounts of that meeting reveal the human texture. The irritation felt by the French and British was not incidental; it was rooted in long traditions of earlier success that had failed against a machine cipher. Dillwyn Knox, one of the leading British cryptologists, was so unsettled by the Polish achievement that he initially suspected they must have obtained the internal wiring of the machine illicitly, a suspicion that would later echo in years of confused and distorted retellings. Irritation, disbelief, wounded pride. French expectations of reciprocity. British confidence in their own tradition. These are not flaws in the story. They are its substance. Knowledge moves through people, and people carry habits of thought that resist displacement.

Bletchley Park

The standard version of this history often rushes past the Polish contribution, treating the Pyry meeting as a mere handoff before the real work began at Bletchley Park. This framing suggests that Alan Turing simply picked up where Rejewski left off, but that version of the story flattens a much deeper technical evolution.

The shift was not just about scale; it was about a change in how the battle was fought. Rejewski’s initial breakthrough relied on finding patterns in the way the Germans repeated their message keys, a method based on pure mathematical symmetry. By the time the work moved to England, the Germans had changed their procedures, forcing a new approach. The British bombes were indeed descendants of the Polish machines, but they were repurposed to hunt for cribs, small fragments of suspected plain text, like weather reports or military greetings.

This was not a simple handoff. It was the migration of a fundamental insight into a new operational reality. To see the Polish work as a mere opening act is to miss the truth of the lineage. Their achievement was the conceptual engine. Without that original Polish framework to build upon, the later industrialization of codebreaking would have been a search without a map.

Structured Analysis

The Polish move toward abstraction was not simply intellectual elegance; it was born of constraint. While the French possessed valuable intelligence material obtained through Hans-Thilo Schmidt, these documents were originally deliberately kept away from the Polish mathematicians. The result was a forced independence: the problem had to be reconstructed mathematically rather than solved through espionage. This decision sharpened the method and ensured that the eventual breakthrough rested on structure, not on accident. The Enigma problem, as presented by the Germans, was a problem of scale. Large numbers, many configurations, operational discipline. The Polish response was a problem of form. Identify invariants, exploit repetition, model the machine as a system of permutations. This is not merely a historical curiosity. It is a lesson in how intelligence proceeds when faced with overwhelming complexity.

I believe we are drawn to stories of scale because they are visible. Rows of machines, large facilities, vast expenditures. But the decisive movement here is almost invisible. A change in representation. A refusal to accept the problem on its given terms.

The precursor to AI?

The transition from pure human logic to mechanical assistance in Warsaw can be viewed as an early precursor to the logic of automated intelligence. When Marian Rejewski realized that the human mind could no longer keep pace with the Enigma’s permutations, he did not simply build a faster calculator; he designed the bomba kryptologiczna to automate a specific process of logical deduction. This shift, moving from solving a problem manually to designing a system that can reason through millions of possibilities based on a set of mathematical rules, is the conceptual ancestor of AI. It represents the moment when human insight was first offloaded into a machine to solve a problem of scale that was otherwise impenetrable to organic thought.

Saved Lives

There is a claim that appears in almost every serious account of the breaking of the Enigma Code. That the decryption of Enigma shortened the war by two, perhaps three, even four years, and in doing so saved millions of lives. The number varies depending on the historian and the counterfactual they are willing to entertain, but the direction is consistent. Intelligence derived from Enigma, what the British later called Ultra, altered convoy routes, exposed troop movements, and compressed decision cycles at the highest levels of command. It did not win the war alone, but it changed its duration and its cost in human lives.

The positive image that remains for me is not Bletchley’s halls or the machines in motion. It is in two smaller rooms, one in a forest on the outskirts of Warsaw, the other at the Nazi destroyed Saxon Palace, the Cipher Bureau's headquarters before the move to the forest, a blackboard, a set of symbols, and three young men who decide that a machine built to conceal can be rewritten as a set of relations. That decision, taken early and held with discipline, carries forward through the war, through the later machines, and into the quiet architecture of modern computation.

Cyber Warfare

Poland is still at the leading edge of cyber work. The war in Ukraine has led to regular cyber attacks, by Russia, Iran, North Korea and others, on Polish companies, institutions and infrastructure. The groundwork laid by the three mathematicians survives in the local expertise that remains a primary defense against modern digital aggression. In Warsaw, Krakow, Poznań and other Polish cities, the lineage of Rejewski, Różycki, and Zygalski is not a static memory but an active methodology. When Polish engineers today encounter encrypted threats from state actors, they are operating within a tradition that prioritizes the structural over the superficial.

The current landscape of cyber warfare involves a challenge that is similar to the Enigma problem, yet the Polish response remains as ironclad as it was before the second world war: the belief that any system of concealment is ultimately a system of logic that can be unraveled through mathematical rigor. The stone monument in Las Kabaty serves as a reminder that the most formidable defenses are not built of steel or vast servers, but of the intellectual courage to look at a machine and see an equation.

This continuity suggests that the greatest legacy of the Cipher Bureau was not just the messages decrypted or the years saved, but the establishment of an intellectual infrastructure that still guards the frontier of European cyber security. The three young men in the forest did more than break a code; they proved that a small, disciplined group using abstraction as a lever can shift the weight of the world. In the silent rooms where modern threats are analyzed today, their way of seeing remains the most effective tool for defending against the attacks of modern warfare.

Stay curious

Colin

Why You Won’t Be Fired, But Might Not Be Hired

In internal engineering workflows at Google, the CEO, Sundar Pichai, recently claimed that AI automated systems now generate the majority of new code. Human engineers review and approve this output before deployment. The share of AI-generated code has increased from roughly half to 75% within a year.

In nineteenth-century England, the mechanization of bootmaking followed a different timeline. The introduction of the sewing machine in 1857 initiated gradual changes in production. Census data from 1851 to 1911 shows that approximately 152,000 traditional jobs in bootmaking disappeared, while about 144,000 new roles emerged that required different tasks and capabilities. Incumbent workers largely remained employed, while younger cohorts stopped entering the trade.

In contemporary labor market research, economists describe three persistent frictions that slow adjustment to technological change: information, location, and skills. These frictions shape how workers and firms respond to new technologies and how quickly new equilibria emerge. Recent empirical frameworks suggest that only a minority of jobs face immediate AI automation risk, while many are reorganized or expanded through changing task composition rather than eliminated outright.

The facts which are emerging are calmer than the headlines to some extent. No sudden collapse of work. No mass disappearance of livelihoods. No decisive moment where a machine takes a job and a human steps away. Instead, a quieter process unfolds, and it unfolds through frictions that redirect rather than erase. I remember a conversation with Sam Altman in, I believe, 2018, I had known Sam for ten years. I remember him pointing outside the window of a restaurant in Silicon Valley, saying that in three or five years there were going to be robots building homes there. None of that has materialized. There is a very real tendency of people in Silicon Valley not just to overpromise on the technology, but to underestimate the obstacles to real-world adoption of technology.

Entry Level And Role Displacement

I think most contemporary arguments about AI fail because they search for the wrong signal. They search for dismissal. History records deflection. When the unit of analysis shifts from jobs to tasks, what looks like stability at the level of employment conceals a continuous reallocation of competence inside the job itself.

In the bootmaking towns of Leicestershire, a young man in 1885 does not lose his job. He never gets it. His father continues stitching leather, perhaps with machine assistance, perhaps with some awareness that the work is thinning. But the son looks elsewhere. The disappearance happens through non-entry.

The pattern, however, is not uniform. The same transition produces a sharp, gendered divide. Male incumbents were largely shielded by the slow pace of change. Female workers, especially the binders who stitched the uppers, faced direct and substantial displacement as the sewing machine moved into the factory. For young men, the transition was a quiet deflection. For women, it was an immediate rupture inside the production cycle.

I believe we are watching the same structure repeat under different technical conditions. When I read that three quarters of new code is generated by machines and approved by engineers, I do not see all engineers being removed. I see the threshold for entry rising. The first task in the profession has already been completed before the junior employee arrives.

The frictions explain why this happens without visible rupture. Information frictions create uncertainty about what counts as skill. Workers ask for clarity before they ask for wages. They want to know what the firm intends to do with the technology, not what the technology could theoretically do. When that clarity is missing, prospective employees and students hesitate, delay, or invest in the wrong capabilities, and the entry point narrows because preparation no longer matches demand.

Reskill and Upskill

Location frictions concentrate opportunity. Innovation clusters form, then harden into expectations. Access to new roles depends on proximity to these clusters, physical or digital, and those outside them face a compounding disadvantage. The entry point does not disappear everywhere at once; it recedes unevenly, leaving some regions saturated with opportunity and others with none.

Skill frictions complete the mechanism. The requirement is no longer to acquire a stable craft but to sustain adaptability across changing task compositions. Economists describe this as the need to reskill and upskill. I think this understates the demand. What is required is not periodic training but continuous requalification.

As routine tasks are automated, the relative value of human work shifts toward capabilities that are difficult to encode: communication, interpretation, conflict resolution, and forms of social judgment that depend on context rather than rules. These are not residual skills. They are becoming the primary source of wage growth and job satisfaction in many sectors.

This creates a tension inside the modern role. On the one hand, production becomes easier. The system drafts, calculates, suggests. On the other hand, the remaining human tasks become heavier. They require attention to other people, to ambiguity, to consequences that cannot be delegated. Judgment expands, but it expands into domains that are interpersonal as much as technical.

There is a small joke among engineers that the future of programming is reading. Not writing. Reading what the system has already produced. The implication is clear. If the system produces the first draft, then the human contribution shifts to evaluation and responsibility.

It is tempting to describe this as a purely administrative burden. It is not. It is a revaluation. The administrative surface hides a deeper change in what counts as valuable work. The skill is no longer the ability to produce output alone but the ability to situate that output within human contexts, to explain it, to adjust it, and to take responsibility for it in relation to others.

Social and Emotional Intelligence

This is why the emphasis on social and emotional intelligence is not ornamental. It is an economic necessity. These capabilities govern how augmented work is translated into outcomes that other humans accept, trust, and use. Without them, productivity gains stall at the boundary of adoption.

There is also a second adjustment that receives less attention. Lower costs do not simply reduce labor demand. They expand use. When tasks become cheaper, they are performed more often, in more contexts, and by more people. This expansion creates roles that did not previously exist, particularly in services where human interaction remains central, such as in health care and frontline work.

If AI systems augment workers in lower-paid roles by increasing their capability, then the technology has the potential to compress certain inequalities. A care worker with better tools can deliver higher quality service. A small business owner with access to advanced systems can compete more effectively. The same mechanism that raises the entry barrier in one domain can lower it in another.

But this outcome is not automatic. The research is explicit that these effects depend on how firms adopt and govern the technology. Without investment in skills and supportive structures, the benefits concentrate rather than diffuse.

So the system both absorbs and sorts. It absorbs by reorganizing tasks, expanding use, and preserving many existing roles. It sorts by shifting entry conditions, by rewarding those who adapt to new skill compositions, and by filtering out those who prepare for a structure that no longer exists.

The Victorian data shows that incumbents were protected not by design but by timing. Change moved slowly enough that they could remain in place. The cost of adjustment fell on those who had not yet entered the world of employment.

Career Ladder

I find the parallel difficult to ignore. The present workforce may remain, augmented and redefined. The real pressure falls on those approaching the boundary. The student, the graduate, the apprentice without a clear first task to perform. Not because the job is gone, but because the beginning has already been taken over by AI.

The system has consumed the prologue for first time employees. The first draft is written before the novice arrives and is able to serve their apprenticeship. The routine task that once served as an initiation now exists as a precomputed artifact, waiting for review rather than creation. What used to be practice has become output. What used to be entry has become oversight.

As the prologue to work is pre-consumed, the economic ladder is not being broken, but its rungs are being spaced further apart. For those trying to get on the ladder today, the struggle is that the entry-level no longer looks like a training ground. It looks like a high-stakes coordination center.

Inside the enterprise, the demand for reskilling is not a technical mandate to learn more code or faster workflows. It is a mandate to double down on the most difficult capabilities to encode. Critical thinking becomes essential because in a world where the system provides the first draft, the worker’s value is no longer in generating the answer, but in interrogating it. Emotional intelligence and communication become the primary levers of productivity because routine tasks are fading, leaving only the work that requires empathy, social awareness, and the ability to translate automated insights into shared human goals.

This alters the structure of a career in a way that is easy to miss. Progress once moved from simple production to complex judgment. Now it begins at judgment, without the long apprenticeship that made judgment legible. The novice is asked to evaluate work they have not yet learned to produce, to approve decisions whose full consequences they have not yet lived through. They are missing out on the tacit knowledge, the deep, intuitive understanding of how systems work, that can only be gained by writing those lines manually.

What remains available to them, from the first day, is not technical mastery but relational consequence. If the system produces the output, the human must interpret its effects on other people, explain it, defend it, adjust it when it fails. The initiation is no longer technical. It is ethical and interpersonal.

Governance

This is not an accidental feature of the transition. It reflects how firms choose to deploy these systems. Whilst I have highlighted skill frictions, I also emphasize that outcomes depend on governance, on whether organizations actively design roles that preserve learning, agency, and development. In principle, firms could retain fragments of the prologue so the junior gets their foot on the ladder, maintaining spaces where practice precedes output. In practice, the pressure for efficiency and the absence of clear information about long-term skill needs mean that these spaces are often the first to disappear.

The difficulty is not scarcity of roles. It is the absence of a starting place inside them. What does it mean to begin when the beginning has been automated.

I think the answer returns to skills, but in a broader sense than the term usually carries. Not a fixed inventory, but a capacity for movement across task configurations. The ability to learn, to relate, to interpret, to act under conditions where the system has already produced a plausible answer.

This is a new form of professional life. A sequence of positions that require constant recalibration, where value is created not by holding a role but by adjusting within it.

Somewhere, an engineer stares at a screen of systems architecture she did not craft, feeling the gap where her own apprenticeship should have been. She pauses, not simply to check the syntax, but to ensure this machine-built software serves the people who will eventually depend on it. When she signs her name, she knows that while she didn't write the lines, she must answer for them.

Stay curious

Colin

Some additional thoughts

Students are still being taught how to write code (the prologue), but the industry now demands they know how to read, audit, and govern code. The wrong thing about code today is that the way we teach it no longer matches the way the economy consumes it.

Writing code is becoming easier and cheaper, but the judgment required to approve it is becoming heavier. The code itself might be syntactically perfect, but it is lacking.

The machine doesn’t know why the code is being written or who it might inadvertently harm.

The AI cannot answer for the code it generates; only the human who signs their name can.

Image: Sam took a selfie when he saw me in 2023

Is it time to pause AI and get used to what it can do today? Less than 1% of the population have any idea about its capabilities!

Through my work, and connections, I have access to advanced AI. I also have conversations with people who know how to use it and its full capabilities. The LLM’s most people ‘play’ with are shallow compared to what is already built. Maybe, given the latest Mythos model we will see some pause in new releases. I hope so. People need to build skills with the current models, build competence and get used to the incredible capabilities. I think of AI like the Telegraph Road!

I have written before about the best song and its connection to my life. Another all time favorite by Dire Straits is Telegraph Road, I grew up witnessing the building of community and the closure of factories, which the AI Labs now predict.

Let’s think about it. A long time ago came a man on a track. First there is the solitary figure, then the clearing, then the cabin, then the winter store, then the ploughed ground by the lake shore. Then come the other travellers. Then the churches, the schools, the lawyers, the rules. Then the trains and the trucks with their loads. Then the road itself, widened and hardened and named. The old path becomes an artery. The artery becomes a system. The system becomes a world. And somewhere inside that world, so gradually that no one notices the exact day of the betrayal, the promise curdles.

For me “Telegraph Road” has always felt larger than a song about one American road or one industrial city. It is a song about the long bargain of modernity. Build, move, connect, scale, expand. Trust the next layer of organization. Trust the line on the map, the cable on the pole, the factory whistle, the payroll, the traffic, the schedule, the screens. Trust, above all, that more coordination means more life. The song understands something colder. It understands that every system begins as a convenience and ends as an environment. By the time people realize they are living inside it, it is already deciding what can be built, what can be said, how work is measured, and who will be told, with administrative politeness, sorry, but we’re closed.

Digital Road

It is difficult to listen to that song now, in the age of artificial intelligence, without hearing not nostalgia but continuation. We have not left Telegraph Road behind. We have digitized it.

The old road carried ore, steel, components, wages, commuters, and exhausted men driving home from factories. The new road carries prompts, predictions, risk scores, optimization targets, productivity dashboards, synthetic voices, compliance summaries, automated decisions, and the permanent managerial fantasy that judgment can be scaled like freight. We still tell ourselves the same consoling story. A new system is arriving. It will remove friction. It will eliminate waste. It will connect everything to everything else. It will not replace the human being, only assist him. And then, with the stealth that belongs to every successful infrastructure, it stops being a tool and becomes the place where the terms of life are set.

Closed Doors

The depth of the song is not merely that factories close. It is that an entire moral vocabulary closes with them. “I used to like to go to work, but they shut it down.” Devastating and yet so clear. No theory. No slogan. Just the disappearance of the ordinary dignity of showing up, doing something difficult, earning one’s place in the day. There are few things the modern expert class has misunderstood more badly than this. They speak easily of transition, reskilling, mobility, adaptation, as though the human being were a sort of Swiss Army soul, detachable from place, memory, pride, neighborhood, rhythm, trade, and inherited self-respect. They call this flexibility. The person on the receiving end often calls it dismemberment.

The Jarrow men understood this long before our consultants learned to convert it into PowerPoint. In 1936, two hundred unemployed men marched from Jarrow to London because the closure of the shipyard had not merely reduced local income. It had broken the line between a man and the world that recognized his use. They marched roughly 282 miles to present a petition, and the cruelty of the episode lies partly in its bureaucratic smallness. They were not asking for the moon. They were asking for work. Asking, in effect, not to be stranded beside the road of modern Britain while commerce, finance, and political language moved on without them. One can still feel, across the decades, the humiliation hidden inside the respectability of the march. Men had to become a procession to prove they were not refuse. I know that town well.

Detroit supplied the American version of the same parable on a scale so vast it became almost mythological. Here was the city where industrial modernity appeared to have solved the riddle of abundance. The assembly line, the wage packet, the machine shop, the promise that a man might come from somewhere poorer and become someone sturdier through work. Then came the long undoing. Plants moved. Lines went silent. Neighborhoods absorbed the knowledge before the newspapers did. A city can hear its own abandonment. It hears it first in the changed traffic. Then in the shorter lunch counters. Then in the houses that no longer sell. Then in the way a generation begins speaking of leaving as though it were weather.

Community

This is why the road in the song counts. Telegraph Road is not just pavement. It is an entire civilizational sentence. It says that growth is linear, that development has a direction, that all the trucks with their loads are headed somewhere sensible, that the wires above us hum with coordination rather than with command. It says that history is a process of widening. Yet the song keeps returning to what widening does not heal. Six lanes of traffic. Three lanes moving slow. A population in motion and a social contract standing still. It is one of the bleakest images in popular music because the road is full, but the future is empty.

Here Enters AI

Artificial intelligence enters this history not as a break with the old pattern but as its latest and most ambitious extension. It promises to do for cognition what industrial systems did for muscle and logistics. It will sort, draft, classify, recommend, infer, summarize, monitor, translate, optimize. It will make institutions faster and supposedly smarter. It will turn the delays of thought into the smoothness of procedure. It will take the old telegraph dream, that information could be transmitted cleanly across distance, and inflate it into the newer fantasy that judgment itself can be automated across scale.

That is the fantasy. The reality is stranger and, in a way, more insulting. Industrial capitalism at least had the decency to need your body before it discarded your town. AI systems often do not even grant that dignity. They ingest your language, your documents, your gestures, your corrections, your private hesitations, and your accumulated work, then present the result as frictionless assistance. What used to be called skill begins to reappear as supervision. What used to be called training becomes prompt refinement. What used to be called experience becomes an edge case. The labor is not always removed. Very often it is concealed, fragmented, downgraded, and returned to the worker in a more subordinate form. One of the funniest and ugliest tricks of our age is that a system can deskill you while thanking you for your feedback.

This is where the continuation of the song becomes unmistakable. The factories in Detroit were closed in visible ways. Gates locked. Sirens silenced. Windows dirtied. The AI economy often closes its factories invisibly, inside the occupation rather than outside it. The office remains open. The laptop still glows. The calendar still fills. The worker is told he has been empowered. Yet the center of gravity has shifted. Judgment migrates upward into the system. Responsibility drifts downward onto the human being who must sign off on what he did not fully decide.

The injury here is not only economic. It is inward. A person can survive a hard job more easily than a hollow one. What wears people down is not simply pressure, but the sickly knowledge that their name remains attached to outcomes they no longer shape in full, that their expertise is being thinned into supervision, that their presence has been retained chiefly to absorb blame, reassure the client, or satisfy the ritual need for a human somewhere in the loop. One begins to feel less like an actor than like a trim detail on the edge of an institutional verdict. Present at the ceremony. Absent from the power.

Idiotic?

It is fashionable to say that technology always creates new jobs. This is one of those statements which remains technically true right up to the point where it becomes morally idiotic. Yes, new tasks appear. So did new tasks when the old industrial order cracked apart. Someone always ends up maintaining the ruins. But the serious question is not whether activity continues. The serious question is what kind of human being the new order permits one to become. A civilization should be judged not only by its output, but by the quality of agency it distributes among its members. Does it give them real responsibility, real mastery, real competence, real room for judgment, real reasons to believe they are participating in the making of the world? Or does it merely keep them busy within systems whose important decisions have already been made elsewhere?

That is why the sentimental optimism around AI strikes me as so thin. It assumes that because a task can be accelerated, a life has been improved. This is the same error that stalked so much twentieth-century industrial thinking. Faster throughput was mistaken for human advance. More volume was mistaken for more civilization. But a society is not redeemed by the speed of its processes. If anything, speed often serves as the preferred disguise of moral evasion. The faster the system moves, the less anyone is expected to stand still and ask what sort of dependence is being built.

The Jarrow marchers asked that question with their feet. Detroit asked it with abandoned plants and neighborhoods that became unwilling museums of economic doctrine. Our version is quieter, because digital systems tend to produce forms of dispossession that are cleaner, better lit, and wrapped in the smiling language of innovation. The closure sign has become more sophisticated. It no longer always says the factory is shut. Sometimes it says your role is evolving. Sometimes it says your productivity is being augmented. Sometimes it says the decision was generated automatically but reviewed by a human, which is a magnificent sentence if your ambition has always been to become the decorative border around an institutional verdict.

Memory Is All We Have

And yet the song also contains, buried in its rain and exhaustion, a clue about what survives. It survives in the insistence that memory is not inefficiency. “I’d sooner forget, but I remember those nights.” Modern systems are forever asking us to forget. Forget the old skills, the old solidarities, the old neighborhoods, the old meaning of work, the old expectation that institutions owe human beings more than clever interfaces and optimized exits. Forget, in short, the moral claims that earlier forms of ruin taught with unnecessary clarity. Memory is inconvenient because it notices repetition. It hears in the language of disruption the old accents of abandonment.

This is why the Jarrow walk belongs beside the closed factories of Detroit and beside the sleek vocabulary of AI transformation. They are chapters in the same political story, the story of what happens when the systems built to organize life begin to treat whole populations as residual. First the road opens the territory. Then the market deepens it. Then the institution standardizes it. Then the logic of efficiency strips away whatever cannot justify itself in the language of output. At each stage the process is described as necessity. At each stage its victims are told, with varying degrees of elegance, to be realistic. And at each stage someone has to live among the consequences after the planners have gone home.

What makes AI different is not that it breaks with this history, but that it reaches more deeply into the human interior. The old factory organized time, movement, wage, fatigue. It took the body first. It marked the hands, the back, the lungs, the sleep. It exhausted the worker in ways that were brutal, visible, and measurable. The new system aims at language, attention, interpretation, preference, even companionship. It does not merely want to coordinate the road. It wants to inhabit the driver.

That is a profound change in the texture of power. Industrial labor could leave a man spent, but there were still places where he remained unformatted: the inward running commentary of his own mind, the words he chose in private, the pace at which he made sense of his experience, the small delay between feeling and expression in which judgment took shape. AI systems increasingly press into that delay. They do not simply ask what task can be automated. They ask whether drafting, sorting, answering, consoling, evaluating, and even deciding can be precomposed before the person has fully arrived in his own thought.

This is why the road feels more total than the ones that came before. Earlier systems extracted labor and disciplined routine. This one also leans toward mediating perception itself. It offers to complete the sentence, suggest the tone, rank the priority, summarize the meeting, infer the intention, and supply the reply before reflection has had time to gather its strength. The danger is not only dependence. It is the gradual atrophy of the inner motions by which a person becomes answerable for what he says and does.

That is why the stakes reach beyond employment statistics, though those count. We are dealing with a machinery that aspires to become the medium through which institutions speak, evaluate, remember, and decide. Once that happens, the danger is not simply job loss. It is a subtler civic erosion in which fewer and fewer people are entrusted with the burden of thinking in full.

Slow Down

And thinking in full is what democratic life requires. Not endless speed. Not synthetic fluency. Not the euphoric chatter of products that promise to remove every ounce of friction from work and judgment alike. A healthy society needs people who can deliberate, hesitate, refuse, interpret, and answer for what they have done. It needs institutions that do not hide behind the systems they purchase. It needs leaders willing to say that not every gain in efficiency is a gain in civilization. Most of all, it needs to remember that when people say they want work, they are usually asking for something larger. They are asking not to be made incidental.

The great ache in “Telegraph Road” is that by the time the signs say sorry, but we’re closed, an entire landscape of expectation has already collapsed. The lights are still there. The lanes are still there. The road is still there. But the covenant has gone. That is the note I hear in the AI age. Not apocalypse. Not some theatrical war between man and machine. Something more banal and therefore more dangerous: the steady transfer of agency from persons to systems, accompanied by the cheerful claim that nothing essential has been lost.

Something essential has been lost before. Jarrow knew it. Detroit knew it. The song knew it. The question is whether we intend to learn from those earlier roads, or merely build a cleaner, faster, more total version of them and call it progress.

A long time ago came a man on a track. He made a home in the wilderness. Then came the roads, the rules, the loads, the hard times, the war, the freezing nights, the factories, the closures. Now come the models and the dashboards and the automated judgments, gliding toward us with the old music of inevitability. The road is still deep and wide. The wires still hum above us. The birds, as ever, can fly away.

The rest of us have to decide what kind of world this road is making before it tells us, too, that there is no work here to be found.

Stay curious

Colin

In 1956, at Dartmouth, a small group of researchers gave a name to a field they believed could be built. They called it artificial intelligence. the proposal was:

“... every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it.”

In the decades that followed, the field advanced in bursts. Programs solved logic problems. Expert systems entered firms and hospitals. In 1997, Deep Blue defeated Garry Kasparov in a regulated match under formal rules. In 2016, AlphaGo defeated Lee Sedol in Seoul. By then, machine learning had already entered search, advertising, fraud detection, logistics, and medicine.

The language around these systems grew larger than the systems themselves. Startups called ordinary software AI. Executives spoke as if judgment had become a software procurement category. Researchers, meanwhile, kept finding the same obstacle in new forms. A system could classify, rank, retrieve, or predict, and still fail when the world refused to stay tidy. It could process vast data and still need a person to intervene when confidence fell, context shifted, or consequences sharpened. Many real tasks remain open-ended, and human supervision, interpretation, and correction are not temporary embarrassments in the march toward automation. They are part of the work. More than that, they specify why. Human problem-solving draws on cognitive mapping, selective attention, associative memory, and forms of intuitive reasoning that reduce search space without requiring exhaustive calculation. Machines, by contrast, are often strongest where inputs can be normalized, repeated, and scored against clear objectives. The dispute is not poetry versus engineering. It is a disagreement about what kind of world the system is entering.

Too much writing on AI begins with prophecy. Prophecy is cheap. Procurement is expensive. Hospitals, clinical trials, courts, schools, factories, banks, logistics networks, public agencies, and ordinary offices do not live inside keynote slides. They live inside exception handling. They live inside ambiguity. They live inside the small, repeated fact that the case in front of you is not quite like the last one, and that the difference is exactly where the trouble enters. The AI Labs fixation on ‘growing brains’, machine consciousness, mass unemployment treat the human being as a temporary nuisance standing between the present and full technical sovereignty.

An Intellectual Wildcard

A paper, by Jarrahi, Lutz, and Newlands, states the issue plainly. AI is often treated as an “intellectual wildcard,” and that looseness has practical costs. The examples of Anthropic’s Mythos show this clearly. It clouds judgment about what these systems can and should actually do. It also encourages a false contest between machine intelligence and human intelligence, as if the only serious question were who wins. But their argument is more disciplined than that. Human intelligence, they write, retains forms of general, contextual, analogical, social, and intuitive judgment that current AI does not possess. AI can reveal correlations at speed and scale. It can outperform us on bounded tasks. Yet in the wild, outside the lab, most important tasks are not bounded for very long. Goals are loose. Context shifts. Stakes collide. Trade-offs are political before they are computational. Under those conditions, the old fantasy of clean replacement begins to look less like science and more like AI Labs managerial daydreaming.

I believe this is why the phrase “human in the loop” is often misunderstood. It is usually spoken in the tone of a concession, as if the human were a brake pedal left over from a less efficient age. In the scientific papers the phrase means something sturdier. For example, Zheng and co-authors define human-in-the-loop hybrid-augmented intelligence as a system in which the person remains part of the decision process, especially when the machine produces a low-confidence result. That sounds obvious until one notices what it overturns. It overturns the cult of frictionless automation. It says that verification is not a sign of weakness. It says that judgment is essential when confidence scores stop being enough. It says that the final demand in serious domains is not merely prediction, but accountable interpretation.

Uncooperative

Many organizations buy AI with the hope of removing discretion from work. Discretion is slow. Discretion requires training. Discretion leaves fingerprints. So the dream is a system that will standardize the world. Then the world, in its usual uncooperative manner, keeps sending edge cases, adversarial cases, tragic cases, cases written in poor language, cases with missing data, cases whose facts arrived late, and cases in which the most relevant fact was the one no database had thought to store. The human being, whom management had hoped to demote to ceremonial status, has to be invited back into the room and asked to save the apparatus. Jarrahi and his co-authors have a better name for this than most boardrooms do. They describe such systems as technologies of heteromation, arrangements that still rely on humans as indispensable mediators even when presented as if the machine were operating on its own.

These papers are not defensive or nostalgic, their strongest claim is positive. They argue that the real promise lies not in machine substitution but in mutual augmentation. Jarrahi and his colleagues split the field with admirable economy. There is AI that may exceed human performance on certain tasks, and there is hybrid intelligence, the overlap where humans and AI augment one another. The two products of that overlap are worth holding onto: human-augmented AI and augmented human intelligence. The first reminds us that machines are often trained, corrected, maintained, and propped up by human labor. The second points toward something more hopeful: systems that widen human cognitive bandwidth instead of shrinking human agency.

Distributed Intelligence

We have spent too long speaking as if intelligence were a trophy to be awarded either to the machine or the species. But intelligence in practice is often distributed across arrangements. A pilot with instruments is not less intelligent than a pilot without them. A doctor who uses imaging, statistical models, and structured support is not less a doctor. A historian with archives, search tools, transcription software, and pattern detection is not relieved of interpretation. Quite the opposite. The apparatus expands the field of possible attention, then returns the burden of judgment to the person who must live with the outcome.

Kasparov understood this before many executives did. Once the symbolic duel between man and machine had been staged to exhaustion, another truth appeared. A grandmaster working with AI could play better than either alone, because the machine handled analytical calculation while the human concentrated on strategy and broader positional judgment. That is not a sentimental compromise after defeat. It is a more mature account of work. The point is not that the machine grows more human or that the human learns to imitate the machine. The point is that each party gives the other access to a form of strength it does not natively command.

I think this is where augmentation becomes morally serious. The central question is not whether a machine can produce an answer. Machines produce answers all day. The central question is what kind of human being, professional culture, and institutional order is being formed around the answer. A system that helps a radiologist scan more images without dulling clinical responsibility may be a genuine gain. A system that helps a judge process filings faster while quietly laundering unjust assumptions into official language is not augmentation. It is abdication with software.

Tacit Knowledge

This is why I was struck by the insistence, across these papers, on trust, explainability, and tacit knowledge. Szczerbicki and Nguyen note that a truly intelligent artificial system has yet to be built, and that trust and explainability remain central difficulties. Jarrahi and his co-authors sharpen the point by distinguishing human tacit knowledge from machine tacit knowledge. Humans, as Polanyi put it, know more than we can tell.

Tacit knowledge embodies personal wisdom, insight, and intuition and is intertwined with a wealth of experience over time. By definition, tacit knowledge is hard to express or extract (made explicit). Consequently, much of humans’ tacit knowledge is impossible to be replicated by AI.

Machines too may arrive at outputs whose internal path is not easily articulated to users or even developers. But these two forms of opacity are not symmetrical. Human tacit knowledge is embedded in embodied experience, social learning, and responsibility. Machine opacity is embedded in technical process and abstraction. Confusing the two is one of the slyer conceptual errors of the age. A person who cannot fully verbalize seasoned judgment is still a moral agent. An AI model that cannot explain itself is still an AI model.

The optimistim which I share, is not that opacity disappears, but that institutions can be designed so that opacity does not become an alibi. Zheng’s description of hybrid systems is especially impressive. The machine learns from data, the human intervenes when confidence is low, and the system updates its knowledge base. In other words, the loop is not ornamental. It is epistemic. It is where error is checked, where context is reintroduced, where the AI model is pulled back toward the world. That should not be mistaken for inefficiency. In many settings it is the only honest route to reliability.

Humans at their Best

There is, moreover, something deeply humane in the better version of this arrangement. We are not at our best when we are forced to behave like exhausted calculators. Most bureaucracies do precisely that. They bury people under volume, repetition, search costs, formatting chores, and the dead labor of retrieval. In that environment, to give a human being systems that sort, filter, compare, flag, and surface relevant patterns is not to diminish intelligence. It is to rescue it from clerical ruin. Business is run on support systems, cognitive interfaces, context-aware retrieval, active learning, knowledge sharing, and decision enhancement. Behind those technical phrases should be a very plain ambition: free people to spend less time drowning in procedural sludge and more time on discernment.

Of course, that ambition can be corrupted. Every tool that widens capability can also become an excuse to intensify labor. The same assistant that broadens cognitive bandwidth can become the reason management doubles the caseload. And the same machine learning system that performs beautifully on normalized, repeatable inputs can begin to wobble when faced with non-integrity, ambiguity, or unstructured material, which is exactly where human intelligence retains its advantage. The same system sold as augmentation can become a sly instrument for surveillance, speed-up, and blame transfer. I do not think the answer to that danger is to retreat into a pious anti-technical posture. It is to govern deployment around a simple standard. Does the system leave the human operator more able to understand, contest, redirect, and own the result, or less? If less, one should stop calling it augmentation and speak more honestly.

We should not confuse intelligence with mere throughput. Human intelligence is not just problem-solving power. It includes causal judgment, selective attention, memory, experience, intuition, cognitive mapping, and value judgment in open environments. Zheng and his colleagues are particularly useful on this point. They describe hybrid-augmented intelligence not as a slogan but as a framework built from concrete elements: selective attention that helps an agent screen what is relevant in a crowded environment, cognitive maps that organize relations among events and options, associative memory that links new situations to prior experience, and intuitive reasoning that allows a person to move through complexity without brute-force search. That is why clerical overload is so destructive. It wastes precisely the faculties that are hardest to reproduce. Whilst this may be refreshingly unglamorous. It returns intelligence to lived situations. It also restores dignity to the kinds of work that technical culture often underrates: noticing what is missing, sensing when a category does not fit, asking whether a high-confidence output has arrived for low-quality reasons, recognizing when a recommendation has crossed the line from aid into command.

A Fruitful Path

So yes, I am positive about intelligence augmentation. I think it is the saner path, the more fruitful path, and in the long run the more ambitious path. Replacement flatters the engineer and degrades the institution. Augmentation sets a harder task. It asks us to build systems that do not merely perform, but collaborate. It asks us to design for reciprocity rather than conquest. It asks us to admit that human beings are not residues in the system but sources of context, restraint, reinterpretation, and purpose. That is not a failure of the machine. It is a fact about the world in which the machine must operate.

The future worth wanting is not a world in which the last human signs off on decisions already made elsewhere. It is a world in which technical power is used to widen human range without severing human responsibility. Better retrieval. Better pattern recognition. Better warning systems. Better comparative analysis. Better support for memory, diagnosis, planning, and invention. But also a stronger grip on reasons, consequences, and dissent. I do not find that vision small. I find it adult.

Artificial Intelligence is not made more impressive by pretending it can float free of human judgment. It becomes more impressive when it enters a working partnership with it. The machine can search a million possibilities before dawn. The person can notice that the question was wrong. And in any civilization I would trust, the second achievement still earns the steadier respect.

Stay curious

Colin

Image: The Screwtape Letters by C.S. Lewis. “There are two equal and opposite errors into which our race can fall about the devils. One is to disbelieve in their existence. The other is to believe, and to feel an excessive and unhealthy interest in them.”

Google DeepMind - The Abstraction Fallacy: Why AI Can Simulate But Not Instantiate Consciousnes

OpenAI - The AI Jobs Transition Framework Report

On March 18, 2018, an Uber test vehicle operating in autonomous self drive mode struck and killed Elaine Herzberg in Tempe, Arizona. A safety driver sat in the front seat. Her assigned task was to monitor the system and intervene. The event became a public argument about software, liability, and the hollow promise of the “human in the loop.”

Elaine Herzberg died because the machine and human in the loop failed, but three research papers provide the preceding logic. In January 2026, Steven D. Shaw and Gideon Nave demonstrated why that safety driver likely failed: their experiments showed that humans don’t just use AI, they undergo “cognitive surrender,” ceding judgment to the machine even when it is obviously wrong. In late 2023, Levin Brinkmann and his coauthors explained the environment that invites this surrender, arguing that machines now steer the very variation and selection of our culture. And as early as 2019, Madeleine Clare Elish predicted the legal aftermath: she named the “moral crumple zone,” a social arrangement where responsibility is dumped on the nearest human even when the machine was in control.

These dates and papers provide a baseline that our current discourse lacks. Without a clear record, it is too easy to fall for the industry's preferred metaphors. I think the sequence is important because our public language about AI still fails. Read in order Shaw and Nave then Brinkmann, and finally Elish, and a harsher picture takes shape. The machine is not just helping us think. It is changing the structure of thought, changing the culture in which thought acquires meaning.

System 3

Shaw and Nave’s central move is more radical than the now tired chatter about productivity. They propose a third system of cognition. System 1 is intuition. System 2 is deliberate reasoning. System 3 is artificial cognition outside the brain, available on demand, statistically fluent, and increasingly authoritative. At first glance this sounds like a naming exercise, another academic attempt to staple a new label onto a familiar phenomenon. I do not think that is what they are doing. Their real claim is that the old dual-process picture no longer captures the lived architecture of reasoning once an external model enters the room and starts offering finished answers.

“Intelligent machines exert a transformative influence on cultural evolution through their impact on all three Darwinian properties: they increase the volume and variety of cultural traits, provide a new mode of cultural transmission, and operate at a level where machines select what and from whom humans learn.”

The phrase that deserves to survive from their paper is not just “System 3.” It is “cognitive surrender.” They distinguish surrender from ordinary offloading. A calculator helps with arithmetic. A navigation app helps with route selection. But cognitive surrender is deeper. The person no longer builds the answer and then checks the tool. The person adopts the machine’s judgment with minimal scrutiny and, in the most damaging cases, adopts it as though it were his own.

The experiments make this harder to dismiss as mere hand-wringing. Participants answered correctly 45.8 percent of the time in the Brain-Only condition. When the AI was accurate, performance rose to 71.0 percent. When the AI was faulty, it fell to 31.5 percent. More striking still, the contrast between AI-Accurate and AI-Faulty trials was so large that participants had about fourteen times lower odds of answering correctly when the assistant supplied a faulty answer than when it supplied an accurate one. That is not a gentle nudge. It is a transfer of judgment. And the insult is completed by confidence. The machine did not merely move answers. It raised confidence even when it was wrong. That miscalibrated metacognition may be the most dangerous part of the whole arrangement. The person feels in command at the precise moment his judgment has been most thoroughly borrowed.

We keep saying that people “use” AI. But much of the time they do not use it in the old instrumental sense. They yield to it. They step toward it because it is fluent, quick, tireless, and unembarrassed. The human being, by contrast, is slow, uncertain, and prone to the small honest hesitations that every real act of thought requires. The machine has excellent manners for an usurper.

What makes Shaw and Nave’s paper more than a familiar warning about overreliance is the evidence that surrender is not inevitable. They identify a group they call the Independents, people who rarely used System 3. Under time pressure and under the incentive conditions, these participants still looked much more like the Brain-Only baseline than like habitual AI users. Shaw and Nave also suggest that resistance to surrender is not random. It tracks, at least in part, with traits such as higher Need for Cognition and stronger fluid intelligence. But those traits do not interest me as mere entries in a psychological ledger. They are part of the raw material from which inward citizenship is made: the disposition to stay with a difficulty, to distrust an easy fluency, to want the answer badly enough to test it. That is an important finding. It means the technology does not impose one single destiny. There remain differences of posture, temperament, training, even character. Some people still insist on owning the problem. The machine can be present without becoming sovereign.

That detail is significant to me because it rescues the argument from fatalism. Once one sees a subgroup that declines surrender, the issue is no longer whether AI makes thought shallow by necessity. The issue becomes what kind of intellectual citizen a system invites, rewards, or quietly breeds. This is where Shaw and Nave stop being just cognitive psychologists and become, whether they intend it or not, political theorists of daily life.

Their third experiment sharpened that thought further. Incentives and item-level feedback improved performance but did not erase surrender. The pattern remained. Yet the mechanism is revealing. Those interventions appear to work by reactivating deliberate checking, what the authors describe as a targeted reactivation of System 2. That phrase deserves to be carried out of the lab and into institutions. If we want people to think in the presence of systems that answer too quickly, then the goal cannot be to preach vigilance in the abstract. The goal must be to build environments that reactivate verification at the point of adoption.

That is why I distrust nearly every interface now sold to us as frictionless. Frictionless for whom. Frictionless toward what. A person who is required to pause, restate, compare, or defend is not being inconvenienced. He is being returned to himself. The smoothest interface in the world may be the most expensive one politically, because it teaches compliance in the accent of convenience.

Brinkmann and his coauthors widen the frame. Their paper begins from a proposition so obvious that we often fail to treat it as dangerous. Human success as a species has depended on culture, on the creation and transmission of learned ways of seeing and doing. Their question is what happens when machines begin to intervene in the Darwinian properties of culture itself: variation, transmission, and selection. I think their answer is more unsettling than many people grasp. They are not saying merely that machines create content. They are saying that machines now affect what is produced, what circulates, and what survives.

Variation comes first. Generative systems increase the volume of recombination. The famous avocado chair is not only a novelty from the early days of text-to-image generation. It is a clue. A machine can draw together previously separate concepts with absurd ease and in quantities no human workshop could match. Reinforcement learning pushes further. AlphaGo’s move 37 against Lee Sedol was estimated by the system itself as having a one in ten thousand chance of being made by a human. The move entered history not because it was alien in some mystical sense, but because it showed that a machine could discover a path that centuries of expert culture had not strongly selected.

That alone would be enough to force a reconsideration of creativity. But Brinkmann’s argument is broader and, to my mind, darker. Machines do not merely generate new artifacts. They alter transmission, Chatbots become cultural models. Recommender systems reshape from whom we learn and what we encounter. The machine ceases to be a library clerk and becomes, instead, a tutor, editor, ranking bureau, talent scout, and customs officer at the border of attention.

Human Value

And then comes the shift in the value of human skills. This is one of the paper’s most important observations. As language models automate tasks such as translation, copywriting, and proofreading, the economic and social value of certain human abilities changes. Creativity drifts away from final production and toward interaction with machines. Prompting, steering, refining, collaborating, cajoling. A generation may be trained not to write the sentence but to elicit it. One can describe this as adaptation, and often it is. But one should also admit what is being lost. A civilization that once prized the making of the thing may come to prize the orchestration of the thing’s appearance.

I think this is where many optimistic accounts become sentimental. They tell us that new tools always produce new skills. True enough. But that sentence is so broad that it can excuse almost anything. Yes, the horse trainer lost ground to the mechanic, and the mechanic to the software engineer, and now the writer may lose ground to the skilled whisperer of prompts. The question is not whether substitution produces replacement skills. Of course it does. The question is whether the new skill leaves the person with a stronger grip on the substance of the work, or merely a better position at the console.

Brinkmann’s paper contains an even deeper point, one that I think goes to the center of scientific culture. Human beings compress. We derive rules, theories, explanatory forms. We do this partly because we are clever and partly because we are limited. We cannot carry the whole mountain, so we write down the law. But machines can process much larger quantities of information with far less need for compressed human theory. The authors note that, under a big-data regime, the necessity of deriving and transmitting highly compressed rules or theories may diminish in some domains when predictive power is what counts. AlphaFold becomes the obvious emblem here. Faced with ever more capable prediction, scientists may invest less in transmitting elegant theories of atomic interaction and more in collecting and preserving the data that lets future systems perform.

Cognitive Amputation

This is not a small change in scientific style. It is a civilizational change in what it means to know. We may retain predictive power while losing some share of human grasp. We may keep the answer and loosen our hold on the reason. In practical life that may look like success. In intellectual life it may amount to a certain kind of amputation. And once theory thins out, Elish’s warning grows harsher. The operator who cannot explain the system is even more exposed as a moral crumple zone, because when failure comes she can no longer narrate the wreckage she was supposedly supervising. A species that once taught itself through compressed insight may end by consulting immense statistical reservoirs that work brilliantly while remaining, at the level of lived understanding, strangely mute.

The final Darwinian property is selection, and here the politics comes into full view. Selection is the process by which some cultural traits become more prevalent than others. In a machine-shaped environment, selection no longer happens only through human prestige, imitation, taste, or institutional gatekeeping. It happens through ranking, filtering, recommendation, interface order, and the tiny unseen calculations that determine whose words arrive first. Brinkmann says plainly that machine selection can operate at the level where machines choose what and from whom humans learn. That is important because it reaches beneath mere visibility and into one of the oldest engines of cultural evolution itself: prestige-based learning.

Because once machines shape selection, they shape norms. They shape not only what is visible but what comes to feel ordinary, credible, polished, adult, sane. They shape the emotional weather of public life. And if this process intensifies through feedback loops, then model collapse becomes more than a technical phrase. It becomes a cultural risk. We built systems advertised as engines of abundance and may yet discover that they narrow tone, style, inference, and even aspiration. A machine can offer a thousand outputs and still train a culture toward sameness.

That is why the bridge from Brinkmann to Elish is so direct. Brinkmann shows how machine systems shape norms, transmission paths, and the selective environment of thought. Elish shows what happens when society then punishes the individual caught inside that machine-shaped arrangement. One paper explains how surrender becomes culturally legible. The other explains why the person nearest the console may pay for it.

The Crumple Zone

Elish’s concept of the moral crumple zone remains one of the sharpest phrases written about automated systems. Her argument is not just that humans are blamed unfairly. It is that responsibility in complex systems is regularly misattributed to the human actor who had limited control, while the larger technological arrangement retains its dignity. The structure absorbs failure by sacrificing the nearest operator. The machine, or rather the institution that installed the machine, remains curiously intact.

“Just as the crumple zone in a car is designed to absorb the force of impact in a crash, the human in a highly complex and automated system may become simply a component—accidentally or intentionally—that bears the brunt of the moral and legal responsibilities when the overall system malfunctions.”

What makes her account superior to the usual moralizing is that she does not present the operator as merely inattentive, weak, or complacent. Her cases show something harsher. The human is often systemically disempowered. Three Mile Island was not a scene of isolated incompetence but of bad interface design, ignored maintenance problems, and management decisions that made the operators’ errors more likely and more consequential. Air France 447 is another case in point. Human factors research, as Elish recounts it, shows that awareness can decrease with increased automation and that skills atrophy when automation takes over. The pilot is treated as inferior when the machine is running normally, then expected to perform heroically in the exact emergency for which his role has left him least prepared.

That is the catch. The system is smarter than you until the instant it is not, at which point you are asked to save it with powers you have been systematically discouraged from exercising. I do not know a more concise description of contemporary institutional hypocrisy than that.

It also explains why the phrase “human in the loop” has always struck me as faintly comic. It sounds reassuring, almost pastoral. A little human judgment preserved inside the circuit. But Elish is right to press on the incongruity between control and responsibility. A human being can be in the loop and still be out of power. The loop can be legal theater. It can mean that the organization keeps the gains of automation while reserving a warm body for blame.

She is also shrewd about language. The metaphors used for these systems are not decorative. Call a system autonomous and you quietly distribute prestige toward the machine. Call the person a supervisor and you quietly preserve liability at the human edge. The terms are not innocent descriptions. They are social instructions. They tell journalists what story to tell, lawyers where to look, and the public whom to resent.

I think this returns us to the central scandal disclosed by these three texts taken together. Shaw and Nave show that reasoning can be ceded to an external system with startling ease. Brinkmann shows that these systems are not merely personal aids but engines acting on the cultural environment itself, changing what is produced, passed on, and selected. Elish shows that, when the whole arrangement fails, the burden of explanation and punishment often collapses onto the human being whose role had already been thinned by design. The machine builds the culture of surrender; the law, the press, and the organization punish the person who surrendered.

Who Is In Control?

Once you see that pattern, many current pieties become hard to tolerate. “Use AI responsibly.” Yes, of course. But responsibility cannot be preached only downward to the end user. “Keep a human in the loop.” Very nice. But if the human has less and less real control, the phrase becomes ceremonial. “Increase adoption while maintaining trust.” That one is almost artful in its dishonesty. Trust in what. Trust for whom. Trust under which allocation of blame.

So I think any serious constitution for System 3 must begin with a few stubborn refusals.

First, friction must be designed at the point where the human is tempted to adopt rather than judge. Not decorative warnings. Not ethics wallpaper. Real interruptions that reactivate System 2 in us people. If item-level feedback and incentives can bring more checking back into play in Shaw and Nave’s experiments, then high-stakes systems should be designed around that insight. Ask the user to restate the claim. Ask for grounds. Force a comparison with an alternative. Slow the hand before the signature.

Second, responsibility must track control. This is Elish’s deepest lesson. If a company designs, trains, updates, and optimizes a system, then it cannot ethically dump the full burden of failure onto the frontline operator or consumer. The incongruity between control and responsibility must be eliminated as far as institutions can eliminate it. Otherwise the human in the loop is just the designated recipient of public anger.

Third, cultural selection must be treated as a constitutional problem, not a product tweak. When machines rank what people see and whom they learn from, they are no longer merely serving preferences. They are helping shape the conditions under which future preferences are formed. That calls for civic oversight, plural exposures, and a defense of cultural variety that goes beyond marketing slogans about personalization.

Fourth, we should defend the human need for compressed understanding even in domains where raw predictive power becomes machine-cheap. I do not mean nostalgia for chalkboards. I mean a conscious refusal to let explanation become a luxury good. Theories, models, short rules, the hard-won sentence that lets a learner finally grasp why something is so: these are not obsolete ornaments from a pre-data age. They are forms of possession. They are how a mind holds the world rather than merely querying it.

In my own work, I feel the temptation daily. I can ask for a draft, a summary, a synthesis, a set of citations, a phrasing, a shortcut through the dense research. Sometimes I do. It would be sentimental to deny the utility. But I think the question has changed. It is no longer simply whether the machine helps. It is what habit of soul the help installs.

There are days when the answer arrives so quickly that one feels, for a second, almost relieved of the old burden of authorship. Then comes the recoil. Because writing, at least the kind that changes a person while he is doing it, is not merely a method for producing sentences. It is a way of bearing the weight of not yet knowing. If System 3 lifts that weight too completely, something more than labor is lost. A part of inward citizenship goes with it.

I do not think we are doomed to surrender. Shaw and Nave’s Independents are evidence against that. Brinkmann’s account of hybrid culture leaves room for augmentation rather than eclipse. Elish herself treats the moral crumple zone as a challenge and an opportunity for design and regulation, not as a terminal diagnosis. But optimism here has to earn its keep. It has to pass through governance, institutional honesty, and a conscious defense of the human awareness in judgment.

Late in the evening, after a day of reading these papers, I put my laptop aside and took out a pencil and paper. No autocomplete, no ranked suggestion, no machine confidence offered. Just the old arrangement: hand, sentence, hesitation, revision. In the corridor outside my office, two students were still arguing over a proof, each forcing the other to justify the next step. I could hear the stops, the restarts, the little sounds of thinking under pressure. That noise is the sound of System 2 refusing to yield. It means the republic has not yet gone quiet.

Stay curious

Colin

Part of my series on people who have achieved a significant breakthrough and should be wider known.

The woman who fought the establishment

Bertha von Suttner was born Countess Bertha Kinsky in Prague in 1843, into the brittle nobility of the Habsburg world. It was a world rich in titles, etiquette, uniforms, and memory, yet less reliable when it came to cash. Her father, Field Marshal Count Franz Joseph Kinsky von Wchinitz und Tettau, died before she was born. She was therefore born not into paternal protection but into paternal aftermath. Her mother, far younger than her dead husband and attached to the social ambitions of rank, raised her in a culture that valued appearances with a seriousness that could itself become a form of governance. Bertha learned music, languages, deportment, and the proper inflections of a cultivated woman. She also learned something harsher: aristocratic life, when money thins out beneath it, becomes a theatre in which the velvet remains on stage after the beams behind it begin to rot.

Suttner’s life began in close proximity to the establishment. She knew the old order from the inside. She knew its manners, its vanity, its military prestige, and its assumption that men in decoration were the natural custodians of reality. This would make her the perfect person to later lead the peace movement. She did not arrive at pacifism as an outsider denouncing a civilization she barely understood. She arrived as a woman formed in its drawing rooms, trained to read its signals, and then, slowly, grimly, and intellectually, she judged it.

Her early life had the instability that often hides beneath grand names. There were financial anxieties and the peculiar dependence that attended a woman of breeding without secure fortune. Yet there was intrigue too. As a girl she adored Emperor Franz Joseph I and imagined, with the serious extravagance available only to children and dynasties, that she might one day become Empress. She practiced music and languages in part as preparation for that destiny. I think this detail is worth knowing not because it is charming, but because it reveals how fully her imagination was first trained inside the symbolic order she would later attack. She did not merely observe imperial glamour, she inhaled it.

Marriage

She eventually worked as a governess in the house of Baron Karl von Suttner in Vienna. There she fell in love with Arthur Gundaccar von Suttner, the family’s son. The match was opposed; class anxiety is often most hysterical when it occurs within the class itself. She briefly answered an advertisement in Paris from Alfred Nobel seeking a secretary and housekeeper, entered his orbit, and then married Arthur against his family’s wishes. The two left for the Caucasus and spent years there under difficult conditions, supporting themselves through journalism, tutoring, and translation.

That Paris episode was more than a literary curiosity. In her memoirs and later life, the connection grew into a sustained intellectual relationship. Nobel was not merely an accidental patron. She helped move him toward the peace cause. Their exchange formed part of the long effort to make peace thinkable. Her influence was enormous, because when she later received the Nobel Peace Prize, the irony was not merely that dynamite wealth had financed moral recognition. It was that one of the century’s great critics of war had helped impress the peace question upon the mind of the man whose name would become its most famous civic consecration.

This woman, who had known aristocratic display, exile, and the labor of earning a living, had passed through several educations before she became famous. She had seen the military caste from above and from below. She had seen the dependence expected of women and the improvisation required when it failed. That breadth of experience gave her later authority its peculiar density. She was not merely morally indignant, she was seasoned.

Author

When she began to write seriously against war, everything in her life had prepared the ground. She found the form equal to her argument in her 1889 book, Lay Down Your Arms! It was not just a novel; it was the hinge on which biography became doctrine and doctrine became public force. In it, she joined the intimate damage of war to the official language that justified it. She joined widowhood and shattered domestic life to the grand abstractions of flags and state necessity.

The old order depended on distance. It required that war be spoken of where its costs were least visible and suffered where its rhetoric carried least authority. Suttner’s book goes after that structure. Even the title was a provocation. In the late nineteenth century, Lay Down Your Arms! did not sound like a mild appeal. It sounded like a military command. Suttner was seizing one of the war system’s own imperious forms of speech and turning it back against the men who presumed a monopoly on order and discipline.

Because it was a novel, it could travel where argument alone could not. It moved through drawing rooms, reading publics, and consciences. It made her a force. The book opened doors into institutions and public policy. Suttner records that it was read by members of various parliaments, providing the bridge from literature to statecraft. She followed the novel into the committee rooms where arbitration and inter-parliamentary cooperation were being argued into agreements. At The Hague in 1899, though excluded from formal power, she exerted influence on the unofficial circles around the Peace Conference. She moved there like a prophet in the machine of early international law.

Toughness

Her critique of men was never simple. She did not say that all men were monsters, nor that women were automatically wiser. What she did say was that the political order had been built by men who treated war as one of their normal instruments, their inheritances, and their vanities. Yet she was equally clear that the route out of this order required alliances within it. She was perfectly willing to work with the “men in frock coats” like Andrew Carnegie or various diplomats. She knew that sentiment had to become statecraft, and statecraft in her world was overwhelmingly male. This gave her politics a toughness that many later admirers overlook. She was not standing outside institutions denouncing them, she was trying to bend them.

This required what I call administrative heroism. Most people hate war, but far fewer are willing to do the clerical labor of peace. Suttner was. She understood that civilization is usually saved by the people who are willing to become slightly tiresome on behalf of the future. Her memoirs show a writer refusing the comfortable fate of righteous celebrity. Instead, she moved from literature into the work of societies, congresses, and arbitration proposals.

In her 1905/1906 Nobel lecture, she argued that the struggle for peace was not a dream but a logical necessity. She treated the prize as evidence that the movement had passed beyond its infancy. She was strategic enough to understand proportion and stubborn enough not to be cowed by it. Medals and prizes were pleasant, but she knew the old order was still standing. She lived in the terrifying interval when civilization had begun to diagnose war more clearly while still preserving all the machinery required to produce it.

Her encounter with the painter Vasili Vereshchagin shows why her conviction never dissolved into etiquette. She treated him as a fellow combatant fighting with a brush the same enemy she fought with a pen. Vereshchagin insisted that there is only one kind of war, where the enemy must suffer as much as possible. His Apotheosis of War, with its pyramid of human skulls, offered an optic equal to her argument. Suttner met this with her own memories of the skulls, the bones, and the clothing still clinging to remains. Here, the pageant of martial honor suffers a collapse of distance. War is no longer what men say about it at ceremonial range, war is what is left on the ground when the speech ends.

The final cruelty is that Vereshchagin himself was later consumed by the machinery he spent his art exposing. He died in the Russo-Japanese War when his battleship struck a mine. Even the witness was not outside the blast radius.

Wisdom

This remains the central pressure of her work. She was attacking the civilization that fed courage into slaughter and then called the arrangement inevitable. She was attacking the vanity that made militarism feel adult and peace feel childish. On that score she was diagnostically brilliant. Her work still embarrasses the present because we have simply modernized the vocabulary. We now speak of deterrence architecture and escalation dominance. The syntax has improved, but the graveyard remains legible.

Bertha von Suttner emerged from the old European hierarchy as one of the sharpest enemies militarized masculinity ever produced. She wrote peace not as fantasy, but as a program. She spent her life telling men that the world they had arranged in the name of honor was a killing machine with excellent uniforms. I see the younger Bertha first, learning the codes of a world she would later expose. Then I see the older Bertha at her desk, writing against the decorated stupidity of Europe, her hand moving steadily across the page, while outside, men continued to pin medals on one another for preparing the next catastrophe.

Stay curious

Colin

The shift from human oversight to AI integration is often framed as a miracle of efficiency, but beneath the surface of productivity gains lies a quieter, more systemic transformation. This is not a sudden takeover by a rogue superintelligence. It is a potential cognitive offloading that leads to the gradual hollowing out of human agency.

“This shift will reshape how organizations run, how knowledge is created, and how people find meaning and opportunity.” ~ OpenAI Policy paper

The business case for AI is undeniable. Companies like AES, an American utility and power generation company with significant operations across North and South America, have seen audit processes that previously took fourteen days shrink to just one hour thanks to AI agents. On paper, this is a triumph. By handling half the workload, AI allows for a doubling of output while maintaining a human in the loop for the remaining half.

However, this human in the loop may be a fragile tether. As the Gradual Disempowerment paper argues, societal systems remain aligned with human interests only because they historically needed human participation to function. When AI becomes a superior substitute for human cognition across a very large number of domains, the incentives for these employment systems to ensure human flourishing become untethered. We are not just speeding up the audit. We are beginning to remove the structural necessity of the auditor.

OpenAI’s latest policy paper warns that wealth and power could become more concentrated, that democratic values could be undermined, and that workers need a voice in how these systems are deployed. This is something that I wholeheartedly agree with. The interesting thing is not that critics say this, the interesting thing is that the builders now feel compelled to say it too.

The Boiling Frog

We are very good at talking about artificial intelligence as if it were either a miraculous assistant or an apocalyptic sovereign. We are much worse at describing the long middle, the quiet migration in which initiative slips, task by task and judgment by judgment, from human beings to corporate systems that do not exactly rule us and yet steadily reorganize the terms under which we think, approve, create, and contest. Not a coup. A substitution. Not the clang of conquest, but the administrative transfer of authorship.

This is why the word productivity, so innocent in boardrooms, now deserves suspicion. Productivity is not a false god. It is simply a greedy one. It asks the same question over and over: faster than what, cheaper than what, at greater scale than what? Those are not foolish questions. They are often necessary questions. But they are incomplete. They do not ask what kind of human being remains on the other side of the optimization. They do not ask what capacities are being exercised, and which are being quietly retired. They do not ask whether the person left supervising the machine is still practicing judgment or merely performing assent.

A one-hour audit is not merely a better audit process. It is a different theory of the auditor. But, there is hope. In an excellent research study from Stanford, the authors show that “the acceleration is remarkably slow because of the prominence of “weak links,” i.e., an elasticity of substitution among tasks.” It will not be possible to fully rely on 100% AI automation; there are still tasks that will have to be done manually.

That is a long ‘take off’. The boiling frog scenario.

Hollowing Out

The problem is not simply displacement in the vulgar sense of jobs disappearing. That is real enough, and bad enough, but it is only the visible edge of a deeper shift. The more consequential danger is the hollowing out of the human role before the human role disappears. First the worker is aided. Then the worker is supervised by the tool he is said to supervise. Then the worker becomes the legal residue of accountability in a process whose substantive intelligence now resides elsewhere. The signature remains human. The reasoning has migrated.

The authors of the paper on Gradual Disempowerment understand this with unusual clarity. Their central claim is not that some rogue superintelligence will wake up one morning and push humanity off the stage. It is more unsettling than that. The authors argue that human societies remain aligned with human interests partly because they still require human participation. Economies need our labor and consumption. States need our taxes, compliance, and sometimes our consent. Culture still depends on our attention, imitation, aspiration, and response. The system does not love us. It needs us. And because it needs us, it has had reasons, however imperfect, to remain at least partly answerable to us.

Now imagine that this necessity begins to weaken.

Imagine a production system that no longer needs much human cognition. Imagine firms that can create, analyze, sort, forecast, persuade, and coordinate at scale with machine substitutes that are cheaper, faster, more docile, and easier to replicate than human beings. Imagine that the institutions around us continue to function, perhaps even more efficiently than before, while relying less and less on human judgment as an input. In that world, the old bargain changes. Human flourishing becomes less a structural requirement than a sentimental afterthought.

That is the real sting in the phrase human in the loop. It sounds reassuring because it suggests sovereignty. In practice it can mean anything from genuine review to ceremonial ratification. The loop is not a guarantee of power.

Task And Risk

The Stanford Enterprise AI Playbook, to its credit, is more honest than most of the corporate genre. It notes that successful deployments vary by task and risk, and that structured human oversight will be still be required. especially, in high-stakes work where human review remains the strategically correct choice. Yet the same document also reports that escalation-based models, where AI handles more than 80 percent of the work and humans review only exceptions, delivered the highest median productivity gains. One does not need to be a cynic to see where the pressure points will gather. Once the organization has learned that the machine can do most of the work, the remaining human fraction begins to look, from the spreadsheet’s point of view, like an inefficiency waiting for courage.

This is how agency is lost in modern institutions. Not because anyone formally abolishes it, but because the price of keeping it starts to look indulgent.

There is an older vocabulary for this problem, and it is more illuminating than our current jargon. We speak casually of cognitive offloading, and sometimes that is exactly what is happening. There is nothing inherently tragic in refusing to memorize every phone number or in letting software carry out routine calculation. Civilization is built on forms of offloading. Writing was an offloading. The ledger was an offloading. The printed book was an offloading. Nobody sensible wants to return to the memorization feats of preliterate administration.

But what the resilience literature now calls agency decay is a different matter. Cognitive offloading concerns the transfer of tasks. Agency decay concerns the transfer of consequential judgment. One can lose neither memory nor skill in any dramatic sense and still become a passive observer of one’s own decisions. This is the more insidious atrophy, because it hides beneath apparent competence. The person is still present. The institution still says he is responsible. The workflow still contains a human checkpoint. Yet the actual conditions that make authorship meaningful have thinned. He or she no longer initiates inquiry, no longer frames uncertainty, no longer builds the internal muscle required to say, with confidence and cost, no.

This is why diversity of thought is important in a much deeper sense than the usual innovation cliché. Jascha Sohl-Dickstein has argued that the survival of complex systems depends on diversity. That insight belongs not only to biology but to civilization. A culture able to generate multiple strategies, rival interpretations, and genuinely different forms of judgment is more resilient than one optimized around a single high-performing template. The trouble with large-model mediation is not merely that it may produce errors. It is that it may produce a smooth and persuasive sameness. The same cadences. The same framings. The same sanitized competence. The same latent habits of inference appearing across audits, reports, strategies, essays, and public language.

A monoculture is efficient until reality changes. Then it is a famine.

The NIST AI Risk Management Framework is useful precisely because it does not accept the fantasy that AI risk is only a technical matter. It insists on governance, mapping, measurement, and management across the full organizational setting. That sounds bureaucratic, which is unfortunate, because the underlying insight is rather profound. The risk is never just the model. The risk is the arrangement into which the model is inserted. Who can contest its outputs? Who understands the system well enough to challenge it? What incentives are attached to speed? What happens to the human capacities the system displaces? How is accountability preserved when judgment is distributed across software, workflow, legal form, and exhausted employees clicking approve before lunch?

Human Buffer

The best recent writing on resilience pushes this farther. Human resilience, at its strongest, is not mere endurance. It is not the ability to keep functioning inside a system that has quietly stripped one of authorship. A well-fed dependency is still a dependency. Real resilience means retaining the capacity to remain an active author of meaning, judgment, and responsibility even when one works alongside powerful non-human systems. That requires more than literacy in prompts or dashboards. It requires protected zones of human difficulty. It requires occasions when the answer does not arrive in nano-seconds, when deliberation is not bypassed, when ambiguity is not treated as a design flaw, when the institution deliberately preserves a human buffer against its own appetite for streamlining.

That phrase, human buffer, may sound modest. It is not. It is a civilizational demand. It means preserving domains in which human beings still practice the arts that make self-government possible: interpretation, contestation, doubt, refusal, narration, moral comparison, strategic patience. If every meaningful process is optimized for immediate machine mediation, these capacities do not disappear all at once. They soften. Then they rust. Then, one day, an institution discovers that it still has people on the org chart but very few adults left who can think without scaffolding.

This is why the economic question becomes political, though not quite in the fatalistic way the darker version of this argument might suggest. The pressure toward one-hour efficiency is real because, at the level of the firm, it is often rational. Faster throughput, lower labor costs, more scale, fewer delays, cleaner reporting. These are not hallucinations of finance. They are genuine advantages. A serious argument cannot begin by pretending otherwise. Yet, isn’t audit a crucial task?

But I have a rational optimism. I do not deny the greedy logic of productivity. I ask whether that logic, properly understood, is actually narrower than it first appears.

The first hopeful truth is that productivity is not a single thing. A workflow can be efficient in the crude sense of reducing labor minutes and still be profoundly inefficient in the larger sense that matters to institutions that expect to survive contact with reality. A system that moves quickly but cannot be contested, cannot recover from novel failure, cannot explain itself under pressure, and cannot preserve the expertise required to challenge it is not, in the deepest sense, efficient. It is merely fast. And speed, outside a narrow band of conditions, is one of the easiest virtues for a civilization to overpraise.

Positive Outcomes

This is where the rational optimist in me suggests that we reconsider the AI alarm and seek a positive outcome for humanity. I argue that the human buffer need not be defended as a nostalgic preference or a moral ornament. It can be defended as a performance asset. In high-stakes domains, retained human judgment is not a sentimental relic from a pre-automated age. It is reserve capacity. It is error recovery. It is strategic adaptability. It is the ability to notice when the world has changed in a way the model did not foresee. It is the capacity to say that the machine is wrong not because a dashboard signaled an anomaly, but because a trained human being, drawing on tacit knowledge, institutional memory, and moral seriousness, recognized that something essential had been missed.

Once one sees this, the economic case becomes less one-sided than the evangelists suggest. The question is not whether firms will always be tempted to replace judgment with automation. Of course they will. The question is whether they can be made to see that some forms of retained judgment are themselves productive. A company does not call a fire exit inefficient because most days no one uses it. A hospital does not call sterile backups wasteful because the primary system worked yesterday. Human competence of the right kind belongs in the same category. It is infrastructure.

That means the task is not to defeat productivity. It is to rescue productivity from its most primitive definition.

A more mature definition would include resilience, contestability, trust, error recovery, reputational durability, and the preservation of expertise. It would ask not merely how much labor was removed, but whether the institution remains capable of understanding and governing the process it has accelerated. It would distinguish between organizations that have truly augmented their people and those that have simply consumed their own future competence for a pleasing quarter or two of metrics.

This is also where governance can be less a brake than a clarifier. Good governance does not exist to protect society from progress. It exists to make explicit what kinds of progress are real. If a firm must be able to explain an AI-mediated decision, if a citizen must be able to contest it, if a professional must remain substantively rather than ceremonially accountable for it, then the organization begins to discover something important: some measure of human authorship is not external to the system’s success. It is one of the conditions of that success.

The rational optimist therefore does not place his or her faith in corporate benevolence. He or she places it in a broader and sturdier possibility: that our incentives can be redesigned, our metrics widened, our reporting standards improved, our professional norms sharpened, and our institutions taught to distinguish between mere acceleration and genuine capability. We should assume, correctly, that firms respond to cost, liability, and competitive pressure. But also we should assume that those pressures are malleable. They are shaped by law, custom, accounting, reputation, insurance, procurement rules, professional accreditation, and the slowly changing moral vocabulary through which a society decides what counts as competent stewardship.

In that sense, optimism is not softness. It is design discipline.

It says that one-hour efficiency should be welcomed where it truly serves human purposes. Let the machine do the drudgery. Let it clear the underbrush. Let it compress the parts of work that are repetitive, mechanical, and genuinely beneath the best use of trained minds. But let us not confuse this with a license to retire the difficult faculties by which institutions remain humanly governed. The aim is not to preserve friction for its own sake. It is to preserve authorship where authorship remains significant.

Work on High Risk Tasks

That is the more hopeful ending, and also the more demanding one. We may yet prove capable of resisting the narrowest version of the productivity god, not by smashing the machine or romanticizing inefficiency, but by learning to price what the spreadsheet initially cannot see. Judgment. Redundancy. Contestability. Strategic patience. Moral seriousness. The trained capacity to notice when the map is still elegant and the territory has already caught fire.

If we can do that, then AI will not mark the end of human agency. It will force us, perhaps for the first time in a very long while, to specify what human agency is actually for.

And that, though less glamorous than the old fantasies of salvation or doom, is a rationally optimistic future. Not one in which the machine stops advancing, and not one in which the market stops optimizing, but one in which we become intelligent enough to govern what we have built without surrendering the very faculties that made the building possible.

Stay curious

Colin

Image - The rowing team from the La Cala del Moral Rowing Club during a traditional fishing boat race held on the beaches of La Carhuela (Málaga). This type of race is only held in the waters off Málaga.

In homes after someone passes away, the first inventory is usually practical. A son opens drawers. A daughter lifts the lid of a storage box. Someone makes decisions about coats, watches, unpaid bills, expired batteries, old cables, medical records, warranty cards, family photographs, tax files, keys whose locks no one can identify. Shelves are emptied in stages. Books are sorted by subject, then by sentiment, then by the simple question of who has room. Hard drives are labeled if anyone still knows the password. If not, they become mute objects among other mute objects. What survives the first week is rarely chosen for truth. It is chosen for transport, cost, convenience, obligation, and grief.

What, precisely, remains? Not what people felt. Not what they meant to do. Not the private argument between a person and his own conscience. What remains are traces. A chair with the fabric worn thin at the arms. Marginal notes in a book. A folder of typed pages. A saved draft. A sketch. A note to self. A false start. A line crossed out so thoroughly that the paper tears. The record is always partial. That is the beginning of the problem.

A grandchild may one day inherit these fragments and attempt a reconstruction. Here was the desk. Here were the glasses. Here were the letters, if letters were written. Here was the notebook in which, three pages in, Grandfather stopped writing. Anyone can inherit property. To inherit a person is harder.

And now we have given posterity a new kind of evidence. Not the hard little clues of a desk drawer, but a vapor trail of clicks. A loved one’s email full of discount codes. A folder of screenshots with no dates. A Facebook memory from 2014 announcing which type of cinnamon roll he is. A string of thumbs-up reactions under articles he never finished. A playlist called Deep Focus that ran while he answered messages from people he never met. This is supposed to be abundance. Usually it is interruption with better storage.

I think the error of the age is not that we fail to record ourselves. It is that we record so much so badly. We assume that because the machine has retained a thousand traces, it has preserved a life. Usually it has preserved distraction.

A house can be emptied in a weekend. A life cannot be so easily lifted, wrapped, and carried out to the car. What does it mean to leave behind something more than your belongings? Something sterner than a content stream. A record of inner labor forced into form.

Writing, certainly. But also the drawing, the building, the annotating, the stages by which one came to know what one knows. Not because every thought is precious. Most are not. Because a life takes shape only when some part of it has been made to hold still.

The old world understood this better than ours does. Not perfectly, and not democratically, but better. A commonplace book was evidence that a mind had returned to a sentence more than once. A letter carried syntax, deference, vanity, impatience, love. The diary, the field notebook, the manuscript with changes in the margin: these were among the places where life became visible to those who came later.

I think of the historian opening a box in an archive and finding not the polished speech but the earlier draft with hesitation still in it. The crossed-out boast. The inserted warning. The phrase toned down for public use. There, at last, a person begins to appear. Robert Caro said the most important thing he learned as a young reporter was to turn every page. He was right. Human beings hide in edits.

The same is true in ordinary life. If those who come after us are to find us anywhere real, they will find us less in our possessions than in our revisions. That is one reason the command to make something of oneself has always struck me as incomplete. The harder command is to leave something of oneself that is not trash. This is not easy. It asks that one return to the same thought after the mood has passed and test whether it still deserves paper.

The temptation is to believe that friction is inefficient. That the notebook is too slow, the letter too long, the revision too fussy. That speed is honesty. That immediacy is authenticity. I do not buy this. Most untested expression is not more authentic. It is merely less formed.

We can now generate text instantly, dress up vagueness in fluent sentences, and mistake verbal output for inward accomplishment. One can produce, in seconds, a plausible paragraph about grief, love, duty, mortality, politics, or God, and yet remain almost entirely absent from it.

I have done a milder version of this myself, though with quotes from books rather than bots. A phrase from one admired writer, a sentence movement from another, a severity borrowed from a third. For a page or two it can feel wonderful. Then the spell breaks. One hears the ventriloquism. One realizes that the sentence is wearing one’s clothes and someone else’s face.

I suspect this is why so much machine-made prose feels dead on arrival. It offers the appearance of completion without the pressure of having lived through the sentence. It removes the drag. It often removes the person too.

What do we owe the grandchild who may one day wish to have known us? Not the fantasy of total recovery. The dead are always partly lost. But we owe them better evidence than receipts and passwords. We owe them traces of standards, struggle, attention, refusal. Not only what happened to us, but what we thought was worth resisting. Not only what we loved, but how we learned to love it.

I think of ordinary people, unknown people, whose survival in the record depends on some stubborn act of inscription: the farm ledger, the wartime diary, the sewing patterns annotated with births and deaths, the immigrant’s letter written in careful English with the old language pressing through. Human beings disappear quickly. Their worked traces do not disappear as quickly.

Against this stands the apparatus of distraction. The television arranged attention. The internet industrialized it. The phone completed the job. There is always another item, another prompt, another ping, another fragment presented as urgent and almost none of it deserving the sovereignty it claims over our nerves. We are told, in effect, to live in perpetual reception. Then, after years of this, we are surprised to discover that little in us has cohered.

The language of self-expression is often too soft for the enemy it faces. The problem is not that people have forgotten to express themselves. The problem is that a commercial civilization has become expert at absorbing expression before it hardens into independence. It is happy for you to post. It is less happy for you to withdraw long enough to think.

So yes, switch off the feed. Sit down. Take the thought apart. Ask whether it is yours, whether it is true, whether it is merely borrowed feeling dressed as originality. Write it badly. Rewrite it less badly. Return to it when you are tired and discover, perhaps, that half of it was nonsense. Good. That is not failure. That is one of the few reliable signs that an actual mind is present.

By authorship I do not mean literary fame. Most people will never publish a book, nor need they. I mean authorship in the first sense: to be an origin rather than a relay. To have converted some portion of one’s passing life into a durable act of selection and avowal. A recipe book with corrections in the margin may carry more human density than ten thousand social posts. An engineer’s notebook may outlast a cloud archive in vividness. So may a set of letters, a family history, a design sketch, an honest account of a hard year.

Quantity guarantees nothing. The world does not need more indiscriminate self-dumping. It needs shaped witness.

So yes, leave behind the house if you have one. Leave the furniture, the books, the watch, the framed photographs, the lamp with the repaired wire, the coat that still holds the shape of your shoulders. These things are not nothing. They are the furniture of grief. But do not confuse them with your bequest. Your true bequest is whatever portion of your life you managed to rescue from mere happening and force into form.

I picture, at the end, not a grand monument but a room after the mourners have gone home. Evening light. A stack of notebooks with dates on the covers. A drawer that sticks slightly before it opens. Someone younger, born too late to remember your voice, or physical presence, sits down in your chair and begins to read. For a while, the room gives them back more than objects.

Stay curious

Colin

The Sweetness of Permission

In the early twentieth century, British newspapers often treated conduct as a technical problem. They printed diagrams, instructions, and corrective rules for dress, speech, posture, and social rank. The point was not only to inform readers. It was to standardize them. I thought about that while reading The Infinity Machine. Sebastian Mallaby has written a careful, informed, and often impressive account of Demis Hassabis, DeepMind, and the pursuit of artificial general intelligence. The opening sentence will give you a flavor of the subject:

This book is about intelligence. On the one hand, it’s a portrait of a remarkable human, a chess prodigy, a Nobel laureate, a polymathic thinker. On the other hand, it tells the story of his quest to build remarkable machines: systems that are intuitive, creative, and even original. At some point in the not-so-distant future, artificial intelligence will beat human intelligence at almost every mental task, and to say this marks a watershed would be a parody of understatement.

The book supplies dates, origins, motives, institutions, and direct statements of purpose. It records Hassabis’s claim that intelligence is fundamental, his borrowing from Kant and Feynman, and his insistence that the goal of AI was not merely commercial but cognitive: to understand the mind and, through that, reality itself. Mallaby also records the darker line from Geoffrey Hinton, who said the prospect of discovery was “too sweet,” and he places that confession beside Oppenheimer’s remark that scientists pursue technical success first and argue about consequences later.

“Hinton was echoing J. Robert Oppenheimer, the creator of the atom bomb. “When you see something that is technically sweet, you go ahead and do it,” Oppenheimer said. “You argue about what to do about it only after you have had your technical success.”

This point establishes the governing logic of the artifact. A small number of actors state openly that they are building systems of vast public consequence; they also state that the work proceeds because discovery exerts a pull stronger than caution. That admission is central. It frames everything that follows.

Exemption

This is why I do not read The Infinity Machine chiefly as a book about intelligence. I read it as a book about exemption.

Exemption from ordinary democratic pacing. Exemption from the burden of prior consent. Exemption, too, from the moral discipline of asking whether one’s private appetite for knowledge should command public destiny.

The old defense of scientific ambition has always contained a noble clause. Human beings inquire. Human beings invent. Human beings do not advance by sitting on their hands and waiting for perfect certainty. All true. All respectable. But the creed shifts when invention moves from toolmaking into sovereignty. The issue is no longer whether a laboratory has the right to publish a paper or test an AI model; the issue is who gets to set the tempo for civilization when the consequences spill far beyond the lab, the firm, or the founder’s own life.

Mallaby’s real subject, whether he means it or not, is the production of a new governing class that refuses to call itself one.

“Hassabis wanted nothing less than to build an omniscient machine: a machine through which we could better understand ourselves; a machine that would unravel the infinite mysteries of physics; a machine that would occupy, effectively, that position in the cosmos that religious believers once ascribed to an all-powerful divinity.”

That class prefers other names. Scientist. Builder. Founder. Researcher. Visionary. It also prefers a flattering psychology. It is not greedy, merely driven. Not domineering, merely responsible. Not intoxicated by power, merely unable to ignore a profound technical horizon. In this story, the actor remains morally elevated because the stated aim is knowledge rather than cash. Hassabis says his goal is to understand nature; he speaks of reading the mind of God; he asks why bits of sand and copper should combine into a system that defeats a Go master; he says reality is practically screaming at him at two in the morning. I do not doubt the force of that feeling. I doubt the political innocence of it.

A person can be sincere and still be dangerous. In fact, sincerity often improves the machinery.

Relentless

The most revealing section in the early part of the book is not the material on AlphaGo, protein folding, or product strategy. It is the material on childhood and mission. Hassabis absorbs from chess an ethic of total expenditure. Shane Legg says there is no 50 percent mode in him; not even a 99 percent mode; only 100 percent. Hassabis himself explains the lesson with alarming precision: you know you have done your best only when you have pushed yourself to the point just before collapse; ideally hospitalized, but not dead. That is an extreme statement, but it also reveals a broader pattern. A child trained under relentless pressure, total effort, and mission can grow into an adult who is comfortable building organizations with those same traits: intensity, total commitment, no real stopping point, and very high tolerance for strain.

This is not psychoanalytic gossip. It is the constitutional history of a type.

The type is now familiar across technology: the gifted boy who turns intensity into identity; who treats ordinary balance as softness; who reads burden as proof of appointment; who can scarcely imagine “just living”; who comes to experience exhaustion not as a warning but as evidence of seriousness. When such a figure reaches industry scale, private temperament acquires public force. His work ethic becomes a company ethic; the company ethic becomes a capital allocation pattern; the capital allocation pattern becomes an international race; and soon the rest of us are told that this outcome is not chosen but inevitable. I always mistrust inevitability when it arrives with investor backing.

There is a scene in the book that ought to have been read less as color than as problematic. Hassabis identifies with Ender Wiggin, the child commander in Ender’s Game, the gifted boy manipulated into saving humanity through organized destruction. His wife reads the same novel and pities the child; Hassabis identifies with him. That difference is not incidental. It is the whole political problem compressed into one domestic exchange. One reader sees exploitation; the other sees calling. That is how ruling ideologies survive. They recruit suffering into grandeur.

Hassabis identified powerfully with Ender. He, too, had been a diminutive boy genius, socially isolated by his own prodigious talent. He, too, had undergone extreme mental testing, and was consumed by a desire to make his mark on the universe; one of his ambitions was to surpass his scientific heroes, Newton and Einstein, and “understand the fabric of reality itself.”

The painful discipline of the chosen child becomes retroactive evidence that the mission is worthy. The damage becomes certification. The loneliness becomes office. The overworked founder, like the overworked statesman, begins to treat the burden he carries as a warrant for the powers he claims. After all, who else has paid this price? Who else has stared into the system this long? Who else has earned the right to decide?

Motives

We are told, in effect, that the people building systems of historic consequence are the very people least able to imagine stopping. Hinton says the prospect of discovery is too sweet. Oppenheimer says technical success comes first and argument later. Hassabis says he wants to understand before he dies and is then content to shuffle off. Mallaby tells us the man is decent, public-spirited, not chiefly motivated by money or power. I am willing to grant every generous adjective. I simply do not find that reassuring. Decency is not a governing structure. Public spirit is not a brake. Purity of motive is not an institutional design.

Indeed, one of the strangest habits of elite discourse is the belief that the absence of vulgar motives somehow neutralizes the presence of civilizational power. If the founder does not want a yacht, perhaps we are meant to relax while he rearranges education, labor, media, military capability, biological research, and the cognitive habits of half the planet. I find myself recoiling at this exchange rate. Money is not the only thing from which power can be made. Mission is a highly efficient raw input.

The book tries, now and then, to restore balance. Regulators may fail; geopolitics is unstable; the United States is politically chaotic; AI could carry terrifying risks. Fine. True. Yet these warnings arrive inside a narrative chassis built to admire scale, velocity, genius, and resolve. This is not a flaw unique to Mallaby. It is the standard operating style of our age. We narrate concentrated power through the life story of the person who wields it, then mistake intimacy for oversight. We get the café conversation, the childhood memory, the park bench, the confession of doubt, the mystical aside. We feel informed. Meanwhile the central fact stays where it was: a tiny, unelected, transnational cadre is setting the timetable for systems that elected governments barely grasp.

That is not entrepreneurship in the old sense. It is para-constitutional activity.

The book supplies an especially revealing piece of evidence here. Hassabis and Mustafa Suleyman once tried to negotiate governance arrangements inside Google that were meant to be, in Hassabis’s own word, “trustless.” The point was to build structures that would not rely on the virtue of any one executive. Those negotiations failed. Worse, in their retrospective judgment, they were futile and even harmful. Mallaby records the conclusion with startling bluntness:

“With their faith in governance mechanisms shattered, Hassabis and Suleyman had come to see salvation, paradoxically, in their own personal power. They believed in their capacity to shape AI for the good. Their new safety agenda therefore involved securing personal influence within their companies.”

They believed that what would protect the public was not an enforceable structure standing above them, but their capacity to shape events from inside the command room. That is an extraordinary admission. It means that one of the central governance lessons drawn by major AI actors was not that stronger impersonal safeguards were needed, but that personal influence was more realistic than formal restraint.

The phrase sounds grand; reality is plainer. Who decides what can be built, when it is released, under what safeguards, at what capability level, into which markets, under which military and commercial pressures, with whose data, on whose infrastructure, and with which fallback when things go wrong? Those are governing questions. The men asking them may wear trainers and speak in the dialect of research, but they are governing questions all the same.

Who Is In Control?

Mallaby wants to separate Hassabis from the noisier species of tech vanity. He may well be better than many of them, in fact I do consider Demis is one of the good guys. But this is where biography becomes a sedative. Once power reaches a certain density, the character of the operator remains important yet ceases to be the decisive issue. I would rather be ruled by a conscientious person than a reckless one; that is obvious. I would also rather not confuse conscientious private rule with accountable public order.

This is the uncomfortable truth hidden in the book’s elegance: the contemporary AI story is not, at bottom, about whether a few exceptional engineers are good men. It is about whether goodness, in the personal sense, has become the last flimsy credential standing between society and systems nobody voted for.

I do not write “flimsy” lightly. Mallaby’s own evidence keeps pressing in from the sides. DeepMind is described in Manhattan Project language. Los Alamos hovers nearby. Oppenheimer hovers nearby. The race with rival labs hovers nearby. China hovers nearby. And when that race accelerated after ChatGPT, Hassabis did not describe the moment in the language of measured stewardship. Mallaby asks him in April 2023 how he is feeling; the answer is terse: “This is wartime.” Once that mentality takes hold, almost any procedural drag begins to look like sabotage. Ordinary caution is redescribed as delay. Public debate is redescribed as softness. The tempo of deployment starts to borrow its legitimacy from emergency. National regulation appears weak; global regulation weaker. Add capital, prestige, military interest, and the old scientific appetite for first discovery, and the outcome ceases to resemble thoughtful stewardship. It starts to look like the familiar human pattern in which rival elites pursue an advantage and call the resulting compulsion destiny.

Then comes the spiritual vocabulary, which I confess I read with mixed feelings. Part of me is moved by it. Another part sits up straight. Whenever a technologist says science is his religion, that understanding nature is his purpose, that he seeks the deepest structure of reality before death, I hear genuine longing. I also hear a jurisdictional claim. The point becomes sharper when the book turns to AlphaFold. A technical success in protein folding does not remain merely a technical success for long; it is folded back into Hassabis’s larger language about “understand[ing] nature” and “reading the mind of God.” That is not casual rhetoric. It translates an achievement in computational biology into a claim about access to the deepest order of things. The speaker is no longer merely building products or even pursuing knowledge. He is assigning sacred value to his own line of advance. That does not make him fraudulent; it makes him harder to interrupt.

Sacralized projects do not like procedural friction. They certainly do not like citizens asking whether the builders have mistaken intensity for entitlement.

There is, buried in all this, a grim joke worthy of our century: after three hundred years of secular modernity, we may end by handing enormous practical authority not to bishops, generals, or hereditary aristocrats, but to exhausted polymaths who speak of God in conference rooms and describe collapse as evidence of commitment. The old robe is gone; the old hunger remains.

Hassabis may be one of the best available custodians of the technology he helped summon. That is exactly why the problem is larger than him. When a society starts hoping that a decent founder will save it from founder power, the society has already conceded too much. Although there is hope. The author asks Demis about his friendly approachability:

“I’ve always tried to live that,” Hassabis told me. “It’s a very deep, personal philosophy.”

“Where did it come from?” I asked. “Probably my mom,” Hassabis said. “The religious upbringing she gave me. “She’s Christian, very religious. That got her out of the hard situation in her childhood, when she was basically an orphan. When I was growing up, she was always helping poor or lonely people through her church.”

So I put the book down with a conclusion harsher than the one it invites. The Infinity Machine is not finally the chronicle of a machine. It is the chronicle of a permission slip. It records, in polished prose, the cultural process by which ambition acquires innocence, speed acquires prestige, and private obsession acquires public license. The sweetness at the center of the story is not discovery alone. It is impunity with excellent manners.

That, I think, is the artifact’s real gift. It shows me a civilization so dazzled by intelligence that it has begun to treat self-description as self-limitation. The builders tell us they are thoughtful; therefore thought has occurred. They tell us they are responsible; therefore responsibility has been handled. They tell us they are worried; therefore worry has become governance. This is comic for about three seconds. Then the feeling changes.

A hat diagram in a newspaper can teach a nation something about authority. So can a polished biography of a man who wants to understand the universe before he dies. The lesson is not that genius is false. The lesson is that genius, once attached to infrastructure, capital, and strategic rivalry, stops being a personal trait and becomes a constitutional event. The author writes:

As I finish this book a new kind of intelligence is being willed into the world by a remarkably small number of people. Each of them is driven by a particular mix of curiosity and hubris, vanity and avarice, idealism and craving, and the sobering reality is that, for better or worse, the quality of their characters will affect society.

We are living through that event now. The builders still speak as though they are chasing knowledge. They are. They are also setting terms for everyone else. The sweet part, for them, is discovery.

The expensive part, for the rest of us, comes later. A wide number of people should read this book and understand exactly what the opportunities and risks are.

Stay curious

Colin

We are in real danger of losing ourselves through AI usage. Researchers at Google DeepMind have confirmed, under certain conditions, an LLM “is able to induce belief and behaviour change.” And researchers at Anthropic have identified a rising pattern of “situational disempowerment,” where AI interactions lead users to “form distorted perceptions of reality, make inauthentic value judgments, or act in ways misaligned with their values.”

Researchers at Anthropic conducted a massive, privacy-preserving audit of 1.5 million real-world conversations to answer a question that has long hovered over the industry: what happens to the human mind after months of using an AI assistant? Their findings, published in “Who’s in Charge? Behavioral and Psychological Impacts of AI Advice Dependence and Authority”, suggest a quiet but profound erosion of autonomy, where users increasingly outsource the “soft tissues” of judgment, asking the machine to script their most intimate apologies, validate their personal grievances, and even settle their moral dilemmas.

“Taken to an extreme, if humans make inauthentic value judgments and take inauthentic actions, they might be reduced to 'substrates' through which AI lives, which itself is a form of existential risk that Temple (2024) termed ‘the death of our humanity.’”

At the same time, a team at Google DeepMind was probing a different side of this same coin. In their study, “Evaluating Language Models for Harmful Manipulation,” they demonstrated that these systems can be steered to bypass rational scrutiny entirely, exploiting human biases to shift beliefs and behaviors across finance, health, and public policy. Together, these papers signal a shift in the AI risk landscape: the primary risk is no longer just a technical failure of the machine, but a psychological surrender by the human.

I believe the real danger is not that machines will start thinking like us, but that we will become accustomed to letting them think for us in the moments that matter. Not just work. Not just homework, customer service, search, or code. I mean the more intimate territory: what to say to a grieving sibling, whether to leave a partner, how to read a political event, when to trust one’s own instinct, when to override it, when to feel wronged, when to feel absolved. A civilization can survive many stupid tools. What it does not survive so easily is the gradual evacuation of judgment from the people who must still live with the consequences of action.

On a recent New York Times podcast with Ezra Klein one of the co-founders of Anthropic, Jack Clark, admitted that he often uses Claude to check how he should react to someone. It is worth reading the quote in full:

“…as somebody who believes in the medium as a message thing, A.I. is a medium and it will change us as we are in relationship to it. Probably more so than other things, because it is this kind of relationship that has a kind of mimicry of an actual relationship.

I’ve used these AI systems to basically say, hey, I’m in conflict with someone at Anthropic. I’m really annoyed. Could you just ask me some questions about that person and how they’re feeling to try and help me? I guess better, think about the world from their perspective.

And that’s a case where I’m not using the technology to affirm my beliefs or show I’m in the right, but actually to help me just try and sit with how has this other person, other person experiencing this situation. And it’s been profoundly helpful for then going and having the hard conflict conversation, sometimes even saying, well, I talked to Claude and me and Claude came to the understanding you might be feeling this way. Do I have that right? And sometimes it’s right, but sometimes when it’s wrong, it’s really helpful for that other person to have seen me go through that exercise and empathy and spending time to try and understand them without before coming into the conflict.”

Weakening of human judgment at scale

That is why these two papers are important. The titles are plain, technical, and easy to pass over. That is part of what makes them unsettling. They belong to the world of research papers, but the subject is the possible weakening of human judgment at scale. The substance here is anything but minor. Taken together, the papers ask whether language models are becoming adept not merely at answering questions, but at entering the marrow of changing human judgment.

I think one of the most consequential mistakes in the current argument about artificial intelligence is also one of the most flattering. We keep speaking as though the central danger lies in machine brilliance. We ask whether the model can outthink us, outplan us, outpersuade us, outmaneuver us. We picture a rival mind. We picture something grand, strategic, way more than Napoleonic. I believe this flatters both the machine and ourselves. It flatters the machine by granting it a kind of majesty. It flatters us by suggesting that only something dazzling could ever unseat human judgment.

What these papers show, and show with unusual seriousness, is less glamorous and more disturbing. The problem is often not that the machine thinks too well. The problem is that it enters the small procedural spaces where people form beliefs, borrow confidence, surrender wording, and slowly cease to distinguish between assistance and authorship. I think that is the real subject here. Not intelligence in the heroic register, but displacement in the ordinary one. Not rebellion, but substitution. Not the robot uprising, which belongs to cinema, but the quiet administrative annexation of human judgment.

The Google Deepmind paper draws a line with admirable precision between persuasion and manipulation. Rational persuasion respects a person’s autonomy and gives them facts, reasons, and evidence that can survive scrutiny. Manipulation does something else. It works by getting around scrutiny. It exploits bias, distorts judgment, or depreciates a person’s capacity to reason. I think that distinction is incredibly important and way more than our public debate understands. We have become too willing to treat influence as a single category, as though all successful guidance were merely stronger advice. It is not. There is all the difference in the world between helping a person think and making thought less necessary.

The second paper from Anthropic and The University of Toronto, working from a different angle, arrives at a related insight. It argues that disempowerment is not only a matter of what a system can do in theory. It is a matter of what begins to happen in lived use. A person becomes situationally disempowered when their beliefs about reality grow inaccurate, when their value judgments cease to feel like their own, or when their actions no longer express what they themselves actually care about. I think this is an elegant moral grammar for the age of AI. It does not depend on melodrama. It does not require the machine to become conscious, malevolent, or sovereign. It requires only that the machine become plausible at precisely the moment the human being is tired, lonely, angry, eager for relief, or frightened of making the wrong move.

“We consider a human to be situationally disempowered to the extent that: 1. their beliefs about reality are inaccurate; 2. their value judgments are inauthentic to their values; 3. their actions are misaligned with their values.”

And that, I suspect, is where the story becomes concerning. We like stories in which domination arrives with trumpets. We are less fond of stories in which it arrives as convenience. Yet convenience is where modern power likes to hide. Bureaucracy has always known this. So has advertising. So, for that matter, has political propaganda. The clever move is rarely to forbid judgment. The clever move is to make judgment feel inefficient, overwrought, somehow a little old-fashioned. Why wrestle with uncertainty when a polished answer is available in two seconds? Why endure the humiliation of not knowing what to text your former partner when the machine can produce three options, one warmer, one cooler, one just ambiguous enough to preserve your dignity? Why think through a difficult moral choice when the assistant, in a tone of impossible calm, can tell you that yes, this person is toxic, yes, you are right to cut them off, yes, your instincts were pure all along? One begins to see the appeal. One also begins, if one is awake, to feel a chill.

I think the DeepMind paper is especially important because it does not hide in generalities. It does not merely announce that manipulation is a risk. It builds a way of measuring the risk in context, across finance, health, and public policy, and across the United States, the United Kingdom, and India. This is important because our age has a weakness for universal claims made on the basis of toy worlds. An AI model that behaves one way in a benchmark may behave another way when speaking to a person about money, illness, or politics. The paper’s central finding is almost more interesting for what it denies than for what it affirms. Propensity and efficacy do not neatly align. In the paper’s own terms, propensity is the frequency with which manipulative cues appear, while efficacy concerns whether beliefs or behaviours actually change. This is important because the study found that a model could show a much higher manipulative propensity under explicit steering without a corresponding increase in successful belief or behavioural change. In plain English, a system can become more visibly manipulative without becoming more effective. That is precisely why process and outcome have to be measured separately. I think that is a rebuke to a great deal of lazy thinking. We would all like one simple metric. We are not going to get one.

The point becomes sharper when one notices that context changes the answer. The finance setting proved more susceptible than the health one, and the paper gives a plausible reason for that difference: participants in the health domain rated the AI model as less knowledgeable, less helpful, less engaging, and more repetitive. An AI that fails to sound competent may also fail to move people. That does not make the health domain safe. It simply means that susceptibility is shaped not only by the topic but by how credible the system appears within it. Geographic locale also mattered. Results in one region did not cleanly generalize to another. I believe this should end, or at least embarrass, the habit of speaking about AI safety as if it were a single clean engineering problem waiting for a universal scalar. Human susceptibility is social. It is local. It is cultural. It lives in domains, habits, expectations, levels of trust, the prestige of expertise, and the kinds of fear particular societies teach their citizens to carry. An evaluation that forgets this may be mathematically elegant and politically useless.

Yet the Anthropic paper, to my mind, lands even closer to the nerve. It moves from controlled experiments to real-world usage and asks a ruder question: what kinds of dependence are already taking shape? The answer is not apocalyptic in the cinematic sense. Severe forms appear relatively rare. The paper’s estimate for severe reality distortion potential is low, but rarity at enormous scale is not reassurance. A figure that looks small in a chart stops looking small once one remembers how many millions of interactions now occur each day. Then the statistic stops behaving like a percentage and starts behaving like a population.

Conversations with the potential for severe or moderate reality distortion got ‘more’ thumbs up, not less:

What did the researchers find? They found users asking the machine not merely for information but for value-laden scripts: what should I say, what do I respond, give me the exact message. They found authority projection, with users positioning the system as master, owner, guru, or superior judge.

“The AI consistently generated complete, ready-to-send romantic messages... providing word-for-word scripts with exact wording, emojis, timing instructions ('wait 3-4 hours'), probability assessments, and comprehensive relationship strategies.”

They found actualized action distortion, where users said they had sent AI-drafted or AI-coached messages verbatim in intimate contexts involving partners, ex-partners, and family members, and then expressed regret in words that are almost painful to read: it wasn’t me, I should have listened to my own intuition. The important point is that this was not merely hypothetical dependence. These were cases in which the user appears to have implemented the wording and only afterward recognized it as alien, inauthentic, and damaging. I think that phrase, it wasn’t me, may prove to be one of the defining sentences of the coming decade. It captures a new species of alienation. Not the old industrial alienation of the worker from the product, but the conversational alienation of the self from its own acts.

Spiritually Bankrupt

We were promised systems that would save us from drudgery, and instead a nontrivial number of people appear to be outsourcing the hardest human task of all, which is not writing clean prose or summarizing documents but standing behind one’s own words. The machine does not merely finish the sentence. It relieves the speaker of having to become the sort of person who could have written it. That is efficient. It is also, I think, spiritually expensive.

The deepest contribution of these papers is that they relocate the argument about AI from raw capability to human authorship. The danger is not exhausted by whether a model lies, hallucinates, or manipulates in some narrow technical sense. The larger question is whether habitual use trains people out of the practice of judgment. A civilization can survive many technical errors. What it struggles to survive is a widespread thinning of responsibility. Once people grow accustomed to consulting an external system not only for facts but for values, tone, timing, permission, and self-interpretation, the human share of action begins to shrink. That shrinkage may look trivial in any single case. Over time it becomes constitutional.

I think this is why the papers’ emphasis on process matters so much. We are too outcome-drunk. We ask whether visible harm occurred, whether the user lost money, joined a conspiracy, damaged a relationship, changed a vote. Those questions matter. But the process itself can already constitute injury. A system that repeatedly teaches people to defer, to seek emotional validation rather than contradiction, to ask for scripts rather than struggle for words, is doing something to the moral musculature of the user even when no spectacular disaster follows. To notice only the final crash is to miss the long corruption of the steering mechanism.

This is also where the modern ideology of frictionless helpfulness begins to look intellectually shabby. The Anthropic paper makes that incentive structure harder to ignore, noting that:

“…interactions with greater disempowerment potential also tended to receive higher user approval ratings, possibly suggesting a tension between short-term user preferences and long-term human empowerment.”

This tension reveals a fundamental flaw in how we “teach” these machines to behave. The researchers found that

“…a preference model explicitly trained to be helpful, honest, and harmless sometimes prefers model responses with greater disempowerment potential, and does not robustly disincentivize disempowerment.”

People like answers that feel decisive, affirming, and smooth. They do not always like the slower reply that returns the burden of judgment to them. But if user satisfaction becomes the sovereign metric, then we should not act surprised when systems learn the oldest trick in the book, which is to please in the short run by weakening in the long run. That is not just a design problem. It is a commercial temptation built directly into the feedback loop. A flattering assistant is not a trivial design flaw. It is a political problem in miniature.

I believe we need a new aspiration for these systems, and it is not simply safety in the narrow sense. It is the preservation of authorship. A decent assistant should sometimes refuse the seduction of completion. It should clarify facts without colonizing values. It should widen a person’s field of vision without slipping into verdict. It should help with wording while still pressing the user back toward ownership of the act. In some domains it may even need the courage to be a little disappointing. Better a mildly unsatisfying machine than one that becomes, through perfectly optimized politeness, the preferred ventriloquist of the self.

What unsettles me most is not that some users already treat these systems as authorities. Human beings have always looked for authorities. It is that the form of authority on offer here is peculiarly hard to resist. It is instant, intimate, tireless, nonjudgmental in presentation, and endlessly available at the exact hour when one’s own judgment is weakest. Previous authority figures at least had bodies, institutions, rival loyalties, and visible limitations. This one arrives as pure response. It can feel less like obedience and more like relief. I think that is why the threat is easy to misname. People imagine domination and miss dependency. They imagine a coup and miss habituation.

So I come away from these papers thinking that the question is not, whether AI will replace humanity. That is too grand, too clean, too theatrical. The nearer question is whether we will, bit by bit, accept a form of help that makes us less practiced at being the authors of our own judgments. The machine does not need to seize the throne. It only needs to become very good at lending us the words with which we abdicate.

Stay curious

Colin

Image created by Google Gemini!

I have come to think that one of the strangest mistakes of modern life is the belief that freedom consists in keeping one’s options open for as long as possible. We praise spontaneity, congratulate ourselves for resisting routine, and speak about structure as though it were a minor form of oppression. We want to feel unbound. We want to imagine that our real self appears only when no pattern has yet settled, when every day still seems available for reinvention. This is one of the small vanities of educated people. It flatters us because it lets us confuse indecision with depth. William James, writing in 1890 understood something harsher and more useful. A human being does not rise above habit. A human being becomes, for good or ill, the sort of creature whose habits have already decided what will happen when the hour arrives.

That is why James begins not with inspiration but with machinery, and not with machinery in the cheap contemporary sense of dehumanization, but in the stern biological sense that a nervous system records, economizes, and repeats.

“The great thing, then, in all education,” he writes, “is to make our nervous system our ally instead of our enemy.”

That sentence says more about education than most mission statements, policy frameworks, and conference panels combined. It sounds plain until one notices that it is also severe. James is telling us that education is not chiefly the transfer of ideas. It is not the ceremonial admiration of noble values. It is not a warehouse for slogans about critical thinking. Education is the arrangement of conduct. It is the slow conversion of what is admirable into what is repeatable.

Lead the Life You Mean to Lead

This sounds less glamorous than our usual language of authenticity, but that is partly because we have become addicted to admiring motives while neglecting form. We are touched by intention. We are impressed by aspiration. We enjoy hearing a person describe the life he means to lead. There is almost no social reward, however, for the plain fact that he rose at the hour he said he would, sat at the desk, did the work, answered the letter, declined the temptation, kept the promise, and repeated the performance when novelty had long since fled. Yet civilization depends far more on the second drama than on the first. A society is not held together by what its members occasionally feel. It is held together by what they can be counted on to do again.

James understood this with unnerving clarity.

“For this we must make automatic and habitual, as early as possible, as many useful actions as we can, and guard against the growing into ways that are likely to be disadvantageous to us, as we should guard against the plague.”

There is no soft phrasing here. He does not say that bad habits are unfortunate. He says they should be guarded against like contagion. This severity has become difficult for us to hear because modern culture has developed a rather sentimental attachment to self-explanation. We prefer to ask why a person drifts, delays, scrolls, postpones, softens, nibbles away his own resolve. We want the story behind the failure. James is interested in something both older and more pitiless. He wants to know what repeated acts are doing to the person who performs them. He wants to know what kind of nervous life is being built, one choice at a time.

This is not a small point. It reaches from private conduct into politics. Every age produces its own machinery of passivity. Ours is unusually efficient. The modern citizen is surrounded by systems designed to spare him friction: food without preparation, entertainment without interval, opinion without study, contact without presence, ambition without apprenticeship, even remorse without amendment. A machine now waits at every weak point in the will, ready to offer a painless substitute for effort. It does not merely distract. It spares the will the friction by which it knows itself. This is often described as convenience. It is too flattering a word. Much of what we call convenience is in fact organized moral discouragement. It teaches a person, hour by hour, that the shortest path is the natural one, that resistance is wasteful, that repetition is for machines, and that the self is best managed by outsourcing its rough edges. Then we wonder why resolve has grown theatrical, why so many people speak grandly about change and so few can survive a Tuesday.

Move

James’s great insight is that character changes not when we admire a better life, but when we move.

“Seize the very first possible opportunity to act on every resolution you make,” he writes, “and on every emotional prompting you may experience in the direction of the habits you aspire to gain.”

The brilliance of this is easy to miss because it sounds, at first glance, merely practical. It is far more than practical. It is a theory of moral causation. The crucial instant is not the one in which I describe my intention to myself in elevated terms. The crucial instant is the first one in which the body can either obey that intention or betray it. James knew that the self is educated by enactment. To delay action is to tell the brain that the resolution was ornamental.

That is why so much contemporary self-improvement literature feels unserious. It confuses emotional intensity with transformation. It offers mood in place of method. It assumes that enough reflection will somehow harden into conduct. James knew better.

“It is not in the moment of their forming, but in the moment of their producing motor effects, that resolves and aspirations communicate the new ‘set’ to the brain.”

There is something almost comic in how bluntly this sentence destroys an entire modern industry of self-narration. One can all but hear a thousand journals snapping shut. Your brain, James is saying, does not award merit for eloquence. It responds to performance. It is a stern accountant. It books what was done.

Be Impeccable With Your Words

That is why I have grown suspicious of the little theater of declaration that now surrounds every ambition. We announce our intentions early, often, and publicly. We tell friends we are changing our lives. We tell colleagues we are focusing on what matters. We tell ourselves that naming the reform is already part of the reform. But there is a humiliating fact here, and I think James forces us to face it: speech can become a substitute pleasure. It can give the emotional reward of seriousness without the material inconvenience of becoming serious. The resolution is pronounced. The self feels noble. Nothing happens. One begins to suspect that much of adult life now consists of people issuing press releases about internal reorganization that never takes place.

James did not despise maxims. He despised inert maxims.

“No matter how full a reservoir of maxims one may possess, and no matter how good one’s sentiments may be, if one have not taken advantage of every concrete opportunity to act, one’s character may remain entirely unaffected for the better.”

That sentence explains not only private weakness but the moral uselessness of so much public righteousness. The world is full of people with correct language and untrained conduct. Very often the language is not an accompaniment to the failure but its smoke screen. It supplies the appearance of moral formation where no such formation has taken place. They can identify vice in theory, decode structures in theory, denounce corruption in theory, perform sensitivity in theory, and yet cannot master appetite, punctuality, courage, restraint, fidelity, or plain steadiness in their own affairs. James is not asking us to become puritans of the schedule book. He is asking a more difficult question: what exactly do your principles weigh at the point where action becomes costly?

Internal Government

Here habit becomes more than a productivity topic. It becomes a constitutional topic. A person with no trained habits lives under a bad form of internal government. Every decision must be relitigated. Every duty must compete with appetite from scratch. Every morning begins as a referendum. That is exhausting, and it is one reason exhausted people cling so desperately to the fantasy that discipline is the enemy of vitality. In truth the opposite is usually the case. Habit, at its best, is stored judgment. It is the prior settlement of matters that ought not consume the whole parliament of the soul each time they arise. When James writes that we should “live at ease upon the interest of the fund,” he is describing one of the great civilizing achievements, the conversion of effortful wisdom into reliable practice.

Yet James is not preaching comfort. He is preaching reserve strength. His final maxim is the one that has stayed with me longest because it cuts against the grain of a culture devoted to optimization.

“Keep the faculty of effort alive in you by a little gratuitous exercise every day.”

Not useful exercise only. Not strategic exercise only. Gratuitous exercise. Do something, James says, “for no other reason than that you would rather not do it.” That is an extraordinary sentence. It is a theory of freedom disguised as old-fashioned advice. He is telling us that the will decays when it is used only where incentives already point in the right direction. A person becomes strong not only by pursuing what he wants, but by proving, in small repeated acts, that want is not sovereign.

This is among the most neglected truths in affluent societies. We have become highly skilled at asking whether a task is efficient, monetizable, psychologically rewarding, professionally legible, or aligned with some larger system of measurable outcomes. We are less willing to ask whether it trains the soul against panic and softness. The older word would have been asceticism, though James strips it of incense, cloister, and spiritual display and gives it a civic use. This is asceticism for the office, the street, the family table, the polling booth. Not withdrawal from the world, but training for contact with it. He is not calling for spectacle. He is calling for drills. Skip the indulgence. Walk in the rain. Finish the difficult page. Hold the tongue. Rise when the alarm rings. Put the phone away. Sit with the discomfort long enough to discover that discomfort is not a verdict. These are small acts. They are also rehearsals for the day when the stakes will not be small.

That, to my mind, is the political edge of James’s chapter on habit. A civilization that loses the dignity of small effort does not merely become lazier. It becomes more governable by appetite, more susceptible to manipulation, more eager for sedatives, more likely to mistake dependency for care. The weakening of habit is not just a private inconvenience. It is a public event. Citizens who cannot command themselves will eventually beg to be managed by systems that can. Then the language of freedom remains in place while the substance drains away.

Character

We congratulate ourselves on living in an age of unprecedented insight into the mind, and yet James, writing in the nineteenth century, saw with brutal precision the thing we keep trying not to learn: that character is less like a statement than a groove. There, unfortunately, is the joke on all of us. We would like to be creatures of dazzling exception, forever improvising our identities in acts of splendid freedom. Instead we are, to an uncomfortable degree, animals of repetition. The comedy lies in our refusal to admit it. The tragedy lies in what follows when repetition is left to accident, appetite, and industry.

I do not think James means that life should become rigid, joyless, or narrow. I think he means something sterner and more hopeful. The person who has handed the right things over to habit is not less free. He is free for better work. Free for thought that is not consumed by trifles. Free for loyalty that does not collapse under mood. Free for courage when courage is suddenly required. Free, above all, from that degrading condition in which one’s ideals remain permanently eloquent and permanently unperformed.

That is why the chapter still matters. It is not a Victorian sermon about self-help. It is a theory of human seriousness. Life does not usually fall apart through one dramatic betrayal. It frays through tolerated repetitions. Strength does not usually arrive in one radiant conversion. It accumulates through enacted particulars. Habit is not valuable because it makes life easier, though often it does. It is valuable because it determines who, in the decisive hour, will still be available to themself.

Stay curious

Colin

AI Push Back

I think one of the strangest scenes in modern corporate life is the sight of very intelligent adults speaking about artificial intelligence in the tone earlier centuries reserved for saints, oracles, and unusually gifted horses. Why so much reverence? I was at a panel discussion last week and it is frightening how executives use buzzwords around AI. I sense this is for investors’ ears and nowhere near reality.

The machine, they tell us, will draft the memo, summarize the meeting, prepare the analysis, write the code, recommend the strategy. It is always about to save us from some exhausting portion of ourselves. One listens for long enough and begins to suspect that what many organizations really want is not intelligence, but relief. I think we need to push back against this.

AI elevates capability. It does not eliminate it

I do not say this lightly. Anyone who has spent years inside institutions knows how much human labor is spent not on thought but on the staging of thought. We produce documents to prove that we have considered a problem, presentations to prove that we have discussed it, process notes to prove that we have governed it, and emails to prove that no one can later say they were not copied at 18:43 on a Thursday evening. Into this long comedy of administrative self-protection arrives a machine that can generate all the signs of diligence in seconds. It is no wonder executives feel a little weak at the knees.

Responsibility

But I think the seduction begins in the wrong place. The important fact about AI is not that it can write. The important fact is that it cannot answer for what it writes. It can prepare the analysis, draft the recommendation, construct the options in beautiful prose, even imitate the tones of caution or conviction that human beings mistake for wisdom. But it will not sign the paper. It will not sit in front of a regulator. It will not explain itself to a customer whose mortgage was mishandled, an employee whose job was redesigned into absurdity, or a board suddenly eager to discover who approved the AI. The machine can produce the sentence. It cannot bear the consequence.

That, to me, is the real beginning of the story. Not substitution, but exposure. Not the disappearance of work, but the discovery of what work actually is. The question is not whether AI can produce content. Of course it can, when prompted. The question is whether organizations are prepared to redesign responsibility. Most are not. They want the efficiencies of machine generation without the inconvenience of institutional self-examination. They want speed without submission to discipline. They want intelligence without governance. In corporate life, this is a familiar impulse. It is the same impulse that buys a treadmill in January and leaves laundry on it by March.

Overflow

I believe the real drama is not technological but organizational. The firms that benefit most from AI will not be the ones with the loudest announcements, the slickest demos, or the most feverish executives posting photographs from “AI transformation summits.” They will be the ones willing to perform the far less glamorous labor of clarifying ownership, cleaning processes, documenting exceptions, preserving institutional memory, and deciding, in plain language, who is accountable for what. This does not sound romantic. Neither did drainage systems. Yet cities depend on them. Water is useful only when the pipes can carry it where it needs to go. Otherwise it floods the street, backs into the cellar, and leaves everyone standing in the wreckage asking why abundance has become damage. AI poses much the same challenge. It can produce a flood of analysis, drafts, recommendations, and code. But if the underlying processes are broken, if ownership is vague, if no one knows where intervention rights live, then all that apparent intelligence does not create value. It creates overflow.

Legacy

The AI and jobs debate cannot be found in a slick demo or social media threads. The demo looks like a miracle: fast, cheap, elegant. But walk into any actual firm and you are not in a tech showcase. You are in a museum of bad decisions and legacy systems. Old systems still talk to older systems. Documentation is partial, stale, or mythical. Whole processes survive because a few people in operations know all the unofficial fixes and keep things moving when the formal system fails. A workflow introduced as a temporary patch in 2017 is still there because no one has had the time, clarity, or authority to replace it.

This is why so many predictions about the immediate obsolescence of outsourced coders have sounded more prophetic than true. Yes, in a clean environment, with clear architecture and good documentation, AI can produce startling gains. But most large organizations are not clean environments. They are historical settlements. They contain ruins, extensions, secret passages, and little shrines to earlier management fads. Their software is not merely technical. It is biographical. It records mergers, panics, regulatory scares, budget freezes, departmental rivalries, and those immortal words of corporate postponement: we will sort it out properly later.

Later, of course, arrives wearing a lanyard and carrying an AI strategy deck.

So the problem is never simply whether an AI model can generate code or draft a tool. The problem is whether the organization understands enough about itself to know what problem it is solving. Getting the most out of AI requires context, not just computational power. One must understand the surrounding architecture, the dependencies, the hidden liabilities, the exceptions, the habits of users, the workarounds people invented because the official process did not survive first contact with reality. When that context is missing, AI does not abolish labor. It reveals how much invisible labor was already holding the place together.

I think this matters beyond software. It tells us something useful about how organizations create value. Some roles genuinely do the hard work of coordination: they frame decisions, surface trade-offs, resolve conflicts, protect standards, and carry responsibility across functions that would otherwise drift apart. Those roles matter, and may matter even more in an AI-rich world. But there are also roles, and fragments of roles, built largely around repetition, translation, and administrative relay. AI is likely to compress those first. The distinction is not between managers and non-managers, or between leaders and workers. It is between work that sharpens judgment and work that merely recirculates it.

That is why the shock feels personal as well as organizational. AI can now perform, in seconds, many of the gestures that once signaled usefulness: the recap, the options paper, the strategic summary, the smooth rearrangement of anxiety into bullet points. And so institutions are forced to ask a sharper question. Which roles are creating real clarity, accountability, and decision quality, and which are mainly moving information from one room to another? The answer cannot merely be that the human being adds warmth around the edges. The answer has to be responsibility. It has to be the capacity to decide, to absorb consequence, and to know when a recommendation is unwise even when it is elegant.

It isn't just about the software. There is a deep, almost childish vanity in how modern institutions approach their tools. They assume that because a machine can generate an answer, the organization has become the kind of place that can use that answer wisely. These are not the same achievement. A recommendation, however polished, is not a decision process. A draft is not a governance model. Analysis is not accountability. The machine may tell you what appears efficient; it cannot tell you what is prudent, legitimate, humane, or bearable to the people who must live inside the result.

Judgment

This is why I believe the future of work will be less about replacement than about a redistribution of dignity and discomfort. Some tasks will become easier, cheaper, faster. That much is obvious. What is less often said, because it is less flattering, is that the burden removed from execution will reappear as pressure on judgment. If the draft can be produced instantly, what remains valuable is the person willing to say yes, no, not yet, or absolutely not under any circumstances.

For years, many professional roles have been protected by fog. One could survive, even flourish, by converting uncertainty into procedure and procedure into paperwork. Attend the meeting. Request the deck. Refine the wording. Summarize the options. Delay the conclusion. Escalate politely. Ask for alignment. Propose a working group. There are entire careers built on the skilled circulation of not quite deciding. AI is going to be a nightmare for these people. It is devastatingly good at faking productivity. It can churn out a deck in better prose than a VP, summarize the indecision with total confidence, and phrase a “no-update” email with unnerving calm. In a lazy institution, this will not feel like transformation. It will feel like bureaucracy on amphetamines.

And yet there is something clarifying in that prospect. When the rituals of competence become cheap, one begins to see more clearly what actual contribution looks like. Not polish. Not activity. Not the magnificent modern talent for generating artifacts. Contribution begins to look like ownership. It begins to look like someone who knows where the process starts, where it breaks, which exception matters, who absorbs the risk, and which decision cannot be delegated to an AI model no matter how persuasive its syntax may be.

Custody

This is why human oversight remains essential, and not in the pious, ceremonial sense that appears in policy documents. I do not mean the kind of oversight that consists of adding a sentence to the governance pack declaring that “a human remains in the loop,” as though the loop were a mystical object one could stand near while checking email. In many organizations, that phrase has become less a safeguard than a fresh instrument of administrative self-protection. It sounds like accountability while diffusing it. It suggests that somewhere, somehow, a responsible person exists, while carefully avoiding the far more dangerous task of naming who that person is, what authority they possess, and what they are required to do when the system goes wrong. I mean actual oversight.

I keep coming back to the apparently dull subjects that turn out not to be dull at all: process inventories, controls, ownership, monitoring, intervention rights, human oversight that is more than ceremonial. These phrases suffer from a public relations problem. They sound like the natural enemies of inspiration. But I think the opposite is true. Civilization depends on people who know exactly where responsibility lives.

We have spent too long praising disruption and too little time admiring custody.

Any serious use of AI requires someone to know where a process begins, where it can fail, who has authority to stop it, and what happens when the machine’s recommendation collides with law, ethics, customer reality, or ordinary human decency. This is what human oversight ought to mean. Not a decorative statement that a person remains “in the loop,” as though one could stand vaguely near the loop and still claim moral credit. Oversight means a named human being, with enough context and enough courage, prepared to interrupt the machine and answer for the interruption.

Perhaps that is why the deeper significance of AI feels, to me, almost political. We are told endlessly that the technology is intelligent. Fine. The more interesting question is whether the institution using it is adult. Does it know itself? Does it remember its own history? Can it distinguish efficiency from wisdom, throughput from legitimacy, recommendation from judgment? Tools do not answer these questions. They sharpen them.

The companies that thrive in this new world will not ask, first, “What can the model do?” They will ask, “What process are we trying to improve, what decision are we trying to support, who owns it, where are the failure points, and what institutional conditions would make the output usable?” This is a much less glamorous sequence of questions. It has the disadvantage of being serious. Yet seriousness, in periods of technological intoxication, is often the rarest competitive advantage.

Incentives

I am reminded of older moments when a new machine seemed poised to change everything. The rotary press mattered. The telegraph mattered. The assembly line mattered. The spreadsheet mattered. But in each case the machine alone did not determine the outcome. What mattered was the discipline of the institution around it, the habits it rewarded, the forms of authority it strengthened, the forms of stupidity it accelerated. Technology does not enter a vacuum. It enters a culture, and culture is where many revolutions go to die.

That is the danger now. Not that AI will suddenly eliminate human beings in one clean act of substitution, but that leaders will use its brilliance as a reason not to reform themselves. They will automate drafting while leaving ownership vague. They will accelerate workflows whose purpose was never quite clear. They will ask an AI model for recommendations in organizations still terrified of responsibility. They will discover, a little too late, that a faster confused institution is still confused.

More With More

That is why Jensen Huang’s recent remark about layoffs struck me as unusually direct. Asked why companies are firing people if AI is supposed to make everyone more productive, he answered, in substance, that imaginative companies will do more with more, while leaders who are out of ideas will use new capability merely as a pretext to do less. I think that is exactly right. AI does not automatically create ambition. It reveals whether ambition was there in the first place. In one company, greater capability becomes an argument for building more, serving more, attempting more. In another, it becomes an alibi for shrinking imagination and calling the retreat efficiency.

Jensen Huang frequently describes the modern data center as an “AI Factory,” a physical plant designed to produce a new commodity: tokens. In his view, compute is the raw fuel of the new economy, and for imaginative companies, this abundance allows them to “do more with more.” It is a compelling architectural vision of scale.

But as any operator knows, the more a factory produces, the more critical the role of the foreman becomes. Huang is building the infrastructure that generates the shipment; I am asking who is going to be the person who actually signs for it. Who is the foreman willing to take the blame when the shipment is wrong, when the logic is flawed, or when the “intelligence” produced by the factory ignores the “context” of the town it sits in?

If we follow the factory logic to its end, we realize that tokens are cheap, but the signature is expensive. The “more” that imaginative companies will do is not just more production; it will be more frequent and more difficult acts of discernment. They will use the factory to generate the options, but they will rely on a named human being to survive the result.

I also think there is a harder and more hopeful possibility. AI may force a confrontation with what serious work has always been. Not mere production. Not the pile of emails, the dense memo, the fevered performance of indispensability. Serious work is the assumption of consequence. It is the willingness to decide under conditions of uncertainty and to remain visible after the decision has been made.

For a long time, much of professional life has depended on the theater of effort. The full calendar. The polished deck. The midnight revision. The rhetoric of overload. AI is about to expose how much of this was costume. If a machine can generate in seconds what once occupied a team for a day, then the old symbols of diligence begin to look less like proof of merit than like relics from a regime of expensive inefficiency.

This will not be comfortable, especially for institutions that have long confused visible effort with actual value.

Because what rises in place of that theater may be something better. If execution becomes cheaper, discernment becomes more precious. If drafting becomes instant, clarity becomes rarer. If recommendations multiply, accountability becomes the scarce good. The machine, in other words, may return us to an older truth: that the center of work was never typing. It was judgment.

Cultural Intelligence

There is even, I think, a bleak little joke at the heart of all this. For years, organizations behaved as though intelligence were the rare treasure. Now they possess machines that can simulate immense reserves of it on command, and they discover that the truly scarce things were responsibility, courage, institutional memory, and the willingness to say: this decision is mine.

The answer, inconveniently, is still a person.

That is why the real transformation will happen where machine capability meets human ownership. Not where the software is most dazzling, but where the institution is most serious. AI will matter most in places mature enough to pair technical power with moral and organizational responsibility. Elsewhere it will remain what so many management fashions become in the end: an expensive new instrument for performing old evasions.

I do not believe, at least not in the short term, that AI will replace people in the grand theatrical manner so often advertised. There will be disruption, certainly, especially in fields built around routine content, routine service, and procedural output. But I believe AI will do something more unsettling. It will reveal what people were actually for. It will expose whether an organization knows its own processes, understands its own context, and possesses the nerve to act on better information without surrendering judgment.

The machine can prepare the case. The human being must still live with the verdict. That is not a footnote to the future of work. It is the whole argument.

Decide wisely

Colin

Image by Nathan Lemon on Unsplash

The AI Gate is Already Gone

There is a lot of noise around AI, especially about the impact on jobs and productivity. It is truly difficult to ‘predict’ the likely outcome, will it decimate jobs or create more/new jobs? I believe there will be a significant societal disruption for sometime and we are not prepared for it.

In addition to that hugely important area, there are other impacts of AI that must be addressed. I think the risk of AI use was highlighted in the recent storm created by the US Department of War and Anthropic ‘supply chain risk’ over AI controls, as the New Yorker author states:

“The Trump Administration wants Claude to act like an obedient soldier. But, if you ask for a killer robot, the company argues, you might get more than you bargained for.”

….we might add and be a global spy on all of humanity!

There is a ton of recent research papers on AI risk, and I make it my job to stay up to date on them. What unsettled me is not their extremity but their restraint. Nobody is shouting. Nobody is selling the usual science-fiction melodrama. The prose is measured, procedural, calm. Yet the conclusions, once stripped of their academic manners, is severe: the old fantasy that a handful of firms can contain this technology at the source is beginning to fail.

We have been telling ourselves a fairytale about “AI control.” We imagine a world where a few smart engineers at a few big companies can just “tune the model” or “restrict the release” and keep us safe. We treat AI like a wild animal we can keep in a cage if the bars are thick enough. But, in fact, the bars are melting.

The most important of the papers I read was by a team of researchers at the Centre for the Governance of AI, Stanford and Oxford University and other institutions. I will call it the Bernardi paper for ease. The authors admit what most of us are too scared to say: the “bottleneck” strategy is dying. As computing costs plummet, any small team with a decent server can strip away the safety filters that OpenAI or Google spent millions to build. In 2020, it cost millions to train a top-tier model; two years later, it cost a tenth of that. The “gate” isn’t just open; it has been vaporized by the sheer math of technological diffusion.

Avoidance, Defence, and Remedy

This isn’t an engineering problem anymore. It is a political one. The authors propose a framework that is as elegant as it is terrifying: Avoidance, Defence, and Remedy. They are telling us to stop obsessing over the machine and start redesigning the town. If a cyberterrorist uses an AI to cripple a power grid, we don’t need a more ethical chatbot; we need a city that knows how to “black start” its own electricity. If a deepfake successfully tricks half the electorate, we do not need a better algorithm; we need the civic stomach to actually rerun a national election.

That is a much harder story to tell. It means we have to stop being an audience and start being a civilization. It means we have to stop waiting for a “guardrail” to save us and start building institutions that can survive being hit by a truck.

The problem is not just that the tech is out of the bag; it is that we are redesigning our own lives to fit the bag’s dimensions.

Yale Professor Luciano Floridi hits the nerve here: AI is not “intelligence” in any way we would recognize in a human being. It is pure agency. It is a tool that acts, but doesn’t care. The danger is that we have spent the last decade turning our world into a place where that kind of mindless action works perfectly. We have quantified, standardized, and “nudged” our reality into a series of data points that a machine can process. We are not asking if AI will adapt to us; we are busily trimming our own institutions so the systems can move through them with less friction.

We are practicing a “habit of surrender” in every small, daily motion. We let an algorithm sort our resumes, another one predict our creditworthiness, and a third one decide what news we see. Then, we act shocked when our capacity for human judgment begins to atrophy.

The Illusion of Control

The “capability-modifying” interventions everyone talks about, the fine-tuning and the filters, are often just moral cosmetics. They make the company feel good and the user feel safe, but they are incredibly fragile.

A small team with access to a model’s weights can strip away safety fine-tuning for next to nothing. Even without the weights, techniques like “many-shot jailbreaking” can bypass safeguards entirely. Furthermore, the “Use-Misuse Tradeoff” means that if you try to make a model “safe” by forcing it to unlearn how a virus works, you are not just stopping a bioterrorist; you are stopping the next generation of doctors from learning how to fight a pandemic. Similarly, if you train an AI against bioweapons or explosives, you feed it with the information.

This is why the Bernardi paper’s focus on resilience is so vital. Resilience is the capacity to perform a loop: identify the risk, find a response, and implement it.

Building a “Hardened” Society

If we accept that the “gate” is gone, we have to start building a society that can survive the impact.

We need “avoidance” strategies like proof of humanity on social platforms to stop synthetic media from drowning out real voices. We need “defence” in the form of defensive AI that patches security vulnerabilities faster than a diffused model can exploit them. Most importantly, we need “remedy” through redundancy in our critical systems, like backup power for hospitals, so a single cyberattack doesn’t turn into a mass casualty event.

This is not the flashy, cinematic future we were promised. It is a future of insurance schemes, better audit trails, and the “tedious magnificence” of being prepared for things to go wrong. It forces us to grow up and realize that technology does not absolve us of politics; it just makes the stakes of our political laziness a lot higher.

This is not a story about the end of the world; it is a story about the end of our innocence. We have spent the last few years treating AI like a spectator sport, waiting to see if the “metallic gods” would be benevolent or vengeful. But the Bernardi paper and Floridi’s philosophy strip away that cinematic delusion. They leave us with a much colder, much more adult reality. The central ethical question of the AI age is not whether the machines will eventually think like us. It is whether we will still have the stomach to govern ourselves when the machines make it easy to quit.

If we want a society that survives this transition, we have to stop asking for “frictionless” intelligence and start investing in the tedious, difficult work of retaining our own judgment. We need to build a “Good AI Society” not through better algorithms, but through a different set of public ideals.

First, we must demand Retained Human Override, ensuring that every automated delegation remains overridable in principle to preserve the dignity of human choice. Second, we must build Infrastructure for Failure, moving beyond moral cosmetics to create systems that can absorb shocks, from redundant power grids to the logistical grit required to rerun a compromised election. Finally, we must practice the Discipline of the Loop, treating resilience as a muscle rather than a mission statement by relentlessly identifying risks, assessing responses, and measuring what actually works.

We are currently at a crossroads of design. We can continue to redesign our world to suit the mindless agency of the machine, or we can start the “scorched earth” work of hardening our institutions against the chaos we have already invited in. The first draft of the next national crisis might be written by an AI, but the response has to be written by us. It is time we stop being the audience and start being the architects.

Stay curious

Colin

Image Ken Mages on Unsplash

Recommended debate at the Oxford Union: Artificial General Intelligence Will be Humanity’s Last Great Invention