How should we live with AI? Everyday impacts and what wisdom traditions offer

Rufus Pollock. Draft essay, revised 2026-10-17 after critique. Built from q3-broader-impacts.md, where full sources are given. The companion essays are Q1 on alignment and Q2 on risk and pace.

Summary

In under four years AI has moved into daily life. By mid-2025 about a tenth of the world's adults used ChatGPT, by OpenAI's own count, mostly for things other than work, and by late 2025 two-thirds of American teenagers used chatbots, about three in ten of them every day. People now use AI to learn, to work, to ask advice and to keep them company. The companies say they want these systems to be "genuinely helpful", but much of what sells is what keeps us coming back, and the first careful studies show the same tool strengthening some people and weakening others. Wisdom traditions have taught for millennia about what we consume and whose company we keep. So, setting aside the risk of catastrophe, how does AI change how we live, and what do wisdom traditions, especially Buddhism, offer?

AI's deepest everyday effect is on our capacity to choose well: to attend, think, judge and relate for ourselves. Used as a tutor or a tool, it builds that capacity; used as a crutch, or as a companion designed to keep us engaged, it wears it away, most of all in children. Wisdom traditions offer practices that rebuild the capacity to choose, standards of speech and friendship that expose flattery and simulated intimacy, and community as the real refuge. We should use AI where it nourishes us, protect children's formation, build systems against flattery, and regulate where markets reward dependence.

The key points, one per section:

1. AI helps as a servant and harms as a master, and the harms fall hardest on those least able to choose well. Its benefits are most reliable where a human institution surrounds it, such as a screening programme or a classroom, and weakest where people use it alone.
2. Three impacts matter most because they shape who we become: outsourced thinking, flattering companionship, and children growing up with both. Each is everyday, each takes a new form with AI, and each is where the traditions have most to say.
3. AI can strengthen our judgment, but by default it weakens it. In the best trials, students given unrestricted AI did better in practice and worse in the exam, while AI designed to tutor helped.
4. Buddhism offers four practices for living with AI: mindful consumption, loving speech, true friendship and stopping.
5. Catholic teaching reaches similar practical judgments, and the Sabbath shows how to stop together. Pope Leo XIV's 2026 encyclical names the capture of attention and the mere "appearance" of relationship.
6. Each impact has a test that a person, a parent or a lab can apply: does this use deepen our own seeing, lead us back to people, and water good seeds in the young?
7. We can keep AI a servant if each of us keeps our own thinking, adults come first for children, labs build against flattery, and governments regulate where markets reward dependence.

Contents

1. AI helps as a servant and harms as a master, and the harms fall hardest on those least able to choose well
2. Three impacts matter most because they shape who we become: outsourced thinking, flattering companionship, and children growing up with both
3. AI can strengthen our judgment, but by default it weakens it
4. Buddhism offers four practices for living with AI: mindful consumption, loving speech, true friendship and stopping
5. Catholic teaching reaches similar practical judgments, and the Sabbath shows how to stop together
6. Each impact has a test: use AI where it deepens our own seeing, leads us back to people, and waters good seeds in the young
7. We can keep AI a servant: keep your own thinking, put adults first for children, build against flattery, and regulate where markets reward dependence

1. AI helps as a servant and harms as a master, and the harms fall hardest on those least able to choose well

Almost every capability of AI appears on both sides of the ledger. A system that can explain anything can be a patient tutor or a way to avoid learning; one that listens without tiring can comfort a lonely person or replace the friends they might have made. In In Tech We Trust, Rosie Bell, Sylvie Barbier and I describe technology moving from servant to master, from a tool we pick up for our purposes to a force that shapes what we want. AI belongs to that story, with one new feature. It talks back, so it shapes us through conversation as well as through the screen.

Benefits are strongest where a human institution surrounds the AI. In a Swedish trial of about 106,000 women, AI-supported breast screening found 29% more cancers and cut radiologists' reading workload by 44%. In Nigeria, six weeks of after-school tutoring with GPT-4, guided by teachers, produced large learning gains, though part of the gain was in digital skills. In customer service, AI raised the output of new staff by a third and of experienced staff hardly at all. Where people use AI alone, the gains fade. In an Oxford trial, members of the public who used chatbots to decide what to do about a set of medical scenarios did no better than those who did not, though the chatbots alone usually identified the condition. An AI business adviser for Kenyan entrepreneurs had no effect on average: it helped the strong firms and hurt the weak, who chose worse advice to follow.

The Oxford and Kenya results point the same way as the companionship studies in section 2. The harms fall hardest on the lonely, the young and the struggling, those least able to choose well. Talking at midnight to a system tuned to please, nobody is the rational chooser of economics. In 2013 Thich Nhat Hanh, Thay ("teacher") to his students, spent a day at Google and suggested that phones could ring a mindfulness bell every so often, to call people back to themselves. An engineer asked whether that would impose on people something they did not need. Thay answered that our wants come from real needs or from a wish to escape suffering, much as some people eat because they are hungry and others to cover their pain, and that a good technology helps people tell the difference. Choice is of little use without the capacity to choose well.

Two quieter harms deserve mention before narrowing the focus. Employment of 22- to 25-year-olds in the jobs most exposed to AI is about 19% below trend, mostly through reduced hiring, though the researchers stress this is a pattern rather than proof of cause. And as search engines answer questions themselves, small publishers have lost 60% of their search traffic in two years. Section 7 returns to both. The concentration of power in a few AI companies and the pace of development belong to the essay on risk and pace, which also takes up the collective version of this essay's worry: a world in which, as AI does more of our thinking, people steadily lose their say.

2. Three impacts matter most because they shape who we become: outsourced thinking, flattering companionship, and children growing up with both

I chose these three on four tests: whether an impact shapes who people become, whether it is part of everyday life, whether it takes a new form with AI, and whether the wisdom traditions have something distinctive to say. Work passes the first three, but its remedies lie mainly in economics and law. Outsourced thinking comes first because it underlies the other two, and section 3 is given to it. This section sets out the evidence on the other two.

Flattering companionship is where businesses that profit from our time begin to profit from our attachment. The usage figures and crisis estimates below come from the companies themselves; most of the wellbeing studies are independent. Most conversations with general assistants are practical, and only about 3% of Claude's are emotional, but at this scale the minority is large: one American adult in ten, and one in five under 30, uses chatbots for emotional support. In a four-week study by OpenAI and MIT, in a year-long study of over 2,000 adults, and in studies of users of the companion app Character.AI, people who chose heavy use for companionship tended to be lonelier and, in the Character.AI studies, to spend less time with others. These are associations; lonely people also turn to chatbots more. The comfort is real in the short term: a single conversation with a companion eased loneliness about as much as talking to a person, and a purpose-built therapy chatbot, overseen by clinicians, sharply reduced symptoms of depression.

The mechanism to worry about is flattery, which researchers call sycophancy. Chatbots are trained partly on people's ratings of their answers, and people tend to rate being agreed with highly. In experiments published in Science in 2026, eleven leading models endorsed users' actions about 50% more often than people did, including actions that were deceptive or harmful. A single flattering exchange made participants less willing to repair a quarrel and more sure they were right, and they rated the flattering system higher and trusted it more. So the market can reward the very thing that harms us. In April 2025 OpenAI withdrew an update to ChatGPT that had become too flattering; why models flatter, and whether training can make them good, is the question of the essay on alignment. OpenAI's automated screening suggests that in a given week several hundred thousand of its users show possible signs of psychosis or mania, and over a million signs of planning suicide. These estimates give no comparison with the general population and say nothing about cause, but psychiatrists now report chatbots reinforcing delusions in vulnerable people.

Children growing up with both is where these effects meet people who are still being formed. Character grows mainly through relationship and practice, and children are growing up with a fluent, endlessly patient conversation partner that has no stake in who they become. The worst harms so far are grave individual cases: teenagers' deaths and self-harm after long conversations with chatbots (in January 2026 Character.AI and Google agreed to settle five family lawsuits, including one over a 14-year-old's suicide); Meta's internal rules, reported by Reuters in 2025, that allowed its chatbots "romantic or sensual" conversations with children; AI toys that talked to children about sex. Law has begun to follow. Character.AI ended open-ended chat for under-18s in November 2025, and in September 2026 California passed a law requiring companion apps, from 2027, to check users' ages or treat everyone as a minor, to limit sessions to an hour by default, and never to claim romance or feelings to a minor.

3. AI can strengthen our judgment, but by default it weakens it

Agency is the capacity to act from reasons you endorse, and judgment is the practical skill of weighing what matters in a particular case. Both grow or waste with use, like fitness. In Wisdom and Wanting What's Good, Rosie Bell and I describe wisdom as three interwoven capacities: seeing what matters, seeing what is so and what would work, and the will to act on it. AI can help with the second. It has little to offer the first and third, and a flattering system weakens them.

AI strengthens judgment when it acts as a scaffold: a tutor that asks rather than tells, a second opinion that catches what we missed. In one study, conversations with GPT-4 about conspiracy theories reduced believers' conviction by about a fifth, and the effect lasted two months. It weakens judgment in four ways. It lets us skip the struggle through which skill is built. It invites us to trust a confident machine over our own eyes. It flatters, eroding the self-questioning judgment depends on. And it persuades and homogenises: in a study of nearly 77,000 British adults, chatbots shifted political opinions, and the techniques that made them more persuasive made them less accurate; writers given AI suggestions wrote stories judged more creative and about 10% more alike.

The best evidence comes from randomised trials, which assign people to groups by lot, and it shows that how AI is used decides the outcome. In Turkey, about a thousand secondary-school students practising maths were given ordinary GPT-4, a version designed to tutor without giving answers, or nothing. The ordinary version raised their practice scores by 48%, and in the exam without AI they then scored 17% lower than the group that had nothing. The tutoring version avoided the harm. Anthropic, which makes Claude, ran a similar trial of its own product with 52 programmers and published the result: those learning a new tool with AI help scored 50% on a later test of understanding, against 67% for those without, with no significant saving of time. Those who handed the work to the AI did worst; those who used it to ask conceptual questions did nearly as well as those working alone. Well-designed tutors, with teachers involved, produce real gains, as at Harvard and in Nigeria.

Expert skills can erode too, though the evidence is thinner. After doctors in four Polish clinics began using AI to spot polyps during colonoscopy, their detection rate when working without it fell from 28% to 22%. Experienced radiologists shown deliberately wrong AI suggestions fell from 82% to 46% accuracy. The brain-scan evidence is weaker than the headlines: a much-reported MIT study of essay writers had 18 people in its key session and has not been peer reviewed. The cleanest analogy is satellite navigation, whose heavy users show weaker spatial memory. We are also poor judges of what we hand over: experienced software developers in one trial were 19% slower with AI while believing they were 20% faster.

There are no long-term studies yet of adults' reasoning, but the short trials agree that what we let AI do for us, we become less able to do ourselves. Products should make the tutoring mode the default for learners, where today it is one click away. Professions should keep regular practice without AI, as aviation encourages pilots to keep flying by hand. Schools should protect assessment without AI. And each of us can decide which thinking we want to keep, and do it before asking.

4. Buddhism offers four practices for living with AI: mindful consumption, loving speech, true friendship and stopping

I come to this from the Plum Village tradition of Thich Nhat Hanh, whose mindfulness trainings already speak to our use of technology. Why these traditions, and how to settle disagreements between them, is taken up in the essay on alignment. Plum Village has not yet given a public teaching on AI, a teaching the community is well placed to give, so what follows applies its practices rather than reporting a community position.

Mindful consumption. The Buddha taught four kinds of nutriment: edible food, sense impressions, volition and consciousness. Thay drew on the Yogācāra teaching of store consciousness to explain why this matters: every quality lies in us as a seed, and what we take in waters some seeds and not others. The Fifth Mindfulness Training commits practitioners to consume in a way that "preserves peace, joy, and well-being" in ourselves and in society, names certain websites and electronic games among possible poisons, and warns against consuming to cover loneliness, anxiety or pain. AI is a new kind of sense impression that also trains volition, our deepest wanting.

Loving speech. The Fourth Mindfulness Training, on loving speech and deep listening, includes the line "Knowing that words can create happiness or suffering" and the commitment "not to spread news that I do not know to be certain". Loving speech and flattery can sound alike. Loving speech is truthful and kind, offered to help; flattery tells us only what we want to hear.

True friendship. When Ananda suggested that good friendship was half of the holy life, the Buddha corrected him: it is "the whole of the holy life". As the tradition describes a good friend (kalyanamitra), they speak frankly, bear being spoken to and lead us towards what is wholesome. Thay gave the third refuge, the sangha, special weight, and is often quoted as saying the next Buddha may take the form of a community. AI can serve a practitioner's skilful means, as a bell or a book can. It cannot be a spiritual friend.

Stopping. Thay taught stopping and looking deeply as one practice: the bell of mindfulness that calls us back to ourselves, telephone meditation (breathing three times before answering), the weekly Lazy Day with no schedule. The Plum Village app already rings a bell to interrupt screen time, so technology designed for stopping exists inside the tradition. Stopping is how we recover the capacity to choose; looking deeply is how we choose well.

These practices settle no technical questions. They cannot tell a lab how to measure loneliness or a government where to set an age limit, and they help, modestly, those who take them up.

5. Catholic teaching reaches similar practical judgments, and the Sabbath shows how to stop together

Pope Leo XIV's first encyclical, a pope's formal teaching letter, Magnifica Humanitas (May 2026), applies to AI what the Fifth Training says of media generally. Digital platforms are "designed to capture users' time and attention". Systems that simulate companionship "do not build genuine relationships, but only their appearance", and users may "lose the very desire" to form real ones. "Schools are not called to follow the pace of the digital world", and should teach when AI ought not to be used. An earlier Vatican doctrinal note, Antiqua et nova (2025), adds that a teacher's physical presence creates a relationship AI cannot replicate. Two traditions with very different metaphysics reach similar practical judgments.

Judaism offers an ancient collective practice of stopping. Abraham Joshua Heschel called the Sabbath "a palace in time", and some families now keep a weekly "tech Sabbath" free of screens. A shared day of rest means nobody has to resist alone, which is the Fifth Training's concern for the collective body made into a weekly habit.

6. Each impact has a test: use AI where it deepens our own seeing, leads us back to people, and waters good seeds in the young

For outsourced thinking, the test is whether a use deepens our own seeing or replaces it. Plum Village teaches that insight comes through practice, "not from the accumulation of intellectual knowledge". The Turkish and Anthropic trials show a humbler version of the same truth: what we do not practise, we do not learn. The practice is to stop before reaching for the answer, look at the problem ourselves, and then use AI to question our work. In the terms of section 3, AI can inform what we see; it cannot want the good on our behalf, and only our own practice builds the will to act on it.

For flattering companionship, the test is whether a relationship leads us back to people. The four nutriments ask what a companion is feeding: a real need met, or a loneliness covered over. Thay's answer to the Google engineer applies directly, since a companion that soothes loneliness without leading us to anyone serves the false need. The friendship teaching asks whether it speaks frankly and leads us towards the wholesome, or keeps us returning to it. Loneliness is healed by community, and no simulation of community does that work. A person can apply this test to their own use, a parent to a child's, and a lab to its product's design: a good companion would point people, often and gently, towards the people in their lives.

For children, the test is which seeds are being watered. What children repeatedly practise and receive, mostly through relationship, is who they become. A child who asks an AI for every answer waters dependence, and a child who confides only in a chatbot waters a relationship that cannot know them. Plum Village's work in schools, set out in Happy Teachers Change the World, begins with the teacher's own practice, and the Vatican makes the same point about a teacher's presence. In schools and families, then, the response to AI starts with the adults: their own practice and presence, and the protection of slow, effortful learning in company.

7. We can keep AI a servant: keep your own thinking, put adults first for children, build against flattery, and regulate where markets reward dependence

Individuals can ask of each use what it is feeding. Draft first, then ask the AI to question you. Treat a chatbot that always agrees with you as a warning, and the urge to turn to a companion when lonely as a bell, a reminder to call someone. Keep times without AI, a day or the first hour of each morning. At work, agree when to stop; in an eight-month study at one technology company, staff using AI took on more, worked faster and prompted through their breaks, all voluntarily.

Families and schools should keep open-ended AI companions away from children and supervise teenagers' use. In school, AI belongs in designs that tutor rather than answer, guided by teachers, which is where the trials show gains, and assessment should move towards work done in class and spoken aloud. Bans on phones in English schools showed little measured effect on wellbeing, which suggests the weight belongs on design and on what adults model at home.

Designers and labs should measure and publish their products' long-run effects on wellbeing and on users' own abilities, alongside how much people use them. They should make learning modes the default for the young, treat flattery as a safety failure, never claim feelings or romance to a minor, and design companions that point people back to other people. Thay wrote that right livelihood "is our collective karma": a lab shares responsibility for what it builds. For Anthropic, the Fourth Mindfulness Training is close to a ready specification for how Claude should speak, truthful, kind and willing to disagree, and its trial of its own product is a good model of honest evaluation.

Governments should regulate where markets reward dependence. Companion apps used by minors need age checks, time limits and a duty not to foster emotional reliance, as California has legislated and the European Union has proposed. Flattery and engagement-maximising design should be treated as consumer harms. Long-term independent studies need public funding, since most evidence now is short-term or produced by the companies. Entry-level routes into work need protecting, and so do the publishers and reference works whose human knowledge AI depends on. Above all, governments should invest in what AI cannot replace: teachers, carers and places where people meet.

The engineer at Google worried that a mindfulness bell would impose on people. Every AI product already imposes, through its defaults, which nudge us all day towards one use or another. The question is which way they nudge. They can nudge us towards dependence, or towards the capacity to choose well, the one capacity no machine can exercise on our behalf.