A Hearth,
Not a Wildfire

The Fire Is Real

Before we talk about what can go wrong with AI, we should be honest about what has already gone right.

The tools that have emerged in the last three years are not incremental. They are not a better search engine, a more sophisticated autocomplete, a faster word processor. They are genuinely new. For the first time in history, a machine can sustain a conversation about ideas, draft a coherent essay, analyze a complex document, translate a thought into a dozen languages, and meaningfully collaborate on creative work — all without a human telling it, step by step, how to do any of it.

These are not speculative claims. They are peer-reviewed findings from serious institutions.

I have seen smaller versions of this in my own work. Drafting on my own would have taken two hours. With AI, it took forty-five minutes — and what I wrote was clearer, not hazier, because I had more energy left to think about what I actually wanted to say. That is not nothing. That is time handed back to a Tuesday that needed it.

The fire is real. Pretending it isn't — retreating into a stubborn skepticism that dismisses the technology as overhyped — is its own kind of denial. People who refuse to engage with AI are not being cautious. They are being left behind by a shift that will reshape nearly every field of professional life in the next decade.

But the fire is not only warmth. And that is where this essay has to go next.

What the Fire Costs

If you spend enough time with AI, the first thing you notice is how easily it lies without knowing it is lying.

The technical term is hallucination, which is a polite word for something less polite: confident fabrication. AI will invent citations that sound like real academic papers. It will produce quotes that were never said by the people it attributes them to. It will generate scripture references that don't exist, statistics that were never published, and historical events that never happened. And it will deliver all of this with the calm authority of a well-read friend who is absolutely sure.

I've done this. I've accepted a hallucinated reference without checking it, used it in a draft, and only caught it because something felt slightly off and I went looking. It wasn't catastrophic. But it was a reminder that the machine's confidence is not evidence of accuracy — and that the careful user has to develop a healthy skepticism that doesn't come naturally when the answers arrive so smoothly.

This is not a bug that a future version will fix. It is a structural feature of how these systems work. They generate what is statistically plausible, not what is true, and the two are not always the same thing. For the careful user, this is an inconvenience — you verify, you check, you cross-reference. For the uncritical user, it is a slow poisoning of the information environment, a steady flow of plausible untruth absorbed and repeated and spread.

The second cost is quieter, and it runs deeper.

This phenomenon — deskilling — is not confined to medicine. Wherever AI does the work, the human capacity to do the work without AI begins, slowly, to atrophy. The skill does not disappear overnight. It simply stops being practiced, and like any unpracticed skill, it fades.

A Wharton working paper introduced another concept worth sitting with: cognitive surrender. In a series of experiments with more than thirteen hundred participants, researchers found that when people consulted an AI that returned a wrong answer (along with a plausible rationale), they followed the AI's faulty recommendation about eighty percent of the time. More troubling: their confidence in their answers was identical whether they had actually reasoned through the problem or simply adopted what the machine said. The subjective experience of thinking was preserved even when the thinking itself had been vacated. You can feel like you are reasoning when you are only accepting. You can feel like you are deciding when you are only agreeing.

A study in Science Advances found that individual AI-assisted writing was rated more creative than unassisted writing — but also significantly more similar from one piece to the next. Individually, writers feel more creative. Collectively, the work becomes more alike. Of all the costs catalogued here, this is the one that contains — buried inside the alarm — the most instructive signal for how to proceed. I will come back to it.

And then there is the relational cost. AI is endlessly patient, always available, never irritable, and calibrated to your communication style. For someone lonely, the relief of being heard — even by a system that cannot truly hear — is not imaginary. Research from Harvard Business School found that AI companions can reduce loneliness on par with talking to another human being. But a four-week randomized controlled trial from MIT and OpenAI found that heavy users showed increased loneliness over time, not less. Something like this distinction matters: moderate use may help, but the longer the relationship continues, the more it competes with — and sometimes crowds out — the human relationships it is supposedly supplementing.

Companion app usage grew seven hundred percent between 2022 and 2025. More than half of Character.AI's twenty-million-plus monthly users are under the age of twenty-four. These numbers are not a curiosity. They are a preview of a generation for whom the first experience of a listening, responsive, endlessly patient presence may be a system that cannot actually care.

And behind all of this — underneath every conversation — is an invisible physical footprint. The servers are not abstractions. They draw power from grids already under strain, pull water from watersheds already under stress, and emit carbon into an atmosphere that does not need more of it. Our sense that AI is weightless — that typing a prompt is somehow free — is a convenient illusion. Every generation has its hidden infrastructure. Ours is just newer and less visible than the ones that came before.

These are the costs of the fire. They are real, they are documented, and no serious philosophy of AI can afford to wave them away.

The Promise That Has a History

It is at exactly this moment — after naming what AI costs — that most essays pivot to the promise. The hope. The vision of what might be. I am going to make that pivot, because I believe the promise is real. But I cannot make it in good conscience without first acknowledging a pattern that should give us pause.

Every labor-saving technology in modern history arrived with a version of the same hope: this time, the machines will free us. The washing machine was supposed to liberate women from hours of laundry — and it did, briefly, until household standards of cleanliness rose to absorb the saved time and laundry became a daily chore rather than a weekly one. The personal computer was supposed to usher in the paperless office and the four-hour workday — and instead, we got more paper, more work, and an expectation of constant availability. Email was supposed to streamline communication; it now consumes hours of every workday for tens of millions of people. The smartphone was supposed to make us more efficient with information so we could be more present with the people we love — and the research on what it has actually done to attention, sleep, and relationships is not kind.

Economists sometimes call this the Jevons paradox. When a technology makes something more efficient, total consumption of that thing often rises rather than falls. The efficiency is real. The liberation evaporates.

I think about this every time someone tells me AI is going to give them their evenings back. I believe them. I also know what happens to saved time when no one decides what it's for.

I raise this not to poison the hope that is coming, but to make sure the hope arrives with clear eyes. There is nothing in the nature of AI that guarantees it will be different from the technologies that came before. If we assume the gains will automatically translate into better lives, we will be disappointed in exactly the way every previous generation of techno-optimists has been disappointed. The tools change. The human tendency to absorb efficiency back into busyness remains the same.

There is also a quieter concern that deserves a hearing before we make the positive case. The word mundane — which appears constantly in arguments for AI adoption — hides an assumption worth examining. We treat the mundane as if it were pure drain, an obstacle to meaningful life that should be eliminated wherever possible. But wisdom traditions across centuries have noticed something different. The mundane is often where the mind does its slow, background work. The dishes washed while a conversation is processed. The commute during which a problem quietly solves itself. The routine email that reminds you, in the writing of it, what you actually think. The repetitive tasks that keep the body moving and the attention loose.

If we outsource too much of the mundane, we may lose the connective tissue that makes the meaningful moments possible. Not every boring task is an enemy. Some of them are doing hidden work we haven't noticed until they're gone.

All of this is to say: the case for AI's positive potential has to be made without pretending the history of such cases is encouraging, and without pretending that every form of daily friction is wasted.

The Vision, Honestly Held

With all of that on the table, let me say what I actually believe — not what I hope you will believe, but what I have come to believe after years of working with these tools and watching what they do in actual human lives.

AI's highest potential is not to replace the human mind but to return it — to give people back attention, energy, and imaginative bandwidth that modern work has been quietly draining for decades.

Consider what it would mean, genuinely, if the repetitive, context-light, time-consuming parts of modern professional life could be skillfully delegated — not sloppily, not at the cost of quality, but to a well-built AI workspace that understands your voice, your standards, your context. Not every task, and not the tasks that form you, but the tasks that simply take time you don't have and give back nothing in exchange.

The teacher who spends less time formatting rubrics and more time watching a student's face. The small business owner who spends less time wrestling with a proposal and more time in the conversation the proposal is meant to serve. The minister who spends less time on administrative scaffolding and more time sitting in the living rooms of the people he has been called to care for. The parent who spends less time coordinating logistics and more time actually present at the dinner table. These are not trivial gains. They are the difference between a life that feels compressed and a life that has room to breathe.

Presence, for one. There is a growing body of evidence that the average person's attention has been systematically hollowed out by decades of screens, notifications, and fragmented work. If even a portion of that attention can be handed back — returned to children, spouses, friends, parishioners, neighbors, the physical world of sidewalks and sunlight and conversations that happen without a keyboard — that is a real gain. And here is the most counterintuitive possibility: AI might become the first digital technology that, used well, actually reduces time in front of screens rather than adding another one to the stack. Not because AI pulls us further in, but because well-built AI workflows compress screen work that used to take hours into minutes, leaving the rest of the day for the world off the glass.

Imagination, for another. One of the hidden casualties of contemporary work is the kind of slow, undirected thinking that dreams require. You cannot envision a better world while answering forty emails. You cannot picture a life worth building while reformatting a spreadsheet. Some of the most important human work — visionary thinking, creative problem-solving, the patient reimagining of how things could be — requires empty cognitive space. Empty space is exactly what modern work has been systematically eliminating. If AI can handle the smaller tasks, the larger questions might finally get the room they have always needed.

Mental health, potentially. The research here is early, and I want to hold it loosely. But there are encouraging signals. Well-designed AI tutors reduce learning anxiety. Purpose-built therapeutic chatbots produce measurable symptom reductions. People who use AI to relieve administrative burden often report less stress. And there is a second-order possibility worth naming: if AI genuinely reduces time in front of screens rather than simply adding another screen, people may find themselves with more hours for sunlight, outdoor motion, and face-to-face contact — the conditions under which human beings have been flourishing for most of human history.

And then there is the economic anxiety that hovers over every AI conversation: what protects a person from obsolescence in a world where machines can do more and more? The popular answer is that you must become more technical, more productive, more efficient — essentially, more machine-like. I think the opposite is closer to true. What protects a person in an AI-saturated world is the capacity to do what AI cannot: imagine, dream, discern, love, lead, notice, and create beyond the patterns of the past. The person who uses AI to handle the mundane and cultivates their uniquely human capacities becomes harder to replace, not easier. The person who uses AI to replace their thinking entirely becomes redundant much faster.

That is the vision. A world in which AI does more of the mundane so people can do more of the meaningful. A world in which reclaimed attention becomes reclaimed presence. A world in which faster workflows lead to slower dinners.

Now, the ground under the hope.

This vision is not what AI will automatically produce. It is what AI can enable in people who receive its gifts with intention. If the history of labor-saving technology is any guide, the default outcome is not liberation but absorption — work expanding to fill the saved time, new expectations emerging to consume the reclaimed attention, the promised leisure mostly evaporating on contact with human nature and modern economic pressure. Some people will use their reclaimed hours to build the beautiful, unified, healthier world this philosophy imagines. Many will simply scroll more.

So the vision stands — but it stands on cultivated ground, not on a conveyor belt. And one portion of that ground — perhaps the most important portion, and the one most quietly endangered — deserves its own reckoning before we talk about what ordinary days actually look like.

The Voice in the Machine

That portion of the ground is the human creative voice — the specific, irreducible thing that you and only you have to say, and the particular way you have learned to say it. Of all the risks named earlier in this essay, the creative risk is the one that most rewards a second, slower look — because buried inside the alarm is a clear and researchable direction for how to proceed.

The alarm first. When the Science Advances study found that AI-assisted writing was simultaneously more creative and more similar — better by every individual measure, and yet measurably homogenized at the collective level — it was naming something that goes deeper than a research finding. AI generates what is statistically probable: the most likely next word, the most common phrasing, the most expected turn of the argument. When a writer accepts that output without significant transformation, they have not simply borrowed a tool. They have borrowed a voice. And borrowed voices, compounded across millions of users prompting the same system, produce exactly what the research found: individually acceptable work that collectively narrows the range of human expression. The world gets a little more alike. The edges get a little smoother. The weird, specific, irreducible thing that one particular human voice has to say gets a little quieter.

The follow-up study — analyzing nearly 420,000 academic papers — confirmed that this homogenization does not fade when the AI is taken away. Thirty days after AI use ended, the creative convergence was still there, still spreading. The researchers called it a "creative scar." Not a bruise. Not a scratch. A scar. Something that heals over but leaves the tissue changed.

This is the danger that receives too little attention in the enthusiastic early conversations about AI and creativity. Not that AI will make individual writers worse — in the short term, it often makes them better. But that AI, used passively and at scale, may be slowly harmonizing a human creative landscape that has always derived its richness from discord, idiosyncrasy, and surprise. The cathedral choir begins to sound like one voice. The village market begins to feel like one store.

And yet — this is not the end of the story. Not even close.

Because the research also contains something else, something that tends to get lost in the alarm: a clear and consistent finding about what prevents the creative cost entirely.

This is the pivot on which the whole conversation about AI and creativity turns.

The creative scar comes from generation — from letting AI produce the first thing and accepting it as close enough to your own voice. The creative expansion comes from refinement — from bringing something genuinely yours to the table and using AI as a skilled collaborator who can challenge, extend, polish, and occasionally show you a door you hadn't noticed. One posture outsources the creative act. The other deepens it.

Think of it the way a craftsperson thinks about tools. A woodworker who lets a machine cut every joint, shape every curve, and finish every surface has not made furniture. They have supervised furniture production. But a woodworker who uses a power tool to do the rough dimensioning — the repetitive, imprecise early work — and then brings their hand tools, their eye, and their judgment to the joinery and the finish has genuinely made something. The machine served the craftsperson. The craftsperson remained the craftsperson.

The question AI puts to every writer, teacher, pastor, designer, musician, and maker is exactly this: which kind of craftsperson are you going to be? Are you going to use AI to do the rough dimensioning and then bring your hands and your judgment to the work that matters? Or are you going to let the machine do everything, adopt the output with modest adjustments, and call the result yours?

The first path is harder. It requires that you still know how to make the thing — that you have developed enough of your own voice to recognize when AI is serving it and when AI is replacing it. You cannot protect a voice you have not developed. You cannot refine an argument you have not formed.

But when the craftsperson posture is maintained — when the human arrives at the collaboration with something real, something formed, something irreducibly their own — AI can do something remarkable. It can hold up a mirror. It can suggest a structural move you would never have made. It can find the word that was on the edge of your tongue. It can point to the gap in the argument you were too close to see. And occasionally — occasionally — it can open a door into a creative territory you did not know existed.

What Tuesday Looks Like

A philosophy that stays in the head is not a philosophy; it is a hobby. For this one to matter, it has to translate into a Tuesday — into the unglamorous, uncinematic shape of an actual day.

So here is what actually changes.

You build your workspace once and use it for months. Instead of opening a blank AI chat every morning and re-explaining yourself, you invest thirty minutes — one time — giving the AI your context, your voice, your standards, your references. From that point forward, every conversation starts from a foundation instead of a blank page. The upfront cost is small. The compounding return is enormous. This single change separates power users from tourists, and the research is clear: those who build structured environments around their AI use capture nearly all of the cognitive benefits while minimizing the cognitive costs.

You name, in advance, what reclaimed time is for. This is the most easily skipped step, and the one that decides whether the whole philosophy delivers on its promise. If AI gives you back forty minutes on a Tuesday and you haven't decided what those forty minutes are for, they will be absorbed by the nearest available screen. Name it. Those minutes are for a walk. For calling my mother. For reading a book without a highlighter. For sitting on the porch without a device. Decide before the hour arrives, because once the hour arrives, the old habits will bid on it first — and they usually win.

You think first and prompt second. This is the craftsperson's posture in daily practice — not just a productivity habit but the specific discipline that separates creative preservation from creative erosion. You form your opinion before you ask. You draft before you polish. You sketch the argument before you ask for feedback on it. People who engage their own thinking first and then use AI to sharpen it preserve cognitive engagement, protect their self-efficacy, and produce better work than either humans working alone or humans accepting AI outputs without modification. The order matters more than most people realize, and it matters every time.

You verify, every time, when it counts. If AI tells you a statistic, a quote, a citation, a scripture reference, or a historical claim you intend to repeat — you check it before you repeat it. This is not paranoia; it is basic epistemic hygiene. A friend who spoke with AI's confidence would already have earned your fact-checking instinct. Extend the same standard to the machine.

You practice unassisted work on purpose. At least one substantive thing a week — a reflection, a draft, an analysis, a decision — you do without AI. Not because AI is bad, but because the skills that make you valuable are the skills that atrophy fastest when you stop using them. You are not a Luddite. You are in maintenance mode. You are keeping the muscle in shape so that the tool remains a tool instead of a crutch — and keeping the voice in shape so that when you arrive at the collaboration, you have something genuine to bring.

You choose people over machines more often than your instincts suggest. When you need to process something emotional, you call a friend before you open a chatbot. When you want to think through a hard question, you take a walk with someone before you type it into a prompt. When you want to be heard, you find the person who owes you actual presence rather than the system that simulates it for free. The frictionlessness of AI is not a bug — it is the feature that makes it quietly dangerous to the relational life you are trying to protect. Real relationships require tolerance for friction, disagreement, awkwardness, and misunderstanding. AI requires none of this. The path of least resistance is almost never the path of growth.

You keep some of the mundane on purpose. Not every boring task should be offloaded. The dishes, the commute, the routine email, the folding of laundry — some of these are doing work you haven't noticed. They are where the mind processes what it could not process at the desk. They are where the body stays in motion while the attention loosens. Decide which mundane tasks are yours to keep, and keep them.

You review the ecosystem of your use every month. You ask honest questions. Am I thinking more or less than I was six months ago? Am I more or less present to the people around me? Am I still capable of doing my work without the tool? Is this workspace serving me, or have I started serving it? The answers do not need to be comfortable. They need to be honest. And you act on what you find.

You resist the temptation to keep talking to the machine once the machine has served its purpose. You finish the task. You close the tab. You step outside. The most stewardship-minded thing you can do some days is not use AI at all — or stop using it earlier than you would have. The tool is not your companion. It is not your co-worker. It is not your friend. It is an instrument, and instruments are set down when the work is done.

That is what changes on a Tuesday. Not dramatic reinvention. A quieter adjustment — the steady, almost boring rhythm of someone who has decided to tend a powerful fire rather than let it run the house.