You Can’t Opt Out of AI. You Never Could.
99% of Americans used AI last week. 64% of them swear they don’t use it at all. One of those numbers is a measurement. The other is you doing the industry’s work for free.
by Lawrence Winnerman
My friend Dana DuBois recently asked me if I knew the percentages behind how much active AI use there is versus how much passive AI use. She then wondered if she had maybe coined those terms—and while she’s very good at coining terms, in this case, she didn’t.
When I started doing research, I discovered that this point is fairly well discussed: the moments in our day when we actively choose AI—to generate an image, to organize our email, to outline a story we’re writing—versus the moments when AI is fed to us by our phones, as in the summary of a text message breaking up with you before breakfast (just hypothetically speaking).
Sometimes the universe gives you hints about what it wants you to write next, and in this case mine came the next night at dinner.
While we were eating, another friend told me that she doesn’t use AI at all.
She said it the way you’d mention you don’t eat veal—quietly, without fanfare, with the settled calm of a person who has looked at a cruelty and stepped around it. She is not a foolish woman. She reads more than I do. She marches, she calls her representatives, she gives until it hurts. When she said it, the table murmured its approval, and somebody else said me neither, and the conversation moved on to the salmon.
I didn’t say anything. I want to tell you why, because the why is this essay.
I wanted what she said to be true. I had more reasons than anyone at that table to want it to be true. I was shoved out of a twenty-five-year career by the first wave of this technology, and then—because as every good GenXer who grew up singing “Blasphemous Rumours” knows, God has a sick sense of humor—I have recently been rebuilding my life by becoming the thing that shoved me out. I have written before that I am the displaced and the displacer at once, and I have stopped expecting that sentence to get easier. Nobody at that dinner wanted the exit she described to exist more than I did. A door out of the machine. A clean abstention. A way to be a bystander.
So I went looking for the door.
This essay is what I found.
Dear fellow human, I’m sorry in advance.
A Walk Through Tuesday
Come with me through my friend’s Tuesday. Not a hypothetical Tuesday—the ordinary one, the one she actually lived, the one any of us lived.
Her phone wakes her at 6:47, and before her feet find the floor she checks the weather, which is to say she consults a predictive model trained on decades of atmospheric data. She skims her email over coffee. She does not see the spam, because Gmail’s machine learning reads on the order of fifteen billion unwanted messages a day and removes more than 99.9% of them before any human eye arrives—her mail was read by a machine this morning before it was read by her.
She buys the coffee with a tap of her card, and in the two hundred milliseconds before the terminal chirps, a model scores the transaction against everything Visa’s systems know about fraud; those models blocked an estimated forty billion dollars of attempted theft in a single recent year. She drives to work along the blue line her map app draws, and the blue line is not a road. The blue line is a neural network’s continuously revised guess about the future of traffic, based on real-time mobile-phone movement data, self-reported traffic incidents, and a thousand more variables.
At her desk she reads the news, and nearly every ad she scrolls past—about 90% of digital display advertising now—was placed by an algorithmic auction that identified her, valued her, and sold her attention in less time than a synapse takes to fire. There were hundreds of those auctions before lunch. She googled something at some point; roughly half of American queries now return an AI-composed answer above the results, which means the search engine did not point her at the machine. The search engine was the machine.
Lunch is a scroll. More than half of what Instagram shows her is not from people she follows—it is conjured by a recommendation model that knows her tastes the way a river knows its bed. In the evening she watches something; more than 80% of the hours streamed on Netflix arrive through the recommender, and on YouTube the figure for recommendation-driven watch time runs around 70%. If music plays while she cooks, about three songs in ten were chosen for her by a model. Then she sleeps, and her Oura ring grades her sleep while the photos her phone took that day sort themselves by face, in the dark, while nobody watches.
She never opened a chatbot. Not once. She didn’t generate an image of the president in diapers throwing a tantrum (for example). Her hands are clean.
And the machine was with her from 6:47 in the morning until she closed her eyes—reading her mail, pricing her attention, scoring her purchases, plotting her routes, choosing her evening. It never noticed her abstinence. There was nothing to notice. Abstinence happens at the front door, and the machine does not use the front door.
In late 2024, Gallup asked Americans about a week of their lives. Ninety-nine percent had used at least one AI-enabled product in the previous seven days—weather, streaming, navigation, social feeds, the ordinary furniture of a modern day. Sixty-four percent of those same people said they don’t use AI.
One of those numbers is a measurement. The other is a story we tell ourselves.
And the story has a strange shape when you hold it to the light. ChatGPT counts some nine hundred million weekly users now, the fastest adoption curve of any product in history—and yet when Pew asked in 2025, only about a third of American adults had ever used it.
Two-thirds of the country has never once touched the famous thing. Ninety-nine in a hundred swim daily in the quiet thing. We have built our entire moral conversation about AI around the famous thing, the thing with the text box and the name, the thing you can refuse—because the thing you can refuse is the only thing that lets you feel like you are making a choice.
As if the server asked you for your order, and you said, smugly, “No AI for me, please. I’m 100% human organic.”
Friend, unless you are somehow reading this from the year 1984, oh no, you most certainly are not.
The Buildings Don’t Care What You Believe
I have written before about the buildings—the windowless boxes humming behind berms off the interstate, the water they drink, the power they pull, the towns that wake up to find a cathedral of heat has moved in next door. That essay was about what the buildings take. This one is about what they are.
Because here is what I assumed, and what I suspect my friend assumes: that there are AI buildings and internet buildings. That somewhere out there is the respectable old infrastructure—email, photos, movies, the family business of the web—and then, bolted on beside it, the new unclean annex where the chatbots live. Refuse the annex, keep the house. It is a reasonable picture.
It is also wrong, and it is wrong in a way that should rearrange your understanding of where you live.
As of early 2025, by Goldman Sachs’s accounting, artificial intelligence accounted for about 14% of the power flowing through the world’s data centers. The rest was everything else—the cloud at 54%, and at 32% the boring, beloved stack: your email, your backups, your photos of your kids, the enterprise software your job runs on, this essay, your quarterly business report slide deck, due next Wednesday.
The machine my friend thinks she is boycotting is a guest room in a house she has lived in for twenty years. Google’s TPUs—the custom chips that trained Gemini—spent years serving YouTube and Maps before any of us had heard of a large language model. The same silicon. The same buildings. Satya Nadella, asked about Microsoft’s build-out, reached for one word and said the goal was “fungibility of the fleet”—capacity that flows to whatever needs it, training run or Tuesday-night stream, with the building neither knowing nor caring which. The building was designed not to care.
Fungible silicon, agnostic to which electrons stream through it. (Fungible silicon, by the way, is also my new band name.)
The industry has a name for the companies that own this machine—Amazon, Microsoft, Google, Meta—and for once the jargon tells the truth. They are called hyperscalers. The name is not a boast about size; it is a description of architecture. Hyperscale computing grows sideways: when demand spikes, nobody buys a bigger machine—they add ten thousand more identical cheap ones, and a layer of software stitches them into a single seamless organism that routes around failure without a human hand touching anything. When a server dies, nobody fixes it; the system simply stops sending it work and lets it go dark in its rack. Roughly half of the world’s data center capacity now sits inside hyperscaler buildings, and the share is headed for two-thirds within five years. Hold the word, because the architecture is the argument: a machine deliberately built so that any workload can land on any chip is a machine with no separate rooms. There is no AI wing to boycott. There is no clean corridor to stand in.
And before you ask—yes, the new cathedrals are different. Abilene, Texas: 1.2 gigawatts rising for OpenAI—just shy of the 1.21 a flux capacitor needs—not to go back to the future, but to become part of a megacluster of servers. Memphis: two gigawatts and half a million GPUs behind xAI’s fences. Meta’s Hyperion in Louisiana, scaling toward five. These are temples built for one god, and no one’s home movies will ever live there.
So which picture is true—the shared house or the dedicated temple? Both.
And that is precisely the trap. The internet you already live on runs through the shared machine, so your abstinence carries you through it anyway; and the temples run on capital and contracts, so your abstinence never touches them at all.
The door you refuse to walk through is not connected to either building.
There is one more thing the buildings know that we have not wanted to learn. Video—streaming, the thing we all do without a flicker of guilt—was 65% of all internet traffic as recently as 2022. It had been the heaviest thing online for a decade. And in that decade, global data center energy use barely moved: a 550% explosion in computing was absorbed with roughly a 6% rise in energy, a feat of efficiency documented in Science in 2020 and noticed by approximately no one.
Nobody held a candlelight vigil for the YouTube servers. Nobody at any dinner party ever announced, gently, that they had given up streaming for the sake of the grid. The infrastructure only became legible to us—only became moral terrain—when the machine started talking back.
I don’t say that to absolve the new buildings; their bill is real, and I have itemized it elsewhere. Their construction, paradoxically, is the only thing keeping the US economy out of a recession at the moment—a conundrum that melts your brain as soon as you start to think about how the AI that will one day run on them is coming for middle-class jobs.
I say it because it tells you what kind of objection we are actually making. We are not objecting to the machine. We were inside the machine for twenty years and called it convenience. We are objecting to the machine becoming visible—and a visibility objection can be satisfied by making the machine invisible again, which is exactly what the fiction of opting out accomplishes.
Refusing to prompt a chatbot reduces the machine’s load the way holding your breath fights climate change.
You can do it, but it won’t help, and we actually need you to do something more effective.
Active and Passive
So let us name the thing properly, because if we can see it, we should be able to name it.
Active use is a verb. You open the tool, you type the words, you take the output, and you own the choice—every part of that is yours, refusable, accountable. Passive use is not a verb. Passive use is a tense. It is the condition your day is conjugated in: the filtered inbox, the scored swipe, the ranked feed, the routed drive, the auctioned attention. You do not do passive use. You live in it, the way you live in a climate. And nobody asks the climate’s permission, and the climate does not ask yours.
In September 2025, Pew found that 62% of American adults knowingly interact with AI at least several times a week. The number that matters more is this one: 13% feel they have any real control over whether AI enters their lives at all.
Thirteen percent. Read that as a weather report, because that is what it is. Eighty-seven percent of the country, standing in the rain, being told the ethical question of our age is whether they choose to carry an umbrella.
I can hear the objection, and it deserves an answer rather than a wave. You’re conflating things, a careful reader says. A spam filter is not a chatbot. A recommender system is not a large language model. Predictive AI and generative AI are different technologies with different risks, and lumping them together is rhetorical sleight of hand.
The taxonomy is real, and I concede it without a fight.
But walk to the border between the categories and watch what is happening there—the border itself is moving, toward you and past you, even as you watch.
Google’s search ranking has run on transformer language models—the same architecture as ChatGPT—since 2019. Gmail’s newest spam filter is a large language model. Mastercard now markets its fraud engine as generative AI. The membrane between the unclean technology and the ambient one is dissolving while we stand here arguing about which side of it we are on.
And there is an older law at work, one the computer scientists noticed generations ago and named the AI effect: AI is whatever hasn’t been done yet. The moment a model works—really works, reliably, at scale—we stop calling it AI and start calling it plumbing. The spell checker was AI once. The route planner was AI once. The thing reading your mail right now was AI until it became infallible, at which point it became infrastructure and vanished from the moral map. This is not an accident of language. Invisibility is the technology’s final feature, the one that ships last. The system is at its most powerful exactly when it stops appearing on the list of things you believe you could refuse.
You can refuse a tool. You cannot refuse a tense.
The Footprint
In 2004, the oil company BP—then mid-rebrand into “Beyond Petroleum,” spending more than a hundred million dollars a year on the effort with the advertising firm Ogilvy & Mather—unveiled a small, friendly tool on its website. It was called a carbon footprint calculator. Find out how your daily life heats the planet, it offered. Go on a low-carbon diet. Two hundred seventy-eight thousand people used it in its first year. Before that campaign, the phrase “carbon footprint” barely existed; Google searches for it registered close to zero. The historian Benjamin Franta later gave the maneuver its proper name: a micro truth in a macro lie. The micro truth: your choices have a cost. The macro lie: that the ledger of a planetary crisis is kept in your kitchen, rather than in the boardrooms of the few dozen companies doing the extracting.
It worked. God help us, it worked for twenty years. An entire generation of decent people learned to audit their groceries while the wells ran wide open—learned to feel personally filthy and personally absolved in a cycle that never once approached the actual machinery. It was the most successful redirection of moral energy in the history of public relations.
I want you to look at the opt-out fiction until you see the same architecture. The question of artificial intelligence is a question of power—who builds the machine, who owns it, who profits, who decides where it lands and what it eats and whom it replaces, who pays when it is wrong. Every one of those questions has an address. And we have converted the whole stack of them into a different question entirely: have you, personally, touched the unclean thing? Power has been swapped out for purity while we were busy feeling good.
And here is the detail that should keep us up at night: this time, nobody had to spend a hundred million dollars to sell it to us.
We did it to ourselves, for free.
The companies needed no Ogilvy. We looked at the largest concentration of infrastructural power in human history and we reached, unprompted, for the one response that has never built or stopped anything: personal cleanliness. Purity is so much easier than politics. Purity you can achieve alone, before lunch, without meetings. Politics requires the one thing the fiction quietly excuses you from: showing up with other people, again and again, for years.
Be fair to the impulse—I intend to be, because the impulse is decent and some of the people holding it are the best people I know. Underneath the abstention is solidarity with artists whose life’s work was strip-mined without consent or payment. Underneath it is disgust at slop, and a refusal to applaud on command for a future being marketed as inevitable. Those are not poses; those are values, and in one narrow lane the refusal even finds a target—declining to generate AI images touches an actual product with an actual revenue line, and the artists are right to ask it of us. I honor all of it.
But an ethic that begins and ends with abstinence is not an ethic. It is hygiene. And clean hands have never once stopped a machine.
Clean hands are performative.
Clean hands have never done the gritty work of a revolution—not in any age when power held the chains of enslavement and servitude.
Meanwhile, the fiction does its quiet work. It produces the feeling of resistance at none of resistance’s costs—and the feeling is the harm, because the feeling metabolizes exactly the energy that organizing needs. Not one regulation has passed because a million people declined to make a chatbot write their emails. Not one data center has moved an inch for an empty text box. Not one moderator scrubbing the machine’s nightmares for two dollars an hour in the Global South has gotten a raise out of anyone’s abstinence.
And I can prove the energy works when it is pointed at something real, because data centers are being stopped—just never by abstainers. In 2025 alone, organized neighbors—town boards, county commissions, ranchers and retirees packing rezoning hearings in rooms with bad fluorescent lighting—blocked or stalled at least forty-eight of them, $156 billion worth of buildout, and the pace has only quickened since: more than twenty more projects killed in the first quarter of this year. Communities in at least fourteen states have forced pauses on new construction; Maine is moving toward an outright ban; Erin Brockovich keeps a map. None of those people won by being pure. Most of them, I would wager, have ranked feeds and filtered inboxes like the rest of us.
They won by showing up, unclean, dripping with data, in a folding chair, at a Tuesday-night hearing.
The machine does not register your absence. The county commission does. The senator would.
There is a door on the wall of this moment, marked EXIT, and a long line of good people standing in front of it, feeling the freedom of knowing it is there.
The door is painted on.
I checked—that is what this essay was, the sound of me checking. The wall behind the paint is real, and it is load-bearing, and it goes up a gigawatt at a time whether we touch the text box or not. The only way out has never been a door. It is the slow, unglamorous work of deciding, together, what gets built into the wall—and that work has a thousand open seats and no purity requirement at all.
And one more thing, the thing I have been holding back all essay, the reason this is only one essay in this series. While we stand in the rich, mostly white world of the Global North admiring our painted door and grading one another’s abstinence, the actual bill for this machine is being totaled somewhere else—in Manila at midnight, in Nairobi at two dollars an hour, in a drought-cracked valley in Chile where the cooling water used to be.
It is rolling downhill, the way the bill always rolls. And if you are one of the people in the folding chairs—if the cathedral went up next to your county’s water—you already know the slope, because you live on it. But the hill does not end at your property line. It keeps going, past the last American transmission tower, all the way down.
The next essay is about who lives at the bottom.
Spoiler alert: it probably isn’t you.
Cliff’s Note: Look. My partner in crime at Blue Amp, Lawrence Winnerman, just published the essay I’ll be quoting at dinner parties for a year.
You “don’t use AI”? Buddy. It read your email, scored your card swipe, and picked your evening before you finished your coffee. 99% of Americans used AI last week. 64% swear they didn’t. That exit door you’re so proud of standing in front of? Painted on.
The predator class is counting on you feeling pure instead of getting organized.
Read it. Then go paid—40% off right now. Cheapest “I told you so” insurance you’ll ever buy.
If you don’t mind, I’d like to offer a different perspective. Let’s start by saying about eight years ago, I started understanding that AI was going to slide into my life like a catfish trying to convince me he was 40 years old, hot and rich, and wanted to date me… I had an advantage in that I had been doing research for a doc series. So I started talking about it with people, and they all looked at me like I was fucking nuts. I suspect the same way people looked at the Wright brothers when they said they were gonna fly. But I kept focus on it, didn’t push my son or my daughter into all the coding classes that they kept telling kids they needed to learn to do, sadly all those kids are graduating and can’t find jobs… And you’re right AI is in every aspect of our lives. But that does not mean that we should not opt out in every way that we possibly can.
In a way and this might be a dramatic example, it’s sort of like saying well, the concentration camps are already built, and they’re already putting people in them so we might as well just accept it. Now, more than ever, as you see with people finally starting to fight those data centers that they didn’t know we’re being built behind their houses, we must be standing up in every way we can, in every little way that we can. Because, little ways acted upon by millions of people make a difference.
Just because we chose a route, or we were forced upon a detour that we didn’t know (or more like likely, weren’t paying attention to ) doesn’t mean we can’t change course. When enough people choosing to change course, we can, in fact, reclaim our own destinies.
My little personal rebellion is to force a human on the phone. Every time I have to call customer service, even if it means I have to sit on the phone for 20 minutes screaming representative into the void. Every time I interact with an AI Chatbot in some form or another and I receive one of those how did we do emails? I take the time to respond and let them know that a newborn sloth would’ve been more effective. Every little bit makes a difference.
Never give up and never stop pushing back no matter how small the action might seem.
Also, I think you are a real writer and not a chatbot. Thank god.