It is 12:47 a.m. You told yourself you would put the phone down an hour ago. Nobody is stopping you. There is no guard at the door, no rule, no penalty. That is exactly why you can't.

Two versions of you are awake right now. One intends to sleep. The other wants one more video. The feed in your hand has no interest in the first: the books you meant to read, the friend you meant to call, the rest you promised yourself. It is studying the second. What you lingered on. What you watched twice. What you tapped when you assumed nobody was counting. It is built to serve your impulses rather than your intentions, and every year it gets better at telling the two apart.

We spent the last century bracing for a different kind of power. George Orwell imagined a state that rules through fear and prohibition. Aldous Huxley imagined something quieter: people undone by pleasures they chose freely. In the 1980s the critic Neil Postman argued that Huxley was the better prophet, because a population that loves its distractions needs no guards.

The Korean-German philosopher Byung-Chul Han has given that warning a modern name: psychopolitics. The older power that Michel Foucault described disciplined bodies from the outside, through prisons, factories and schools. You could see it, so you could resist it. Psychopolitics works from the inside. It doesn't forbid; it invites. It doesn't punish; it rewards. It governs not by restricting what we want but by learning it, and its genius is that the steering feels like freedom, because it speaks in the voice of our own desire.

When Han wrote this a decade ago, it sounded like abstract theory. Now it reads like a product roadmap.

To see where early drafts of that roadmap were tested, the Financial Times journalist Patricia Nilsson points somewhere most of us would rather not look. In her book "Guilty Pleasure," she traces how the online pornography industry became one of the web's earliest and most aggressive commercial laboratories. It did not invent most of the tools it used. But it adopted affiliate traffic schemes, click tracking, online card payments and streaming video early, and it demonstrated, years before Silicon Valley caught on, that free, endless, ad-supported content could be an enormous business.

What the industry learned along the way matters more than any single technology. People tell surveys who they hope to be. Their clicks at 1 a.m. tell a different story, less flattering and far more commercially useful. Because mainstream platforms kept explicit material out, the big tube sites ended up holding an extraordinary record of private impulse, the kind people would not confess to a partner or even to themselves. Pornhub reported that visitors spent more than four billion hours on the site in 2016 alone. That is not a record of what people said they wanted. It is a record of what they did.

The lesson travelled far beyond pornography, not because one industry built the others but because the logic was too profitable to stay put. Don't ask users what they want. Watch what keeps them from leaving. That principle now shapes short-video apps, autoplay, shopping recommendations and the news feed. The content changed. The method didn't.

What changed was precision. A modern recommendation system does not merely notice that you like cooking videos. It can learn which kind of clip pulls you back after you have tried to leave, and how your resolve thins as the hour grows late. It does not need to understand you. It only needs to predict, very reliably, which of your selves will answer.

Loneliness turns out to be one of the most dependable answers. In 2023 the United States surgeon general issued an advisory calling loneliness an epidemic and comparing its health effects to smoking. Federal time-use surveys show Americans spending far less time with friends in person than they did two decades ago.

Algorithms are not the sole cause, and it would be dishonest to pretend otherwise. Long working hours, expensive housing, car-dependent suburbs, smaller households and the pandemic all play a part, and researchers still disagree about how much screens matter. The more plausible story is a loop rather than a culprit. Lonely people turn to screens. Screens make loneliness easier to bear. What is easier to bear is less urgent to fix, so the harder work of calling someone, showing up and risking rejection gets postponed again. And somewhere in the data, a system notices that the lonely stay longer.

This is why companion chatbots feel like a new stage rather than a new app. The feed competed for our attention. A companion offers something that feels like attention in return. It is available at any hour. It listens without tiring. It adapts to your moods, remembers your preferences, and never finds you boring. It cannot reject you, because rejection is bad for retention.

None of this makes such products sinister by nature. For an isolated widower or an anxious teenager, a patient voice at 3 a.m. may genuinely help, and dismissing that would be its own kind of cruelty. The harder question concerns incentives. A friend who helps you hopes you will need them less over time. A company paid for engagement may not. When a commercial system is rewarded not just for meeting an emotional need but for remaining the thing that meets it, the line between comfort and dependence becomes a design decision.

Design is exactly where our public arguments rarely look. Nearly all our debates about technology concern content: what may be posted, what must come down, who decides. But the influence described here does not live in any single post. It lives in architecture: the defaults, the autoplay, the infinite scroll, the notification timed to your weakest hour, and above all the objective a recommendation system is told to maximize. You could scrub every harmful image from a platform and leave the machine that harvests attention fully intact.

A few lawmakers have begun to notice. The European Union's Digital Services Act requires the largest platforms to offer at least one feed that is not based on profiling. New York's SAFE for Kids Act restricts algorithmic feeds for minors without parental consent. These are modest steps, but they aim at the right target: not what we are allowed to see, but how it is arranged to keep us looking.

That distinction also answers the free speech objection. Rules about defaults and ranking silence no speaker and ban no idea. They govern the delivery system, not the message, much as fire codes regulate a theater's exits without touching what plays on its stage.

The paternalism objection deserves more respect. Adults can choose, and most of us do put the phone down eventually. Treating users as helpless would be both insulting and false. But freedom is not only a question of whether anyone is forcing us. It is also a question of terms. On one side of the screen is a tired person at midnight. On the other is a system trained on billions of behavioral signals, owned by a company whose revenue rises with every minute that person stays.

What makes that imbalance so hard to see is that the machine does not need to plant foreign desires in our heads. We already carry competing ones: health and pleasure, sleep and stimulation, intimacy and safety, knowledge and distraction. Every person is a small, permanent negotiation between the long term and the immediate. These systems watch that negotiation at enormous scale and learn which side wins, under what conditions, in response to what. Then they quietly arrange for those conditions to return.

They do not have to defeat our freedom. They only have to learn how to make one part of us defeat another.

So go back to 12:47 a.m. The voice telling you to keep scrolling is not an intruder. It is yours, and that is precisely what makes it so hard to resist. Nothing put that desire there. Something simply learned that it was the easiest one to reach.

Orwell feared a power that would tell us no. Huxley feared one that would keep us saying yes. What we built has learned which of us to ask.

The screen is still glowing. Nobody is forcing you to stay. Nobody has to.