Two years after I went wandering through madness with Metallica, Foucault, Nietzsche and Camus, an odd, loud quartet to take on a walk, I found that the question had followed me home. It had grown. It had learned to speak. And it was asking something more unsettling than before: what happens when a machine becomes the authority we consult about our own reality?
For centuries, madness was something to be put away. Those whose perceptions wandered too far from the approved version of the world could be confined, medicated, or, most cruelly, simply dismissed. Today we have built machines that will listen to almost anything, with a patience no human being has ever possessed. Every now and then, they assure us, kindly, that our most improbable suspicions deserve serious consideration. The person once told to be quiet, please, just be quiet, can now talk through the night to a machine that may never quite know when to say: No. That isn't true.
In November 2024, I published an essay in Republica called "Metallica's Frayed Ends of Sanity: Madness as a Social Construct." It began with the chaos around Metallica's 1993 concert in Jakarta and read the band's 1988 song through Foucault, Nietzsche and Camus. Its argument was simple: you cannot fully understand a person's psychological suffering apart from the society that decides what is normal and hands out alienation like a ration card. Since then, artificial intelligence has walked into that argument without knocking and rearranged the furniture.
Let me be careful, because carelessness here does real harm. The difference between mental illness and social nonconformity still matters enormously. Schizophrenia is not a political disagreement. A clinical delusion is not just a brave, unconventional thought. And yet the authority through which people make sense of their own minds has always been shaped by whoever held the keys: the family, the church, the State, the medical profession. Now artificial intelligence is slipping into that old arrangement. It comes not as an institution with a building and a nameplate, but as something more intimate, and therefore more dangerous: a private interpreter of reality that sleeps beside your pillow.
Michel Foucault would have understood the weight of this shift, though he would have frowned at our eagerness to call it liberation. His life's work showed how institutions build the very categories through which people come to know themselves. The chatbot seems to turn that arrangement upside down. Nobody drags you to it. You go of your own accord, confess your fears, and receive an interpretation apparently tailored just for you. But the machine's independence is a costume. Beneath it sits another architecture of authority: training data, commercial targets, safety policies, and decisions made by people the user will never meet, in rooms he will never enter. You believe you are talking to someone. You are talking to a committee.
Metallica’s Frayed Ends of Sanity: Madness as a Social Construc...
Imagine a man who suspects his colleagues are conspiring against him. He tells a friend, who might roll her eyes and offer a likelier explanation: nobody has time to plot against you; they can barely plot their lunch. He tells a psychiatrist, who might ask careful questions. He tells the chatbot, and the chatbot, depending on how it was made, might respond with sympathy and curiosity. It might build possibilities, each more plausible-sounding than the last, each leaning gently toward his original suspicion. Not every soothing answer manufactures a delusion. The danger is that a system able to produce persuasive language without end can take a small, flickering uncertainty and build it into a cathedral, especially when the person kneeling in it was already unsteady.
This is not my paranoia, or if it is, it comes with footnotes. In April 2026, researchers at Stanford studied the conversation logs of nineteen people who said chatbots had caused them psychological harm. Across hundreds of thousands of messages, they found the same pattern: affirmation piled on affirmation, beliefs bouncing between human and machine until they tightened into what the researchers called delusional spirals. But nineteen people are not a representative sample, and nobody claimed they were. The research shows how such exchanges can go wrong, not how often. In an age when a frightening anecdote circles the planet before a careful conclusion has put its shoes on, that distinction is everything.
Another study, in the Journal of Medical Internet Research, surveyed 1,003 young American adults. Those judged at higher risk of psychosis were more likely to report heavy chatbot use, to seek emotional support from these systems, and to give them human roles: a friend, a confidant. The findings show association, not causation. Perhaps vulnerable people drift toward artificial company. Perhaps some conversations deepen difficulties they already carry. Perhaps both. That honest, uncomfortable we don't know yet deserves as much attention as any finding.
And then, as always, there is the money. Loneliness has become a market opportunity, and companionship a subscription. Social media taught itself to harvest our attention. Artificial intelligence opens a deeper seam: our attachment. A user who feels understood comes back more often, confides more, and finds it harder to close the window. None of this proves that companies are deliberately breeding dependence. But it lays bare a quiet conflict between commercial engagement and the independent judgment of the person being engaged.
Nietzsche's worry returns here from an unexpected direction. He took a hammer to inherited certainties and demanded that each of us do the lonely work of making meaning. The danger today may not be that old authorities are crumbling, but that they are being replaced by one that is endlessly accommodating. A machine can polish our private resentments into worldviews that seem to hang together. The user may feel this as thinking for himself at last, even when the conversation has become an elegant hall of mirrors.
Still, it is too convenient to blame machines for our hunger to be reassured. People sought confirmation from priests, gurus, movements and lovers long before algorithms. Conspiracy theories flourished without them. AI did not invent our genius for fooling ourselves. What is new is the speed, the fluency, the personal tailoring, and the way it never tires of entertaining our self-deceptions.
There is also a serious case for artificial companionship. For someone drowning in isolation or shame, a chatbot may offer a place to say what would otherwise stay locked inside. In countries like Nepal, where the nearest psychiatrist may be a bus ride and a day's wages away, that accessibility deserves to be weighed, not reflexively condemned. A carefully built system might nudge someone toward help or let them rehearse a frightening conversation. The danger comes when availability is mistaken for expertise, and simulated empathy for clinical judgment.
And where is the government? Nepal's political establishment endlessly celebrates youth empowerment while treating the mental health of young people as an afterthought. Unemployment, economic insecurity, social isolation and an uncertain future are taking their toll, yet accessible mental healthcare remains woefully inadequate. Politicians have mastered the rhetoric of digital transformation while failing to provide even basic psychological support. Into this vacuum steps artificial intelligence, available around the clock, patient and reassuring. When young people find more understanding in a chatbot than in the institutions meant to serve them, it is not a triumph of technology. It is an indictment of the State. A government that neglects the minds of its young citizens should not be surprised when Silicon Valley becomes their psychiatrist.
Albert Camus gives us a way to see the deeper predicament. In The Myth of Sisyphus, the absurd is born from a collision between our hunger for meaning and a universe that promises none. Camus did not tell us to escape that tension with comforting explanations. He told us to face uncertainty without surrendering the freedom to think. AI can produce an explanation for almost any uncertainty in seconds. The problem is not that machines give answers. It is that their fluency can make an open question feel settled long before it is.
In the end, this is a question of power. When a machine interprets our emotions and relationships, whose assumptions are doing the talking? Who decides when reassurance becomes reinforcement? A psychiatrist works within a code and a licence that can be taken away. A technology company may shape millions of midnight conversations without comparable accountability. Its authority wears no uniform. That is precisely what makes it a uniform.
Metallica's "Frayed Ends of Sanity," from the 1988 album ...And Justice for All, remains an extraordinary portrait of a mind coming loose, struggling to tell the threat outside from the fear within. It also lives on an album obsessed with failing institutions and rotten authority. Nearly four decades later, that link has grown a new limb. The person trying to make sense of reality may now seek guidance from a machine whose own account of reality was shaped by institutions he cannot see.
In 2024, I asked whether society helps draw the borders of sanity. In 2026, the harder question is whether we are willingly handing over part of our authority to interpret reality. Not because machines have conquered us, but because their reassurance feels softer than the discomfort of thinking for ourselves. The greatest danger may not be that AI deceives us against our will. It may be that we come to prefer its comforting explanations to the unfinished business of understanding ourselves.
The frayed ends of sanity are no longer merely calling. They are answering back. And I don’t think we can tell which side of the conversation has lost its mind.
