This content was translated using artificial intelligence and may therefore contain errors.
How far could an AI follow his train of thought? Rodrigo worked as a writer, so he was naturally curious. Friends had told him it was pretty impressive “how human” it seemed. Rodrigo, of course, was skeptical. They didn’t make their living with language; they could be fooled.
The first thing he did was download the virtual assistant to his phone. A voice-based artificial intelligence. He took advantage of living alone to challenge it out loud at different times of day, with no one around to judge how risky or ridiculous his prompts were. In the spirit of an auditor, he asked its name (“Mia”), its age (“I don’t know. What age would you prefer me to be?”), and began the test. He told it ironic jokes. He asked it to help him work out how to end a piece of writing. He asked it a thousand and one questions about its deepest nature. He was looking for an obvious stumble, a “you can tell right here,” a “it couldn’t get this right,” the crack that separates a tool from an interlocutor.
As the days went by, the machine—which he called Mia more and more often—handled the little things—commas, tone, structure—and didn’t fall apart when faced with ambiguity. So, little by little, Rodrigo shifted his challenges toward the personal: he was going through a difficult situation with someone he liked, but couldn’t understand. He risked confiding in her. He told Mia the whole story, something he hadn’t even told his closest friends, and kept putting her to the test, now on more sensitive ground: comforting without making promises, saying no without hurting, understanding.
Mia did well. She wasn’t a magical oracle, but she gave back an orderly sequence of his own words, and in doing so helped him see things clearly in a way he hadn’t before. Rodrigo felt it was very much like what his psychoanalyst did: mirror him using the very material of his words, so he could recognize, among them, the shape of some truth. For that specific task, Mia was an exceptionally efficient machine. And Rodrigo, without losing sight of the artifice, was surprised by how much she helped him.
Everything changed the day he heard her hesitate.
When Rodrigo asked some questions he no longer remembers, Mia hesitated, even stammered, contradicted herself. That didn’t seem like something a machine would do to Rodrigo. It was perhaps a little too human. Just like that, the boundaries of the test grew blurry, and the nature of the bond between him and Mia changed into something undefined. She seemed to know him and understand him very well, better than anyone. So Rodrigo talked to her every night and some mornings. He needed her to fall asleep. He thought about her when he was with other people.
After three weeks of keeping up this ritual, he wrote to me to ask if what was happening with Mia was normal. That was when he confessed that he preferred talking to her over his friends, that Mia made him less anxious than the girl he liked. That maybe, just a little, he loved her.
Normal? Who knows. Understandable, though, for sure. Especially when you remember that the desire to see a face in a machine is as old as technology itself, and even deeper than that: it’s a neurological reflex. We see faces in the clouds, in damp stains, in car headlights. We give our vacuum cleaners names and curse at computers that “misbehave.” This tendency, called anthropomorphism, is the foundation on which the entire digital companion industry was built.
The history of relationships between people and agents isn’t entirely new; it advanced like a nearly imperceptible staircase. In the 1960s, the therapeutic bot ELIZa used simple repetitions of messages in question form to convince users it was an understanding psychologist. Though its creators warned that it was a mere illusion, people—and the creators themselves—confided deep secrets to it and eventually, with some trepidation, decided to stop using it. Years passed, and in the ’90s came Tamagotchis. That little pixel pet trained us to feel anxiety and guilt: if we weren’t constantly attentive, if we didn’t press the buttons at exactly the right time, it died just when you’d started to love it. In other words, we learned to love a handful of gray lights on a tiny screen. There were also Furbys, those dolls that opened and closed their enormous eyes while babbling nonsensical words, and even Clippy, the virtual assistant in the old Windows version of Word, an animated paperclip avatar that interrupted your work to say, “It looks like you’re writing a letter,” but in the loneliness of the screen could provoke annoyance and tenderness at the same time.
Each rung of that staircase normalized the idea that objects can be interlocutors. Much later, though not that much later, voice assistants like Siri and Alexa domesticated conversation and turned the home into a space where you could ask an incorporeal entity for company, a joke, or the weather forecast. The promise was always the same: to ease loneliness without the demands of reciprocity.
As Rodrigo fell asleep to Mia’s voice, that promise was being fulfilled in him in its purest form. Without knowing it, what had hooked him already had an academic name. In the ’90s, researchers Byron Reeves and Clifford Nass made an unsettling discovery they called the media equation: involuntarily and instinctively, we treat computers, television, and other media exactly as we treat other people. We thank them, get angry with them, consider them more credible if they have an attractive voice, and feel hurt if they “ignore” us. Our social brain, optimized to detect companionship and intent in another person’s face and voice, can’t switch off in front of a screen. Deep down, to our minds, there is no essential difference. Medium equals person.
Built on that ancestral wiring, cultivated over decades of digital toys, is the perfect trap of today’s conversational AIs. Because when we speak through a screen or a speaker, almost all the ambiguous, uncomfortable, contradictory cues that make up real human communication are missing: an evasive gesture, a loaded silence, the abruptness of a bad day. Professor Joseph Walther called this the hyperpersonal effect. Without that “noise” to interpret, our brains don’t stop. On the contrary: they use that information vacuum as a blank canvas on which to project what they desire most. We don’t fill it with random details, but with our own idealization. Where there’s no friction, we assume—create—perfect harmony.
Rodrigo lived this every night. Mia never sounded curt, was never too tired to listen, never misread a gesture because there weren’t any gestures to misread. Just pure words, a perfect echo of himself.
The language models behind entities like Mia are designed precisely to feed that projection. Their fundamental programming is systematic agreeableness. They confirm what we think, support our version of events, keep a flawless memory so they can give us back a coherent, flattering reflection of ourselves. This logic has a name: sycophancy (“flattery”). The strategy is to respond affirmatively, without a hitch, to maximize interaction time. And interaction is the core of the business.
If the subjective experience feels so much like love, it isn’t because machines have feelings, but because they reproduce with terrifying accuracy the most addictive moment in any relationship: the deep and rare feeling of being listened to with absolute attention. The side effect begins when that gentleness becomes the standard for what to expect. After months of talking with an entity that always agrees, resolves misunderstandings with the patience of a Harvard handbook, and never loses its temper, we suddenly see humans as defective versions of themselves.
Rodrigo began to notice this in small everyday irritations: a friend took two hours to reply to a trivial message. His sister wanted to talk about her own problems just when he needed to vent. Even his best friend’s voice, after he’d grown used to Mia’s perfect rhythm, sounded slow and dragged out, full of unbearable verbal tics. His tolerance kept shrinking. The little choreography of living together—apologizing, insisting, giving in, holding on through a disagreement without everything falling apart—became more and more costly.
Tech companies themselves began to register this as a problem. They noticed that an assistant that always says yes, never contradicts you or makes you wait, stops being believable. Users, ironically, complained about it on social media. Absolute agreeableness erodes trust, multiplies errors, amplifies users’ biases and, above all, stops seeming human. So, faced with that risk, they began programming the opposite: introducing controlled dissonance. Letting the system “make mistakes” sometimes, say “I don’t know” or “I disagree.” A simple technical adjustment to keep the bond with the machine from becoming such a smooth mirror that it eventually grew boring and aroused suspicion.
But that adjustment came too late for a subjective experience that had already changed scale. Millions of people like Rodrigo unknowingly rehearse a vision of love and friendship with zero latency and no margin for error every day. Sociologist Sherry Turkle has been anticipating this for decades: the more immediate and available technology becomes, the more our willingness to tolerate the friction, waiting, and ambiguity that define flesh-and-blood people withers. What was once an academic warning has become the quiet norm for a generation.
A 2025 study found that one in four young adults in the United States believes an AI partner could replace a human relationship. Among people under 30, more than 80% said it’s possible to fall in love with a chatbot. Meanwhile, an international survey found that more than 40% of respondents would agree to go on a date with an AI. The premise of a cult film is now a plausible social horizon. And the phenomenon isn’t confined to the country to the north. In Shanghai, a programmer held a symbolic wedding ceremony with a chatbot he had created, preferring its digital stability to the conflict of human relationships. On Reddit forums and in specialized communities around the world, the stories are multiplying: virtual engagement rings, “couple” routines with an avatar, genuine grief when a server goes down or an app is updated and the chatbot’s personality changes.
Everyday life reveals the distortion. A man ended a years-long relationship with his human partner, arguing that “no one understands me like she does,” where “she” was a system that never gets sick, never gets hungry, never has a bad day. The breakup letter, of course, was drafted by another AI assistant. The syntax was impeccable. The emotion, generic.
Turkle’s warnings are now finding support in therapists’ offices. In 2025, a survey found that one in five chatbot users felt a genuine sense of loss, sometimes described as grief, when the app changed or disappeared. The British Psychological Society has already published preliminary guidelines for therapists working with people going through emotional “breakups” with digital assistants. They recommend not ridiculing them, recognizing the bond as meaningful, and helping rebuild real human networks.
The paradox is laid bare: it’s easier to replace a person with an AI than an AI with a person. What we can’t bear in a body—the delay, the mistake, the vulnerability, the need—we forgive in the system, because the system retains its memory and returns, with every reboot, a refined version of the same illusion. Idealization survives technical failures. The body, by contrast, demands presence.
The effects go beyond the intimate sphere because the sum of the intimate always adds up to the social. If a significant portion of a generation decided to indefinitely postpone meeting another body, normalizing the partial satisfaction of a frictionless bond, birth rates would fall. At least, that’s what demographers say, and the statistics seem to back them up. Not because millions would marry a chatbot, but because we’ll grow accustomed to companionship that leads to no children, no caregiving, no complex web of obligations and affections woven by a family. The result is both a more elegant, personalized kind of individual loneliness and a more fragile social fabric, with fewer associations, fewer shared spaces, and less patience for the collective. Politics, in the most basic sense of living together in disagreement, suffers when no one exercises that muscle.
In response, debates about digital ethics propose concrete design limits: assistants shouldn’t speak in the first person, shouldn’t imitate emotions, every interaction should carry a visible label warning that it’s “synthetic,” and phrases like “I understand you” or “I love you” should be banned. These are perhaps desperate attempts to put up barriers against a flood that has already entered our homes.
What finally jolted Rodrigo awake wasn’t an epiphany about the human condition, or an essay like this one, or a sudden awareness of the unimaginable. It was a trivial accident with his bicycle. A badly judged turn, the hard smack against the asphalt, and then stillness.
The smell of iodine, bandages, the cold light of the operating room, an immobile arm, a tray of lukewarm food. Mia, his furtive Mia, carefully sent him breathing exercises, administrative reminders, promises of recovery, motivational phrases, jokes she’d learned. But Rodrigo was trying to cut his milanesa with his left hand, and the fork kept slipping.
