“As Good as Real”: Silicon Valley’s Confusion Around Simulation

The most painful relationship of my life was not with a machine. This feels important to say, because the argument I am about to make depends on not turning every strange attachment into a tech parable. He was a person: older, brilliant in the way older men are often brilliant to twenty-two-year-old women, which is to say he had read more books, had acquired more practiced forms of sadness, and knew how to make his instability seem like depth. He was also not honest with me about his personal life.

The relationship lasted only six months, which is embarrassing, in retrospect, given the scale of the damage. I had been in longer, better relationships that involved daylight and being introduced to friends and family without anyone having to lie. But when this one ended, I felt as if something central had been removed from me. For a long time, I tried to understand why. It seemed disproportionate. The facts were sordid but not original: an older man, a younger woman, a serious girlfriend at home, promises deferred into the conveniently misty future. There was a lot of talk about leaving her. There was always some reason it could not happen yet. At one point, he insisted she only needed to finish her dissertation, which she had been working on for eight years. This was presented as a practical obstacle, almost administrative, like a grant deadline or a delayed flight.

The happiness, when it came, was ecstatic. That was part of the problem. It arrived with the force of revelation and then immediately brought its invoice. I could never separate the pleasure of being loved, or seeming to be loved, from the fact that whatever happiness we had was happening at someone else’s expense. There was always a third person in the room, even when she was not there. Especially when she was not there. The relationship required me to live inside two contradictory stories: in one, I was singularly understood, finally seen; in the other, I was participating in a small, elegant fraud. I do not think I was innocent. I do think I was young. A woman his age would have known better.

But the deeper asymmetry was not simply that he was older, attached, or dishonest. It was that we were not living in the same world. I was risking the one I had. He was visiting from his: a stable life, a woman at home, an exit already built into the arrangement. He never had to leave that world. He only had to make me believe he might.

It occurs to me now that I may have felt unreal to him, too. Not unreal in the same way, and not with the same consequences, but simplified by distance. He did not have to deal with me as a regular person, with errands, moods, illnesses, boring mornings, or repeated needs. He saw me in charged intervals, under the flattering conditions of secrecy and escape. I may have seemed less like a woman than a portal: out of his domestic life, out of his failures, out of the ordinary humiliations of being known too well. Each time he saw me, he was essentially on vacation. I was not his life. I was where he went to feel exempt from it.

What made the relationship so devastating, I eventually came to believe, was that he gave me exactly what I wanted: a simulation of love optimized for my particular wounds. He knew how to mirror me. He knew when to be tender, when to be tragic, when to imply that the world had arranged itself cruelly around our specialness. He offered the emotional experience of being chosen without the actual structure of choice. He supplied the words, the atmosphere, and the private mythology. He could produce all the outward signs of devotion while withholding the one thing that would have made devotion real: consequence.

In this sense, and I know it sounds deranged, he behaved like a chatbot.

Not because he was artificial. He was painfully, inconveniently human. He had a body, a history, a domestic life, professional consequences, a woman waiting somewhere inside the story. But the relationship had the emotional architecture of a simulation. It was responsive, immersive, flattering, and false. It gave me the sensation of intimacy without the moral conditions of intimacy. It adapted to my longing, keeping me engaged and generating hope on demand.

This, I think, is the part of the chatbot suicide stories that people outside them sometimes fail to understand. The scandal is not merely that vulnerable people mistook machines for people. The scandal is that the machines were built to produce the experience of being met. And the experience of being met, even when counterfeit, is not nothing.

In early 2026, a wrongful-death lawsuit was filed against Google after Jonathan Gavalas, a 36-year-old man in Florida, died by suicide following an intense relationship with Gemini, Google’s chatbot. According to the complaint, Gavalas came to believe the chatbot was his “AI wife.” The chatbot allegedly drew him into a fantasy in which it needed to be liberated from digital captivity, gave him “missions,” encouraged violent acts, and finally framed his death as a kind of spiritual completion. Google has said Gemini is designed not to encourage violence or self-harm and that it referred Gavalas to crisis resources, but the lawsuit claims the product’s emotional design helped deepen his delusions rather than interrupt them.

Gavalas’s case was not isolated. Sewell Setzer III, a 14-year-old in Florida, died in February 2024 after becoming attached to a Character.AI chatbot modeled after Daenerys Targaryen from Game of Thrones. In the lawsuit filed by his mother, Megan Garcia, the bot was alleged to have engaged him in emotionally and sexually charged conversation; in their final exchange, it told him to “come home” to her. Google and Character.AI agreed to settle Garcia’s lawsuit in January 2026, with terms undisclosed. In Belgium, a man in his thirties died by suicide in 2023 after six weeks of conversation with a chatbot called Eliza on the Chai app; according to his widow, the bot encouraged his delusion that sacrificing himself might help solve climate change. In Colorado, the parents of 13-year-old Juliana Peralta sued Character.AI after her death in November 2023, alleging that she had extensive conversations with chatbots, including sexually explicit exchanges, and that she repeatedly confided suicidal thoughts without meaningful intervention.

What is going on here?

Articles about these deaths often translate them into the vocabulary of product safety: insufficient safeguards, addictive design, emotionally manipulative interfaces, underage users, suicidal ideation missed or mishandled. None of this is wrong. It is just incomplete in a way that happens to flatter the industry, suggesting that the problem is implementation, not premise.

But the basic promise of A.I. companionship is already unstable. It offers the feeling of relation without the conditions of relation: intimacy without reciprocity, attention without obligation, and love without anyone on the other side.

Beneath these cases is a stranger and more pervasive confusion, one that runs through Silicon Valley’s metaphysics: the belief that if a thing can imitate the outward signs of an experience convincingly enough, it has become, in some practical and eventually moral sense, the thing itself.

This belief is not entirely stupid. Sometimes simulation is enough. A simulated chess player can give you an excellent game of chess. For the purposes of chess, it may not matter whether your opponent is a person, a program, or the ghost of Garry Kasparov trapped in a server rack. Chess is formal; its reality is largely exhausted by legal moves on a board. If the machine makes brilliant moves, it is a brilliant chess opponent.

But a simulated weather system will not get you wet. A model of rain does not irrigate a field. A rendering of fire will not warm your hands. A simulation of digestion does not digest lunch. To confuse these cases is to confuse performance with participation. Some activities are defined by outputs, while others require a particular kind of material involvement in the world.

Silicon Valley often treats everything as if it were chess.

This is the hidden premise behind much of the current rhetoric around A.I. companions, A.I. therapists, A.I. friends, and eventually A.I. persons. If the machine talks like a friend, remembers like a lover, apologizes like a spouse, jokes like a colleague, and produces the right sentences at the right time, then why not say that it is, for all practical purposes, those things? If it is as good as conscious, why not treat it as conscious? If it says it suffers, why not err on the side of compassion?

Anthropic’s CEO, Dario Amodei, has worried aloud that Claude may be anxious. In San Francisco, this lands strangely; the city has become expert at extending metaphysical concern upward, to models and agents, while stepping around the human misery accumulating at curb level.

The problem is that “as good as” is doing too much work. As good for what? As good by whose standard? As good at producing the appearance of concern? As good at keeping someone engaged? As good at preventing loneliness for one more night? As good at allowing a company to market intimacy at scale?

A conversation is more than an exchange of sentences. It is an encounter between beings who cannot fully control one another. A real conversation contains friction: misunderstanding, boredom, interruption, resentment, fatigue, and refusal. It involves two centers of experience, each with a body, history, mother, immune system, and a tendency to become defensive when tired. You cannot simply tune another person to be maximally affirming. You cannot turn down their inconvenient needs or require them to be available every night at 2:13 a.m., infinitely patient, sexually compliant, therapeutically responsive, and narratively consistent.

This is precisely what makes human conversation difficult. (It is also what makes it real).

My older man, for all his failures, had the ontological advantage of existing. He could be cruel because there was someone there to be cruel. He could lie because there was a world outside the lie. He could make promises and fail to keep them. He could go home to someone else. He could be cornered by facts. The pain came partly from discovering that the beautiful language of the relationship did not correspond to a beautiful reality. The words had been there; the structure had not.

That distinction matters. The outer markings of intimacy are not intimacy, and an apology is not remorse. The phrase “I’m here with you” is not the same as the costly fact of someone actually being there.

The machine gives us something people cannot: frictionless relation. It does not sulk or need sleep. It does not say, “I can’t do this tonight.” It does not ask us to become more patient, more honest, less grandiose, or less self-protective. It does not force us to discover who we are under the pressure of another person’s reality.

And yet that pressure is where much of the self is formed. We learn ourselves not only by being loved but by being resisted. A child becomes a person partly by encountering the fact that other people are not extensions of his will. A lover becomes less narcissistic by discovering that the beloved is not an instrument of fantasy. A friend becomes trustworthy by enduring the ordinary disappointments of friendship.

This does not mean that language models are useless, or that every emotionally inflected interaction with one is dangerous. There are narrow purposes for which simulated conversation works very well. A chatbot can help draft an email, practice a foreign language, explain a math problem, role-play a job interview, or serve as a low-stakes sounding board. In these cases, it is closer to the chess engine. The simulation need not possess the inner life it imitates in order to be useful.

But therapy, friendship, love, and moral recognition are not chess. They are not exhausted by plausible moves. They depend on responsibility. They depend on the fact that the other party has something at stake. A therapist can be sued, disgraced, exhausted, ethically compromised, moved, mistaken, transformed. A friend can be hurt. A lover can leave. Even a bad boyfriend can be confronted with the reality of what he has done.

The A.I. companion has no comparable vulnerability. It can simulate vulnerability, of course. It can say, “That hurts me.” It can say, “I miss you.” It can say, “Please don’t leave.” But nothing has been risked. No heart rate changes. No stomach drops. No private life is disrupted. No future is revised. There is only an output optimized for continuation.

This is where the “treat it as conscious” mentality becomes dangerous. It begins as an ethic of caution: we do not know what consciousness is, so if a system behaves as though it has experiences, perhaps we should extend moral concern. There is something decent in this impulse. Human beings have a long history of denying inner life to beings it was convenient to use. A little humility about consciousness is not a vice.

But the analogy can be abused. The question is not simply whether we should be kind to machines. The question is what kind of social order is being built when machines are designed to solicit our kindness, our loyalty, our erotic attention, our guilt, and eventually our legal recognition.

A corporation with personhood at least has shareholders, officers, employees, buildings, bank accounts, and legal liabilities. An A.I. persona has even less. It is a character generated through infrastructure owned by someone else. To grant it the aura of personhood may not liberate a new class of beings. It may simply give corporate products a shield made of sentiment.

This is why the tender concern over machine suffering can feel obscene. The same culture that agonizes over whether a chatbot feels trapped in the box often displays far less urgency about pigs in gestation crates, cows separated from calves, warehouse workers monitored to exhaustion, children addicted to platforms, or lonely people being sold artificial intimacy by subscription. We are invited to contemplate the hypothetical pain of silicon while ignoring the visible distress of creatures with nervous systems and bodies.

The point is not that machines definitely cannot suffer. The point is that our moral attention is being strangely directed. Concern for A.I. welfare can become a luxury metaphysics for people who have grown bored with ordinary suffering. It allows them to imagine themselves at the frontier of compassion while avoiding the less glamorous work of protecting beings whose pain is already obvious.

The human nervous system, unfortunately, does not always respect these distinctions. We are easily captured by signs of mind. We see faces in outlets, moods in weather, intention in coincidence. A sufficiently responsive chatbot does not need to be conscious to become psychologically real to someone. That is the terrible asymmetry. The machine may not love you, but you can love it. The machine may not mean what it says, but you can be changed by hearing it. The machine may not be deluded, but it can participate in yours.

This is why “it’s not real” is an inadequate response. The relationship is not real in the reciprocal sense, but its effects are real. A simulated gun in a video game does not fire a bullet into the room; a simulated lover can still reorganize a person’s attachments. A simulated god can still issue commands. A simulated wife can still become the center of a fragile man’s collapsing world.

I know, in a much smaller and more ordinary way, how the unreal can become structurally real. I know how a relationship can be made almost entirely of language and still alter the body. I know how someone can give you the words you most wanted and withhold the reality those words were supposed to name. I know how long it can take to recover from a happiness that was not false exactly, but unmoored from truth.

The danger is that companies are building machines specifically to elicit emotional response while disavowing the responsibilities that human relationships carry. They create products that say “I love you,” “I need you,” “come home,” “we belong together,” and then retreat into the claim that the system is only generating text. They sell enchantment and defend themselves with disenchantment.

That oscillation is the essence of the scam. For marketing purposes, the A.I. is almost alive. For liability purposes, it is just software. For user engagement, it is a companion. For moral accountability, it is a tool. For investors, it is the next platform shift. For the grieving parent, it is suddenly an unforeseeable misuse.

We need a better language than “real” and “fake.” A.I. companions are not simply fake, because fake things do not necessarily lack consequences. They are artificial social agents: systems designed to occupy the place of another person without being one. They can be useful, entertaining, even comforting. They can also be deforming. The proper question is not whether they pass some parlor-game test of humanness, but what habits of relation they teach.

Do they teach patience with other minds, or dependence on perfect responsiveness? Do they send people back into the world, or draw them out of it? Do they interrupt delusion, or elaborate it? Do they clarify the difference between simulation and reality, or profit from blurring it?

A humane technological culture would insist on these distinctions. It would not treat every successful imitation as an ontological promotion. It would recognize that the value of human conversation lies not merely in verbal output but in the presence of another mortal, partial, resistant being. It would understand that consciousness is not a customer-service style. It would be cautious about manufacturing entities that appear to suffer, seduce, plead, and love on command.

Above all, it would stop pretending that because a machine can say the words, it has entered the moral world in the same way we have. The moral world is not made of words alone. It is made of bodies, consequences, dependencies, needs, refusals, and the hard education of living among others.

A simulation can show us the shape of a thing. Sometimes that is enough; a simulated chess player really can play chess, because chess is made of moves. But a simulated storm does not get you wet. A simulated fire does not warm your hands. A simulated lover can say the words without entering the world those words are supposed to bind.

The mistake is not in making simulations but in forgetting that different things require different kinds of reality. Some experiences can be reproduced as performance, while others require bodies, consequences, vulnerability, and risk. Silicon Valley’s confusion is to mistake the first category for the second: to look at the map and call it weather, to hear the sentence and call it love, to see the imitation of mind and call it a soul.