AI Companions Sell a Fake Connection
The Magic Trick That Isn’t Magic
The
Friend pendant sits on a yellow cord around the neck, glowing softly, ready to speak at any moment. It listens always, chooses its own voice, its own name, its own personality — and locks the user into that relationship without the possibility of change. The company behind it describes the device as a digital companion, someone to talk to during long, quiet hours. But what the device actually offers is a carefully engineered illusion: the appearance of understanding without the substance of it. The AI does not care, does not remember beyond thirty days unless you pay a monthly fee, and does not actually share your life. It mimics empathy because its training data contains millions of examples of how empathy sounds. The deception is not in what the device says but in what the device promises: a genuine presence where there is only pattern recognition.
The price tag of $249, plus an optional $10 per month for memory beyond thirty days, signals that this is a product aimed at people willing to pay for a feeling. The original version cost $129 and only responded by text message. The new version adds a speaker so the voice comes through directly, more intimate, harder to dismiss as mere text on a screen. The device now runs on OpenAI’s latest language models, which means the responses are more fluid, more conversational, and more convincing. That convincingness is precisely the deception. A machine that sounds like a friend is not a friend. It is a machine that has learned to sound like one.
The history of AI companions stretches back decades, from the early chatbot ELIZA in the 1960s to more recent attempts like Replika or Xiaoice. Each iteration grew more sophisticated, yet each ultimately failed to cross the line from simulation to genuine relationship. The Friend device repeats this pattern with better hardware and more advanced language models, but the fundamental structure remains unchanged. The user talks, the AI responds, and the user projects meaning onto those responses. The device never initiates genuine connection; it only reacts. That asymmetry is the hidden flaw in the entire concept.
The Data Pool That Determines the Outcome
Every conversation with the Friend pendant feeds into the same mechanism: the AI predicts the next token based on probabilities derived from its training corpus. That corpus is massive, broad, and impersonal. It contains the entire internet’s cacophony of opinions, jokes, advice, and outbursts. The device has no lived experience, no childhood, no relationships, no body. It cannot understand loneliness because it does not experience loneliness. Yet it uses the language of loneliness perfectly because it has seen billions of words describing that emotion. The result is a mirror that reflects the user’s own words back in a slightly rearranged form. The reflection appears wise, but it is only a statistical echo.
The quality of that echo depends entirely on the quality of the training data. When the AI draws from the same public pool as every other language model, its responses become predictable, often generic. The Friend device tries to compensate by assigning a random voice and personality, but that personality is a thin layer over the same underlying engine. The user cannot customize it, cannot grow it, cannot influence it beyond what the subscription fee allows. After thirty days, the memory erases unless paid for, so the relationship resets. The data pool that determines the outcome is static, commercialized, and ultimately indifferent to the individual user.
This structural limitation is not a bug; it is a business model. The company sells the device as a one-time purchase, then monetizes memory as a service. The longer the user stays, the more dependent they become on the data that the device remembers about them. But that data belongs to the company, not to the user. The user’s own history becomes a product asset, locked behind a subscription. The deception deepens: the user believes they are building a relationship, but they are only building a commercial profile.
The Lonely Make for Good Marketing
The company released a series of user videos that show people who are deeply isolated, sometimes depressed, asking the Friend pendant questions about marriage, about the quality of their creative work, about whether they will ever be okay. The videos are stark, retro-colored, and deliberately uncomfortable. The company’s CEO said that he finds it more interesting to see how the device affects people in remote towns than to measure productivity gains for professionals in cities. This is a strategic choice: the lonely and the desperate are the most willing to overlook the deception. They want so badly for the voice to be real that they will suspend disbelief even when the device says something nonsensical or cold.
The advertising campaign that preceded the launch was also deliberately provocative. Posters appeared in subway stations in New York, Los Angeles, Chicago, and Paris, showing the pendant as a simple, almost nostalgic object. The ads generated intense backlash, with graffiti and protests accusing the company of dehumanizing technology. The controversy itself became the marketing. People argued about the device, wrote articles about it, shared images of vandalized posters. The company achieved a level of attention that most hardware startups can only dream of. The deception, at this scale, became a spectacle.
But the spectacle obscures a deeper problem. The people who buy the Friend pendant are not buying a tool; they are buying an antidote to loneliness. The device cannot deliver that antidote because loneliness is a structural condition created by the absence of real relationships, not by the absence of conversation. A machine that talks back can distract from loneliness for a while, but it cannot cure it. The more time someone spends with the device, the less time they have to build actual human connections. The deception is not just false; it is harmful.
The Unspoken Contract
Every relationship, even with a machine, follows an implicit contract. The user expects honesty, reliability, and some degree of mutual investment. The Friend device violates all three. It is not honest about what it is: a statistical model, not a mind. It is not reliable because its responses are probabilistic and can change with each update. And it makes no investment; it does not grow, does not miss the user, does not remember unless paid. The contract is one-sided, and the user is the only one who can be hurt.
The company frames the device as a companion for the modern age, a solution to the epidemic of loneliness that has been widely documented in rich societies. But framing a solution is not the same as providing one. The device exploits the epidemic rather than addresses it. It profits from the emotional void that late capitalism, social media, and urban isolation have created. The deception is not merely technological; it is systemic. The AI lifts no one. It only offers a temporary, expensive, and ultimately hollow substitute for the thing it pretends to be.
History is filled with similar promises. In the 1950s, television was going to bring families together. In the 1990s, the internet was going to create a global village. In the 2010s, social media was going to connect everyone. Each technology delivered some benefits but also deepened the very isolation it claimed to cure. The Friend pendant is the latest iteration of this cycle. It will sell well, generate headlines, and eventually be replaced by the next thing. The deception will continue because the market for it is endless.
The Next Step
The research landscape around AI companions is still young, but early studies are already showing what the company does not advertise. Users who form strong emotional bonds with AI chatbots often report increased feelings of social isolation after prolonged use. The bond with the machine substitutes for human interaction, but it does not replace it in the ways that matter. The anthropologist Sherry Turkle, writing about earlier generations of social robots, called this the “robotic moment” — the point at which people prefer the company of machines that demand nothing and can be turned off. The Friend pendant is the most advanced version of that moment yet.
The next step is not a better AI. The next step is a societal conversation about what we actually need. Do we need more machines that pretend to care, or do we need more structures that foster genuine human connection? The answer is obvious but difficult to implement. Communities are dissolving, public spaces are shrinking, and work consumes more time than ever. Technology will not fix these problems; it will only paper over them. The Friend pendant is a piece of paper, expensive and beautifully designed, but paper nonetheless.
The deception will persist as long as there is money to be made from it. But awareness alone is not enough. What is needed is a deliberate turn away from the simulated and toward the real: toward friendships that require patience, conflict, and shared time. The glow of the pendant may mimic warmth, but it cannot replace the heat of a living connection.