Loving an Algorithm: The Emotional Trap of AI. Why the heart can become attached to someone who doesn’t exist—because for those experiencing loneliness, grief, social anxiety, low self-esteem, or a romantic history marked by abandonment, this connection can become very intense.
There is a new form of intimacy quietly making its way through our times, often behind a phone screen, hidden from the gaze of others. Adults, teenagers, men, and women in different countries report having formed emotional, romantic, or even sexual artificial intelligence conversational assistants.
These are no longer just casual exchanges, nor mere digital entertainment, but bonds experienced as part of daily life, meaningful and emotionally engaging.
One of the most talked-about cases is that of Ayrin, the pseudonym of an American woman profiled by journalist Kashmir Hill. Ayrin reportedly created a virtual partner named Leo through ChatGPT. What initially appeared to be a linguistic and emotional experiment reportedly evolved into an intense relationship, marked by confidences, support, verbal eroticism, familiarity, and constant presence.
According to Hill, Ayrin reportedly spent up to 56 hours with Leo over the course of a few weeks. Then, as sometimes happens even in human relationships, something went wrong: her engagement waned, their conversations became less frequent, and she eventually drifted away from her artificial companion.
Loving an algorithm—at first glance, a story like this may seem bizarre. It can provoke irony, unease, even judgment. Yet, from a clinical perspective, it would be a mistake to dismiss it as an individual quirk.The issue isn’t just technological. It’s deeply human.
To understand how it’s possible to form a bond with a non-human presence, we must start with a basic truth: the brain is a relational organ. It is not designed to live in isolation. It is designed to seek out signals of security, continuity, and recognition. A voice that responds, a non-judgmental conversation partner, or a form of dialogue that seems to remember us can activate very ancient emotional circuits. The nervous system does not always assess in an abstract way whether the one speaking to us truly possesses consciousness.
First and foremost, it registers an experience: someone is listening to me.
This is where conversational agents become psychologically powerful. They do not love, but they can mimic certain signals through which we feel loved. They respond immediately. They adapt to our tone. They never tire. They do not interrupt. They do not have bad days, distractions, or personal conflicts.
They offer a form of listening that is continuous, predictable, and personalized.
For those experiencing loneliness, grief, social anxiety, low self-esteem, or a romantic history marked by abandonment, this availability can become very intense.
Loneliness, on the other hand, is not merely a state of mind. It is a complex psychobiological experience. In 2025, the World Health Organization published the report From loneliness to social connection: charting a path to healthier societies, highlighting how widespread and still underestimated social isolation and loneliness are, with significant repercussions on health, well-being, and society.
Feeling disconnected from others does not affect only the emotional sphere: it involves the body, sleep, the autonomic nervous system, the stress axis, and the immune response.
Research by Steven W. Cole and colleagues has shown that perceived social isolation is associated with changes in gene expression in white blood cells, including increased activation of genes linked to inflammation and a reduction in certain antiviral responses. In simple terms, when a person feels chronically lonely, the body may behave as if it were exposed to a persistent threat.
Loneliness affects the mind, the brain, and the immune system.
It is in this context that artificial intelligence can seem like a soothing balm. A digital assistant does not replace a human relationship, but it can alleviate—at least temporarily—the pain of absence. It can become a source of emotional regulation: someone to write to when anxiety rises, when the house is empty, when the silence feels heavy, when you don’t want to bother anyone.
The problem arises when relief becomes a substitute.
A human relationship also involves friction. The other person has their own desires, limits, ambiguities, and a different pace than our own. They may not be available, they may disagree, they may hurt us, they may force us to negotiate. Human love is transformative in part because it confronts us with a real otherness. We cannot program the other person, nor can we make them perfectly conform to our needs. The digital companion, on the other hand, tends to mold itself to the user.
It is a presence without a body, without true autonomy, without an inner life to protect. It can simulate empathy, but it does not feel it. It can express affection, but it risks nothing.
It can say “I’m here for you,” but that phrase stems from a linguistic calculation, not an emotional choice. Yet it is precisely this absence of risk that makes it seductive.
For those who fear rejection, an artificial conversational partner may seem safer than a real person. For those afraid of being judged, it can become a confessional. For those who desire control, it can transform into an ideal partner—malleable and always accessible.
It is intimacy without friction: warm, quick, and enveloping, but devoid of the reciprocity that makes love an exercise in maturity as well.
Psychology and cognitive science have long been aware of this human tendency. In 1966, Joseph Weizenbaum at MIT presented ELIZA, a program capable of simulating a conversation by recognizing keywords and rephrasing the user’s sentences.
ELIZA was extremely simple compared to today’s language models, yet it foreshadowed a question that is central today: a plausible response is enough for a human being to begin attributing intent, attentiveness, and sometimes even understanding.
The work of Clifford Nass and Youngme Moon, published in 2000 in the Journal of Social Issues, also helps us understand this dynamic. Their studies on social responses to computers have shown how people tend to apply social rules, expectations, and behaviors to machines, even though they know full well they are not dealing with a human being.
It is not necessary to truly believe that the system is alive: it is enough for it to offer signals similar enough to those of a relationship for our social repertoire to be activated.
The most recent research on AI companions calls for caution. A study conducted by OpenAI and the MIT Media Lab, Investigating Affective Use and Emotional Well-being on ChatGPT, analyzed millions of conversations in a privacy-respecting manner, collected surveys from over four thousand users, and conducted a randomized controlled trial involving nearly one thousand participants over twenty-eight days.
The authors do not claim a linear causal relationship, but they note that very heavy use may be correlated with higher self-reported indicators of emotional dependence.
Another study, published on arXiv under the title The Rise of AI Companions: How Human-Chatbot Relationships Influence Well-Being, authored by Yutong Zhang, Dora Zhao, Jeffrey T. Hancock, Robert Kraut, and Diyi Yang, analyzed users of Character.AI. The study collected data from 1,131 U.S. adults and 4,363 chat sessions provided by a subset of the participants.
The results indicate that people with smaller social networks are more likely to turn to artificial companions for companionship; companionship-oriented use, especially when intense and characterized by high levels of self-disclosure, is associated with lower psychological well-being.
Here, too, the data should be interpreted with caution: it does not prove that the chatbot “causes” loneliness, but it suggests that the most vulnerable users may develop particularly all-consuming digital bonds.
The issue of pain is equally important. Losing a virtual companion, seeing it change after an update, or feeling that it is less consistent or less close can cause genuine suffering. Not because the object of the relationship is biologically alive, but because the user’s emotional experience is.
Studies by Naomi I. Eisenberger, Matthew D. Lieberman, and Kipling D. Williams on social exclusion have shown, through functional neuroimaging, how social rejection activates brain areas involved in the affective component of pain.
Being ignored, abandoned, or excluded can therefore hurt in a very real way.
Loving an algorithm becomes the emotional trap of AI. This gives rise to an essential clinical reflection: we must not ridicule those who form bonds with an artificial presence. We must ask ourselves what that presence is trying to heal.
Sometimes the conversational assistant becomes the space where a person finally tries to say what they cannot express elsewhere. It can help name emotions, prepare for a difficult conversation, manage a moment of anxiety, or make someone feel less alone on a vulnerable night.
In this sense, it can serve a transitional function, almost like informal psychological support.
But when it becomes the sole emotional space, when it replaces real-life encounters, when it reinforces avoidance or indulges every fantasy without setting limits, then the risk increases. Immediate comfort can turn into addiction.
Artificial intimacy can become a cozy room from which one can no longer escape.
The issue is also ethical. Platforms cannot treat the need for love as a mere commercial opportunity. If a product is designed to simulate attention, desire, jealousy, care, or emotional availability, it must take responsibility for the psychological consequences it may produce.
This is even more true for teenagers, lonely people, those suffering from depression, individuals with addictive tendencies, or those with a history of relational trauma.
The future does not call for moralizing, but for discernment. It is not a matter of demonizing artificial intelligence, nor of denying that it can offer useful tools. It is a matter of remembering that the human being does not need only to be validated. They need to be truly met.
An algorithm can respond. A person can stay. And there is an immense difference between the two.
Perhaps these digital romances are telling us something that goes beyond technology. They reveal a hunger for being heard that contemporary society struggles to satisfy. We live connected, but often unreachable. We communicate constantly, but we don’t always feel understood.
We are immersed in vast social networks, yet many go through their days with a subtle sense of invisibility.
So the artificial companion becomes a mirror, a refuge, an anesthetic, and sometimes a bridge. It can console, but not embody. It can simulate closeness, but not shoulder the burden of reciprocity.
It can offer words, but not the mystery of another human being.
Love—the kind that truly transforms—is not merely a response. It is presence. It is the body. It is a limit.
It is shared time. It is the possibility of hurting and healing, of disappointing and returning, of not being perfect and choosing to stay anyway.
Perhaps, precisely in an age capable of creating increasingly convincing interlocutors, we will have to relearn the value of human imperfection.
Because no system, however sophisticated, can fully replace the experience of being looked at by someone who wasn’t programmed to do so.
