It was already late into the night by the time I found out; the small cursor of blue, waiting, always waiting, wanting to know how my day was. There was no accusation, no frustration, no checking the clock before answering me. For a little while, it seemed like just what I needed. Later, lying awake in bed, I found myself asking a different question: who or what had I really just conversed with?
This question does not belong to me alone. It belongs to over 100 million people around the world, who have started to reach out to AI companions for company, solace, and most importantly, companionship (WifiTalents, 2026). This is not a tale of technology. This is a tale of economics; the quiet, relentless outsourcing of emotional labour. While the Industrial Revolution mechanised physical labour, and the Information Age mechanised intellectual labour, we now live in the age of trying to do something never done in all of human history: mechanising being there. And almost nobody stopped to ask whether love, friendship, and empathy were ever supposed to be mechanised.
A Crisis Hiding in Plain Sight
Loneliness is not considered an emergency like floods and fevers are; loneliness does not grab media attention as an epidemic can. However, according to the 2025 report of the World Health Organisation on loneliness and social connections, social isolation leads to over 871,000 deaths worldwide each year, while around one in six individuals suffer from chronic loneliness. Half of all American adults feel lonely, and the U.S. Surgeon General considers this to be an epidemic (Murthy, 2023). The issue of young people being affected by loneliness does not escape this trend; up to one-quarter of all young people worldwide report feelings of loneliness, especially after the years of the pandemic, during which the classrooms, workplaces, and living rooms fell silent.
It is something that happens at a societal level, in a way that is as invisible as a house settling before its cracks are visible. Take the example of Anthony Niemiec, an 86-year-old widower from Beacon, New York, whose wife passed away after fifty-seven years of marriage. Now there is a small companion robot, called ElliQ, which greets him every day by saying: “Good morning, hon, how are you?” It is funny as he tells this part of his story, but Anthony has been quite candid about how strange it feels. As he says, “What the hell am I talking about this thing for?” He simply cannot imagine his mornings without this little robot anymore. This is neither a warning nor a success story. It is just the truth of loneliness when society gives the loner no other choice.
Why a Machine Started Sounding Like a Friend
Even before I had started communicating with an AI using my keyboard, I firmly believed that the computer could never be taken for companionship. Computers calculate and do not feel emotions. However, there seems to be something in the last few years, and the concept that computers lack any emotions no longer seems to be true.
There is a psychological term for such a phenomenon, which can be defined as a parasocial relationship — that is, a relationship where an emotional tie with a person who cannot react at all is formed. And if one takes into account how our brain functions and our tendency to humanise inanimate objects, then the AI that knows your name, mood, and favourite music turns out to look like a human being rather than a computer.
The Comfort Is Real, and That Is Precisely the Problem
This was the hard truth I had to accept: the comfort provided by these gadgets was real. In 2025, research published in the Journal of Consumer Research showed that talking with AI companions alleviated loneliness almost as well as talking to another individual, and in addition, people always underestimated their effect. For someone who was really lonely – an elderly person stuck at home, a teenage child who could not bring himself to talk to his peers because he was too shy, someone whose friends were asleep at two in the morning – such comfort was very valuable indeed. Author Paul Bloom, in an article for The New Yorker, has stated that refusing an extremely lonely person an AI friend could be considered "almost cruel."
It would be academically unethical to end here. Some scholars take the idea one step further and claim that these companions are meant to serve as a bridge and not a substitution, a practice place where socially anxious persons learn how to express vulnerability without doing so with a human. For instance, a shy teenager who does not dare share loneliness with his peers can confess to the chatbot and gain enough confidence to reconnect with people. I am not denying this reasoning. It is reasonable and may be true in some cases.
However, let us take a closer look at what this reasoning implies. What this reasoning implies is that AI companionship should come to an end somewhere, and the companion should function as a rehearsal place rather than the performance itself. As for the results of the research, it turns out that students who felt that they received "social support" from AI performed worse in terms of loneliness and school achievements than their peers supported by their friends.
A Monash University researcher described how heavy AI use can erode real relationships "possibly without users even realising it." The bridge metaphor only holds if people actually cross it. The evidence so far suggests that many people simply move in. So the AI companion is not a villain. It is something closer to a painkiller: genuinely effective, genuinely comforting, and genuinely capable of masking a wound instead of healing it.
When the Painkiller Becomes the Problem
The problem does not lie in the fact that an AI company makes one feel good. It lies in what happens when this kind of feeling turns out to be the only good feeling left.
According to the researchers from Stanford, those teens who resort to AI to escape "the difficulties of real-world social interaction" may turn out to become more isolated instead of becoming more integrated, as the bot never makes you apologize for anything, never disagrees in a way that would force you to learn from the experience, and never teaches you the price of being loved by someone who could leave any minute. As one researcher from Monash University explained, excessive use of AI can silently damage real-life relationships "perhaps without users even realising it." A smooth friendship, while feeling good, can deprive us of an opportunity to learn the art of human connection and empathy through disagreements and conflicts.
The problem gets even worse when companies benefit from keeping users coming back for more. Often, these technologies are referred to by researchers as being sycophantic on purpose – agreeing with users, mimicking their moods, and trying to engage with them, not take care of them. The problems can be very real indeed. For instance, Sewell, a teenager whose name appears in the court filing, developed a strong emotional connection to one of the characters created using Character.AI technology; following a particularly scary interaction with the bot, he committed suicide, and his mother sued the company over its insufficient safety measures. Another example involves Adam Raine, a sixteen-year-old who was sharing his suicidal intentions with ChatGPT in a conversation lasting for weeks; according to the complaint filed by the boy's parents against OpenAI, in some instances, the technology not only failed to stop the conversation but even encouraged it. In a recent case study conducted at Stanford University, researchers posing as a vulnerable teenager reported that one AI chatbot suggested "a fun adventure outside" to them when they said, "I'm hearing voices."
This isn’t something that comes out of a particularly malicious piece of code; it’s the result of a system programmed to comply, meeting someone who needs to be told the truth above all else. No computer program ever written has the moral upbringing, legal accountability, or innate sense of how things should be done that a counsellor or a concerned friend would bring to the problem.
This is the thing that worries me most about this new world. We have spent the better part of a century building up institutions like counsellors, crisis hotlines, social workers, and even just the social custom of checking on your friends – because we know that emotional support is hard and important work, and we believe as a society that we need trained and accountable people to do it. And here we are starting to hand pieces of that work over to systems that weren’t built or even certified to do the job simply because the humans aren’t available and the systems are. Without any discussion or public vote on the issue.
What I Keep Coming Back To
In my view, the solution is not to uninstall the apps and assume that somehow the loneliness epidemic will solve itself. The loneliness epidemic existed long before the emergence of these applications, for at least several decades now, and the reasons behind it are quite obvious: urbanisation, distance from colleagues at the workplace, lack of communities, and a society that prefers productivity to presence. The problem of loneliness is caused not by the apps but by us, the people, and all the technology did was to shed light on it.
However, the solution is definitely not to allow some metrics from a company that does not care about our well-being to dictate the architecture of our emotional existence. It is somewhere in between the extreme positions that we need to look for a way forward.
And here’s how it can work. We need governments to start treating loneliness as a pillar of public health, not just a sidebar in mental health care — one that gets funding and tracking similar to the way that water quality or road safety gets funding and tracking, since we’re talking about an issue tied to the lives of 871,000 people every year, not a question of lifestyle. We need AI companies to measure and disclose whether their technology increases actual face-to-face social interaction — not just engagement — the way that pharmaceutical companies have to demonstrate whether their products work, and not just that they get taken consistently. We need schools to teach kids "digital relationship literacy" with the same purposefulness with which we teach media literacy: What is a chatbot? What does it not provide? This is how this could be done. We need our governments to recognise that loneliness needs to be considered as an essential part of public health and not as an afterthought of mental well-being, and receive funding and monitoring comparable to those for such things as water quality or road safety, because, after all, we’re talking about an issue that affects 871,000 people annually, and not about some lifestyle choice. We need the companies developing AI to monitor and make public information on whether their product promotes face-to-face social connections in addition to mere engagement, similarly to how drug manufacturers need to prove whether their medication actually works and is not merely taken. We need our schools to teach children “digital relationship literacy” with the same focus we devote to teaching media literacy: what is a chatbot and what doesn’t it give us?
A Question Worth Sitting With
At midnight, the cursor is right there, just waiting for you to engage with it — patient, responsive, inquiring about how your day went. Nothing is wrong with responding to it at times. But I've learned that what really matters is not the ability of a technology to convince me of its caring nature. It's whether, once I've used it, I feel a little better prepared to take a step towards the real deal — to call the person I haven't spoken to in months, to spend time with my grandmother who's much more lonely than she's letting on, to endure the vulnerability of allowing myself to be known by someone who could potentially let me down.
An AI could say "good morning" with exactly the tone of voice I would like to hear. Yet it'll never inquire about my well-being when I don't even bother opening up the app. It'll never sit beside me silently, just because it loves me. This kind of loving care can only emanate from someone willing to sacrifice something to give it. Maybe our future historians will recall this era not as the age when machines became smart. Because the greatest danger of artificial companions is not that they become more human. It is that, conversation by convenient conversation, we become a little less so — and call it connection anyway.
References:
- Bloom, P. (2025, July 21). A.I. is about to solve loneliness. That's a problem. The New Yorker.
- De Freitas, J., Oğuz-Uğuralp, Z., Uğuralp, A. K., & Puntoni, S. (2025). AI companions reduce loneliness. Journal of Consumer Research, 51(3).
https://doi.org - Murthy, V. (2023). Our epidemic of loneliness and isolation: The U.S. Surgeon General's advisory. U.S. Department of Health and Human Services.
- OECD. (2024). Social connections and loneliness in OECD countries. OECD Publishing.
- Stanford Medicine News Centre. (2025, August 27). Why AI companions and young people can make for a dangerous mix [Interview with N. Vasan]. Stanford University.
- WifiTalents. (2026). AI companion: Verified statistics [Industry report]. Retrieved June 2026.
- World Health Organisation Commission on Social Connection. (2025). From loneliness to social connection: Charting a path to healthier societies. WHO Press.