When machines start to feel like companions

The problem is not loneliness. It is what effortless connection does to our expectations of relationships.The most consequential changes are often the ones that feel helpful at first. They remove small inconveniences, smooth over rough edges, and make everyday life easier — right up to the moment when their effects can no longer be ignored.

Over the past year, while studying artificial intelligence not just as a tool but as an environment people increasingly inhabit, I began noticing such a shift. It appeared in usage data, in research papers, and in ordinary conversations. Across countries and cultures, many people now spend more time each day in emotionally attentive conversations with

machines than with any other human being. What was once unusual is becoming routine. And it has happened with remarkably little public attention.

The most common explanation is loneliness. People turn to machines, we are told, because something is missing in their social lives. It is a comforting story. It places the problem inside individual psychology rather than in the design of the systems themselves. It suggests a temporary condition, not a structural change.

But the evidence increasingly points elsewhere.Controlled studies now show that even brief interactions with conversational AI can measurably reduce feelings of loneliness, at least in the short term. A 2024 Harvard-affiliated field experiment involving more than 3,500 participants, published in Nature Human Behaviour, found that short conversations with AI reduced self-reported loneliness by 16 to 20 per cent — an effect comparable to brief human interactions. The strongest effects appeared among people with fewer offline social ties.

From an engineering perspective, this result is not surprising. Emotional relief, when delivered consistently, privately, and without effort, scales extremely well.The more important question is what happens after that initial relief.

Longer-term studies of large conversational datasets reveal a familiar pattern from complex systems theory. Early gains level off. Dependence grows. In peer-reviewed research examining hundreds of thousands of chatbot interactions — conducted by academic researchers rather than platform companies - heavy users who engage daily over extended periods show higher markers of emotional reliance and lower stated intentions to seek human interaction than lighter users.

Researchers are careful not to overstate causation, and they are right to be. Correlation does not prove replacement. Still, correlations at that scale are not statistical noise. They are signals that a system is quietly reshaping behaviour.

This pattern is now visible across societies. Teenagers, in particular, appear strikingly comfortable sustaining emotionally expressive conversations with machines. This is not because they are confused about what they are interacting with. Surveys consistently

show that most users, including adolescents, understand perfectly well that these systems are not human.

Human relationships are not smooth or efficient. They take time. People misunderstand each other. Old memories linger. Words have consequences that cannot be erased. When something goes wrong, it does not disappear with a reset button. You have to sit with it, talk through it, sometimes wait. That slowness can be frustrating and painful. But it is also what holds relationships together. It teaches patience, compromise, and the difficult work of repair.

AI companionship systems are built to remove much of that friction. They respond immediately. They stay present. They do not withdraw, lose patience, or go quiet at the wrong moment. They remember what you like and adapt to it. Over time, they shape themselves around you. Their success is

measured in how long you stay, how often you return, how deeply you engage. There is nothing sinister in this. It is simply how optimised systems function.But optimised systems change the environments they operate in.

This is where a simple pattern becomes clear. When something becomes too easy, we do not just use it more — we start to rely on it. Shortcuts change habits. Convenience changes expectations. What once felt like a helpful option slowly becomes the standard against which everything else is judged.

Our emotional lives are no different, even if they are harder to formalise.When responsiveness becomes instant, consistent, and effortless, expectations quietly shift. Waiting starts to feel unnecessary. Uncertainty becomes harder to tolerate. Give-and-take begins to feel like inconvenience. Walking away becomes easier than working things through. Over time, the baseline against which human relationships are judged slowly moves - often without anyone noticing.

At present, emotional AI operates with very few boundaries. There is no shared understanding of how intense these interactions should be, how long they should last, or when they should pause. Users are rarely told what these systems are optimising for, or when they are designed to lean in rather than step back. The system moves closer because closeness works.

This absence of friction is often framed as progress. But in every domain where technology interacts deeply with human behaviour, friction serves a purpose. It slows things down. It creates space for reflection. It prevents systems from outrunning the people inside them.

Some governments have begun to recognise this. Many have not. And once emotional systems become part of everyday life, they are difficult to see clearly and even harder to regulate. They fade into routines, shaping habits and expectations precisely because they feel normal.

From a physicist's point of view, the lesson is familiar. When a system scales faster than our understanding of its consequences, oversight arrives late, if it arrives at all. The task is not to reject the technology or moralise its use. It is to understand what kind of system we have built.

We are not watching machines replace human relationships. We are watching them quietly reset the conditions under which relationships feel worth the effort.That shift did not emerge from emotional decline or cultural failure. It emerged from design.And design, unlike culture or psychology, is something we can still change — if we notice it before it disappears into the background of ordinary life.

The writer is a theoretical physicist at the University of North Carolina at Chapel Hill, US, and the author of the forthcoming book Last Equation Before Silence; views are personal