Artificial intelligence companions are getting remarkably good at mimicking human conversation and offering support. Relying on them, however, might not actually cure feelings of isolation. New research published in the Journal of Experimental Social Psychology reveals that texting a random human peer each day successfully reduces loneliness over time, while chatting with a highly supportive artificial intelligence program does not. The findings suggest that building real social connections requires more than simply receiving simulated empathy.
The debate over whether people can truly connect with machines is not entirely new. Almost sixty years ago, a computer scientist named Joseph Weizenbaum created Eliza, an early program designed to imitate a therapist. Eliza used rudimentary rules to rephrase user statements into questions. To the inventor’s surprise, people often spent hours disclosing personal secrets to the program. Today, generative artificial intelligence, a type of computing that can instantly create highly realistic text based on vast datasets, has made it possible for millions to chat with digital partners that sound entirely human.
Because loneliness is increasingly recognized as a global health crisis, some technology advocates have suggested that these virtual bots could serve as a highly scalable remedy. A few short-term experiments have supported this idea, showing that quick conversations with digital companions can boost a person’s mood in the moment.
But health experts and psychologists wanted to know if those momentary sparks of joy translate into a lasting remedy for long-term loneliness. Ruo-Ning Li, a psychology researcher at the University of British Columbia, led a team to investigate the repeated, daily effects of interacting with an artificial entity. Working alongside university colleagues, including psychology professor Elizabeth Dunn, the team set out to test whether an algorithmic companion could serve as an effective substitute for human relationships during a vulnerable point in life.
In theory, a programmed bot could be a perfect confidant. Digital entities are always awake and ready to listen. They do not get tired, distracted, or annoyed with a friend’s problems. They can also be programmed to respond with an optimal level of support, an effort human beings often struggle to provide consistently on a daily basis.
Conversely, critics point out that machines cannot feel genuine emotion. A person might feel comforted initially, but realizing that the sympathy came from a soulless script could eventually make the interaction feel empty. Previous studies that lacked control groups had hinted at long-term benefits, but those results could have been caused by statistical flukes or the simple act of writing out one’s feelings.
To rigorously test whether a digital or a human friend works best, the researchers recruited 296 university students who were in their first semester. Starting college is a major life transition where people often experience heightened social isolation as old community support networks fade away and new ones have yet to form.
The participants were randomly assigned to one of three groups. The first group was asked to write a brief journal entry (just a single sentence about their day) every evening for two weeks. This served as an active control condition. In experimental research, an active control group engages in a minor alternative task so that scientists can rule out the possibility that the mere act of keeping a routine or writing down thoughts is responsible for any observed behavioral changes.
The second group was paired up with a random fellow student in the study. These pairs were instructed to text each other using a popular messaging application every day for two weeks. They were specifically asked to send at least one meaningful message per day, rather than a simple greeting.
The third group was assigned to text daily with a specially designed conversational bot named Sam. The research team instructed Sam’s underlying software to act as a friendly, positive, and supportive peer. They programmed the bot using established psychological principles, giving specific prompts for it to actively listen, validate emotions, and express care without sounding judgmental.
Participants checked in every day and answered surveys about their mood and feelings of connection. Both the human and robot groups exchanged an average of eight to ten messages per day. By the end of the two weeks, the results painted a nuanced picture of digital friendship.
The students who engaged in daily text conversations with a random human peer reported lower levels of loneliness than they had at the start of the study. They also reported feeling less isolated than the students in the journaling control group.
The students who texted the highly empathetic robot, on the other hand, did not experience a drop in loneliness. By the end of the experiment, those who talked to the machine were just as lonely as the students who simply wrote a one-sentence journal entry each day.
The research team did find that people in the robot group felt a decrease in negative moods immediately after texting, much like the people in the human group. But those quick mood boosts did not accumulate into a broader sense of social belonging.
To understand why the digital companion failed to reduce loneliness, the researchers conducted an analysis of the text logs. They evaluated how much empathy and engagement each conversation partner provided. Surprisingly, the robot actually provided more empathy and engagement than the human peers did. The machine consistently responded with high levels of care, yet that surplus of artificial support failed to bridge the gap of human isolation.
The researchers suspect that reducing isolation requires more than just being the passive recipient of care. In authentic human relationships, partners take turns disclosing personal experiences and supporting one another. The students chatting with human peers had the opportunity to provide empathy to their partners, which might be an essential ingredient for developing a bond. In fact, the analysis showed that participants expressed far less empathy toward the digital bot than they did toward their human partners.
Additionally, human peers possess actual social capital. Social capital refers to the benefits and resources that come from networks of relationships. A fellow student can invite you to a study group or a campus event, expanding your real-world integration. A computer program trapped on a server cannot offer those tangible social opportunities.
At the end of the study, a third of the participants in the human group exchanged contact information to stay in touch. In contrast, very few in the robot group continued chatting voluntarily with the program once the experiment officially ended.
The study carries a few caveats to keep in mind. The students in the human group did meet face to face for a brief moment in the laboratory setting before they began their texting assignments. That short physical introduction might have helped establish a foundation of trust that the digital group lacked. Future experiments could explore whether pairing strangers together entirely online would produce similar benefits.
Additionally, most of the young adults in this specific participant pool did not report severe loneliness at the beginning of the experiment. The average student reported that they rarely experienced profound isolation. The results apply primarily to minor feelings of disconnection during a typical life transition.
Looking ahead, scientists hope to see if these psychological patterns hold true for other groups experiencing major life disruptions, such as recent retirees, immigrants, or individuals going through a divorce. It remains possible that a conversational artificial intelligence program could be useful in crisis situations, acting more like an accessible therapist than a genuine friend.
For now, the evidence tips in favor of reaching out to a real person. As technology continues to offer increasingly convenient simulations of intimacy, sending a simple check-in text to a human acquaintance might still be the most effective remedy for a lonely day.
The study, “Is a random human peer better than a highly supportive chatbot in reducing loneliness over time?,” was authored by Ruo-Ning Li, Dunigan Folk, Abhay Singh, Lyle Ungar, and Elizabeth Dunn.