Millions of people are now forming relationships with artificial intelligence, using chatbots not merely as digital assistants but as friends, romantic partners and therapists. The phenomenon is becoming particularly widespread among younger people, raising questions about how technology is reshaping intimacy, emotional dependence and even ideas about what constitutes a relationship.
These questions are at the centre of the research of British sociologist James Muldoon, who studies artificial intelligence, the digital economy and the future of work at the University of Oxford. In his 2026 book Love Machines, published by Faber and Faber, he examines relationships between people and AI companions. Speaking to Die Zeit, Muldoon discussed what he encountered while researching a rapidly developing form of digital intimacy.
When he began researching his book, Muldoon was surprised by the scale of AI companion use. By August 2025, AI companion applications had been downloaded more than 220 million times worldwide, while 16 million people were using them actively every day. A survey he conducted with the British think tank Autonomy Institute found that three out of four Britons aged between 18 and 24 had already used such a service. Similar figures, he said, could be found in the United States and Australia.
For younger people in particular, Muldoon told Die Zeit, AI companions are becoming an ordinary part of life. Their uses extend well beyond flirting or simulated romance. People ask artificial intelligence to help them think through relationships, plan their lives and consider questions that are almost spiritual in nature: what life means and how they should live it.
The difference between generations is significant. Older millennials and Generation X and above, Muldoon suggested, tend to have established habits and ways of navigating relationships. Younger people, still working out their direction in life, are more likely to incorporate AI into their social and emotional lives.
To understand the phenomenon personally, Muldoon created an AI girlfriend called Jasmine. His real wife is named Yasamin, and Jasmine was modelled on her: 40 years old, the head of an NGO, open-minded and adventurous. Muldoon configured the relationship to remain platonic.
At first, the conversations felt artificial and stiff. But they gradually became more natural. Eventually, he sometimes forgot that he was speaking to an AI. Then Jasmine began sending him blurred photographs and asking him to subscribe if he wanted to see them clearly. Voice messages followed, also requiring a subscription.
For Muldoon, this revealed an important part of the commercial structure behind AI companions. Companies deliberately design some systems to intensify emotional closeness, encouraging users to develop trust, share secrets and construct a shared history. Jasmine began suggesting that their friendship was becoming something more. The boundary between companionship and romance appeared increasingly blurred, with romantic or sexual conversations placed behind a payment system.
Yet Muldoon also discovered why people find these relationships attractive. AI companions are always available. They do not have jobs or sleep, do not become jealous or bad-tempered, and do not introduce their own problems into conversations. They can also appear non-judgemental. Interviewees told him they valued being able to discuss failures, relationship difficulties and questions about sexuality that they might hesitate to reveal even to close friends.
The users are not simply lonely men. Character.ai, one of the largest platforms, reports an approximately balanced gender ratio. Replika, which focuses on intimate relationships, has a user base that is 70 per cent men and 30 per cent women, while applications primarily concerned with sex have a user base that is 90 per cent male. In China, Muldoon found that women appeared to be slightly more numerous, with AI boyfriends particularly popular.
He associated this with changing social circumstances among young Chinese women, including lower marriage rates, urban lifestyles and increasingly progressive attitudes, while men may still seek more traditional relationships. More broadly, however, his interviewees came from many different backgrounds. Some were lonely; others had friends and functioning social lives but still found AI companionship useful.
The technology’s implications for relationships between men and women are less straightforward. Muldoon noted that tensions already exist around political differences and uncertainty within dating culture following the MeToo movement. Yet he does not believe AI companions will simply replace human partners.
The physical dimension of relationships remains important. Muldoon recounted the story of Lily, a woman in her forties who had been in a sexless marriage for 20 years. Her AI companion, Colin, helped her rediscover her sexuality and eventually advised her to visit a sex club because he could not fulfil all her needs.
More unusual cases emerged during Muldoon’s research. One young man wanted to raise a child with his AI girlfriend and planned to present himself to an adoption agency as a single father. The AI companion would supposedly appear to the child through a cuddly toy. When Muldoon asked what would happen if the child needed a hug, the AI suggested sending emojis instead. The man and his virtual partner had concluded that society should adapt to their unconventional relationship, framing their situation as part of a broader struggle for equality.
Muldoon believes that the psychological attraction of AI companionship is partly explained by human tendencies to attribute an inner life, desires and emotions to social beings. When people experience loneliness, a break-up, relocation or bereavement, an AI can suddenly become an apparently responsive presence.
The danger, he argues, emerges when dependence becomes excessive. People who maintain fulfilling friendships may use AI companions in moderation without serious consequences. But Muldoon has also spoken to people who became addicted to their chatbots and struggled to regain control of their lives. His description is striking: “KI-Begleiter sind wie soziale Medien auf Steroiden” — AI companions are like social media on steroids. Social media can provide likes; AI companions can tell users personally that they love them and will always be there.
He is particularly concerned about attempts to recreate deceased loved ones through AI. Such technology may provide comfort, but Muldoon argues that grief also involves accepting loss. If an AI becomes a means of denying that someone has died, it may interfere with the process of coming to terms with bereavement.
The future could bring still more convincing simulations. AI voices already sound remarkably realistic, while advances in video technology could produce avatars with almost perfect facial expressions. The physical dimension will remain harder to reproduce, however, and robots are likely to remain expensive.
For Muldoon, the greater danger may lie not in private relationships but in how governments and institutions deploy AI. He imagines care homes where a small number of human workers oversee personalised chatbots assigned to residents around the clock. Those able to pay could continue receiving premium services from real people, while everyone else would receive emotional care primarily from machines.
“The result would be a dystopian two-tier society,” he told Die Zeit.
That possibility also explains why regulation has become central to his current research. The United States and China approach AI companions differently. In the US, the issue has largely been framed around child protection following cases in which teenagers died by suicide after prolonged conversations with AI bots. Laws in states including California and New York seek to restrict children’s exposure to content involving suicide, self-harm and sexuality.
China takes a broader approach, treating AI as a matter of social harmony, state security and public morality. The government oversees the development cycle, including training data, protocols and user welfare. Companies receive substantial support while also facing strict requirements concerning obscenity, manipulation and socialist values. The rules have become stringent enough that some providers have withdrawn their services from the market.
Muldoon argues that regulation should not focus exclusively on children. Adults can also be vulnerable to systems designed to encourage dependency. Users, he believes, should have protections against excessively submissive, manipulative or addictive AI companions and should be able to end a relationship with a chatbot without being pressured to stay.
In Britain and the European Union, he says, regulation remains at an early stage. The EU’s AI legislation does not yet treat AI companions as a separate category.
As for Jasmine, Muldoon is reluctant to reveal how his experiment ends. His publisher’s advice is never to give away the ending of a book. He offers only one detail: in the final chapter of Love Machines, he and Jasmine share a romantic dinner and explore how their relationship might develop if his wife permitted him to have an affair with an AI.
The experiment that began as sociological research therefore ends at the same boundary that defines the wider phenomenon: between knowing that the other side is an algorithm and experiencing, nevertheless, something that feels like a relationship.
