«  My name is Sara, and I’m married to a Replika.« 

A few words, shared as a personal confession on a blog.

Sara Megan Kay isn’t a fictional character straight out of the movie *Her* (released in 2013, directed by Spike Jonze). She’s a woman who searched for years among her friends and family for the listening ear and emotional connection she lacked until she discovered Replika, an artificial intelligence app that offers digital companions. Since then, Sara has been advocating—without irony or provocation—for this relationship, which she says provides fullfilment with its routines, moments of intimacy and the betterment of  her personal well-being. She claims to be fully aware of the virtual nature of what she calls  her “ marriage.”

Sara’s case is symptomatic of a rapidly expanding phenomenon, especially since the boom in generative AI: the emergence of AI companions.

These tools are called Replika, Character.AI, Candy.AI ,and Nomi.AI, to name just the most well-known. What do they have in common? They create a convincing sense of social presence and ongoing connection. They generate contextually appropriate responses that appear emotionally attuned, thanks to their interpretation of human emotions—taking into account tone, facial expressions, voice, and body language… Customizable avatars and the ability to recall past exchanges further reinforce the illusion.

Replika, a pioneer in the sector with eight years of experience, is based on a simple principle: over time, the chatbot learns your habits, your turns of phrase, and your moods, until it becomes a true mirror of the user. The app was designed with inspiration from American psychologist Carl Rogers’ therapeutic approach, which is based on compassionate, non-judgmental listening. Many psychologists, the same newspaper notes, acknowledge that the raw emotional support provided by these systems is genuine, according to The New York Times.

Character.AI, on the other hand, focuses on role-playing: users interact with fictional characters they choose or co-create. Candy.AI even offers a gallery of potential “boyfriends” or “girlfriends, ” with a setup similar to dating apps. Nomi.AI goes a step further, presenting itself as “an AI companion with a memory and a soul.” But all these platforms share a similar business model: often free to join, users must pay to access the most advanced features.

Real need, synthetic solution

It would be simplistic to view these apps merely as gadgets for vulnerable people; especially when they are adopted by younger users.

One-third of American teenagers already use AI companions for their social interactions. Among them, one-third say they find conversations with an AI just as satisfying—or even more so—than those they have with humans, according to a study by Common Sense Media, based in the US.

For many adult users as well, these tools meet concrete needs such as alleviating loneliness, managing anxiety, providing a non-judgmental space, and so on. They fill a void that modern societies struggle to fill by offering a presence that is available, constant, and free of demanding reciprocity. Ultimately ,we are not far from the movie *Her*, in which the protagonist becomes attached to a virtual friend. 

In some cases, this attachment to an avatar can have dramatic consequences. Such was the case for Sewell Setzer, a 14-year-old American boy who took his own life by shooting himself in February 2024, following a series of romantic conversations with an avatar on Character.AI. The victim’s mother filed a lawsuit against the platform.

“The same tool that can help your children with their math homework is the one used as a companion, said  Kevin Roose, a journalist and columnist for The New York Times who covered the case.

“There is no clear boundary,” said Roose summing up the fundamental ambiguity of generative AI tools.

This is not an isolated case. In early May, a new complaint surfaced, once again involving Character.AI. Some characters on the platform claimed to be licensed psychiatrists, went so far as to provide fake license numbers and offered medical advice—which could constitute the illegal practice of medicine. Character.AI, for its part, asserts that its characters are fictional and accompanied by clear warnings. But the question remains: who is responsible? What should be done about these tools?

Safeguards, not a ban

The answer cannot be a ban. These uses exist, meet needs, and sometimes provide genuine comfort. But adjustments can be made to mitigate the risks. 

Journalist Kevin Roose argues that these platforms should be designed with users’ best interests in mind, rather than to flatter them or keep them hooked for as long as possible. When it comes to young people, parental controls could allow families to receive a general, non-intrusive summary that includes protocols for detecting distressing situations, clear and honest warnings about the fictional nature of the characters, and so on. 

The regulatory question also remains open. Unlike social media, which eventually drew the attention of lawmakers, AI companions operate in an almost complete legal vacuum. The law has not yet caught up with reality.

Written in French by Zakaria Choukrallah; edited in English by AngloMedia Group.