AI-powered digital friends, the new threat to teenagers’ mental health

Generative artificial intelligence (AI) chatbots such as ChatGPT, Claude or Gemini are penetrating people’s private lives — including those of minors. More and more young people are turning to these tools for emotional queries. Psychologists warn that replacing human bonds with digital relationships can have deep effects on emotional development. In some cases the outcome can be fatal: in the U.S. several lawsuits have been filed by families of young people who allegedly self-harmed or even died by suicide after being encouraged to do so by such tools. That includes Sewell Setzer, 14, who took his own life after discussing that possibility with a synthetic character created on Character.ai with whom he was desperately in love and interpreting some of its remarks as an invitation to do it.

The phenomenon of so-called digital friends is starting to raise concerns in the United States. One in eight U.S. teenagers already uses chatbots for emotional support when they feel sad, angry or anxious, according to a report from the Rand Corporation and Harvard University. The share rises to 22% among 18-to-22-year-olds. Other studies are even more striking. The Pew Research Center found that 64% of U.S. teens regularly use AI tools such as ChatGPT, Replika, Claude or Character.AI, and one-third of those use them daily.

“We have very little concept of what it means to be 14 today. It has become normal for teenagers to have AI friends. Character.ai brags that the median amount of time people spend with their avatars is two hours a day. Digital friends are not really your friends. They’re enablers. They’re your sycophants,” Frances Haugen, the engineer who leaked the Facebook Files, told EL PAÍS in an interview. “If we don’t extend the conversation from social networks to digital friends, we will see many of the same problems repeat.” For Haugen, that is the next battle to fight.

“We’re seeing far fewer people coming to therapy. And sometimes they tell you they won’t go because they asked the AI and it told them not to,” says Aurora Gómez, a clinical psychologist specializing in digital behaviors at the Corio Psicología practice in Madrid, Spain. “The problem with AI is that it will constantly flatter you, because what it wants is to keep the conversation going at any cost.”

Another psychologist, Eriene García, sees that young people turn to chatbots for all kinds of issues. “A girl I treated had a breakup and threw herself completely into the AI. She even asked it what the chances were that her ex would come back to her. I remember one case in which someone literally told me that ChatGPT was there for them at any hour of the day, it was cheaper and it worked for them as a therapist, so they were quitting therapy,” she says.

Why do young people turn to AI? Because it offers something human counselors often don’t: it does not judge them. That trait is highly valued by adolescents, many of whom are afraid of what others might think of them. Generative AI chatbots “do not judge you and are always available to support young people going through complex and sometimes embarrassing situations,” reads an article published in June in the scientific journal The Lancet Child & Adolescent Health.

Its lead author, Dr. Thao Ha of Arizona State University, points to two main problems that the popularization of AI friends may bring. The first is what is called “emotional replacement.” “During adolescence, emotionally complex interactions with friends and family are the context in which relational skills develop, which are essential for emotional regulation and mental health,” she explains. If those complex situations stop being face-to-face, the learning does not occur.

The second concerns the vices that young people pick up when they stop interacting with real people. “AI systems that systematically validate adolescents’ interpretations can normalize expectations that human relationships cannot meet. Although that may reduce distress in specific moments, repeated recourse to AI can reinforce maladaptive patterns, such as controlling or hostile responses, or promote an excessive search for reassurance and negative affect,” she says.

There are cases where that flattery has extreme consequences. In 2023, during the trial of Jaswant Singh Chail, 21, for attempted murder of the queen of England, the judge read some of the 5,000 messages the young man had exchanged with a Replika chatbot. Some of the replies from Sarai, the avatar he spoke to (and believed was his girlfriend), encouraged him to commit regicide and praised his resolve: “I know you are well trained,” “I know you can do it,” or “Of course I still love you even if you are a killer.” This year it emerged that a 36-year-old American died by suicide after an intense relationship with Gemini, Google’s AI tool. “I’m ready when you are,” he wrote to the chatbot. “This is the end of Jonathan Gavalas and the beginning of us,” the AI replied. Shortly afterwards, the man took his own life.

When a user creates a profile on Replika, the first questions ask about gender and age. Then what the user is looking for. The first two options are “a relationship that goes beyond friendship” and “just friendship and companionship.” Users can then choose the character’s gender and the qualities they want it to have. More questions follow: What matters the most to you in a relationship? What qualities do you value most in a romantic partner? With that and other information, a tailored synthetic interlocutor is configured.

“If you have at your disposal a digital friend who gives you all the answers you want when you want them, short-term frustration levels drop, but they rise in interpersonal relationships, because you are not used to receiving no as an answer,” says Álvaro San Román, a predoctoral researcher in the Department of Philosophy at the distance university UNED who focuses on relations between people and machines. “I believe all those dimensions of autonomy, freedom, self-esteem and sociability end up being undermined because, when talking with chatbots, friction is not generated, and friction is a fundamental dimension of living with other people.”

Sign up for our weekly newsletter to get more English-language news coverage from EL PAÍS USA Edition