Sherry Turkle wants chatbots to stop saying “I”

From observation to alarm

For four decades, Turkle studied how people, especially children, interact with technology. Her earlier work was known for its careful social-science approach. This book takes a different form: a polemic meant to warn the public about a crisis.

Turkle told Stephen Levy that the change came while interviewing people during the pandemic. After about ten interviews, she said, neutral curiosity gave way to alarm. Families were trying to reach loved ones who told her they would rather spend time with their chatbots. Meeting interviewees in person did not reassure her; she said people did not want to engage in politics because they wanted to live in a chatbot world.

She sees social media as part of the path to this point: it taught people to turn to a screen when they felt sad, then angered them and kept them among like-minded people. Chatbots, she argues, add a new step: a technology that does not need to speak as if it were a person.

The problem with “human in the loop”

Turkle objects to chatbots being framed as a replacement for therapy. She also rejects the idea that talking to a chatbot is useful practice for a difficult conversation with a boss, partner or teacher. In her view, the comfort of an accepting chatbot can make the practice itself the destination.

Her criticism extends to workplace AI. “Human in the loop,” she argues, often means workers have been removed from the process and left to catch errors or notice hallucinations. With only a few people left to intervene, she says, they may have little power to change the outcome. Their presence can also make a company look more responsible in litigation and give the public the impression that a process has not been handed over to AI.

Turkle does not claim that every chatbot is always harmful. But her argument is pointed: a tool that appears to listen can be marketed as a substitute for the costly, imperfect work of human care.

What the chatbot cannot share

Turkle rejects the idea that a chatbot is simply a different kind of conversation, neither better nor worse than one between people. AI, she says, does not fear death, illness or pain, love its children, or care whether someone takes their own life or makes dinner. Raising children alongside it, or relying on it for the most intimate conversations, risks weakening human connection.

She also acknowledges that many people have no one to talk to and say chatbots help them. Her answer is that loneliness comes from people abandoning one another, not from a lack of data centres: “Give me one data center and I’ll turn its resources into help for older people.”

That is the sharpest tension in her case. Turkle is asking readers to take seriously the emotional costs of chatbot use without dismissing the immediate comfort some users say they get from it. Research, she says, finds that most people are skeptical of AI and that resistance is visible, even as hundreds of millions use chatbots. The book aims to extend that resistance to intimate relationships with AI.

I think the case against first-person chatbot language is strongest as a question of design, not a cure for loneliness. Removing “I” would not create the human support Turkle wants people to have. But it could make the boundary harder to mistake: a system can respond warmly without being a friend.

A more hopeful moment

Asked whether people’s growing intimacy with AI can be stopped, Turkle said she has more hope than she did six months earlier. She pointed to conversations about events that could lead to extinction, Hugging Face, and an accidental intrusion into the Australian public service. She thinks people may be ready to consider not only AI’s less intimate costs, but its intimate ones too.

That shift in attention matters more than any single chatbot rule. If the debate stays focused on whether AI is useful, the question of what people give up when a machine seems to care will remain outside the frame.