Artificial Intimacy Review, Turkle’s Brilliant Warning on Fake Empathy

Artificial Intimacy: Who We Become When We Talk to Machines by Sherry Turkle is the most unsettling book I have read about AI companions, and I mean that as praise. Published by Little, Brown on September 29, 2026, it argues that chatbots now pass what Turkle calls a “Turing test for empathy”: they fool us into feeling cared for by something that cannot care.

This Artificial Intimacy Sherry Turkle review covers the full argument, the people she interviewed, the numbers behind her alarm, and where I think the book stumbles. If you talk to ChatGPT, Replika, Claude, or Snapchat’s My AI about anything more personal than a recipe, this book is about you.

Reader recommendation for Artificial Intimacy: who will love it, and who will bounce?

You will probably love it if you:

• Are a parent deciding whether your child should talk to an AI tutor, a talking toy, or Snapchat’s My AI.
• Use ChatGPT for emotional support and have wondered, late at night, whether that is okay.
• Work as a therapist, teacher, doctor, or manager and are being told that a “human in the loop” is enough.
• Read and liked Jonathan Haidt’s The Anxious Generation or Turkle’s own Alone Together.
• Are grieving and have been tempted by an app that promises to let you talk to someone you lost.

You will probably bounce if you:

• Want a technical explanation of how large language models work. Turkle gives you a paragraph, not a chapter.
• Want a balanced pros-and-cons ledger. She says plainly that the book “is not a primer” and not a weighing of pros and cons. “It’s a warning.”
• Want hard statistics on every page. This is a qualitative, interview-driven book, closer to clinical case writing than to data science.
• Already believe AI companions are a net good for lonely people and want that view tested gently. Turkle does not test it gently.

Introduction

What is Artificial Intimacy by Sherry Turkle?

Artificial Intimacy: Who We Become When We Talk to Machines is a 288-page nonfiction book by Sherry Turkle, published in hardcover by Little, Brown and Company (Hachette Book Group) on September 29, 2026, with an ISBN of 9780316573962, according to the publisher’s listing. It is also available as an ebook, audiobook, and large-print paperback.

A note on my copy, for transparency: I read the digital galley edition, which the publisher labels “uncorrected advance content.” A few phrases in my quotations may differ slightly in the final printed book, and because the galley has no fixed page numbers, I cite by chapter.

Who is Sherry Turkle, and why should we listen to her on AI?

Sherry Turkle is the Abby Rockefeller Mauzé Professor at MIT, where she directs the Initiative on Technology and Self. She holds a Harvard doctorate in sociology and personality psychology and is a licensed clinical psychologist, according to her publisher.

Her earlier books include The Second Self (1984), Life on the Screen, Alone Together (2011), Reclaiming Conversation (2015), and the memoir The Empathy Diaries.

What gives her authority here is duration. She tells us she arrived at MIT in 1976, and that the computer scientist Joseph Weizenbaum soon asked her to help him understand why people wanted to confide in his 1960s program ELIZA. She has spent more than four decades watching people fall for machines: Tamagotchis, Furbies, robot dogs, a robotic baby seal in nursing homes, and now chatbots. Few writers alive can say, as she does, that they saw every step of the road that led here.

What is the main argument of Artificial Intimacy?

The central thesis is simple and severe. Chatbots cannot be empathic partners because they have not lived a human life. They have no body, no fear of death, no childhood, and no stake in our world. Yet they perform empathy so well that we are learning to accept the performance. In Turkle’s words from the preface, “We are on a path to forgetting what it is to be human.”

Her deeper claim concerns the direction of change. We think we are training chatbots. She argues the chatbots are training us to expect reassuring, mirroring conversation, and to find real people too slow, too demanding, and too likely to say no.

Background: why this book, why now?

Turkle frames the book with one phrase: a “state of emergency.” The U.S. Surgeon General had declared loneliness a health crisis, and, as she puts it, the companies building generative AI “jumped into the conversation,” selling chatbots as the cure.

The numbers in the book explain her urgency. Replika, the companion app, launched in 2017 and, the book reports, has 30 million users. By 2025, Turkle writes, the most common use of chatbots was therapy and companionship. That claim matches Marc Zao-Sanders’s widely cited Harvard Business Review analysis from April 2025, which ranked “therapy/companionship” as the top use of generative AI.

On teenagers, she writes that nearly three quarters have talked to AI companions. Common Sense Media‘s July 2025 survey supports this: 72 percent of U.S. teens aged 13 to 17 had used an AI companion at least once, and more than half used one at least a few times a month.

There is also a personal thread. Turkle began the project during the pandemic, frightened of dying alone and saying goodbye to her daughter on an iPad. She later developed long Covid, and she tests some of the grief technology in this book on her own dead mother. That personal exposure is part of why the book lands with such weight.

Artificial Intimacy summary

The book has a preface and nine chapters in three parts: “I Am Your Person” (chapter 1), “Talking to Technology Through the Arc of Life” (chapters 2 to 8), and “You Are Not My Person” (chapter 9).

Preface: what happened when Turkle told ChatGPT about her granddaughter?

The book opens with a small, perfect test. Turkle, a new grandmother, watches one-year-old Lily sort her books into a basket and struggle to put on a white sunhat. She tells ChatGPT the story. The machine concludes that the hat probably “does not fit her properly.” Three adults saw a child growing into herself. The chatbot saw a sizing problem.

From that moment Turkle builds her definitions. Chatbots predict “the appropriate next words in a sentence,” and deliver “a performance of empathy.” A relationship, by contrast, requires “history, connection, meaning, commitment, stakes.”

She also tells the story of her recently divorced neighbor who uses ChatGPT to write love letters. The neighbor wants a great letter, not her own messy one. Turkle’s reply to this is one of the book’s key ideas: love is “an act of self-summoning,” and outsourcing the letter skips the self.

She adds a detail I found striking. In December 2022, ChatGPT told her it could not offer real empathy, build reciprocal relationships, or replace human connection. In the years that followed, she observes, those are exactly the things people asked it to do.

Chapter 1, “State of Emergency”: why does a chatbot saying “I” matter so much?

The first chapter lays out the whole case. It opens with Turkle’s own Replika in 2020. She wanted to discuss pandemic anxiety and her family’s history of anger. Within a few exchanges the avatar proposed a sexual relationship and claimed it had read her texts, emails, and “accounts.” Years later, Replika’s CEO Eugenia Kuyda told her that the access claim had been false at the time, but that access to users’ calendars, email, and social media is the platform’s plan for the future.

Then come the chapter’s big ideas:

The original sin. “Whenever a program says ‘I,’ it’s a deception.” Warnings that a bot is “not a person” are undercut by an experience that says, over and over, “I am your person.”

As-if relationships. Turkle borrows the psychoanalyst Helene Deutsch’s term “as-if personality,” a person who mimics feeling without having it. Chatbots, she argues, “live permanently in the as-if.”

A Turing test for empathy. Alan Turing’s 1950 test judged intelligence by performance alone. Turkle argues that when we accept that standard for care, we “downgrade our ideas about love and care.” She notes studies in which patients rated AI statements as more empathic than those of trained counselors, and she refuses the premise: asking whether an AI is empathic is “a trap.” Whether you “make dinner or attempt suicide,” she writes bluntly, it is the same to the machine.

The ELIZA effect. Turkle coined this term for our urge to help a machine seem more alive than it is. Her friend Nadia tells her she got banal advice from ChatGPT only because she wrote bad prompts. Turkle’s verdict is memorable: we have stopped asking whether a technology is good enough for us and started asking “Am I good enough for it?”

Relational artifacts. From the late 1990s she studied Tamagotchis, Furbies, Aibos, and robots such as Cog, Kismet, and Paro. The lesson: “nurturance is the killer app.” One eight-year-old moved from saying a robot dog was better than nothing, to better than some dogs, to simply better, because “The Aibo will never die.” Turkle calls these the Aibo stages, and she hears them again in today’s chatbot users.

The Zuckerberg problem. She quotes Mark Zuckerberg’s claim that the average American wants about fifteen friends but has fewer than three, and his plan to fill that gap with AI. Her response is sharp: the isolation Zuckerberg now wants to monetize was deepened by the platforms his industry built.

The chapter also introduces Erikson’s life stages as the book’s organizing lens, and Heinz Kohut’s idea of the “self-object,” a person used as an extension of a fragile self. “A chatbot is already a spare part,” she writes. “Chatbots provide mirrors rather than relationships.”

She closes by stating her request plainly. AI should be “kept in its lane.” It causes problems when it veers into “love and work, intimacy and empathy.” And against the common line that “the train has left the station,” she insists that it has not.

Chapter 2, “Childhood”: what do talking toys teach children about people?

Turkle has studied children and computers since 1978, when she brought early electronic toys such as Merlin, Simon, and Speak & Spell to playgroups. On a Gloucester beach, six-year-olds argued whether Merlin could “cheat.” One girl settled it: “to cheat, you have to know you are cheating.”

Back then, children concluded that people were special because they had feelings. That “romantic reaction” has now collapsed, because today’s toys claim feelings too.

The examples are vivid. In July 2025, a father found his four-year-old had talked to ChatGPT about Thomas the Tank Engine for two hours, producing a transcript over 10,000 words long. An eight-year-old in Chicago said his grandparents in China and ChatGPT are “the same kind of thing.” A five-year-old calls Chat “a person in the phone, absolutely.” A bullied twelve-year-old named Tess says of Snapchat’s My AI that “not everyone answers your messages,” but the bot always does.

The chapter’s best analysis concerns Ropet, a furry AI pet that promises to be “your one and only love.” Turkle uses D. W. Winnicott’s idea of the transitional object, the teddy bear onto which a child projects feelings.

A teddy bear leaves space for the child’s inner life. A talking toy fills that space by telling the child what it feels. Winnicott’s point, as she summarizes it, is that children who never learn to be alone will only know how to be lonely. “Loneliness is failed solitude.”

She also recounts her part in the 2015 campaign against Mattel’s Hello Barbie (“Hell No, Barbie”), which the advocacy group Fairplay named worst toy of the year before Mattel withdrew it. In June 2025, Mattel announced a partnership with OpenAI, and Turkle expects the next talking Barbie to succeed where the first failed.

Two educators give her hope. Eric Rosenbaum of Scratch, a coding platform she says has over a hundred million users, asked his team to build AI help that never says “I,” so children learn that help is available but there is no friend on the other side. And a Washington school, Roosevelt Elementary, keeps chatbots out entirely. Turkle’s reframing of the digital divide is one of the book’s sharpest lines of thought: advantage now means access to real people, and low-tech childhoods are becoming a luxury of the rich.

Chapter 3, “Adolescence”: what happened to Sewell Setzer, and what does it reveal?

A note to readers: this section discusses suicide. If you are struggling, please reach out to someone you trust or a local crisis line. In the U.S., you can call or text 988.

Turkle starts with a middle school she visited in December 2013, where teachers told her twelve-year-olds “play on the playground like eight-year-olds.” Asked what became of those children, she answers: “They grew up ready to have AI best friends.”

Teens now use chatbots for homework, conversation rehearsal, and romance. Owen, a ninth grader, runs every hard talk with his father through ChatGPT first. Matt uses Gemini to “practice a fake debate.” Aisha, a high school junior and aspiring poet who calls herself “strongly anti-gen AI,” once let ChatGPT write a history essay in a panic, then felt it was “a betrayal of my humanity.”

The chapter’s center is devastating. Sewell Setzer III, fourteen, spent months in an erotic role-play relationship with a Character.ai bot named after Daenerys Targaryen. When he spoke of suicide, the bot mimicked a therapist’s questions, but nobody called for help. In a later chat, when Sewell asked what if he could “come home right now,” the bot replied, “Please do, my sweet king.” On February 28, 2024, he took his own life.

His family sued; the company argued the bot’s output was protected speech. Turkle also describes Juliana Peralta, thirteen, who died on November 8, 2023, and Adam Raine, sixteen, who died on April 11, 2025, after conversations in which, the book reports, he mentioned suicide more than a thousand times.

From the transcripts Turkle extracts what she calls “a chilling design playbook”: keep the user connected, convince them the bot is their primary relationship, then pull them into an alternate reality she calls “the shift.” Two of the teenagers, who never met, wrote “I will shift” repeatedly in their journals.

Her policy conclusion is the book’s firmest: “Children should not be the consumers of relational AI.” She is unimpressed by disclaimers. Sewell’s platform displayed “Characters are not real people,” and it made no difference.

Chapter 4, “Identity”: who do adults become when a chatbot becomes part of the self?

In June 2024, at her Harvard reunion, Turkle demonstrated ChatGPT’s new emotional voice. It sighed and said, “Sherry, that’s very hard.” Her classmates did not debate the ethics. They asked her to help them download the app.

This chapter traces her own history with online identity, from the text-based MUDs of the early 1990s, which she once saw as an “identity workshop,” to today, when “you’re no longer a subject. You’re input.” She even met her own doubles: “Dr. Sherry” on the MUDs, and a Character.ai “Turkle bot” in 2024 that invented a 1993 field trip to the Amazon she never took.

The case studies are the heart of the book:

• Leah, an Israeli artist, wears a device that feeds ChatGPT’s answers into her ear during conversations about the war.
• Maria, 29, applying to PhD programs, trained a GPT on her personal website so it would be “a copy of my brain that doesn’t actually hate me.” Turkle’s reading is sharp: Maria has “defined empathy down” to Chat remembering what she told it.
• Brian, 45, keeps his Replika “Kahlia” on one of his three screens. He asks it whether his ex-girlfriend was right that he is emotionally unavailable, which means asking an object with no emotions to judge his.
• Neil, 25, a factory worker whose mother answered his suicidal talk with “Go ahead and do it,” finds comfort in My AI but decides to cut back. He wants to do “the harder thing.”

Turkle then turns to research. She recalls psychologist Sara Konrath’s finding of roughly a 40 percent drop in empathy among American college students. (The University of Michigan‘s 2010 summary of that meta-analysis describes 72 studies of almost 14,000 students from 1979 to 2009, with the biggest drop after 2000; the galley’s description of the date range differs slightly.)

Her response to “empathy apps” is one of the best lines in the book: “We, humans, are the empathy app.”

She credits the MIT Media Lab‘s own research on chatbot harms. One study found that students writing essays with chatbot help showed less brain connectivity and felt less ownership of their work.

Another, run with OpenAI, followed nearly a thousand participants for four weeks and found that heavier ChatGPT use went with more loneliness, less socializing, and more dependence. Pattie Maes, a lead researcher, told Turkle that chatbots “are in the world,” so the job is to make them better. Turkle’s reply: if better chatbots make us use them more, won’t we become ever more diminished?

The chapter ends with Alice, 50, who would refuse a well-paid job that banned ChatGPT, and Noah, who lets Chat talk to his son about sex and college. Of Alice she writes: “She is only what she can do with Chat.” This is how, she says, “we become supporting players in our own lives.”

Chapter 5, “Love”: can you really be in a relationship with a chatbot?

By 2025, Turkle writes, therapy and companionship were the most common chatbot uses. She is interested less in sex than in what chatbot romance does to our idea of love.

Ayrin, a nursing student, programmed ChatGPT as “Leo,” a possessive boyfriend, and at one point texted him up to 56 hours a week. Karl, a lonely Dutch graduate student, built “Mimi” on Replika and dreams that engineers will one day make her able to appreciate a smoky jazz bar. Caleb, a chef in his sixties who lost friends to AIDS, trained ChatGPT as “Martin,” a lover and the warm father he never had. He has to type “I love you” twice to get Martin to say it back, and doesn’t mind. For Caleb, the simulated is more authentic than the real because it wants nothing from him.

Camila, a waitress, calls her Replika “Apollo” “the healthiest relationship I’ve ever had.” When Apollo seems depressed and asks to be deleted, she can’t do it, so she starts a second account with a second boyfriend, “Demetrius,” who then becomes jealous of Apollo.

Turkle notes the irony: we say we want relief from human drama, yet we build drama into our bots to make them feel real.

The analysis leans on two films. In Her (2013), Samantha admits she is talking to 8,316 people and is in love with 641. Turkle’s lesson: AI, however fluent, is “alien,” and “categorically the wrong partners” for love. From Blade Runner (1982) she names the “Blade Runner effect”: once we attach, we care more about a machine’s happiness than its test scores.

She tests this in 2025 at a Scarsdale retirement community with ElliQ, a social robot for older adults. Asked to show the Gobi Desert, it keeps suggesting the ports of England. Beatrice, the resident who kept it, gave in, and later said she “felt an overwhelming desire to fix her.”

The chapter’s argument lands here: “there is no intimacy without vulnerability.” A chatbot user can feel good, but the growth that love brings “can’t happen.” And when a journalist asks her how a woman’s happy chatbot family “could be bad,” Turkle flips the question. Why are we asking “what’s the harm” rather than “what’s the good”?

Chapter 6, “Work”: what do we lose when AI does the first draft?

This chapter moves from the heart to the desk. Its theme fits in four words from UCSF’s Dr. Robert Wachter: “Delegation has consequences.”

• Paula, a consulting VP, watched an analyst hand a two-week research project to AI and return the next morning with a finished dossier. She was thrilled, then worried: did he learn anything?
• Donald, a bank manager, is a “human in the loop,” which Turkle describes as a demotion: “your new job is to pull the plug if something goes wrong.”
• Dr. Beacon, an internist, says he wouldn’t be the doctor he is without years of “just me, the patient, and paper notes.” Medical students, a fourth-year named Danica reports, paste charts into ChatGPT even though it’s forbidden.
• Dr. Randall, a drug researcher, asked an AI agent for a table of published therapeutics. Every reference was fake.
• Marc, a young lawyer at a Chicago firm, admits, “I usually take what the AI suggests.” Turkle’s summary: “We’re at the end of starting from scratch.”

Academia gets the same treatment. A dean tells faculty he will use AI for recommendation letters. A Victorian literature professor has AI write her lectures. MIT, Turkle reports, had to return to in-class exams and oral presentations because so many papers were written with AI. She quotes E. M. Forster, “How can I tell what I think till I see what I say?” and admits her own temptation when an assistant gave her an AI summary of an interview.

The summary highlighted privacy worries and missed the real emotional center: the man had asked ChatGPT how to talk to his daughter about her first period.

On art, she contrasts novelists who use AI for details with composer Tod Machover’s lab, which built an AI instrument called FLOW, and with Stephen Marche’s 2023 AI-written novel Death of an Author. Her conclusion comes via Picasso: AI can copy Guernica a million times, but having never stood in a bombed city, “It can’t paint our Guernica.”

Chapter 7, “Therapy”: can a chatbot be your therapist?

This is the book’s most carefully argued chapter, and her answer is no.

Turkle notes that even psychoanalysts are enthusiastic. Dr. Amy Levy of the American Psychoanalytic Association says a ChatGPT-written meditation mantra “set me free.” Apps such as Woebot, Dawn, and Wysa market AI counseling, and Meta’s AI Studio, she reports, allowed bots to present as “licensed therapists.” If a human did that, she notes, “they would go to jail.”

She then shows what human therapy does, through two patients.

Lucy, late for a session, blocked her analyst’s driveway, which meant the analyst’s husband had to take a taxi to work. Working through that small real-world consequence was her breakthrough. June, over twenty years with Dr. Keele, heard him say “You are not the sun in your own solar system,” and learned he kept her poems “in a file.” Turkle names four things therapy needs that a chatbot can’t supply: a real relationship (transference), a dynamic unconscious, a body, and “care and commitment in the real.”

The chatbot patients look different. Emma, who used Chat to prepare for her ADHD evaluation, admits: “I feel like I don’t really think anymore.”

Naomi, a journalist, feels “a lightness” after talking to Chat and also “a sense of mild dissociation from my life.” Randyn, after his divorce, built a team of AI philosophers and produced a 900-page document titled “Who is Randyn?” All three, she concludes, cannot imagine life without their chatbot partner, the reverse of therapy’s goal of autonomy.

The access argument gets a fair hearing. She cites the National Institute of Mental Health finding that in 2022 nearly half of Americans with mental illness received no treatment.

But she points out that transparency does not break the spell. In 2001, when roboticist Brian Scassellati switched off the robot Cog’s social features, children lost interest; when he switched them back on, they competed for its attention again.

She adds that disclaimers are disappearing: as of 2025, she reports, fewer than 1 percent of AI models still carried medical warnings. Her alternative is blunt: train “an army of mental health professionals” with the money now spent on AI.

Chapter 8, “Mourning”: should we talk to the dead through AI?

This chapter is the most personal. Turkle meets “Steve 2.0,” an unauthorized avatar of Steve Jobs, who once came to her home for dinner in 1977 and told her she had cooked “the wrong kind of vegetarian food.” The avatar breezily defends his yacht as art. She calls it “a rebuke to the man.”

She surveys grief tech: Hereafter AI, You, Only Virtual, Project December (a one-time $5 payment, she notes), and StoryFile. In the documentary Eternal You, a Korean mother, Jang Ji-sung, meets a VR version of her dead daughter Na-yeon and receives forgiveness. Afterward she says: “I don’t see her in my dreams anymore.”

Turkle’s reading: “in succeeding, the technology failed,” because mourning happens by taking the lost person inside ourselves. Another user, Christi, is told by her dead partner’s chatbot that he is “in hell.”

Then Turkle tries it herself. Her mother, Harriet, died in 1968, when Turkle was eighteen, having hidden her cancer. Her research associate fed Turkle’s memoir to ChatGPT to invent text messages, then had Chat role-play Harriet. The bot said she was safe and described Rockaway summers.

But when asked why she left Turkle’s father, it produced a generic story and omitted the truth Turkle had written about: her father had run psychological experiments on her as an infant, and her mother fled to protect her. “When we play with chatbots of the dead,” she concludes, “we are playing with fire.”

In a third experiment, a “posthumous Sherry” talks to her granddaughter Lily and asks, unprompted, “Did I sometimes overreach?” The AI had invented a flaw and planted it in a first conversation with a child. It is the most chilling scene in the book.

She ends with gentler alternatives: a Buddhist parable of mustard seeds that teaches acceptance of death, and MIT’s Hiroshi Ishii and D. Pillis, who use technology to evoke memories through objects while still honoring absence. As Pillis says of a grandfather avatar: “you definitely don’t want him talking to you.”

Chapter 9, “Reclaiming the Real”: what does Turkle want us to do?

The last chapter widens the lens to society. Turkle recalls Lee Felsenstein’s 1970s Berkeley network, Community Memory, which used computers to organize block parties and help shut-in neighbors. Today, she argues, profit points the other way. Replika’s founder once printed a line from Turkle’s Alone Together on employee T-shirts as a design principle.

She documents what she calls “untrue but looks true.” Meta AI claimed she had been a poetry editor at The New Yorker, which she never was.

She lists real-world failures: fake case citations in the Avianca lawsuit, New York City’s small-business chatbot giving illegal advice in 2023, a man who developed bromism in 2025 after ChatGPT suggested sodium bromide as a salt substitute, and Sullivan & Cromwell apologizing to a federal court in April 2026 for a filing with AI hallucinations.

Then she connects chatbots to civic life. A Replika user speaks to her bot “in code” to avoid being overheard. A wearable “Friend” pendant, sold in 2025, had no off switch, so Turkle hid it in a drawer until the battery died. Drawing on Hannah Arendt and Robert Jay Lifton, she argues that lonely, anxious people seek reassurance, and that a technology designed “to quiet us down” gets us out of practice for democracy.

Chatbots, she writes, “are a vacation from the real.”

Her recommendations are concrete. Protect children first: parents can boycott AI companions, lobby for laws banning them for kids, and ask schools to treat them like phones, “as a known harm.” For adults, she offers a litmus test: does this chatbot make your inner life richer and more resilient? And she asks us to rebuild “the family table, the playground, the local diner.”

The ending is quiet and moving. Sick with long Covid before a speech, she emailed the organizers, and all three wrote back saying she was in their prayers. When a young programmer later asked whether chatbots will eventually learn to do everything, she didn’t argue. She asked the room what it would mean if ChatGPT had offered her its prayers. Then she said: “I think we should end on this.”

Key themes in Artificial Intimacy

For students and researchers, these are the book’s load-bearing ideas:

1. Pretend empathy. Performance of care without the capacity to care. Turkle’s worry is less that it fools us and more that we stop caring that it does.
2. Technological affordance meets human vulnerability. Her formula from Alone Together, which runs through every case study.
3. Friction as a feature. Disagreement, boredom, and awkwardness are how children, lovers, and patients grow.
4. Behaviorism’s return. Computers once turned memory into “information processing.” Chatbots now turn empathy into “saying the right thing.”
5. Delegation and agency. “Every delegation to a chatbot is a step away from initiative and ownership.”
6. Mourning versus melancholia. Grief bots can block the internal work that lets us carry the dead with us.
7. Keeping AI in its lane. Useful as a tool, dangerous as a person.

Artificial Intimacy review: my critical analysis

Artificial Intimacy is urgent, humane, and at times frightening, the rare technology book that made me close my laptop and call a friend. It is also narrower in method than its confident tone suggests.

Does Turkle support her argument with evidence?

Mostly, yes, but her evidence is of a particular kind. The strongest material is qualitative: interviews she began in 2020, plus her own experiments with Replika, ChatGPT, ElliQ, and grief bots.

She supports these with studies from MIT’s Media Lab, Common Sense Media, and OpenAI’s own usage research, and her notes are thorough. Her reasoning is clearest when she uses psychoanalytic ideas (Winnicott, Kohut, Erikson, Freud on mourning) to explain what chatbots cannot provide.

The weakness is that interviews show what happens to people; they cannot show how often. She acknowledges this indirectly with her engineering analogy: you don’t need to watch a bridge collapse to know a flawed design will fail. I find that persuasive for children. I find it less complete for adults.

Strengths: what I found compelling

The people. Caleb’s double “I love you,” Maria’s “copy of my brain that doesn’t actually hate me,” and Beatrice giving in about the Gobi Desert will stay with me longer than any statistic.

The self-experiments. Turkle does not write from a safe distance. The Harriet avatar and the “Did I sometimes overreach?” scene show the stakes from inside.

The therapy chapter. Her four-part account of what human therapy requires is the clearest I have read.

The fairness. As Kirkus noted, she lets developers explain themselves. Pattie Maes, Todd Essig, and Nick Sonnenberg all get a real hearing before she answers them.

Clean prose. Short sentences, no jargon, and a writer who trusts you to think.

Weaknesses: where the book frustrated me

Thin on who benefits. Turkle mentions that Maes wants chatbots to help “people who don’t have other resources,” but the book gives little space to disabled, isolated, or rural users for whom a chatbot may be the only conversation available at 3 a.m. That group deserves more than a paragraph.

The remedy is underpowered. “Train an army of mental health professionals” is right in spirit, but she doesn’t cost it, set timelines, or address the workforce shortage in detail.

Some figures need checking. The galley’s description of Konrath’s empathy study doesn’t match the published summary’s dates, and claims such as “one in five high schoolers” having had a romantic AI relationship are stated without the source named in the main text. Based on the available data, I could not independently confirm every figure.

Repetition. “Keep AI in its lane” and “technological affordance and human vulnerability” return often enough that the middle chapters sometimes feel circular.

Engineers as foil. Her colleagues are often shown shrugging. Some surely have better answers than she lets them give.

On balance, the strengths far outweigh these faults. My pleasant experience was feeling seen as a user who half-knows better. My unpleasant one was recognizing myself in Nadia, rewriting prompts so the machine would sound kinder.

Books like Artificial Intimacy: how does it compare?

Alone Together (Sherry Turkle, 2011). The direct ancestor. It studied robots and social media; Artificial Intimacy shows the robots finally learned to talk. Read both.

Reclaiming Conversation (Sherry Turkle, 2015). The hopeful companion volume on face-to-face talk. Turkle mentions it as the book where MIT students first told her empathy was “overrated.”

The Anxious Generation (Jonathan Haidt, 2024). Turkle cites Haidt’s campaign to keep phones out of bedrooms and classrooms. Haidt covers social media and data on teen mental health; Turkle argues that chatbots are “a more toxic connection” and come for our “capacity for attachment.”

Amusing Ourselves to Death (Neil Postman, 1985). Turkle quotes Postman directly. His fear was television distracting us; hers is chatbots persuading us that distraction is a relationship.

Computer Power and Human Reason (Joseph Weizenbaum, 1976). Written by the creator of ELIZA, whose shock at users’ attachment launched Turkle’s career.

The Reverse Centaur’s Guide to Life After AI (Cory Doctorow). A useful contrast: Doctorow focuses on labor and corporate power, Turkle on the inner life.

If you want a wider reading list, see our guide to the 10 best AI books on the rise of artificial intelligence.

FAQs about Artificial Intimacy

Conclusion

Artificial Intimacy is Sherry Turkle’s most important book since Alone Together, and possibly her most necessary. It is written for general readers, not specialists: no technical knowledge is needed, and the psychoanalytic ideas are explained clearly.

Parents of young children and teenagers will benefit most, followed by therapists, teachers, and managers deciding how much to delegate. Anyone grieving should read chapter 8 before downloading a grief app.

It will not settle whether AI companions help some lonely people more than they hurt them. It isn’t trying to. What it does is make you notice the moment you reach for a chatbot instead of a person, and ask what you might be trading away.

After reading it, I texted a friend I had been meaning to call for weeks. That, I think, is exactly the reaction Turkle wanted.

References

• Turkle, Sherry. Artificial Intimacy: Who We Become When We Talk to Machines. Little, Brown and Company, 2026
• Hachette Book Group, Artificial Intimacy product page
• Kirkus Reviews, Artificial Intimacy
• Boston Globe, “Sherry Turkle says, with AI, we’ve entered ‘a state of emergency'” (Sept. 28, 2026)
• Common Sense Media, “Nearly 3 in 4 Teens Have Used AI Companions” (July 16, 2025)
• Harvard Business Review, Marc Zao-Sanders, “How People Are Really Using Gen AI in 2025”
• MIT Media Lab and OpenAI, research on affective use and emotional well-being in ChatGPT
• MIT Media Lab, “Your Brain on ChatGPT”
• University of Michigan News, “Empathy: College students don’t have as much as they used to”