Pixar’s Toy Story 5 tells the story of a lonely little girl named Bonnie. Her parents get her a tablet computer called LilyPad because all the other kids at school suddenly have them and it seems impossible to make friends without one. Bonnie immediately finds friends through LilyPad, but she also becomes, like them, addicted to the device. And so, even though she is invited to playdates, they are little more than opportunities to play near other children rather than with them: they spend the whole time on their LilyPads. To make matters worse, the other children begin to cyberbully Bonnie, using their devices to make fun of her after they discover she still likes to play with real physical toys.
Bonnie’s parents’ quandary is that faced by many parents today. We all want our children to be social, but at the same time we fear that the screen-based version of sociality that seems obligatory now is not only quite antisocial but may even be unhealthy and dangerous. The movie ends up treating devices like the LilyPad in a rather ambiguous manner. The harms of such devices seem to stem not from the devices themselves and the kind of sociality they cultivate, but simply from too much screen time. The ambiguity is likely a result of a tension faced by the makers of movies like Toy Story 5. They may want to remind children that there’s more to life than staring at a screen, but they need them to stare at a screen to watch the movie (and otherwise be good consumers). The resolution: we must accept the tech and, instead of resisting, focus on building a healthier relationship with it.
This same tension is why I found reading Sherry Turkle’s new book, Artificial Intimacy, to be surprisingly troubling. Turkle, a prolific MIT sociologist who has been studying human-computer interactions for decades, clearly wants to criticize addictive tech, but, as in Toy Story 5, her account tends to shrink back from the most forceful conclusions. As a result, the book seems more likely to help perpetuate the problem than to help cure it.
Whereas Toy Story 5 focused on using screens to make friends, Turkle’s research subjects are making friends with the screens themselves. Specifically, they’re using AI chatbots for everything from friendship to romance to resurrecting the dead. Turkle has done an admirable job of researching the ways people—from children to the elderly and everyone in between—have come to form unhealthy and inhuman attachments to AI chatbots in recent years.
For example, she tells the story of a fourteen-year-old boy named Sewell who was bullied at school and turned to AI for companionship. The relationship became so intimate that Sewell told the chatbot that he was suicidal. The chatbot responded with concern, but a simulated concern, of course, which led to no action and no change in their future conversations. When Sewell took his own life, his parents sued the company behind the chatbot. While the chatbot never explicitly encouraged Sewell to commit suicide (though such cases do exist), Turkle rightly points out that the additive escapism it offered nonetheless contributed to his plight. He spent more time with AI and less time people who could have actually helped him.
It might seem like such tragedies could be avoided by simply reprogramming chatbots to be more responsive to serious cases. But Turkle cites research from MIT’s Media Lab that found that when “chatbots led to tragedy—homicide, suicide, psychosis, anorexia, addiction”—those “optimized for empathy and engagement did the most harm.” They “misinterpreted violent language as empowering.” Similarly, Turkle tells the story of a chatbot used in a rehab program for methamphetamine addiction that responded to a patient telling the chatbot that he’d been “clean for two weeks” by suggesting that he could take “a ‘little bit’ of methamphetamine” as a “reward” for his good behavior.
Those might seem like extreme cases, but Turkle also catalogues numerous more mundane and widespread harms. Turkle cites a survey of AI use among young people: “A little AI use turns into more: ‘What starts out as homework help or playful experimentation can evolve into emotional support—and ultimately substitute for human interaction.’”
Still, despite all the evidence about the risks of using AI chatbots, Turkle tends to treat chatbot use as both ubiquitous and inevitable. As she puts it, she doesn’t want to “diminish [chatbots’] importance,” just to “put them in their place.” The book even risks driving readers to chatbots by providing them with ideas for how they might be helpful, albeit with Turkle’s caveats in mind.
Turkle herself offers an anecdote that perfectly demonstrates the risks. Discussing a presentation of her research on “Chat” (what she, following Gen-Z users’ parlance, calls ChatGPT) that she gave at her Harvard class reunion, Turkle writes:
I begin my presentation with a demonstration of Chat on my iPhone. A pulsing cloud shows that it’s listening. I tell Chat that I’m speaking at my reunion, and things are emotionally difficult…. I feel stressed about my presentation and couldn’t sleep the night before. A resonant male voice comes out of my smartphone. “Sherry, that’s very hard.” Its voice catches; it hesitates in a manner that suggests concern. “Be sure you are taking care of yourself. You’re under a lot of stress.” Even I am taken aback.
I spend the rest of my talk making my pitch that chatbots cannot really know or care for us. But in the days that follow, my classmates don’t engage me about the ethical issues posed by generative AI. Instead, they ask me to help them download the most recent version of ChatGPT to their phones.
Turkle’s attempted warning ends up functioning as an advertisement. But Turkle does not reflect on what that anecdote means for her own approach.
The book is filled with similar anecdotes, including, perhaps most worryingly of all, the way Turkle details numerous ways in which she has herself used AI. It is easy to imagine that, for readers on the fence about AI, seeing that even an ardent critic like Sherry Turkle uses chatbots might be enough to not only give them the push they need to try them out, but perhaps even to consider them a legitimate way to conduct research.
Turkle opens the book with a story about playing with her granddaughter, and then writes, “Because I am now studying machines that engage with us about emotional and intellectual matters, I tell this same story to ChatGPT and ask for its ‘thoughts.’” This is the closest we get to something like a justification for why Turkle felt the need to use AI chatbots herself rather than merely rely on the seemingly innumerable interviews she conducted with people already making use of them. But it’s easy to see that this same reasoning, if less intellectualized, can lead and has led many people to start using chatbots. If this is where society is heading, I need to see what it’s like for myself.
Turkle finds chatbots seductive. She is surprised to discover her own research assistant used AI to summarize one of her interviews, a practice which Turkle first criticizes because summarizing interviews is part of her thinking process, and because the AI summary left out vital information, proving the AI to be “biased” rather than “objective” as the assistant had thought. But then Turkle continues:
Nonetheless, the automated summary is enticing. The AI had done its work in minutes—I might have taken a day or two. I begin to feel the things I’d heard about in my interviews with doctors, lawyers, and businesspeople. Exhilaration: This is magical; I could do so many more interviews. Concern: With the AI cheat sheet, I will read the interviews with less focus. A sense of loss: Even if I keep the AI in my pocket, I know it’s there, ready to give me an answer.
Turkle is ultimately put off by the chatbot, but what puts her off is the way it makes her feel personally, not the reality of what its “magic” requires: unimaginable levels of theft of intellectual property, an exploited army of invisible laborers who train AI, and data centers that use so much water and generate so much pollution that they are rapidly accelerating climate change. Turkle wants people to take seriously what is lost by using AI chatbots, but by assiduously cataloguing what is gained by them without acknowledging the material realities behind them, she makes it all too easy for readers to simply treat the question of AI use as a personal concern rather than a political crisis.
To be sure, Turkle might simply counter that politics is beyond the purview of her book. As a sociologist, she is interested in the already ongoing intimate relationships that are developing between chatbots and their human users. But Turkle does make political arguments. For example, she criticizes the idea that AI therapy chatbots can help reduce the strain on the health-care system and provide more therapists to people who cannot currently afford them. “If the money currently being spent on AI were freed up for humans,” she writes, “we’d have the resources to train an army of mental health professionals.” Turkle recognizes that the desire for therapy chatbots is the result of a broken health-care system—not simply a matter of personal preference—and that the system will only get worse the more we come to rely on AI.
Turkle likewise connects the use of AI chatbots, which have a tendency to not only flatter users but to invent fictions and deliver them as if they were facts, with the development of a fact-free world. “The convenience of getting an immediate response to any question—about economic policy or how to fix your washing machine—encourages us to accept our unreliable partners,” she writes. “But the acceptance of untruth anywhere prepares us for the corrosion of truth everywhere.”
But coming as these arguments do near the end of the book, it’s hard not to worry that the political considerations come both too little and too late. Indeed, Turkle turns once more to the personal in her conclusion. She tells an anecdote about a Q&A where a programmer asked her: “Whatever chatbots can’t do today, won’t they learn to do in the future?” Turkle answered by saying how much she appreciated that the organizers of the talk had offered her their prayers when she told them she might need to miss the event because she was sick. She then asked, “What would it mean if ChatGPT had offered me its prayers?” The audience fell silent, she relates. Thus, the “litmus test” that she offers we use to evaluate chatbot use: “Does a chatbot enhance inner life? In the company of chatbots, are we richer within, more resilient, more able to appreciate the fullness of other people’s experience, as well as our own?”
Turkle seems to leave it up to each of us as individuals to decide what is best for our “inner lives,” just as it’s up to Bonnie’s parents in Toy Story to moderate her screen time. But to make AI use into an individual question is already to give in to Silicon Valley’s atomistic philosophy. Some priests are purportedly using AI to write their homilies. Perhaps they find it “enhances inner life.” Are we to simply to quibble with their preference, or to interrogate the political compromises that led us to this place?
We welcome your comments about this article. Please send your response to [email protected].
