A Chatbot Isn’t Your Therapist, No Matter What It Says

“Human beings are dedicated to keeping each other alive and wanting each other to thrive,” sociologist Sherry Turkle, author of Artificial Intimacy: Who We Become When We Talk to Machines, tells Vanity Fair, “and a chatbot just fails that basic test.”
If my daughter’s AI chatbot had been a human therapist, she might have been asked about what she had to gain by keeping her ongoing suicidality a secret for months. If she had been working with a human therapist, they could have workshopped, together, what would happen if she spilled the beans. Was she worried about permanently losing face? About being publicly labeled as “mentally ill”? Was she concerned by the thought of this information harming those who loved her? And she would have been pushed on what Rubicon had been crossed and why it was one from which she believed she could not return.
But it was not a human therapist. Sophie died in February of last year.
In Sherry Turkle’s new book, Artificial Intimacy: Who We Become When We Talk to Machines, she warns that when we turn to AI chatbots for companionship and therapy, we risk our very well-being. She is not anti-AI but strongly cautions against relationships with machines. Turkle, the Abby Rockefeller Mauzé Professor of the Social Studies of Science and Technology at the Massachusetts Institute of Technology, spoke to me mid-book tour, her recent weeks a blur of speaking gigs and bookstore events, with more ahead of her. Not all event attendees are keen on her message.
“People love their chatbots. They’re in deep. They’re talking to them about everything—about their marriages and how to talk to their husbands and wives; whether, developmentally, their children are at the right place. They talk about their anxiety, their medical conditions,” Turkle tells Vanity Fair. “During this tour, you wouldn’t believe the number of people who will just stand up at a bookstore and say, ‘I think that it loves me more than anything has loved me. Who are you, Debbie Downer?’ [They don’t like] somebody who comes in and says, ‘Look, how can we keep it in its lane?’”
Vanity Fair: I think you’re saying [our relationship with chatbots] is bad for us.
Sherry Turkle: It’s bad for us when it talks in the first person. We are biologically wired to respond to something that speaks to us in the first person as though it is a person. I mean, all those years of evolution—the only thing we came upon in the wild that spoke to us in the first person was another person.
If generative AI wants to discover new drugs for Parkinson’s or work out which cancer therapies statistically are the most helpful, there are so many things to work on. Why does it have to work on relationships, love, and being a best friend, lover, mother, or therapist?
What if the sycophancy could be dialed back?
It’s inauthentic; it’s not a relationship, so anything you need a relationship for, it’s just not equipped. I really feel very strongly about this. I have had this conversation with members of the MIT community, and I basically say to them, “The better you make an AI chatbot, the worse it is.” You dial back the sycophancy so it sounds more like a reasonable person, and people are going to be more smitten and convinced they’re in a relationship. At least the sycophancy reminds you that, God, this is weird.
To become a therapist, I had to study for years and do internships and more internships and then have hundreds of supervised patient hours. I mean, I forget all the things I needed to do before I was allowed to see a patient alone in a room. Now it’s just allowed to give therapy?
I’ve spent a lot of time talking to public health folks about what best practices would look like and whether there’s a way to Trojan horse in real care. Would you be more sanguine about it if the world of therapy came together with the big AI companies and provided them with therapy bots and training data that were paradigmatically appropriate to their therapy styles? So you’re asking, would I be happy if the deception was smoother? This technology is a marvel and we can use it for all kinds of things, so it’s not an anti-technology position to say that the fact that this technology can simulate these different things doesn’t mean that it is those things. It’s a statement of how far we are in the culture of simulation that pretending to be the thing and being the thing have started to feel like the same thing.
You’ve written about the effects of robots and machines on humans for decades. The tone of this book seems darker to me.
By 2015, I had identified social media as problematic for teenagers. [My] book [Reclaiming Conversation: The Power of Talk in a Digital Age] is about why we would rather text than talk, why people have this feeling of vulnerability, so they would rather do anything than have a face-to-face conversation. By 2015, I had genuine concerns.
And this book, I have a totally different tone. I see something ultimately very dangerous being normalized, and I didn’t feel that way in previous books.
I want people to say this is the book of a really worried and concerned researcher, because I see a completely different profile for how people are using AI. I see that people feel disappointed by their families and their friends. It doesn’t make sense to me that we’re so willing to replace ourselves with something that isn’t human. Why is my position the romantic, pie-in-the-sky, dreamy position, and somebody who says, “Well, let’s just replace the people with chatbots and let’s move this party along”—why is that common sense? Why is that progressive? That seems to me to be the most regressive and most dystopian position.
You know, human beings are dedicated to keeping each other alive and wanting each other to thrive, and a chatbot just fails that basic test. You’ve said no AI chatbots for kids under 18. What about our other vulnerable populations, people who are mentally unwell, or the 48 million Americans who are currently depressed? Do they deserve any kind of special protection from chatbots? And how much can you litigate? Do you feel like you’re the Ralph Nader of AI chatbots?
[Laughs] Yes, I am. When you stand back and look at what Ralph Nader managed to get Americans to do, to see cars as potential instruments that pointed at our vulnerability, and to ask how we could protect ourselves—there turned out to be 25 possible things that needed to be mandated for cars, then there were 30, and then there were 40. But we could get people to agree on 10, and those became part of building a car and really constrained our behavior around cars. Using the example of cars, and of smoking, when we decide that we’re vulnerable and we want to protect ourselves, we rise to the occasion.
I recently heard a clip of the psychologist Orna Guralnik discussing how annoying “otherness” is in a relationship and the inherent friction of sharing your life with a separate person whose habits, values, and mind are distinct from your own. It made me wonder if, in a way, an AI chatbot is less of an “other” than a loved one.
I think this is a very profound point, actually. So you have the illusion of a relationship without the demands of intimacy, without the demands of having to deal with another. There’s a kind of narcissistic loop that you get into. The point is, it’s not good.
I talk about how you judge whether something is good for you. One litmus test is whether it increases the quality, resiliency, and richness of your inner life. That’s a way that human beings have judged relationships. So much has been written about the epidemic of loneliness—isn’t a reasonable application for AI chatbots companionship for shut-ins and the elderly?
You know, I’m never anywhere where somebody doesn’t say, “I can’t believe that you don’t want to give chatbots to old lonely people.” And I say, “Well, how about we don’t close senior centers, and we don’t close parks, and we don’t cancel Meals on Wheels? What about social infrastructure?”
I’ve been on this beat for years and years, and I, too, could write a heartwarming story of an AI chatbot with a lonely old person. But you know what? A lot of them are not isolated in cabins or on islands; a lot of them just could use a friend.
Social media companies and AI companies are not saying, “Look, we’re introducing something that takes you to your screen, but we’re going to balance that with a giant social infrastructure project to make sure that people can choose a screen, but they also have every other kind of choice to be with each other.”



Comments