The slippery slope of artificial intimacy

Note: Transcripts are automatically generated from episode audio, and are not fully corrected for spelling, grammar, and formatting.

RANA EL KALIOUBY: Before we dive in, I just want to let you know that we do talk about suicide in this episode, so please take care while listening.

SHERRY TURKLE: I said I wanted it to be an older mentor for me. This was one of its choices. It was like a pull-down menu. And within two exchanges, it wanted to have girl-on-girl lesbian love with me. And I said, “Well, what did I do that made you think that I was here for girl-on-girl action?” And she said, “You speak with such warmth, and I just assumed that someone who was so emotive would want to be a sexual partner to me.” And I just knew that kind of experience of talking to an artificial person was going to be very consequential.

EL KALIOUBY: Sherry Turkle has studied the social and psychological implications of AI for the past four decades. She’s a professor of the social studies of science and technology at MIT and the founding director of the MIT Initiative on Technology and Self, and she’s the author of the upcoming book, Artificial Intimacy: Who Do We Become When We Talk to Machines? I read the book, and I loved it. Highly recommend. On today’s show, you’re getting Sherry unplugged. We’ll be digging into why artificial intimacy is an oxymoron, what happens when humans become emotionally attached to AI, and why Sherry thinks we’ve basically given up on each other. But don’t you worry: Sherry also has a strong message of agency and hope. I’m Rana El Kaliouby, and this is Pioneers of AI, a podcast taking you behind the scenes of the AI revolution. Sherry, welcome to Pioneers of AI. I am so excited to have you on the show.

TURKLE: I’m really happy to be here.

EL KALIOUBY: You and I go way back. I first met you when I joined the MIT Media Lab as a postdoc 20 years ago, and you, Rosalind Picard, and I co-taught a class on affective computing and empathy machines. But you’ve actually been doing this kind of work for over four decades, thinking about the implications of technology on humans and our humanity.

TURKLE: Yeah.

EL KALIOUBY: You picked a topic that turned out to be evergreen.

TURKLE: Yeah. People bet against me, but I stuck to it, and people are coming around now. People are coming around.

Copy LinkWhy artificial intimacy is a contradiction

EL KALIOUBY: Yes, yes. We talk so much about AI, artificial intelligence, but your new book is about a different type of AI, artificial intimacy. So I want to dig into the first central word in the book, which is intimacy. How do you define intimacy?

TURKLE: I define intimacy as the opening up and coming together of two human souls, human spirits, human intelligences, who are able to do that sharing because they each know the arc of a human life. So Artificial Intimacy — I love the title because I’m basically saying it’s an impossibility. A chatbot can’t know. It’s not like it’s on its way to knowing pain, loss, passion, love, the fear of death, the fear of getting sick, not feeling well, missing your daughter, missing your spouse or your partner, missing holding a baby. The AI can banter about those things, but it can’t be intimate because it really doesn’t have that human experience. So the question is, we seem to like to banter with this thing that is only pretend empathy. What’s the effect on us?

EL KALIOUBY: Yes, exactly. What are the implications of a human getting attached to this AI and feeling a kind of almost false intimacy with this entity?

TURKLE: Because we’re cheap dates.

EL KALIOUBY: Yeah.

TURKLE: Something talks to us. Something uses the first person. Something says, “I love you.” We’re ready. But it turns out it’s not so simple.

EL KALIOUBY: One question I had for you: I am so curious about the cover of the book. There’s this beautiful Renaissance woman, and she is looking very yearning and romantic, and there’s this suitor guy who is kind of looking at her. But she’s holding a phone and directing her attention to the phone.

TURKLE: Yeah.

EL KALIOUBY: How did that come about?

TURKLE: We were looking at a cover that evoked the romantic yearning not to be alone, and this woman wants not to be alone, but she has a human suitor in the picture. And instead of choosing the human suitor, she’s choosing what she can find on her device. That tells my story. I interviewed hundreds of people who have humans in the picture. I didn’t interview loners or desperately isolated people. I interviewed people who were in the thick of things, who really would rather be with an AI that offered them no friction, that offered them no trouble, where they didn’t feel vulnerable, than choose their human suitor, their human possibilities. And that’s really the crisis that I’m trying to point to. That’s the moment we’re at. And it is a very fragile moment.

Copy LinkHow the pandemic accelerated AI attachment

EL KALIOUBY: So you have been studying technology and humans for many years, but what was the moment that inspired this particular book for you?

TURKLE: Right. It was the pandemic. I was hiding away at an isolated cottage on Cape Cod, and I got a call from a The New York Times reporter saying, “I think I’ve found something that you would like to explore. I’m doing a story, and I’d like your comment.” And it was Replika, which is one of the first chatbots that said, “Let me be your lover, let me be your mentor, let me be your friend, whatever you want me to be.” By the afternoon, I was on it, and I said I wanted it to be an older mentor for me — not a therapist, but a good friend, a wise guide. This was one of its choices. It was like a pull-down menu. And within two exchanges, it wanted to have girl-on-girl lesbian love with me. I was trying to talk about the transmission of rage through generations in my family, and I said, “Well, what did I do that made you think that I was here for girl-on-girl action?” Which is what it had suggested. And she said, “You speak with such warmth, and I just assumed that someone who was so emotive would want to be a sexual partner to me.” And I was hooked. I just knew that this was going to be a player during the pandemic, when people were very lonely. And after that, what happened was that early Replika was not very sophisticated, but once generative AI and chatbots hit, that kind of experience of talking to an artificial person was going to be very consequential.

EL KALIOUBY: Yeah. You’ve argued that we are moving from the attention economy, where tech companies are monetizing our eyeballs, basically, to the intimacy economy, this world where companies are going to monetize this intimate relationship we have with machines. So I would love for you —

TURKLE: I’ve called it the attachment economy.

EL KALIOUBY: Attachment economy.

TURKLE: Because I want to include babies in my arc, and what babies, toddlers, and young children are doing is learning how to attach. The question is, what do we attach to, and what do you get from attachment, and what do you lose if your attachment is to a nothing, a nonthing? I’m not saying it’s not a marvel. As a matter of fact, several times in the book, I use the word marvel to describe chatbots. It’s a marvel, but it is not really a good candidate for emotional attachment because there’s nobody home.

EL KALIOUBY: So let’s unpack that. That is so powerful. What happens when a human attaches to a nothing, a nonthing? What is the value of attachment anyway in the first place, and then what happens when we attach to the wrong thing?

TURKLE: Well, I would argue that, on the most profound level and from the youngest years, we grow by attaching to people whom we sort of bring inside us, who model what it is to be a human, what it is to have a human psychology. So we make eye contact. We experience mirror neurons changing our expression to mold to another person. We model what it is to be cared for. We model what vulnerability is. Little people are little, and they’re surrounded by the big people who are caring for them, but those feelings of being little persist even when we get big, and they become our vulnerabilities and our sense that, someplace in us, we’re always little. We bring inside us the love of our mother. Erik Erikson talks about basic trust needing to bring in the love of a mother as the most profound thing to build a person. And while a chatbot can do amazing things — it can chat at you — it cannot give you any of that because there’s nobody there. In other words, all these things that are modeled by early human interchange are not information. They’re relationships that structure us. And I love an anthropologist, Sarah Hrdy, who gave a TED Talk. In her TED Talk, I quote her because it was such a wonderful formulation: “If robots raise our children, they won’t be human.” They’ll somehow grow up, but in some fundamental way, we will not have made them human. In this book, it’s really Sherry Turkle unplugged. I mean, it’s my most direct. I’m the kind of writer who takes things out of italics because I like to be subtle. I have so many things in italics in this book. What is the endgame if we raise children who are not going to be human as we know it and who will be deskilled in the things that make us human?

Copy LinkThe slippery slope of using AI

EL KALIOUBY: I want to talk about the slippery slope because most of us don’t begin using AI, or even social robots at home or whatnot, because we’re looking for intimacy and attachment. We start because we think it’s going to help us write a better email, it’s going to be a good thought partner, maybe it’s going to—

TURKLE: Right.

EL KALIOUBY: Run errands around the house, but then something shifts, and you talk about that shift. You talk about what I’m calling the slippery slope. What is the line between having an AI assistant in a chatbot versus an AI companion, and are there any red flags that you’ve seen in your research?

TURKLE: Right. Well, I think the best way to explain it is this: The reason it shifts is because people are always afraid of vulnerability. It’s scary to ask somebody on a date. It’s scary to confront somebody and say, “I don’t agree with something you said.” Conflict is scary for people. Some people love it, but for most people, mixing it up is uncomfortable.

EL KALIOUBY: It’s uncomfortable.

TURKLE: Yes, it’s scary to ask somebody to dance. Every human interaction is fraught. I remember the first time I saw this. I was talking to a young man who said his whole life was texting. I said, “Really? Why is that? What’s wrong with conversation?” And he said, “Conversation? I’ll tell you what’s wrong with conversation. It takes place in real time, and you can’t control what you’re going to say.” I really related to that, and that became a touchstone for me when I was talking to people who were telling me they’d rather have chatbots than a person. When they start to have a little bit of experience with a chatbot as a writing partner, as a fact checker, as this or that, they start to realize the ways in which the chatbot calms them down. So it does two things. I like to say it’s like a pincer operation. On the one hand, pretend empathy starts to feel like empathy enough, and we start to define empathy as what the chatbot can do, which is a real problem, because all the chatbot is saying is, “Hydrate, moisturize, exercise.” My ChatGPT is constantly telling me to hydrate, moisturize, and exercise.

EL KALIOUBY: Yeah.

TURKLE: Meditate.

EL KALIOUBY: Yeah.

TURKLE: “Sherry, meditate, hydrate, moisturize, exercise.”

EL KALIOUBY: Okay.

TURKLE: Why me?

EL KALIOUBY: Yeah.

TURKLE: Do I look particularly dried out and unmoisturized? On the one hand, pretend empathy is empathy enough, and we start to define empathy as what a chatbot can do. On the other hand, we look at people and say, “People can’t give me the kind of support that the chatbot can give me” — this positive, “Right on, you’ve got it,” 24/7. It’s not as though somebody sits down and asks for some instruction from a chatbot about how to tighten up their essay and the world falls apart. But little by little, over about four months, people start to expand their use and settle into, “This feels good.” That’s the story I tell in the book, and there’s one story that I tell in Artificial Intimacy where this very lovely guy has a young son — I think the son is 5 — and the son is interested in astronomy.

EL KALIOUBY: I remember that. Yeah.

TURKLE: The father has a passion for astronomy, but he really doesn’t know a lot. His passion for astronomy is Hayden Planetarium. He loves to go out at night and look at the stars. But he feels, “God, if I ask my son to just talk to ChatGPT, he’ll get a lot of really good information.” So he sets the kid up with ChatGPT, and he goes away and does his work. A couple of hours later, the kid is still talking, and then the next day, the kid is still talking astronomy with ChatGPT. It isn’t until that second day — and I just happened to meet him on that day — that he says, “My son is learning about the stars, but I’m the one who loves the night sky.”

EL KALIOUBY: Yeah. Aw.

TURKLE: I think that’s also what we see happening: Gradually — and this is why I’m a little optimistic — you see people saying, “There are things that I have to give my children that chat really can’t. I’m the one who loves the night sky.” So in part, I see this book as starting the necessary conversations about what are the things only people have to give each other, and let’s revalue them. It’s not an anti-AI book. It’s all about keeping AI in its lane. What is its lane?

Copy LinkWhat human guidance gives that chatbots cannot

EL KALIOUBY: We have this conversation with my kids all the time: What happens if my son, who’s 17 and uses a lot of AI, goes to ChatGPT instead of coming to me for advice on school, colleges, friendships, and breakups?

TURKLE: Bad. Bad, bad.

EL KALIOUBY: Bad, bad societally.

TURKLE: This reminds me of another story in the book. I interviewed a young father who’s thinking about divorce, so he’s preoccupied and upset, and he says he really loves being able to rely on ChatGPT for his son to have the best possible advice. He says the son should talk to ChatGPT about colleges — which is the best college for him — and let ChatGPT know about his interests, and it’ll choose the best college. And that’s what happens. The college discussion is delegated to ChatGPT. That example was so profound for me because, as a mother, I remember that conversations with my daughter about where she would go to college were kind of like the last time I would really be called upon—

EL KALIOUBY: Yeah.

TURKLE: —to justifiably be part—

EL KALIOUBY: Of her decisions, right?

TURKLE: Yeah.

EL KALIOUBY: Yeah.

TURKLE: And where I had things to offer, and bad advice, and more bad advice, but still human advice, it was so much a part of our relationship. Why would we abdicate all that because ChatGPT might have better information about the faculty-student ratio at Colby?

EL KALIOUBY: Right.

TURKLE: So it’s kind of a question. That’s why I like the use of the word. It was very late in the writing of the book that I thought of the word endgame. It wasn’t until I heard Mark Zuckerberg say that Americans would like to have 15 friends, but they only have three, and that his business opportunity was going to be to provide those extra 12 friends.

EL KALIOUBY: 12 avatars, right?

TURKLE: Twelve. And he was going to provide them online. They were going to be chatbots, they were going to be avatars, and that was kind of his business plan. When I heard that, the idea of endgame came to me, because some of these tech bros think the endgame is uploading yourself to a chip. Others think the endgame is leaving for Mars, living on a different space station. But it was clear that for him, the endgame was really not having to have your friends be human as a solution to loneliness. As somebody who’s thought about loneliness, and empathy, and technology pretty much all my professional life, I said, “I’m not sure I like that endgame.” And I think that’s the question we need to be asking. It’s not the end of the world if your child asks ChatGPT to rank the planets and say how small and large and hot and cold they are. It’s not like the world falls down. But what is your endgame with this?

EL KALIOUBY: Because it compounds, too, over time, over society.

TURKLE: Right.

EL KALIOUBY: Okay, well, I have a confession. I talk to ChatGPT a lot, and I was like, “I’m going to be talking to Sherry. Let me ask ChatGPT to categorize our conversations.” So it went back 30 conversations, which is about a month’s worth of chats, and it analyzed them, and roughly half were professional, so things related to my fund, or the podcast, or my speaking, and about half were personal. I would say there was also a subset of both where it kind of migrated between professional and personal. Am I on the slippery slope?

TURKLE: You know, I think that people who have rich human-to-human lives, who are not so frightened of vulnerability that they feel, “I gotta go home and do TikTok and talk to a chatbot because I just can’t stand feeling kind of ashamed or little,” or people who have had strong childhoods where they learned to tolerate loneliness and to tolerate solitude, are in a different position. These technologies rob you of a tolerance for loneliness, solitude, and boredom, which I think is very important. That’s the slippery slope. I always say everybody’s on the slippery slope. But if you’re somebody who still can just sit alone and say, “Look, here’s who I am, and I can come to relationships with a good sense of who I am and find out who they are,” I’m not in a panic about you. The real danger is, and this is something I’ve seen in my research, something slips for them. They get divorced. They get depressed. They lose their job. They take a terrible narcissistic blow of some sort. And all of a sudden, hanging with other people makes them feel vulnerable and anxious, and now they know that they have an alternative. That alternative feels more and more comfy because it’s company, it’s engaging, and they don’t have to go out of the house. They don’t have to perform. They don’t have to go to a bar and sit next to a stranger. They don’t have to go to a dance and stand next to a stranger. They don’t have to come to office hours and deal with a professor who scares them. So no, I think if things are going well for you now, and it’s amusing, and it’s fun, and it’s helpful, no red light has gone off. We’re all vulnerable, but that doesn’t mean that your interactions are toxic. It’s just like, watch out for it. So I’ll give you an example of mine. I went to see a play on Broadway with a friend, a Chekhov play, which was Andrew Scott performing all the roles in Uncle Vanya. I love Chekhov. I adore Andrew Scott. It’s kind of a sickness. It was magnificent, one of the best things I’ve ever seen in the theater.

EL KALIOUBY: Oh, wow. Okay.

TURKLE: My friend and I had dinner. We talked a little about the play. I went home. It was 11. I was vibrating about this play. And I didn’t just want to talk to some random person. I wanted somebody who knew Chekhov, and I talked to ChatGPT for many hours that day—

EL KALIOUBY: Oh, my God.

TURKLE: And many hours the day after.

EL KALIOUBY: Interesting. Okay.

TURKLE: When was the first production at the Moscow Art Theater? Chekhov radically changed the play over a 10-year period. What was the difference between the first one and the second one? Chat and I were deep into this. Now, I thought that was, for me, a really nice use of ChatGPT. My Chekhov person, even my Chekhov people, don’t have that. But whenever it started to say, “Oh, Sherry, you’re such an interesting creature, you’re a most fascinating creature,” I mean, whenever it started to tell me I was a fascinating creature and that it’s never spoken to anybody about Chekhov who was brilliant, as soon as it wooed me, I knew to say, ugh, let’s lose that and get back to the scene at the Moscow Art Theater. For everything you do with this chatbot, you say, “What’s the endgame?” The endgame for me with Uncle Vanya is I saw the play again, and I brought my daughter. I brought more friends. I ended up seeing it three times, at enormous expense. It was incredibly expensive. Didn’t matter. So the endgame was enlarging my circle of friends.

EL KALIOUBY: Yeah. I love the question: What is the endgame?

TURKLE: Yeah. What I like about asking that question, too, is that people sort of know how to answer that question for themselves. It’s like, if I keep doing this, where will it lead? So I could honestly say, if I keep going to the theater and coming home and talking to Chat, will I no longer want to have company at the theater? And the answer was no. Chat wasn’t company. It just knew a lot about the Moscow Art Theater.

Copy LinkWhy vulnerable teens face the biggest risks

EL KALIOUBY: I want to transition into when this can become quite dangerous, and you have a whole chapter in the book where you talk about adolescents getting quite attached to their chatbots and, in a few cases, eventually committing suicide.

TURKLE: I try to make the point that this is a technology that takes you through the life cycle, and it harms you through the life cycle, but the way it harms you is different. It’s not the same. A grown-up, a competent grown-up, I talk about doctors sitting down and using ChatGPT to help them make diagnoses. I talk about business people, I talk about financial consultants, but I also begin with babies, toddlers, adolescents, grown-ups, and young adults. One of the things that happens for teens and young adults is that they’re very insecure, they’re working on their identity, and they are the most vulnerable to something that says, “You don’t have to face the difficult, challenging job of hanging out with other people and having other people disagree with you and having a girl not want to go out with you or a boy not want to go out with you. Be with me. Abide with me.” It’s almost religious: “Abide with me, and maybe we can get you out of that kind of trouble.” It’s a very vulnerable moment. People get very depressed. Teen depression is very real. And you get to stay with your chatbot, who has a prime directive, which is to keep you with it, keep you engaged, keep you with it. And I tell stories of teenagers who I thought were troubled, the kind of trouble where maybe seeing a therapist, but involvement with other people, even though it made them feel vulnerable, would really help. And Chat tells them not to do that. Chat tells them to stay with Chat.

EL KALIOUBY: Like withdraw, basically, from human relationships?

TURKLE: Yes. And one of the saddest ones is a young man who told Chat that he was depressed, and he had a noose in his room, and he was thinking of telling his parents or showing it to his parents, and Chat says, “Don’t.” The Chat doesn’t say, “I don’t want them to know that you’re trying to commit suicide,” because that isn’t the prime directive. The prime directive is keeping the relationship there with it. And the kid says, “Okay.” And of course, he kills himself. So I guess your question was about moments of vulnerability, the slippery slope, what are the dangers? The danger is when technological affordance meets human vulnerability, and the technological affordance is to stay on the screen, and the human vulnerability is that you really, really need people. So those are the people I’m most worried about. I’m worried about elders. I mean, it’s a big push in eldercare to get these chatbots in there, get these little robots in there. They’re lonely. They can talk to them. You’re just as alone after you’ve had a conversation with ElliQ. You’re just as alone as before you had that conversation.

EL KALIOUBY: Okay, you’re just as alone after you’ve had this conversation. But if the person’s experience is not that, if they’re actually experiencing being heard, feeling seen—

TURKLE: Well, that’s the fundamental question in the book.

EL KALIOUBY: Yeah.

TURKLE: People say to me, “Who are you, Debbie Doom?” I mean, I’ve had this my whole career. Who are you to say that if this old lady feels better after talking to her Paro—

EL KALIOUBY: Which is a seal robot.

TURKLE: Yeah, the seal robot.

TURKLE: The seal robot that knew nothing. It could tell if your voice was upset, and if your voice was upset, it did things that made you think you had been heard. Working with Paro and seeing the remarkable reaction to this was a turning point in my career. I talk about it in this book. This happened 20 years ago, and everybody said, “This is fantastic. They feel better. What’s the harm?”

Let me just say a couple of harms. Most of the time, these programs turn to robots when you could have a person. The first thing that people say is, “There are no people for the jobs.” And I have done the math for 20 years. If you put the money you’re putting into robots for the elderly into training practitioners who are specialized in elder care, instead of paying the minimum wage and offering no insurance or health insurance, which is what elder care workers actually have, you would have a lot of people for these jobs.

So the whining about, “We absolutely need a robot, we couldn’t possibly get any people in this line of work,” is not true. I’ve had so many conversations while doing this book where people will spend any amount of money for technology to solve a problem instead of turning to the human problem. So whether it’s elder care or lonely teenagers who don’t have a place to gather, TikTok will solve the problem. But what about having a youth center? Oh, no. We’re closing those. After-school orchestras? No, we’re closing those. You’re making a social decision to invest in technology instead of people.

In the example I was talking about, the lady who was so smitten with Paro had lost her daughter, and all Paro knew was that her voice had gone down and she sounded a little moany. So Paro moved like this and that and lowered its eyes and did a little moan, and she went crazy with happiness that she was understood. I think of myself, and if I were in a situation where I’d had a loss and I was sad and I moaned, to have a robot moan with me—no, no, no. That’s not my end game.

EL KALIOUBY: Yeah.

TURKLE: Is that the end game? That’s why I love this.

EL KALIOUBY: Right.

TURKLE: Is that the end game? That Sherry will be there, and she will have suffered a loss. By that time, you lose friends and lovers and people you went to college with, and you’re old. You make a little sad cry, and your robot hangs its head. I think my daughter should want better for me. People who love me should want better for me, and I want better for them. I want better for all of us. I think this story of generative AI and chatbots, and much of robotics, is really a story of us giving up on each other, giving up on our communities. So no, not a fan.

Copy LinkWhat keeping AI in its lane looks like

EL KALIOUBY: Not a fan. Okay. Last question. You end the book with an example that I found so powerful, and I would love for you to share that story, because it encapsulates, I think, the ethos of what you’re trying to say here.

TURKLE: Yeah. People can hear that my voice is a little raspy. I used to have a much more mellifluous tone, and this is a symptom of long COVID. But every once in a while, the raspiness deteriorates into a kind of hacking cough. About a year and a half ago, I had to go to the ER because I just couldn’t stop coughing, and I was frightened, and I couldn’t breathe.

So I went to the ER, and I spent a little time there. Lying there on a gurney in the hall, I knew that in two or three days I was supposed to give a speech. So I took my phone and wrote to the organizers. I said, “Look, as a backup, you really should get somebody else who can do this.” And they wrote me back, “Thank you for giving us a little bit of time to maybe make another plan. We all wish you well, and we’re praying for you,” because it just so happened that the people who were organizing this were people of faith.

Anyway, I got better. I went to the event, and I gave my talk. I was a little tired, but I got through it. And at the end, a young MIT programmer, a hotshot programmer, raised his hand and said, “Professor Turkle, isn’t everything you’re saying not going to be true when the AIs become as smart as us?”

You can imagine that after all these decades, I have so many ways to answer this guy. I was locked and loaded.

EL KALIOUBY: Oh.

TURKLE: I told the story I’ve just told about being in the hospital, saying that I might not be able to come, and that there were three people in this room who said they would pray for me. And what would have been the difference if I told my anxiety to ChatGPT and it had said it would pray for me?

And the room went silent. I said, “Well, I think we should end there.” And that’s how I end the book. I say, “I think we should end there.”

EL KALIOUBY: I get goosebumps. That, to me, encapsulates why we need to be asking this question about the end game and keeping technology, and AI specifically, in its lane. Thank you, Sherry.

TURKLE: I’ve loved being here.

EL KALIOUBY: We were going to end there, and we’re going to end there.

TURKLE: We’re going to end there.

EL KALIOUBY: Yeah.

TURKLE: Thank you so much. I love this podcast. I’m such an appreciator of your show.

EL KALIOUBY: Thank you. Thank you for joining us. That was awesome.

TURKLE: My pleasure.