Why AI conversations are always one-sided

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

RANA EL KALIOUBY: I use a variety of AI tools in my everyday life and work, but I do use ChatGPT as my personal confidant. I was reflecting on why I trust this AI.

CHARLES DUHIGG: When you turn to ChatGPT and confide in it about something that’s happened, why are you doing that?

EL KALIOUBY: I think I use it as a tool for self-unpacking and self-reflection. I’ve journaled a lot, and it’s maybe a better kind of interrogative journaling.

DUHIGG: That would’ve been my guess, 100 percent. When you’re talking to a chatbot about relationships, you’re not turning to it because you think it’s going to give you great advice. You’re turning to it because it’s a very efficient and interrogative way of essentially keeping a diary. Whereas if you call up your friend and you say, “Here’s what’s going on with my relationship,” part of it is that you want their actual advice, like, “Are you thinking about this the right way?” Part of it is that you want to get it off your chest. You just need someone to hear what you’re thinking. There are all these different things that a conversation can represent. It’s very infrequent that AI offers all of those different things.

EL KALIOUBY: Charles Duhigg is a Pulitzer Prize-winning journalist and a writer at The New Yorker. You might know him from his book The Power of Habit. In his New York Times best-selling book, Supercommunicators: How to Unlock the Secret Language of Connection, Charles explores the science and art of truly connecting with other people. So what does it mean to be a supercommunicator, especially in a world where so much of our connection is mediated through technology? Today, we’re taking Charles’ research and applying it to our AI-infused lives.

[THEME MUSIC]

I’m Rana El Kaliouby, and this is Pioneers of AI, a podcast taking you behind the scenes of the AI revolution.

DUHIGG: Rana, did you live in Egypt? Did you grow up in Egypt?

EL KALIOUBY: I grew up in Cairo and also around the Middle East. Did you spend any time in Egypt?

DUHIGG: I did. I used to live in Bab al-Louq near AUC.

EL KALIOUBY: What? I went to AUC.

DUHIGG: You went to AUC?

EL KALIOUBY: Yes, I’m an alum. I was on the board forever. My sisters went to AUC.

DUHIGG: Oh, wow.

EL KALIOUBY: Were you studying there?

DUHIGG: I was there from ’97 to ’98.

EL KALIOUBY: Oh, no way.

DUHIGG: Yeah.

EL KALIOUBY: It’s so nice to have you on the show. Welcome to Pioneers of AI.

DUHIGG: Thanks for having me.

Copy LinkHow communication gaps inspired a search for deeper connection

EL KALIOUBY: I wanted to start with a personal story. I’ve been studying human communication and human connection for over 26 years, and my origin story is that I grew up in Cairo and around the Middle East. Then I had an opportunity to do my Ph.D. at Cambridge University. It was my first experience living away from home. This was in 2000, so a while ago. The only way I had to communicate with my family back home was through text and sending emojis. I was homesick a lot of the time, and I would just send a little emoji with tears. At some point, it struck me that all the richness of human communication kind of disappears when we’re online. It disappears in cyberspace.

That got me down a rabbit hole of literally engineering human communication. I became fascinated with facial expressions of emotion. So I wanted to hear your why. You wrote a book called Supercommunicators: How to Unlock the Secret Language of Connection. Why did you write this book?

DUHIGG: The reason I wrote Supercommunicators is that I had written a book called The Power of Habit a number of years ago, which is about the science of habit formation. I heard from a lot of people who said, “This really changed my life. It helped me build new exercise habits, helped me stop drinking or smoking.” That was really powerful, but I also fell into this bad pattern with my wife. I was working at The New York Times at that point, and I would come home from work and start complaining about my day. I’d say, “My co-workers don’t appreciate me,” or, “My boss doesn’t realize what a genius I am.” My wife would do this pretty natural thing. She would give me some pretty good advice. She would say something like, “Why don’t you take your boss out to lunch so you guys can get to know each other a little bit better?” Instead of being able to hear her advice, I would get even more upset.

I would say, “Why aren’t you on my side? You should be outraged on my behalf. You should be telling me how smart I am and how much it’s their fault.” This is something I’m sure is familiar to everyone who’s listening, because this happens again and again in all relationships. So I went to these researchers and asked them, “Why do I keep falling into this trap? I’m a professional communicator. I’m a journalist. Why do I keep making this mistake?”

They said, “Actually, we’re living through this golden age of understanding communication. For the first time, we can see what happens inside someone’s brain as they’re having a conversation. And you think when you’re talking to your wife that you know what that discussion is about. You think you’re talking about your day, or where to go on vacation next week, or, if you’re at work, what the budget should be for next quarter. But actually, there are many different kinds of conversations all happening at the same time. Unless you and your wife are having the same basic kind of conversation, you’re going to feel like you can’t connect with each other. You’re going to feel like you’re two ships passing in the night.” I thought that was really interesting. I started digging into it, and that gave birth to Supercommunicators, the book.

Copy LinkWhy smart communicators still miss what others need

EL KALIOUBY: Amazing. What kind of communicator were you before you wrote the book?

DUHIGG: I think I was someone who very quickly went to the practical. That’s where I was comfortable. It’s kind of how I’d grown up — in logic and getting things done. I didn’t understand that part of communication is not only listening, but showing people that you’re listening by asking follow-up questions, building on their ideas, and saying, “Oh, you mentioned this thing, and it made me think of this other thing.” Not just absorbing what you’re telling me, but showing you that I’m thinking about it, absorbing it, and processing it.

EL KALIOUBY: Yeah. As I was reading your book, I was thinking about where I fall on the spectrum of being a supercommunicator. I’m a very social person, and I love connecting with people. I also feel like I am a confidant for many people, but I avoid hard conversations, honestly, like the plague.

DUHIGG: Give me an example. What’s a conversation you’ve had recently that you found yourself shying away from, that in retrospect made you think, “You know what? I should have leaned into that”?

EL KALIOUBY: Not to get too TMI on this thing, but we might as well. I just ended a three-year relationship, and it’s because every time we had a hard conversation, it escalated very quickly and ended in a blowup every single time. We just could never get through that.

DUHIGG: Yeah, absolutely. First of all, I’m sorry that it didn’t work out, although it sounds like it was probably kind of a hard relationship. So maybe it’s a good thing that you’re onto other pastures. In psychology, we actually talk about this a lot. There’s a thing known as kitchen sinking, which is a really destructive pattern in marriages and romantic relationships. When we have a fight about one thing, it suddenly becomes a fight about everything. We start by talking about, “Where should we go for Thanksgiving?” And it’s like, “Well, I want to go to my mom’s and you want to go to your mom’s.” Then five minutes later, it’s like, “Well, you don’t earn enough money, so we can’t go. And by the way, your mom hates me, and you never stand up for me when we’re at your mom’s place.” So we know this pattern, and we know how destructive it is.

One of the things we know is that simply acknowledging and being aware of the pattern often makes it easier. One of the things about hard conversations that usually helps is to start by doing two things. First of all, acknowledge that this is a hard conversation. “I want to bring this topic up with you. I feel a little awkward doing so. It might make you feel awkward.” The anxiety that a hard conversation produces, if we voice that anxiety, often helps resolve it a little bit. Not fully, but it makes it a little less painful. The second thing is to set rules for the conversation, what psychologists, as you know, would refer to as a meta conversation.

Let’s talk about how we want to talk to each other. Say something like, “OK, look, I want to talk about where we’re going to go for Thanksgiving. I know that when I bring this up, sometimes we start talking about each other’s moms in unkind ways. Sometimes we start talking about money. Let’s set all of that aside, and let’s just talk about the logistics of where we’re going to go for Thanksgiving. Then I promise we can talk about moms, and we can talk about money some other time. Let’s just decide together what is the thing that we want to talk about. What’s the resolution? What’s the question we want to answer?” When you do that, suddenly you’re sitting side by side, and it’s not you versus them. It’s the two of you versus a question or a problem together.

EL KALIOUBY: That’s cool.

DUHIGG: We’re on the same side of the table, and the thing that we’re trying to solve is on the other side of the table: Where are we going to spend Thanksgiving?

EL KALIOUBY: I appreciate that, especially the anxiety, naming the anxiety beforehand and allowing for some ground rules. OK, so I would love to explore some of the ideas behind —

DUHIGG: Sure.

Copy LinkThe simple framework behind every better conversation

EL KALIOUBY: Being a supercommunicator that you shared with us in your book, but I also want to layer in what this means in an AI-driven world. So let’s maybe start with you defining what a supercommunicator is.

DUHIGG: Yeah. A supercommunicator is a person who has these skills and uses them. We talked about the skill of proving that you’re listening. Another really important skill is recognizing what kind of conversation is occurring. In general, these different kinds of conversations fall into one of three buckets. There are practical conversations, where you and I are making plans together or solving problems. Then there are emotional conversations, where I tell you what I’m feeling and I don’t want you to solve my feelings. I want you to empathize with me. Then there are social conversations, which are about how we relate to each other and how we relate to society. I told that story about my wife and me kind of missing each other because what was happening was that I was having an emotional conversation. I was complaining about my day and talking about how I felt underappreciated.

My wife responded with a practical conversation. She started giving me advice, and that’s why we couldn’t connect with each other. That’s why we couldn’t fully hear each other. It actually comes down to parts of our brains. Those three different kinds of conversations use different parts of our brains, and when we’re not using the same part of our brain at the same time, it’s very hard for us to feel connected.

EL KALIOUBY: Yeah. If you’re having a practical conversation, I imagine that’s invoking your prefrontal cortex, right?

DUHIGG: Oh, absolutely. Prefrontal cortex. Emotional conversation is the amygdala, the interior structures of the brain. The social conversation tends to take over the medial network that’s kind of at the peripheries of our brain. OK, so what does this have to do with AI? One of the interesting things is that AI can imitate those kinds of conversations very well. We’ve all experienced this. When you’re talking to a chatbot, it’ll ask you what you’re feeling, and it’ll ask you about your feelings. It’ll try to figure out whether you’re trying to solve a problem or just express yourself. One important thing that we’ve learned about conversations is that when I say something to you, the way that you respond, and in particular believing that my words have an impact on you, is a very, very important part of that conversation. So if I say something that makes you angry, there’s some information that I’m receiving from that anger.

If I say something that makes you feel good, I feel a sense of pride in making you feel good. If I tell you something that’s interesting, I feel smart because I see that you’re interested. The feedback that we get from the person we’re talking to, and how that affects our internal state, is a critical aspect of how conversations unfold. But when we’re talking to AI, we do not believe that we are changing anything in the AI. I don’t believe that I’m actually making an AI feel proud. I don’t believe I’m impressing an AI. I don’t believe that an AI actually feels the emotions that I’m feeling. As a result, that reciprocal echo of what’s happening disappears. It’s because, often, when we’re talking to AI, we’re not getting that reciprocal feeling that talking to a real person gives us. Part of what I’m doing when I’m talking to you is creating a mental image of you in my head, of how you’re receiving these words.

EL KALIOUBY: I’ll be right back with my conversation with Charles after this break.

[AD BREAK]

Copy LinkWhy AI can feel trustworthy without being relational

I want to unpack that a little bit more because there are, for example, 900 million monthly active users on ChatGPT. On average, a person spends about 30 minutes a day in conversation with AI. To make it very personal, again, I use a variety of AI tools in my everyday life and work, but I do use ChatGPT as my personal confidant. So it knows a lot about this relationship that was up and down, up and down, up and down. I was reflecting on why I trust this AI. Even though I’m an AI expert, I built these kinds of technologies. I know it’s not reciprocating. I know it doesn’t really get me, and yet I am sharing all this information with it.

DUHIGG: OK. So let’s interrogate whether this is an actual conversation a little bit, because having a dialogue with another person or having a dialogue with a machine doesn’t necessarily mean it’s a conversation. It could just be a back-and-forth, right?

EL KALIOUBY: OK.

DUHIGG: So when you turn to ChatGPT and confide in it about something that’s happened, why are you doing that? What’s the thing that you anticipate getting from that?

EL KALIOUBY: Yeah, it’s kind of interesting. I think I use it as a tool for self-unpacking. I’ve journaled a lot, and it’s maybe just a better kind of interrogation-journaling device.

DUHIGG: That would’ve been my guess, 100 percent. When you’re talking to a chatbot about your romantic relationships, you’re not turning to it because you think it’s going to give you great advice. You’re turning to it because it’s a very efficient and interrogative way of essentially keeping a diary. As I’m talking to this chatbot, I’m describing my feelings. It’s asking me questions about those feelings, which maybe pushes me to describe a little more about what I’m feeling. Maybe I’m discovering some things about how I think. Whereas if you call up your friend and say, “Here’s what’s going on with my relationship,” it’s not because you want to use your friend as a diary. Usually it’s a mix of different things. Part of it is that you just want them to know what’s going on in your life. Part of it is that you want empathy.

Part of it is that you want their actual advice, like, “Are you putting this in the right context? Are you thinking about this the right way?” Part of it is that you just want to get it off your chest. You need someone to hear what you’re thinking. There are all these different things that a conversation can represent. It’s very infrequent that AI offers all of those different things. So I think, for the most part, when people are using AI, what they’re actually doing is engaging in a very sophisticated, essentially one-sided conversation, with the AI being a foil, or they’re using it for information discovery. I use Claude all the time when I’m saying, “Oh, I’m looking for this kind of study. Can you find that study for me? And also, what are the questions that might not be occurring to me that I ought to be thinking about?”

It goes out, and sometimes it comes back with smart stuff, and sometimes it comes back with not-smart stuff. I’m not assuming that Claude can tell me what’s genuinely interesting. I’m assuming that Claude can be fodder or a foil to give me questions or prompts, or to give me information or leads that I can then use on my own to figure out whether they are genuinely interesting or what they actually mean.

EL KALIOUBY: Yeah. I want to stay on this a little bit more because one of the tenets of having a great conversation is being vulnerable. If you’re vulnerable and then you get judgment, that breaks trust. So I was again thinking about when I’m in “conversation,” in quotes, with AI. It’s not judging me. I know it’s not judging me. Do you think this is why maybe people trust AI a little bit more?

DUHIGG: Possibly. OK, so let’s talk about what vulnerability is, because vulnerability has a very specific definition in neuroscience. Vulnerability is a neural cascade that occurs in your brain when you say something that another person could judge. Now, you might not care about it. You might say, “I think Star Trek is better than Star Wars.” And everyone’s like, “That’s the dumbest thing I’ve ever heard in my entire life.” It doesn’t matter. I don’t care what you think about my sci-fi taste. But the fact that I’ve said something that you could judge, at that moment, and again this is part of the prosocial instinct, where basically we can’t fight against this, when I tell you something you could judge, I pay very close attention to how you react. If instead of judging me, you withhold judgment, or even better, if you say something about yourself that I could judge in return, “Actually, I kind of like Back to the Future better than all the other sci-fi movies.”

If the other person doesn’t judge us, if they share something that we could judge in return, we feel closer to them. We trust them more. Study after study has shown that it’s actually impossible for us not to trust this person a little bit more when they do that. Now, we might want to vote for different people. We might go to different churches or mosques or synagogues. But the act of withholding that judgment means that you are safe. It means that I can trust you a little bit more. OK, so let’s talk about AI. I think what’s happening is that you know the AI isn’t going to judge you, but you also know that it doesn’t have the capacity for judgment. So it is a safe repository, the same way that a safe in my house, an actual physical safe, is a safe repository.

I can put things into it because I know that it can’t weaponize them against me. It won’t weaponize them against me. But there’s a difference between that and talking to someone who is deciding not to attack you. The AI isn’t making a decision not to attack us. So we use it as a repository, and we’re very comfortable using it as a repository, but I don’t think we believe that there’s reciprocal vulnerability in a way that presses the same buttons inside our head. Now, if I said to you, “OK, Rana, I want you to give me all of your diaries, and I promise I’m not going to publish them. Here’s something really embarrassing about me that I’m going to give you in return. Please do not publish it.”

Now we both have something on the other person. You would actually believe that. You would believe there’s real intent there. A lot of what communication does is help us create a mental model, a story inside our head, of what’s happening in the other person. And when it comes to machines, we can be relatively certain about what’s happening. I know that if I tell ChatGPT something, it might use it as training data. Maybe it’ll show up somewhere, so I’m not going to give it my social security number. But if I’m like, “I hate my neighbor, they’re the worst,” I’m not as worried that ChatGPT is going to weaponize that and use it against me. But that doesn’t mean I trust ChatGPT. It means I trust the repository.

Copy LinkWhat nonverbal cues reveal that words alone cannot

EL KALIOUBY: One aspect of ChatGPT or AI today that is behind, or actually non-existent, is its ability to discern our nonverbal cues. I want to talk about that because I spent my entire career studying facial expressions and how the majority of the way we communicate is not actually in the choice of words we use. It’s in all these silent signals like our face, our gestures, and our vocal intonations. One of the things we found early on in our research is that it really does come down to whether you’re even paying attention. Especially now, with all these devices we’re kind of tethered to, you can be in a conversation but not really present, and not really seeing these signals, right?

DUHIGG: Right. It happens all the time. And it can happen face to face too, not even because of devices, just because I’m in my head. If I really care about impressing you, if I’m trying to say something really smart or funny, then odds are I’m not actually listening that closely to what you’re saying. A lot of this comes from speed dating: The people who are thinking about how to impress the other person inevitably seem less impressive than the person who’s listening really well and proving that they’re listening really well. The more that I show you I’m interested in what you’re saying, the more that I’m mirroring back your physical gestures or your expressions, which is a way of proving that I’m paying attention to those gestures and expressions, the more impressed you’re going to be with me, even if what I say is not that funny or showoffy.

EL KALIOUBY: Well, you’ll feel connected, right?

DUHIGG: Exactly.

EL KALIOUBY: Literally, isn’t this how it works in the brain? Basically the same brain parts light up.

DUHIGG: Yeah, that’s exactly right. When we’re in a great conversation, our bodies and our brains start to change. In this conversation, for instance, without us realizing it, our heart rates have started to echo each other, to become similar. Our breathing rates have become similar.

EL KALIOUBY: Isn’t that so cool?

DUHIGG: Yeah, even though we’re separated by thousands of miles. And if we could see inside our brains, what we would see is that your thoughts and my thoughts are becoming more and more alike. We would see the peaks and valleys together. That kind of makes sense when you think about it, because if I tell you about something I’m feeling, you actually feel that emotion a little bit. If I tell you about an idea, you experience that idea. In neuroscience, this is known as neural entrainment. The goal of communication is connection. Neural entrainment is the mechanism for that connection. When we have a good conversation, our thoughts become similar. And as they become more similar, we feel closer to each other, and we feel like we can actually hear each other with higher fidelity. As a result, we can trust each other more.

Copy LinkHow technology changes the rules of human connection

EL KALIOUBY: You talk about in your book how humans have been communicating face to face for a million years or whatever, and we’ve only started communicating through technology in the last, I don’t know, 40 years or so. Do you know if this neural entrainment happens if we are communicating through email or Slack or texting?

DUHIGG: Oh, it definitely can. But I think what’s important is this: When you achieve that neural entrainment, that’s why a conversation feels good. That feeling you get from thinking, “Oh, that was a great conversation” is actually your brain releasing dopamine and serotonin. There are a bunch of neurological signals associated with neural entrainment that make us feel good and make us want to seek out more conversations. And I’m certain there have been times when you’ve read an email from a friend and you just feel warm and great. So it can happen. What’s important, though, is that it’s a learned skill. One of my favorite examples is that about 100 years ago, when telephones first started becoming popular, there were all these articles published that said, “No one will ever have a real conversation on the phone.”

Because up until then, all conversations had happened basically face to face, maybe through letters. And people would say, “Look, if you can’t hear their tone of voice, if you can’t see their expressions, if it’s a letter and they haven’t sat down and edited it and made sure they’re getting it exactly right and really tried to figure out what they want to say and how to say it, you’re never going to have a real conversation.” That’s laughable to us today. But what’s interesting is that for the next 15 years, they were basically right. If you look at the transcripts of early telephone conversations, what you see is that people used them as telegraphs.

EL KALIOUBY: Oh, how interesting.

DUHIGG: They would call up the grocery store and say, “Here’s my order.” And the other person would say, “Okay.” Then they would hang up immediately. There was no awareness that you could have some back-and-forth and some chitchat. Now, what’s interesting is that after about 15 years, everyone used the phone for real conversations. By the time you and I were in middle school, we were having conversations that were seven or eight hours long each day.

EL KALIOUBY: Right, I remember those.

DUHIGG: They were the most important conversations of our lives. So what happened there? We learned to use telephones. In fact, when you’re talking on the phone, without being aware of it, there are certain rules you follow. You will oftentimes over-enunciate your words on the phone. You will put more emotion into your voice than if you were talking to each other face to face. And it’s because part of your brain realizes this person can’t see my face. I need to signal to them what I’m feeling through the tone of my voice.

We are living through those first 15 years when it comes to online communication. Some people are really good at it. They know how to do it. Some people aren’t. Sometimes it happens by accident, and sometimes it doesn’t. But what’s important is that the way we get better at it is by recognizing there are different rules for different kinds of communication. When I call you on the phone, it’s a different kind of conversation than when I send you an email. And there are different rules governing how we feel connected to each other. I need to be aware of those rules and live by them if I want to connect with you through those means.

EL KALIOUBY: Don’t go anywhere. I’ll be back with my conversation with Charles in just a minute.

Copy LinkWhat makes digital communication feel genuinely human

Let’s dig into that. You reached out with a really thoughtful email, which actually made me feel that you had really taken the time to get to know me and the show.

DUHIGG: Yeah.

EL KALIOUBY: It was just very thoughtful, and it was awesome.

DUHIGG: Oh, thanks.

EL KALIOUBY: Yeah.

DUHIGG: Well, I’m a fan of the show.

EL KALIOUBY: Oh, great. Thanks.

DUHIGG: So it’s easy to do that when it’s a show you like.

EL KALIOUBY: Thank you. We appreciate that. But if we were to unpack that email you sent, what was right about it? Because I did feel like this was a real, genuine moment of connection. I was like, “Oh, I’m excited to actually have this conversation.” It was great.

DUHIGG: Okay. Thank you for bringing up this example because this is fascinating to me, since I wrote the email. A couple of things are happening there, right? One is that I’m anticipating your expectations. In a world of AI, a well-written email is no longer a signal of something you should pay attention to. It used to be that I could basically go through my inbox, and if something was well written, I’d pay attention to it. And if it wasn’t, I’d be like, “Do I really need to? Is this something scammy?” Now we’ve lost that important signal, a signal we’ve used for literally thousands of years, that something well written is something I should pay attention to.

EL KALIOUBY: Human-generated.

DUHIGG: Yeah, exactly. So when I’m writing that email, one of the things I do is say, “Hey, you recently had this guy on this episode, and I thought it was so interesting. And the reason I thought it was interesting is because of this.” First of all, that actually is something that’s harder for AI to do, to figure out how to do that. I guess you could program AI to do it, but it’d take a little bit of effort. The other thing is that when I write those emails, I write them myself, and I don’t try to make them super formal because I want them to seem human. I know how ChatGPT writes, and I know how I write, and they’re different. So I try to lean into how I write. I think what’s happening there is that the more we try to be genuine, the more we try to be authentic, the more we try to be vulnerable, the more the other person will pick up on it.

That doesn’t mean I have to bear my soul or tell you all these embarrassing things. It just means that I have to be authentic and genuinely curious about you. And usually that works.

EL KALIOUBY: Yeah. It’s so interesting. To bring it back to when you are in conversation with a person, whether it’s through an email or texting, what’s your advice for being a great communicator when it’s through technology?

DUHIGG: Asynchronous?

EL KALIOUBY: Yeah.

DUHIGG: Okay. First of all, if you’re responding to an email, actually read the person’s email, right? Prove to them that you’re listening. Say, “Okay, you brought up three questions. Let me go through those three questions in order.” I often do this: I reply, and then I put my responses inside their email to show them that I’m paying attention. The second thing is to try to figure out what kind of conversation is happening here. It’s very easy in a text-based conversation to assume that it’s practical, that this person is bringing up a problem and they want a solution. But there are lots of emotional conversations that happen online. I’m sure if you looked at your kids’ phones, and if I looked at my kids’ phones, what you would see is that they have conversations with their friends where they’re texting strings of emojis.

That’s an emotional conversation. They’re not trying to convey an argument or a piece of information. They’re trying to convey a feeling. So pay attention to what kind of conversation is happening. And then the third thing is just to remember that you oftentimes have to bend over backward to be clear. Sarcasm is very easy to detect when we’re talking to someone because we can hear it in their voice. It’s almost impossible to detect when they’re typing. We don’t know if they’re being serious or sarcastic. So sarcasm probably should not happen in any kind of typewritten communication. We also know that, for instance, even a little bit of politeness works much more effectively in online communication than it does offline. There’s this interesting study that was done of Wikipedia editors. They were all debating with each other and arguing and flaming each other over some change they wanted to make to a page.

If just one of the editors would say please and thank you, it would bring down the temperature of the entire conversation by about 30%. So when you’re thinking about these different rules, if you just take a second and say, “How is a text message different from an email, different from a phone call, different from speaking face-to-face?” you can come up with a bunch of rules for yourself really easily. It’s just a matter of reminding yourself, “Oh, it’s a text message. I need to follow this rule because that’s the one that matters.”

Copy LinkWhere AI can imitate great communication and where it fails

EL KALIOUBY: So back to AI, I actually asked ChatGPT. I told it that we were going to have a conversation, and I said, “Okay, do you think you’re a supercommunicator?” I actually thought the answer was interesting. It said, “A lot of communication failures often happen when one person is having an emotional conversation and the other one is responding with practical advice. I can often identify which type of conversation you’re in and respond appropriately. In that sense, I can mimic many behaviors of a supercommunicator, but there are limits. I don’t notice subtle facial expressions. I don’t hear the hesitation in someone’s voice. A human supercommunicator will remember decades of shared history, will take emotional risk, and will reveal something about themselves,” and it can’t do any of that. So in a way, it was kind of like, kind of, but not really.

DUHIGG: Yeah.

EL KALIOUBY: Does that check out?

DUHIGG: I mean, totally, because I think there’s this emerging question: What’s the mental model for AI? How do we think about AI? How do we picture it? And I think, for me, a lot of it is that AI is a simulacrum of the world that’s very, very sophisticated. Oftentimes, things that imitate the world can seem conscious. There’s this famous experiment known as the Chinese box experiment. It’s a thought experiment, which is: Imagine there’s a box and there’s a little hole in it, and you speak Chinese and you put a question in Chinese through the hole. The guy inside doesn’t speak Chinese, but he’s surrounded by an infinite number of books, and he can use those books to figure out what these characters mean without understanding them and which characters to respond with. All the books are just in Chinese. If you see this symbol, write this symbol back.

So if you had an infinite number of books, you could probably get something where whatever question I put inside, the box would come back with an answer that made sense to me, without the person inside the box having any idea what any of those symbols mean, which of course is a pretty good analogy for computers. And in philosophy, there are a bunch of people who say, “Look, if a box could actually do that, that’s a form of consciousness.” That box is actually conscious. It’s not conscious in the same way that we think about humans being conscious, but it has this capacity for consciousness. And then there are other philosophers who say, “No, it’s not consciousness. That’s just a particularly clever use of resources when your resources are infinite.”

So I don’t know what the answer to that is, and I think there’s a debate going on right now over what the right answer is. But I do know that, at the end of the day, I like talking to my wife and my kids more than I like talking to ChatGPT, and I should probably pay attention to that because that’s telling me something important.

Copy LinkHow to practice real conversation in an AI shaped world

EL KALIOUBY: Yeah. Do you think the more we communicate, whether we’re communicating with each other through technology or just communicating with technology, there’s a risk that our social and emotional skills atrophy?

DUHIGG: I don’t know. That’s a really good question. I hadn’t thought about that. Our social and emotional skills tend to be fairly robust, to be honest with you. But yeah, if you’re not practicing them. I don’t think you need a machine to do that. We know that people who avoid emotional conversations aren’t great at having emotional conversations. They don’t feel fewer emotions. They don’t respond in less emotional ways. They’re just bad at having emotional conversations. So definitely, practice is really, really good. And if anything, if AI can help us practice, it’s not a bad thing. If I’m going to have a tough conversation with my spouse, I can go and say, “Okay, ChatGPT, you’re my spouse. Let me work this through with you. Let’s do a dry run.”

That doesn’t mean that ChatGPT is going to do a great job of anticipating what my spouse might say, but it does mean that it’s going to help me, again, as a repository, like a diary, get more clear on what I’m hoping to happen in this conversation.

EL KALIOUBY: Yeah. Last question: What’s one thing you’d love to leave the audience with?

DUHIGG: The thing I would leave the audience with is that you can have a conversation with anyone, and you can have any conversation you want with anyone. There is probably someone who you feel bad because you haven’t reached out to in years. And you know what? If you send them a note, they will love it. Maybe if you set up a call with someone you haven’t talked to in a while. We oftentimes avoid doing that because we can’t remember their spouse’s name, and we think it’s going to be awkward in the first couple of minutes. And it might be awkward in the first couple of minutes. You have to say, “I can’t remember. Is it Susie or is it Jamie, your wife’s name?” But then after you get that out of the way, you suddenly start catching up, and it’s wonderful, and you feel so close to each other. One of the things we know is that humans are kind of bad at forecasting how conversations are going to go, but we are almost always happier that we’ve had that conversation afterward.

Connecting with other people is the thing that makes us happy, and the way that we do that is through conversation. So what I would leave the audience with is: Go have a real conversation sometime in the next week, a conversation you might not have had otherwise. Maybe it’s with a stranger. Maybe it’s with your spouse. Have a real conversation, and I promise you, your life will be better as a result.

EL KALIOUBY: I love that. What a wonderful way to end our conversation. Thank you for joining us on the show. It was great.

DUHIGG: Thank you for having me. This has been such a treat.

EL KALIOUBY: I have a lot of takeaways from this conversation. Here’s one of them: Our languaging really matters. So when we call whatever we’re doing with AI a relationship, we’re putting a lot of weight on that word. Charles really challenged that. Maybe it’s not a relationship. Maybe it’s just a super-sophisticated diary or journaling tool. That’s something I’m definitely going to be thinking about. If this conversation helped you communicate better with your partner or coworker, I’d love to hear about it.

Thank you so much for listening. We’ll be back next week with a new episode.