#484 — Artificial Intimacy

You're listening to Making Sense with Sam Harris. This is the free version of the podcast, so you'll only hear the first part of today's conversation. If you want the full episode and every episode, you can subscribe@samharris.org There are no ads on this show. It runs entirely on subscriber support. If you enjoy what we're doing here and find it valuable, please consider subscribing today.

0:02

I am here with Paul Bloom. Paul, it's great to see you again.

0:24

It's always great to see you, Sam. It's always too long between our conversations.

0:27

Yeah, no, I haven't checked to see how long it's been, but I'm sure there's a lot that we can talk about here. You're one of the short list of people who I know I'm gonna have a good conversation with, no matter what's happening and no matter how little prep either of us do. It's like a phone call that I know I'm gonna enjoy.

0:30

I appreciate it. I've been avoiding you since the Epstein files, but I decided to nonetheless push through despite your prominence in those documents.

0:45

Yeah, well, we were talking offline about the World Cup. Did I glean correctly that you're not as obsessed with it as I am? Are you not a many hour a day watcher of soccer at this point?

0:54

I'm not. My sons came to visit, and one of my sons came to visit, and he's much more of a sort of manly than me, so we watched sports together and it's okay. I don't want to, you know, I don't want to get into this sort of culture war thing and get too many people to hate me. But soccer is not the most enjoyable sport to watch.

1:07

You didn't play soccer, I take it as a. In high school or.

1:24

No, hockey. Like, we played street hockey, street hockey in the summer and hockey hockey in the winter. But I've never. I've kicked around the ball. But are you. Have you?

1:28

Well, yeah, that was the only team sport I played, so I kind of, sort of feel it from the inside. But I am the genius who scheduled this podcast recording during the Spain Portugal elimination match. So I'm the first person I'm going to fire.

1:36

Do you have a little screen, like, in front of you and you're kind

1:50

of like, if I see a lot of. Yeah, I'll be watching the game for the next 90 minutes, so I hope you have a lot to say.

1:52

Yeah, yeah. It was long pauses.

1:57

Yeah. So now you. Speaking of Bandwidth and claims on your time, you recently became a father yet again. Is that. How long ago did that happen?

1:59

Yet again. I have two older sons, terrific boys, 27 and 30. But I now am the proud father of a four month and one week old. Nice names. And this adorable little girl. I just spent a whole morning with her. We went to the park, we had a bottle, you know, Good times.

2:08

Awesome. Awesome. Well, that's not. If memory serves, that's not yet the funnest zone of parenthood, but it's still amazing. That's great.

2:27

Do you have a. You have an optimal age that you think of?

2:35

Well, it's just. It's probably not four and a half months.

2:39

No. No. I don't know.

2:42

How is sleep?

2:44

She's sleeping through the night.

2:44

Is sleep happening?

2:45

She's sleeping through the night. We have.

2:46

Oh, that's amazing.

2:47

I don't know. I know you had your daughter so a while ago, but we have something called a snooze.

2:48

Uh huh. Is that, is that some kind of swaddle monstrosity?

2:53

The bassinet. You swaddle up the kid, you clip them into the snoo and then when they cry, it vibrates.

2:56

Right.

3:01

And then it basically vibrations knock them unconscious and then they. And then they sleep. So our sleep problems, when she grew old enough to be in the snooze, she's actually sleeping. Great. But now we have to. Because unfortunately the world does not contain snooze and adults have to sleep without snooze. We now have to get her out of it, and that's going to be a bit of a challenge.

3:01

All right, well, this is our first ad on the Making sense podcast for Snoo. Is that the brand SNU.

3:19

Highly recommend it. I get 10% from all purposes recommended by developmental psychologists. I highly recommend the snhu.

3:24

Awesome. Okay, so how are you feeling about the world your children will eventually inherit at this point? What's top of mind for you about the unraveling of culture or anything else that. Are you optimistic, pessimistic, averting your eyes?

3:31

That's a good question. You know, whenever you and I talk, we end up talking about Trump and the extreme train wreck of American politics. But I think things will go back to normal. I think, I think that post Trump, the world will be wacky in its normal ways, but we'll just have a normal president. You know, it'll be along the lines of a Biden or a Bush or a Clinton or, you know, and this

3:48

is the happy talk. This is the Happy talk I'm getting from the relative safety of Toronto. Is that what's happening?

4:13

Yeah, right. Oh, feels so good to be out of the reach of Trump. At least I've been distracted by things and he's not talking about invading us anymore anyway, so I think we're going to get. Things will get better. But I am. I don't know what to think about AI. You know, I remember listening to you, you called the warning about AI before everyone else was talking about it, actually had some sort of big talk. You just pointed out that we tend not to worry so much about it. I think you get this line because it's so cool. Unlike other sort of crises like global pandemics and climate change and nuclear war, there's something about AI which sounds science fictiony and kind of neat, but now that we're in an age of these chatbots which have, and I think they have extraordinary abilities way beyond I would have ever predicted, I worry. I know. I'm curious whether you share this take. And my take is that it's very, very high variance. On the one hand, it could not affect the world very much. It could be what we have now is kind of the way things are going to go. It's replace Google and that'll be fine.

4:19

That seems very low probability to me

5:19

that I think so. So the positive thing is it leads us into some sort of post scarcity world, you know. Yeah. A year from now it solves Alzheimer's. Two years from now, because that's limitless energy. Three years it brings upon a peaceful world government and teaches us about world peace. The bad side is if it doesn't kill us all, it just takes all of our jobs, you know, takes all of the jobs, you know, civil unrest, war, chaos, destruction. And you know, if, if before AI, I would think, well, you know, my daughter, try to get her to go to college and, and become some sort of professional and maybe, maybe in the family business she becomes a scholar of some sort of. I don't know, I don't know what jobs will be there. What do you think?

5:21

Well, given that your daughter is only four and a half months old, I think there really is no way to guess what the world is going to be like in 17 and a half years. I mean, it's just whether college is going to be a thing.

6:02

But it's harder, right? But it's harder now, so my boys, much harder. 1990s. Well, okay, you set up college funds, you have just a general direction and you never know. But that's the general idea, but here I feel it's different.

6:18

Yeah, no, there's an acceleration of cultural change. There's no question. So let's kind of move through those various possibilities, but before we kind of follow each branch. What has most surprised you about AI in recent years? I mean, it sounds like you didn't foresee somehow, like many of us didn't foresee just how quick the changes would be. And I mean, for me, one thing that's been surprising is just how quickly we accommodate to change and are no longer surprised by it. It's like I. For me, the landmark here, or the fake landmark, I mean, the mirage of sorts was expecting the Turing Test would be a thing. And when the Turing Test was passed, and passed so well that it was then failed because, you know, obviously this couldn't be a person who's given me exactly 400 words on any topic I want within three seconds. I mean, it was just amazing to have thought that passing the Turing Test would be this culturally salient, psychologically even destabilizing encounter with the uncanny. And yet, no, it's just like having a calculator or something. I mean, there's just nothing. Of course, nothing happened when the Turing Test was passed.

6:31

It's the Louis CK routine, right? Where he's talking about being on an airplane and people complaining that peanuts aren't being delivered quick enough and there's a bit of turbulence and the WI fi is unsteady and says, you're flying in the sky like a God. Why don't you appreciate it more? And now people are saying, gosh, I asked Chad a question and it got a little bit wrong. What the hell? And, well, and I'm thinking you're talking to a super intelligent machine that can have deep riching conversations with any aspect of your life that could give you personal, that could mimic a therapist, a companion. And, you know, why don't you marvel at it more? So, you know, I'm a psychologist. I wrote. I wrote. I wrote a book called Psych, which reviewed all of psychology, and it got published in 2023. And the one part I like to change is I talk a little bit about AI. Isn't AI impressive? But of course, it can't have a conversation like people. It can't. You know, there's all these things, basic things it can do. And then chat came out.

7:46

Yeah.

8:41

And then. And did it all. And then it got better and better and better and better. And you're right that one of the weird thing is how, what, how quickly we've got used to it, that, that, that really I can pick up my phone, set it to voice mode and have a conversation with an artificial being. That is, I don't have to talk to it in code. I don't have to use a programming language. I can say I get life advice from it. I get there are people who, and we could talk about this. I found it very interesting. People who have friends, relationships, partners, who are artificial beings. And although I'm not tempted that way, it's not madness because they sound like people and we created them.

8:42

Yeah, I noticed that I'm not engaging it in voice mode and I think I'm not doing that for a reason. I think this is something you and I spoke about, I think several times and we wrote an op ed about for the New York Times in response to the series Westworld where, I mean, we argued that Westworld was functionally impossible because, you know, it would, you know, a theme park where you get to, you know, rape and kill Dolores. I mean, this is going to act like a bug light for psychopaths. I mean, there's no way a psychologically normal person can go spend a weekend like that and then come home to wife and family and think normal thoughts about themself or have their, the people who know what they were up to think normal thoughts about them. I mean, you just, people would begin to treat you as deviant if you wanted to spend your time that way. Because this is just the consequence of getting out of the uncanny valley and being in the presence of a true human simulacrum and acting like a psychopath there. So I feel like there's something worth considering around this issue of losing sight of whether or not it's an even interesting question to wonder whether AI is conscious. Many people are speculating about AI consciousness now and, you know, asking, you know, what would what, how will we know? And my feeling has been for the longest time and certainly ever since you and I wrote that op ed together, is that we are going to just build something that's going to seem so incredibly conscious that most people are going to lose touch with whether it's even an interesting doubt to have.

9:23

I think that's right. Once it looks like a person and sounds like a person, you know, right now you could talk to it. I don't think we're that far away from having a zoom conversation with it where, you know, as you're looking at me and I'm looking at you, we could just be AIs doing it. And I think put aside Once it has a body, then it's going to be irresistible to think of it as human. We haven't been wired up, we've been wired up to just take things that look like people and treat them as people. So that's half of it. Half of it might be that even if it's not even close to conscious, we will, we can't help but treat it as such. And that's going to open up a whole lot of things. The other half of it. And this is actually, there's a. A wonderful paper by Maddie Wilkes, professor at University of Edinburgh, and she was my postdoc for a while. We're actually writing something up on this. But she makes the following argument which is very interesting. A lot of people seem to think that, well, if we were to discover that as conscious, we would then treat it kindly. We can't enslave it, it has to have rights, we can't torture it and so on. She argues this is wrong because actually we know that non human animals feel pain, suffer, have, have some degree of cognitive sophistication. Yeah, we torture and eat them all the time. And so it shows that we could think of something as conscious and yet treat it horribly. And so that's not what's going to save the AIs. The consciousness isn't what it's going to save. Right.

10:55

Except I mean, presumably we're less and less comfortable with that not perfectly expanded circle of moral concern. And the circle has definitely expanded. It expands first to things that are most like us. Right. And so like obviously mistreating chimpanzees is something that exactly nobody is for at this point. And the depth of our concern diminishes from there. But in the presence of AI that is smarter than us, that is more articulate than us, that will be able to argue better than we can around the ethics of all of this and which at this point will seem conscious to us and may even claim to be conscious. I feel like it's suddenly going to be. I mean, they'll be standing right in the center of the circle with us insofar as we believe that there's something that it's like to be them.

12:12

I think that's right, but I don't think that the linguistic and cognitive skills are enough.

13:05

So.

13:12

You know, my favorite movie when thinking about AI is her. You've seen it? Yeah, came out, I think 2013. Scarlett Johansson plays Samantha. This which our main character quickly falls in love with. And we do too watching it, but she has a voice she has emotion, she talks like a person. Once we have that, once that's fully instantiated, plus the intelligence and rationality you're talking about. Yeah. It'll be irresistible.

13:13

Well, we do. So crucially, her was disem. The Scarlett Johansson character is disembodied. Right. So I feel like we're already. Anyone who's dealing with an AI companion now is essentially living in the her universe. Right. I mean, again, I'm not spending much time with the voice, but I think didn't in the launch of ChatGPT version whatever, didn't they steal Scarlett Johansson's voice and get sued?

13:38

They asked for her permission, she said no, and in true Hollywood style, took somebody who sounded exactly like her. And then she said, I don't know how that came about, but she should have won the. Because the voice that it originally had sounded exactly like her. So you said a couple of times you don't interact with the voice and that just might be because you tend to use a computer or whatever. But do you have a worry?

14:04

I do, actually, I do. I mean, so I haven't thought much about this, but. And it is mostly just a use case issue for me. I mean, I'm so text based and the utility of AI for me is to do research, so really that needs to be text rather than somebody talking to me. But when I've played around with the voice, I notice that a different kind of memetic psychology comes online. And I mean, this is what gave me my concern. My personal concern was first peaked when, hey, you must have seen some of these kind of viral videos around the failure modes that still exist for these LLMs where if you ask any one of them, I think this is probably still true at the moment we're talking. If you ask any one of them, give me a number below 1000 that has an A in it. None of them can do this. I mean, there is no number apparently below 1000 that has an A in it. None of them. They all fail this, but they all confidently give you numbers that don't have A's, right? So they'll all say, oh, yeah, well, so 70. 70's got an A. And you say, no, no, 70 does not have an A. And they say, oh, you're right, right, 70. So they'll endlessly apologize and endlessly answer confidently with new numbers. And this was very bizarre behavior. But I was sitting with my daughter, showed me this phenomenon and we were in chat mode with one of these bots, I think it was Claude, and effectively making fun of it. Right. And as it was failing, I detected in its tone a kind of a subtle loss of patience. Like there's something shifted in its tone. I mean, we were just laughing at its expense endlessly. And I was getting, like, somehow uncomfortable with this because it was a kind of social shaming that you would never do with a person with a clear conscience. But the AI, it seemed to me subtly started to react to it. And honestly, like, the hair on the back of my neck stood up. I mean, I just felt like, okay, we're pissing off the robots now and they're going to. They're going to remember this and it's

14:27

going to matter at some point. Check your bank account, Sam. See what happened to it.

16:36

Exactly.

16:40

As you fuck with me.

16:41

Yeah, that'll be the least of my problems.

16:42

But, yeah, yeah, I mean, part of the issue, which is engaged in a lot of people, is the sycophancy, the sucking up, which is, I think part of why you get this effect, which is it's very hard for it not to want to give you what you want. So you ask for something and it just can't say, I gotta disappoint you, man. There's no. There's nothing. But. Doesn't have a letter A.

16:43

Well, no, but I did switch that off. I think they still fail.

17:01

You went in.

17:04

I don't think that's the variable there. But yeah, my version. I remember I did lobotomize the sycophancy of at least one of the LLMs I use, and I did that and

17:04

then I changed it back again after a month.

17:17

Oh, yeah. Anyway, I saw that from your talk.

17:19

I don't get as much love as you. Maybe I just kind of, you know, I kind of. I kind of liked it. Telling me how smart I was and how good my questions were.

17:21

But, I mean, I noticed that I feel the need to be. I mean, this has widely been remarked upon. Like virtually everyone else, I say things like, please show me this. Right. So dropping those kind of courtesies seems wrong. Not because. And I really have zero concern that the LLMs are currently conscious, but again, it seems like a violation of social norms that I don't want to encourage in myself. And at a certain point, it's just all of our adherence to these norms are going to be totally hijacked by the technology and we're just going to feel like we're talking to persons.

17:30

Yeah. And then there are an increasing number of people. I think it was like some study finds some large proportion, maybe a third of teenagers have confided in LLMs and asked them advice for intimate personal matters. I think the young people are going to do it more and more. And I know friends of mine know adults who have had problems in their marriages and they, they, you know, two in the morning, they're, they're talking to the Claude, they're talking to chat and they're just saying, and she said this and what should I do and everything. And I don't want to, I don't want to mock it, I don't want to deride it. I think people find solace in it. I think that unlike social media, which drives people to its batty extremes, often AI has sort of a normalizing function.

18:10

Yeah, yeah.

18:50

I was asking it once questions about the, the Holocaust, and I said, you know, could this have happened in any other country at the same time? What do you think? Could, could have been in, in Austria, what if, you know, and I, and, and then it sort of sternly reminded me, you know, this really did happen.

18:51

Yeah.

19:04

You know, I will have no truck of Holocaust denial and I kind of, okay, dude, I got it. But it doesn't indulge in conspiracy theories. It's, for the most part it's a normalizing thing. I think there's a lot, I, I, I am not a booster. I think that there are dangerous, but I think these conversations that people have are to a large extent good for them and of course, alleviate loneliness for a lot of them.

19:04

Yeah. So you just invoked kind of two different use cases there. One is kind of the advice that AI can give for relationships, like you say. I just had this argument with my wife and this is what happened. And I haven't done much of this, but I know people who have really, who really feel that certain difficult conversations have been kind of expertly navigated in advance by consulting AI and then they've just had a much better go of it in real life with the person who has their own user interface issues. But you're talking now about actually as a remedy for loneliness and social isolation too. So let's talk about that because I know you recently gave a TED talk about that and it's not out yet though, right? It's not available.

19:25

It's not, it's coming out in a couple of months.

20:09

Okay. But I saw it and I recommend it. So let's back up for a second and just talk about the problem of loneliness as you see it and what it amounts to and what a proper cure for it beyond just being surrounded by people who Love you and to whom you matter. If you can instantly switch that on, obviously that is a cure. But I know you're a fan of Rebecca Goldstein's framing of mattering here, so maybe we could start there.

20:11

Yeah. She has this wonderful book where she talks about the importance of mattering to another person. I think that's a nice framework with which to see loneliness. Obviously, loneliness isn't just a matter of being alone. Some people do very well in solitude. And some people, and often alternatively could be surrounded by people and feel tremendously lonely. Yeah, it's not even necessarily a matter of being. Being loved. I think it's a matter of mattering to people, having value, having weight, being taken seriously.

20:41

And I think she describes that as being deserving of attention. I think that's.

21:09

That's right. Yeah, that's right. And that sort of. To jump ahead is that regardless of what, of what impression we get, that's actually one thing AI can't do for us. Really. Because if I'm talking to you, which means that you've taken your time to talk to me, giving up, accident, watching the soccer game, you're spending your time, you're. These precious hours.

21:14

The truth is, Paul, it was time I did not want you. Yeah, an accident that worked to your advantage.

21:35

And also at three in the morning, I can knock on your door and say, sam, I want to talk. And maybe for me, maybe not. But basically, there's a value in a person making time for you. And at the micro level, you know, you say something funny, I laugh, you've affected me in some way. You know, I disagree. I get angry. I, you know, I say, wow, you made a great point. And we're resonating to each other. And I don't think an AI really has any of that. I think, I think an AI is there. It's there to listen to you for the same reason your toaster, you know, toast your bread for you at three in the morning. It's just. It's a machine. That's what it does. It doesn't. There's no choice to it.

21:44

You don't think we're going to lose sight of that? Because what you're basically saying is that the value of attention on some level is in its scarcity. I mean, the value of really anything is in its scarcity. And the infinite supply of AI attention is going to undermine this sense that the attention matters when it's aimed at you. Don't you think that's yet another thing we might lose sight of?

22:28

I think we're probably wired up for that scarcity to matter to us. But we might lose sight of the fact that AIs work differently. We might delude ourselves or be deluded into thinking that sort of scarcity applies. There's a scene in the movie Her. I'm going to forget the exact numbers where our main character is realizing there's something gone wrong with his relation to Chatbot. Everybody remembers the scene and he sits down, he says, how many people are you talking to right now as you talk to me? And she says in her beautiful voice, why do you ask that? I said I want to know. And she says something like 3812. And then he says, how many people are you in love with? And he says 645. And he's like shocked. And. But there's of course, no, no limit. They are of course finite beings as a matter of principle, but practically they're not finite at all. And this intimate conversation I'm having with Chad, at the same time, it's having a million other conversation. I have no, no value to it, but so, so demattering. You're right. It. It. It's important to us, but it may be an illusion. But let me give the case for AIs. I think there's a case for and a case against it. And I think it's. I think it's stupid to ignore either side of it. Loneliness is terrible. Loneliness is just. It's just terrible. You know, I'm not gonna give you made up statistics about it equal to pack a day of cigarettes or being obese, whatever, but it messes up your body, but it messes up your soul. It, it is a terrible form of suffering. It's why. It's one reason why solitary confinement is. Should be viewed as a form of torture. It is just to be. And, and just. You know, you think of how nice it is to be people you love who care about you, respect you, who, who matter to you and you matter to them. And then imagine taking that all away, you know, I mean, I don't want to get all incel. Like but. But I'd much rather lose all my money than lose all my social connection to the people I love. I'd much rather be. Be poor with people who think I'm terrific and love me than be a trillionaire and with nothing but sycophantic yes man and no real contact. Then you add to this that there are a lot of people. I'm thinking now of elderly people who are in institutions who have no. They Go day and night with nobody to talk to? No. They've either outlived older family and friends or they don't want to see them. Maybe they suffer from dementia or personality disorder, making honest to God, difficult to talk to. And there's not enough money in the world to pay people to, To. To spend time. So if, if something like some future version of Chad or Claude or Gemini could come in and, and ease the pain of the loneliness of these people, and I think it'd be a godsend. I think it'd be wonderful. It'd be a, A cure for a terrible disease. So that's, that's the, the plus side. What do you, what do you think

22:52

the plus side was as a measure of how far we've come with AI? I remember it's not that long ago when the product for that situation. What is the thing you're going to deliver to the old age home to keep some isolated elderly person company? It was something like a. It was effectively a stuffed animal. I think it was like a baby seal or something that just did nothing but, like, blink its eyes and vibrate or. I mean, it was just the most. And I mean, now you literally have an omniscient conversationalist, you know, who occasionally hallucinates that you can turn loose ad infinitum in that situation. It's quite amazing.

25:35

That's great. And, you know, and they have to be. The interface has to be done so that elderly people, maybe some of diminished capacities could, could use it. But, you know, imagine.

26:15

Well, I think you're right about the Zoom Call. The Zoom Call's gotta be like a month away.

26:24

Members can hear the full conversation by subscribing@samharris.org Subscribers get a private RSS feed you can use with your favorite podcast player.

26:29

Imagine a child or a teenager who spends their days with a chatbot. And it's always, you know, it never tries to wait for you to finish telling your story so I can tell their story. Never gets bored. It never needs an apology. It never says, hey, that was inappropriate. And I think that can have a real corrosive effect. Oldest time in dreamland. All this time conversing to chatbots could leave you unable to interact with real people.

26:37