WEBVTT - Ep123 "Will AI cure loneliness?" with Paul Bloom 0:00:05.120 --> 0:00:09.680 Will AI cure loneliness? This is actually a tough question, 0:00:09.760 --> 0:00:13.240 and it's quite nuanced because, on the one hand, AI 0:00:13.320 --> 0:00:19.079 companions are increasingly amazing at soothing our sense of isolation. 0:00:19.239 --> 0:00:21.640 They can be here for us twenty four to seven, 0:00:22.040 --> 0:00:24.520 ready to listen and to help. But does that come 0:00:24.640 --> 0:00:29.960 at a cost, because more than simply pain, loneliness is 0:00:30.000 --> 0:00:36.200 a biological signal that pushes our behavior towards improving ourselves socially. Also, 0:00:36.400 --> 0:00:40.760 does everyone have the same need for curing their loneliness? 0:00:40.800 --> 0:00:44.720 In other words, will having an AI relationship to cure 0:00:44.840 --> 0:00:49.120 loneliness mess up our teenagers or should we be thinking 0:00:49.159 --> 0:00:53.920 about it as providing a critical lifeline to our seniors. 0:00:54.400 --> 0:00:56.760 So in this episode, we'll be joined by my friend 0:00:56.760 --> 0:01:03.880 and colleague Paul Bloom to discuss loneliness. Welcome to Inner 0:01:03.960 --> 0:01:07.559 Cosmos with me David Eagleman. I'm a neuroscientist and author 0:01:07.600 --> 0:01:11.000 at Stanford and in these episodes we sail deeply into 0:01:11.040 --> 0:01:15.319 our three pound universe to understand why and how our lives. 0:01:15.000 --> 0:01:15.880 Look the way they do. 0:01:33.400 --> 0:01:36.800 Today's episode is about loneliness, which is one of the 0:01:36.840 --> 0:01:41.040 most painful experiences that humans face. Many of us think 0:01:41.040 --> 0:01:44.759 of loneliness as a kind of passing sadness, but from 0:01:44.800 --> 0:01:49.720 a biological perspective, it's a kind of internal alarm system, 0:01:49.920 --> 0:01:53.560 like hunger or thirst or pain. For most of our 0:01:53.640 --> 0:01:58.160 evolutionary history, being cut off from others was more than 0:01:58.200 --> 0:02:03.200 a sting. It was dangerous. Isolation meant vulnerability. To survive, 0:02:03.680 --> 0:02:07.400 we needed each other, and so through an evolutionary lens, 0:02:07.920 --> 0:02:12.640 the emotion of loneliness is like a pain that pushes 0:02:12.720 --> 0:02:16.639 us towards something, in this case, back into social connection. 0:02:16.880 --> 0:02:20.519 But here's the question we face today. What happens if 0:02:20.560 --> 0:02:25.399 we mute that signal? What happens if technology, in its 0:02:25.520 --> 0:02:31.000 growing sophistication, takes away this sting of loneliness by offering 0:02:31.360 --> 0:02:36.679 a digital substitute. We've all seen how AI can mimic empathy, 0:02:36.840 --> 0:02:39.959 as I've talked about in several previous episodes, and AI 0:02:40.200 --> 0:02:44.880 companion never gets tired or distracted, they don't lose patience, 0:02:44.919 --> 0:02:48.000 they don't get angry, they don't get snarky, and they're 0:02:48.040 --> 0:02:51.360 there for you twenty four to seven with endless attention, 0:02:52.040 --> 0:02:55.680 and they're good. One study found if you compare doctor's 0:02:55.800 --> 0:03:00.880 responses to patients against AI's responses to patients, people rate 0:03:00.919 --> 0:03:05.639 the AI responses as more empathic. Now, just imagine someone 0:03:05.639 --> 0:03:11.120 who's isolated, who feels invisible. The kind of attentiveness AI 0:03:11.200 --> 0:03:14.400 gives can feel like a salvation, and I think that 0:03:14.440 --> 0:03:18.440 it can actually be salvation. One study out of Stanford 0:03:18.480 --> 0:03:23.280 showed that having a companion bought for young people mitigated loneliness, 0:03:23.639 --> 0:03:27.440 and three percent of the respondents said it stopped their 0:03:27.520 --> 0:03:32.040 suicidal ideation. But there's a tension here, and my colleague 0:03:32.040 --> 0:03:34.400 Paul Bloom recently wrote about it in The New Yorker. 0:03:34.520 --> 0:03:37.000 And that's back to this issue that the pain of 0:03:37.200 --> 0:03:42.040 loneliness is a forcing function for growth, That pain pushes 0:03:42.120 --> 0:03:45.839 us to do the hard work of making ourselves understood, 0:03:46.400 --> 0:03:50.880 or of building bridges to others, or of repairing broken bonds. 0:03:51.440 --> 0:03:56.680 If AI relationship makes loneliness disappear, do we also lose 0:03:57.200 --> 0:04:02.360 the social effort, the striving that loneliness once demanded of us. 0:04:02.560 --> 0:04:05.360 So at this fast moving moment in history, we're stuck 0:04:05.400 --> 0:04:10.800 with this central question. Will artificial companions rescue us from 0:04:10.840 --> 0:04:15.600 needless suffering or will they erode something essential about the 0:04:15.720 --> 0:04:19.960 meaning of this biological signal? Will they soften the edges 0:04:20.040 --> 0:04:22.960 of life? In a way that helps us or in 0:04:23.000 --> 0:04:25.960 a way that hollows us out. I think we should 0:04:25.960 --> 0:04:29.440 all avoid easy answers to this question, and happily, our 0:04:29.480 --> 0:04:31.960 guest today is a colleague who's been thinking about these 0:04:32.040 --> 0:04:36.240 questions with subtlety and depth. Paul Bloom is a professor 0:04:36.240 --> 0:04:39.359 of psychology at the University of Toronto and a professor 0:04:39.360 --> 0:04:43.120 emeritus at Yale. He's the author of psych The Story 0:04:43.160 --> 0:04:46.440 of the Human Mind, and many other influential books, and 0:04:46.600 --> 0:04:49.000 in his recent essay for The New Yorker, which I'll 0:04:49.000 --> 0:04:52.440 link to in the show notes, he explores the promise 0:04:52.600 --> 0:04:56.240 and peril of AI as a cure for loneliness. So 0:04:56.360 --> 0:05:00.800 let's step into this conversation at the crossroads of psychology, technology, 0:05:01.080 --> 0:05:08.080 and what it means to be a human. So, in 0:05:08.120 --> 0:05:12.159 your recent New Yorker article, you describe loneliness as a 0:05:12.360 --> 0:05:16.200 toothache for the soul. So let's just dive into loneliness 0:05:16.200 --> 0:05:18.760 for a moment. Talk about what it is and what 0:05:18.839 --> 0:05:20.279 people experience there. 0:05:20.640 --> 0:05:23.279 Well, let's start off with the obvious, which is loneliness 0:05:23.400 --> 0:05:27.799 is awful. Loneliness is terrible. It's I know everybody listening 0:05:27.839 --> 0:05:30.240 to this will have experienced some degree of loneliness. Some 0:05:30.279 --> 0:05:33.200 people will have experienced a transient loneliness. You know, you're 0:05:33.200 --> 0:05:36.120 a weigh on a trip, you know, no friends around. 0:05:36.760 --> 0:05:38.560 Some you know, but it goes away. You know people 0:05:38.560 --> 0:05:40.280 will love you. On the other side of things, some 0:05:40.320 --> 0:05:44.320 people will experience real loneliness for long periods of time. 0:05:44.440 --> 0:05:47.280 You're more likely to be lonely if you're old, maybe 0:05:47.320 --> 0:05:51.160 everybody you know has died, maybe you're strange from your family. 0:05:51.400 --> 0:05:53.839 Some studies find that past age like sixty, but half 0:05:53.839 --> 0:05:58.240 of people say that that they're lonely. So loneliness is awful. 0:05:58.560 --> 0:06:01.400 It is in as simple as sense. You know, a 0:06:01.520 --> 0:06:04.560 lack of contact with other people, but you can go 0:06:04.680 --> 0:06:08.800 deep here. You could be lonely, but with people. I 0:06:08.839 --> 0:06:11.280 have sometimes lonely, the loneliest times of my life when 0:06:11.279 --> 0:06:13.400 I was surrounded by people and I didn't feel or 0:06:13.480 --> 0:06:15.440 was the connection. I didn't feel it was love. You 0:06:15.480 --> 0:06:17.760 could sometimes even be loved and feel lonely because you 0:06:17.760 --> 0:06:20.800 feel people don't understand you, that there's no connection. But 0:06:21.600 --> 0:06:24.839 it's a serious problem. It's a serious form of human suffering. 0:06:25.160 --> 0:06:27.800 And what are you supposed to the evolutionary purpose of 0:06:27.839 --> 0:06:28.720 loneliness is. 0:06:29.240 --> 0:06:31.239 That's a good question. I mean, you know, we should 0:06:31.240 --> 0:06:34.080 think like we should think like evolutionary biologists, and when 0:06:34.080 --> 0:06:36.800 we talk about something, even something awful, we should say, 0:06:36.800 --> 0:06:40.280 what's it there for? And I think that just about 0:06:40.320 --> 0:06:43.360 all of these experiences, these feelings are there for a reason. 0:06:43.640 --> 0:06:45.919 You know, hunger's there to motivate you to seek out food, 0:06:46.520 --> 0:06:49.279 boredom's there to motivate you to find something interesting in 0:06:49.279 --> 0:06:53.600 your life. And loneliness is there, I think, to motivate 0:06:53.600 --> 0:06:57.159 you to seek out human contact. And once you have 0:06:57.279 --> 0:07:00.840 that human contact, to work at it, you know, to 0:07:00.880 --> 0:07:04.800 try to be interested in another person, try to connect 0:07:04.800 --> 0:07:08.560 with them. One way to answer your question is what 0:07:08.600 --> 0:07:10.680 would it be like if we go into your brain 0:07:11.240 --> 0:07:13.560 and make it so you never feel lonely. You might 0:07:13.600 --> 0:07:16.200 say that that's great, loneliness is awful. My life would 0:07:16.200 --> 0:07:18.800 be much improved. But then you price spend all your 0:07:18.800 --> 0:07:21.800 time by yourself. You'd have no motivation, no care to 0:07:21.800 --> 0:07:25.640 connect with other people. You wouldn't you wouldn't reproduce, you 0:07:25.640 --> 0:07:28.880 wouldn't fall in love, you wouldn't develop friendships. You'd be 0:07:28.960 --> 0:07:31.720 fine by yourself. And everybody wants to be a little 0:07:31.720 --> 0:07:33.560 bit fine by themselves, but too much of it, I 0:07:33.600 --> 0:07:36.720 think leads to estrangement, So loneliness serves a purpose. 0:07:37.120 --> 0:07:39.800 So we're in this moment now where suddenly we have 0:07:40.200 --> 0:07:44.360 AI that can solve the problem, and we'll talk about 0:07:44.440 --> 0:07:48.320 what we mean by solving the problem here. But you know, Paul, 0:07:48.360 --> 0:07:49.960 you and I have been in the field for a 0:07:50.000 --> 0:07:52.240 long time, and I don't think we would have ever 0:07:52.360 --> 0:07:55.440 guessed this one decade ago that we would be talking 0:07:55.480 --> 0:07:58.520 about AI as a real. 0:07:58.400 --> 0:08:01.120 Solution for loneliness. So let's talk about that. 0:08:01.480 --> 0:08:04.000 What do you think the pros and cons are of 0:08:04.160 --> 0:08:06.640 having AI companions in this context? 0:08:07.120 --> 0:08:10.240 So just just to agree with you on something, AI 0:08:10.480 --> 0:08:13.720 was the biggest surprise in my career. If you had 0:08:13.760 --> 0:08:15.560 asked me, I'm going to ask you what you felt, 0:08:15.640 --> 0:08:19.920 But I'll tell you. A month before chat GBD came 0:08:19.920 --> 0:08:21.960 out and you said, when will we have a machine 0:08:21.960 --> 0:08:23.560 you could just talk to a different or a person 0:08:23.640 --> 0:08:26.280 I say, ten years, twenty years, thirty years, and then 0:08:26.320 --> 0:08:28.680 boom it came out and stunned me and stun in 0:08:28.680 --> 0:08:31.560 the world. What was your reaction. You're closer to these things, 0:08:31.600 --> 0:08:33.360 so maybe you were less surprised. 0:08:33.520 --> 0:08:36.440 Now it's exactly the same I spent really most of 0:08:36.480 --> 0:08:39.920 my career as a neuroscientist sort of snickering at AI 0:08:40.559 --> 0:08:43.160 and thinking, well, it's nothing like the brain, it's never 0:08:43.240 --> 0:08:45.559 going to get there, And then suddenly it was there, 0:08:45.600 --> 0:08:48.760 and it's changed all our lives so enormously that you know, 0:08:48.800 --> 0:08:52.240 I do this neuroscience podcast and about a quarter or 0:08:52.280 --> 0:08:55.840 a third of my podcast episodes are about AI nowadays. 0:08:55.679 --> 0:08:58.240 Sooner or later, by law, every conversation we have has 0:08:58.240 --> 0:08:59.839 to be about AI. You're a teacher, how do you 0:08:59.880 --> 0:09:01.600 do AI? What are you going to do with AI? 0:09:03.000 --> 0:09:07.080 So let's get to the loneliness thing. My view is 0:09:07.120 --> 0:09:10.000 annoyingly nuanced so that no one's really happy with it. 0:09:10.240 --> 0:09:11.760 I think there are a lot of people who are 0:09:11.760 --> 0:09:17.080 really lonely and their loneliness is severe, and maybe they're 0:09:17.120 --> 0:09:20.240 in a state where an AI companion is really the 0:09:20.280 --> 0:09:23.320 best they're going to get. And you know, suppose you're 0:09:23.360 --> 0:09:26.080 eighty years old, you're in an institution, you have no 0:09:26.280 --> 0:09:29.800 friends and family visiting you. Maybe you have dementia which 0:09:29.840 --> 0:09:32.880 makes you very difficult to talk with, and you're not 0:09:32.960 --> 0:09:36.080 a multimillionaire, so you can't pay people to entertain you. 0:09:36.840 --> 0:09:39.360 If it could turn out that chat GPT could be 0:09:39.600 --> 0:09:44.160 our claud or whatever could be a rewarding, satisfying companion 0:09:44.480 --> 0:09:49.160 to you. I think it's monstrous to deny it to you. 0:09:49.720 --> 0:09:51.800 It would be like telling people you can't have pets 0:09:51.840 --> 0:09:56.520 because we have decided pets are not sufficiently rich interacting partners. 0:09:56.720 --> 0:10:01.439 So whatever limitations these AIS have, you know, we should 0:10:01.480 --> 0:10:04.640 make it available the people who really need it and 0:10:04.720 --> 0:10:07.959 people who who's who's suffering is intense and there aren't 0:10:07.960 --> 0:10:12.240 any alternatives. So that's the pro AI side. The anti 0:10:12.320 --> 0:10:14.520 AI side is kind of the rest of us, where 0:10:14.559 --> 0:10:17.560 I think that there are serious problems where we move 0:10:17.679 --> 0:10:22.320 to AI companions as a replacement for human companions, and 0:10:22.360 --> 0:10:24.680 I'll tell you one of them. Maybe The main one 0:10:24.679 --> 0:10:27.600 that bothers me is that it goes back to your 0:10:27.679 --> 0:10:31.000 question about the function of loneliness, where the struggles we 0:10:31.120 --> 0:10:34.040 have when dealing with people inform us and make us 0:10:34.040 --> 0:10:37.040 better people. I tell you a story and you find 0:10:37.040 --> 0:10:39.240 it kind of boring. Next time I tell a story 0:10:39.240 --> 0:10:42.000 a bit more interesting. You find talking to me frustrating 0:10:42.040 --> 0:10:43.920 and don't want to talk anymore because I don't listen. Well, 0:10:44.000 --> 0:10:47.120 I'll listen better next time. By dealing with people, I 0:10:47.200 --> 0:10:50.040 become more sensitive. This is a story we've all gone 0:10:50.080 --> 0:10:52.800 through our lives. We've all been we've all been teenagers 0:10:52.800 --> 0:10:56.319 who have been awkward, who've been terrible at flirting, who 0:10:56.600 --> 0:11:00.360 have been borish and inconsiderate. And through the feed we 0:11:00.400 --> 0:11:02.840 get from people, we have become better. We have become 0:11:02.840 --> 0:11:05.600 better people, not just better conversational. It's not just better company, 0:11:05.600 --> 0:11:10.520 but better people. Imagine you take that away. And one 0:11:10.600 --> 0:11:14.559 thing ais do very well is they make you feel 0:11:14.720 --> 0:11:19.400 like you are wonderful. Everything I tell CHAGGYBD is brilliant. 0:11:19.440 --> 0:11:22.200 All my questions are wonderful, All my paper drafts are brilliant, 0:11:23.000 --> 0:11:26.680 My stories are amusing, my jokes are hilarious. It is 0:11:26.720 --> 0:11:29.960 always available to me at any time, at any second, 0:11:30.480 --> 0:11:33.920 and it will never sort of wait for me to 0:11:33.960 --> 0:11:36.000 finish talking so it could get its own words in. 0:11:36.920 --> 0:11:39.000 And in some way that's wonderful. But in some way 0:11:39.040 --> 0:11:42.360 that's terrible. I think it has a sort of corrosive power. 0:11:44.280 --> 0:11:47.520 Can I make a distinction, though, which is you're adjusting 0:11:47.520 --> 0:11:50.280 two points here. One is that it's always there and 0:11:50.360 --> 0:11:53.800 paying attention to you, which is wonderful. The other is 0:11:53.840 --> 0:11:56.960 that it's sycophantic and telling you that your ideas are great. 0:11:56.960 --> 0:11:59.720 But the second one I think can evolve and will 0:11:59.720 --> 0:12:00.640 eve time. 0:12:01.040 --> 0:12:02.600 You know, Chat GPT. 0:12:02.440 --> 0:12:06.040 Released a version that was overly sycophantic and then nobody 0:12:06.080 --> 0:12:07.360 liked that and they got rid of that. But I 0:12:07.360 --> 0:12:10.880 think with time, it certainly seems possible that companies will 0:12:10.880 --> 0:12:15.840 make better and better hard love ais that give you 0:12:15.920 --> 0:12:19.640 feedback that tell you that you're you're wrong, or there's 0:12:19.679 --> 0:12:22.240 a different way you can think about it. But the 0:12:22.280 --> 0:12:26.680 first part, which is that it never gets angry or 0:12:27.080 --> 0:12:30.480 you know, loses attention or has other things that it 0:12:30.480 --> 0:12:32.960 needs to do besides talk to you, that part's permanent, 0:12:33.000 --> 0:12:35.239 and that part I think might be really positive. 0:12:35.679 --> 0:12:37.599 I mean, you can imagine fiddling with it, right, you 0:12:37.600 --> 0:12:39.760 can imagine making it that. You know. I three in 0:12:39.760 --> 0:12:42.720 the morning, I ask Chat a question and it says, dude, 0:12:42.720 --> 0:12:44.160 do you know how late this is? And do you 0:12:44.160 --> 0:12:46.120 know what a stupid question that is? I am not 0:12:46.120 --> 0:12:47.720 going to answer it. Go back to sleep and talk 0:12:47.760 --> 0:12:50.320 to me in six hours. You know no reason why 0:12:50.360 --> 0:12:52.360 we can't build the machines that way. We probably won't 0:12:52.400 --> 0:12:52.800 want to. 0:12:53.120 --> 0:12:55.920 The reason I think we will is because even something 0:12:56.000 --> 0:12:59.240 like TikTok, which is there to grab your attention and 0:12:59.320 --> 0:13:02.120 keep it even They have started implementing things. If you 0:13:02.200 --> 0:13:04.679 surf too long, it pops up a video that says, hey, 0:13:04.720 --> 0:13:07.400 you've been surfing for X hours. Why don't you put 0:13:07.400 --> 0:13:10.600 this down and come back later. So it certainly is 0:13:10.720 --> 0:13:13.200 plausible that we can get to a point where we 0:13:13.280 --> 0:13:15.360 will have AI that gives us the right kind of 0:13:15.360 --> 0:13:16.120 feedback like that. 0:13:16.520 --> 0:13:20.080 There's certainly no technological barrier to it. In fact, right now, 0:13:20.160 --> 0:13:23.040 with some prompt engineering, you could tell your AI stop 0:13:23.120 --> 0:13:25.680 sucking up to me so much. Don't tell me I'm 0:13:25.679 --> 0:13:31.000 a genius anymore, just answer my damn questions. I wonder, though, 0:13:32.000 --> 0:13:34.520 if people will really want this, if we'll be in 0:13:34.559 --> 0:13:38.319 a situation where it's technologically fully possible to make these 0:13:38.360 --> 0:13:43.240 machines less frictionless, give us more pushback, call us out 0:13:43.920 --> 0:13:46.600 when we mess up, But we won't want them to 0:13:46.640 --> 0:13:50.280 do that. We will insist on machines that make us 0:13:50.320 --> 0:13:53.160 feel good about ourselves. I mean, what do you think 0:13:53.320 --> 0:13:56.000 imagine this, say four years from now, five years from now, 0:13:56.040 --> 0:13:57.880 when you get better and better and better, do you 0:13:57.880 --> 0:14:01.640 think we'll be talking to machines that will give us pushback? 0:14:01.920 --> 0:14:05.160 Or do you think we'll find it irresistible to deal 0:14:05.200 --> 0:14:08.120 with machines that just make us feel great about ourselves. 0:14:08.160 --> 0:14:10.600 I think it will become boring pretty quickly to have 0:14:10.640 --> 0:14:13.520 a machine that's sycophantic, and in fact, I happen to 0:14:13.559 --> 0:14:18.080 be very optimistic about AI relationships even among young people. 0:14:18.160 --> 0:14:20.160 Will come back to this about the difference between the 0:14:20.200 --> 0:14:23.440 elderly and the young, but I think it could serve 0:14:23.520 --> 0:14:26.760 as a sandbox where young people get to make all 0:14:26.800 --> 0:14:30.360 their dumb mistakes. If you have an AI companion that's 0:14:30.400 --> 0:14:32.960 giving you the right kind feedback and saying, hey, that 0:14:33.040 --> 0:14:35.320 hurt my feelings that you said that, or that didn't 0:14:35.320 --> 0:14:38.040 seem like you were thinking about me when you said that. 0:14:38.480 --> 0:14:42.480 So I think people can get better by using these 0:14:42.520 --> 0:14:46.600 things if the company makes these correctly with giving hard feedback. 0:14:46.960 --> 0:14:50.040 I totally agree with you. Back in the day, I 0:14:50.160 --> 0:14:52.400 found it difficult to talk to girls when I was 0:14:53.160 --> 0:14:55.520 You may find see me so smooth and say, how's 0:14:55.560 --> 0:15:00.200 that possible? I was somewhat awkward, somewhat shy, And to 0:15:00.320 --> 0:15:03.680 have something a machine I could practice on, I guess 0:15:03.760 --> 0:15:05.560 you have to work on talking to it. We can 0:15:06.160 --> 0:15:09.320 forget about just more generally, anybody who some social problems 0:15:09.320 --> 0:15:12.720 would benefit a lot from a sandbox. From practice, and 0:15:12.760 --> 0:15:16.040 then having practice enough, you'd unlaunch yourself into the real world. 0:15:16.800 --> 0:15:18.920 But the problem is, what if it's more fun to 0:15:18.920 --> 0:15:22.440 play in the sandbus than the real world. What if 0:15:22.480 --> 0:15:26.080 you find your AI companions just better than real companions? 0:15:26.400 --> 0:15:27.040 So okay, good. 0:15:27.120 --> 0:15:29.160 So this brings us back to the issue about age. 0:15:30.320 --> 0:15:34.520 We agree that for let's say, an elderly person who's 0:15:34.560 --> 0:15:36.960 alone in the same way as you point out in 0:15:36.960 --> 0:15:39.480 the article that in the same way they might we 0:15:39.600 --> 0:15:42.480 might give them opiates. We wouldn't want to do that 0:15:42.680 --> 0:15:46.400 with a teenager, And so we wouldn't want to give 0:15:47.200 --> 0:15:50.680 false cures for loneliness to a teenager who needs to 0:15:50.800 --> 0:15:54.200 go out and do the work to get better at 0:15:54.240 --> 0:15:58.520 what they're doing. Do you see a line that we 0:15:58.600 --> 0:16:01.120 would ever be able to where we say, okay, look, 0:16:01.200 --> 0:16:04.160 it's okay for people over here who are really lonely 0:16:04.240 --> 0:16:06.960 to have issues that are going to make them lonely 0:16:07.160 --> 0:16:09.840 into the future, versus someone who should go out and 0:16:09.920 --> 0:16:10.400 do the work. 0:16:11.240 --> 0:16:13.880 I don't think there's a bright line, and it's going 0:16:13.960 --> 0:16:17.280 to be somewhat arbitrary, But I like the analogy of opiates, 0:16:17.760 --> 0:16:20.760 and you know, we will give very powerful drugs to 0:16:20.840 --> 0:16:22.800 people near the end of their alliance, people who are 0:16:22.840 --> 0:16:26.600 suffering from terrible pain. Drugs we would never give to 0:16:26.880 --> 0:16:29.360 a seventeen year old who hurt us back, because we 0:16:29.400 --> 0:16:31.320 don't want a seventeen year old to become addicted. We 0:16:31.480 --> 0:16:34.680 worry a lot less about a nine year old. And 0:16:35.240 --> 0:16:38.040 there's not a single bright line here. But there's going 0:16:38.120 --> 0:16:43.160 to be cultural decisions made, policies, laws, and I think 0:16:43.200 --> 0:16:46.160 there's a decision we're going to make with AIS now 0:16:46.720 --> 0:16:49.120 that there are an end in some way liberty issues, right. 0:16:49.680 --> 0:16:51.320 You know, I don't know how you're going to feel 0:16:51.560 --> 0:16:54.040 when the government starts restricting your use of chatbots. 0:16:54.360 --> 0:16:55.120 Yeah, so what do you think. 0:16:55.200 --> 0:16:58.640 Do you think there will be legislation around this or 0:16:58.720 --> 0:17:00.440 will this be a laisse a fair issue. 0:17:00.800 --> 0:17:03.800 I'll make a prediction here, which is that they will. 0:17:03.880 --> 0:17:06.520 There will not be legislation. People want it too much. 0:17:07.440 --> 0:17:11.080 You know, you made the connection with TikTok and Twitter. 0:17:11.160 --> 0:17:13.280 You know, if if the government stepped in, you know, 0:17:13.560 --> 0:17:16.240 I don't know Trump or Biden or whoever whoever's going 0:17:16.280 --> 0:17:18.560 to be the next present and said we're going to 0:17:18.640 --> 0:17:21.560 limit your use of social media, I mean some way 0:17:21.600 --> 0:17:25.080 probably a good thing, but it really is a violation 0:17:25.160 --> 0:17:28.159 of our freedom. And I don't know. Well, let me 0:17:28.200 --> 0:17:30.080 ask you how many hours a day, I mean, if 0:17:30.119 --> 0:17:32.200 at all, do you interact with AI's. 0:17:33.280 --> 0:17:34.720 Forty minutes I have. 0:17:35.480 --> 0:17:38.760 I have a tesla and there's AI in it now, 0:17:39.080 --> 0:17:41.120 and so when I'm driving along, you know, I talk 0:17:41.200 --> 0:17:43.080 to it. I ask you questions anything I'm curious about, 0:17:43.119 --> 0:17:45.880 how to fix this piece that I'm trying to fix 0:17:45.960 --> 0:17:48.800 on my door, or how to think about you know 0:17:49.040 --> 0:17:51.040 which philosophers have said X y Z. 0:17:51.800 --> 0:17:52.280 I find it. 0:17:52.640 --> 0:17:56.080 You know, it's it's the most intelligent companion that I have, 0:17:56.280 --> 0:17:57.720 and it's really it's really wonderful. 0:17:57.920 --> 0:18:00.680 Right now, imagine the government step in and then says, well, 0:18:00.920 --> 0:18:03.440 a man of your age and you know, success and 0:18:03.800 --> 0:18:06.879 good social connections, that's too much. We'll go to leive 0:18:06.920 --> 0:18:09.320 me that half hour half hour a day, you know, 0:18:09.480 --> 0:18:12.800 unless you get a doctor's prescription. I don't think people 0:18:12.800 --> 0:18:15.639 would stand for that. I think these machines are too available, 0:18:16.320 --> 0:18:35.560 too popular, too addictive for us to control it. Now, 0:18:35.680 --> 0:18:36.320 have you seen the. 0:18:36.359 --> 0:18:39.440 New Grock Avatar AI companion? 0:18:40.880 --> 0:18:41.560 It's I have not. 0:18:41.800 --> 0:18:48.160 Oh, okay, so it's a it's a sexy anime female. 0:18:48.800 --> 0:18:51.080 You talk to her and you see her on your phone, 0:18:51.400 --> 0:18:54.160 and she's moving around and when you say something nice, 0:18:54.440 --> 0:18:58.080 little hearts pop up from her and she's very, very flirtatious. 0:18:58.640 --> 0:19:01.800 And I do worry when I saw that for the 0:19:01.880 --> 0:19:05.440 first time, I felt a not in my stomach for 0:19:05.520 --> 0:19:08.280 the first I'm very cyber optimistic about most of the stuff, 0:19:08.359 --> 0:19:10.639 but it did make me worry about what it'll mean 0:19:10.680 --> 0:19:15.240 for young people to have such a magnetic AI companion. 0:19:16.160 --> 0:19:18.000 What's your take on whether it can go too far? 0:19:18.480 --> 0:19:19.479 Have you seen a movie her? 0:19:19.800 --> 0:19:19.960 Yes? 0:19:21.160 --> 0:19:23.119 To me, this is a really important movie. So if 0:19:23.400 --> 0:19:25.760 those people haven't seen it, just the short summary is 0:19:26.080 --> 0:19:27.880 it's the near future. In fact, i've heard this twenty 0:19:27.920 --> 0:19:30.320 twenty five. It was done like ten years ago, and 0:19:30.440 --> 0:19:34.320 the future was at twenty twenty five. And AI companions 0:19:34.480 --> 0:19:36.400 who you deal with over your phone like we deal 0:19:36.440 --> 0:19:40.320 with AI right now, or your computer become popular. And 0:19:40.760 --> 0:19:43.320 it's about this guy who falls in love with Samantha, 0:19:43.960 --> 0:19:48.720 this wonderful AI companion voice by Scarlett Johanson. And the 0:19:48.840 --> 0:19:51.640 thing about it is it's so well done that you're 0:19:51.680 --> 0:19:54.080 watching it and you yourself fall in love with her, 0:19:54.440 --> 0:19:57.320 and the movie ends by kind of movie ends. You 0:19:57.440 --> 0:20:00.840 see everybody walking talking excitedly on their phone, and everybody 0:20:01.200 --> 0:20:03.600 has an AI compaign at diurnal love with because it's 0:20:03.680 --> 0:20:09.479 infinitely preferable to human companion. Now, where this falls apart, 0:20:09.960 --> 0:20:11.680 and this is something you know a lot more about 0:20:12.080 --> 0:20:16.200 than me, is that these machines are basically at disappoint abstract. 0:20:16.280 --> 0:20:20.840 They're not robots. You can't interact bodily with them. And 0:20:20.960 --> 0:20:24.240 it always struck me. I know you're working act lot 0:20:24.240 --> 0:20:27.760 of these issues that we have done much better at 0:20:27.840 --> 0:20:32.280 solving the computational problems at the abstract level than building 0:20:32.400 --> 0:20:34.679 machines that could deal with three dimensional world. I mean 0:20:34.720 --> 0:20:36.600 to put it a different way. It might have surprised 0:20:36.600 --> 0:20:39.680 a lot of people, but we very quickly came to 0:20:39.760 --> 0:20:43.200 develop machines that could beat grand masters at chess. We 0:20:43.280 --> 0:20:45.200 don't yet have a machine that could load a dishwasher 0:20:46.080 --> 0:20:50.720 right lit alone serve as a suitable romantic or sexual partner. 0:20:50.960 --> 0:20:52.480 So how far do you think we're away from that? 0:20:52.880 --> 0:20:55.840 Interestingly, you know, robotics is really taking off right now. 0:20:56.840 --> 0:21:03.080 Essentially people are using LM large language models in analogy 0:21:03.160 --> 0:21:05.280 to physical motions. So they say Look, if I had 0:21:05.359 --> 0:21:07.639 just moved my hand here and here and here, the 0:21:07.800 --> 0:21:10.040 next token should be that I'm going to move my 0:21:10.080 --> 0:21:14.320 hand over here. So it's actually it's speeding up enormously. 0:21:15.000 --> 0:21:18.240 But what's interesting is I think even with you know, 0:21:18.400 --> 0:21:20.960 just a purely chat ai, or even when we get 0:21:21.480 --> 0:21:24.879 sex robots, we're still going to have a very strong 0:21:25.080 --> 0:21:29.280 drive to have a girlfriend and take her out with 0:21:29.400 --> 0:21:32.960 our friends, or have her meet our parents, or take 0:21:33.000 --> 0:21:35.080 her to the movies or have a nice romantic dinner. 0:21:35.400 --> 0:21:38.280 So that's why I'm slightly less worried that we'll have 0:21:38.359 --> 0:21:43.000 a takeover of you know, chatbots or robots in terms 0:21:43.040 --> 0:21:47.000 of our relationships, because we have these very deeply embedded 0:21:47.160 --> 0:21:51.720 drives to be part of a social world and be 0:21:51.960 --> 0:21:53.600 with our companions in that world. 0:21:53.960 --> 0:21:56.960 I think that's right. I think that. I mean this 0:21:57.080 --> 0:21:59.760 turns on another issue which I think we both thought about, 0:22:00.119 --> 0:22:03.120 is at what point will this become conscious? If these 0:22:03.200 --> 0:22:07.080 machines become conscious and embodied, then to all intents and purposes, 0:22:07.119 --> 0:22:10.760 they're just more people, very smart people, people created in 0:22:10.880 --> 0:22:13.240 a lab and not a normal biological way. But they're 0:22:13.320 --> 0:22:18.000 just people. But if they're not conscious, having a relationship 0:22:18.080 --> 0:22:20.840 with an AI is no different in kind than having 0:22:20.920 --> 0:22:22.760 a relationship with a chair or a toaster. 0:22:23.480 --> 0:22:26.800 It's just a thing, but it's a thing that gives 0:22:26.840 --> 0:22:29.560 you feedback and reflects things to you. 0:22:30.160 --> 0:22:30.560 I wonder. 0:22:30.680 --> 0:22:34.000