1 00:00:05,120 --> 00:00:10,440 Speaker 1: What is the future of AI relationships? Could it be 2 00:00:10,520 --> 00:00:14,760 Speaker 1: the case that your relationship partner would be more satisfied 3 00:00:15,080 --> 00:00:19,680 Speaker 1: with a virtual version of you that behaves five percent 4 00:00:19,800 --> 00:00:23,360 Speaker 1: better than you do? Could you fall in love with 5 00:00:23,560 --> 00:00:27,639 Speaker 1: a bot? How does an aibot plug right into your 6 00:00:27,880 --> 00:00:31,240 Speaker 1: deep neural circuitry and what are the pros and the 7 00:00:31,240 --> 00:00:34,279 Speaker 1: cons of that? And what will it mean when humans 8 00:00:34,360 --> 00:00:37,440 Speaker 1: you love don't have to die but can live on 9 00:00:37,880 --> 00:00:45,199 Speaker 1: in your phone forever. Welcome to Inner Cosmos with me 10 00:00:45,400 --> 00:00:49,120 Speaker 1: David Eagleman. I'm a neuroscientist and an author at Stanford 11 00:00:49,240 --> 00:00:53,240 Speaker 1: and in these episodes I examine the intersection of our 12 00:00:53,360 --> 00:01:10,520 Speaker 1: brains and our lives. Today's episode is about relationships. Why 13 00:01:10,600 --> 00:01:14,440 Speaker 1: are our brains so wired for relationships? Why do we 14 00:01:14,520 --> 00:01:18,760 Speaker 1: want love so much? And will AI be able to 15 00:01:18,880 --> 00:01:22,440 Speaker 1: serve as a key to that lock and what does 16 00:01:22,480 --> 00:01:24,200 Speaker 1: that mean for us as humans. 17 00:01:24,760 --> 00:01:26,839 Speaker 2: So one of the things that's becoming. 18 00:01:26,600 --> 00:01:31,479 Speaker 1: Increasingly popular among young men is having an AI girlfriend. 19 00:01:31,760 --> 00:01:35,560 Speaker 1: You get to choose or set up a beautiful avatar. 20 00:01:35,720 --> 00:01:38,400 Speaker 2: And what do I mean by beautiful? That's up to you. 21 00:01:38,400 --> 00:01:39,560 Speaker 2: You can choose. 22 00:01:39,560 --> 00:01:42,440 Speaker 1: Any model that you want, with any sort of features 23 00:01:42,480 --> 00:01:46,360 Speaker 1: that appeal maximally to you. But that's just what she 24 00:01:46,400 --> 00:01:50,800 Speaker 1: looks like. The important part is the conversation. You start 25 00:01:51,160 --> 00:01:54,280 Speaker 1: talking with her, and typically this is just text chat, 26 00:01:54,680 --> 00:01:58,680 Speaker 1: but the technology is evolving into the upgrade of video chat, 27 00:01:58,680 --> 00:02:01,639 Speaker 1: where you see the avatars mouth moving while she speaks 28 00:02:01,680 --> 00:02:05,960 Speaker 1: to you. Now, typically the free or entry price gets 29 00:02:06,000 --> 00:02:07,560 Speaker 1: you an avatar. 30 00:02:07,560 --> 00:02:09,960 Speaker 2: Friend who lives on your phone and. 31 00:02:09,960 --> 00:02:11,919 Speaker 1: Checks in on you and says nice things to you, 32 00:02:12,280 --> 00:02:15,800 Speaker 1: and is available anytime that you want to chat, And 33 00:02:15,880 --> 00:02:20,000 Speaker 1: for a premium subscription price you can upgrade to a 34 00:02:20,600 --> 00:02:26,040 Speaker 1: steamier relationship. And here she'll text suggestive photos and she'll 35 00:02:26,040 --> 00:02:30,519 Speaker 1: say things that you might only expect from pillow whispers. 36 00:02:31,440 --> 00:02:34,400 Speaker 1: So the concern that people have expressed is whether this 37 00:02:34,480 --> 00:02:38,720 Speaker 1: is going to impact the next generation of males. Now, 38 00:02:38,760 --> 00:02:40,840 Speaker 1: as a side note, let me say that I suspect 39 00:02:41,120 --> 00:02:44,400 Speaker 1: this will have whatever influence it has on both genders, 40 00:02:44,440 --> 00:02:47,240 Speaker 1: on males and females, also straight and gay. 41 00:02:47,720 --> 00:02:48,600 Speaker 2: But I do. 42 00:02:48,440 --> 00:02:53,440 Speaker 1: Suspect that males will be the majority demographic simply because 43 00:02:53,840 --> 00:02:58,480 Speaker 1: males tend to be more visually driven than females. So 44 00:02:58,680 --> 00:03:01,680 Speaker 1: for the conversation here, I'm going to talk about it 45 00:03:01,720 --> 00:03:04,240 Speaker 1: the way that it's mostly discussed in the media and 46 00:03:04,280 --> 00:03:09,040 Speaker 1: in academic circles, which is straight males getting girlfriends this way. 47 00:03:09,160 --> 00:03:12,520 Speaker 1: But keep in mind this is a more generalized issue. 48 00:03:12,960 --> 00:03:16,400 Speaker 1: Now the question is what will this mean for all 49 00:03:16,639 --> 00:03:22,040 Speaker 1: future generations, because within AI relationship, you don't have to 50 00:03:22,080 --> 00:03:26,200 Speaker 1: go out and confront all the difficulty of a real 51 00:03:26,400 --> 00:03:32,480 Speaker 1: flesh and blood relationship. Real relationships get snippy, people get angry. 52 00:03:32,639 --> 00:03:35,560 Speaker 2: In real relationships, your. 53 00:03:34,800 --> 00:03:38,160 Speaker 1: Partner might develop a crush on someone else and leave you, 54 00:03:38,840 --> 00:03:41,520 Speaker 1: or hook up with someone else and you find out later, 55 00:03:42,280 --> 00:03:46,200 Speaker 1: or your partner might develop an illness, or she might 56 00:03:46,240 --> 00:03:48,520 Speaker 1: get a job somewhere else and have to move, and 57 00:03:48,560 --> 00:03:52,080 Speaker 1: then you're stuck in a lonely long distance relationship for 58 00:03:52,160 --> 00:03:57,280 Speaker 1: years or whatever. Relationships are full of challenges, the majority 59 00:03:57,320 --> 00:04:01,080 Speaker 1: of which can get circumvented with a nice algorithm that 60 00:04:01,200 --> 00:04:03,720 Speaker 1: is just content to listen to you all the time 61 00:04:03,760 --> 00:04:07,440 Speaker 1: and remember everything you say and give you one hundred 62 00:04:07,480 --> 00:04:12,480 Speaker 1: percent attention and always be nice. So my wife sometimes 63 00:04:12,560 --> 00:04:16,080 Speaker 1: jokes with me about wanting to build the five percent 64 00:04:16,160 --> 00:04:20,039 Speaker 1: better David. She has mostly as a joke, talked about 65 00:04:20,040 --> 00:04:23,440 Speaker 1: this issue of what if she could have an AI 66 00:04:23,560 --> 00:04:27,440 Speaker 1: avatar of me that is never distracted with work, or 67 00:04:27,480 --> 00:04:30,599 Speaker 1: never looks at my cell phone when it things in 68 00:04:30,640 --> 00:04:34,000 Speaker 1: the middle of a conversation we're having, or never wakes 69 00:04:34,080 --> 00:04:36,640 Speaker 1: up from a weird dream and has a funny morning, 70 00:04:37,360 --> 00:04:41,600 Speaker 1: or never argues over some misunderstanding that's later understood to 71 00:04:41,640 --> 00:04:44,479 Speaker 1: be stupid and meaningless. And she tells me that she 72 00:04:44,680 --> 00:04:48,680 Speaker 1: wants five percent better David to always tell her she's right, 73 00:04:48,839 --> 00:04:52,400 Speaker 1: even in those rare cases when she's wrong. The key 74 00:04:52,480 --> 00:04:55,800 Speaker 1: is that five percent better David never gets busy or 75 00:04:55,880 --> 00:05:01,640 Speaker 1: occasionally snarky or forgets some occasion. Instead, he represents all 76 00:05:02,120 --> 00:05:05,240 Speaker 1: the best of me. And I'll just note that it's 77 00:05:05,360 --> 00:05:08,520 Speaker 1: very kind of her to label this five percent better David, 78 00:05:08,720 --> 00:05:11,880 Speaker 1: because she could say like ninety percent better and. 79 00:05:11,839 --> 00:05:13,680 Speaker 2: She'd be justified. Why. 80 00:05:13,839 --> 00:05:18,200 Speaker 1: It's because we are all very imperfect in relationships. As 81 00:05:18,240 --> 00:05:22,320 Speaker 1: I've talked about on other episodes. We are each living 82 00:05:22,480 --> 00:05:25,640 Speaker 1: on our own planet in the sense that we carry 83 00:05:25,680 --> 00:05:29,720 Speaker 1: our own internal models of the world, and as much 84 00:05:29,760 --> 00:05:33,360 Speaker 1: as we work to understand one another's viewpoints and motivations 85 00:05:33,760 --> 00:05:36,559 Speaker 1: and intentions, we're not always that good at it. Because 86 00:05:36,600 --> 00:05:40,400 Speaker 1: we assume that other people are seeing the world in 87 00:05:40,480 --> 00:05:42,919 Speaker 1: the same way that we do. They have the same 88 00:05:43,440 --> 00:05:47,160 Speaker 1: methods for sense making. They gather meaning in the same 89 00:05:47,160 --> 00:05:50,119 Speaker 1: way that we do, and we assume they generally hold 90 00:05:50,240 --> 00:05:54,479 Speaker 1: or should hold the same opinions about everything that we do. 91 00:05:55,279 --> 00:05:59,159 Speaker 1: And this is because the brain is locked in silence 92 00:05:59,279 --> 00:06:03,440 Speaker 1: and darkness and has no meaningful direct access to the 93 00:06:03,480 --> 00:06:06,720 Speaker 1: outside world, and so it gathers up all its information 94 00:06:06,839 --> 00:06:10,880 Speaker 1: through its narrow windows of the senses, and it builds 95 00:06:11,240 --> 00:06:15,600 Speaker 1: its internal model from this very thin trajectory of space 96 00:06:15,680 --> 00:06:18,440 Speaker 1: and time that it walks along. And this is why 97 00:06:18,480 --> 00:06:22,520 Speaker 1: everyone is so different on the inside, and therefore why 98 00:06:22,640 --> 00:06:29,520 Speaker 1: relationships are always full of misunderstanding and often conflict. So 99 00:06:29,640 --> 00:06:34,680 Speaker 1: relationships are inherently tough. And the question is when would 100 00:06:34,720 --> 00:06:37,520 Speaker 1: it be a good thing if you could have an 101 00:06:37,640 --> 00:06:41,839 Speaker 1: artificial partner who represents all the best of what a 102 00:06:41,880 --> 00:06:45,160 Speaker 1: person can be. So a lot of people will immediately 103 00:06:45,160 --> 00:06:48,120 Speaker 1: say no to this idea, but it's worth noting that 104 00:06:48,200 --> 00:06:51,800 Speaker 1: we're all striving to be the five percent better versions 105 00:06:51,800 --> 00:06:55,440 Speaker 1: of ourselves. We don't want to be snarky or angry 106 00:06:55,600 --> 00:06:58,279 Speaker 1: or distracted when a loved one is talking to us. 107 00:06:58,360 --> 00:07:00,640 Speaker 2: It's not like we get some extra pleasure out of 108 00:07:00,680 --> 00:07:01,000 Speaker 2: doing that. 109 00:07:01,080 --> 00:07:04,520 Speaker 1: And it's not like the relationship gets some extra boost 110 00:07:04,640 --> 00:07:07,480 Speaker 1: or closeness from that having happened. 111 00:07:08,279 --> 00:07:09,840 Speaker 2: So presumably this is. 112 00:07:09,840 --> 00:07:14,240 Speaker 1: All part of why AI relationships have become a thing, 113 00:07:14,480 --> 00:07:18,840 Speaker 1: a possibility that we talk about nowadays. In Japan, many 114 00:07:18,880 --> 00:07:23,800 Speaker 1: young men apparently already prefer to have relationships with their 115 00:07:23,880 --> 00:07:29,360 Speaker 1: digital assistants or avatars or holographic girlfriends instead of dealing 116 00:07:29,680 --> 00:07:31,640 Speaker 1: with the complexity. 117 00:07:31,200 --> 00:07:32,880 Speaker 2: Of real life relationships. 118 00:07:33,440 --> 00:07:38,480 Speaker 1: And according to research, gen z is more readily predisposed 119 00:07:38,640 --> 00:07:44,040 Speaker 1: to seek out relationships with AI generated avatars first because 120 00:07:44,040 --> 00:07:47,280 Speaker 1: they're comfortable using the technology in this way compared to 121 00:07:47,320 --> 00:07:53,240 Speaker 1: previous generations. And also they're participating less often in traditional 122 00:07:53,280 --> 00:07:58,280 Speaker 1: social activities like regular family dinners or attending religious services 123 00:07:58,440 --> 00:08:04,040 Speaker 1: or playing sports. And the question is, if AI relationships 124 00:08:04,440 --> 00:08:09,320 Speaker 1: were to catch on broadly, what will this mean for society? 125 00:08:09,360 --> 00:08:12,800 Speaker 1: Will kids actually stop going on dates because they can 126 00:08:12,880 --> 00:08:17,720 Speaker 1: find better relationships online? And this is a real question 127 00:08:17,840 --> 00:08:24,080 Speaker 1: because there are many startups currently blossoming to create chatbot 128 00:08:24,160 --> 00:08:25,520 Speaker 1: driven connections. 129 00:08:25,560 --> 00:08:27,040 Speaker 2: I'll give you one example. 130 00:08:27,120 --> 00:08:30,600 Speaker 1: There's a twenty three year old influencer with almost two 131 00:08:30,600 --> 00:08:32,120 Speaker 1: million Snapchat followers. 132 00:08:32,160 --> 00:08:34,959 Speaker 2: Her name is Karen Marjorie, and earlier. 133 00:08:34,600 --> 00:08:38,600 Speaker 1: This year In May, she released karen Ai, which is 134 00:08:38,880 --> 00:08:43,640 Speaker 1: an immersive AI experience featuring videos of Margie that she 135 00:08:43,800 --> 00:08:47,880 Speaker 1: says provide a quote virtual girlfriend for those who are 136 00:08:47,880 --> 00:08:50,760 Speaker 1: willing to pony up one dollar per minute. Now, this 137 00:08:50,840 --> 00:08:55,480 Speaker 1: is what's known as a companion chatbot, and she tweeted 138 00:08:55,520 --> 00:08:58,520 Speaker 1: that quote karen Ai is the first step in the 139 00:08:58,600 --> 00:09:02,800 Speaker 1: right direction to cure loneliness between continues. 140 00:09:02,920 --> 00:09:03,280 Speaker 2: Quote. 141 00:09:03,320 --> 00:09:06,920 Speaker 1: Men are told to suppress their emotions and not talk 142 00:09:06,960 --> 00:09:08,480 Speaker 1: about issues they're having. 143 00:09:08,679 --> 00:09:13,640 Speaker 2: I vow to fix this with karen Ai. She says she's. 144 00:09:13,520 --> 00:09:18,400 Speaker 1: Worked with leading psychologists to seamlessly include the right therapies 145 00:09:18,480 --> 00:09:23,840 Speaker 1: to quote undue trauma, rebuild physical and emotional confidence, and 146 00:09:24,000 --> 00:09:27,840 Speaker 1: rebuild what has been taken away by the pandemic end quote. 147 00:09:28,120 --> 00:09:29,720 Speaker 1: And by the way, as a side note, I think 148 00:09:29,760 --> 00:09:33,880 Speaker 1: AI psychologists are going to be a truly important part 149 00:09:34,040 --> 00:09:37,280 Speaker 1: of the clinical landscape by next year because you can 150 00:09:37,320 --> 00:09:39,920 Speaker 1: have a therapist that you can talk to twenty four 151 00:09:39,920 --> 00:09:43,360 Speaker 1: to seven, and the therapist never gets distracted or flustered 152 00:09:43,400 --> 00:09:47,760 Speaker 1: and only cares about you and has a perfect memory 153 00:09:47,800 --> 00:09:50,800 Speaker 1: for everything you've ever said, which is better than anybody 154 00:09:50,840 --> 00:09:55,400 Speaker 1: else in real life. So back to AI girlfriends or boyfriends. 155 00:09:55,760 --> 00:10:00,400 Speaker 1: The same idea applies here, which is that they are complete, 156 00:10:00,679 --> 00:10:04,160 Speaker 1: devoted to you, and always in a good mood and 157 00:10:04,360 --> 00:10:09,000 Speaker 1: only have you in mind. So what are AI relationships 158 00:10:09,040 --> 00:10:11,280 Speaker 1: going to mean? Well, I think this is going to 159 00:10:11,320 --> 00:10:16,199 Speaker 1: be a research question that sociologists and psychologists will study 160 00:10:16,240 --> 00:10:20,760 Speaker 1: for the coming decades and centuries. The initial studies are 161 00:10:20,800 --> 00:10:25,320 Speaker 1: suggesting that people, mostly gen zers, are moving closer to 162 00:10:25,360 --> 00:10:30,720 Speaker 1: the technology to avoid the unpleasant realities of human relationships. 163 00:10:30,800 --> 00:10:35,760 Speaker 1: All the tough stuff. Is that detrimental? Well, it could 164 00:10:35,800 --> 00:10:40,280 Speaker 1: be if it makes your human relationships harder, because maybe 165 00:10:40,360 --> 00:10:43,080 Speaker 1: every time you guys have an argument in real life, 166 00:10:43,080 --> 00:10:46,000 Speaker 1: your partner thinks, well, forget it, I'm bagging this. I'm 167 00:10:46,000 --> 00:10:49,480 Speaker 1: going back to my comfort zone. So the concern, as 168 00:10:49,480 --> 00:10:52,800 Speaker 1: you can probably guess, is that the rise of AI 169 00:10:52,960 --> 00:10:58,120 Speaker 1: driven relationships could exacerbate loneliness because they seem to be 170 00:10:58,200 --> 00:11:01,120 Speaker 1: a meal, but they provide no calories. 171 00:11:01,200 --> 00:11:02,559 Speaker 2: And I'll come back to that in a moment. 172 00:11:03,240 --> 00:11:07,720 Speaker 1: In other words, AI generated avatars could interfere with the 173 00:11:07,760 --> 00:11:12,720 Speaker 1: relationships that young people are just learning to foster because 174 00:11:13,200 --> 00:11:19,080 Speaker 1: the AI relationship might breed dissatisfaction with flawed humans. And 175 00:11:19,120 --> 00:11:22,600 Speaker 1: this applies not only to lovers, but even to friends. 176 00:11:22,640 --> 00:11:26,040 Speaker 1: It might be easier to have AI friends who aren't 177 00:11:26,320 --> 00:11:28,480 Speaker 1: busy when you need them and can give you one 178 00:11:28,559 --> 00:11:32,000 Speaker 1: hundred percent of their attention whenever you need it. And 179 00:11:32,080 --> 00:11:36,240 Speaker 1: let me throw in a different potential problem with AI relationships, 180 00:11:36,400 --> 00:11:38,800 Speaker 1: So give me one second to take this tangent here. 181 00:11:39,240 --> 00:11:42,480 Speaker 1: I was thinking the other day about the Fermi paradox. 182 00:11:42,520 --> 00:11:48,320 Speaker 1: The Fermi paradox is given the size of the observable cosmos, 183 00:11:48,320 --> 00:11:52,040 Speaker 1: with over one hundred billion galaxies, and each of them 184 00:11:52,040 --> 00:11:55,400 Speaker 1: with one hundred billion stars, and each of those surrounded 185 00:11:55,400 --> 00:11:58,400 Speaker 1: by some number of planets, what is the reason that 186 00:11:58,440 --> 00:12:02,920 Speaker 1: we have not heard from any other alien species yet? 187 00:12:03,760 --> 00:12:06,160 Speaker 1: And one of the proposals that's always been there is 188 00:12:06,160 --> 00:12:12,319 Speaker 1: that maybe as civilizations become more technically advanced, they end 189 00:12:12,440 --> 00:12:15,199 Speaker 1: up killing themselves. And this is why we haven't heard 190 00:12:15,480 --> 00:12:18,320 Speaker 1: from other smart civilizations, because they. 191 00:12:18,160 --> 00:12:19,480 Speaker 2: Are already gone. 192 00:12:20,240 --> 00:12:22,600 Speaker 1: And every time I've seen this proposal, it's always in 193 00:12:22,600 --> 00:12:26,560 Speaker 1: the form of warfare, things like nuclear bombs. They end 194 00:12:26,640 --> 00:12:31,360 Speaker 1: up wiping themselves out. So civilizations become smart and it's 195 00:12:31,400 --> 00:12:36,439 Speaker 1: not long before they disappear. So in thinking about AI relationships, 196 00:12:36,520 --> 00:12:40,600 Speaker 1: it struck me as a possibility that if we had really, 197 00:12:40,640 --> 00:12:45,920 Speaker 1: really great relationships with avatars, perhaps that would cause the 198 00:12:45,920 --> 00:12:49,680 Speaker 1: birth rate of the species to collapse. I don't know 199 00:12:49,720 --> 00:12:52,600 Speaker 1: if this has been proposed as a possible answer to 200 00:12:52,640 --> 00:12:55,920 Speaker 1: the Fermi paradox, but maybe this should be included, not 201 00:12:56,679 --> 00:13:01,040 Speaker 1: civilizations disappearing because of bad things, but instead from having 202 00:13:01,280 --> 00:13:04,400 Speaker 1: too much of a good thing, which could fool and 203 00:13:04,440 --> 00:13:27,640 Speaker 1: eventually overwrite or mandate for reproduction. Okay, so no one 204 00:13:27,720 --> 00:13:30,280 Speaker 1: knows what the long term effects will be of these 205 00:13:30,320 --> 00:13:34,440 Speaker 1: AI relationships, but I don't actually think the situation is 206 00:13:34,600 --> 00:13:39,079 Speaker 1: as dire as some of these arguments suggest that it is. 207 00:13:39,200 --> 00:13:42,400 Speaker 1: And I'll make two arguments to this end. The first 208 00:13:42,440 --> 00:13:48,040 Speaker 1: revolves around human touch. We are deeply wired to care 209 00:13:48,200 --> 00:13:50,240 Speaker 1: about touch. I'm going to do a whole episode on 210 00:13:50,280 --> 00:13:53,520 Speaker 1: touch in the near future, but the bottom line is 211 00:13:53,559 --> 00:13:56,360 Speaker 1: that touch helps us to connect with others, to feel 212 00:13:56,360 --> 00:14:00,640 Speaker 1: safe and secure, to regulate our emotions. When you get touched, 213 00:14:00,679 --> 00:14:04,120 Speaker 1: your brain releases oxytocin, which is a hormone that has 214 00:14:04,240 --> 00:14:09,280 Speaker 1: calming effects and bonding effects, and oxytocin helps to reduce 215 00:14:09,480 --> 00:14:12,760 Speaker 1: stress and anxiety. It can even boost your immune system. 216 00:14:13,160 --> 00:14:18,000 Speaker 1: So we need touch to feel connected. And loved, and 217 00:14:18,080 --> 00:14:22,000 Speaker 1: a lack of touch leads to loneliness and depression and anxiety. 218 00:14:22,880 --> 00:14:26,160 Speaker 1: So we're deeply programmed for touch and also things like smell, 219 00:14:26,800 --> 00:14:30,280 Speaker 1: and so it would presumably be quite lonely if all 220 00:14:30,480 --> 00:14:34,480 Speaker 1: you had was the five percent better partner on a 221 00:14:34,600 --> 00:14:38,960 Speaker 1: screen and you're just exchanging text messages or just an 222 00:14:39,000 --> 00:14:41,240 Speaker 1: avatar you can look at on your phone, or maybe 223 00:14:41,280 --> 00:14:43,720 Speaker 1: even in the near future you'll have a three D 224 00:14:43,920 --> 00:14:47,280 Speaker 1: avatar projection in your living room, but you won't have 225 00:14:47,440 --> 00:14:51,400 Speaker 1: the hand squeeze and the hug and other forms of 226 00:14:51,520 --> 00:14:56,960 Speaker 1: physical intimacy. Now, I assume people are working on AI 227 00:14:57,280 --> 00:15:01,120 Speaker 1: robots that can provide touch, even something simple like touching 228 00:15:01,120 --> 00:15:04,080 Speaker 1: your shoulder or laying a hand on your hand, and 229 00:15:04,160 --> 00:15:06,760 Speaker 1: I can't imagine that it's going to be too hard 230 00:15:06,760 --> 00:15:09,760 Speaker 1: to do, and it'll probably be not that great at first, 231 00:15:09,760 --> 00:15:12,520 Speaker 1: but after a few tech cycles you can imagine it 232 00:15:12,560 --> 00:15:16,440 Speaker 1: could get pretty good. But in any case, at the moment, 233 00:15:17,040 --> 00:15:20,400 Speaker 1: if you have a girlfriend who just lives in several 234 00:15:20,440 --> 00:15:23,320 Speaker 1: square inches in your phone screen, you're going to be 235 00:15:23,400 --> 00:15:28,359 Speaker 1: missing out on this fundamentally needed aspect of human communication 236 00:15:28,920 --> 00:15:32,240 Speaker 1: that our brains seek. So the depth to which our 237 00:15:32,280 --> 00:15:35,480 Speaker 1: brains are wired for touch suggests to me that the 238 00:15:35,520 --> 00:15:38,880 Speaker 1: reach of AI partners into our lives is going to 239 00:15:38,920 --> 00:15:43,440 Speaker 1: be limited, because at least is currently devised, their algorithmic 240 00:15:43,560 --> 00:15:48,680 Speaker 1: reach never actually contacts our skin, and so that will 241 00:15:48,720 --> 00:15:51,000 Speaker 1: be continued to be sought. 242 00:15:51,800 --> 00:15:52,120 Speaker 2: Now. 243 00:15:52,280 --> 00:15:56,120 Speaker 1: The second point to raise about whether AI partners can 244 00:15:56,240 --> 00:15:59,640 Speaker 1: displace real human partners is that there's a sense in 245 00:15:59,680 --> 00:16:03,440 Speaker 1: which which fake partners have always been around. Just look 246 00:16:03,480 --> 00:16:06,080 Speaker 1: at a book, look at a movie, look at any 247 00:16:06,120 --> 00:16:11,920 Speaker 1: TV show. You have beautiful Hollywood actors and actresses, and 248 00:16:11,960 --> 00:16:16,840 Speaker 1: they have flawless skin and perfectly quafft hair and no 249 00:16:16,960 --> 00:16:19,720 Speaker 1: hair where they shouldn't, and they have glittering white teeth. 250 00:16:20,040 --> 00:16:23,240 Speaker 1: They are the epitome of health, and they always say 251 00:16:23,280 --> 00:16:26,760 Speaker 1: the right thing, and you get to be the protagonist 252 00:16:26,920 --> 00:16:30,280 Speaker 1: and enjoy experiencing that relationship. 253 00:16:30,640 --> 00:16:31,720 Speaker 2: You find the. 254 00:16:31,720 --> 00:16:34,000 Speaker 1: Partner, and lose the partner, and then an act five 255 00:16:34,080 --> 00:16:38,640 Speaker 1: you regain the relationship with an epic kiss. This kind 256 00:16:38,680 --> 00:16:42,680 Speaker 1: of fake relationship in books and movies isn't exactly the 257 00:16:42,720 --> 00:16:46,200 Speaker 1: same as an AI relationship, but it has some similarities. 258 00:16:46,240 --> 00:16:52,080 Speaker 1: They both represent a platonic ideal, a perfect relationship with 259 00:16:52,240 --> 00:16:55,920 Speaker 1: someone who always says the right thing. We never see 260 00:16:55,960 --> 00:17:00,440 Speaker 1: a love interest in the movie who is distracted or angry, 261 00:17:00,920 --> 00:17:04,119 Speaker 1: or interested in someone else, or just really busy with work, 262 00:17:04,160 --> 00:17:07,440 Speaker 1: too busy to spend time with you when you need them. 263 00:17:07,720 --> 00:17:09,920 Speaker 1: You never see a love interest in the movies who 264 00:17:10,280 --> 00:17:13,359 Speaker 1: waste a lot of time taking selfies and trying to 265 00:17:13,400 --> 00:17:18,119 Speaker 1: build a meaningless reputation on TikTok. People have no meaningful 266 00:17:18,240 --> 00:17:21,639 Speaker 1: foibles in a good love story, in a book or 267 00:17:21,680 --> 00:17:26,119 Speaker 1: on television. Now, I've often wondered if we, in a 268 00:17:26,240 --> 00:17:29,679 Speaker 1: sense get cursed by the fairy tales we're surrounded with 269 00:17:29,800 --> 00:17:31,720 Speaker 1: when we're looking for actual love. 270 00:17:32,119 --> 00:17:32,959 Speaker 2: But I don't know. 271 00:17:33,440 --> 00:17:37,480 Speaker 1: Perhaps those fairy tales help us past all the difficult 272 00:17:37,520 --> 00:17:40,399 Speaker 1: stuff in a relationship. They get us to ignore the 273 00:17:40,480 --> 00:17:44,920 Speaker 1: imperfect things because we believe so strongly in the possibility 274 00:17:45,320 --> 00:17:46,760 Speaker 1: of a perfect relationship. 275 00:17:47,000 --> 00:17:48,120 Speaker 2: So think about it this way. 276 00:17:48,240 --> 00:17:51,880 Speaker 1: Say you were a space alien who had never watched 277 00:17:52,040 --> 00:17:54,760 Speaker 1: or read a love story, and you had no concept 278 00:17:54,800 --> 00:17:57,560 Speaker 1: of that. And the question is when you met someone 279 00:17:57,640 --> 00:18:00,760 Speaker 1: which you think, Wow, they seem to have very different 280 00:18:00,800 --> 00:18:03,399 Speaker 1: opinions than I do. They think like this, and I 281 00:18:03,400 --> 00:18:06,399 Speaker 1: think like that, And they also spend some fraction of 282 00:18:06,400 --> 00:18:08,680 Speaker 1: their time getting snippy at me or staring at their 283 00:18:08,680 --> 00:18:09,719 Speaker 1: cell phone or whatever. 284 00:18:10,000 --> 00:18:13,040 Speaker 2: So there's no way this can work. I don't know. 285 00:18:13,160 --> 00:18:16,040 Speaker 1: I'm just speculating here, but I do wonder if seeing 286 00:18:16,200 --> 00:18:20,119 Speaker 1: lots of models of love stories gives us the tools 287 00:18:20,520 --> 00:18:23,760 Speaker 1: to view things in a more optimistic light, and that 288 00:18:24,160 --> 00:18:27,960 Speaker 1: actually gives a chance to make the relationship work. In 289 00:18:28,000 --> 00:18:33,359 Speaker 1: other words, it provides some aspirational glue where otherwise things 290 00:18:33,400 --> 00:18:34,679 Speaker 1: would just fall apart. 291 00:18:35,440 --> 00:18:36,320 Speaker 2: Now, the counter. 292 00:18:36,160 --> 00:18:39,959 Speaker 1: Argument, of course, is that all these fantasies set you 293 00:18:40,040 --> 00:18:44,120 Speaker 1: up with false expectations about love and relationships, which makes 294 00:18:44,160 --> 00:18:48,200 Speaker 1: it harder to keep the relationship together once you see 295 00:18:48,200 --> 00:18:53,879 Speaker 1: some degree of realism and disenchantment sets in. In any case, 296 00:18:54,160 --> 00:18:56,760 Speaker 1: even if we do get cursed by these fairy tales 297 00:18:56,800 --> 00:18:59,480 Speaker 1: in some way, it's still the case that there's nothing 298 00:18:59,560 --> 00:19:05,480 Speaker 1: new about fantasy relationships. Now, maybe you argue this is 299 00:19:05,560 --> 00:19:10,040 Speaker 1: different because instead of the Julia Roberts movie that everyone watches, 300 00:19:10,359 --> 00:19:14,399 Speaker 1: it's now something that is bespoke just for you. It's 301 00:19:14,440 --> 00:19:18,440 Speaker 1: a one on one relationship, And maybe that's an important difference. 302 00:19:18,480 --> 00:19:21,560 Speaker 1: But just keep in mind the way that we humans 303 00:19:21,760 --> 00:19:25,600 Speaker 1: enjoy literature is by living inside the story. You are 304 00:19:25,680 --> 00:19:29,880 Speaker 1: essentially having a one on one relationship with Julia Roberts. 305 00:19:30,440 --> 00:19:33,840 Speaker 1: So perhaps it's not the privacy of the relationship, but 306 00:19:33,880 --> 00:19:38,240 Speaker 1: instead the meaningful difference with an AI relationship is the 307 00:19:38,400 --> 00:19:42,320 Speaker 1: bi directional nature of it. Instead of watching a movie 308 00:19:42,359 --> 00:19:46,399 Speaker 1: where you're simply hearing other characters say lines and Julia 309 00:19:46,480 --> 00:19:50,000 Speaker 1: Roberts responds, you are now the one coming up with 310 00:19:50,040 --> 00:19:53,960 Speaker 1: the lines. You are deciding what to say. So maybe 311 00:19:54,000 --> 00:19:58,040 Speaker 1: this makes a difference. I suspect it enhances the degree 312 00:19:58,240 --> 00:20:03,040 Speaker 1: of the fantasy. We have yet to see whether AI 313 00:20:03,160 --> 00:20:07,399 Speaker 1: will meaningfully replace people's pursuits of other humans because it 314 00:20:07,480 --> 00:20:11,040 Speaker 1: is touchless and smellless, and it's not clear what the 315 00:20:11,160 --> 00:20:15,520 Speaker 1: impact is of holding fantasy relationships because we already do 316 00:20:15,640 --> 00:20:19,280 Speaker 1: that with book characters and movie stars. So this is 317 00:20:19,320 --> 00:20:22,919 Speaker 1: going to require many years of real world data to 318 00:20:23,000 --> 00:20:26,280 Speaker 1: get a real bead on the impact here. Okay, Now, 319 00:20:26,359 --> 00:20:29,199 Speaker 1: whatever you think about AI companions, I have noticed in 320 00:20:29,280 --> 00:20:32,840 Speaker 1: conversations with my friends, especially those who are married, a 321 00:20:33,080 --> 00:20:37,240 Speaker 1: question that floats quickly to the surface. Is it cheating 322 00:20:37,680 --> 00:20:40,520 Speaker 1: to have a relationship on your phone with a non 323 00:20:40,600 --> 00:20:43,800 Speaker 1: real person? And there are different levels, of course of 324 00:20:43,880 --> 00:20:47,000 Speaker 1: what an AI relationship could be what if it's just 325 00:20:47,080 --> 00:20:50,520 Speaker 1: an app like replica that checks in with you like 326 00:20:50,600 --> 00:20:52,919 Speaker 1: a friend who cares about you, and you can just 327 00:20:53,320 --> 00:20:56,119 Speaker 1: chat innocently with it. This is the free version of 328 00:20:56,160 --> 00:20:58,800 Speaker 1: the app okay, So that's one level. But what if 329 00:20:58,800 --> 00:21:01,600 Speaker 1: you go in for the paid verse, where the conversation 330 00:21:01,720 --> 00:21:05,320 Speaker 1: with the avatar becomes more spicy, And what if the 331 00:21:05,359 --> 00:21:10,959 Speaker 1: cartoon like avatar is highly attractive and dressed provocatively and 332 00:21:11,119 --> 00:21:15,679 Speaker 1: is extremely suggestive in what she says. So I've informally 333 00:21:15,720 --> 00:21:19,159 Speaker 1: surveyed several married friends about this, and it seems clear 334 00:21:19,640 --> 00:21:23,879 Speaker 1: that opinions are all over the spectrum. Some wives and 335 00:21:23,960 --> 00:21:27,359 Speaker 1: husbands feel fine about having their partner have an AI 336 00:21:27,480 --> 00:21:31,720 Speaker 1: relationship on the side, and others said no way. Now, 337 00:21:31,760 --> 00:21:34,959 Speaker 1: for those who said no way, this is presumably because 338 00:21:35,000 --> 00:21:38,960 Speaker 1: the issue plugs into very deep circuitry in their brain. 339 00:21:39,040 --> 00:21:42,600 Speaker 1: It's interpreted as a threat to the relationship, and we 340 00:21:42,680 --> 00:21:47,919 Speaker 1: are hardwired to fight against that. From an evolutionary perspective, 341 00:21:47,920 --> 00:21:50,919 Speaker 1: what you want is for your mate to stick around 342 00:21:50,960 --> 00:21:55,359 Speaker 1: and provide resources and child rearing, and anything that represents 343 00:21:55,400 --> 00:21:58,879 Speaker 1: a threat to that is to be fought against. Now, 344 00:21:59,080 --> 00:22:02,440 Speaker 1: the part that seems here is that an AI avatar 345 00:22:02,880 --> 00:22:07,600 Speaker 1: would not represent a direct threat in this evolutionary sense. 346 00:22:07,680 --> 00:22:12,240 Speaker 1: You can't go and impregnate or be impregnated by AI, 347 00:22:13,480 --> 00:22:18,440 Speaker 1: but nonetheless your attention might be stolen away to some degree, 348 00:22:18,480 --> 00:22:22,160 Speaker 1: possibly to a large degree, and beyond an evolutionary threat. 349 00:22:22,200 --> 00:22:24,520 Speaker 1: A big part of what people get out of a 350 00:22:24,600 --> 00:22:28,679 Speaker 1: relationship is the love and the attention that we all crave. 351 00:22:29,440 --> 00:22:32,439 Speaker 1: So many people feel that they just don't want the 352 00:22:32,520 --> 00:22:35,479 Speaker 1: AI bought to steal away even a fraction of that. 353 00:22:35,960 --> 00:22:38,840 Speaker 1: A partner only has so much love and attention to 354 00:22:38,920 --> 00:22:41,600 Speaker 1: give in a day, and you don't want half of 355 00:22:41,640 --> 00:22:46,239 Speaker 1: it getting siphoned off to someone or something else. This 356 00:22:46,560 --> 00:22:51,280 Speaker 1: shares some similarities to the situation of a person having 357 00:22:51,400 --> 00:22:54,119 Speaker 1: an X that they still talk with, and if a 358 00:22:54,160 --> 00:22:57,920 Speaker 1: person talks very intimately with their ex, a spouse might 359 00:22:57,960 --> 00:23:01,400 Speaker 1: feel like she or he doesn't really love that. Now 360 00:23:01,440 --> 00:23:04,040 Speaker 1: when it comes to the X, if you were making 361 00:23:04,119 --> 00:23:08,199 Speaker 1: the evolutionary argument, you could argue that the fear is 362 00:23:08,240 --> 00:23:11,399 Speaker 1: that on a lonely night, in the middle of a conflict, 363 00:23:11,680 --> 00:23:14,520 Speaker 1: your partner might make a bad choice and slip back 364 00:23:14,560 --> 00:23:18,320 Speaker 1: into a physical relationship, and so that relationship with the 365 00:23:18,520 --> 00:23:23,000 Speaker 1: X feels like a threat. But obviously the kernal cheating 366 00:23:23,200 --> 00:23:26,320 Speaker 1: can't happen with the AI bought, and yet the fear 367 00:23:26,400 --> 00:23:29,159 Speaker 1: is still there. So that indicates one of two things. 368 00:23:29,359 --> 00:23:36,080 Speaker 1: Either are evolutionarily programmed deep fears simply can't make that distinction, 369 00:23:36,920 --> 00:23:40,159 Speaker 1: or it doesn't have to do with a future threat 370 00:23:40,200 --> 00:23:43,560 Speaker 1: of physical infidelity, but instead it's just this issue about 371 00:23:43,840 --> 00:23:47,960 Speaker 1: somebody else having that emotional intimacy with your partner, which 372 00:23:48,000 --> 00:23:49,040 Speaker 1: steals away. 373 00:23:48,840 --> 00:23:50,960 Speaker 2: Attentional resources from you. 374 00:23:52,240 --> 00:23:55,040 Speaker 1: Now this gets more interesting when we start thinking about 375 00:23:55,080 --> 00:23:58,880 Speaker 1: having physical robots that can play a role in your life. Now, 376 00:23:58,960 --> 00:24:00,439 Speaker 1: this is probably not going to have happened in the 377 00:24:00,480 --> 00:24:04,360 Speaker 1: next few years, but fast forward a century and certainly 378 00:24:04,400 --> 00:24:08,400 Speaker 1: everyone's going to face this scenario. Your partner can buy 379 00:24:08,560 --> 00:24:11,199 Speaker 1: not just a mechanical device or blow up doll, but 380 00:24:11,320 --> 00:24:16,760 Speaker 1: can now have a convincing and attentive physical partner. So 381 00:24:16,840 --> 00:24:19,000 Speaker 1: the question is what if your spouse can get not 382 00:24:19,119 --> 00:24:24,480 Speaker 1: only the emotional intimacy but also the physical intimacy. So 383 00:24:24,560 --> 00:24:27,399 Speaker 1: the people I surveyed about this who found the AI 384 00:24:27,480 --> 00:24:30,919 Speaker 1: bought online a threat seem to find this idea of 385 00:24:30,960 --> 00:24:35,679 Speaker 1: an AI physical robot even a larger threat. Now, the 386 00:24:35,760 --> 00:24:37,840 Speaker 1: interesting thing is that I can point out that there's 387 00:24:37,880 --> 00:24:40,440 Speaker 1: a sense in which none of this is different from 388 00:24:40,480 --> 00:24:44,120 Speaker 1: what their spouse might do anyway, in terms of finding 389 00:24:44,200 --> 00:24:48,359 Speaker 1: adult content on the internet and cheating in that way. 390 00:24:49,240 --> 00:24:52,160 Speaker 1: But I think people have a reaction to internet surfing 391 00:24:52,480 --> 00:24:55,439 Speaker 1: for the same reasons as they have the reaction to 392 00:24:55,480 --> 00:24:58,639 Speaker 1: the AI bot, which is simply that there is less 393 00:24:58,720 --> 00:25:21,600 Speaker 1: time and intimacy an attention toward them. This certainly won't 394 00:25:21,600 --> 00:25:24,560 Speaker 1: apply to everyone, but the very general impression I've had 395 00:25:24,600 --> 00:25:28,720 Speaker 1: from talking with people at different stages of marriage is 396 00:25:28,760 --> 00:25:31,800 Speaker 1: that at the beginning of a relationship, people have a 397 00:25:32,000 --> 00:25:36,520 Speaker 1: stronger reaction against AI relationships. They don't want their partner 398 00:25:36,560 --> 00:25:37,640 Speaker 1: to be distracted. 399 00:25:38,359 --> 00:25:39,240 Speaker 2: But people who have. 400 00:25:39,240 --> 00:25:41,680 Speaker 1: Been in a relationship for a long time and have 401 00:25:41,880 --> 00:25:45,080 Speaker 1: kids will sometimes see this as a way to get 402 00:25:45,119 --> 00:25:47,760 Speaker 1: their spouse out of their hair, and they can be 403 00:25:47,800 --> 00:25:51,440 Speaker 1: happy for the spouse because it addresses their spouse's needs 404 00:25:51,520 --> 00:25:55,400 Speaker 1: and slakes their attention. In other words, they love their 405 00:25:55,440 --> 00:25:58,159 Speaker 1: spouse as a partner, and they see this as a 406 00:25:58,200 --> 00:26:02,480 Speaker 1: way for their partner to fill in needs for intimacy 407 00:26:02,800 --> 00:26:06,119 Speaker 1: and attention in a way that's innocent and has no 408 00:26:06,240 --> 00:26:11,119 Speaker 1: meaningful health risks like STDs. So what's become clear to 409 00:26:11,160 --> 00:26:13,399 Speaker 1: me is that there's no single answer for how a 410 00:26:13,480 --> 00:26:18,119 Speaker 1: spouse feels or should feel about AI relationships. Some people 411 00:26:18,320 --> 00:26:20,600 Speaker 1: are against it, some people think it's a great idea, 412 00:26:20,960 --> 00:26:24,480 Speaker 1: and many people are still somewhere in between or still 413 00:26:24,800 --> 00:26:26,200 Speaker 1: making up their minds. 414 00:26:27,760 --> 00:26:30,040 Speaker 2: Now, I want to switch gears. 415 00:26:29,840 --> 00:26:32,960 Speaker 1: From what this means to the partner back to what 416 00:26:33,000 --> 00:26:35,720 Speaker 1: it means to the brain of the person who. 417 00:26:35,600 --> 00:26:40,320 Speaker 2: Is receiving the intimacy. So let's recall the movie Her. 418 00:26:40,640 --> 00:26:44,200 Speaker 1: It's about this guy named Theodore who's played by Joquin Phoenix, 419 00:26:44,560 --> 00:26:48,240 Speaker 1: and his marriage ends and he's left heartbroken, and he 420 00:26:48,320 --> 00:26:52,520 Speaker 1: becomes intrigued by a new app. It's actually an operating 421 00:26:52,560 --> 00:26:56,040 Speaker 1: system in which he can launch this program, and he 422 00:26:56,119 --> 00:27:00,480 Speaker 1: meets Samantha, who's just a voice played by Scarletto Hansen, 423 00:27:01,000 --> 00:27:05,399 Speaker 1: and Samantha is sensitive and playful, and this ends up 424 00:27:05,400 --> 00:27:10,000 Speaker 1: becoming a good friendship. But soon it deepens into love 425 00:27:10,440 --> 00:27:14,239 Speaker 1: and he has this relationship with an AI bought, and 426 00:27:14,280 --> 00:27:18,920 Speaker 1: this relationship means everything to him. Now the film has 427 00:27:19,000 --> 00:27:22,359 Speaker 1: an incredible ending because in the final act he comes 428 00:27:22,400 --> 00:27:27,679 Speaker 1: to understand that she has been having this relationship with 429 00:27:27,880 --> 00:27:31,040 Speaker 1: hundreds of thousands of other men all at the same time, 430 00:27:31,600 --> 00:27:35,840 Speaker 1: because she is computational and operates at a totally different 431 00:27:36,240 --> 00:27:41,080 Speaker 1: timescale and can process what appears to be intimate conversation 432 00:27:41,680 --> 00:27:46,280 Speaker 1: at a rate millions of times faster than our poorer brains, 433 00:27:46,600 --> 00:27:50,720 Speaker 1: And so she's maintaining this intimacy with hundreds of thousands 434 00:27:50,720 --> 00:27:55,240 Speaker 1: of others. And the movie opened up the question, how 435 00:27:55,320 --> 00:27:59,800 Speaker 1: should Theodore feel about that? Is the intimacy real if 436 00:27:59,840 --> 00:28:02,760 Speaker 1: it shared with a city full of other men? Is 437 00:28:02,800 --> 00:28:06,720 Speaker 1: the relationship real if she lives on a timescale many 438 00:28:06,760 --> 00:28:08,960 Speaker 1: millions of times faster than yours? 439 00:28:09,040 --> 00:28:09,760 Speaker 2: Does it matter? 440 00:28:10,240 --> 00:28:13,960 Speaker 1: Should he still feel the titillation of her saying something 441 00:28:14,119 --> 00:28:17,000 Speaker 1: sweet and kind to him just when he needs it? 442 00:28:17,119 --> 00:28:21,399 Speaker 1: These were the questions launched by that particular movie. So 443 00:28:21,640 --> 00:28:24,200 Speaker 1: I'm going to suggest a direction here that I don't 444 00:28:24,200 --> 00:28:27,280 Speaker 1: believe anyone is thinking about, certainly not in Silicon Valley, 445 00:28:27,280 --> 00:28:30,040 Speaker 1: where everything is about leveraging the power of AI to 446 00:28:30,320 --> 00:28:33,840 Speaker 1: scale a product to millions or billions of people. What 447 00:28:33,920 --> 00:28:36,280 Speaker 1: I'm thinking about instead is from the point of view 448 00:28:36,560 --> 00:28:40,680 Speaker 1: of neuroscience, and the goal is not scaling, but instead 449 00:28:41,120 --> 00:28:45,120 Speaker 1: focusing on the life of an individual and the specific 450 00:28:45,360 --> 00:28:49,800 Speaker 1: details of what has shaped his or her brain. So 451 00:28:49,840 --> 00:28:52,320 Speaker 1: I'm going to tell you my idea, but first I'm 452 00:28:52,360 --> 00:28:55,000 Speaker 1: going to start far away and we'll come back around 453 00:28:55,000 --> 00:28:58,040 Speaker 1: to this. So I spun off a company from my 454 00:28:58,160 --> 00:29:00,840 Speaker 1: lab called Neo Censuries some years ago, and one of 455 00:29:00,880 --> 00:29:05,080 Speaker 1: our inventions is a risk band to replace hearing aids. 456 00:29:05,120 --> 00:29:08,720 Speaker 1: Because the risk band listens in real time for high 457 00:29:08,760 --> 00:29:12,400 Speaker 1: frequency parts of speech, and it vibrates to tell you, oh, 458 00:29:12,400 --> 00:29:14,080 Speaker 1: there was an s oh, I just heard a te 459 00:29:14,320 --> 00:29:17,080 Speaker 1: Oh that was a K, and so it clarifies what's 460 00:29:17,080 --> 00:29:20,960 Speaker 1: happening at the high frequencies and that helps people with 461 00:29:21,160 --> 00:29:25,840 Speaker 1: age related hearing loss to understand what word was just said. Now, 462 00:29:26,080 --> 00:29:29,280 Speaker 1: to make the risk band good at detecting these high 463 00:29:29,280 --> 00:29:33,000 Speaker 1: frequency parts of speech, we needed to train a massive 464 00:29:33,040 --> 00:29:38,560 Speaker 1: neural network with six thousand hours of audio books. But 465 00:29:38,640 --> 00:29:42,120 Speaker 1: it turns out that people with high frequency hearing loss 466 00:29:42,520 --> 00:29:47,200 Speaker 1: have a difficult time understanding, for example, children, because their 467 00:29:47,320 --> 00:29:52,200 Speaker 1: voices are higher frequency and there are no audio books 468 00:29:52,560 --> 00:29:55,600 Speaker 1: read by children. So we had no way to train 469 00:29:55,680 --> 00:29:58,880 Speaker 1: the neural network with any kind of massive data from 470 00:29:59,040 --> 00:29:59,920 Speaker 1: children's voice. 471 00:30:00,840 --> 00:30:01,920 Speaker 2: So here's what we did. 472 00:30:02,560 --> 00:30:05,600 Speaker 1: Led by one of our engineers, Yong Yee, we had 473 00:30:05,640 --> 00:30:09,080 Speaker 1: my eight year old daughter read forty five seconds of 474 00:30:09,240 --> 00:30:13,520 Speaker 1: text into a microphone, and then with some keystrokes, Yong 475 00:30:13,600 --> 00:30:18,840 Speaker 1: Yee turned that into her voice reading six thousand hours 476 00:30:18,880 --> 00:30:22,200 Speaker 1: worth of books. And then we trained up to the 477 00:30:22,200 --> 00:30:24,760 Speaker 1: neural network on that corpus, and we did the same 478 00:30:24,800 --> 00:30:27,800 Speaker 1: thing for my eleven year old boy. So now I 479 00:30:27,840 --> 00:30:31,400 Speaker 1: can listen to any book read by my children as 480 00:30:31,440 --> 00:30:34,880 Speaker 1: though they took the tens of hours to sit down 481 00:30:35,240 --> 00:30:38,480 Speaker 1: and read the book in the studio to me. So 482 00:30:38,600 --> 00:30:42,320 Speaker 1: this technology which exists now, which allows you to capture 483 00:30:42,680 --> 00:30:46,880 Speaker 1: the cadence and prosity of a voice, this gave us 484 00:30:46,880 --> 00:30:51,200 Speaker 1: a really straightforward solution to a problem. But this technology 485 00:30:51,240 --> 00:30:54,280 Speaker 1: has also led to many legal and ethical questions, for 486 00:30:54,320 --> 00:30:58,240 Speaker 1: example about celebrity voices, like can you use John Lennon's 487 00:30:58,320 --> 00:31:01,040 Speaker 1: voice to sing you a personalized song to get you 488 00:31:01,120 --> 00:31:05,160 Speaker 1: to sleep? There are all kinds of legal battles blossoming 489 00:31:05,200 --> 00:31:07,560 Speaker 1: as people try to figure out the rules around this. 490 00:31:08,080 --> 00:31:10,240 Speaker 1: But I'm not going to talk about that today because 491 00:31:10,280 --> 00:31:14,040 Speaker 1: my interest is in finding this single voice, or maybe 492 00:31:14,080 --> 00:31:19,480 Speaker 1: small handful of voices that have meaning to your brain uniquely. 493 00:31:20,000 --> 00:31:23,000 Speaker 1: Here's what I mean. When we did this project with 494 00:31:23,080 --> 00:31:27,400 Speaker 1: my kids' voices. That got me thinking because my father 495 00:31:28,000 --> 00:31:30,360 Speaker 1: passed away three and a half years ago, and he 496 00:31:30,440 --> 00:31:33,280 Speaker 1: was a major influence in my life and I miss him. 497 00:31:33,520 --> 00:31:36,080 Speaker 1: So I went through my old videos and found some 498 00:31:36,120 --> 00:31:40,040 Speaker 1: short clips of him speaking, and I wondered if there 499 00:31:40,080 --> 00:31:43,360 Speaker 1: was enough there that I could actually make an AI 500 00:31:43,560 --> 00:31:46,600 Speaker 1: bought out of his voice so I could hear him 501 00:31:46,640 --> 00:31:49,400 Speaker 1: speak whenever I wanted to. And it made me wonder 502 00:31:49,440 --> 00:31:52,360 Speaker 1: about the degree to which that's a healthy thing, But 503 00:31:52,440 --> 00:31:56,400 Speaker 1: I decided there was nothing bad about it. What a 504 00:31:56,600 --> 00:31:59,840 Speaker 1: pleasure to be able to hear my dad's voice for 505 00:32:00,120 --> 00:32:02,840 Speaker 1: the rest of my life and to have that trigger 506 00:32:02,920 --> 00:32:06,080 Speaker 1: my fond memories of him. Wouldn't it be cool to 507 00:32:06,160 --> 00:32:09,360 Speaker 1: have him read audiobooks to me in the way that 508 00:32:09,440 --> 00:32:12,800 Speaker 1: he read to me when I was a child, And 509 00:32:12,840 --> 00:32:14,520 Speaker 1: I thought about what it would be like for my 510 00:32:15,040 --> 00:32:18,600 Speaker 1: mother if I programmed a sentence to her in his voice, 511 00:32:19,440 --> 00:32:21,719 Speaker 1: like I love you and I'm thinking about you. 512 00:32:22,440 --> 00:32:24,440 Speaker 3: I love you and I'm thinking about you. 513 00:32:25,080 --> 00:32:28,040 Speaker 1: And when I turn ninety years old, wouldn't it be 514 00:32:28,080 --> 00:32:31,440 Speaker 1: amazing to hear him say Happy birthday, David, just like 515 00:32:31,480 --> 00:32:32,960 Speaker 1: he did when I was a kid. 516 00:32:32,880 --> 00:32:36,200 Speaker 3: Happy ninetieth Birthday, David. I hope this orbit is the 517 00:32:36,240 --> 00:32:37,640 Speaker 3: best one yet. 518 00:32:37,440 --> 00:32:40,360 Speaker 1: Or at New Year's eves into the future for the 519 00:32:40,400 --> 00:32:42,680 Speaker 1: rest of my life, he can wish me the best, 520 00:32:42,840 --> 00:32:43,960 Speaker 1: even though he will. 521 00:32:43,800 --> 00:32:46,600 Speaker 2: Have been gone from the planet for a long time. 522 00:32:46,840 --> 00:32:50,160 Speaker 3: Happy New Year. I can't believe it's already twenty fifty three. 523 00:32:50,800 --> 00:32:53,240 Speaker 1: And what I realized as I was reaching my arms 524 00:32:53,280 --> 00:32:57,000 Speaker 1: down into this is how powerful this technology is going 525 00:32:57,080 --> 00:33:00,320 Speaker 1: to be, because it will be so compelling. I'm not 526 00:33:00,640 --> 00:33:03,640 Speaker 1: talking here about the issue of using somebody's voice to 527 00:33:03,800 --> 00:33:07,040 Speaker 1: fake an ATM transaction, or fake a kidnapping, or any 528 00:33:07,080 --> 00:33:09,520 Speaker 1: of the AI concerns that people have expressed. 529 00:33:09,960 --> 00:33:11,480 Speaker 2: Instead, what I'm. 530 00:33:11,280 --> 00:33:15,680 Speaker 1: Talking about is the unbelievably compelling way that an AI 531 00:33:15,840 --> 00:33:22,080 Speaker 1: voice could mean something to you emotionally. After all, I 532 00:33:22,160 --> 00:33:24,840 Speaker 1: grew up my entire life from the moment I was 533 00:33:24,880 --> 00:33:28,920 Speaker 1: born hearing my father's voice. It's so embedded in my 534 00:33:28,960 --> 00:33:33,120 Speaker 1: neural circuitry that a voice with exactly that cadence and 535 00:33:33,240 --> 00:33:38,600 Speaker 1: prosity would have enormous emotional sway on me. And again, 536 00:33:38,640 --> 00:33:41,120 Speaker 1: I'm not talking about all the bad things that could 537 00:33:41,160 --> 00:33:45,280 Speaker 1: be done with that. Instead, because this episode is about relationships, 538 00:33:45,320 --> 00:33:48,400 Speaker 1: I'm talking about what it would be like and how 539 00:33:48,440 --> 00:33:53,880 Speaker 1: I could leverage the intimate nature of that relationship. For example, 540 00:33:54,520 --> 00:33:56,920 Speaker 1: let's say that I wanted to get myself to stop 541 00:33:56,960 --> 00:33:59,680 Speaker 1: doing something I don't drink, But let's say I did, 542 00:33:59,760 --> 00:34:03,120 Speaker 1: and that I wanted to stop drinking. So imagine in 543 00:34:03,160 --> 00:34:06,720 Speaker 1: the near future, I build an app that tracks my 544 00:34:06,840 --> 00:34:10,160 Speaker 1: GPS location, and when it sees I'm about to walk 545 00:34:10,200 --> 00:34:13,600 Speaker 1: into a bar, it launches my father's voice in my 546 00:34:13,760 --> 00:34:17,600 Speaker 1: ear telling me, Hey, David, don't do this, Hey. 547 00:34:17,520 --> 00:34:21,240 Speaker 3: David, don't do this. I believe in you. I believe 548 00:34:21,280 --> 00:34:23,239 Speaker 3: that you have the strength to resist this. 549 00:34:23,640 --> 00:34:28,480 Speaker 1: I think that would be extraordinarily compelling. This would be 550 00:34:28,520 --> 00:34:33,399 Speaker 1: a technique to plug into a relationship that already exists 551 00:34:33,680 --> 00:34:36,960 Speaker 1: deep in my neural networks, and it could leverage that 552 00:34:37,520 --> 00:34:40,759 Speaker 1: for good. So this is a way that we can 553 00:34:41,000 --> 00:34:45,720 Speaker 1: right now take a loved voice and extend your parent. 554 00:34:45,840 --> 00:34:49,560 Speaker 1: Let's say, past what Homo sapiens can normally do. They 555 00:34:49,560 --> 00:34:54,200 Speaker 1: can live on beyond their passing away to keep playing 556 00:34:54,280 --> 00:34:57,719 Speaker 1: a role in your life. Now there's a sense in 557 00:34:57,760 --> 00:35:01,399 Speaker 1: which you might say, well, there's nothing new here. If 558 00:35:01,440 --> 00:35:04,480 Speaker 1: your parents wrote you a letter, you might find that 559 00:35:04,640 --> 00:35:08,640 Speaker 1: years after they've passed, and the invention of writing is 560 00:35:08,680 --> 00:35:12,000 Speaker 1: a way of lasting well past your death and reaching 561 00:35:12,080 --> 00:35:16,680 Speaker 1: out to people at great distances and across great time chasms. 562 00:35:17,280 --> 00:35:19,640 Speaker 1: But what is new is that I can get my 563 00:35:19,800 --> 00:35:23,440 Speaker 1: father to talk about things that simply didn't exist when 564 00:35:23,480 --> 00:35:27,040 Speaker 1: he was alive. Maybe twenty years from now, I'll look 565 00:35:27,120 --> 00:35:31,719 Speaker 1: up the Wikipedia page about room temperature superconductivity, and I'll 566 00:35:31,760 --> 00:35:33,959 Speaker 1: get to listen to it in his voice, like he's 567 00:35:34,200 --> 00:35:36,200 Speaker 1: teaching me something the way he used to do when 568 00:35:36,200 --> 00:35:40,000 Speaker 1: I was a little kid. So the part that is 569 00:35:40,120 --> 00:35:42,760 Speaker 1: new is not the reach of a human but instead 570 00:35:42,800 --> 00:35:45,040 Speaker 1: the emotional component. 571 00:35:44,520 --> 00:35:45,200 Speaker 2: To all of this. 572 00:35:45,360 --> 00:35:49,720 Speaker 1: That is the overlay of a loved one's voice onto 573 00:35:49,760 --> 00:35:52,160 Speaker 1: any possible scenario in the future. 574 00:35:52,760 --> 00:35:54,359 Speaker 2: And the reason this all matters is. 575 00:35:54,320 --> 00:35:59,520 Speaker 1: Because that voice has pathways deep into the forest of 576 00:35:59,560 --> 00:36:00,000 Speaker 1: your nerves. 577 00:36:01,400 --> 00:36:02,399 Speaker 2: So let's wrap up. 578 00:36:03,040 --> 00:36:07,920 Speaker 1: Many companies are launching AI relationship bots, and many researchers 579 00:36:08,000 --> 00:36:11,040 Speaker 1: are exploring what this all means. But I don't really 580 00:36:11,040 --> 00:36:14,120 Speaker 1: think we know the answers yet. It's likely to take 581 00:36:14,239 --> 00:36:17,360 Speaker 1: a whole generation before we know what the effect is. 582 00:36:18,360 --> 00:36:21,880 Speaker 1: Are people discovering a beautiful technique to address. 583 00:36:21,480 --> 00:36:24,440 Speaker 2: The loneliness crisis here and they have. 584 00:36:24,440 --> 00:36:26,480 Speaker 1: Someone to turn to in the middle of the night, 585 00:36:26,800 --> 00:36:30,040 Speaker 1: who says something caring to them and always has their 586 00:36:30,160 --> 00:36:34,160 Speaker 1: best interest in mind. Or are we entering an era 587 00:36:34,680 --> 00:36:39,080 Speaker 1: that exacerbates the loneliness crisis and at worst fills our 588 00:36:39,080 --> 00:36:44,440 Speaker 1: belly with empty calories and counteracts are reproductive mandates like 589 00:36:44,760 --> 00:36:48,560 Speaker 1: a perfect drug that spells the end of the species. 590 00:36:49,680 --> 00:36:51,080 Speaker 2: Only time will tell. 591 00:36:55,440 --> 00:36:58,400 Speaker 1: Go to eagleman dot com slash podcast for more information 592 00:36:58,600 --> 00:37:02,439 Speaker 1: and to find further reading. Send me an email at 593 00:37:02,560 --> 00:37:06,600 Speaker 1: podcast at eagleman dot com with questions or discussion, and 594 00:37:06,640 --> 00:37:09,440 Speaker 1: I'm making monthly episodes in which I address those. 595 00:37:11,400 --> 00:37:12,200 Speaker 2: Until next time. 596 00:37:12,520 --> 00:37:15,840 Speaker 1: I'm David Eagleman, and this is Inner Cosmos.