Kate Darling: Social Robots, Ethics, Privacy and the Future of MIT | Lex Fridman Podcast #329

Kate Darling 0:00 ... I think that animals are a really great thought experiment when we're thinking about AI and robotics, because again, there's comparing them to humans, that leads us down the wrong path, both because it's not accurate, but also I think for the future, we don't want that. We want something that's a supplement. But I think animals, because we've used them throughout history for so many different things, we, we domesticated them not because they do what we do, but because what they do is different, and that's useful. And I, it just, like whether we're talking about companionship, whether we're talking about work integration, whether we're talking about responsibility for harm, there are just so many things we can draw on in that history from these entities that can sense, think, make autonomous decisions, and learn, that are applicable to how we should be thinking about robots and AI.

Lex Fridman 0:51 The following is a conversation with Kate Darling, her second time on the podcast. She's a research scientist at MIT Media Lab interested in human-robot interaction and robot ethics, which she writes about in her recent book called The New Breed: What Our History with Animals Reveals About Our Future with Robots. Kate is one of my favorite people at MIT. She was a courageous voice of reason and compassion through the time of the Jeffrey Epstein scandal at MIT three years ago. We reflect on this time in this very conversation, including the lessons it revealed about human nature and our optimistic vision for the future of MIT, a university we both love and believe in. This is the Lex Fridman Podcast. To support it, please check out our sponsors in the description. And now, dear friends, here's Kate Darling.

Lex Fridman 1:46 Last time we talked a few years back, you wore a Justin Bieber shirt for the podcast.

Kate Darling 1:51 (laughs)

Lex Fridman 1:51 So now looking back, you're a respected, um, researcher, all the amazing accomplishments in robotics, uh, you're an author. Was this one of the proudest moments of your life, uh, proudest decisions you've ever made?

Kate Darling 2:05 Definitely. You handled it really well, though. It was cool 'cause I walked in, I didn't know you were gonna be filming. I walked in and you're in a-

Lex Fridman 2:12 Right.

Kate Darling 2:12 ... fucking suit.

Lex Fridman 2:13 Yeah.

Kate Darling 2:14 And I'm like, "Why are you all dressed up?"

Lex Fridman 2:16 Yeah. (laughs)

Kate Darling 2:17 And then you were so nice about it. You, like made some excuse. You were like, "Oh, well, I'm interviewing some art..." Didn't you say you were interviewing some military general afterwards to like-

Lex Fridman 2:24 Oh yeah, that was a-

Kate Darling 2:25 ... make me feel better?

Lex Fridman 2:26 ... CTO of Lockheed Martin, I think.

Kate Darling 2:28 Oh, that's what it was.

Lex Fridman 2:29 Yeah.

Kate Darling 2:29 You didn't tell me, oh, I was dressed like this.

Lex Fridman 2:31 (laughs) Are you an actual Bieber fan, or was that like one of those T-shirts that's in the back of the closet that you use for painting?

Kate Darling 2:40 I think I bought it for my husband as a joke, and-

Lex Fridman 2:43 Oh, yeah.

Kate Darling 2:43 ... yeah, it was, we were, got renovating a house at the time, and I had worn it to the site.

Lex Fridman 2:48 Got as a joke, and now you wear it. Okay. Have you worn it since? Is this a one-time?

Kate Darling 2:55 No, like how could I touch it again? It was on your podcast.

Lex Fridman 2:58 Yeah, it's hold-

Kate Darling 2:58 Now it's framed.

Lex Fridman 2:59 It's like a wedding dress or something like that. (laughs) You, you don't, you only wear it once. You are the author of The New Breed: What Our History with Animals Reveals About Our Future with Robots. You opened the book with the surprisingly tricky question, "What is a robot?" So let me ask you. Let's try to sneak up to this question. What's a robot?

Kate Darling 3:20 That's not really sneaking up.

Lex Fridman 3:22 It's just asking it.

Kate Darling 3:24 Yeah.

Lex Fridman 3:25 All right, well... (laughs)

Kate Darling 3:26 (laughs) What do you think a robot is?

Lex Fridman 3:29 What I think a robot is, is something that's, has some level of intelligence and some level of magic, that little shine in the eye, you know, that allows you to navigate the uncertainty of, uh, of life. So that means, like autonomous vehicles to me in that sense, uh, are robots because they navigate, uh, the uncertainty, the complexity of life. Obviously social robots are that.

Kate Darling 4:02 I love that. I like that you mention magic 'cause that also... Well, so first of all, I don't define robot definitively in the book because-

Lex Fridman 4:11 Yes.

Kate Darling 4:11 ... there is no definition-

Lex Fridman 4:13 There's no definition.

Kate Darling 4:13 ... that everyone agrees on. And if you look back through time, people have called things robots until they lose the magic because they're more ubiquitous. Like a vending machine used to be called a robot, and now it's not, right? So I do agree with you that there's this magic aspect that, that, that which is how people understand robots. If you ask a roboticist, they have the definition of something that is, well, it has to be physical usually. It's not an AI agent. It has to be embodied. Um, they'll say it has to be able to sense its environment in some way. It has to be able to make a decision autonomously and then act on its environment again. I think that's a pretty good technical definition even though it really breaks down when you come to things like the smartphone, because the smartphone can do all of those things. And most roboticists would not call it a robot. So there's really no, (laughs) no one good definition. But part of why I wrote the book is because people have a definition of robot in their minds that is usually very focused on a comparison of robots to humans. So if you Google image search robot, you get a bunch of humanoid robots, robots with a torso, and head, and two arms, and two legs. And that's the definition of robot that I'm trying to get us away from because I think that it trips us up a lot.

Lex Fridman 5:38 Why does the humanoid form trip us up a lot?

Kate Darling 5:40 Well, because this constant comparison of robots to people, artificial intelligence to human intelligence, first of all, it doesn't make sense from a technical perspective because, you know, the early AI researchers, some of them were trying to recreate human intelligence. Some people still are, and there's a lot to be learned from that academically, et cetera. But, um, that's not where we've ended up. AI doesn't think like people.... we wind up in this fallacy where we're (laughs) where we're comparing these two, um, and we're, when we talk about what intelligence even is, we're often comparing to our own intelligence. And then the second reason this bothers me is because it doesn't make sense. I, I just think it's boring to recreate intelligence that we already have. I see the scientific value of understanding our own intelligence, but from a, like, practical, "What could we use these technologies for?" perspective, it's much more interesting to create something new, to create a skill set that we don't have that we can partner with in what we're trying to achieve.

Lex Fridman 6:44 And it should be, in some deep way, similar to us, but in most ways different, because you still want to have a connection, w- which is why the similarity might be necessary. So-

Kate Darling 6:56 That's what people argue, yes, and I think that's true. So the two arguments for humanoid robots are people need to be able to communicate and relate to robots, and we relate most to things that are like ourselves, and we have a world that's built for humans, so we have stairs and narrow passageways and door handles, and so we need humanoid robots to be able to navigate that. And so you're speaking to the first one, which is absolutely true, but what we know from social robotics and a lot of human/robot interaction research is that you, all you need is something that's enough like a person to, for, for it to give off cues that someone relates to, and that, but that doesn't have to look human or even act human. You can take a robot like R2-D2 and it just, like, beeps and boops, and people love R2-D2, right? Even though it's just, like, a trash can on wheels. And they like R2-D2 more than C-3PO, who's a humanoid. So-

Lex Fridman 7:51 Yeah.

Kate Darling 7:51 ... there's lots of, there's lots of ways to make robots even better than humans in some ways and make us relate more to them.

Lex Fridman 8:00 Yeah, it's kind of amazing the variety of cues that can be used to anthropomorphize the thing, like a glowing orb or something like that-

Kate Darling 8:08 Yeah.

Lex Fridman 8:08 ... or just, just a voice, just, uh, just subtle basic interaction. I think people sometimes over-engineer these things. Like, simplicity can go a really long way.

Kate Darling 8:19 Totally. I mean, ask any animator and they'll know that.

Lex Fridman 8:22 Yeah. Yeah, those are actually... So the people behind Cozmo, the, um, the robot, the right people to design those is animators, like Disney type of people.

Kate Darling 8:36 Yeah.

Lex Fridman 8:36 Versus, like, roboticists. Robotcists, quote unquote, "are mostly clueless."

Kate Darling 8:42 (laughs) Well, no-

Lex Fridman 8:43 It seems like-

Kate Darling 8:43 ... they just have their own discipline that they're very good at and they didn't, don't have...

Lex Fridman 8:47 Yeah, but that, that don't, don't, you know... I feel like robotics of the early 21st century is not going to be the robotics of the later 21st century.

Kate Darling 9:01 Oh, no.

Lex Fridman 9:01 Like, if you call yourself a roboticist, it'll be something very different. Uh, 'cause I, I think more and more, you'll be like a, maybe, like, a control engineer or something, c- controls engineer, like a... You separate, because ultimately all the unsolved, all the big problems of robotics will be in the social aspect, in the interacting with humans aspect, in the, uh, perception, interpreting the world aspect, in the brain part, not the, not the, not the basic control level part.

Kate Darling 9:37 You call it basic. It's actually really, (laughs) really complex.

Lex Fridman 9:40 It's very, very complicated, yes.

Kate Darling 9:41 And that's why... But, like, I think you're so right, and, and, and what a time to be alive-

Lex Fridman 9:46 Yeah.

Kate Darling 9:46 ... because, for me, I just... We've had robots for so long and they've just been behind the scenes, and now finally robots are getting deployed into the world and-

Lex Fridman 9:58 They're coming out of the closet.

Kate Darling 10:00 Yeah! And, and we're seeing all these mistakes that companies are making because they focus so much on the engineering and getting that right and getting the robot to be, even be able to function in a space that it shares with a human.

Lex Fridman 10:12 S- see, what I feel like people don't understand is to solve the perception and control problem, you shouldn't try to just solve the perception and control problem. You should teach the robot how to say, "Oh, shit. I'm sorry. I fucked up."

Kate Darling 10:28 Yeah, or ask for help.

Lex Fridman 10:29 Or for, ask for help or be able to communicate the uncertainty. Yeah, exactly, all of those things, because you can't solve the perception and control. We humans haven't solved it. We, we're really damn good at it, uh, but w- the, the magic is in the, the self-deprecating humor and the self-awareness about where our flaws are, all that kind of stuff.

Kate Darling 10:52 Yeah, and there's a whole body of research in human/robot interaction showing, like, ways to do this. But a lot of these companies haven't... They don't do HRI. They... Like, the, have you seen the grocery store robot in the Stop & Shop?

Lex Fridman 11:06 Yes.

Kate Darling 11:07 Yeah. The Marty, it looks like a giant penis. It's, like, six feet tall. It roams the aisles.

Lex Fridman 11:11 I will never see Marty the same way again. Thank you for that (laughs) .

Kate Darling 11:14 (laughs) You're welcome. But, like, they, th- these poor people worked so hard on getting a functional robot-

Lex Fridman 11:21 Yeah.

Kate Darling 11:22 ... together and then people hate Marty because they didn't at all consider how people would react to Marty in their space.

Lex Fridman 11:30 Does everybody... I mean, you talk about this. Do, do people mostly hate Marty? 'Cause I, I like, I like Marty. I (laughs) I feel a little less-

Kate Darling 11:38 Yeah, but you like Flippy.

Lex Fridman 11:39 Yeah, I do.

Kate Darling 11:39 And actually, like-

Lex Fridman 11:40 There's a, there's a parallel between the two?

Kate Darling 11:43 I believe there is. So we were actually gonna do a study on this right before the pandemic hit and then we canceled it because we didn't want to go to the grocery store and neither did anyone else. Um, but our theory... So this was with a, a student at MIT, Daniela Dipaola. She noticed that everyone on Facebook in her circles was complaining about Marty. They're like, "What is this creepy robot? It's watching me. It's always in the way."... and she did this, like, quick and dirty sentiment analysis on Twitter, where she was looking at positive and negative mentions of the robot, and she found (laughs) that the biggest spike of negative mentions happened, um, when Stop & Shop threw a birthday party for the Marty robots, like with free cake and balloons. Like, who complains about free cake?

Lex Fridman 12:26 Yeah.

Kate Darling 12:26 Well, people who hate Marty, apparently, so.

Lex Fridman 12:29 Yeah.

Kate Darling 12:29 And so we were like, "That's interesting," and then we did this, like, online poll. We used Mechanical Turk, and we tried to get at what people don't like about Marty, and a lot of it wasn't, "Oh, Marty's taking jobs." It was, "Marty is a surveillance robot," which it's not. It looks for spills on the floor. It doesn't actually, (laughs) like, look at any people. Um, it's, it's watching, it's creepy, it's getting in the way. Those were the things that people complained about, and so our hypothesis became, is Marty a real-life Clippy? Because I know, Lex, you love Clippy, but many people hated Clippy. (laughs)

Lex Fridman 13:07 Well, uh, there's a complex thing there. It could be like marriage. A lot of people seem to w- like to complain about marriage, but they secretly love it. So it could be-

Kate Darling 13:16 (laughs)

Lex Fridman 13:16 ... the, the relationship you might have with, uh, with Marty is like, "Oh, there he goes again, doing his stupid surveillance thing," but you g- grow to love the, um... I mean, b- bitching about the thing, that kinda releases a kind of tension, and there's... I mean, s- s- some people, a lot of people show love by sort of, uh, busting each other's chops, you know? Like, making fun of each other. And then if... I think, I think people would really love it if Marty talked back.

Kate Darling 13:50 (laughs)

Lex Fridman 13:50 (laughs) And (laughs) , and, and like, uh, the... there's so many possible options for humor there. One, you can lean in. You can be like, "Yes, I'm an agent of the CIA monitoring your every move," like mocking people that are concerned, you know what I'm saying? Like, (laughs) "Yes, I, I'm watching you because you're so important with your shopping patterns-"

Kate Darling 14:11 (laughs)

Lex Fridman 14:11 "... I'm collecting all this data." Um, or, or just, you know, any kind of making fun of people. I don't know.

Kate Darling 14:18 But I think you hit on what exactly it is, because when it comes to robots or artificial agents, I think people hate them more than they would some other machine or device or object, um, and it might, and it might be that thing (laughs) ... it might be combined with love or, like, whatever it is, it's a more extreme response because they view these things as social agents and not objects. And that was, um... so Clifford Nass was a big human/computer interaction person, and he... his theory about Clippy was that because people viewed Clippy as a social agent, when Clippy was annoying and would like, bother them and interrupt them and like, not remember what they tolds- him, that's when people got upset, because it wasn't fulfilling their social expectations. And so they complained about Clippy more than they would've i- if it had been a dif- like, not an... not a, you know, virtual character.

Lex Fridman 15:14 So is complaining to you a sign that we're on the wrong path with a particular robot? Or is it possible, like, again, like marriage, like family, that there still is a path towards that direction where we can find deep, meaningful relationship?

Kate Darling 15:32 I think we absolutely can find deep, meaningful relationship-

Lex Fridman 15:35 With Marty.

Kate Darling 15:36 ... with robots. And, well, maybe with Marty. I mean, I just would... I would've designed Marty a little differently.

Lex Fridman 15:42 Like how?

Kate Darling 15:43 Um, well-

Lex Fridman 15:43 Isn't there a charm to the clumsiness, the slowness? Like, I'll sometimes just-

Kate Darling 15:47 There is if you're not trying to get through with a shopping cart and a screaming child. You know? There's... I think, I think you could make it charming. I think there are lots of design tricks that they could've used, and one of the things they did, I think without thinking about it at all, is they slapped two big googly eyes on Marty.

Lex Fridman 16:04 Oh, yeah.

Kate Darling 16:05 And I th- I wonder if that contributed maybe to people feeling watched, um, because (laughs) , 'cause it's looking at them, and-

Lex Fridman 16:14 (laughs)

Kate Darling 16:14 (laughs) ... so, like, is there a way to design the robot to do the function that it's doing in a way that doesn't... that people are actually attracted to rather than annoyed by? And there are many ways to do that, but companies aren't thinking about it. Now they're realizing that they should've thought about it.

Lex Fridman 16:31 Yeah. I wonder if there's a way to w... if it would help to make Marty seem like an entity of its own versus, um, the arm of a large corporation. So there's some sense where this is just the camera that's monitoring people versus this is an entity that's a standalone entity, it has its own task, and it's... has its own personal- like, the more personality you give it, the more it feels like it's not sharing data with anybody else. Like, when we see other human beings, our basic assumption is, "Whatever I say to this human being, it's not, like, being immediately sent to the CIA," or whatever that says.

Kate Darling 17:17 Yeah, what I say to you, no one's gonna hear that, right? (laughs)

Lex Fridman 17:19 Yeah. That's true. That's true.

Kate Darling 17:21 No, no, I'm kidding.

Lex Fridman 17:22 Well, you forget it. I mean, you do forget it.

Kate Darling 17:23 Yeah.

Lex Fridman 17:23 I mean, I don't know if the... even with the microphones here, you forget that, that that's happening. But there, f- for some reason, I think probably w- with Marty, um, I think when it's done really crudely and crappily, you start to realize, oh, this is like PR people trying to make a friendly version of a surveillance machine. But, um...

Lex Fridman 17:47 I mean, that reminds me of the slight clumsiness, or significant clumsiness, on the initial releases of the avatars for the metaverse. I don't know. What, what are your... what are your actually thoughts about that? The, the, um, the way, uh, the avatars, the way, like, M- Mark Zuckerberg looks in that world. You know, the, in the, the, the metaverse, the virtual reality world where you can have, like, virtual meetings and stuff like that.

Kate Darling 18:14 Yeah.

Lex Fridman 18:14 Like, how do we get that right? Do you have thoughts about that? 'Cause that's a kind of, uh...... it's a... It- it feels like a similar problem to social robotics, which is how you design a digital virtual world that is c- compelling, uh, when you connect to others there in the same way that physical connection is.

Kate Darling 18:38 Right. I haven't looked into... I mean, I've seen people joking about it on Twitter and, like, posting (laughs) like, the whatever.

Lex Fridman 18:44 Yeah. But the... I mean, have you seen it? 'Cause it- it... There's something you can't quite put into words, uh, that, um, doesn't feel genuine-

Kate Darling 18:52 Yeah.

Lex Fridman 18:53 ... about the way it looks. And so the question is, if you and I were to meet virtually, what should the avatars look like for us to have similar kind of connection? Should it be really simplified? Should it be a little bit more realistic? Should it be cartoonish? Should it be more, um... Better capturing of expressions, uh, in interesting, complex ways, versus, like, cartoonish oversimplified ways?

Kate Darling 19:24 But haven't video games figured this out? I'm not a gamer, so I don't have any examples. But I feel like there's this whole world in video games where they've thought about all of this and depending on the game, they have different, like, avatars, and a lot of the games are about connecting with others. I just... The thing that I don't know is... And again, I haven't looked into this at all, um, I've been, like, (laughs) shockingly not very interested in the metaverse, but they must have poured so much investment into this, um, Meta, and, like, why- why is it so- why are people... Why is it so bad? (laughs) Like-

Lex Fridman 20:02 Well, uh-

Kate Darling 20:03 There's gotta be a reason. There's gotta be some thinking behind it, right?

Lex Fridman 20:07 Well, (laughs) I talked to Carmack about this, uh, J- John Carmack, who's a part-time, um, Oculus CTO. I think, uh, there's several things to say. One is, as you probably know, that- I mean, there's bureaucracy. There's large corporations, and they often... Large corporations have a way of killing the indie kinda artistic flame that's required to create something really compelling. Somehow they make everything boring, 'cause they- they run through this whole process through the PR department, through all that kind of stuff, and- and it- it somehow becomes generic through that process. because there's, like-

Kate Darling 20:49 They strip out anything interesting because it could be controversial? Is that... Or...

Lex Fridman 20:53 Yeah, right. Exactly. Like, um, like, what... I mean, we- we're living through this now, like, with, uh, (laughs) with a lot of people with cancellations and all those kinds of stuff. People are nervous, and nervousness results in, like, the... Like usual, the assholes are ruining everything. But, you know, the magic of human connection is taking risks, of making a risky joke, of- of- of... Like, with people you like, who are not assholes, good people. Like, some of the fun- some of the fun in the metaverse or in video games is, you know, being edgier, being interesting, revealing your personality in interesting ways. Um, in the sexual tension or in, uh... Like, they're definitely paranoid about that, because-

Kate Darling 21:38 Oh, yeah.

Lex Fridman 21:39 ... like, in metaverse, the possibility of sexual assault and sexual harassment and all that kind of stuff, it's- it's obviously very high, but they're, um... So you should be paranoid to some degree, but not too much, because then you remove completely pers- the personality of the whole thing. Then everybody's just like a vanilla bot that, uh... Like, you have to have ability, um, to be a little bit political, to be a little bit edgy, all that kind of stuff. And, uh, large companies tend to suffocate that. So I... But in general, just forget all that. Just the ability to come up w- really cool, beautiful ideas. Um, if you look at, um... I think Grimes tweeted about this. She's very critical about the metaverse. Is that, um, you know, the, uh, i- independent, uh, game designers have solved this problem of how to create something beautiful and interesting and compelling. They- they do a really good job. So you have to let those kinds of minds, the small groups of people design things and let them run with it, l- let them run wild and do edgy stuff. Yeah. But w- otherwise you bec- you get this kind of... You get a Clippy type of situation, right? Which is, like, a very generic looking thing. Um, but even Clippy has some, like... That's kinda wild that you would take a- a paper clip and put eyes on it.

Kate Darling 23:08 And suddenly people are like, "Oh, you're annoying, but you're definitely a social agent."

Lex Fridman 23:12 (laughs) And I just feel like that wouldn't even... That Clippy thing wouldn't even survive Microsoft or Facebook of today-

Kate Darling 23:21 Mm-hmm.

Lex Fridman 23:21 ... Meta of today, 'cause it would be like, wha- there'd be these meetings about, "Why is it a paper clip?" (laughs) Like, "Why don't we... It's not sufficiently friendly. Let's make it..." Uh, you know. And then all of a sudden, the artist that- with whom it originated is killed, and it's all PR marketing people and all of that kind of stuff. Now, they do important work to some degree, but they kill the creativity.

Kate Darling 23:47 I think the killing of the creativity is in the wh- Like, okay. So what I know from social robotics is, like, obviously, if you create agents that... Okay, so take for an example, you'd create a robot that looks like a humanoid, and it's, you know, Sophia or whatever. Now suddenly, you do have all of these issues where, are you reinforcing an unrealistic beauty standard? Are you-

Lex Fridman 24:12 Yeah.

Kate Darling 24:12 ... objectifying women? Uh, why is the robot white? So you ha- But the thing is, I think that with creativity-... you can find a solution that's even better, where you're not even harming anyone and you're creating a robot that looks like, not, not humanoid, but like something that people relate to even more. And now you don't even have any of these bias issues that you're creating. And so how do we create that within companies? Because I don't think it's really about, like I, 'cause I, you know, maybe we disagree on that. I don't think that edginess or humor or interesting things need to be things that harm or hurt people or that people are against. There are ways to find things that everyone is fine with. Why aren't we doing that?

Lex Fridman 25:01 The problem is, there's departments that look for harm in things.

Kate Darling 25:05 Yeah.

Lex Fridman 25:06 And so they will find harm in things that have no harm. That's the-

Kate Darling 25:09 Okay.

Lex Fridman 25:09 That's the big problem, because their whole job is to find harm in things. So what you said is completely correct, which is edginess should not hurt, doesn't necessarily, doesn't need to be a thing that hurts people. It, it obviously. Great humor, great, uh, personality doesn't have to, uh, like Clippy. (laughs) But yeah, I mean, it, it, but it's tricky to get right. And I, I'm not exactly sure. I, I don't know. I don't know why a large corporation with a lot of funding can't get this right.

Kate Darling 25:42 I do think you're right that there's a lot of aversion to risk. And so if you get lawyers involved or people whose job it is, like you say, to mitigate risk, they're just gonna say no to most things that could even be in some way... Yeah. Yeah, you get the problem in all organizations. So I think that you're right that that is a problem.

Lex Fridman 26:00 I think what's the way to solve that in large organizations is to have Steve Jobs types of characters. Unfortunately, you do need to have, I think, um, from a designer persp- or maybe like a Jony Ive, that is almost like a dictator.

Kate Darling 26:17 Yeah, you want a benevolent dictator.

Lex Fridman 26:18 Yeah. Who rolls in and says, like, the, the, cuts through the lawyers, the PR, but has a benevolent aspect. Like yeah, that has a good heart, and makes sure... Like I, I think all great artists and designers create stuff that doesn't hurt people. Like if you have a good heart, you're going to create something that's going to actually, um, make a lot of people feel good. That's what like, uh, people like Jony Ive, what they love doing is creating a thing that brings a lot of love to the world. They imagine like millions of people using the thing and it instills them w- with, with joy. That's, that, you could say that about social robotics, you could say that about the metaverse. Uh, it shouldn't be done by the PR people. It should be done by the designers.

Kate Darling 27:03 (laughs) I agree. PR people ruin everything.

Lex Fridman 27:06 Yeah, all the fun. Uh, (laughs)

Lex Fridman 27:09 in the, uh, in the book you have a picture. This, I just have a lot of ridiculous questions. You have a picture of two hospital delivery robots with a caption that reads... By the way, see, your book, I appreciate that it, it keeps the humor in.

Kate Darling 27:23 (laughs)

Lex Fridman 27:23 You didn't run it by the PR department.

Kate Darling 27:25 No, no one edited the book.

Lex Fridman 27:27 (laughs)

Kate Darling 27:27 It got rushed through.

Lex Fridman 27:28 (laughs) Uh, the-

Kate Darling 27:30 Would be a bad thing.

Lex Fridman 27:30 ... the caption reads, "Two hospital delivery robots whose sexy nurse names Roxy and Lola made me roll my eyes so hard they almost fell out." Um, what aspect of it made you roll your eyes?

Kate Darling 27:42 (laughs)

Lex Fridman 27:43 Is it the naming? Is-

Kate Darling 27:44 It was the naming. The form factor is fine. It's like a little box on wheels. The fact that they named them, also great. That'll let people enjoy interacting with them. We know that even just giving a robot a name, people will, uh, it, it facilitates technology adoption. People will be like, "Oh, you know, Betsy made a mistake. Let's help her out," instead of, "This stupid robot doesn't work." But why Loly and, Lola and Roxy? Like...

Lex Fridman 28:10 Those are too, too sexy?

Kate Darling 28:12 I mean, there's research showing that a lot of robots are named according to gender biases about-

Lex Fridman 28:22 Yeah.

Kate Darling 28:22 ... the function that they're fulfilling. So, you know, robots that are helpful and assistants and are like nurses are usually female gendered. Robots that are, you know, powerful, all-wise computers like Watson usually have like a booming male, uh, coded voice s- and, and name. So like why, uh, like that's one of those things, right? You're opening a can of worms for no reason. For no reason.

Lex Fridman 28:48 So you can avoid this whole can of worms?

Kate Darling 28:49 Yeah, just give it a different name. Like why Roxy? It's because people aren't even thinking. So to some extent, I don't, I don't like PR departments, but getting some feedback on your work from a diverse set of participants-

Lex Fridman 29:04 Yes.

Kate Darling 29:04 ... listening and taking in things that help you identify your own blind spots, and then you can always make your good leadership choices and good... Like you can still ignore things that you don't believe are an issue. But having the openness to take in feedback and making sure that you're getting the right feedback from the right people, I think that's really important.

Lex Fridman 29:28 So don't unnecessarily propagate the biases of society-

Kate Darling 29:32 Yeah. Why?

Lex Fridman 29:33 ... in the design. But, uh, if you're not careful though, when, when you do the research of like, you might, if you ran a poll with a lot of people, of, of all the possible names these robots have, they might come up with Roxy and Lola as, as, as names they, um, would enjoy most. Like that could come up as, uh, as the highest.

Kate Darling 29:58 As in you do marketing research and then... Well, that's what they did with Alexa. They did marketing research-

Lex Fridman 30:04 Yeah.

Kate Darling 30:05 ... and nobody wanted the male voice. Everyone wanted it to be female.

Lex Fridman 30:09 Well, what do you, what do you think about that? Like what... I-

Kate Darling 30:12 I mean-

Lex Fridman 30:12 ... if, if I, if I were to say I think the role of a great designer, again, to go back to Jony Ive, is to throw out the marketing research.

Kate Darling 30:22 ... yeah.

Lex Fridman 30:23 Like, take it in, do it, learn from it. But, like, if everyone wants Alexa to b- to be a female voice, the role of the designer is to think deeply about the future of social agents in the home and think, like, "What does that future look like?"

Kate Darling 30:40 Yeah.

Lex Fridman 30:40 And try to reverse engineer that future. It's like, in some sense there's this weird tension, like, you want to listen to a lot of people, but at the same time you want to... Y- you're creating a thing that defines the future of the world, and the people that you're listening to are part of the past. So, like, that weird tension.

Kate Darling 31:02 Yeah. I think that's true. And I think some companies, like Apple, have historically done very well at understanding a market and saying, "You know what our role is? It's not to listen to what the current market says. It's to actually shape the market and shape consumer preferences." And comp- companies have the power to do that. They can be forward-thinking, and they can actually shift what the future of technology looks like. And I agree with you that I would like to see more of that, especially when it comes to (laughs) existing biases that we know... Or, or, uh, you know, that, that... I think there's the low-hanging fruit of companies that don't even think about it at all and aren't talking to the right people and aren't-

Lex Fridman 31:43 Yeah.

Kate Darling 31:43 ... getting the full information. And then there's companies that are just, like, doing the safe thing and, and giving consumers what they want now. But to be really forward-looking and be really successful, I think you have to make some judgment calls about what the future is gonna be.

Lex Fridman 31:57 But do you think it's still useful to gender and to name the robots?

Kate Darling 32:01 Yes. I mean, gender is a minefield but people... (sighs) I, it's really hard to get people to not gender a robot in some way. So if you don't give it a name or you give it a, like, ambiguous voice, people will just choose something.

Lex Fridman 32:22 Mm-hmm.

Kate Darling 32:22 And maybe that's better than just, like, uh, you know, entrenching something that you've decided is best. But I do think it, it can be helpful on the, like, anthropomorphism engagement level to give it attributes that people identify with.

Lex Fridman 32:39 Yeah. I think, uh, a lot of roboticists I know, they, they don't gender the robot. They don't... They even try to avoid naming the robot, or naming it ain't something that is, uh, can be used as a name in conversation kinda thing. And I think that's actually... That's, um, irresponsible because people are going to anthropomorphize the thing anyway. So you're just, uh, removing from yourself the responsibility of how they're, they're going to anthropomorphize it.

Kate Darling 33:10 That's a good point.

Lex Fridman 33:11 And so, like, you want to be able to take... Like, they're going to do it. You have to start to think about how they're going to do it. Even if the robot is like a Boston Dynamics robot that's not supposed to have any kinda social component, they're obviously going to project a social component to it.

Kate Darling 33:28 Yeah.

Lex Fridman 33:28 Like, that arm... I worked a lot, a lot with, uh, quadrupeds now with, with, with robot dogs. You know, that arm people think is a head immediately.

Kate Darling 33:39 Yeah.

Lex Fridman 33:39 It's supposed to be an arm, but they start to think it's a head. And you have to, like, ac- acknowledge that. You can't... I mean, uh-

Kate Darling 33:45 They do now.

Lex Fridman 33:46 They do now?

Kate Darling 33:47 Well, they've deployed the robots and people are like, "Oh my god, the cops are using a robot dog." And so they have this PR nightmare, and so they're like, "Oh. Yeah. Okay. Maybe we should hire some HRI people."

Lex Fridman 34:02 (inhales deeply) Well, Boston Dynamics is an interesting comp- or any of the others that are doing similar thing because their, their main source of money is, um, in industrial applications, so, like, surveillance of factories and, uh, doing dangerous jobs. So, to them, it's almost good PR to, for people to be scared of these things 'cause it's, it's... For some reason, as you talk about, people are naturally for some reason scared, we could talk about that, o- of robots, and so it becomes more viral, like, uh, playing with that little fear. And so it's almost like a good PR because ultimately they're not trying to put them in the home and have a good social connection. They're trying to put them in factories. And so they, they, they have fun with it. If you watch Boston Dynamics videos-

Kate Darling 34:50 Yeah.

Lex Fridman 34:50 ... they're aware of it.

Kate Darling 34:52 Oh, yeah.

Lex Fridman 34:53 They're... I mean, I was-

Kate Darling 34:54 The videos for sure-

Lex Fridman 34:55 It-

Kate Darling 34:55 ... that they put out.

Lex Fridman 34:57 It's almost like an unspoken tongue-in-cheek thing. They, they're aware of how people are going to feel when you have a robot that does, like, a flip. Now most of the people are, uh, just, like, excited about the control problem of it, like how to, how to make the whole thing happen, but they're aware when people see.

Kate Darling 35:18 Well, I think they became aware. I think that in the beginning they were really, really focused on just the engineering. I mean, they're at the forefront of robotics-

Lex Fridman 35:26 Yeah.

Kate Darling 35:27 ... like locomotion and stuff. Um, and then when they started doing the videos, I think that was kind of a labor of love. I know that the former CEO, Marc, like-

Lex Fridman 35:37 Mm-hmm.

Kate Darling 35:37 ... he oversaw a lot of the videos and made a lot of them himself.

Lex Fridman 35:40 Mm-hmm.

Kate Darling 35:40 And, like, he's even, like, really detail-oriented, like, there can't be, like, some sort of incline that would give the robot an advantage. They're very... Like, he, he was very, um, had integrity about the authenticity of them. Uh, and... But then when they started to go viral, I think that's when they started to realize, "Oh, there is something interesting here that..." You know, I don't... I don't know how much they took it seriously in the beginning other than realizing that they could play with it in the videos.

Lex Fridman 36:10 Yeah.

Kate Darling 36:10 I know that they take it very seriously now.

Lex Fridman 36:13 What I like about Boston Dynamics and similar companies, it's still mostly run by engineers. But, you know-... I've had my criticisms. There's a bit more PR leaking in. But those-

Kate Darling 36:29 (laughs)

Lex Fridman 36:29 ... those videos are made by engineers 'cause that's what they find fun.

Kate Darling 36:33 Mm-hmm.

Lex Fridman 36:33 It's like testing the robustness of the system. I mean, they, uh, they're having a lot of fun there with the robots.

Kate Darling 36:41 Totally.

Lex Fridman 36:42 (laughs)

Kate Darling 36:43 Have you been, have you been to visit?

Lex Fridman 36:45 Yeah. Yeah. Yeah.

Kate Darling 36:46 Yeah?

Lex Fridman 36:46 Yeah. It's-

Kate Darling 36:46 So cool.

Lex Fridman 36:47 ... it's one of the most incry- like, I, I, I, uh, I mean, 'cause I, I have, um, eight, uh, robot dogs now. Uh-

Kate Darling 36:56 Wait, you have eight robot dogs?

Lex Fridman 36:58 Yeah.

Kate Darling 36:58 What?

Lex Fridman 36:58 (laughs)

Kate Darling 36:59 Are they just walking around your place? Like, where do you keep them?

Lex Fridman 37:00 Yeah, I'm working on them. Uh, that's actually one of my goals, is to have, at any one time, always a robot moving.

Kate Darling 37:09 Oh.

Lex Fridman 37:10 I'm far away from that.

Kate Darling 37:10 That's an ambitious goal.

Lex Fridman 37:13 Well, I have, like, more Roombas than I know what to do with.

Kate Darling 37:15 (laughs)

Lex Fridman 37:15 The r- room- that are programmed.

Kate Darling 37:17 (laughs)

Lex Fridman 37:17 So the, the programmable Roombas.

Kate Darling 37:20 Nice.

Lex Fridman 37:20 And, um, I have a bunch of little, like I bu- I built a, well, I'm not finished with it yet, but bought a robot th- from Rick and Morty. I just have a bunch of robots everywhere. But the thing is, what happens is you're working on one robot at a time, and, uh, that becomes like a little project, it's actually very difficult to have just a passively functioning robot always moving.

Kate Darling 37:46 Yeah.

Lex Fridman 37:46 And that's a, that's a, that's a dream for me 'cause I- I'd love to create that kind of l- little world. So, uh, the impressive thing about Boston Dynamics to me was to see, like, hundreds of spots. And like, there was a, the most impressive thing that still sticks with me is, um, there was a, a spot robot walking down the hall seemingly with no supervision whatsoever, and he was wearing, he or she, I don't know, was wearing a cowboy hat.

Kate Darling 38:15 (laughs)

Lex Fridman 38:16 And just, it was just walking down the hall, and nobody paying attention, and it's just, like, walking down this long hall, and I'm, like, looking around, does, is anyone, like, what's happening here? So, um, presumably some kind of automation where it's doing the mapping, I mean, the- the whole environment's probably really well-mapped. But I (laughs) it was just, it gave me a picture of a world where a robot is doing, doing its thing-

Kate Darling 38:40 Yeah.

Lex Fridman 38:40 ... wearing a cowboy hat, just going down the hall, like getting some coffee or whatever, like I don't know what it's doing, what's the mission. But, uh, I don't know, for some reason that really stuck with me. You don't often see robots that aren't part of a demo or that aren't, uh, you know, like a, with a semi-autonomous or autonomous vehicle, like, directly doing a task. This was just chilling.

Kate Darling 39:03 Yeah.

Lex Fridman 39:03 Just walking around. I don't know.

Kate Darling 39:05 Well, yeah, you know, I mean, we're at MIT, like, when I first got to MIT, it was like, "Okay, where's all the m- where's all the robots?" And they were all, like, broken or, like, not demo-ing.

Lex Fridman 39:14 Yeah.

Kate Darling 39:14 So yeah. Uh, and- and- and what really excites me is that we're about to have that. We're about to have so many moving ro- about to, well, it's coming. It's coming in our lifetime that we will just have robots moving around. We're already seeing the beginnings of it. There's delivery robots in some cities on the sidewalks. And I just love seeing like the TikToks of people reacting to that, because yeah, you see a robot walking down the hall with a cowboy hat, you're like, "What the fuck?"

Lex Fridman 39:43 (laughs)

Kate Darling 39:43 "What is this?" (laughs)

Lex Fridman 39:44 Yeah.

Kate Darling 39:44 "This is awesome and scary and kind of awesome."

Lex Fridman 39:47 Yeah.

Kate Darling 39:47 And people either love or hate it. That's one of the things that I think companies are underestimating, that people will either love a robot or hate a robot and nothing in between.

Lex Fridman 39:56 Mm-hmm.

Kate Darling 39:56 So it's just, again, an exciting time to be alive.

Lex Fridman 40:00 Yeah. I- I think kids almost universally, at least in my experience, love them. Lo- love legged robots especially.

Kate Darling 40:08 If they're not lo- my- my son hates the Roomba because ours is loud.

Lex Fridman 40:13 Oh, that, yeah. No, but the legs, the legs make a difference.

Kate Darling 40:16 Oh, yeah.

Lex Fridman 40:16 'Cause I- I don't th- th- your son, um, do- do they understand Roomba to be a robot?

Kate Darling 40:23 Oh, yeah, my kids, (laughs) that's one of the first words they learned. They know how to say beep, boop, and yes, they think the Roomba's a robot.

Lex Fridman 40:29 Does- do they project intelligence out of the thing?

Kate Darling 40:32 Well, we don't really use it around them anymore for (laughs) the reason that my son is scared of it.

Lex Fridman 40:37 Make noise. Hmm. Yeah. That's really interesting.

Kate Darling 40:41 They- I think they would. Like, even a Roomba, because it's moving around on its own, I think kids and animals view it as a- an agent.

Lex Fridman 40:52 So what do you think, if we just look at the state of the art of robotics, what do you think robots are actually good at today? So if we look at today?

Kate Darling 40:59 You mean physical robots?

Lex Fridman 41:01 Yeah, physical robots.

Kate Darling 41:04 Well...

Lex Fridman 41:05 Like, what are you impressed by? So, uh, I think a lot of people, I mean, that's what your book is about is, uh, maybe a not a perfectly calibrated understanding of where we are in terms of robotics, what's difficult to robotics, what's easy in robotics.

Kate Darling 41:22 Yeah. We're way behind where people think we are. So, what's impressive to me, so, uh, let's see, oh, one- one thing that came out recently was Amazon has this new warehouse robot, and it's the first autonomous warehouse robot that can- is safe for people to be around. And so, like, it's kind of, uh, most people, most people, I think, envision that our warehouses are already fully automated, and that they're just, like, robots doing things. It's actually still really difficult to have robots and people in the same space because it's dangerous for the most part. Robots, you know, because, especially robots that have to be strong enough to move something heavy, for example, they can really hurt somebody. And so until now, a lot of the warehouse robots had to just move along, like, preexisting lines, which really restricts what you can do. Um, and so having, I think, that- that's- that's one of the big challenges and one of the big, like, exciting things that's happening, is that we're starting to see more-... cobotics in industrial spaces like that, where people and robots can work side-by-side and not get harmed.

Lex Fridman 42:35 Yeah. That's what people don't realize are the physical manipulation tasks with, with humans. It's not that the robots wanna hurt you. (laughs) I think that's what people are worried about, like, this malevolent robot-

Kate Darling 42:49 (laughs)

Lex Fridman 42:49 ... gets a mind of its own and wants to destroy all humans. Now it's, you know, it's actually very difficult to know where the human is.

Kate Darling 42:56 Yeah.

Lex Fridman 42:56 And to, to respond to the human in, dynamically, and collaborate with them on a, on a task, especially if you're something like an industrial robotic arm, which is extremely powerful. So you-

Kate Darling 43:09 Right.

Lex Fridman 43:09 ... there's some of the, some of those arms are pretty impressive now that you can just, you can move, you can grab it, you can move it. So the, um, the collaboration between human and robot in the factory setting is really fascinating.

Kate Darling 43:23 Yeah.

Lex Fridman 43:25 Um, do you think they'll take our jobs?

Kate Darling 43:28 I don't think it's that simple. I think that there's a ton of disruption that's happening and will continue to happen. Um, you know, I think, uh, speaking specifically of the Amazon warehouses, that might be an area where it would be good for robots to take some of the jobs that are, you know, where people are put in a position where it's unsafe and they're treated horribly and, you know, probably it would be better if a robot did that, and Amazon is clearly trying to (laughs) automate that job away. So, uh, I think there's gonna be a lot of disruption. I do think that robots and humans have very different skill sets. So while a robot might take over a task, it's not gonna take over most jobs. Um, I think just things will change a lot. Like, I don't know, one of the examples I have in the book is mining. Um, so there you have this job that is very unsafe and that requires a bunch of workers and puts them in unsafe conditions. And now you have all these different robotic machines that can help make the jobs safer, and as a result now people can sit in these, like, air conditioned remote control stations and, like, control these autonomous mining trucks. And so that's a much better job, but also they're employing less people now. So it's, it's just a lot of... I think from a bird's eye perspective, you're not gonna see job loss, you're gonna see more jobs created because that's, I, I, I think the future is not robots just becoming like people and taking their jobs. The future is really a combination of our skills and then the supplemental skillset that robots have to increase productivity, to help people have better, safer jobs, to give people work that they actually enjoy doing and are good at. Um, but it's really easy to say that from a bird's eye perspective and, um, ignore kind of the, (laughs) the rubble on the ground as we go through these transitions because of course specific jobs are going to get lost with it.

Lex Fridman 45:39 Yeah. If you look at the history of the 20th century, it seems like automation constantly increases productivity and improves the average quality of life. So it's, it's been always good, so, like, thinking about this time being different is at le- would need to go against the lessons of history.

Kate Darling 46:01 It's true.

Lex Fridman 46:03 And, uh, the other thing is I think people think that the automation of the physical tasks is easy.

Kate Darling 46:10 (laughs)

Lex Fridman 46:10 I was, I was just in Ukraine, and the interesting thing is, um... I mean, there's a lot of, uh, difficult and, uh, dark lessons just about a, a war zone. But one of the things that happens in war is there's a lot of mines that are placed. Um, that's th- that's one, one of the big problems for years after a war is even over, is the entire landscape is covered in mines. And so there's a demining effort, and you would think robots would be good at this kind of thing, or, like, your intuition would be like, well, say you have unlimited money and you wanna do a good job of it. Unlimited money. You would get a lot of really nice robots. But y- no. It's humans are still far superior-

Kate Darling 46:57 Or animals.

Lex Fridman 46:57 ... at this kind of tech. Or animals. But e- right. But humans with animals together.

Kate Darling 47:02 Yeah.

Lex Fridman 47:02 Like, you can't just have a-

Kate Darling 47:03 That's true. (laughs)

Lex Fridman 47:03 ... dog with a hat.

Kate Darling 47:05 (laughs)

Lex Fridman 47:06 (laughs)

Kate Darling 47:06 That's fair. (laughs)

Lex Fridman 47:08 (laughs) But yes. And but figuring out, uh, also how to, uh, disable the mine. Obviously, the easy thing, the thing a robot can help with is to, uh, find the mine and blow it up.

Kate Darling 47:25 Mm-hmm.

Lex Fridman 47:25 But that's gonna destroy the landscape. That, that, that really ha- does a lot of damage to the land. You want to, uh, disable the mine and to do that because of all the different, all the different edge cases of the problem. It requires a huge amount of humanlike experience it seems like, so it's mostly done by humans. They have no use for robots. They don't want robots.

Kate Darling 47:45 Yeah. I think we

Kate Darling 47:46 overestimate what we can automate.

Lex Fridman 47:50 In the, especially in the physical realm.

Kate Darling 47:51 Yeah.

Lex Fridman 47:52 It's wo- it's weird. I mean, it's continues that the st- the s- the story of humans, we think we're shitty at everything in the physical world, including driving. We think everybody makes fun of themselves and others for being shitty drivers, but we're actually kind of incredible at them.

Kate Darling 48:07 No, we're incredible and that's why, like, that's why Tesla still says that if you're in the driver's seat, like, you, you are ultimately responsible. Because the ideal for... I mean, I mean, you know more about this than I do, but hu- like, robot cars are great at predictable things and can react faster and more precisely than a person and can do a lot of the driving. And then the reason that we still don't have autonomous vehicles on all the roads yet is because of this long tail of just-... unexpected occurrences, where a human immediately understands that's a sunset and not a traffic light, that's a horse and carriage ahead of me on the highway, but the car has never encountered that before. So, like, in theory, com- combining those skillsets is what's gonna really be powerful. The only problem is figuring off the, figuring out the human/robot interaction and the handoffs. So, like, in cars, that's a huge problem right now, figuring out- out the handoffs. Um, but in other areas, uh, it- it might be easier. And that's really the future, is human/robot interaction.

Lex Fridman 49:12 What's really hard to improve... It's- it's- it's terrible that people die in car accidents, but I mean, it's like 70, 80, 100 million miles, one death per, uh, 80 million miles. That's, like, really hard to beat for a robot. That's- that's, like, incredible that hum- like think about it, like the... how many people? Like, just the number of people throughout the world that are driving every single day, all of the st- you know, sleep-deprived, drunk, uh, distracted, all of that, and still very few die relative to what I would imagine. If I were to guess, back when the horse... See, wh- when I was like, in the- in the beginning of the 20th century riding my horse, I would talk so much shit about these cars. I'd be like, "This is gonna... this is extremely dangerous. These machines traveling at 30 miles an hour," or whatever the hell they were going at. (laughs) "This is irresponsible, it's unnatural, and- and it's going to be destructive to all of human society." But then it's extremely surprising how humans adapt to the thing, and they know how to not kill (laughs) each other. Um, I mean, th- that a- ability to adapt is incredible, and to mimic that in a machine is really tricky. Now that said, what Tesla's doing... I mean, I wouldn't have guessed how far machine learning can go on vision alone.

Kate Darling 50:38 Yeah.

Lex Fridman 50:38 It's really, really incredible. And people that are, uh, at least from my perspective, people that are kinda, um, uh, you know, critical of Elon and those efforts I think don't give enough credit to how much progress we've made, some- how much incredible progress has been made in that direction. I think most of the robotics community wouldn't have guessed how much it can do on vision alone. It's kind of incredible, um, because we would be... I think it's that approach which is relatively unique, has challenged the other competitors to step up their game. So if you're using LIDAR, if you're using mapping, um, that challenges them to do better, to scale faster, and to use machine learning and computer vision, uh, as well to integrate both LIDAR and vision. So, um, it's kind of incredible. And I'm- I'm not... I don't know if I even have a good intuition of how hard driving is anymore. Maybe it is possible to solve. So all the sunset-

Kate Darling 51:46 I think it is.

Lex Fridman 51:46 ... all the use cases you mentioned-

Kate Darling 51:47 I think we're gonna get there.

Lex Fridman 51:48 Yeah, the question is when.

Kate Darling 51:49 Yeah, I think it's not happening as quickly as people thought it would, because it is more complicated.

Lex Fridman 51:55 But I wouldn't have... I- I, uh, agree with you. My current intuition is that we're gonna get there.

Kate Darling 52:02 I think we're gonna get there too.

Lex Fridman 52:04 But I didn't... Before, I wasn't-

Kate Darling 52:07 You didn't think so?

Lex Fridman 52:07 ... sure we were gonna get there without, like, with current technology. So, you know, I- I was kinda... this is, like, with vision alone, I, my intuition was, you're gonna have to solve, like, common sense reasoning. You're gonna have to, you're gonna have to solve some of the big problems in artificial intelligence, not just, um, not just perception.

Kate Darling 52:35 Yeah.

Lex Fridman 52:35 Like, you have to have a deep understanding of the world is m- was my sense. But now I'm starting to, like, well this... I mean, I'm continuously surprised how well the thing works.

Kate Darling 52:44 Yeah.

Lex Fridman 52:44 Obviously, Elon and others, others have stopped, but Elon continues, you know, saying, "We're gonna solve it in a year." I used to-

Kate Darling 52:51 Well, yeah, that's the thing.

Lex Fridman 52:52 (laughs)

Kate Darling 52:53 Bold- bold predictions, though. (laughs)

Lex Fridman 52:54 Yeah. Well, y- everyone else used to be doing that, but they're kinda like, "All right, we-" (laughs)

Kate Darling 52:59 Yeah, maybe won't. (laughs)

Lex Fridman 53:01 (laughs) Maybe let's not promise we're gonna solve, uh, level four driving by 2020-

Kate Darling 53:05 (laughs)

Lex Fridman 53:05 ... let's, uh, let's chill on that.

Kate Darling 53:07 But people are still trying silently. I mean, the UK just committed 100 million pounds to research and development to speed up the process of getting autonomous vehicles on the road. Like, everyone is, everyone can see that it is solvable and it's going to happen and it's gonna change everything, and they're still investing in it.

Lex Fridman 53:28 And, uh, like, Waymo low-key has driverless cars in, uh, in Arizona. Like, you can get... you know, there's, like, robots. It's s- it's weird. Have you ever been in one?

Kate Darling 53:42 No.

Lex Fridman 53:43 It's so weird. It's so awesome 'cause, uh, the- the most awesome experience is, uh, is the wheel turning.

Kate Darling 53:50 Ah, yeah.

Lex Fridman 53:50 And you're sitting in the back. It's like... I don't know. It's, um... it feels like you're a passenger with that friend who's a little crazy of a driver.

Kate Darling 54:01 (laughs)

Lex Fridman 54:01 It feels like... shit, I don't know. Are- are you all right to drive, bro? You know, that kind of feeling. but, but then you kinda... that experience, that nervousness, um, and the excitement of trusting another being, in th- in this case it's a machine, is- is really interesting. Um, just even introspecting your own feelings about the thing.

Kate Darling 54:26 Yeah.

Lex Fridman 54:27 Uh, uh, they're not doing anything in terms of making you feel better-

Kate Darling 54:34 (laughs)

Lex Fridman 54:35 ... about (laughs) like, at least Waymo. I think they went with the approach of, like, let's not try to put eyes on the thing.... was, it's a- it's a wheel, we know what that looks like. It's- it's just a car. It's a car, get in the back, let's not, like, discuss this at all. Let's not discuss the fact that this is a robot driving you, and you're in the back, and if the robot wants to start driving 80 miles an hour and ru- run off of a bridge, you have no recourse. Let's not discuss this. You just get in the back. There's no discussion about, like, how shit can go wrong. Um, there's no eyes, there's nothing, there's, like, a map showing what the car can see. Like, you know, what happens if it's like a- a HAL 9000 situation? Like- (laughs) I'm s- like, I'm sorry, I can't... Y- you have a button you can, like, call customer service? (laughs)

Kate Darling 55:24 Oh, God. Then you get put on hold for two hours?

Lex Fridman 55:26 (laughs) Yeah, probably.

Kate Darling 55:29 (laughs)

Lex Fridman 55:29 Um, but, you know, currently what they're doing, which I think is understandable, but, you know, the car just can pull over and stop and wait for help to arrive, and then a driver will come, and then they'll actually drive the car for you. But that's, like, you know, what if you're late for a meeting or all that kind of stuff?

Kate Darling 55:48 Or, like, the more dystopian, isn't it The Fifth Element where... Is Will Smith in that mo- who's in that movie? No. Bruce Willis?

Lex Fridman 55:55 Bruce Willis.

Kate Darling 55:55 Oh, yeah. And he gets into, like, a robotic cab or car or something-

Lex Fridman 56:00 Yeah.

Kate Darling 56:00 ... and then because he's violated a traffic rule, it locks him in?

Lex Fridman 56:03 Yeah.

Kate Darling 56:04 And he has to wait for the cops to come, and he can't get out. So, like-

Lex Fridman 56:06 Yeah.

Kate Darling 56:07 ... we're gonna see stuff like that maybe.

Lex Fridman 56:09 Well, that's...

Kate Darling 56:10 (laughs)

Lex Fridman 56:11 I- I- I believe that the companies that have robots, the- the only ones that will succeed are the ones that don't do that, meaning they respect privacy.

Kate Darling 56:25 You think so?

Lex Fridman 56:26 Yeah. Because people... 'cause- 'cause they're gonna have to earn people's trust.

Kate Darling 56:31 Yeah, but, like, Amazon works with law enforcement and gives them the data from their Ring cameras, so...

Lex Fridman 56:37 Really?

Kate Darling 56:38 Why should, uh... Yeah. Oh, yeah. Do you have a Ring camera?

Lex Fridman 56:41 (laughs) Uh, no.

Kate Darling 56:43 Okay.

Lex Fridman 56:44 No, no. But, you know, basically any security camera, right? I have, uh, a Google s-... Whatever they have.

Kate Darling 56:51 We have one that's not... The data, we store the data on a local server because we don't want it to go to law enforcement. 'Cause all the companies are doing it. They're doing it. I- I bet Apple wouldn't.

Lex Fridman 57:03 Yeah.

Kate Darling 57:03 Apple's the only company I trust in I don't know how much longer.

Lex Fridman 57:08 I don't know. I... Maybe that's true for cameras. But with robots, I d- no- people are just not gonna let a robot inside his- their home. Where, like, one time where somebody gets arrested because of something a robot sees, that's gonna be, that's gonna destroy a company.

Kate Darling 57:28 You don't think people are gonna be like, "Well, that wouldn't happen to me. That happened to a bad person"?

Lex Fridman 57:33 I think they would.

Kate Darling 57:35 Yeah?

Lex Fridman 57:35 'Cause in the modern world, people, like, get... Have you seen Twitter? They get extremely paranoid about any kind of surveillance.

Kate Darling 57:43 But the thing that I've had to learn is that Twitter is not the modern world. Like, when I go, you know, inland to visit my relatives, like, they don't, the- that's a different discourse that's happening. I think, like, the whole tech criticism world-

Lex Fridman 57:58 Yeah.

Kate Darling 57:59 ... it's loud in our ears 'cause we're in those circles.

Lex Fridman 58:02 D- y- you think you can be a company that does social robotics and not win over Twitter?

Kate Darling 58:07 That's a good question.

Lex Fridman 58:09 I feel like the early adopters are all on Twitter. And it feels like you have to win them over.

Kate Darling 58:15 Feels like nowadays you'd have to win over TikTok honestly. I don't-

Lex Fridman 58:19 TikTok. Is that, is that a website?

Kate Darling 58:22 (laughs)

Lex Fridman 58:22 (laughs) I need to, I need to check it out.

Kate Darling 58:25 (laughs)

Lex Fridman 58:28 Um, (laughs) yeah, and that's an interesting one because China is behind that one.

Kate Darling 58:31 Exactly.

Lex Fridman 58:34 Uh, so if it's compelling enough, maybe people would be able to, uh, give up privacy and that kind of stuff. That's really scary.

Kate Darling 58:43 I just, I mean, I'm worried about it. I'm worried about it. And I'm... There have been some developments recently that are, like, super exciting, like the large language learning models. Like, wow, I did not anticipate those improving so quickly. And those are gonna change everything. And one of the things that I'm trying to be cynical about is that I think they're gonna have a big impact on privacy and data security and, like, manipulating consumers and manipulating people because suddenly you'll have these agents that people will talk to, and they won't care or won't know, at least on a conscious level, that it's recording the conversations. So kinda like we were talking about before. Um, and at the same time, the technology is so freaking exciting that it's gonna get adopted.

Lex Fridman 59:31 Well, it's not even just the collection of data, but the ability to manipulate at scale.

Lex Fridman 59:37 So, um, what do you think about the AI, the engineer from G- uh, from, uh, Google that thought LaMDA is sentient? You had actually a really good proposed from somebody else, I forgot her name, is brilliant. I can't believe I didn't know about her. Thanks to you-

Kate Darling 59:55 Janelle Shane?

Lex Fridman 59:56 Yeah. From Weird AI.

Kate Darling 59:57 Oh, yeah. I love her book.

Lex Fridman 59:58 Oh, she- she's great. I'll have to note for myself to reach out to her. She's amazing. She's hilarious and brilliant and just a great summarizer of the state of AI. But she has, um... I think that was from her, where I was looking at, uh, AI explaining that it's a squirrel.

Kate Darling 1:00:18 Oh, yeah. Because the transcripts that the engineer released, LaMDA kind of talks about the experience of human-like feelings and I think even consciousness. And so she was like, "Oh, cool, that's impressive. I wonder if an AI can also d- describe the experience of being a squirrel." And so she interviewed, I think she did GPT-3, uh, about the experience of being a squirrel, and then she did a bunch of other ones too like, "What's it like being a flock of crows? What's it like being an algorithm that powers a Roomba?" And, like-

Lex Fridman 1:00:50 ... yeah.

Kate Darling 1:00:51 You can have a conversation about any of those things, and they're very convincing.

Lex Fridman 1:00:53 It's pretty convincing, yeah.

Kate Darling 1:00:55 Yeah.

Lex Fridman 1:00:55 Even GPT-3-

Kate Darling 1:00:56 Yeah.

Lex Fridman 1:00:56 ... which is not, like, state of the art-

Kate Darling 1:00:58 Right.

Lex Fridman 1:00:58 ... is convincing of being a squirrel. It's like wha- what it's like... (laughs)

Kate Darling 1:01:02 (laughs) It's so cute.

Lex Fridman 1:01:03 I mean, you shou- you should check it out, 'cause it, it really is. It's like, yeah, that probably is what a squirrel would, would ta- would, sound like.

Kate Darling 1:01:10 (laughs) Would say.

Lex Fridman 1:01:11 (laughs) Are you excited? Like, wha- what's it like being a squirrel? Oh, it's fun, whatever it is.

Kate Darling 1:01:14 Oh, yeah. I get to eat nuts and run around all day. (laughs)

Lex Fridman 1:01:17 Uh, like, w- how do you think people will feel like when you tell them that you're a squirrel? Um, you know, or, like, I forget what it was. Like, a lotta people might be scared to find out that you're a squirrel or something like this, and then the, the system answers pretty, like-

Kate Darling 1:01:32 Yeah.

Lex Fridman 1:01:32 ... pretty well. Like, yeah, like, um, I hope they'll... Like, what do you think the, when they find out you're a squirrel? (laughs) Um, I, I hope they'll see how fun it is to be a squirrel-

Kate Darling 1:01:44 (laughs) Yeah.

Lex Fridman 1:01:44 ... or something like that.

Kate Darling 1:01:45 What do you say to people who don't believe you're a squirrel?

Lex Fridman 1:01:47 Yeah, right. (laughs)

Kate Darling 1:01:48 I say, "Come see for yourself."

Lex Fridman 1:01:49 (laughs) Yeah, right.

Kate Darling 1:01:49 "Uh, I am a squirrel."

Lex Fridman 1:01:50 That's great. I-

Kate Darling 1:01:52 Well, and s- and I think it's really great because it, it, like... The two things to note about it are, first of all, just because a machine is describing an experience doesn't mean it can, it actually has that experience. But then secondly, these things are getting so advanced and so convincing at describing these things and talking to people that that's... I mean, just the implications for health, education, communication, entertainment, gaming. Like, I just, uh, like, all of the applications, it is mind-boggling what we're gonna be able to do with this, and, and that my kids are not gonna remember a time before they could have conversations with artificial agents.

Lex Fridman 1:02:32 Do you think they would... Because to me, this is, um... The focus in the AI community has been, well, this engineer surely is hallucinating. The thing is not sentient. But to me, I... First of all, w- it doesn't matter if he is or not. This is coming-

Kate Darling 1:02:52 Yeah.

Lex Fridman 1:02:52 ... where a large number of people would believe a system is sentient, including engineers within companies.

Kate Darling 1:02:59 Yeah.

Lex Fridman 1:02:59 So it's in that sense, you, you start to think about a world where, like, your kids aren't just used to having a conversation with a bot, but used to believing, kinda having an implied belief that the thing is sentient.

Kate Darling 1:03:14 Yeah. I think that's true, and I think that one of the things that bothered me about all of the coverage in the tech press about this incident... Like, obviously, I don't believe the system is sentient. Like, I, I think that it can convincingly describe that it is. I don't think it's doing what he thought it was doing and actually experiencing feelings. But a lot of the tech press was about how he was wrong and depicting him as kind of naive, and it's not naive. Like, there's so much research in my field showing that people do this, even experts. They, they might be very clinical when they're doing human/robot interaction experiments with a robot that they've built, and then you bring in a different robot, and they're like, "Oh, look at it. It's having fun. It's doing this." Like-

Lex Fridman 1:03:58 Mm-hmm.

Kate Darling 1:03:58 ... that happens in our lab all the time. We are all this guy, and it's gonna, it's gonna be huge. So I think that the, the goal is not to discourage this kind of belief or, like, design systems that people won't think are sentient. I don't think that's possible. I think you're right. This is coming. It's something that we have to acknowledge and even embrace and be very aware

Kate Darling 1:04:24 of.

Lex Fridman 1:04:25 So, one of the really interesting perspectives that your book takes on a system like this is to see them... Not to compare a system like this to humans, but to compare it to animals, of how we see animals. Can you kinda try to, uh, again sneak up, try to explain-

Kate Darling 1:04:43 Mm-hmm.

Lex Fridman 1:04:43 ... why this analogy is better than the human analogy, the analogy of robots as animals?

Kate Darling 1:04:48 Yeah. And it gets trickier with the language stuff, but, y- we'll get into that, too. Um, I think that animals are a really great thought experiment when we're thinking about AI and robotics, because, again, this comparing them to humans, that leads us down the wrong path, both because it's not accurate, but also, I think, for the future, we don't want that. We want something that's a supplement. But I think animals, because we've used them throughout history for so many different things, we, we domesticated them not because they do what we do, but because what they do is different, and that's useful. And I... It just, like, whether we're talking about companionship, whether we're talking about work integration, whether we're talking about responsibility for harm, there are just so many things we can draw on in that history from these entities that can sense, think, make autonomous decisions, and learn that are applicable to how we should be thinking about robots and AI. And, and the point of the book is not that they're the same thing, that animals and robots are the same. Obviously, there are tons of differences there. Like, you can't, you can't have a conversation with a squirrel, right? But the, the point is that-

Lex Fridman 1:05:56 I do it all the time.

Kate Darling 1:05:57 Oh, really?

Lex Fridman 1:05:58 By the way, squirrels are the cutest. I project so much on squirrels. I wonder what their inner life is. Um, I suspect they are much bigger assholes than we imagine.

Kate Darling 1:06:09 Really?

Lex Fridman 1:06:10 Like, if it was a giant squirrel, it would fuck you over so fast if it had the chance. It would take everything you own. It would eat all your stuff-

Kate Darling 1:06:18 (laughs)

Lex Fridman 1:06:18 ... but 'cause it's small and the furry tail... The furry tail is a, is a, is a weapon against human consciousness and cognition. It wins us over. That's what cats-

Kate Darling 1:06:29 Oh, yeah.

Lex Fridman 1:06:29 ... do, too. Cats out-out-competed squirrels.

Kate Darling 1:06:33 And dogs. Like-

Lex Fridman 1:06:35 Yeah. W-

Kate Darling 1:06:35 Dogs are-

Lex Fridman 1:06:35 No, dogs have love. Cats are, have no soul. They... No, I'm just kidding.

Kate Darling 1:06:40 (laughs)

Lex Fridman 1:06:40 People get so angry when I talk shit about cats. Uh, I love cats. Anyway, uh, so y- uh, uh, yeah. You're, you're describing all the different kinds of animals that get domesticated.... and it's a really interesting idea that it's not just to pets, there's all kinds of domestication going on. They all have all kinds of uses.

Kate Darling 1:07:00 Yes.

Lex Fridman 1:07:00 Like the, uh, ox that you proposed might be the... at least historically one of the most useful domesticated animals.

Kate Darling 1:07:09 It was a game changer because it revolutionized like what people could do economically, et cetera. So, I, I mean, just like robots, they're gonna cha- they're gonna change things economically, they're gonna change landscapes, like cities might even get rebuilt around autonomous vehicles or drones or delivery robots. Like, uh, I think just the same ways that animals have really shifted society, and society has adapted also to, like, socially accepting animals as pets. Um, I think we're gonna see very similar things with robots. So I think it's a useful analogy. It's not a perfect one, but I think it's... it helps us get away from this idea that robots can, should, or will replace people.

Lex Fridman 1:07:49 If you remember, what are some interesting uses of animals? Ferrets, for example.

Kate Darling 1:07:54 Oh, yeah, the ferrets. They- uh, they still do this, they use ferrets to go into narrow spaces that people can't go into, like a pipe or, like, they'll use them to run electrical wire. I think they did that for Princess Di's wedding. There's so many weird ways that-

Lex Fridman 1:08:07 Work is work.

Kate Darling 1:08:08 ... we've used animals and still use animals for things that robots can't do. Like the dolphins, the- that they used in the, in the military. I think the... I think Russia still has dolphins and the US still has dolphins in their navies. Um-

Lex Fridman 1:08:25 For what?

Kate Darling 1:08:26 Uh, mine detection, looking for lost underwater equipment, some rumors about, like, using them for weaponry, (laughs) which, which I think Russia's like, "Sure, believe that," and America's like, "No, no, we don't do that." Who knows? Uh, but they started doing that in, like, the '60s, '70s. They started training these dolphins because they were like, "Oh, dolphins have this amazing echolocation system that we can't replicate with machines, and they're trainable, so we're gonna use them for all the stuff that we can't do with machines or by ourselves." And they've tried to phase out the dolphins. I know the US has, like, invested m- a lot of money in trying to make robots do the mine detection. But, like you were saying, there is some things that the robots are good at and there's some things that biological creatures are better at, so they still have the dolphins.

Lex Fridman 1:09:13 So there's also pigeons, of course.

Kate Darling 1:09:16 Oh, yeah. Pigeons. Oh, my gosh, there's so many examples. The pige- I mean, the pigeons were the original hobby photography drone. They also carried mail for thousands of years, letting people communicate with each other in new ways. So, the thing that I like about the animal analogy is they have all these physical abilities, but also sensing abilities that we just, we don't have, and, like, that's, that's just so useful. And that's, that's robots, right? Robots have physical abilities, they can help us lift things or do things that we're not physically capable of. They can also sense things. It's just... Yeah, I just feel like, I still feel like it's a really good analogy and-

Lex Fridman 1:09:54 Yeah, it's really strong.

Kate Darling 1:09:55 And it works because it's- people are familiar with it.

Lex Fridman 1:09:58 What about companionship? And when we start to think about, like, cats and dogs, like pets that seem to serve no purpose whatsoever except the social connection.

Kate Darling 1:10:08 Yeah. I mean, that's kind of a newer thing. At least in the United States, like, dogs used to have... like, they used to have a purpose, they used to be guard dogs or they had some sort of function, and then at some point they became just part of the family.

Lex Fridman 1:10:24 Mm-hmm.

Kate Darling 1:10:25 Um, and I... it's, it's so interesting how there's some animals that we've treated as workers, some that we've treated as objects, some that we eat, and some that are parts of our families, and that that's different across cultures. And I'm convinced that we're gonna see the same thing with robots, um, where people are gonna develop strong emotional connections to certain robots that they relate to, either culturally or, or personally, emotionally, and then there's gonna be other robots that we don't treat the same way.

Lex Fridman 1:10:58 I wonder, does that have to do more with the culture and the people or the robot design? Is there interplay between the two? Like, why did dogs and cats out-compete ox and, I don't know, what else, like farm animals, to, to really get inside the home and get inside our hearts?

Kate Darling 1:11:18 Yeah. I mean, people point to the fact that dogs are very genetically flexible and, um, they can evolve much more quickly than other animals, and so they... evolutionary biologists think that dogs evolved to be more appealing to us, and then once we learned how to breed them, we started breeding them to be more appealing to us too, which is not something that we necessarily would be able to do with cows, although we've bred them to make more milk for us. So, it, uh, but part of it is also culture. I mean, there are cultures where people eat dogs still today, and then there's other cul- cultures where we're like, "Oh, no, that's terrible. We would never do that." And so I think there's a lot of different elements that play in.

Lex Fridman 1:12:01 I wonder if there's good... 'cause I understand dogs because they use their eyes, they're able to communicate affection, all those kinds of things. It's really interesting what dogs do. There's, uh, whole conferences on dog consciousness and cognition and all that kind of stuff. Now, cats is a mystery to me because they seem to not give a shit about the human.

Kate Darling 1:12:22 But they're warm and fluffy and cute.

Lex Fridman 1:12:24 But they... but they're also passive-aggressive, so they're... at the same time, they're like, they're dismissive of, of you in some sense.

Kate Darling 1:12:33 I think some people like that. Some people like that about people.

Lex Fridman 1:12:37 Yeah. They want, they want the push and pull of a re- of a relationship with something like that.

Kate Darling 1:12:40 They don't want loyalty or unconditional love.

Lex Fridman 1:12:43 Yeah. Yeah.

Kate Darling 1:12:43 That does... that means they haven't earned it.

Lex Fridman 1:12:45 Yeah. Yeah. (laughs) And maybe that says a lot more about the people than it does about the animals.

Kate Darling 1:12:53 Oh, yeah. We all need therapy.

Lex Fridman 1:12:54 Yeah. (laughs) So, I'm judging harshly the people that have cats, or, or the people that have dogs. Maybe the people that have dogs need, are, are desperate for attention and unconditional love, and they're unable to s- to, um, to sort of struggle, uh, to earn, uh, meaningful connections. Um, I don't know.

Kate Darling 1:13:21 Maybe people are talking about you and your robot pets in the same way.

Lex Fridman 1:13:26 Yeah, that's, uh... (laughs) It is kind of sad. There's just robots everywhere. But it is, I mean, I'm joking about it being sad, 'cause I think it's kind of beautiful. I think robots are, uh, beautiful in the same way that pets are, even children, in that, like, they capture some kind of magic of, uh, social robots. M- they have the capacity to have the same kind of magic of connection. Um, I don't know what that is. Like, um, when they're brought to life and they move around, the way they make me feel, I'm pretty convinced, is, as, as you know, they will make billions of people feel. Like, I, I don't think I'm like some weird robotics guy. I'm not. Uh-

Kate Darling 1:14:13 I mean, you are. But not in this way.

Lex Fridman 1:14:15 Not in this way. I mean, I just, I can put on my, like, hu- normal human hat and just see this, oh, this is, like there's a lot of possibility there of something cool, just like with dogs.

Kate Darling 1:14:27 Yeah.

Lex Fridman 1:14:27 Like, what is it? Why are we so into dogs or cats? Like, it's, like, that, it's way different than us. Like-

Kate Darling 1:14:35 It is.

Lex Fridman 1:14:35 It's like drooling all over the place with its tongue out. It's like, what, it's like a weird creature that used to be a wolf. Why are we into this thing?

Kate Darling 1:14:43 Well, dogs can either express or mimic a lot of emotions that we recognize, um, and I think that's a big thing. Like, a lot of the magic of animals and robots is our own self-projection, and the easier it is for us to see ourselves in something and project human emotions or qualities or traits onto it, the more we'll relate to it. And then, you also have the movement, of course. I think that's also really... That's why I'm so interested in physical robots, because that's, I think, the visceral magic of them. I think we're... I mean, there's, there's research showing that we're probably biologically hardwired to respond to autonomous movement in our physical space because we've had to watch out for predators, or whatever the reason is. Um, and so animals and robots are very appealing to us as these autonomously moving things that we view as agents instead of objects.

Lex Fridman 1:15:40 I mean, I, I love the moment, which is, I've, I've been particularly working on, which is when a robot like the cowboy hat, uh, is doing its own thing, and then, uh, it recognizes you. I mean, the way a dog does. And it, it looks like this, and the, the moment of recognition, like, you're walking, sa- say you walk into an airport or on a street, and there's just, you know, hundreds of strangers. But then, you see somebody you know, and that like, where you wake up to like, ah, that excitement of seeing somebody you know and saying hello and all that kind of stuff, that's a magical moment. Like, uh, I think, especially with a dog, it makes you feel noticed and heard and, and loved, like, that somebody looks at you and recognizes you, that, to, that it matters that you exist.

Kate Darling 1:16:33 Yeah, you feel seen.

Lex Fridman 1:16:34 Yeah. And that's a cool feeling. And I, I honestly think robots can give that feeling too.

Kate Darling 1:16:39 Oh, yeah. Totally.

Lex Fridman 1:16:41 Currently, Alexa, I mean, one, one of the downsides of these systems is they don't... They're servants. They, like, uh, part of the, you know, they're trying to maintain privacy, I suppose. Uh, but y- I don't feel seen with Alexa, right? It's-

Kate Darling 1:17:00 I think that's gonna change. I think you're right, and I think that that's, that's the game-changing, changing nature of things like these large language learning models. And the fact that these companies are investing in embodied versions that move around of Alexa, like Astro.

Lex Fridman 1:17:18 Can I just say, yeah. Astro, I haven't... Is that out? Is, is it out?

Kate Darling 1:17:22 I mean, it's out. You can't just like buy one commercially yet-

Lex Fridman 1:17:25 Yeah

Kate Darling 1:17:25 ... but you can apply for one.

Lex Fridman 1:17:27 Yeah.

Lex Fridman 1:17:28 My gut says that these companies don't have the guts to do the personalization. This goes to the, because it's edgy. It's dangerous. It's gonna make a lot of people very angry. Like, it, in the way that, you know, just imagine... Okay. All right. If you do the full landscape of human civilization, just visualize the number of people that are going through breakups right now, just the amount of really passionate, just even if we just look at teenagers, the amount of deep heartbreak that's happening, and like if, if you're going to have Alexa have more of a personal connection with a human, you're gonna have humans that like have existential crises. There's a lot of people that suffer from loneliness and depression. And like, you're now taking on the full responsibility of being a companion to the, the, the, the rollercoaster of the human condition. As a company, like imagine PR and marketing people. They're gonna freak out. They don't have the guts. It's gonna have to come from somebody from a new Apple, from those kinds of fol- like a, a small startup.

Kate Darling 1:18:41 And it might.

Lex Fridman 1:18:42 Yeah.

Kate Darling 1:18:42 Like, they're coming. There's already virtual therapists. There's that Replika app. I haven't tried it, but-

Lex Fridman 1:18:47 Okay.

Kate Darling 1:18:47 ... Replika is like a virtual-

Lex Fridman 1:18:47 It's great.

Kate Darling 1:18:48 ... companion. Like, it's coming, and if big companies don't do it, someone else will.

Lex Fridman 1:18:53 Yeah. I think the neg- the future, the next trillion dollar company will be those personalization, 'cause if you think, um...... if you think about all the, the AI we have around us, all the, the smart phones and so on, there's very minimal personalization.

Kate Darling 1:19:11 You don't think that's just because they weren't able?

Lex Fridman 1:19:14 No.

Kate Darling 1:19:14 Really?

Lex Fridman 1:19:15 I don't think they have the guts.

Kate Darling 1:19:17 I mean, it might be true but I have to wonder, I mean, Google is clearly gonna do something with the lang- I mean ...

Lex Fridman 1:19:24 They don't have the guts.

Kate Darling 1:19:25 (laughs) Are you challenging them?

Lex Fridman 1:19:28 Uh, partially, but not really, 'cause I know they're not gonna do it.

Kate Darling 1:19:32 I mean-

Lex Fridman 1:19:32 They don't have to, it's bad for business in the short term.

Kate Darling 1:19:35 I'm gonna be honest, like, maybe it's not such a bad thing if they don't just like roll this out quickly because I do think there are huge issues.

Lex Fridman 1:19:44 Yeah.

Kate Darling 1:19:45 And, and there's, and not just issues with, like, the responsibility of, like, unforeseen effects on people, but what's the business model? And if you are using the business model that you've used in other domains then you're gonna have to collect data from people, which you will anyway to personalize the thing, and you're gonna be somehow monetizing the data, or you're gonna be doing some, like, ad model. I, it just, it seems like now we're suddenly getting into the realm of, like, severe consumer protection issues, and I'm, I'm really worried about that. I, I see massive potential for this technology to be used in a way that's not for the public good, and not, I mean, that's in, in an in- individual user's interest maybe, but not in society's interest.

Lex Fridman 1:20:35 Yeah, see I think, I think that kind of personalization should be, like redefine how we treat data. I think you should own all the data your phone knows about you, like, and be able to delete it with a single click and walk away. And that data cannot be monetized or used or shared anywhere without your permission. I think that's the only way people will trust you to give, to, for you to use that data.

Kate Darling 1:21:07 But then how are companies gonna... I mean, a l- a lot of these applications rely on massive troves of data to train the AI system.

Lex Fridman 1:21:18 Right, so you have to, um, opt in constantly, and opt in not in some legal, "I agree," but obvi- like show exa- like, um, in the way, um, I opt in to tell you a secret. Like, we understand, like, that, like, I have, I have to choose like how well do I know you and then I say, like, "Don't tell this to anyone." (laughs) And then I have to judge how leaky that, uh, like, how good you are at keeping secrets. In that same way, like, it's very transparent in, uh, which data you're allowed to use for which purposes.

Kate Darling 1:21:56 That's what people are saying is the solution, and I think that works to some extent, having transparency, having people consent. I think it breaks down at the point at which, and we've seen this happen on social media too, like people are willingly giving up their data because they're getting a functionality from that, and then the harm that that causes is on a, like maybe just someone else and not to them personally. So ...

Lex Fridman 1:22:21 I don't think people are giving their data. They're not being asked. Like ...

Kate Darling 1:22:25 But if, if you were asked-

Lex Fridman 1:22:26 It's not consensual.

Kate Darling 1:22:27 If you were like, "Tell me a secret about yourself and I'll give you $100." I'd tell you a secret.

Lex Fridman 1:22:34 No, not $100. Uh, first of all, you wouldn't. Uh, you wouldn't trust, like, why are you giving me $100? Um-

Kate Darling 1:22:41 It's a bad example. (laughs) But, yeah.

Lex Fridman 1:22:43 But, but like I, I need, um, I would ask for your specific, like, fashion, uh, interest and to, you know, to give recommendations to you for shopping, and I'd be very clear for that. You could disable that, you can, uh, delete that. But y- but then you can be, have a deep meaningful rich connection with the system about what you think you look fat in, what you look great in, what, like, the full history of all the things you've worn, whether you, uh, regret the Justin Bieber or enjoy the Justin Bieber shirt, all of that information that's mostly private to even you, not even your loved ones. That, a system should have that, 'cause then a system, if you trust it to keep control of that data that you only can walk away with, that system could tell you a damn good thing to wear.

Kate Darling 1:23:36 It could, and the harm that I'm concerned about is not that the system is gonna then suggest a dress for me that is based on my preferences. So I, I went to this conference once where I was talking to the people who do the analytics in like the big ad companies, and like literally a woman there was like, "I can ask you three totally unrelated questions and tell you what menstrual product you use." And so what they do is they aggregate the data and they map out different personalities and different people and demographics, and then they have a lot of power and control to market to people. So, like, I might not be sharing my data with any of the systems because I'm like, "I'm on Twitter, I know that this is bad." Other people might be sharing data that can be used against me. Like it's, I think it's, it's way more complex than just I share a piece of personal information and it gets used against me. I think that at a more systemic level, and then it's always, you know, vulnerable populations that are targeted by this. Um, you know, low income people being targeted for scamming loans or I don't know, like, I could get targeted, like someone, uh, not me because I ha- someone who doesn't have kids yet and is my age could get targeted for, like, freezing their eggs, and there's all these ways that you can manipulate people where it's not really clear that that came from ...... that person's data. It came from all of us, (laughs) so all of us opting into this.

Lex Fridman 1:25:11 But there, there's a bunch of sneaky decisions along the way that could be avoided if there's transparency. So that... So one of the ways that goes wrong if you share that data with too many ad networks, don't... Run your own ad network. Don't share with anybody.

Kate Darling 1:25:30 Okay. And that's something that you could regulate.

Lex Fridman 1:25:32 This data, you b- that belongs to just you and all the ways you allow the company to use it. The default is in no way at all. And you are consciously constantly saying ex- exactly how to use it. And, um, and also it has to do with the recommender system itself from the company, which is, um, freezing your eggs. If that doesn't make you happy, if that idea doesn't make you happy, then the system shouldn't recommend it. It should very, be very good at learning. So not the kind of things that the category of people it thinks you belong to would do, but more you specifically, what makes you happy,