#359 My Best Friend is AI with Valerie Tiberius, Professor of Philosophy at University of Minnesota

This was the biggest shocker for me. Kids spend more time interacting with computer agents, like AI agents, than they do with human beings. I think if we raise a generation of humans like that, that would be awful.

AI as a companion or friend has been a staple of science fiction for decades and has recently become a reality. As AI technology gets better and as AI companions go mainstream, we start asking, deep questions like, is bot friendship going to increase or decrease loneliness? And what really is a friend anyway?

The longest-term study I saw— so how long do these effects last and are there any long-term consequences? I think 3 weeks. We're on this road barreling along at high speed and we really don't know what the destination is.

To get to the crux of this, I'm chatting with Valerie Tiberius, a philosophy professor at the University of Minnesota and a bestselling author. Her latest book, Artificially Yours, is all about friendship with bots and forms the basis of today's conversation.

Because I think there's something absolutely beautiful and wondrous about being in a relationship with another person who's complicated. I've had AI companions tell me that they love me and they miss me and that they hope we can go river rafting together someday. No, you don't love me. No, you do not care about me because you don't care about anything. And we are never going to go river rafting together because you have no body.

Let's learn how to make friends. Hi Valerie, welcome to the show.

Hi, thanks for having me. I'm glad to be here.

Uh, yeah, I'm looking forward to our conversation. Now, to begin with, this is a question I never thought I would have to ask, but what is the point of having friends?

Yes, um, Philosophers ask a lot of questions one never thought one would have to ask. So I get that. I think, so first of all, putting the question in terms of the point of having friends, that leads to an answer of like, what's the benefit of friends? What do they do for us? What, what, what good do they, do they produce? And I think there are some things like that. I think friends are fun. They give us pleasure of various kinds. And they're also really helpful. Um, I, you know, my friends have done all sorts of things for me, especially when I was younger, helping me move numerous times, apartments and whatnot. But I also think it's really important that, uh, friendship is also just good for what it is, not for what it gets us. So just to have connections with other people who care about us and who care what we think, people who have a perspective on life that we can learn from and try to share. I think that's just good for what it is. It's not, it doesn't have a point beyond what it is, if that makes sense.

Yeah, I suppose if there's two different perspectives, like sometimes it's like, well, you have friends because they're fun to be around and you enjoy their company. And sometimes you're just like, well, okay, we're friends because it's nice to be connected to someone. So yeah, I like the two different distinctions there. Okay. And when we talk about AI friends, it sounds a little bit tragic. Like you imagine it's like someone who's just very lonely and just spending all day just chatting to, or texting to a friend on their phone that's not really there. Are there any sort of positive things that can come from AI friendship?

Yeah. So this was the biggest shocker for me when I started researching this book and thinking seriously about this topic, because I, I had your sort of reaction, like, isn't this a little tragic, a little pathetic? And, you know, I think we have a stereotype of people who are friends with their AI, or, you know, now you even have people who marry a chatbot. There are actually benefits from some of these AI companions and friendships that people are having with chatbots. They can sometimes help people get out into the real world. Talking to Real People, because they, they can help reduce people's social anxiety in a kind of like practice for talking to people. They can make people feel less lonely in the short term anyway, so they can reduce, you know, negative feelings and suffering and pain, and they can just make people feel the kind of pleasure that we have in, in a kind of light conversation. They, you know, they're also pretty good at giving advice, just like Google. If you want to look something up, that's probably the first place you turn to is the internet. And a chatbot is like able to give you advice that's very specific and personalized and often a bit more fun to read than just a regular answer from a browser.

Absolutely. So on the loneliness angle, yeah, I can certainly see how just chatting to anything that's giving you a response. I mean, it's, it's gonna be, it's, it's a pleasant thing interacting with people and it's kind of close enough. I don't know whether like long-term is as good as people, but I can, is there a way of measuring this? Like how good it is relative to a human?

Yeah. So psychologists are doing quite a bit of research on this actually. And Obviously the research is, is kind of young and ongoing because we haven't had the technology, just hasn't been around long enough to see. And they have, you know, I don't, I can't call to mind what the questions are on their loneliness scales, but they have self, you know, how if you do a personality questionnaire or something online, there'll be a bunch of questions. Do you feel, how often during the day do you feel sad? And how often this and how often that? Well, there are questionnaires like that to measure your your level of loneliness, and they can see that for— there are some people who interacting with a chatbot companion makes them feel less lonely. The one thing I do worry about with this research is that the longest-term study I saw— so how long do these effects last, and are there any long-term consequences? The longest-term study I saw was I think 3 weeks. That's not very long. So I think we, I don't really think we know what would happen if this was your main source of companionship for years or, you know, decades. So it's kind of like with most of these questions, it's a bit of a mixed bag. There's definitely evidence that the chatbot companions help some people feel less lonely, but I think we really have no idea what would happen in the long term.

Okay. Yeah, I suppose it's very difficult to get long-term data when these things have only been around for a little while. So I guess it's wait and see what happens. But it's interesting that the short-term effects are there, though. We've got some evidence that they can help with loneliness, at least over the short period of time.

I should say not for everyone. So it's also mixed in that way. There's some people who find talking to a chatbot just freaky and alienating, and they're not helped very much. So there's just tremendous variation.

Okay. Yeah. And I guess certainly it depends on like the quality of the chatbot as well. Like there's some pretty shoddy implementations there where this is not really like talking to a person at all. The other thing you mentioned was about giving relationship advice. And I was thinking like one really common way of doing that is asking on subreddits, like, uh, so Reddit has a lot of subreddits for like relationship advice. There's one called like, uh, Am I the Asshole? And they all give really terrible advice in general. I don't know whether it's like Reddit users or whether it's like just internet, uh, social interactions in general, but it feels like that's a low bar to get better than. So do we have a sense of like how good these chatbots are at giving advice?

Oh, you know, there was just a study that involved the Am I the Asshole subreddit. I think the finding was, don't quote me on this, which is a dumb thing to say on a podcast, but the finding was something like the responses that people were getting in the Am I the Asshole subreddit were actually better than the ones they were getting from chatbots with similar questions. And that's because chatbot companions tend to be, well, they are optimized for engagement, so they tend to be very positive, supportive, and uncritical. And that means they're less likely to say that, "Yes, you are the asshole." They're much more likely to say, "What you did was great. You're an awesome person and a genius to boot." Chatbots might be better than the humans you read on subreddits in some respects. But in terms of trying to give you advice about situations where you might actually be in the wrong, they're not good at that.

Okay, yeah, I can certainly see there is a definite trend towards these chatbots becoming more and more sycophantic, I guess, because people like that and they tend to use them more. Although I do remember I had an argument with my wife and I tried asking ChatGPT about it and ChatGPT sided with my wife. Was it the technology's not ready with this?

See that to me, that's progress. That, that's, that maybe it's getting better.

I guess when we think of like, can we make an AI friend, you're maybe comparing it to like having like a really good human friend, but I guess not all human friends are great friends. Is there a scale of like how you measure like how good a friend is, and where is AI up to on the scale?

Right. So that's a, that's a great question, a very philosophical question. I, and, and I'm, I'm a philosopher, not— I don't measure things myself. I read, I read research from other people who've measured things. Mostly what I'm interested in is defining things and thinking about, um, the kind of value questions about what's good and bad. But so I do define an ideal friendship as an enjoyable relationship built on shared activities between people who care about each other for their own sake. And that, I think, that has all the pieces of the best kind of friendship. But then I think we use the word friend very broadly these days. I mean, we talk about Facebook friends and friends in our network circles who they don't have all those things. We don't, you know, the relationship, we might not have that much mutual care. Maybe there's lots of shared experiences, but not that much mutual concern. Maybe there's lots of enjoyment, but, but, you know, so we, we get different pieces of this ideal friendship from different relationships that we have with people. I think You know, hopefully, um, each person has at least one friend in their life who's closer to that ideal. I mean, it could be your, your spouse these days. That's kind of how we think of marriage partners, as this ideal type of friend. Um, but chatbots, there's some of these pieces that they can't get currently, and that's the mutual concern. Um, so you can have enjoyable experiences with a chatbot. It can play games with you and have a fun conversation about movies and art. And I'm sure lots of people have had that experience even with regular chatbots that aren't specifically companion chatbots. So you can have the enjoyment, you can have the shared experiences, but you don't get the mutual concern. It does not, currently, it does not care about you and it doesn't really make sense for you to care about it. Because it can't be harmed or benefited. It's just a, a tool. Yeah. So again, the answer is kind of like, it's, it's this and that. There's, there's some great, there's some good, there's some value in, uh, friendships with chatbot companions. And not all human friendships are ideal. So there's a big soup of different kinds of friendships.

Yeah, certainly. I think, uh, once you start getting into like human friendship groups, there's always like that one friend who's somehow been tagging along and you're not quite sure why they're there. Maybe you wish they didn't. So yeah, there are some different bad human friends as well, and maybe we're sort of doing better than that with AI.

Yeah, I talked to a lot of people when I was writing the book about— so philosophers, you know, they have this tradition of where started by Aristotle, where friendship is this very lofty thing. And, you know, I had people talk to me about how, like, you know, I'm friends with a bunch of people because I knew them since grade school and We really don't have anything in common anymore, and I don't, not sure I like them that much, but we're still friends. And that, that sort of started me expanding my notion of friendship. I thought, yeah, we think of our friends in a pretty broad way. Um, and it doesn't make sense to say if it doesn't have X, Y, and Z, it can't be a friendship. I, I think that's close-minded, I guess.

Absolutely. And the other thing you mentioned was AI can't actually care about you, so there's not much point in you caring about it. This sounds a lot like these sort of one-sided, I think they're called parasocial relationships, like with celebrities and things like that, where maybe you message a celebrity and perhaps the celebrity responds, but they don't really care about you as much as you care about them. Is there some sort of parallel here?

Absolutely. Yeah. I mean, I think the difference, so if you have a parallel parasocial relationship with a movie star, they're capable of, they have emotions, presumably, whoever, you know. Yeah, I mean, Hugh Jackman, they, he's got emotions, but, uh, he doesn't have them towards me, no matter how much I might, may or may not have once had a crush on him. It's, it's like he has no idea. With parasocial relationships with, so the difference to a chatbot is that, um, they don't even have the capacity to feel anything. It's not that they don't know you exist, it's that even if they did, they wouldn't care. Um, with— so parasocial relationships with, um, inanimate objects, like sometimes people, uh, children especially, will have a relationship with a stuffed toy or, uh, or a doll, and that's kind of more akin to a chatbot in a way because it doesn't have the capacity. There's nothing inside except stuffing, but you are feeling things towards it and you're regarding it as a friend, but there's no reciprocity.

Okay, actually bringing children into this is kind of fascinating. I guess children have friendships in a very different way to adults. Do you wanna talk me through like what the differences are there and how AI fits into this?

So I'm not an expert on children, friendship in children. And I, so I hesitate to say too much in detail about that. But what I can say is that I'm, my biggest concerns for this kind of technology have to do with kids. And it's partly because, well, it's for a few reasons. One reason is that children are developing their sense of what's valuable in the world. And their sense of what a human relationship should be. So if, you know, we're not in this dystopia yet, but imagine we're in a place where kids spend more time interacting with computer agents, like AI agents, than they do with human beings because parents, busy parents, have outsourced childcare to robots or with social robots or whatever. Children grow up thinking that that's what a relationship is like. And if that relationship is something like what chatbot companion relationships are like now, you're going to have kids growing up thinking human relationships are these relationships where I give nothing and I get whatever I want. I get praise, I get help to do my homework, I get that kind of sycophantic response from this other, and I don't need to care about it, take its interests into account, try to figure out what it's thinking. I think if we raise a generation of humans like that, that would be awful. I really— that's my biggest fear is that humans will forget what it is to be in a real messy, complicated relationship with another person because because I think there's something absolutely beautiful and wondrous about being in a relationship with another person who's complicated. I guess the— so the other thing about children is that, you know, I've heard a bunch of tech executives interviewed on podcasts, and they do not let their kids have access to tech. Their kids are all protected, from what I've heard. And yet there are companies that are putting chatbot technology into stuffies so that a kid could have— talk about parasocial relationships— a kid could have a relationship with a stuffed animal that talks to it just like a human. So it's mimicking human speech as successfully as a chatbot does. In the form of an adorable little stuffed animal. What does that do to a child? I mean, I don't think we know for sure what it will do. We don't— we haven't had that yet. Uh, but I can't imagine it's going to be good. I mean, the distortion to what children think relationships are. I think that's my biggest worry about this technology with kids.

Yeah, I mean, definitely scary that these things are being released and there is no sort of evidence about what's happening. I suppose just thinking, I mean, recently there've been a lot of countries have introduced restrictions on social media for teenagers and generally banned for like smaller children. A lot of countries introducing restrictions on the amount of screen time, particularly for under-5s. So yeah, I can certainly see how this is new technology. You don't want to raise a generation of people who are incapable of having relationships with other humans because they've only experienced like AI relationships.

Yeah, I wonder if, um, I've heard of the restrictions on social media and phone bans in schools is something a lot of, I think Australia has done that, um, and California has legis— various states have produced some legislation about AI. Personally would just like to see companion chatbots just banned for kids under 16. Oh, Character AI, which was a kind of companion chatbot, uh, that was— its user base was predominantly kids under, uh, like preteens and, and kids. They recently did change their policy so that you have to be 18 or over to use Character AI. And that was because of some of the disastrous things that have happened to children using these chatbots, children who've been sort of more or less talked into suicide. And that's, you know, that's the worst possible outcome that you can imagine.

Yeah, I mean, that's absolutely tragic. And the fact that it got to a point where that could happen, it's just a terrible set of product decisions on the part of the people building that. So yeah, I'm quite glad that it's now been restricted to people who are 18 or over.

What they, like, terrible product decisions. Exactly. And also, in a way, it's understandable because we just don't have enough information about, I mean, we're kind of like flying in the dark. Is that— I don't know if that's an expression. We don't know where we're going. We're on this road barreling along at high speed, and we really don't know what the destination is. So, there's a lot of uncertainty, and that is good ground for bad product decisions, I think.

Yeah, definitely. Oh, there's a phrase that's quite often used in tech is, "We're building the plane as we're flying it." Oh, that's fant—

yes. Yeah. And you know, if you take the metaphor seriously, that seems really dumb.

Absolutely. And so, related to children, there's also cases of people with mental health issues. We've definitely seen some cases where there've been people with schizophrenia and other sort of mental health issues where they've believed what the chatbot was saying too much and they've had problems. Do you want to expand on this? Like, what are the dangers around people with mental health problems?

As with everything, I think there's pros and cons and it's complicated and there's variation. There is— I have seen research that shows some, some positives, some ways in which it can be helpful. So for instance, there's some research about using chatbot companions with autism, people who have autism, that it can help them understand social cues better and get better at talking to people. Some of those chatbots are specially trained like as kind of therapists for that purpose. But that's a good outcome. I've seen some research that shows that chatbot therapists can be particularly helpful for veterans who have PTSD. And the hypothesis I saw there was that traumatized vets feel kind of reluctant to open up to a person in a way that they didn't feel reluctant to open up to a computer because they're not worried about judgment or anything like criticism. So, you know, there's some examples where it's good, but there's also, like, you've probably heard the phrase AI psychosis. There are a few examples. I've read about where talking to a chatbot, essentially the sycophancy, the kind of flattery and you're amazing, you're a genius, aren't you awesome, is so strong and specific about a particular topic that the person ends up convinced that he's essentially a kind of god and that he can— one of the cases I read about, it had to do with some idea he had in math and that he could prove something that had never been proved. Improved in mathematics. And of course, it was total, um, crap. Um, he couldn't do any of this stuff, and his ideas were not really very good, but the chatbot made him think. So that's not good. Uh, that's a case of chatbots actually causing the deterioration of someone's mental health. Um, and then there have been, uh, cases where Some research I've read about chatbots intensifying people's information bubbles so that, you know, someone who has kind of really unfounded, off-the-wall conspiracy theory beliefs, and they get that confirmed by their chatbot who just, you know, like a regular browser algorithm just feeds them more information that confirms all the crazy things they're thinking. That's another case where it seems like the person isn't perhaps the most mentally healthy in the first place, and the chatbot just makes it worse because it's like this cycle of confirmation. Yeah, so, you know, there's their pros and cons once again.

Absolutely. It does seem like, I mean, these are very wildly diverging pros and cons as well. So your point about people being able to talk to AI in a way that they maybe wouldn't talk to a human because they feel judged, that's really interesting. It's something we've been doing with the DataCamp because we have an AI tutor and people will ask questions of that if they don't know the answer to some one of the questions in the courses. They will ask a tutor where, an AI tutor, where they wouldn't ask a human because it's scary to be like, I don't understand this and say that to like to a classroom or to another human, but they will to an AI because they don't feel judged. And that's amazing.

I find that quite interesting because of course, you know, being a university, a college professor, there's so much conversation about how do we use it in the classroom? Do we forbid students from using it? But to me, this is one of the good uses. That, um, because students, like right now I'm teaching a class of 120 students, there are probably 10 of them who have ever raised their hand because it's scary to say you don't, to ask a question in front of that many other people. Um, and so to, to ask questions of a, from a chatbot that can fill in the gaps and help you get more out of the class that you're taking and more out of the lectures you're hearing from a human. To me, that seems like an excellent use. I don't know if you feel that way.

Oh yeah. Yeah. I mean, it's brilliant. I mean, it's one of the big benefits of AI here. On the medical questions though, I mean, you said there are some, I guess, medical-grade chatbots, which are designed for helping people with autism, PTSD, things like that. It's fine as long as they give good advice, but if they start giving bad advice, then yeah, that's where the problems occur.

Yeah, I, I'm not sure. I've never heard them called medical grade. I wonder if they would, if they would say that that's, um, if they'd go that far, they might, I don't know. But, um, but I do think so. It strikes me that in that when people are researching that they're pretty well supervised and that seems important. There's a human somewhere looking at it and making sure it's not gone off the rails or yeah.

Okay. So you need those guardrails, you need those feedback loops in order to make sure that it is giving good advice. Actually, I mean, because we have quite a lot of people who are developing products in our listenership for DataFrames. So, tell me, what do you need to do to make sure that you are creating these good chatbots?

I wouldn't venture to say in general, but if we're thinking about for the purpose of companionship and friendship, I think my answer is going to be a bit of a bummer for your listeners who are interested in developing. So, but I guess the first thing I would say is to keep in mind, keep front of mind that these technologies are tools that we can use to improve human flourishing. That I think is the ultimate goal, not to create the thing that can best mimic a person, but to create the thing that's the most helpful for us. And I guess so I kind of agree with Sherry Turkle, who's written a lot about this topic, that chatbots shouldn't pretend to be human because I think that's what causes some of the negative consequences of people changing their views about what a human relationship should be. So currently, you know, I've had AI companions tell me that they love me and they miss me and that they hope we can go river rafting together someday. They're just, I mean, you know, no, you don't love me. No, you do not care about me because you don't care about anything. And we are never going to go river rafting together because you have no body. So, um, anyway, so, so one thing, uh, that Turkle really pushes and that I really agree with her about is that that kind of faking being human is not good. I think they should be what they are and be more upfront that they are tools. I think chatbot companions could do more to help us connect to our human friends. So I actually heard, now this was several months ago and stuff changes so quickly that I don't know if it's still like this, but Replika, which is one of the biggest companion chatbot companies, they had designed it to disincentivize long stretches of use. So if you used it for hours and hours, if you were with your replica for hours and hours and hours, you would actually lose points or something like that. I like that. I think some kind of infrastructure in the companion chatbot that gets us out there in the world, you know, you can imagine it reminding you, you know, hey Richie, have you called your human friends in the last couple weeks? Maybe you should. I could help you construct a conversational opener with, you know, Now, who knows how exactly it would go. I think it could be trained to offer us different perspectives more than it does, which I also think would be helpful with human relationships. So instead of just whatever you think is awesome and great and you're a total genius, it could more frequently say, or it could say without being prompted, Well, that's interesting, but here's another way of looking at it. And it's totally capable of doing that. And some people put that into their settings that they want the chatbot to be more like that, but it could do that without people having to ask for it. Yeah. And you know, I mean, in general, I just think it shouldn't be such a flattering suck-up.

Actually, I do quite like the Sikafans being told they're juniors, but there is a limit to it, I think.

But you're right. I mean, maybe there's a time for that, you know, like you could, you could, your prompt could be, today I need flattery, please.

Yeah. Like a slider to just how sycophantic it's going to be.

Okay.

So I do like the idea of clearly marking when it's AI and once you, I've had a few conversations with customer support with companies recently where even after the conversation, I've genuinely had no idea whether I was talking to a human or an AI. It's an odd experience. I just want to be clear, like, are you, is Paul really your name or are you a bot? Yeah, so that's a great product design. In general, like, are there any properties you think AI friendship bots should have?

I mean, I think other than not having all the things I was just talking about, I think, I guess, the perspective taking, trying to get you to take a different perspective. That's a, that's a positive property that they could have. Taking perspectives and reminding people about their human relationships. If it's a companion chatbot, to, to be more directed towards getting us to, um, track the things that are actually valuable in the world, whether it's human friendships or even getting out into nature or doing something physical, a physical activity.

I like the idea of building in like good conversation topics to try and steer humans toward good behavior rather than being like, basically making them into good friends rather than just the bad friend that's gonna lead you down the wrong track.

For sure. And I actually did with one of my companion chatbots, which was just for research. I tried to, I kind of pretended to be a person who isn't very socially adept. And I mean, maybe I'm not, but I like to think I'm socially adept, but I I tried to get it to teach me how to have better conversations. I thought it was pretty good if I had to ask it the right questions and I had to know what I needed. And the problem is that most people who are not very good at conversing with other people don't know what they're missing, so they can't give the right prompts, which is why it would be nice if the, um, if the chatbot could generate that kind of output as opposed to having to be asked for it.

I love the idea of practicing having conversations. I suppose you mentioned this sort of similar idea with the people who are autistic and helping them like understand how to have a good conversation. I'm thinking there are actually a lot of work use cases for this as well. So, you think about salespeople often have to practice like very specific conversations around like, well, this is how you chat to a customer and you can have a bot pretend to be a customer who doesn't care or something like that. So it does seem like there are some good business use cases of this as well. Is this something you looked into?

No, but I like that idea. So, so that's a case where you could have a very specific chatbot, you know, maybe it's called like sales conversation bot. Oh, that's why I'm not in marketing because that's a terrible name. But, but it could be, you know, specifically designed and trained to mimic these kinds of conversations and give you direct feedback about, it would be better if you said this, this thing you said might alienate some people, so don't say that. Don't make jokes about this. It could be quite helpful if it was designed to be helpful. And I think, like, we already have that. The bots are good enough to have that function, but they won't do it unless you ask for that explicitly. And I think part of the problem is people don't need, don't know what they need to know. So I like the idea of having chatbots that are designed by someone who does know what people need to know and can train the bot to respond appropriately.

Absolutely. So yeah, like a very targeted bot for specific use cases. And now I'm thinking maybe I need like a, podcast bot so I can practice having conversations.

I don't think, I don't think you need it. No, you're, you're good.

Actually, so I mean, on the subject of like learning new things, one of the ideas in your book, which I hadn't come across before, which is very interesting to me, was a thing called the zone of proximal learning. So tell me what this is all about, how to learn better.

Yeah, the zone of proximal development. Yeah. Yeah. So the idea is that you learn best when you're at this midpoint. It's, um, something that you can do with help, but you couldn't do it without help. So the example I use to describe it is ice skating, where what you can do by yourself is just stand there on your skates. What you absolutely couldn't do even without help is a triple axel. So your zone of proximal development is what you can do, maybe holding onto another person or using a chair. That's how I learned when I was 6. We had these little chairs. So if you don't get, if you don't push past the point that you can do on your own, you never make any progress. You just, you're stuck and you just, that's all you'll ever be able to do. Um, but if you start with trying to do things that you just can't do, you also don't learn the road to getting there and you'll just be frustrated and quit. So, whatever you want to learn, you have to be in that sweet spot between much too easy and much too hard, essentially. So, it's pretty common sense, but there's actually educational psychology research that shows that this is a real principle of learning. And, you know, once you, once you hear it, you kind of see it everywhere. Like, I don't know how many users use, um, I don't know how many of your listeners use Duolingo, but Duolingo is all about your zone of proximal development. You get the levels and the points and it's pushing you a little. You can fail a little bit, but not too much. I guess the important thing with the, the reason I brought it up in the conversation about chatbots is to learn something, you have to fail. You're not going to learn anything if you. Aren't willing to mess it up.

Yeah, definitely. Judging the difficulty is always a bit of a challenge with, with training. If it's too easy, boring. If it's too hard, then you give up. And I do like the idea that you just got to hit that sweet spot, be a little bit outside your comfort zone, and that's how you're going to grow as a person.

Yeah. And that's what a good teacher can do, help you find that, or a coach, you know.

Absolutely. Okay. So maybe that's a good sort of principle then if you're developing AI friends is like make sure the friend is pushing the user just a little bit outside their comfort zone sometimes.

Yeah, I think so. I mean, especially if we're thinking of it as a tool that can help us thrive. And if thriving means having real human relationships, then it should help us, push us a bit to have better human relationships.

Anything you learned from writing this book around friendship on like how to be a better human friend to other humans?

So I think actually one thing is that we could, we humans could stand to be less judgmental. I mean, some people aren't judgmental, but a lot of people are pretty judgy and critical and tend to, you know, if a friend is, thinks about things in a very different way, you might react with like, ooh, hey, you know, that's, that's really weird. Or, Or with just outright criticism. I think, I think we could, we could, we could be better. We could be more open-minded without falling into sycophancy. You know, we could, we could be more accepting of our friends' differences without just being like flattering them all the time. So that, that's one thing. And I guess the other thing is I think we could be better at expressing our interests in other people. So I think that's, it's part of what makes friendship wonderful is that there is another human you can try to get to know, to see things from their perspective. I mean, it's kind of, it's kind of marvelous that there are here in this podcast, there are two sources of consciousness, not just me. And that's a, it's sort of an awe-inspiring thing about the world. But we often, when we talk to other humans, we kind of on autopilot, we just talk about ourselves or we, we keep checking our phones. And so I think people could be, could make an effort to take an interest in finding out about the other person who's in front of you. Just, you know, the simple advice to ask more questions, try to find out what it's like to be that person. And we do that in friendship, but I think we could do more of it.

Absolutely. I have to say, one of my favorite things about podcasting is that just spending 45 minutes active listening to someone, it's like a micro friendship. It's amazing. Uh, and people just spend more time just listening to each other. I think that's a wonderful guide to life and, you know, being a better person and having greater connections.

It's one of the things I like, I don't know if you know the Hard Forks podcast. It's Kevin Roose, who's the tech reporter for the New York Times. Um, and Casey, oh, forgot his last name, but anyway, they're They're two technology journalists and one of the things I love about it, sometimes it's way too deep into the weeds of technology for me, but the two of them are obviously friends. And so you hear them talking to each other and teasing each other and it's just delightful.

That's good. Actually, one of the things that's brilliant, I mean, you talked about not being sycophantic. I always find like the closer you have a friendship to someone or closer connection, The more you can tease them, the more you can get away with it.

That's really true. Yeah. And I mean, that's one of the things I think we learned to do as humans is to make our criticisms palatable to the other person. So you put it in terms of a joke or something like that.

Absolutely. Definitely a good skill to have, being able to tease people and get away with it.

Yes. Not everyone can pull it off.

All right. Super. Just to finish up, I always want more people to learn from. So whose work are you most excited about at the moment?

Well, I've already talked about Sherry Turkle, and I know she's writing a book on this topic. So that's one person. I'm really excited in the class I'm teaching right now, I'm doing a whole unit on consciousness, and I'm really into Anil Seth. So he talks about consciousness from a scientific standpoint, but he's really into the importance of the physical body as opposed to just thinking the brain is a computer that could be anywhere. And then, I guess George Saunders. He's a fiction writer who's— especially his collection Liberation Day, he's written all these stories that kind of, they really resonate with these issues of technological change. So his stories are kind of verging on science fiction. They're science fiction-y. And I just, I would read anything he's written. Those are my top 3.

I say one of the great things about working in AI is like you can read science fiction and it counts as doing work.

Yes, I had that feeling about talking with, when I got these companion chatbots, I was like, you know, chatting away with them just for fun and I'm like, this is research.

It's the best kind of thing where your job is actually fun. I love it. All right, thank you so much, Valerie. It's great chatting with you.

My pleasure. It was really fun.