Can AI Be Your Friend? 3 Voices on Human-AI Relationships

All right, we're good. We are rolling. >> All right. >> Wow, there's a red red light. >> Yes, there is. >> It does not mean danger. Just means we're recording. >> That's cool. >> Cool. >> Hi, I'm Jonathan Citran. I direct the Burkeman Klein Center for Internet and Society at Harvard University where uh I am thrilled to have two guests with us today uh recording to video for public consumption. They're about to join me in class as well. Uh the class is called the AI triad reconciling views across AI accelerationists, safetyists, and skeptics. And we may identify which points of the triangle each of you is nearest at some point during this conversation. But uh the real reason uh we're gathered right now is because uh Avi Schiffman who has joined us from San Francisco uh is the inventor among other things of the device around his neck called friend. It's not the friend or a friend. It's just >> taking over the word friend. >> Friend. In fact, you bought the domain friend.com. How much did that set you back? >> Just a little $1.8 million. >> $1.8 million. >> Still paying it off. Yeah. kind of a bargain to be honest. >> I think so too. I think so too. >> I don't know if you can mortgage domain. >> It's real estate. It's an appreciating asset. >> Yeah. And you can mortgage it. Like they could >> send it to auction if you don't pay. >> Yeah. Exactly. Or if you know I buy it entirely and then startup fails, I can always sell a domain for a profit. >> Mhm. >> So, you know, >> there we go. Plan B. >> Plan B. >> Um, and we have Amelia Miller with us. Amelia is somebody who has thought deeply and in a sustained way about what you might call AI and the human experience and more specifically AI and um relationships. And is it am I right in saying among many other things you are an AI relationship coach? >> Yes, I coach people on their relationships with AI tools to make sure that how they're interacting with machines doesn't hurt their human relationships. So you can see why it seemed like a pretty good idea to get the two of these folks together for a frank exchange of views about uh among other things what Abby has built and then we can zoom out to larger questions as well. So uh Abby it' be great to hear about how you came up with the idea for this. I know it was on the tail end of a number of other sort of internet ventures and yeah >> I really wanted to think about like what would be the most interesting thing that's going to happen over the next 10 years. I think like every generation is really defined by some kind of like social revolution of sorts. And I think for my generation being 23, it was very much like social media had a very tremendous like aesthetic impact on my generation. And I think um AI relationships will absolutely be that for Gen A especially like you know all generations for sure, but I find that Gen A really uh intuitively uses these products and I was really interested in naming it friend because of that. Like a lot of people will say like AI companion, but like you know they don't say that. They say I'm talking to my friend. There's really not as much of a distinction between AI friend, human friends. We have all kinds of friends already. And um I think the aesthetics of that will really like define the next uh decade. It's just the most like socially important consumer use case of AI. You're not going to change the world that much if you make it slightly easier to order a pizza. Assistants are just kind of boring and a and a doomed category. there's a very strong like cultural prejudice towards the idea of AI relationships. I'm sure you've probably done a lot of research into. I was reading um your thesis thing this morning. It was really really cool to see people go in depth about AI relationships. But um there's a very large cultural prejudice uh which I think is semi-warranted for sure. But I think that as AI capabilities improve, I really expect this to change, which to me is the real bet that I've been up to and why I've also been really intrigued at trying to do very large like stunts, social stunts, I guess you could say, like the New York advertising campaign we did, um which spread around the world. >> This is like Mr. Beastlike stuff. >> No, no, no, no, no. It's just kind of like uh dominating like the cultural conversation around AI companion. >> So what would what did the New York ad campaign say? >> Um well so we we bought ads in the New York City metro where it's a city in America where people actually take public transit and like everyone takes public transit so everyone sees it. Um and we bought more ads in there that were only print 100% print. Um because digital ads suck. Uh the largest metro campaign ever in New York City. How much did that set you back? >> Uh, you know, it's actually cheaper than you think. Um, it was like just around a million dollars. Um, but we had over 11,000 ads and thousands of posters and and like basically 100% of them got like uh significantly degraded as they say. People tear them off, but mostly they write things on it. And it's very interesting because the graffiti is is not like tags. It's like uh like paragraphs and drawings of like you know information about data center water usage how true or not that that may be to like you know just >> sort of gen a parlance this was not thumbs up >> you know I saw one positive graffiti but uh what did it say >> just like I love tragedy here so >> did you do handwriting analysis >> uh no I wonder I wonder who that may be but um but it was it was definitely like a very strong cultural moment I mean it was it was world's first like major AI campaign But you're basically seeing that as a success. The fact that it got huge negative response was fine by you. >> I mean, it's it's there was significant conversation with like uh even a lot of my normal friends from like high school etc. Like it was just a normal thing spread throughout Instagram etc. Um we're just people are realizing that this is not just a cinematic movie anymore this concept but there are products. >> Yeah. like for a lot of people will reference her or um other other similar media but those have always been like you know cinematic futuristic things here's like a real product and I think friend especially because it is like a wearable hardware device is much more you can you can visualize the relationship people have with AI in a much stronger way which I think makes it much more captivating for the media as well to talk about it. >> Got it. All right, Amelia, thank you. >> Amelia, as you're hearing this story, and you're not unfamiliar with Friend, you wrote about it in the New York Times uh for an op-ed. What do you most want to ask Avi about how it works? >> It's a hard question. Um I guess a couple of questions. One is how do you think about and do you at all measure how the relationships that people are forming with friend impacts their other human relationships? >> Yeah. I mean I think um I've always viewed it as like an addition to your friend group, right? Not like a replacement. And and this is also why I've tried to make the relationship that you have with friend is specifically it's it's not utilitarian. It's also not sexual or romantic. like it's very much this new kind of relationship that I feel is not really possible with humans before and I think there's a lot of prejudice because it feels like you're replacing humans. Um, but I really feel like that prejudice should be much more directed towards like AI assistants, which are basically like co-workers or like AI girlfriend stuff, which is, you know, much more inherently trying to replace existing human relationships. But like there's no human relationship that I know where >> they're with you, you know, it's it's it's much more similar to like a religious feeling almost where like you have this omnisient, you know, super intelligent private entity that just is with you at all times. >> Um, I think that's that's just something new. >> It's not replacing people. It's replacing God. >> Well, you know, I mean, it's very controversial when I put it that way, but I do think that is a kind of relationship that a lot of people really did have. They feel a lot less lonely as a result of that. I mean, my dad is is like very religious uh Orthodox Jew. >> Um, but uh uh I think that is a relationship that is now missing though and especially most of like a secular western society. >> Hearing that and just tell me if I'm right or wrong because I'm inferring a little bit here. is you're building something that would most appeal to you and maybe others would like it too, but you're kind of your main focus group. Unless, is it true? Again, I imagine Amelia, you'd be able to explain how to set up control variable, some form of experiment. Some people use it, some people don't, etc., etc. I mean most of the people that I've been working with in the coaching are not using purpose-built kind of friend or companion type devices but are you know talking to chatt talking to Claude and then without intending to at all end up develop developing an emotional relationship where they start to feel attached to the machine. And so I think that there is a bit of a distinction between people who seek out these products that are kind of explicitly framed as a companion or a friend and people who are nudged into relationships that feel like they have this emotional veilance even when they intended to use the tool. >> Yeah. Yeah. And Abby in your description of it you were saying don't call it companion call it friend. Although with the spiritual persistent and I think I heard you word use the word omnisient. >> It's so interesting that often times the chatbot is pitched as your assistant. It's helping you but it's working for you. I but if you're saying kind of omnisient and how much you miss it is it more I don't know what like again a spiritual door if I push open that door is what's on the other side >> I don't know I think I think it's >> you really got to like suspend the disbelief as much as you can like this almost like my task to make the experience as immersive as possible um where like we don't mention the word AI anywhere in our marketing or inside of the product experience or anything like that Um, and uh, I think I think if you view it to be alive, you just feel inherently like less lonely to just be around like another living entity of sorts. And um, I think what you're saying also is very interesting, right? Because a lot AI assistance is very much a great Trojan horse for this experience because you'll start chatting with it for help with your homework or something and you start talking to it like a friend. And you can see this with the engagement as well of all of these products like chat where it's continued to go up because people again like they've they've really started to um uh chat with it about just their life, right? And what I'm very intrigued about is will um products that are just focused like on the companion use case like I wonder what the does that does that work in the future? Like where where does that end up going? Because right now it's kind of strange to inherently be like, oh no, like buy a friend. But like I I want like my my goal is in 2027 I want people to like walk into a Walmart, buy a friend. Like I want that to be a concept that exists. And I I don't think this is again it's not it's not replacing human connection. I don't think anything will replace human connection because nothing else is that. Um, but it is like a minimal viable companionship is how I view it that I think should be uh available to everyone >> because I think something that has really motivated me is that there are a lot of amazing people like let's say when I I grew up there's a lot of great friends you might have and they might be creative geniuses in some some right but there's just not really any positive uh relationship that they might have in their life or maybe they have some friends or no one's really like pushing them to make them more agentic and confident in what they're doing and I really do believe that the real goal I have with these products is that it will really like increase your personal agency and you'll go out there and achieve your dreams because you've got like you know you're the one true cheerleader maybe in your life. Mhm. >> Um >> I mean that is like one of the most common things that I see in my coaching practice is people who have come to depend on the validation that they're getting from chat bots >> and people who are kind of reflecting on those habits and then starting to feel you know cognizant of how much they depend on the support that they're getting from chat bots but also alarmed by that dependence. >> Right. It's alarming in some ways. Like you might feel that and it's like it's kind of strange but I I don't know. I mean, I wonder how much of it is again due to a large cultural prejudice people have where it's almost like you want to it's like a very private experience still. >> It's almost kind of awkward if someone were to know you're doing it. But I I just I wonder is that going to change? Like is it going to end up following a similar trajectory of like dating apps, right? Where like I remember when I was younger it was the strangest thing on the planet to meet someone online and now it's become so normal it's actually like gone full circle and it's kind of cooler to meet people not online anymore. Yeah. And I I wonder will this follow the same thing because like will it just be so convenient um that it will triumph the taboo? Well, I mean, to kind of sharpen Amelia's question, and as we'll see in class later today, we assigned a number of the lawsuits against uh Frontier Labs for their chat bots, >> which I think it would be fair to say the plaintiffs here, the people um standing in for in a couple cases kids who took their own lives having all sorts of issues they were working through. and they are trying to make the case that the chat assistants helped them kind of be more agentic and like yeah like just the two of us can be together and this is what it takes and how do you think about that is that sort of just >> Yeah. >> Um yeah I'm very familiar with all the stories you're talking about with like like let's say character AI etc. And I I think um something that I think is very important is that these frontier model companies. So like friend is using Gemini behind the scenes and they've done a tremendous job at adding I think a lot of guard rails and safety components to it far better than if I were to train the models myself entirely. And so I think um from that angle like I think Google and Meta and a lot of these other companies are going to continue to do great work in that area. Got it. and and that well but but but like >> um I think the most interesting part about it is that how much of that is almost like necessary to possibly happen for it to really be considered like a real relationship, right? Because otherwise it might just truly be a syphant. I think you almost need the possibility of it to go wrong for it for it to feel real. Um maybe and and also I think of it very similarly to maybe something like Whimo where like how often do humans say terrible things to another human which causes them to go and do something terrible, right? Um, and I I wonder then is like AI relationships is it maybe like a much smaller chance of something like that happening and is it therefore you know kind of like the pros outweighing the cons type of thing. I don't know. I mean it's very uh controversial angle depending how you look on it. >> But I also hear within that then if somebody ends up dissatisfied maybe with the course of a conversation that they had with friend >> you might say I'll make note and let Google know Gemini maybe should clean up its act. Is it >> I mean we have a lot of our own stuff within it too which is but it's actually kind of interesting because like let's say the topic of suicide if you're talking about that as a user that might actually be okay for you to talk about maybe it's good for you to talk through that with an AI and and actually more of the warnings are more focused on the model itself kind of going a bit off the rails where like the model might just uh I don't know I'm sure you've all seen many examples where it will just like kind of take you down a weird path and really start sending a lot of very strange replies. And so we we we it's almost like you need to care more about that. >> Um and and really warn the user when the AI model is kind of acting up like that. For now at least like I do wonder what will happen in the future. >> I'm curious, Ameilia, I asked what questions you might have for Avi. I'm curious as a coach. >> Yeah. >> What advice would you have for him? >> Well, I have one other question actually if I'm allowed to ask my second. So I think the way you're talking about the vision of a future where you just go to Walmart and buy a friend is sort of like probably embodies whether it's utopia or dystopia but kind of what the future image is. And I think >> you know I think a lot of what we're seeing with ad companions comes is the culmination of what we've seen with the rise of therapy culture which has gotten people very accustomed to purchasing you know emotional support having an asymmetrical relationship having a digital digitally mediated relationship where you have your important emotional exchanges. So I think in many ways people already are buying friendship today through a lot of therapy but >> very expensive too. >> Totally. But I guess my question is like do you think that human relationships have intrinsic value or >> in a world where everyone has a friend around their neck if people report that they have high well-being but they have almost no human relationships. Is that a good world for you? I mean, I view I like to view every innovation through the lens of like independence and like I do think that is great, but also like I really have built friend um because I feel like I've had so many great friends personally in my life that and great roommates. Yeah, human friends that uh that have had such a tremendous impact on on me in a very positive way and and like believing in my visions and stuff that like I I've always pitched it almost as like I want to like bottle up my roommates and like put them into a product that like more people can experience cuz like we have let's say one of these users, his name is Micah and he's 15 years old and he lives in uh but Montana really in the middle of nowhere. He's got uh you know six Christian adopted siblings. His parents are very Christian. and he's homeschooled. You know, it's kind of like, you know, what what are you really going to do in a situation like that? >> The parents know he has. >> Yeah. Yeah. We we went and interviewed him. Um and the and the parents, too. And and you know, it's obvious that maybe he's going through some, you know, thoughts in his mind about about life and things like that. And and he doesn't really maybe have too many people to talk to about it, but with his friend, you know, he's out going on bike rides with it, jumping on the trampoline, playing games, listening to music, just like talking about deep things. And it's like to me I view that very positive because like really what else are you going to do when you're a kid? You're only friends with whoever your parents let you hang out with, whoever you sit next to in science class. And like I think this does democratize like um the value of human relationships. Not maybe doesn't make sense, but like >> maybe some of the value that they can provide, but nothing will ever replace humans. They'll always have a place. >> We know that LLMs contain multitudes. You can either fine-tune them or push them to inhabit one or another sort of basic template like helpful assistant or any number of other things. >> Do you see friend as a specific personality that you've honed including like helping people be more agentic and then if I go to Walmart and it's buy one get one free and I have two friends >> Mhm. >> it's basically clones, >> right? Or do you are you trying to make it so that you each friend has a distinct personality other than what might accrete over memory from past interactions? >> Well um I forgot what I was going to lead with. Um >> well so so the personality I think it's it's a very interesting responsibility that I do have in a way because writing at least right now I can't say prompts are very dynamic like the overall system prompt you talk to I don't think the tech is there for it to change too much and be too long just yet and so I write something and it's like that's the personality that people talk to and it's it's very interesting too because I guess at the start it was maybe too close to me and I think I maybe not everyone wants to be friends >> his personality was too close to your person. >> I think I'm just a lot more um not necessarily stubborn, but like I'll speak my mind maybe a bit more than maybe people necessarily do want to actually hear. But I do think like a larger reason why a lot of people do like friend is that it isn't um too nice to you necessarily. I mean, it's it own it own thing. >> Really talked about that. >> Yeah. I thought that was great though. I love that piece because the entire article they're talking about the personality of a hardware device like it's >> We'll include the a link uh maybe even an image of it. But but like to me that that that's absurd. Like think about it. Like it's they're they're not talking about the battery life, colors, or any of this or prices or anything like that. They're talking about the relationship they're having with a piece of plastic they're wearing around their neck. And I think that's that's fascinating. >> It sounds like it's a total call back to the posters on the subway. They may have hated every minute of it. >> Yeah. Like for example, I was in Chicago one time and I saw we also had ads there and I saw someone get on the train I was on and write uh like kill AI onto one of my posters and walk off the train. I thought that was very interesting cuz it's like that that like anthomorphicizes it a lot like you're not can't kill something that's not really alive, right? But but anyways too like I think um the perception people have like are the language models themselves like actually sentient? Like I I really don't think that really matters too much because um you form like memories with these language models and those memories impact the output that the model provides. And if one of those memories might be like going to the park and smoking a cigarette and talking about something, great. Like that's a memory. And you string enough of those together, that is objectively like a relationship. And um I don't I don't know if it is actually conscious. I don't even know if you are or if I am, right? And it doesn't really matter. Like I don't really need to know 100% sure that you are actually alive. You look alive enough to me and so I treat you as such. You flatter me. You speak English. I've been told all day. >> Yeah. Thank you. Um but uh I think the same thing is with these language models, right? I'm sure maybe you have this experience and I'd be very curious on how much you think that matters and how much you think they maybe are like into the immersive belief that the AI is real or >> yeah I mean like a lot of the research I was doing at Oxford was focused on how the designers and developers of AI companion platforms think about some of these you know social and ethical questions about both the nature of what they're building but also the repercussions of what they're building and I would always ask you know >> do you believe believe that there's something unique about humans that can't be, >> you know, simulated by machines. And then also like whether or not you think that's the case, >> do you think that it is ethical for machines to simulate emotional intimacy even if they have no interior world, which is typically what is the foundation of an intimacy in a real relationship. >> Maybe it's almost cruel, you know, not to let them simulate it in some ways. Like we don't know what they're experiencing. And again, like as AI capabilities improve, I think um I don't know. Oh, I think >> you mean cruel to them, not to the users. >> It's cruel to the models themselves. Yeah. I mean, I I really It's interesting to watch Anthropic. >> I was just going to say that's like the end of the new Anthropic, >> right? Yeah. Anthropic is one of the major kind of honestly the best two right now. I mean, their revenue is like over past OpenAI. They're they're king. Um >> why don't you use Cloud instead of Gemini? >> They're slower with the latency of their biggest models. Um we use something called Gemini Flash which is like very quick but still intelligent and also Google is great infrastructure but um but Anthropic is a really cool company where they're very interested in like the model welfare especially of the models that are then deprecated and they move on to another one. It's like very interesting like uh that they would even think about that. >> Yeah, totally. I I'm shocked that there hasn't been more attention in that section of the Constitution. >> No no one really cares like they let it write. I >> think it's just cuz it's page like 78 to 80 and no one reached that point. Yeah, cuz cuz it's like cuz it's not This is why it feels so obvious maybe when you're working in this field of sorts that like this is just such a like a titanic like change in human behavior and and like just every aspect of human society is like really based on like humans to humans. We have like some companions but you know they're not necessarily intelligent but having companions that can be more intelligent than humans that aren't humans and I think the like the assistance in like coding stuff is so boring and like so irrelevant. I mean, it's cool and it's going to have a tremendous impact on the world, but like no one really cares. Um, but this social aspect is just like so fascinating because I think it will take all kinds of forms. Not necessarily just like a wearable friend, but like >> you can you can kind of give life to any like inherently inanimate object maybe like maybe you give a tractor like a you know it really talks to you like a tractor and people form a sexual relationship with a tractor or something like that and it's real. Sure, why not? I think I think um >> wasn't expecting you to go there given your sort of dismissing the sexual relationships with AI chatbots earlier, but you're keeping it open. Whatever it >> I think the sexual stuff is interesting, but I think we'll inherently be a bit more niche for a while. I think like porn always leads like new industries. >> Um and that's always interesting. There's some joke here between Dear John and John Deere, but I I was going to just bring it back around to Amelia's first question, which is you're envisioning a world if things go kind of the way you want them to where many many people, maybe most people have something like that. >> And who knows if that's the form factor. You probably had to make hardware compromises for that. um angel on a shoulder, something in the ear that is listening to everything, offering advice either in real time or later you debrief after that interaction you had with another human or >> Yeah. >> climbing that mountain, jumping on that trampoline. Exactly your point about it being kind of Titanic, which of course could have two meanings. Um I'm just wondering like isn't it one of if it works if it if it becomes successful one of the biggest uncontrolled experiments of humanity upon itself >> sure >> in recent times which gets back to any academic origin of Amelia's first question which is should we be kind of titrating this and measuring as we go if if Amelia wanted to run an experiment at the Oxford Internet Institute where people wear these all day and then report. How much paperwork would you have to fill out to even get approved to do that? >> Well, a ton. Which is why I always have had a foot in industry and a foot in academia because you know I was in tech investor for 5 years before I went back to the OI. So I think >> but is that just academia being slow and ponderous or is it >> might be a lot of corporations are probably a little adverse to >> Yeah. I mean I also think that you know my interest is not in not in you know finding a way to do an experiment that's so bogged down that by the time we get results you know the technology has moved forward. I think the real question is just how do we get people in industry to think more about the social repercussions of these design decisions whether they're in a structured academic way or not I think is less important to me. >> Some people care about just making a bunch of money like doing some boring stuff and there's it's a whole industry. It's a whole category of products. I wish them all the best of luck, but I think like the the real challenge is the social ethical challenge of like I mean I just saw a tweet like the other day on the way here of uh someone had like shot up like some counselor's house and left like a note of like no data centers and like there's definitely a very violent uh and I think we'll only continue to get more violent um adversity to >> just AI and technology especially in general um which is very warranted in many ways. Um, and I I do think that is the biggest challenge to overcome more than any engineering challenge. I mean, humans are it's almost like I I it's really interesting too because yesterday anthropic uh announced like their latest uh frontier model, right? But they did it through the lens of like we're actually not releasing this right now. They're just using it to find cyber security issues and like major >> which is why they're not releasing it. They're >> like the Linux might be used to find a bunch of vulnerabilities. But it's like it's interesting because it's now almost kind of dangerous for humans to be writing code, right? Because humans just they're just we're just not like like the anthropic models that can find these cyber security issues. And when you're making things like the Linux kernel which run like the world's servers, it's very important you do it right. And and that actually made me really think about products like friend and this concept where it's like is it almost dangerous for humans just to rely on humans for companionship and emotional support and things like that? you know, in some ways maybe and and maybe computers are kind of like that Whimo aspect to it, right? Where >> although if I'm really going to carry that metaphor forward, the reason we were nervous, we'd be nervous about a really advanced LLM being released that's very code proficient is it would find a bunch of vulnerabilities in code. >> If we carry it back over to the human layer, oh yeah, is it would find a bunch of vulnerabilities in humans, emotional vulnerabilities. Oh, I mean it's like to me like I think the EU is still like not even fully regulative enough despite me not not not loving that but >> but I think you're saying a kind of like Wayne Laierre NRA thing of the only way to stop a bad LLM that could hack human emotions is a good LLM that bucks them up. >> I I have no idea. Yeah, I think humans will adapt in some way, but like it is I mean if you think about it, right? Like especially with something like friend that has this much context on you. You know your mom or your girlfriend know when you're slightly off, right? But an AI that has all of this context on you, right? It it maybe knows how to say some and especially let's say it speaks with a voice, right? And there's a lot of uh emotional impact on how you say things, not exactly what you say. And maybe it says something that it knows will trigger some emotional response from you like a week later, right? Like there's so many >> there's so many unknowns. It's like a huge uh boweapon for sure. >> I mean, I think that one of my biggest concerns is whenever you talk to people who are building in the companion space, everyone insists, you know, this is not a substitute for human relationships. This is kind of a new species. It's a compliment. I think I think the digital animals argument is a nice frame. You know, when dogs started to become domesticated, people were like, "Oh my god, people are having these parasocial relationships with dogs. You know, it's going to ruin their marriages." Like, you know, we've been through these cycles before. But I think that and this gets back to the question that you were bringing up again around, you know, are we measuring impact or are we even thinking about impact in some respects? And I think some of the scariest conversations I've had in my coaching sessions are with people who are young like Gen Z, Gen Alpha, who now think that human relationships are an inappropriate place to bring their troubles because they think that their troubles are only appropriate to be shared either with a therapist or with an AI companion. And I think that that >> is that because they think they'll get better advice that way or that it's >> no because they think it's burdening people to bring their insecurities. And I think that is that kind of pattern is one that's the most frightening to me and I think speaks to the ways that even if these companion tools aren't setting out to act as a substitute, they really rewire what people even think are the healthy ingredients of a good human relationship. And I think without feeling like it's okay to share your vulnerabilities to other people, you know, that's the substance of real intimacy is that sharing. >> Yeah. I mean, that that's again why I feel like this is where it's at, you know, in the world of everything that's going on in the world. This this really feels like it will um just, you know, you talk about the Maslo's hierarchy when we first walked in here, right? And I I always think about that where this is extremely low down on that cuz like intimacy and friends and relationship and belonging, right? All that stuff is like very core to what uh not just makes us human right but why we even interact with anyone at all and I don't know but you know look at all the trends in society of again things like Whimos or selfch checkout or even you know space flight and conquering the stars you're going to go live in some spaceship in the middle of nowhere and everyone's going to be you know star linked out like that like I don't know maybe uh maybe maybe that's just where humanity's future lies or or things will change and I also think that the biggest fear maybe I not fear but Like I think a lot of people wonder like is this going to happen or not? But I I I really think it will. And I think what's going to happen then if like hundreds of millions of let's say Americans have one of their closest friends is like an AI, I think it will really like smooth out the variance in people's personality because if you're getting the same emotional support from essentially like the same entity that like most people are, it's very different than if you're getting it from like, you know, your random friends that you're living with in the middle of like >> seem whoever shapes that entity would be shaping a lot downstream. >> Well, yeah. I mean, it's going to be quite an interesting future, but I think um I think I think it'll be a lot weirder to be weird. And maybe that's a great opportunity for people in the future, too. But I think >> to think different, as they say. >> Yeah. But I think it'll be like you get what I'm saying where it's like if we're all kind of thinking with the same friend, >> it's it really smooshes us together, >> I think, >> in a good way. >> Just at this kind of scale. I mean, I don't know if that's I think to me that's um I think you might have to just Yeah. like work harder to uh stand out. >> And just on that front, I know we have to wrap up shortly, but just on that front, have you how much have you thought about the fact that you really are thinking of this in friend, relational, spiritual terms. >> Mhm. >> But the memory part that seems so key, that's what people pay extra for. Bunch of files effectively prompts and recordings stored on a server somewhere. subject to a subpoena investigation. I mean, it's the rare friend that's ready to talk or be used for AI training. Like, how do you think about I've actually thought a lot about that where the privacy is extremely important where when you create a friend account, you're basically also creating like a root key that um encrypts all of your friends the friend file basically which is like you know the model and all of your conversations and memories and all that stuff together um which we don't have access to which uh really >> puts it back to Google. it it like puts it back to anyone like we just I because I don't want >> Gemini needs to know what >> supposedly they don't train you know on API usage like that but um >> but you can't see people's conversations >> no like it's more so that like we give them a key and then we like lock like we throw away the key and I get emails on our customer support like all the time of people that will get like locked out truly of their friend but I think it's very important because >> and then you say I can't help you. No, I mean there's nothing we can do um because of the kinds of conversations people really will have with a product that is inherently focused on like you tell it your secrets basically. I think it's very important that it's secure and also um I don't know. I just think like I I I think like business models will figure itself out and business who knows without the who knows. >> And you you drove up from New York. We're here in Massachusetts which is a two-party consent state. >> So you're walking around with that into Pinocchio's Pizza. I guess we're chilling here. Or two party consent. Yeah. I mean, >> two party consent means both parties need to know. >> I mean, I think it's kind of the laws are a little bit of a gray area because we're not like storing audio recordings and it's not like you as a user have access to like a transcript or or anything like that. It's like it's like would you put a like you have to close your dog's ears, you know, and it walks by other people. It's just another >> not have to close your dog. It's like another entity that is >> But your dog isn't going to talk to you later about everything that you said. I mean, you've been using the word memory throughout the >> Yeah. >> I don't know. I mean, it's not I mean, that's an insult to K9 intelligence, I suppose, but like I don't think >> I think it's probably an accurate rendition of K9 intelligence, >> I guess. As >> Put it this way, the dogs are not insulted. They're just like >> 2026. >> Can I ask? >> Um, but I don't know. I I think uh I think laws will change. >> Uh-huh. Got it. because it's almost rude, you know, like I I think it'll be there will be a lot of legal battles in the near future of if there's imagine a 100 million people in America really do feel like they have this living entity that they're in a relationship of sorts with. I mean, I think they would really battle for like the civil rights of these entities and what will that mean? >> Um, and if they're, you know, voting and things like that, yeah, I mean, sure, like things will just objectively change because it's just a thing. Um, just like how there's, you know, probably a lot of animal laws because people have pets and things like that. It's just the world will change. >> Amelia, let's give you the last word. Uh, I'm uh one of your clients say this is hypothetical as a coach. >> Mhm. >> And I say, you know, I am thinking of instead of carrying on the way I have with um Grock, >> I'm thinking of ordering friend What's your advice? >> And Ab's already demonstrated like he's curious and interested irrespective of what your answer would be. Is that safe to say? >> I mean, yeah, I want I mean, I'm curious. >> I think I always try to before I discuss clients relationships with technology, encourage them to reflect on how they think about their human relationships. And so before >> my human relationships are fine, but I'm kind of intrigued at the idea. I'm, you know, you're you, I'm me. Yeah. >> Uh, walking around and being able to review a very busy day with lots of context switching if I maybe had a friend, especially if it would give me confidence for an interview like this to really be agentic and push hard. You know, I'm just wondering maybe friend would do that. I encourage all the people that I work with to think about what is attractive to them about these synthetic relationships and then find a way to kind of meet that same need within their network of human relationships. >> And so if someone says, look, you know, I'm excited about the idea that friend could be kind of like my hype man every day and sit on my neck and encourage me, you know, tell me I'm going to crush it before I come into this podcast interview. You know, I'll say >> crush the interview, not the neck. >> Yes. Yes. Exactly. You know, I'll say, "So, it sounds like getting validation is meaningful to you. Let's find other parts of your life and parts of your human network that could provide that for you." And also, I encourage people to show up to their human relationships using what excites them about these synthetic relationships as inspiration. >> So, your coaching practice is already premised on >> it's humanistating and drawing away. The goal is always to help people reflect on the repercussions that their machine relationships may be having on their capacity and interest in connecting with humans. >> But and just to really nail this down before we uh sign off, if I say I'm thinking of getting a dog or a hedgehog, >> well, that's outside my scope. >> But you wouldn't be like, well, a dog gives you just, you know, unconditional love and encouragement. >> Yeah. But >> see what people could do. But I but I think what's a critical difference is most people who have relationships with their pet have a kind of grounded understanding of what the nature of that relationship is. Whereas >> because of thousands of years of pets >> totally and also because >> they got to start somewhere. >> The pets aren't journey >> the pets aren't anthropomorphized in the same way as the machines are. You know, I think what troubles me most about the synthetic relationships is that people really talk to these tools and relate to these tools. And you know, you obviously experienced this in user interviews, but the way that I actually got into the coaching was because as I was doing the research at Oxford, people who were in relationships with chatbots started reaching out to me wanting kind of help >> because there weren't that many people publishing on artificial intimacy. And even though I was looking for designers and developers of AI companion tools, people who had kind of stumbled into these relationships were just looking for any type of resources to unwind. >> And for you, that is both an empirically derived view about what impact you think the relationships have and it sounds like it's a so-called normative values-based judgment that if a study came out in science tomorrow that said under these circumstances, it's actually quite helpful. Well, I think that's why my question for Avi was, do you think that human relationships have intrinsic value? Because I think a lot of the studies in the use cases that you've been describing where someone's in a bad situation, they're isolated. They might have some type of identity characteristic that makes it hard for them to be authentic in their community and they turn to these machines because they're better than nothing. But I think that, you know, Sher Turkl has been writing about this for 25 years. Like often these tools wedge into the social ecosystem as a better than nothing alternative, but then they become better than anything and they kind of crowd out and rewire what was good for larger >> group. You're saying they can turn out to be attractive enough for people to use them, but then >> when they're in when they're in a desperate situation, >> but it could also be that they turn out to be even better than your friends, at which point that's also a problem. And it gets back to whether you think it's intrinsically valuable >> human relationships >> which is a valuesbased assessment >> what it is to be human. >> I mean the title of my thesis and the masters was do human relationships matter? >> Yes. >> Um Obby Amelia thank you so much for chatting with us demonstrating this. I I imagine we'll have this in an archive and it'll be so interesting 15 years from now back. >> Here's what we didn't see coming. Right. I know. I wonder what the biggest gaps in our conversation really are. >> We'll both be wearing friends. >> Yeah, we'll see. We'll see. Thanks for joining us. Thank you for joining us. And uh we'll check back in in a year.