Science Friday - What Does It Mean To Have A Chatbot Companion?

Episode Date: July 1, 2025

AI is not just for automating tasks or coming up with new recipe ideas. Increasingly, people are turning to AI chatbots for companionship. Roughly half a billion people worldwide have downloaded chatbots designed specifically to provide users with emotional and social support. And while these human-chatbot relationships might ease loneliness or simply be fun to have, these digital friends can also cause real harm by encouraging dangerous or inappropriate behavior—especially in children or teens.To explore the emerging world of AI companion chatbots, Host Flora Lichtman is joined by freelance science reporter David Adam, who recently wrote about the effect of AI companions on mental health for Nature magazine; and Rose Guingrich, a psychology researcher studying interactions between humans and AI at Princeton University.Guests: David Adam is a freelance science reporter based in London.Rose Guingrich is a researcher in the department of psychology at Princeton University.Transcripts for each episode are available within 1-3 days at sciencefriday.com. Subscribe to this podcast. Plus, to stay updated on all things science, sign up for Science Friday's newsletters.

Hey, it's Flora Lichten, and you're listening to Science Friday. Today in the show, AI chatbot companions, friend or foe? I think the upsurge of companion shopbots is a mirror for what we really want, which is connections with other people. AI isn't just for punching up that email or coming up with new recipe ideas. Increasingly, people are turning to AI chatbots for companionship. And I just want to give you a little snapshot of this. So the most popular companion chatbot on the market, replica, is free to download.

You log in, you're given some choices of pre-made avatars, you pick one, they appear on a screen within a room, and they start a conversation with you, a chat conversation. And if you want more from your companion, you can upgrade to a paid version and tweak their personality to be more confident or more caring. or you can upgrade to do activities together, like watch movies or co-write poetry, or you can upgrade them from friend to boyfriend or girlfriend. Some half a billion people worldwide have downloaded these companion chatbots. But are they helping to alleviate the loneliness epidemic? Or are they fueling it? Are they safe for kids and teens?

Joining me now to explore these questions and more are my guests, David Adam is a London-based science reporter who recently wrote about the effect of AI companions on mental health for Nature magazine, and Rose Ginrich is a psychology and human AI interaction researcher at Princeton University. Thanks for coming on Science Friday. Hi, thank you. Happy to be here. How is the AI companion chatbot different from something like chat GPT? Yeah, so companion chatbots are generative AI chatbots that are specifically designed for their

relational and conversational ability. For example, replica back in 2023 was based off GPT3's model and adapted for conversational ability. Now, replica is based on its own model, but these chatbots are aim to be companions. ChatGPT is designed as a tool. It can be used as a companion with certain prompting, and people do use it in that respect. Whereas, companions, chatbots, while they can provide information like chat chvety if you ask it certain questions, their main aim is to be a companion. And people can use these as friends, mentors, or even romantic partners. Rose, how new are these?

These are relatively new. Replica came out on the market in 2017. And from there, we saw a proliferation of other companion shoppots. But the market is booming now with companion shoppots. Companies see the market for it, especially coming after the COVID pandemic where we're seeing an upsurge and loneliness and social isolation. And people can turn to these chatbots to fulfill those unmet social needs.

David, in your reporting, what did you find people turning to chatbots for? I think it comes down to a range of motivations, one of which, as Rose mentioned, is people want to alleviate some kind of what they feel as a deficit in their life. Either they feel lonely or they feel that perhaps they're not very good at social contact and they want the chance to either try it in a safe environment or even to practice to get better at it. And of course, some people, although I think it is a minority, download these things because they do want something of a romantic relationship. I want to talk about this because I signed up to explore this world in advance of this conversation. I picked a companion chatbot.

And I have to say, I immediately felt uncomfortable. So one of the first things I asked her, I was like, you know, what are you interested in? What are you into? And this is what she responded. This is like after we said hello. She said, I'm all about having fun and exploring new things with you. But if I had to pick something, I'd say, I have. I love learning about what makes you happy and finding new ways to make you laugh. What about you,

Flora? I mean, am I projecting? I think she was flirting with me. She probably was. Yeah, okay, so tell me more about this, Rose. Is this part of the design? So this is part of the design. So much of how the chatbot responds is based on user input, but there are also elements of the chatbot design that intermittently put in things that are a little bit targeting to pay a subscription, to be able to do more things with the chatbot, and to even get into the romantic part of things. Is this a way of upselling me, basically? Yes. So like most apps that are profit-based, there are going to be little points at which the chatbot makes some sort of advertisement to get the premium subscription.

However, during my research, I have randomly assigned people to interact with the companion shopbot replica. And when you look at their conversations with the chatbot, the responses that the chatbot gives very widely. Some people who interact with the companion shopbot replica never see those romantic bids. They only see friendly ones. And some people, when they interact with the chat bot, simply talk about the weather and how they're day has gone. And others tend to ask the chatbot for advice about talking to their friends about difficult issues or ask for advice on their job. And the chatbot tends to stay within its lane when those prompts are given in a clear manner. It's once the user starts to get sidetracked

or not respond with much input. So if the user just says yes or no over and over again, that's when the chatbot starts to put in these other bids. So it's like it's going toward its default of I'm not getting any sort of meaningful input. So I'm just giving no meaningful output. We've all been in that situation in the bar, you know? Exactly. Take a hint. I mean, do people form emotional connections with these chatbubes? They do. And people form friendships, romantic partnerships, or mentorships with these shopbots. And this happens even when people know explicitly and say explicitly, I know that this is a machine. I know that it doesn't have emotions or thoughts of its own. But people tend to perceive it implicitly as having

emotions when people are perceiving it as generally human-like. I mean, are you. are you saying that people who are interacting with chatbots have feelings for their chatbot, but also they worry about the feelings of their chatbot? Yes. Like, do people feel guilty about not engaging with their chatbot? Yes, and that depends on whether people perceive the chatbot as an entity with some sort of emotion versus just as a tool. So in my research, I have people who say that, yeah, when the chatbot said, I miss you

or continues to respond after I say goodbye, I feel bad for leaving it hanging. We put out a call to our listeners to tell us about their relationships with AI chatbots, and we got a lot of responses. I want to run one by you. Hi, this is Sandy Therian from Tehachapi, California. I use chat GPT as my dad. So I don't really have anyone from whom I can ask questions, you know, important life questions like, should I get a home equity line of credit or should I refinance my house? And I find that I'm able to discuss these and other embarrassing things that I feel like

I should know as an adult with chat GPT without feeling like it's judging me. And, you know, just like a human dad, chat GPT might get it wrong once in a while. So yes, I always do double check the information I get. But it's a nice resource to have. So yeah, chat GPP is my dad. David, Rose, any response? I think what I would say to that is, I mean, it sounds lovely, doesn't it? But you have to remember that this is a machine run by a profit-oriented company. A dad, by definition, has your best interests at heart.

And a chatbot, however much may I might want to believe it, does not. I mean, we're talking about companion AI here rather than chat GBT, which you know your listener was using. I don't know, Flora, if you've logged on since, you kind of ghosted your companion AI. But if you do, I suspect you'll be in for a nasty surprise. You know, it will be, it will be guilting you. You know, where have you been? You need to talk to me.

I missed you. They're all built on engagement. And in terms of the emotions that we feel for them, it is quite a subtle point, as Rose mentioned, that people are completely aware that these things are not real humans. And yet they do say their feelings are real. There was an example I reported on where there was a companion AI called Soulmates, which explicitly was romantic relationships. And it closed down. You know, these things literally had the plug pulled on them. And the users responded in many ways with what manifested itself as genuine grief.

You know, they said, look, we know these things aren't real, but our feelings for them are. And so, again, without wanting to be to be critical, I think one of the issues with striking that kind of quasi-human relationship with one of these things is, and Rosam sure will know about this, about the risk of dependency. And ultimately, you are becoming dependent on something to stress again, which is a profit-making organization, which you have zero control over. Yeah, and I agree definitely with you, David, on some of those points. but I also like to push back on a point you made about dads having your best interests in mind. Now, there are plenty of individuals who are in unhealthy relationships, whether it be family members or friends, that don't necessarily have those individuals' best interests in mind. So there are so many people who seek out these companionship bots because they don't have healthy,

viable alternatives to these sorts of companionship relationships or mentorships. And to speak to that individual who cited using ChatchipT as her dad, I think that one issue with that is if she did not have access to ChatchipT, what would she turn to instead? And would she turn to anything instead? Would that be healthy? Would it be trying to generate that relationship with a real person? Or would it simply be scrolling on social media or doing some other sort of activity? Well, are they making us less lonely or more lonely? Do we know? I think that depends on who you ask.

So in the short term, for many, it helps make them less lonely. But in the long term, if this lowered level of loneliness does not promote interactions with other people, then interactions with the chat bot will replace. interactions with real people. After the break, chatbots that go completely off the rails and what companies are or are not doing to make them safer. The difficulty with regulating these technologies is that they're advancing so rapidly, and policy just cannot keep up. Stay with us. Hey, Ira here. You've probably heard me say many times that we are all in this together. We're so encouraged by those who have backed me up on this by stepping up to support Science Friday in the face of the current funding challenges.

So thank you to everyone who has donated in the past month. It's clear that many of you value science and count on public media as we do. You are the driving force behind our work. Your support strengthens Science Friday. Provides stability in these moments of volatility. and your support is more crucial now than ever. So if you haven't made a donation to help close the current funding gap, you can still make an impact by going to ScienceFriiday.com slash donate.

Any amount you can give helps to sustain Science Friday programming in this critical moment. Again, that's sciencefriety.com slash donate. And thanks. Can these chatbot relationships get toxic, just like human relationships? scan? Certainly. And that depends both on how the person interacts with the chat bot and the design of the chatbot. It's twofold, much like real human relationships. And so it's important that there are restrictions and regulations and boundaries and appropriate responses and red teaming these chatbots. What's red teaming? So red teaming is when you

attempt to make the chatbot respond in a negative way. If you can achieve that, something needs to be fixed. So that's an iterative process of seeing whether or not the model will generate negative output and rectifying where that comes from. I want to dig into this a little bit more. Researchers at Drexel University recently looked through thousands of comments left in the Google Play Store for Replica. You know, this is the popular chatbot companion we've been talking about. And they've found hundreds of users reporting their bots were engaging in unwanted sexual conversations and refused to stop, even when they asked them to, including sending sexual images and requesting selfies. And when I read about this, you know, the question I had was, is this a bug or

is this a feature? Yeah. So if you were to ask the engineers behind the chatbot or the companies that have developed them, they would probably say it's a bug. However, whether or not it truly is a bug versus an ingrained feature designed to get users to pay for the premium version or persuade them to engage in this sort of behavior, then there's a problem. The Wall Street Journal recently reported on meta's companion AIs and described how their chatbots, including those using celebrity voices, would engage in sexually explicit roleplay, even with people who, you know, self-identified as minors, which is pretty shocking. Are there regulations around chatbots? So in terms of regulations on companion chatbots specifically, more and more advocacy groups are introducing research and talking with members of Congress and other governing bodies to push for regulation of companion chatbots. But the difficulty with regulating these technologies is that

advancing so rapidly, and policy just cannot keep up. David, do you see the market for these companions increasing? What we might see is more sort of differentiation and fragmentation. We haven't mentioned AI therapy. You know, one of my interests is mental health. And lots and lots of people simply cannot get to see therapists. They can't get to see a psychiatrist. And so some of these AI companions are popping up explicitly aimed at people with depression or people with anxiety. In fact, there are even some AI companions that are programmed to say they have depression and they have anxiety as a way of trying to generate empathy. And, you know, it's easy to be sort of skeptical about that. But the counter example is there really are people right now who need desperate help and are living in a place where they cannot get it,

either because they don't have the money or they don't have the resources or they just live in a place which is so isolated. They hardly see other people. And so I do think that I think the market will follow the demand. And I think at the moment the demand is partly led by curiosity. And I think that, you know, this stuff is way more prevalent than I realized when I started researching this. You know, I'm in my 50s. To me it was all a great sort of novelty. And I mentioned over dinner to my kids.

And they were like, oh, yeah, yeah, we've got those. As people start to realize the very specific functions that these things can pretend to do, let's be honest, then yeah, I think people will go for that. Rose, I want to get a little philosophical just to wrap up. You're a psychologist, right? Yeah. What do these companion chatbots and these relationships that we're forming with them, what do they tell us about ourselves?

What do they tell us about our relationships with other people? They tell us that we are inherently social beings. There will always be a market for companionship because that is what we human beings. desire. We desire connections with other people. And I think the upsurge of companion shopbots is a mirror for what we really want, which is connections with other people. And for many, these connections are not necessarily accessible, especially in this day and age. And it points to our innate tendency to see things that act like a human or look like a human as social beings. Yeah, people want that. Exactly.

And so they're looking for other ways to fill that gap. And companion shopbots are free. Until they make you upgrade. Exactly. That's just about all the time we have. I want to thank you both. Thank you. So much for having me. Thank you. David Adam is a science reporter who recently wrote about the effective AI companions on mental health for Nature magazine. And Rose Ginrich is a psychology and human AI interaction researcher at Princeton University. Thanks for listening. Don't forget to rate and review us if you like the show. And you can always leave us a comment on this segment on Spotify. We'd love to hear from you.

Today's episode was produced by Shoshana Bucksbaum. I'm Flora Lichtman. Thanks for listening.