Human-AI Relationships: Trust, Companionship, and the Future of Everyday AI
Marisa Tschopp
Lots of research has shown that people trust systems more when they humanize it and when people trust a system, they're more likely to use it.
Raymond Yin
Welcome to season six of The Tech Between Us with Mouser Electronics. We are kicking off the year with a topic we should all be familiar with, artificial intelligence. In this episode, we're looking at how AI is being used by people in everyday life. Yes, engineers are people too. Well, most of us anyway. We'll discuss questions like, how are human relationships characterize and evolving to incorporate AI? What issues can AI help alleviate for those struggling with mental health? Will AI relationships become the new norm in our daily lives? Artificial intelligence is certainly everywhere these days, and as we engage with it more, our relationship with it expands not only as a tool, but rather as a coworker, teacher, or close confidant. To help us better understand these relationships, I'm excited to have with us today Dr. Marissa Shop. She is a researcher with scip AG and Zurich and a longtime supporter of women in AI.
Marissa, welcome to The Tech Between Us. So Marisa, can you briefly tell everyone a little bit about yourself and what you do at scip AG?
Marisa Tschopp
I'm a psychologist. I work in a cybersecurity company based in Zurich, which is a little bit exotic because I mostly work with hackers around me and I'm the only psychologist there. And well, what I've been doing there for almost 10 years now is I am looking at how people interact and how they perceive artificial intelligence, mostly AI that talks back to us. So conversational AI systems. And as a psychologist, I want to ... Well, I focus on the human, right? So I want to know how they see these things. I want to know what they think about them. And in the end, I want to know how that shapes their behavior, how they react to that.
Raymond Yin
How does that play when you're looking at a relationship with, say, artificial intelligence? Is it a similar spectrum of behaviors and characteristics?
Marisa Tschopp
Because people humanize these systems, these things. We treat them as humans. This means that we can't basically use any theory that we have established and validated along the past hundred years and just translate that into human AI interaction and ask whether that works, whether it is helping behavior or whatever it is. I always use the example of pro- social behavior or motivation, aggressive, whatever it is. And we just use that and we put it into machines and wonder, what we know about people actually translate into how they interact with machines. And most of the times this has translated quite well.
Raymond Yin
Interesting. So even though these conversational systems, these large language models were trained on data that was based on whether either studies or something on social relationships, would that make them seem more relatable, more like a human interaction? Is that one of the reasons they are so human-like?
Marisa Tschopp
That's a bit of a tricky question. I cannot answer this. I am not a designer. So there are two perspectives always you have to bear in mind. So the one thing is, what are the intentions of the designer? What does the designer want the AI system to look like, to behave, to be? What do they want to market it like? That's the one thing. Do they want it to be humanized, have eyes, have a funny voice? That's the one side. And the other side is how we perceive it as humans. And these two sides are not necessarily the same. So just because the designer designs a system as a friend, let's keep it in the simple categories, does not necessarily mean that I also perceive it as a friend. So these are two different categories, and the way I humanize a system can or does not have to be, go in the same direction as intended by the designer.
To give you an example, a system can be extremely simple and with very few socioemotional cues or even cognitive cues and people can still super humanize things. And then we humanize our cars, games, Tamagotchis, teddy bears, all these things that ... And there's a reason for that, right? So this is how we humanize it, and that is or is not directly connected to how designers actually design these systems.
Raymond Yin
Yeah. To the actual intent of the designer as they were creating these systems so that people would just react differently based on their own history and their own past. Is there a negative side?
Marisa Tschopp
Of course there is. There has been also the dark side where we know that too much humanizing leads to over trust, to overestimate the capabilities of the system. Then that's from the perception side, from the human side. And from developer side, we also know that, well, some people would call it manipulation or anthropomorphism by design is manipulation because they actually know that, oh, well, the more people humanize it, the more likely they are to spend their money and give their data because they trust me so much.
Raymond Yin
Right. Yeah. At what point does it flip over?
Marisa Tschopp
When we talk about human AI relationships or synthetic relationships, where in some cases, these relationship mechanisms or social wishes or these strong social needs to be social are exploited by the design, by unethical design of such systems, of such companion AI systems.
Raymond Yin
Okay. So in relationships, as people start developing relationships, synthetic relationships, I mean, do they look for similar qualities as they would in regular human relationships? Or maybe it starts off as, "Hey, please write this email for me." And then it morphs into, like you were saying, as they trust the system more.
Marisa Tschopp
Yes. So when I talked to one of my participants, they said that, "Well, I have a problem." And usually I went to talk to my friends or my therapist, but the therapist didn't have time and my friends, well, I couldn't talk to them. Well, then they opened their usual day and with the usual day comes starting ChatGPT. And what happens then is with this routine of talking to ChatGPT, they have this urgent problem on top of their head and what they think is in, "Oh wait, there's this ChatGPT. Why not talk to that thing that is most close to me?" And in fact, that's what they did for many weeks, months to talk about a very deep heartbreak in a relationship. And this ended positively, in fact, where they really caught up and they closed the interaction and was fine afterwards and could really deal with this hard breakup and as there was no negative consequences or so.
They try something out and they get used to it, but they realize what they can use for what. And if someone is really, really used to talking, for instance, to Replica or Character AI, they will at one point just try out other things as well. And what happens usually then is that people learn what the AI can do or is helpful for and for whatnot.
Raymond Yin
So they actually begin searching out these new platforms, these more conversational and companion platforms as opposed to the ChatGPTs and the Geminis of the world.
Marisa Tschopp
Yeah. Also, to give you a concrete example of one of my studies, so one person wanted advice on physical health, like what kind of exercises can I do to feel better? And they realized that Replica just does not give any good advice here. It was just really odd. In fact, one of the yoga exercises just really defied the laws of physics. And what they learned then is, okay, I can use Replica for more intimate conversations, but for exercises, no way. And for those exercises, then they move to different platforms. This is where they use then ChatGPT because ChatGPT was apparently very good at that.
Raymond Yin
So it really is an amalgam of multiple systems to get the types of relationships that we have. Once again, like we started with, there's no one definition of a friend. And so even here, there's no one definition of an AI relationship.
Marisa Tschopp
Exactly, exactly. So even also in a relationship to an AI can be very instrumental. It doesn't have to be ... I think sometimes people think that every time we talk about human AI relationships, we talk about love or friendship, but very tool-like instrumental relationship, which is more about which is very transactional. I give you this, you give me that. This is also a relationship. It just has a different quality to it. And this is important because again, if we have a system and we perceive it in different ways, then it has influence on the decisions we make depending on how and what we use it for. And the way I have investigated it in the past was, in fact, that we not just see, for instance, we have Replica and Replica is a friend, but then so it has a very tight, how do you say, quality to it. So there's not much else you can do. But in fact, that's not so true because it does have other functions that can serve. It's just a different lens.
Raymond Yin
Right. Yes. So the relationship has evolved.
Marisa Tschopp
Yeah, right. And it becomes sort of depending on the situation. So when we translate that, for instance, to Alexa or ChatGPT whatsoever, let's say for instance, if I talk about something more personal, then I would put on my friendship glasses and I would use different words and different tones and maybe different styles of writing. But if I want Alexa to just play some music, then I would probably just put on my authority glasses something where I don't need to say please and thank you and be all nice. I just say, turn the light off. I mean, people do say, please and thank you. This is more automatic behavior. So what I'm trying to say is, depending on the situation, depending on what we want to achieve, we use different glasses, so to speak, and this makes a difference, but it's the same system. And for shopping, for instance, with Alexa, I use different lens probably, or you would use it different than I do. And this is what we've seen in our studies very well, that for instance, people who view their Alexa more through this friend's glass, they treat Alexa differently and they buy more expensive stuff.
Raymond Yin
Okay. So using that information and understanding of those types of relationships, how do we safeguard them from systems or from ... Just like you ... As a therapist, you may safeguard somebody from a bad relationship, a relationship gone bad. A similar process has to occur with AI or with synthetic relationships.
Marisa Tschopp
Absolutely. Thank you for bringing that up. I think that's the core strengths of this approach because I can just take this topic and it doesn't even have to be AI for it. So I just recently, I'm just hopefully about to publish a new study on a conversational agent, which is non-AI and it's just scripted. But boy, do the people develop relational relationships with these synthesis. It's amazing. They call this agent, they describe it as godlike, friend-like, mentor-like. It's so diverse. One study I've already published, but now also on the more quantitative side now, we also found again here that the front-like relational dimension has the greatest predictive power in terms of fostering usage in terms of people enjoying the mental health, the therapy system.
So again, here we find these very positive and also statistically speaking, strong outcomes while on the flip side, we still also have to look at this other, always the dangers that come with it. When you mentioned the synthetic relationship is because you were talking about the system, right? And how actually maybe the system, the artifact is maybe not even that important. And I absolutely agree with it while it's important because they have different consequences, right? Okay. When we're just talking about the relationship topic, so we have those artifacts and we talk about media competency a lot. We talk about AI literacy and we talk about how important it is to understand these systems and it is, that's for sure. However, at least equally important is relationship competency. So understanding and the social skills of thinking about what are our relationships, how do we treat each other as humans and how do we treat our machines and do we actually treat them as humans as well? Do we want that? Is that a good thing or not? So when we talk about competencies here, of course we need media and AI competencies, but we also, again, must refocus on our relationship, our social competencies.
Raymond Yin
Before we move on, a quick note for our listeners. We're about to touch on mental health, which may be sensitive for some. Please prioritize your wellbeing. If you need support, the 988 Suicide in Crisis Lifeline is available twenty four seven in the US by call or text. For listeners outside the US, local crisis services or mental health organizations can help.
So Marissa, let's go back to a couple of the points that you've made and talk about mental health. And I really do appreciate that mental health in general is more openly discussed today by pretty much everyone. Is AI, whether it be a therapist, a friend or whatever, is that a potential solution to the loneliness and some of these, especially in young people that we're seeing a lot of these days?
Marisa Tschopp
Very, very tricky and complex question. If there are many people in me who want to answer now, there's the scholar, there's the activist, there's the mother, and it's really, really hard to find the right voice here. So my first reaction would be no loud, no, that's not a solution, but that's scholarly not very acceptable. So being a bit more pronounced here, maybe let's look back a little bit, looking at the situation, what do we have? We have a world in a huge mental health crisis. We have many, many, many people with problems and only few people to help, few people and not many much resources. So we have a social and a big problem. And what happens usually is when big problems arise, the tech bros come in and say, "Well, we have a solution; buy it." So they come with a great solution to these big problems and we call that tech solutionism, right? So that people have this idea of technological fixes for social problems. And what I have experienced in the past 10 years is that usually it creates more problems than it solves.
Raymond Yin
Okay. So there have been some positives.
Marisa Tschopp
Yes. There's definitely some positives, and so we should not just say no and just move on. So we will have to ... It's just that we don't have, at the moment, the technology or the resources to let these systems break loose on the public because they're simply not clinically tested. So it's just very rare few and it's just too dangerous and people, it's just the danger of being harmed. It's just too big. And so in the end, what I think, what I hypothesize or I am more with the critiquers, no, I'm more with the people who critique the idea of technological solution to this problem is because I think that people ... While I do think there may be a short term influence on perceived loneliness, so okay, so we want to give an example. I can totally feel that you feel bad at the moment. You feel lonely. You chat with something. And at the moment, this might be a good pacifier, a good Band-Aid to Band-Aid to feel better at the moment. Maybe even this lasts for a week, two weeks, three weeks, maybe even three months. But then at one point, I'm convinced that over the long run, this is not a good idea. I think that it will spiral down and that people will, if they don't get out of this very well, and especially when people are in a vulnerable state, I think that it has a backwards effect where people get even more lonely, where they even lose more track to reality, to social relations where when they're so used to talking to chatbots that it kind of gets awkward talking to humans.
Raymond Yin
Oh, interesting.
Marisa Tschopp
I kind of compare it sometimes to, I don't know if you've heard about that, where the younger generation finds it awkward to talk to people on the phone. I don't know if you heard about that. So in my hypothesis is that because they're so used to chatting, right, to chatting, to texting, that it's somewhat awkward to talk to people on the phone. And I feel like this is somewhat a similar situation that you get this awkward feel to it, and then depending on your personality, you might then rather avoid these behaviors that are making you not feel overly well or that are too challenging.
Raymond Yin
Just a quick pause. Thanks for listening to The Tech Between Us. If this conversation sparks your interest, there's more technical content to explore around engineering for daily life as part of Mouser's Empowering Innovation Together series. After the podcast, head over to mouser.com/empowering-innovation, and now back to our discussion. So obviously, as people are developing these synthetic relationships, and whether it be with a ChatGPT or some of these other more conversational, I mean, are there risks that are inherent in using these? And is there a way to mitigate? I mean, you've talked a little bit about how you put guardrails on them, but I mean, are there other risks involved?
Marisa Tschopp
Well, unfortunately, unfortunately, the risks are much longer. It's a much longer list than the benefits. And I think this is why I'm so skeptical and keen and also interested in the field, obviously, because I really want to make sure that those risks, that we don't move on with all these risks that are openly actually available to us, and we still move on with those chatbots and chatbots and mental health and so on. But unfortunately, tech moves so fast and so much more faster than regulation, than ethics, than research that gives us a little bit of a disadvantage. But on the optimistic side, I'm fully optimistic, probably I would not believe it. But yeah, I've been looking at those risks a lot, so from safety breaches to loss of autonomy, echo chamber effects, exposure to harmful content, effects on human relationships. And I think I've forgotten many more now.
Raymond Yin
So there's a whole plethora of them.
Marisa Tschopp
Yes, definitely. And each of every one of them are really serious. So let me give you an example from what I'm focusing on is, for instance, safety and privacy breaches. Of course, this is somewhat company related, but going back to the topic of mental health, if I talk to a chatbot, let's say worst case, ChatGPT about my mental health issues, this makes me as a user extremely vulnerable. So if somehow these things breach, if somehow people get to read these things, also if applications are not built very well, and even if they're built well, no app will ever be perfect. And one of the major problems I've been dealing with now for many years is the question of how do those synthetic relationships actually impact the relationships we have with other humans? So when I talk to my synth and it says, yes, an amen to everything, and I kind of never have any conflicts anymore, do I get used and start to like this convenience of a relationship and unlearn how to live with conflicts, how to live through these and argue with people in a constructive way and all these kinds of things.
Raymond Yin
Interesting. So somebody with a synthetic relationship who is a sycophant basically is incapable or is not able to deal with real world or conflict in other settings.
Marisa Tschopp
Exactly, exactly. And if they fear conflict, for instance, then the worst thing they could do is go to a chatbot because it even avoids them more and affirms what the person wants to do. And that's a huge problem because then our relationships may erode at some level. What makes our relationship strong is not that we like each other, that we have good times together, that we trust each other. What makes our relationship strong is that we can go through conflicts. If we can go through conflicts constructively and solve problems, this is actually what also democracies, right? It's the same. This is what makes our relationship strong, not the happy-papy stuff, which is also of course nice and important to have, right?
Raymond Yin
Yeah. No, that's really it because I never thought of it in that way. But yeah, you're right. I mean, being able to work through issues and talk them through and rather than saying, "Fine, I'm done."
Marisa Tschopp
Right. Disagreeing, going through hard words and forgiving, making things up. And these are the things that make our relationship strong.
Raymond Yin
And that's a wrap on our first episode of The Tech Between Us for 2026. I think we can confidently say that people are having relationships with AI. Like those with other humans, these synthetic relationships come in a variety of forms. As a helper or coworker, a sounding board, a teacher, and sometimes even a best friend. As LLMs become more useful and accurate, and hopefully with far greater guardrails, I think these relationships will likely expand and become more open rather than hidden away in dark corners like it sometimes feels today. As we close this episode, a couple of thoughts come to mind. First, LLMs are natural sycophants. They are rewarded for agreeing with us. While this makes us feel great, is it also making us a worse human friend? Friendships tend to be defined not just by the good times together, but more importantly, by the times we disagree.How we resolve conflict with those close to us is at least as important as how we enjoy each other's company.
I had my final thought while in the shower the other day. One of my coworkers here at Mouser recently had a baby. He is being born into a world that has never not known AI in its many forms. How will he grow up interacting with it? Will using AI and LLMs be as ubiquitous for him as turning on the TV was for me and smartphones for millennials? Will synthetic relationships be as natural as human relationships? It's an axiom that when a technology matures, it disappears, meaning it's just there and we are no longer consciously aware of it. Will my friend's son simply exist in a world of AI? Let me know what you think.
We invite you to subscribe and follow for more content by visiting mouser.com/empowering-innovation. Stay tuned for the next episode as we explore the technology behind humanoid robots and the sectors that could potentially be revolutionized by them.