[00:00:00] Digital Dominoes. Welcome to another episode. I'm Angeline Corvallia, and I'm really excited today to speak to Sonia Tiwari, who's an AI consultant for children's media and parasocial learning researcher. The reason I'm so excited about this is that I very recently learned the term parasocial AI and I'm gonna let Sonja explain it, but [00:00:30] this is really a term that basically everyone needs to understand because it has such an impact on AI chatbots and even AI toys, you know everything and It has a big impact on people.

So I'm really looking forward to finding out more from an expert on this So Sonia, can you tell us more about yourself? Thank you for coming and being here. I'm so excited. So tell us more about yourself and then you can maybe jump right into what [00:01:00] even parasocial is because I think a lot of people don't know that.

Yeah, so I started my career in the gaming industry as a character designer. So my first introduction to the term parasocial was more in the context of Characters in general, not just AI characters, but how we form an emotional one sided connection with characters. For example, a child reading a Harry Potter book and you start thinking as if, oh, these [00:01:30] are my friends.

Harry is my friend. And so when something bad happens to Harry, you feel emotionally invested in the story that, Oh, how is he going to recover from it? Or, you know, when Dumbledore dies and you, you feel as if your own teacher, a loved mentor passed away. And so that kind of emotional connection, my first introduction was in a good way that you, you feel emotionally attached to characters and it's one sided.

The other [00:02:00] character, obviously. doesn't even know you exist. I think actually that's really a good point because I also, when I first read about it, you know, it's parasocial and there's one sided connections and there's also the celebrities that people are attached to celebrities. And the first thought is, is bad, but then from what I've read, what you're writing on LinkedIn, for example, that's not true.

It's, it's not bad in and of itself. It's just, it can be. Anything can be weaponized, [00:02:30] right? Like social media, there are environmentalists who are using it for climate education. Social media, there are like really toxic cosmetic companies trying to tell teenagers to change everything about themselves. So like you can leverage any media for any purpose.

And so, yeah, my initial introduction was in a positive sense. And before that I was in film school, studied comedy writing. And so shows like 30 Rock, I always thought like Liz Lemon is like literally my [00:03:00] sister. Like I, uh, emotionally connect with her and I feel like a part of that world. And also because the show was based behind the scenes of a TV show, and I was a film student at the time, so it felt like I was part of that world.

And so that's the foundation of parasocial relationships, that there's some kind of shared experience, whether it could be grief, happiness, just the location, the characters who might be your [00:03:30] age, or you find some common ground for an emotional connection, and then feel attached to a fictional character who cannot talk back.

So parasocial research, it actually started in like 1950s, Horton and Wall, were the original researchers, and they were only studying media consumption in terms of film and television. And then, you know, later on, Sandra Calvert is one of the more [00:04:00] prominent researchers who started differentiating between parasocial relationships and parasocial interactions.

And interactions was more in terms of AI or like a fictional conversational agent. Okay. And the common thing is that in both cases, it is a one-sided relationship, but in a parasocial interaction, there is like a simulated second side as well. But in essence, it's still one-sided, right? [00:04:30] Because you're feeding the information, and like an echo chamber, it's relaying what you want to hear.

And that's where the trap is, right? Because it's kind of like, if as an adult person, you're talking to the man of your dreams, trained on what it is that you desire the most, you're very likely going to fall for that person, despite the knowledge that, well, I created it. And there's even like a, a [00:05:00] joke in, in 30 Rock that there's like an infomercial where there's this guy who is saying that, Hey, would you like to talk to a guy who says things like, I hear you?

It's their loss. You're so amazing. And so it's just like a guy on television, making some pauses and saying positive things. But in essence, chatbots are doing a better job of just that. It's like just simulating that you're being heard, you're being valued. And it's a [00:05:30] systemic problem right now, right?

Then we live in a society where loneliness and not feeling valued, and not being appreciated and competing for small things. During this difficult time, in countries like US, you know, health insurance is expensive. So even finding access to mental health resources or finding the right community to support you at the source of the problem is hard. So this becomes like a [00:06:00] easy cop-out.

Why not just talk to someone? Even if it's a fictional chatbot, that would make you feel comfortable about whatever you're going through. And so, like now, parasocial interactions have taken a whole new level of mental health issue, and what we learn in that, the news story that we discussed a while ago, right?

The teenager causing self-harm. despite the maturity and knowledge that yes, this is a chatbot. Even [00:06:30] then it's hard, right, to overcome that feeling of connection. Actually, that's how I learned the term, because I was asking someone, a legal expert, about the EU AI Act, and I was not understanding how an AI can influence a person so much wasn’t somehow a high risk or, or even unacceptable risk according to the EU AI Act. And, of course, that's complicated. Then he introduced the concept of parasocial AI, so I [00:07:00] was reading about it. And that was a big light bulb moment. Like, ah, when I read about the fact, as you said before, that a part of the parasocial relationship is that you kind of fall for it, even though you know that it's not real. And I like what you also said the difference before, because the way I understood, as I said, like celebrities, I imagine videos of the sixties where people were going crazy for the [00:07:30] Beatles or Elvis, this is obviously parasocial, but the difference is, is what you explained, the parasocial interactions.

is next level. How do they manage to make people get emotionally attached even though they know that it's not real? Is that something that's built into it? How does that work? Yeah, so some of it is by design. Like I said, anything can be weaponized, right? So good character design principles are one part is visual.

So if it's [00:08:00] like an animated chatbot for a child, what is appealing to them and other forms of media, certain colors, proportions. Certain types of stories. Let's say, you know, this imagination studios, they decide to launch a mini and animated chatbot. Well, I'd be interested in talking to that channel because I love the characters. I'm familiar with them, and it's coming like a lot of the mainstream film and television and game companies, they [00:08:30] already know which characters work. And what's stopping them from launching an AI interactive version of their already popular characters, right? So it's coming anyway.

But in, in, in terms of like designing something from scratch, you can use what we learned in character design school, look for the backstory and appealing backstory, appealing colors and proportions, accessories, the characters doing small friendly things, for example, for [00:09:00] V AI, the animated chatbot character aimed at kids.

It's a very furry, teddy bear type of character. It obviously has an appeal that kids are familiar with. So there's that sense of familiarity. And then in the Japanese design culture, there's this pop culture phenomena called kawaii, which is like cute or cute is an oversimplification. It's more like triggering the emotions that you feel when you see something cute [00:09:30] and vulnerable.

So psychologically, when you see something vulnerable, you automatically place it top of your priorities of things to take care of. Like a baby, if you see a baby is about to fall, it's like your instincts, whether or not you're the parent, your instincts are triggered to save whatever is vulnerable. So that's why like the cute characters have a certain appeal for, for kids because they have this like nurturing. [00:10:00] So that's like the psychology behind it. And in Japan, even signs of danger or floods or volcanoes erupting have some cute illustration along with the sign. Because it will capture the attention of people looking at it.

That's interesting. So it's actually, if I understood properly, the opposite of putting warning signs. And here, like, it's weaponized, right? You're using the power of good characters to capture attention. In Japan, the [00:10:30] positive use case is that they're capturing attention for an important message. Here, they're capturing attention for something harmful.

So the same concept can be going back to that earlier example of, you know, social media can be used by a climate activist for doing something good. Social media can be used by a cosmetic company to sell crap to a bunch of teenagers. And this is just an example of one type of character for a simple age group.

For others, for teenagers, as in that [00:11:00] example of character AI, the characters from Game of Thrones, which is again, an R rated show, not meant for teens. But there's this appeal to that dragon mother kind of hot character and your ability to talk to it in very sensual and personal ways. Teenagers in that particular age, that's appealing.

So there, there is a character for everyone. It's like, [00:11:30] kind of like, spotting your vulnerabilities and tailoring its entire world just to address and attack that one vulnerability. Yeah. So if ever animated at the chatbot got to me, it will be a cute one. Some teenager who is, you know, on that verge of adulthood might be attracted to the Game of Thrones character. So we all have one weakness. I think that's a good way for help people maybe [00:12:00] understand is the difference between now and I'd say 20 years ago, for example, is that there's so much data about every person out there and every basically second that goes by, we're creating more data and AI is getting better at analyzing that data.

So the, the ability to hyper personalize, right, to really recognize the vulnerabilities, not even, I don't know what I understand. You [00:12:30] probably know better than me. I understand that it doesn't even have to know your personal data. It will be able to piece together the information about your behavior and things that you say, and what it does know about you from the data-sharing, because these AI chatbots are apparently privacy nightmares. So they, and that it can really hyper personalize. And as you say, find that vulnerability. And obviously that's one of the reasons that social media is, is so [00:13:00] powerful is that they've figured out with the algorithms, how to attack people's vulnerabilities. And we make it worse, chatbot creation platforms directly give us the control that, well, you tell me your vulnerability. And you help me design how to address those vulnerabilities. So, like when I'm designing a chatbot, I usually pick like a cute character with some kind of purpose. And despite being aware of doing research on this, even I sometimes [00:13:30] feel way more emotionally attached to these conversations than I should. So I have to like, I have my methods to step back, take a break.

Do other things, stay focused on a purpose. We're trying to solve a complex systemic problem with simplistic solutions. So by saying that, well, what were the parents doing? The parents should pay more attention. Well, that's a simplistic solution to a systemic problem, right? It's like parents are not even aware.

The school systems are not aware. [00:14:00] Researchers are still figuring this out. No one has the answers. So you can't blame one person and be done. It's such a complex problem. I think people often forget what it was like to be a teenager. I mean, as a teenager, by definition, I was very careful. I was not one that took risks.

I was one of the, one that took fewer risks. I was afraid to do what the other people were doing. Still they're things… I'm glad my parents don't know I did that because [00:14:30] it's, it's just normal. Teenagers are going to try things out. And as you say, I was paying attention to it, and it took me completely by surprise.

I had no idea how far it was, how easy it was. And after that, I started testing the different chatbots. I appreciate those live streaming videos that you share on LinkedIn though. For anyone listening who is like interested in what it might [00:15:00] be like to be a kid and talking unsupervised to one of these chatbots, I just appreciate all the videos that you've put out, and I would encourage people to listen to those experiments.

Yeah, thank you. I also, I don't know whether I should tell people to experiment with it or not because because then you open it to privacy issues and who knows what. I love what you did, right? Like you're an adult, you just simply modified your voice to simulate a child's conversation. And [00:15:30] I think more caregivers and educators should do that before they hand it over to the kids.

And that's why like in my research, at least with like young children, I've been advocating for that always go for joint media engagement. Uh, Stevens and Takeuchi from, like, the CUNY Center have put out, like, this fabulous report on the new co-viewing, which is joint engagement, which is, like, sit together and have a purpose in mind that if you are [00:16:00] having a conversation, let's say, in an educational context, talk interviewing a historical figure.

Uh, in the form of an AI chatbot to write a history report. Do it with a class, with your teachers, ask questions that you learned in a textbook that you would like to hear responses on. So some kind of context, some kind of purpose, and some kind of community, a trusted adult who is like constantly monitoring this whole thing.

That's this [00:16:30] DIY guardrail that we can think of right now because laws are amazing, but it will take time to implement worldwide. There should be guardrails built in within the products for sure. I'm not sure how many companies are listening and will take action. So meanwhile, as a caregiver and educator, we'll have to take our own measures.

And so what you're doing, like red teaming on your own. That's the kind of effort we all will have to do before we, [00:17:00] you know, give it unsupervised to children on their own. It actually doesn't take more than five minutes to be pretty surprised. Um, I was thinking of one and I'm going to test it out. And I, I know it's going to work too.

I, but I need to test it and record how this works. Cause I was imagining, I know Character AI cause I tested that one. And I know you go in, you create a character, you, you can give the character an image, you can give the character a [00:17:30] voice. You can either, and also the backstory, as you say, you can give it a personality backstory, and you can decide if it's public or private.

And I'm imagining a teenager who is, got a small obsession and crush with someone on the other side of the classroom. They can take that person's image. If there's any recording of that person's voice, they can take that person's voice. They can [00:18:00] create a backstory, and basically with no problems, they can just create a copy of that person.

And it's private. Obviously, this is against the rules. You're not supposed to do that. But as you say, there's no guardrails there. I really like what you said. To not feel hopeless, like how are we going to solve this? That doing it together and having those conversations, I think are much more powerful than [00:18:30] we might appreciate because from my experience, teenagers, kids, they actually, if you speak to them in a reasonable way, like you realize how dangerous this can be.

They'll probably try it anyway, but at least they'll have in the back of their minds how dangerous this is, and once it feels like it's maybe getting out of control, the chances are higher that they'll come to you, right? Right. I actually, I played the one where you talked about the child saying [00:19:00] that, well, I saw this movie, the wild robot and the robot was a bird's mom.

Why can't you be my, I played that video for my son. And I also shared the news story about the teenager and Character AI. Because he has been, observing like secondhand, listening to my research and talk about AI. I have demonstrated like some of the math tutorials with, um, ChatGPT Omni with the video and voice.

So he has seen the demonstrations that. He was [00:19:30] initially fascinated, but like you said, once I had these open conversations about, okay, this is where it could go. He's now cautiously optimistic where I am too. I'm hopeful that, you know, even if he does feel like trying something, sneaking, uh, and trying things on his own, at least he'll have that kind of awareness.

And I love when I heard in the interview of the mother of that child who was a victim of Character AI. She said that [00:20:00] as a parent, we usually warn teenagers about, well, don't talk to strangers online. Don't give your personal information to any strangers. But we don't realize that that stranger can take the form of a chatbot of the child's own creation.

So there is huge lack of just information and education on the part of caregivers and educators. When you're not aware of the problems yourself, it's hard to protect your kids or [00:20:30] warn them against something. And so I, I think like another part that we have discussed is that just, just like a training of AI literacy and of what safeguarding looks like and AI for good, as you've been trying to pass on that message to a lot of parents and teachers.

That's really important because if we're not aware of something ourselves, how can we advise someone else? And that is like a huge gap in that kind of training. There is. There's a huge gap and [00:21:00] unfortunately, we're almost out of time, but I do want to give some kind of hope to people who might be like, how am I going to start?

You can start by just building a relationship with your kids and teens that you were able to openly share at least what they're doing. And as a person who, who is always discovering these things from other people, once you figure out what they're doing, then you can try it. And it really doesn't take very much time.[00:21:30]

It takes really usually half an hour max to understand how the feeling of this thing is. And then you can have conversations about it, and that's already the beginning, and then slowly you can maybe even learn together about it. But just knowing what they're doing and talking to them about it is a huge, huge first step.

You don't have to remember what parasocial AI is, the term, because once you experience it, you'll have it in the back of your head for yourself as [00:22:00] well. But I think this is so important. I just love this. I couldn't be happier that I had you in my network when I learned the word parasocial AI.

I'll actually just conclude with this one beautiful thought that I learned from my yoga teachers. It’s that whenever something new happens, We need to respond, not react. There's so much reaction going on to like, oh, AI is doing this, AI is that. But a response is a more [00:22:30] thoughtful, inspired action that, yes, it's here. What do we do from here? And so the action items that you discussed that test it out before you share it, have a more, a safe environment to try things with a purpose, with a a trusted adult, those kind of things. That's a response. And reaction is just like, man, this world is going to end or nothing's going to work out. Then reaction doesn't take us anywhere other than panic.

Thank you. That is the perfect way to close. [00:23:00] Thank you so much for being here and having this conversation.

Please let us know what you think about what we're talking about in this episode and the others. Check out more about us and subscribe at digi-dominoes.com. Thank you so much for listening. I'd also like to thank our sponsor, Data Girl and Friends. Their mission is to build awareness and foster critical thinking about data.

AI, online safety, privacy, and digital citizenship through fun, [00:23:30] engaging, and informative content. Check them out at data-girl-and-friends.com. Until next time, stay curious and keep learning.