Loading summary
Transcript49 lines
- [00:01]
Liberty Vittert
Hello and welcome to the Harvard Data Science Review Podcast. I'm Liberty Vittert, feature editor of the Harvard Data Science Review. In today's episode, along with my co host and editor in chief, Shali Meng, we are exploring emerging technologies and how they are impacting our love lives. How are AI tools and algorithms changing the way people meet online? Can an app actually help you land more dates? And what does that mean for the future of dating and relationships? Our guests, Seong Kim, head of AI and machine learning for the dating app Riz, and Katherine Caduto, a communication science researcher and author from Boston University, will unpack these big love life questions and more in today's episode. But I have to mention before we start that this is our 50th episode of the Harvard Data Science Review podcast. In February 2021, we started our first podcast on love, interviewing John and Julie Gottman of the Love Lab, probably the most famous love and relationship researchers in the world. The word AI didn't come up once in our conversation, so we can't help but reflect on just how much the world of data science has changed in just a few short years. And we can't wait to see where it goes when we finally do our hundredth episode on love.
- [01:40]
Shali Meng
Sean, I gotta start with you. I know from my students what the word, the new slang word Riz means, but what exactly how do you all define Riz and what does your company do?
- [01:52]
Sean Kim
So Riz is short for charisma, and so essentially, Riz is an AI dating assistant. So what that means is that it helps you send that first text or you're stuck in a conversation or you don't know what to say, or you see someone's Instagram story you really like, or on a dating app, you have no idea what to say or how to continue the conversation. Riz is there to help you out and be like, okay, what can you possibly say? Let's just say you have a goal of, hey, I want to slide into this girl's DMs. She's in Rio de Janeiro. How do I have the best opener? So Riz is really there to be that wingman or wing woman and really just help you out. Say that first message and get the spark going. And so you can continue to get on that first date.
- [02:29]
Shali Meng
So it can be used across any platform. Text message, Instagram, an actual dating app, like any of these things, it can be used across.
- [02:38]
Sean Kim
Yeah, correct. So we are very intentional about that. Instagram actually is the biggest dating app that we see on our end. So we see a lot of Instagram, we see a lot of WhatsApp, we see a lot of iMessage. And so if there's a huge subset of our users are also submitting screenshots from dating apps, but maybe you have a high school classmate that you know, you have a crush on or someone that you met at a party, but you don't know, you know, find them on like a dating app or whatnot. Riz is really there to help you out and, you know, allow you to have an opener whenever you need. Agnostic. The fact that you match on a.
- [03:09]
Shali Meng
Platform, hopefully an opener better than like WhatsApp or something.
- [03:13]
Sean Kim
Yeah, really, you want to stand out. You want to actually have something meaningful to say instead of like, hey, what's good looking? Or like something like that. You want to really have a good question or a good opener to kind of spark a conversation.
- [03:25]
Katherine Caduto
Well, thank you, Sean, that I learned the word Riz this morning from Liberty. I said, what is that? I do not even know how to pronounce it. And I know that for me, that this whole thing is entirely new. Right. I've been out of the dating scene for many, many years, and in my years back in China, it's entirely different. So I do have lots of other questions I'm going to ask later for both Sean and Katie in terms of this whole idea of using AI to help humans in terms of the emotion side. But because this is a data science podcast, I want to first get into a little bit of more the technology side. So the question for Sean is what technology you actually use and specifically what data you were using to train these models and what factors influence the generation of the responses here.
- [04:10]
Sean Kim
So the technology behind this, obviously AI and LLM, is very, very popular. Right? It's no secret that we use OpenAI, Mistral, Grok, Llama, Deepseek, you name it. We test out all the state of the art models every single time they come out, once we deem it's ready for use. But I think what people miss oftentimes is input to the LLM, which is not true, actually. So because the input is primarily visual, we rely really a lot on computer vision models to really understand the visual context of the screenshot that the user is sending in. Right. So like I said, is a conversation coming from imessage or is it a dating app? Right. Or is it Instagram? There's so many contextual clues that would influence the response and not just like, you know, what platform is coming from, you know, Right. Is it an image what is the image actually of? Maybe it's a picture of someone at a beach versus, you know, on the street, you know, that's going to influence the reply of the conversation. So really we try and extract all the visual information and really try and get a human level understanding of, you know, what you would reply. Right. So, like, you always have, you know, that friend who's looking over your shoulder telling you, it's like, hey, you should do this, because X, Y and Z. And they kind of give you a list of things, right? And so on Instagram, they'll be like, oh, yeah, like, you know, she's at the beach, she's, you know, holding a coconut or something like that. Why don't you make a comment out of that? And like, in those, like, few seconds, those are all, all visual processing. When we really look for these visual cues of understanding of how to start conversation, there's a big difference. When it comes from Tinder or Instagram, if it's Tinder, you guys have matched. You already know there's some kind of level of agreement that you may entertain the other person versus Instagram. You don't know. It could be a classmate. Right. It could be someone you met at a party. You don't know. So we're going to have to tailor the responses for that. And so computer vision is really often something that we overlook when we talk about the technology behind that. And in terms of the responses, Right. So at the very beginning, when Riz started, you know, we got female dating coaches, male dating coaches, and we took a bunch of replies, scenarios and whatnot and fine tuned the replies or like, what they would say on these various data sets of, you know, like, here's a certain scenario, here's a certain age group, here's a certain demographic. Write a reply or write what you would do in this scenario. And we fine tuned it on a lot of these dating coaches who have offered to help. Yeah.
- [06:22]
Shali Meng
So, Katie, can you tell us a little bit about your research? Like, how did you get involved in the love business?
- [06:29]
Katherine Caduto
I love that. Yes. So I started getting interested in dating apps right around when they were coming out and really taking off. So in 2012, with Grindr and Tinder taking off, I was just kind of intrigued by those. And then I was actually working in advertising at the time and was working as a researcher in an advertising agency. And I thought it would be really, really cool to be able to take my research skills and apply those to this kind of growing industry. And so, like, I joke about it, but, like, I really went back to get my PhD because of online dating and dating apps. Like I was just so intrigued by what I was seeing, you know, anecdotally like watching friends swipe, watching them try and find that perfect relationship. And so I saw these things really taking off and I was like, how can I learn more about this? And so I have been now formally studying these since about 2016, so almost 10 years, which has been really fascinating to watch. The very rapid evolution from match.com Desktop based dating to now this focus not just on mobile, but with the incorporation of AI. And I can tell you that just in the last year, the number of interviews I've done around AI, around people getting these kinds of, you know, helps, has really, really grown exponentially.
- [07:59]
Shali Meng
Yeah, it's funny, we were actually, Shelly and I were talking about it earlier. This is our 50th episode and we did our first episode ever with the Gottmans of the Love Lab. And the word AI never even came into the conversation. Like, this is all so brand new and happening. It's really, it's a cool thing to be researching that you've probably seen change drastically in the last 10 years. But could you tell us a little bit, you know, how you think the use of AI for communication on dating apps really influences the quality of conversations, the relationships that are formed. I mean, I'm sure it matters like how much AI is used versus the person, but what does your research sort of tell you?
- [08:36]
Katherine Caduto
So it's been really interesting again, especially in the last year because what I think we're seeing is consumers overall are just more aware of AI. They're thinking about it more. When I was first doing this research, people might mention something like the algorithm, and obviously that's also a form of AI, but they weren't thinking of it like that. Now in the last year, people are really thinking about how can I utilize AI to really help me with that opening message. It's just like Sean said earlier that what people tend to worry about is how do I hook someone, how do I say that thing that helps me to stand out. And what's also really interesting in this are some of the gender differences, particularly when we look at cisgender men, cisgender women. Because there's lots of evidence about this that there tend to be more men than women on dating apps. And so men in particular are really trying to find a way to stand out. And when I interview people about this, when I survey them about this, they say like, I don't really want to respond to a message that just says like, hey, you know, I want some evidence, some proof that, you know, someone looked at my profile and really thought about what they were going to say. And in that case, people are pretty optimistic about AI. They're very open to it. Where it starts to teeter is when they feel like they're actually just conversing with the AI and not a real person. And we can get so much further into this because there's a lot of research, not just my own, when we look at online dating about deception overall. And so it's not that this is an AI only problem by any means. It actually, to me it fits in with other trends that have permeated long before AI was so prominent.
- [10:12]
Sean Kim
Totally.
- [10:13]
Shali Meng
You have like filters where, you know, what does the person actually look like, all that kind of stuff.
- [10:18]
Sean Kim
Totally.
- [10:18]
Liberty Vittert
That makes sense.
- [10:19]
Katherine Caduto
Well, that was the conversation that Liberty and I, we got ourselves into this morning, right? Because I was thinking first of ask myself, like, if I were part of this process, why would I interested in knowing that someone is responding to me or open the line basically was not that person or either he or she, Right. How would I emotionally respond to that? I would almost want to avoid doing this thing. But then I realized I was asking Liberty. I said, oh, you know, I realized the one problem I did not really understand, I still don't understand is what is the purpose of this dating app? Because you know, in my generation trying to dating that means you want to develop a long term relationship, eventually get married. Now I realized by saying, so probably completely date me, right? Because probably that's not what the current. Okay, so now you guys have educated me here because I think are you saying that this kind of app or whatever you develop is not going to serve the purpose like what I have in mind, but rather serving some other purpose. And I think for at least the older generations like myself, it might be very useful to know what is the purpose being served and the wine that would work on the people to respond to those things. Knowing this may not be genuinely from a person.
- [11:40]
That's a, I think a really great and important contextual question. So when you look at dating apps, right? And when we think of the word dating, we think of long term relationships, people, you know, getting together, romance, all of that. But you're right that for a lot of people, they use dating apps not just for romance or for very different forms of romance. And so in my own work, I will often ask people what kind of relationship they are looking for. If they're looking for a variety so that ranges from the more traditional, like, yes, I do want a stable long term relationship to people who are looking for casual sex for a one night stand. But then you also do have people who are downloading dating apps because they're looking for friends. So it's a really, really wide range. And within that then I think that does change the openness to an AI message or the interpretation of an AI message. I think there's also a trade off where, how and when a person reveals that AI message. Definitely having an impact. Because I think if you are looking for a long term relationship and you're really nervous about the opening line, I think if you're having a good connection with someone and you say like, hey, my opening line, I had help from something like Riz, I don't think that's a deal breaker for a lot of people. I think it's when again it kind of goes on and on and on, then it's way later and it's like, oh yeah, I used AI. I think that's more of a violation. And again, I think that would then change when you think about someone who is looking for casual sex. And research hasn't really gotten into this yet, but I would be very curious to see how if you're looking for like a one night stand, you probably aren't really that worried about if it's AI or not because you're trying to find those kind of matched goals and it's not about that deeper kind of connection.
- [13:29]
Shali Meng
Sean, that makes me think, you know, for Riz, when you're using Riz, is there information that the user puts in that determines how the AI is going to act? For example, whether you're gay or straight, whether you're looking for a long term relationship or a one night stand or friends. Like, is there any information the user gives ahead of time?
- [13:46]
Sean Kim
Yeah, for sure. Yeah. So like, you know, we factor in a lot of stuff, right? It's like, what's your gender, what are you looking for? You know, what's your age range, right? These all affect impact, the style replies that we give, right? We have looking for a relationship, you're already in a relationship. Actually a big subset of our users are already in relationships and they still want to use risk to spice things up. Maybe, or maybe they're just looking for casual or you just don't know yet. And also the lines between something a 60 year old might say in the dating scenario will be a little different than what an 18 year old would say. So we factored a lot of These user inputs in users can interchange. Maybe they're feeling that's like, oh, maybe I'm ready for a relationship now. So then they can change to that. We always prioritize the user input and what they're looking for. Yeah.
- [14:27]
Katherine Caduto
What is your process to kind of humanly validating these responses generated by the computer? Is that where you say, ah, that line is to be kind of the right line? Because humans wouldn't have respond this way or they just created something which is very exotic. Who knows? So what is the process there for you as the algorithm developer to understand how much this AI doing is actually better than a human or maybe maybe just replace human, but much larger scale. What's the process there?
- [15:01]
Sean Kim
So we at RIZ really try to embrace randomness, right? Like people will have different styles of replying or what preferences for the replies or, you know, that we generate. So we really try and embrace different styles and we try and give users, you know, everything. Essentially it's like, oh, here's, here's a possible thing. We kind of view Riz as a collaborator rather than someone who replaces your texting style for you. So it's kind of like a friend looking over that shoulder. Again, we want to give examples to people. Here's what you could say and here's a different way you could say it or you can edit this one that we give you. We really want to give power to the user to really control the responses because ultimately they're the ones in control of sending this message. And so when we try and think of is an AI better at replying than a human at times and for different users, probably yes. As in there are people in this world who do not know how to date well, and frankly, AI could be better than them. But the whole idea is that we want to teach users, here's what's a good response and whatnot. And so internally for us, when we look at these replies, there's always a North Star metric, like, get a date with this girl. Continue the date with this girl. Maybe you're ghosted, maybe you want to instigate another response from her. There's always a North Star scenario. And we essentially get the replies and we judge the reply. It's based on the likelihood that you would get this North Star metric. And so we kind of like compare that scenario to the reply, basically get a probability of like, hey, like, maybe this is. This would be likely to be the best result for this certain scenario. And we try and judge the responses based on that. But it's Very, very dependent on the user. Right. So again, maybe reply would be better suited for someone in a. Someone looking for a relationship rather than a one night stand. The objective or the North Star is going to be very different. So it's very, very dependent on the user intention. Yeah.
- [16:46]
Shali Meng
I have to say, Sean, I downloaded Riz right before this and I used it like I took screenshots of my messages with my husband and then sent him. It's so good. It's so fun. You know, it's funny, when I did a story on this about a year ago, my initial reaction to it wasn't Riz, but like the concept of it was like, that's cheating. But then I thought about it and I was totally wrong because I mean, when I put on makeup, I literally don't recognize myself sometimes. So like, is that cheating or you know, as we talked about, Katie, like the filters that people use on their apps all the time. And Sean, what you said like a friend over your shoulder. Like when I was dating, I would totally send my like screenshots of my conversations to my friend group and be like, what should I say? So that's the same concept as something like Riz, it's like an aide or a friend or a helper. But you know, it also does beg the question of how does you know, this AI generated messaging really impact the authenticity and depth of questions on these dating platforms, like, do users feel more comfortable or authentic when interacting? And Katie, I know that you touched on it a little bit, but maybe if you could go a little bit deeper into that.
- [17:49]
Katherine Caduto
Yeah, absolutely. And I think you hit all the main points, right, that there are so many layers of what makes someone authentic, particularly when we look at their online presence. And so this is why I think this AI kind of debate in this space has been so interesting because I also often compare it to just like having a friend help. And I've surveyed people about how often do you have friends help you when you're choosing photos, when you're writing a biography. And so many online daters absolutely rely on friends and they usually. And it's not just one friend, it's multiple. And so there's a whole kind of, you know, group of people that's helping you get set up to find those dates. The other thing, like you said, when people put on makeup, when women wear makeup, it often changes, right? Like how they look. And it is very different from what they might look like on the day to day. And something that I've looked at a lot is the pictures that People choose to put into their dating profiles and how then, you know, does that translate to what they are when you meet with them in person? And there's a lot of research into this, going back to desktop dating, that of course people choose the best photos of themselves. They will often choose slightly older photos if they look a little better. And so there's actually, in this area, there's a framework called the Profile is Promise framework, which basically says that what you put into your dating profile is going to approximately be who you are. It's essentially a promise to your potential matches or your potential in person dates that you'll be pretty close to that, but you might not be exactly that. And so then when I look at AI, particularly when I think about AI in opening lines and getting the conversation going, to me it feels like another tool or another way to kind of manage that self impression. Because that's what a lot of this is, right? It's not about saying like, oh, the AI is going to help me fall in love. The AI is helping me to craft that self impression that gets me, like Sean said, to that kind of next step, that North Star, whether that's maybe not even going on a date. It could just be, you know, getting them to add you on Instagram. Because that's something I see a lot too is, you know, people look for clues of authenticity at so many stages of this process. It's so rare that you would have someone match on Tinder and then immediately go on a date. Instead they're trying to go from Tinder to Instagram, maybe to Snapchat. So there's kind of a collection of authenticity that goes on throughout this whole process. And in my mind, AI is again, it's adding a tool into that kind of toolbox for assessing.
- [20:29]
I was just thinking about, okay, fundamentally, and I agree AI is a tool, but this is a bit different tool. This tool has generated first is, you know, seems now it's almost everywhere, right? This is like fundamentally is kind of changing the human and the human interaction and because it's generating emotional response for us, right? Because we. Now I find myself like I was, you know, prompted in prompted ChatGPT. I got very frustrated. Then I realized, wow, the machine actually engaged me. It generated emotions on my side. So the question is really, what are the current researches on how these technologies are really changing the human's emotional process? Are we become a lot more skeptical because everything could be presented differently, or are we more impatient because we should be able to find things right away? What are the research on that. I mean, this obviously big than love perspective, but I'm thinking fundamentally is about a human's emotion response. Right. So love to hear your thoughts on whatever, you know.
- [21:36]
So there's. This is a really interesting and probably one of the biggest growing areas in this research field, in part because it takes a while to do this kind of research where you're actually having people respond to and have that emotional response to artificial intelligence. That's actually a study that I'm working on right now. So we're measuring people's responses to AI over time. So we're having them talk to ChatGPT and see how those emotions are changing. So more to come on that front. But I think there are kind of two pieces to this because on the one hand, like you said, you see people who are, you know, maybe they feel frustrated when they're working with ChatGPT or they don't necessarily love the response that they're getting. But on the other hand, I don't think that humans are just taking what an AI tells them without necessarily editing it. The thing I've kept thinking about is that, you know, if you're using something like Riz to help you get a response, you could still edit that and you still have a lot of control. And I think the AI conversation is really interesting to me because I think we always take teeter on taking away human agency in these conversations. And I do think that humans still have a lot of agency and a lot of control. Right. So if an AI, if ChatGPT tells you to say something to someone, you don't have to do what it says. Or you could say like, okay, this is a starting point, but I'm going to adjust it from there. And so these are the things that research is really starting to get into right now. So this is, I would say, like probably the most kind of cutting edge research that is out there because we're only now starting to understand how people manage the AI and what their emotional responses are to it.
- [23:23]
So back to the question about using technologies that I have two more questions. They're related, One is to Sean as really about how do you protect users data privacy. And the one thing you said, which you remind me, you just said, well, you said people might want to spice things up even they're in relationship. I guess they probably don't want other people to know if they're doing. So how do you make them trust your app? Because everything online eventually is online as well now. What are the processes you are taking there?
- [24:01]
Sean Kim
Yeah, that's a great question. So we get the question a lot. We don't permanently store your data and we don't ever directly read the conversations. Right. That would be very bad for us. It's an invasion of privacy. So all that happens is we encrypt all the messages. So hopefully no bad actors can get into our systems or databases and really we get the responses and you can generate the replies but after a certain amount of time we delete that data from our database and so it's only stored on your device and so there's that. But I think really we obviously want to be a privacy forward company. It's really a balancing act of how can we get meaningful insights without invading our users privacy. How can we add noise to our data and get the same scenarios without individually going into each column, each row and looking at that. So we always take a proactive approach. Not only are we encrypted, but we want to obviously use the best privacy methods possible. And I think or making efforts on our side to try and do that. Yeah.
- [24:59]
Katherine Caduto
How do you ensure, I think this is not just for you. I think for most of these apps. How do you ensure you do not getting to bad actors just play with. You can generate lots of images not related, all kinds of terrible things, do bad things using your app to seduce people. They shouldn't do these things. What is your safety mechanism there?
- [25:21]
Sean Kim
Yeah, so we obviously have safety guidelines. So we get pinged really if there's a safety incident, if there's like certain words that are being said that we deem offensive, obviously we would get flagged and then we take action on basically not letting the person use our app. And so we have those mechanisms involved and really we try our best and be like hey to the AI it's like hey, there's a certain. These are the scenarios we want to avoid, obviously like anything like underage, any abuse or whatnot. We want to be able to get flags and be like hey, these are the scenarios we don't want to look for. These are the keywords. We don't want users generating any like slurs or whatnot that are like very, very harmful. We want to avoid. You don't want to see on our platform. So we do go and do a pretty good job about like detecting those and getting those out and then automatically banning those users.
- [26:10]
Katherine Caduto
Katie, what are the general research in those areas? Do you get into this research to ensure the safety use of any of these AI tools?
- [26:20]
Absolutely. Safety comes up a lot even when we're not necessarily directly asking about it, because I do think for especially online daters, but most social media users, there's a lot of concern about safety, about bad actors. And I think this is just particularly heightened in online dating because you have people who are potentially going to meet people in person who they've never met before. And I think that is where AI becomes a little scarier for some people because there's a sense of, well, if I've been messaging with them through AI and then I meet them, what if they really aren't like that at all? And that's, I think, the biggest safety issue. That again, it organically comes up. It's not something that I necessarily ask about, but it is something that users are very, very aware of. And again, it ties back into other research into online dating, where when, you know, people posted photos that weren't exactly them and then they meet up with that person and it's like, oh, you don't really look like what I thought you would. And we see this particularly among women. Women in particular tend to have higher concerns around the use of AI when it comes to them meeting someone in person because they are typically at greater risk. I just collected data that actually shows this, that in comparing men to women in my sample from about a week ago, men were twice as likely to feel optimistic about AI. And I think that's not the most surprising thing to me, just again, given some of these safety concerns. And it also ties back into authenticity, right? Just trying to verify that someone is who they say they are, that they're being kind of their true self.
- [28:01]
Shali Meng
So I got to ask, you know, Riz is three years old. The concept of AI and dating is so new that even just a couple years ago, when we started this podcast, it wasn't even mentioned. What's the next frontier?
- [28:16]
Sean Kim
So I think a lot of people don't realize that AI has played a pretty big role already pre LLMs to dating apps, right? You know, matching algorithms or whatnot. You know, what are your best photos? You know, they have all, they have all like the machine learning algorithms for this already. So I think like really the next step is like a two prong thing, really. So if you want to talk about, you know, the consumer expectation, I think users are going to get used to AI in their dating apps or whatnot or in their dating lives because it's already been a thing. And I think people are naturally going to accept that LLMs are just a natural. Another feature added on to an already AI centric application, Tinder, the matching algorithm they one could already consider AI, but I think in terms of new features, matching algorithms aren't perfect. AI matchmaking is a frontier that a lot of people are trying to figure out. It's very, very hard. And also there's arguments to be made that's like, okay, even if AI were able to match you to your perfect theoretical partner, is that the best thing for you? So let's just say at 18 years old, you put me in front of my girlfriend, probably would not be the best thing. I'd probably mess it up somehow. But because I met her later in life and I had all this dating experience now that we can have such a great relationship. So I think really, like, the next frontier is AI matchmaking. You know, how can we create better algorithms so that users are more satisfied with the quality of their matches? I think there's also going to be a huge push for, you know, like, assistance, right? Like, you know, something like riz, but really, like, you know, it's like something empathetic, right? People have a lot of dating struggles, right? You know, why am I getting ghosted? Why am I not able to get a second date? You know, someone to listen to their dating struggles in a really private way, right? It's embarrassing to say, you know, certain things that happen to you on dates or, like, in your relationships. And I think people are going to really, really start relying on these assistants to give them guidance on, like, hey, here's a scenario, right? My girlfriend got mad at me. Or like, hey, this girl is being wishy washy about committing to me as her girlfriend. You know, I think people are going to see a lot more of that. Yeah.
- [30:19]
Katherine Caduto
I would also say that I think AI matchmaking is kind of the next real frontier. And I think, again, just to add to that, one of the things that I've also noticed, both in my own research, but in other research in this area, is that there has been an increase in dating app burnout and fatigue. And so I think the promise of AI matchmaking, the promise of potentially improved matching algorithms, the idea that you do get help when you're sending messages is really appealing for people because I think when you look at the dating app market right now, it's very saturated. And there's a feeling that no matter which app you're on, you're kind of going to have the same experience. And so whichever of these companies can figure out that kind of next level of actual AI matching and saying, kind of like Sean said, right? It's not necessarily that it's your perfect person. But this is a really good match for you, and we can kind of guarantee it. I think that is something that consumers are already looking for and are eager for.
- [31:21]
Shali Meng
So we always end with this magic wand question. So if you guys could wave your magic wands and create your own dating app from the ground up, what would be its key feature? And we might have already hit on it, but let's see if there's anything else. Sean, you want to go first?
- [31:36]
Sean Kim
Oh, gosh, where do I start? I think if I were to create a dating app from the ground up, I would fundamentally try and create a dating app where it kind of ensures, like, because we're very visual creatures. Right. But it's very hard to. For, you know, for, you know, hinge or anything to really figure out, like, who you're attracted to, just, like, visually speaking. Right. I think we know in almost instantly, in a few seconds, you know, if you find someone attractive or not. So I think that as a guarantee of, like, okay, we're going to serve you individuals we know you're attracted to. So not only visually speaking, but obviously that's a huge part of it, but, like, you know, as a person. Right. So I think trying to figure out how can we. How can we really align the user incentive and the business incentives of, like, you know, we want to serve you good people, essentially, and keep you on our app and be happy about, you know, the people you're being served. Maybe the person you were served was not your type, but, you know, was not compatible with you. But, you know, at least, you know, visually speaking, you were attracted to them. So users are not discouraged. It's like, oh, my gosh, like, you know, I'm in the dumps, or I'm at, like, the 0elo score. I think, fundamentally speaking, like, I would want to ensure that users are always, always, you know, at least maybe not as a person, but, like, visually speaking, they're attracted to the matches. And we can guarantee that at, like, a 99% rate. Yeah.
- [32:59]
Shali Meng
I have to ask, what's Elo?
- [33:02]
Sean Kim
So Elo is like a video game term, right? It's like, you know, what's your. What's your rank at? It's very interchangeable. And so the common theory is that there's an ELO score for these dating apps of your attractiveness score internally. It's like, what's the rate of people swiping right on you? And that's impacted by what their ELO score is? Or they're ranked essentially. And so that's the Common theory.
- [33:24]
Shali Meng
Shelly, you're going to have to ask your students what their ELO score is.
- [33:28]
Katherine Caduto
I will.
- [33:30]
Liberty Vittert
Katie, what about you?
- [33:31]
Katherine Caduto
So this is something I think about a lot. And if I could, like, wave my magic wand, have a dating app, I would have a dating app that I don't want to know if I want to say made, but strongly encouraged people to meet in person. I think that is one of the other, like, really big pain points in this area is just people get stuck swiping, they get stuck messaging, and then they're not moving on to something else. And so I think an app that said, like, cool, you're matched, you have 72 hours to get on Instagram or go on a date. That, to me, is what I think is really missing in this particular market right now. And I think that would help overcome a lot of the burnout, a lot of the fatigue. And I think that's where AI can come in, because I think that's another area where people struggle to say, well, how do I say, let's go on a date? How do I say this in a way that gets people excited and isn't just like, oh, I'm trying to rush things, but I think when I think about the ultimate dating app, that's really what I think about.
- [34:32]
Shali Meng
I love that idea. I had a friend that used to say that if a guy didn't ask her out within three back and forths, she was just done with, she unmatched.
- [34:40]
Katherine Caduto
That's honestly such a healthy strategy.
- [34:44]
Shali Meng
I'll let her know. Yeah. Well, thank you both so much.
- [34:49]
Katherine Caduto
And thank you from me because I'm so glad that I don't really have to think about dating anymore because I don't know these words, the risks, elo, whatever it is. And thank to both of you for helping our readers to understand both kind of how dosing works, but also on the technology side, thank you very much.
- [35:11]
Liberty Vittert
Thank you for listening to the Harvard Data Science Review Podcast to stay updated with all things HDSR, you can visit our website at HDSR, mitpress, mit.edu, or follow us on X and Instagram @the HDSR. A special thanks to our executive producer, Rebecca McLeod, producers Tina Toby Mack, Arianwyn Frank Gavin Yang, and Belle Reilly. If you liked this episode, please leave us a review on Spotify, Apple, or wherever you get your podcasts. This has been the Harvard Data Science Review. Everything Data Science and Data Science for everyone.