Taha Yasseri: On Tinder, both sides swipe on each other. And then when there is a mutual interest, they can talk to one another. It’s symmetric by design. But in practice, we see that 80% of conversations are initiated by males. And even in those cases, the 20% of conversations that females start to talk and take the initiative, they are punished for that.
Noshir Contractor: Welcome to this episode of Untangling The Web, a podcast of the Web Science Trust. I am Noshir Contractor and I will be your host today. On this podcast we bring thought leaders to explore how the web is shaping society and how society in turn is shaping the web.
My guest today is Taha Yasseri. You just heard him talking about the gender gap that exists when it comes to who starts conversations on the dating app, Tinder. Taha is an associate professor at the School of Sociology and a Geary Fellow at the Geary Institute for Public Policy at University College Dublin, Ireland. He has been a Senior Research Fellow at the University of Oxford, a Turing Fellow at the Alan Turing Institute for Data Science and AI, and a Research Fellow at Wolfson College at the University of Oxford. He has studied the dynamics of social machines on the Web, online collective memory — and my favorite: online dating. Welcome Taha.
Taha Yasseri: Thank you very much. Noshir, I didn’t know you’re on the market.
Noshir Contractor: I am not in the market which is exactly why I like to be an observer. But there are lots of people who are in the market for online dating. And as you mentioned in one of your recent articles, it’s over a $2 billion business just in the United States and is expected to continue growing in the foreseeable future.
Taha Yasseri: A lot of things that we do have changed due to internet-based technologies and web technologies. But to me, one of the most important things that have been revolutionized over the past 10 to 20 years is dating and mating — the way that we meet and we choose our partners for shorter relationships and for longer relationships, sometimes lifelong relationships.
Noshir Contractor: I remember when online dating first began, and there was still a stigma against it. And people would not even be willing to admit that they were going online to look for dates. That changed. Tell us why you think it changed. What brought about that change? And what got you interested in doing research on this topic.
Taha Yasseri: Any new technology has its own stigma. Particularly online dating was seen as a tool or an environment for you know, not committed relationships or behavior that are promiscuous, which is something in general, something that the societies are less judgmental about it at the moment, but also, people realized, no, actually people can find partners online and through this apps or websites that they can buy, you know, settle down with and get married to and create lovely and happy families.
Noshir Contractor: One of the things that we have been asking ourselves well before online dating is what are the kinds of traits that men find attractive about partners they are seeking, as well as females find attractive about partners that they are seeking? What is your research on online dating told us about differences or disparities in user behavior between male and female users?
Taha Yasseri: One of the most striking things that we have seen is the imbalance or the gender gap in initiation of the conversation. These modern technologies, they try to give equal weight to both genders. For example, let’s say on Tinder, both sides swipe on each other. And then when there is a mutual interest, they can talk to one another. It’s symmetric by design. But in practice, we see that 80% of conversations are initiated by males. And even in those cases, the 20% of conversations that females start to talk and take the initiative, they are punished for that. They receive less responses, compared to conversations that are initiated by males. As if collectively, we judge them because of cultural biases and the baggages that we as a society, still are dealing with. So in a newer project, We looked at 10 years trends in online dating, and we were hoping to see this gap is actually getting closed, and the balance is increasing. However, it wasn’t the case, actually, we realized that over 10 years, the gap in initiation rates has increased. And that simply tells me that it’s not only the technology, we have cultural baggages and we have things that we want to move on. And only having a shiny website is not the only solution. We require other things as well, which might not be even easy to attain through that technology.
Noshir Contractor: And indeed, online platforms are reifying the norms that preceded the platform’s in terms of males being the ones who were expected to initiate these kinds of interactions.
Taha Yasseri: That’s very true and we can look at those trends at a very large scale. People click and people send messages. And we look on into the logs generated by these activities rather than asking people because of course when it comes to dating and mating, people’s behavior can be very, very different what they say on a survey or on a questioner. And I think web science methods and computational social science methods are particularly adequate to to address these questions, and to look at the traits in mating — our preferences and our behavior.
Noshir Contractor: Well, one of the things that we know historically, or at least, It’s socially been circulated that women put more emphasis on income and education when it comes to potential partners. And there is always the debate about the importance of physical attractiveness. What does your research show about these questions and if these criteria are changing, since online dating first began?
Taha Yasseri: You’re absolutely right. It has been predicted or reported based on a small scale studies that females put more emphasis on societal features like income or education, and males with more emphasis on physical attractiveness. This was something we observe in our analysis as well. But what we saw and it was interesting was that the emphasis on education and income is decreasing. People are more accepting of differences between their own education and income level and their potential partners, particularly when you look at female users. And it could have many reasons. Of course, one important factor here is women are much more independent today compared to 10, 15, 20 years ago. And that makes the income or education level of their potential partners less relevant so they can focus the interest into other factors.
When it comes to physical attractiveness, one of the things that I find fascinating coming out of the analysis was that we looked at the popularity of profiles, measured through the number of messages that people receive, versus the self reported attractiveness. If I tell you Noshir I’m a 10 out of 10, you would think I would be very popular on this online websites. Well, actually, in practice the most popular male users or profile owners on online dating websites, and the ones who actually think they are at 10 out of 10, are not receiving as many messages. This is something we call the douchebag effect, because you know, someone who thinks “I’m a perfect 10,” particularly a man who thinks I’m a 10, out of 10, probably is lacking some other personality features that are attractive to female users. But when we look at the female users, the higher they rated themselves, the more messages they received, the ones who thought are 10 out of 10, were actually the ones who received the most messages.
The other factor why very attractive males do not receive a message could have to do with self confidence and self esteem of female users. They might think, oh, that guy is out of my league, I might not even try. Whereas men don’t have this understanding of their capabilities. You know, even if they are sure that the potential female partners out of the league, they still try.
Noshir Contractor: There’s also been a perennial debate about whether dating and partnerships and romance is more likely to succeed when birds of a feather flock together. Or the other saying, which is opposites attract. What did you find about the similarity between profiles and the extent to which it might have predicted future success in terms of online dating?
Taha Yasseri: One of the things that people have credied online dating for is there a higher ratio of interracial marriages and relationships today compared to 10, 15 or 20 years ago. It is very difficult to argue that this is primarily due to the surge of online dating or it’s something that happens anyway, parallel to online dating. But I do think online dating provides us with much more diverse of a pool of potential partners. When we looked at data, however, we realize that homophily, or similarity between potential partners is not a very strong predictor of success. We couldn’t measure success after relationship, of course, because the irony of online dating is that if it works, you lose your customers. But we could see, for example, if people exchanged phone numbers, or if people carry on chatting for a while, we took measures of success like that. And we realized homophily plays very small role, this could be a reflection of a bigger change in our society is that now we are more curious and more accepting of people who are different to us.
Noshir Contractor: I want to move this to another part of your research where you have argued about whether we can use algorithms and intended bias injected within these algorithms, to move us away from our natural tendencies for homophily, for creating echo chambers or creating fragmentation. And so tell us a little bit about how you got interested in this notion, how we tend to naturally move towards segregated networks in segregated societies. And then you have a very depressing message, you say that, even if we are to use positive algorithms to try to break away from these tendencies for homophily and echo chamber — your research shows that we’re not likely to be successful.
Taha Yasseri: We all agree that we have become very fragmented in our political opinions, particularly in the US, I would say, in the UK, some countries that have gone through a lot of trouble in recent years just because of the divide in the society. So in that sense, bubbles have formed and echo chambers are there. I had heard and I had read that people say it’s up to the platforms to break the bubbles. They have to use algorithms to mix people up and connect people from other opinion camps. And we thought okay, well let’s let’s see if that works. So we developed a mathematical model, and we realized, as long as we have homophily in the network, as long as there is a slightest tendency for an individual, to prefer a connection to like-minded people, over a connection to someone dissimilar to them, no matter how much algorithmic bias we introduce, bubbles will form. We might postpone them, but we never can break them. Because that tendency — that homophily tendency — is so strong that we basically practically need huge amounts of algorithmic intervention, which, of course, takes all of the joy out of the online social networks, right? We do not go on Facebook or Twitter just to fight. And that’s actually what social network companies have capitalized on. Because if you’re happy there, we interact with people who are like minded and support our opinions, we spend more time there and we are more likely to click on the ads and so on. So confrontation is not something social networks advocate for, and combining that with homophily, and the ease of disconnecting from people who are different to us on social networks, all this together make the formation of echo chambers and bubbles inevitable. It sounds very grim. I agree. But we also propose a couple of solutions. It’s not that that’s the end of the story.
You know, before the internet, I live in a village, not everyone thinks like me. My neighbor might vote differently, might think differently. It’s not that I just move next day, you know, I still go to the same church or to the same strip club, depending of my interests. And I interact with people who are different to me, and through this interaction, I might not completely change my opinion. But at least I appreciate the differences. I learned to understand and acknowledge the existence of other opinions. On Twitter and Facebook, we are encouraged to block people, but we shouldn’t just block others or unfollow others, because we don’t agree with them. This is such a new web thing that we just don’t talk to the person so easily, and web gives us the opportunity not to see that person ever again. Whereas in that village, I had to see that neighbor anyway. We somehow have to introduce mechanisms, which encourage people to keep the interaction on and carry on interacting with people who are not exactly the same as themselves. And can I think of an example, of course, Wikipedia, that’s where these conflicts and these clashes of opinion happen. And I have spent years studying edit wars between editors of Wikipedia. One thing that we realize is that the more conflict and the more interaction between opinions and an article there is the quality of the article increases.
Noshir Contractor: And so what I’m hearing you say is that if Facebook were to take work to make the algorithm make interventions that you think might help, that unlike what you or the algorithm might hope, which is that I will look at this and consider other points of view and it will broaden my perspective. Instead, what the user would do is simply walk away from Facebook because it’s not feeding them what they want to hear.
Taha Yasseri: Either that could happen. So that explains why social media platforms might not even try. The other thing and that is based on research that Chris Bale and his colleagues have done in their control experiment. People who are exposed to content from a different opinion have become more extreme in their own opinion because that wasn’t necessarily a interaction between humans, it was me seeing some content supporting the opposite opinion. And I never had the chance of having an active interaction with some human of that opinion. So I don’t think content sharing, meditated by algorithms is the solution. All we need is human to human direct interaction. And it is not comfortable. We all know that. And the cost for the platform could be that people walk away and their revenue might go down.
Noshir Contractor: So you already spoke about the fact that as as an example of good engagement, where editors and Wikipedia would go off to one another. And the more they argued with one another and debated one another, the higher the quality of the final Wikipedia page that they were debating. But you also talked about the role of not just human editors of Wikipedia battling with one another. But bots, in Wikipedia, battling with one another.
Taha Yasseri: Yes, they do. They’re not doing much creative work there, but they do a lot. In some Wikipedia editions, more than half of the edits are coming from bots. But because they never asleep, they’ve worked 24 hours a day. And they do very little things. They fix typos, they add commas, as we continue with Wikipedia. And as we develop the technology, bots now do more sophisticated things. They detect vandalism, they even create articles based on a structured information they are fed with. As I said, we were studying conflicts among humans. And it was a very long shot for me to think maybe we should also look between if there are edit words between bots. And my hypothesis was, there wouldn’t be any because bots are not emotional. They don’t take things personal. But then as soon as we looked into the data, we realized there have been pairs of bots undoing each other’s contributions for more than three years. And no one have noticed, because no one is actually looking at bots, we trust these machines, because they’re predictable with that. Yes, they’re predictable at the individual level, to some extent. But if you have learned one thing from complex system studies is that system behavior is very different to individual’s behavior.
Noshir Contractor: How do you think that humans will play a role in brokering or mediating these kinds of arguments that emerge and don’t seem to end amongst bots?
Taha Yasseri: Ss long as we know the system, and we can predict its behavior, the good thing about sociology of machines, as opposed to sociology of humans is that we have full power, and we have all the agency that we need. Whereas if we understand the problem in a society, we might not be able to come up with an immediate solution, even if we come up with the solution, they might not be able to implement it. But the good thing is that those bots have no agency and they are serving the purpose of the owners and the society they’ve worked for. In that sense, I think things are easier. However, the difficulty comes from the fact that we have zero history of sociology of machines. We just arrived to this land and we just discovered this creatures or started to build them and embed them at every corner of our bedrooms and living rooms and the streets. We are creating the systems, it’s already a bit late to start analyzing, studying the social behavior. But as soon as we do that, and we understand how they behave, coming up with a solution and implementing it, I think should be easier than the long lasting problems we have in our own societies.
Noshir Contractor: Web science scholars have for a while thought about the web as being a social machine. And what you’re highlighting is that given that the web is a social machine, or was a collection of social machines, we need to come up with a new sociology of these social machines.
Taha Yasseri: That’s a very elegant way of putting this, that is true. The thing that I might propose to change here is to turn machine to machines. Because we have different machines coming from the fact that we have different actors. One of the sad things we learned during the pandemic was that this utopian image of a global society is not relevant. Neighboring countries blocked each other’s purchases, because of competition. When it was about the masks and the tests, and then the vaccines and so on. Therefore, the social machine of the web is not just one entity. There are competing entities. And when we saw complexity in behavior of Wikipedia bots, are very, very nice and good. I can only imagine how things could go bad and wrong when we have competing interests among not very good and not very well behaving, automated bots that are fighting for the benefits of the owners.
Noshir Contractor: I want to end by taking you to yet another issue that you have been doing some really exciting research, and that is on the topic of collective memory. One of the things that has been argued about the web in general is the fact the internet doesn’t forget. We have the archives that is allowing us to go back and look at rewind. At the same time, the European Union has to lead the way in terms of regulation that gives individuals the right to be forgotten, or at least for some of their actions to be forgotten. Tell us a little bit about how the web is able to advance our understanding of what our collective memory is, how we socially generate these common perceptions of any event on the web, and how those perceptions might change over time.
Taha Yasseri: Collective memory is not a new term, people have been talking about it at least for a 100 years. But it’s the first time we can measure it, we can put a number on it, we can look at an airline crash, and measure how many people read the Wikipedia page about this event, how many people googled it on on Google Trends data. And then 10 years later, look at the same rate and see how this number has declined over years. This is very materialistic and very operationalized, maybe oversimplified way of measuring memory. But It’s a good starting point. We have taken a similar approach, as I just described, and looked at logs of pageviews on Wikipedia and Google Search volumes, and so on, one of the first things we realize is that well, our attention is biased. We are much more attentive to things that are closer to us that are benefiting us and that are related to us. But then we also realize our memories are very much biased, to be remember past events only if they are somehow connected to us. Web science allows us to study these patterns.
Noshir Contractor: What did your research show us about any difference in generations when looking back at events in the past? Were there differences in how one generation might view a set of events compared to others?
Taha Yasseri: One of the limitations we have to admit that web science has is it doesn’t give us much of historical view. In our analysis, we of course, had data for the last 20 years, but we couldn’t say how much our results generalized 200 years ago, based on the data that we had from recent years, one thing that we could say is that are tied to time and scale of collective memory is around 40 years.
Noshir Contractor: But Wikipedia does have entries for events that happened centuries ago.
Taha Yasseri: That’s very true. And that’s exactly why we could see how people react to those events in the last 20 years, and how people reacted to events more much more recent in the last 20 years. And these are people who are using Wikipedia. What we cannot talk about is how people would have reacted to those pair of events 100 years ago, because we simply didn’t have any tool to measure their behavior.
Noshir Contractor: Exactly. Well, there are many things that web science can do and others that we may recognize our limitations, at least science at this point in time. So again, I want to thank you so much for talking with us about how the web has changed online dating or maybe hasn’t changed online dating, the extent into which algorithms may or may not be able to help us confront the challenges we faced with echo chambers — the sociology of machines, as he said about how we might be looking at bots fighting with bots, mediated by humans, and then again, how all of this shapes our collective memory, I want to thank you again for taking time to talk about this. You’ve been such an exciting scholar at the forefront of web science. And we all look forward to seeing continuing research come from you and your team of collaborators. So thank you again, for joining us today.
Taha Yasseri: Thank you very much. It’s been a great pleasure. Thank you for having me.