Nabeel Gillani, Ann Yuan, and Martin Saveski, MIT Labs, Authors of Me, My Echo Chamber, and I: Introspection on Social Media Polarization
Nabeel Gillani, graduate student and research assistant in the Laboratory for Social Machines at the MIT Media Lab, using data science/machine learning and software engineering to try and promote equality of opportunity in education.
Prior projects include mapping political tribalism and information bubbles on social media platforms like Twitter and experimenting with tools to help build bridges between politically and culturally disparate groups of people.
Ann Yuan, is a software engineer at Google. At the MIT Media Lab Ann worked alongside data scientists, designers, and journalists to create data tools and visualizations that help people make sense of cutting-edge NLP and machine learning research
Martin Saveski, is a PhD student in the Laboratory for Social Machines at the MIT Media Lab advised by Deb Roy. He loves studying data and people. His research interests lie at the intersection of Causal Inference, Machine Learning, and Social Network Analysis. He is passionate about (i) drawing insights from data, (ii) designing interventions, and (iii) running randomized experiments. He spent the summer of 2016 at LinkedIn working on randomized experiments in networks.
Prior to coming to MIT, he interned at Yahoo! Labs in Barcelona and the Amazon Machine Learning team in Berlin. He spent one year in Paris and one year in Barcelona to do a M.Sc. in Data mining and Knowledge Management. He received his B.Sc. from Staffordshire University with a First-Class honors in Computer Science.
Season 2, Episode 2
Although there is a lot of conversations about the role of social media and polarization, few have investigated it as this team from MIT Labs. Using Twitter and a Social Mirror experiment, we explore what this team learned about how social media polarization happens and what we can do about it.
Join the conversation with Nabeel Gillani, Ann Yuan, and Martin Saveski, MIT Labs, authors of “Me, My Echo Chamber, and I: Introspection on Social Media Polarization.”
Listen to more: Activate World
Social Media Polarization: What Happens? What Can We Do?
Nabeel Gillani, Ann Yuan, and Martin Saveski, MIT Labs
Season 2, Episode 2
Jon Mertz: 00:02 Welcome to the Activate World Podcast, a series on how business leaders have more power to solve societal issues than any elected official. We explore business activism with substance and depth of thought. In this episode, we’re excited to have a great team joining us to explore social media’s role in polarization. Our guests are the authors of a research paper entitled, Me, My Echo Chamber and I: Introspection on Social Media Polarization. We welcome the research team here from MIT Labs. For each of the three, introduce yourselves, tell us a little bit about your background and your role in the research.
Nabeel: 00:41 Sure, I can start. Thanks for having us, appreciate the opportunity to speak with you also. My name is Nabeel and I’m a PhD student here at the MIT Media Lab and the Lab for Social Machines. I have a background in data science, machine learning, and I’ve been working with several members of the team including Ann and Martin who are here today, to try to better understand polarization on social media and explore interventions to see how we can potentially mitigate the issue.
Jon Mertz: 01:10 Great. Martin?
Martin Saveski: 01:12 Hi Jon. I’m Martin Saveski, I’m also a PhD student at the MIT Media Lab. I have a background in computer science, in my past research I have spent a lot of time developing machine learning methods that actually made it for recommendation, building recommended systems. But since I moved to the Lab I got very interested in how those systems actually influence people’s behavior online, and in general how people connect with each other on social media, which is what started my interest in the work that sort of led to this project, looking at polarization online, especially during the election.
Jon Mertz: 01:56 Okay, great. Ann?
Ann: 01:58 Hi Jon. I’m Ann, also I was a student at the Lab for Social Machines with Martin and Nabeel. I just finished the master’s program, and my background is in software engineering and I’m pretty new to machine learning but this project was sort of a fun introduction to ways to apply some of the methods we’ve developed in the lab.
Jon Mertz: 02:21 So what sparked this interest in social media polarization?
Nabeel: 02:25 I can start and then Ann and Martin please jump in. Our lab had been working on this project called the Electome, which was basically an attempt to monitor the public discussion about the 2016 US Presidential Election as it unfolded on Twitter. So the project started with mapping the topics that are being discussed, getting a sense of what’s the relative share of conversation on Twitter, guns versus immigration versus some of these other issues occupy, and pretty soon after that, once we started mapping the topics, we pretty soon went to actually mapping the social networks of people who are doing the talking. And, in particular, we started to look at how did people who are discussing politics, and in particular the election on Twitter, how are they actually connected to each other? Who follows who? And how do those connections differ according to the candidates they support, or the political parties that they’re affiliated with?
Nabeel: 03:20 And as you can imagine, what we found was the political discourse landscape on Twitter and social media was highly polarized, and this is something that several other researchers and other folks have been talking about for a while. But as we started to look at these social graphs where you have a lot of Trump supporters who are primarily connected to other Trump supporters, Clinton supporters who are really mostly connected to other Clinton supporters and not really talking to Trump supporters, and even Sanders supporters that are in the mix talking amongst themselves and very little to people outside of their own political tribe. A natural question we had was, “Is there anything that we can potentially do? Is there any way that we can intervene to see if we can sort of chip away at this issue?” I mean, it’s obviously a big problem and I don’t think any of us thinks that we have this silver bullet. But is there any way that we can use software and some of the same tools that might have created or intensified these divides, to see we can bridge them in some way.
Nabeel: 04:15 And so, one hypothesis we had … And there’s a family of interventions we’ve tried and Martin’s been working on. Another one that we could chat about as well but this one, in particular, for this paper, one hypothesis we had was, “Hey, you know, as researchers we’re able to build these social networks and analyze data, and really have this bird’s eye view on how polarized the social media, social landscape really is”. And one question we had was, “What if we actually give people that view? What if we could somehow pull people out of their very local experience of social media where, when you’re on Twitter you really only have a view of what your immediate connections are sharing with you, but if we could sort of zoom you out and give you a different perspective on this space you occupy in this broader social media universe, and in particular how tied that echo chamber that you exist in is, how you really live in a very small part of the network, could that affect, one, your beliefs about who you’re connected to, how homogenous the information you consume really is?
Nabeel: 05:19 And then two, possibly even motivate you to try to get out of your echo chamber, to start to try to follow people who’d you follow at different, or share different perspectives, and have different political opinions.” That was some of the motivation behind this particular study.
Martin Saveski: 05:35 Yes, that is right. Yes.
Jon Mertz: 05:37 In your research, you focused only on Twitter, correct?
Martin Saveski: 05:39 Primarily on Twitter, yes. We have now recently started looking at other sources, like talk radio but that’s still in early stages but this project in particular was mainly focused on news published about the election on news websites, news outlets and Twitter and how the two intersect.
Jon Mertz: 06:02 As you dug in to start this research, how did you structure it and set up those experiments around social media polarization?
Ann: 06:08 Yeah, so actually that was one of the more challenging components of the experiment was basically reaching out to people so that they would actually use the tool. I don’t know if you had a chance to explore it or not but it’s … The tool basically is designed for people who we sort of identified over the summer as being active participants in the social media discussion surrounding the 2016 election. And so if you were part of that cohort of individuals then we sort of … We had designed a separate experience for you, so we really wanted to target the people who we already identified and so in order to do that, we simply direct message people on Twitter. Yeah, which was … It was successful in some cases, a lot of people were really receptive and gave us really good feedback about the tool and engaged with it, but also a lot of people pushed back.
Ann: 07:05 First people were skeptical about this new account on Twitter reaching out to them for any reason at all, but then also a lot of people reacted negatively to the tool itself because they … Some people challenged the premise of the tool, which is that you should be engaging with people who you might not ordinarily come into contact with, and we got to a lot of pushback on that premise, which was actually really interesting to explore everyone’s emotional reactions to our request. Some of that we discuss at the end of our paper. So yeah, in terms of recruiting users, it was really just a brute force, direct message approach.
Jon Mertz: 07:48 How did you solicit the different participants? They are on Twitter correct?
Ann: 07:52 Yeah. So the ones who agreed to participate, they would then click the link that we included in our message and people would go through the experience sort of to varying degrees. The entire experience has multiple steps, not everyone made it all the way through. But the ones who did are the ones, whose data we considered for the study.
Jon Mertz: 08:12 So as you collected the data on the ones who stayed with the experiment, what were some of the things that you found?
Nabeel: 08:17 Basically, we invite people to use this tool, the Social Mirror tool, so they sign on with their Twitter account. And then what we do is, we walk them through an experience that shows them first their own … The goal is to show them their position in a polarized social network [inaudible 00:08:35] a subset of their social connections on Twitter. And so, we start by showing them just their own … a node that represents them, and then sort of the next step is we show them their immediate connections and then after that we sort of lay out the network itself, I think probably out a couple degrees from their position in the network. And so, by the end of it you have this 3D network visualization that you can then sort of rotate, you can explore, you can get a sense of what were people talking about, people in this part of the network what were they talking about? People in this part of the network what were they talking about?
Nabeel: 09:08 And the idea there is to really try to orient the user to this sort of social topology so to speak that we have sort of introduced them to. And then what we ask them to do, is we ask them to try to guess which node is that. So up until now you had a chance to explore, you’ve probably started to get a feel for, “Hey, people in this part of network are talking about stuff that I don’t agree with at all, and people on this part of network seem like my crowd or my kind of people.” And so, at that point we ask people to sort of guess which node represents their account. After they guess, we show them … we do this, the reveal, and we show them which node actually is there, sort of how far off, how many hops away, network hops away were they from their actual guess.
Jon Mertz: 09:50 Surprising that, right? When people saw where they fit into the typology.
Nabeel: 09:54 That’s right, yeah. I mean people as you can imagine are not very good at guessing sort of where they were. And so there’re certainly I think people who … We did look at exact numbers and how far off they were, but I think on average probably at least a couple degrees away from where they thought they were, which was actually pretty interesting. So then after that, I think is where the core of the experience comes in. Once we show them their actual position, we actually then show them a score that quantifies their political diversity of their connections. And so, if they get a score of 0%, what that means is that, they’re connected people of a single political ideology, so all of their connections are left leaning people, and all of their connections are right leaning people. And a score of 100% means that they actually have a pretty good balance of left and right connections in their neighborhood.
Nabeel: 10:40 So for some people that’s where the experience ends, and we’ll talk about sort of what … And then we do like a post test and talk about some other results. For another set of people, we add this additional step, where we then actually made suggestions for accounts that they can follow if they want to increase the diversity of their connections. So the idea is we’re trying to nudge people into basically changing their Twitter graph, deciding to follow people that might introduce new perspectives or opinions into their social media diets, and so for some people we actually make their suggestions and then after that we sort of observe what effect that has on the diversity of their network connections moving forward.
Jon Mertz: 11:18 Was there a median where most people fell? Either having a pretty diverse network or not a diverse network at all?
Nabeel: 11:23 Yeah, it’s a good question. I mean the median was definitely not 50%. The median was much, much lower which basically means that most people’s connections were pretty much homogenous.
Jon Mertz: 11:34 When you made suggestions, the people that would make their network more diverse, what was the reaction to that?
Nabeel: 11:40 So for that first batch of people that we didn’t make account recommendations to, we actually found that through the experience people seem to be more likely to admit they’re in an echo chamber. So something about this experience of zooming out, exploring … seeing yourself as one small piece in this broader sea of connections and realizing that you probably are in a pretty polarized part of the network, but what’s interesting is that those people didn’t really seem to do anything about it, meaning … One thing we did was we actually measured one, two and three weeks after these folks used the tool, we actually looked at who they’re connected to on Twitter and we computed the diversity of their network to see if there’s any change after they used the tool.
Nabeel: 12:23 And so, while these people seemed to admit that they’re in an echo chamber, they didn’t do anything about it, they didn’t seek out more diverse accounts to follow, they didn’t really seem to have any change in the kinds of political information they’re consuming week over week. Now it’s interesting, and this is what was counterintuitive for us, was that second batch of folks, the people that we actually did make account recommendations to, those folks actually did end up following more diverse accounts afterwards, and they weren’t the accounts that we recommended. Now, unfortunately those effects didn’t last for long. So when we computed the diversity of the second batch’s neighborhood one week after treatment, we saw this change, this significant difference in political diversity, but then when we continued to measure two and three weeks out, that effect diminished. It went away.
Jon Mertz: 13:16 So as you look at the midterm election coming up, is there a way to keep that nudge in place or sustain the diversity of views?
Nabeel: 13:23 Yeah. We’re all stumped. What we’ve talked a lot about is the goal to change people’s minds or attitudes and we’ve always arrived at, “No, like that’s not the goal. We’re not trying to manipulate people, we’re not trying to brainwash people with these interventions.” I think with Social Mirror A goal was, definitely exposure to diversity but also kind of exposure to your own behaviors, sort of this like self-reflective, self-educative process of learning more about what you do consume.
Jon Mertz: 13:53 Is there something more that social media companies could do to encourage that self-reflection in their social media platforms?
Martin Saveski: 14:01 In some sense in my view this is like having a fitness tracker that tells you how much you have actually walked today, or maybe counting the number of calories that you have consumed today and whenever we have a problem, sort of behavioral problem we assume that, if people only knew about how much they eat, how much they exercise or how their social networks look like, it will change their behavior. What this experiment tests is whether that’s really the case and to what extent, basically showing people this kind of information will work. And as Nabeel mentioned, we see some change but not too much.
Martin Saveski: 14:44 I think our job as researchers is to really come up with prototypes that would show that some of these interventions may work but this was sort of an experience that was separate from the platform itself, but hopefully will be a motivation for other platforms to think about similar experiences that they can actually make part of the platform itself, that can keep the users informed over time. And then in some sense nudge them from time to time, to make sure that there are more long-term effects, rather than just things wearing off after a few weeks.
Jon Mertz: 15:22 If I want to get more reflective about who I’m following, is Social Mirror available to me to tap into and use?
Martin Saveski: 15:28 The website is socialmirror. media.mit.edu.
Jon Mertz: 15:32 So what role can journalists play in trying to diffuse some of this polarization? What can social media companies do?
Martin Saveski: 15:39 We have thought a lot about it. I would say one thing that we have been thinking a lot about is polarization [inaudible 00:15:47] somehow your question, it seemed like it’s something we’ll have to mitigate, but very often polarization can be actually good and can allow some ideas that would not be mainstream to develop into something that would be very important, basically take over society as well. I’m trying to think of an example but often a lot of issues or opinions that are now mainstream, started really from small circles of people talking to each other and exchanging opinions on that topic before the ideas are actually developed enough to be shared or maybe adopted by others in the [inaudible 00:16:28]. So in some sense polarization is not necessarily good or bad.
Nabeel: 16:34 What’s interesting is … Building on Martin’s point in your example Jon. Marshall Ganz made this interesting distinction between polarization and activism, where polarization often ends up being about the person, so at some point things become ad hominem and they lose the essence or the focus on values that activism often has. So when you think about activism in its purest, best form, it’s not about the individual actors or people involved, it’s about the essence or the values that you’re trying to espouse and trying to push and I think that … We think about that is a distinction. What happens oftentimes in political discourse is the conversation becomes too much about individual actors or what they said or what they did or what they didn’t do, and those are important but I think the real challenge is to somehow focus and point the conversation towards values and towards the essence effectively of whatever it is that we’re trying to promote or achieve.
Jon Mertz: 17:30 So maybe focus more on the change rather than the actors?
Nabeel: 17:33 That’s right. And I think that’s still going to lead to factions at times, or it’s still going to lead to groups. But I think to our points earlier and to Martin’s points, groups are not bad, I think groups we’re born in communities, we live in communities, we form groups. I think the challenge though is going back to what you said Jon, how do we have groups that still know how to communicate in a productive way with each other?
Jon Mertz: 17:55 Is there something that social media companies can do to help activists movements and facilitate change, or are there design elements that they should be taking a closer look at?
Nabeel: 18:04 The lab for social machines like the name itself indicates, “Yeah, we use machine learning and we use these other tools but ultimately the question is how do we create these human machine symbiotic systems where each component or each person or each part is playing its role in order to achieve some greater good?” I think when we think about what social media companies can do, and how [inaudible 00:18:30] with business models I think is another question. But how can social media companies leverage the power of digital, but then not make the experience end at digital, like how do we actually use maybe digital as a starting point to then spark action in the real world.
Nabeel: 18:44 So Pokémon Go [inaudible 00:18:46] fascinating, it’s a great example of something that I think actually did this very well, which is you’re using the digital realm to I guess augment reality to some extent, and to do what it’s good at which is maybe simulating new realities or envisioning new realities but then at the end of the day that experience is very much felt in the real world with other people.
Jon Mertz: 19:07 It still had a digital aspect to it obviously, but it also had a very strong, “Get out in your community and find these things” element too. So how if you noticed your own digital habits change since the research has been completed.
Ann: 19:19 I mean, I imagine that the research I’ve been involved with is a little bit responsible for the fact that I’ve tried to listen to podcasts and read news, news organizations that are [inaudible 00:19:31]. When I came to the lab I was like very traditionally liberal and I only read The New York Times. I only listened to very squarely liberal podcasts. And we talk about this stuff so much that I thought, “I really ought to try to listen to at least one conservative commentator.” So I’ve managed to do that and that’s been a great experience. Obviously, I’m just scratching the surface here. I should try to make some conservative friends, there lot more that I could be doing but my digital habits have definitely changed and I think that some of the research that I’ve been involved with is at least partly responsible.
Jon Mertz: 20:15 That’s great. How about you Martin?
Martin Saveski: 20:18 I have actually started to restrict my social media usage. It sort of relates to the question that you asked previously, like how I actually feel that social media platforms are really trying to get us addicted to this and maybe it’s not directly related to polarization but I’ve started restricting myself to 10 minutes of social media per day. And I’ve deleted all the social media apps from a phone, and I time the time that I spend on Facebook and Twitter. That is challenging sometimes because I do research on Twitter so I have to open the website a lot, and that counts for my time, but that has been one effect.
Martin Saveski: 21:00 But beyond that, I think actually these few projects that we’ve [inaudible 00:21:05] really made me reflect myself about the choice of people and organizations that I have decided to follow and actually how the … sort of I have my small mini-world that I get information from. And being able to see this from sort of more higher level, helped me realize that actually there is a lot out there that I’m not even aware of. And so, at least made me a little bit more humble and challenged a little bit my views, because I think I’m sort of operating under different information basically, and I’m drawing my conclusions from a very curetted set of sources.
Nabeel: 21:51 I think one thing that, just looking at some of these homogenous networks has made me do is also just question the homogeneity of my own social networks and not just online but even in the real world to the point of when I’m in a certain setting like I think it’s nice if I am able to remember, “Hey, like everything that I’m seeing is like one small part of everything I could be seeing.” There’s a lot of other stuff that’s out there and there’s also a lot of groups of people and types of connections that are out there that I know that I’m not experiencing right now, because I’m in my own cocoon.
Jon Mertz: 22:26 What’s your best advice for citizens as we get more into this election season?
Nabeel: 22:30 I guess just, even when I feel like I’m right or I completely understand something, just maybe try to hedge that a little bit or just keep this healthy skepticism or doubt in my mind, because I think that question of, “Hey, what if I’m wrong?” Or like, “What if I really don’t understand this issue or this candidate or this person as well as I think I do?” I think that oftentimes leads to the kinds of information seeking and exploration that … And I guess humility in its best form that could be productive in these settings.
Martin Saveski: 23:02 If I was to summarize that in that sentence, I would say if you disagree with somebody or some views or issues, try to understand why the people who hold the opposite opinion, why do they hold it? Why is it that you disagree with them? Where are they come from? I think that helps a lot.
Ann: 23:20 Someone told me once that when someone convinces you that you’re wrong, it means that you learn something and I felt after I heard that I thought might be … It was sort of a nice way to think about being [inaudible 00:23:36] I guess.
Jon Mertz: 23:38 Well, I really appreciate your time. The research that you’ve done is sparking the conversation, I hope. I know this conversation here has been very helpful for me as I think through this as well.
Nabeel: 23:47 Thank you so much for having us.
Martin Saveski: 23:48 Yeah.
Ann: 23:48 Thanks.
Jon Mertz: 23:55 Listeners, we’d love to hear from you.
- Do you make an effort to stay open to other points of view?
- How have you successfully navigated differences of opinion on social media?
Send your perspective to me at Jon @ activateworld.com. That’s Jon without an H, J O N @ activateworld.com. Write it out or record it, send it my way. We want to hear and share your thoughts. Activate World podcast. Encourage them to subscribe, listen, and share from their favorite podcast platform, Apple, Google, Spotify, RadioPublic, and others.
Let us know howwe’re doing by leaving a review. Your reviews mean a lot to us. And join usnext time as we talk with Susy Schultz, president of public narrative. We’ll explore journalism and democracy.
Activate World is a team endeavor, special thanks to Kaela Waldstein and Kent Nutt. Music by Jason Goodyear. For Activate World, I’m Jon Mertz.