Model Suggests Digital Media Contributing to “Maelstrom” of Societal Division

Justin Hendrix / Oct 13, 2022

Audio of this conversation is available via your favorite podcast service.

Regular users of social media platforms are well aware that they often produce toxic discourse. But while social media executives such as Meta’s Mark Zuckerberg and Nick Clegg continue to make straw man arguments designed to undermine the contention that products such as Facebook play a role in exacerbating societal divisions, communications and political science scholars continue to produce results that bring clarity to the mechanisms by which digital and social media exacerbate partisan and identity-based conflict. A better understanding is crucial for keying in on what platforms should be held responsible for, devising better policy, and potentially designing solutions.

A new peer-reviewed paper from Petter Törnberg, a researcher at the University of Amsterdam Institute for Social Science Research, contributes to this understanding by developing a computational model that “suggests that digital media polarize through partisan sorting, creating a maelstrom in which more and more identities, beliefs, and cultural preferences become drawn into an all-encompassing societal division.”

Drawing on research on the phenomenon of affective polarization, digital media and opinion dynamics, the model indicates that “the diverse and nonlocal interactions of digital media drive plural conflicts to align along partisan lines,” and that this phenomenon may “intensify affective polarization by contributing to a runaway process in which more and more issues become drawn into a single growing social and cultural divide, in turn driving a breakdown of social cohesion.” While prior hypotheses had suggested that “echo chambers” played a role in polarization, Törnberg presents empirical evidence that digital media is in fact exposing users to a greater diversity of “individuals, perspectives, and viewpoints, often in contentious ways.”

Digital media may in this way disturb the balancing mechanism of plural societies, by pushing conflicts and cleavages to align, creating a maelstrom in which additional identities, beliefs, and cultural belonging become sucked into a growing and all-encompassing societal division, which threatens the very foundation of social cohesion.

When this phenomenon is combined with some fresh thinking on the “central puzzle” as to how the sorting and opinion dynamics can be linked to affective polarization, Törnberg sees an opportunity to thus propose “an emergent mechanism through which social media may drive affective polarization through partisan sorting.” In other words, social media doesn’t drive polarization by isolating us in echo chambers; rather, it disrupts the way we engage with people outside our “local bubble.” (Törnberg references a paper by Joe Bak-Coleman, et al, Stewardship of Global Collective Behavior, that posits the ways in which digital media is scrambling how humans interact should be regarded as a ‘crisis discipline.’)

“Coming from physics,” Törnberg told me in an interview, “I tend to think of it as positive or negative feedback cycles. But we have ended up in a situation where the political system isn't disarming or calming conflicts or somehow channeling into something constructive, but quite the opposite. It's taking conflicts, like masks, for instance, that weren't originally somehow intense conflicts, and then it manages to actually transform them into very unconstructive and very dangerous conflict.”

This happens on every issue, as what unfolds is what the paper calls “the crystallization of conflicting identities and the intensification of polarization, driven by a process in which sorting begets sorting and polarization begets polarization.” Political discourse is not characterized by differences ofideas or opinion, but rather by identity. The feedback loop on social media is vicious. This will come as no surprise to those who feel a tinge of anxiety every time they log on to a social media site, or to the armies of people who actively engage in partisan or identitarian conflict on social media every day, often in increasingly organized ways.

But perhaps Törnberg’s equations may be useful to the social media firms, assuming their implications do not conflict too much with profit. When Zuckerberg’s own researchers (whom he and Clegg curiously never cite in their arguments against social media’s role in driving division) presented him with ways to make Facebook less divisive, he shut them down.

Let’s hope others will look at the math.

- - -

What follows is a lightly edited transcript of my discussion with Petter Törnberg.

Justin Hendrix:

So tell me about your research practice. What have you worked on and what brought you to the subject of social media, digital media and polarization?

Petter Törnberg:

I have an unusual background, to a certain degree, because my studies were actually in computer science and physics looking at complex systems. So the dynamics of interaction and systems with a lot of agents. So that led me towards an interest in societal dynamics and understanding how feedback dynamics and interaction between large number of agents and different systems can lead to unexpected outcomes. And I think social media is a good example of that. And I think in particular polarization dynamics on social media is a good example of that. It's full of dynamics that are not necessarily in anyone's interests or intentions, but due to the collective effect it leads to certain macro level outcomes.

My approach is to use big data and data from platforms to study interaction using natural language processing and similar methods, social network analysis for instance, and then to try to capture these certain mechanisms using computer simulation. So basically we know that people interact in a certain way. We know that social identity has certain dynamics. So what happens when we create a system with agents that are operating under that logic? What are the macro level outcomes? And so this paper is an example of the latter, which is based on a simulation to try to capture a certain mechanism.

Justin Hendrix:

The paper you're referencing is, of course, How digital media drive effective polarization through partisan sorting, which appeared in PNAS just a couple of days ago, published to their site. Can you tell me a little bit about what literatures you drew on for this paper? There's a mishmash of ideas here and I see you maybe breaking some new ground and the way that you're able to combine some of these with the simulations that you mentioned.

Petter Törnberg:

So the literature that I'm to some degree speaking to and most clearly for me is maybe the opinion dynamics literature, which is a physics-based literature and very mathematical, and trying to understand, as I was saying, how micro interactions can lead to unexpected macro level outcomes. And so the focus there for a long time on polarization and trying to understand how polarization can occur on a societal level on the basis of individual interaction.

And in particular this sort of question of if people, normally when they interact one on one, they tend to become closer in their opinions, how can we get polarization on the micro level or can we get polarization on the micro level? So that's been one mystery that's been central for that literature for decades. And so my answer to that is to draw in this literature on affective polarization, which is basically stating that we should maybe not focus so much on divergence of opinions but rather on social identity.

That polarization is characterized, not so much about you disagreeing a lot or having very different opinions, but rather by having a strong partisan identity that you really identify as on basis of your political affiliation. And in turn it emphasizes that the reason partisan identity is becoming strong has a lot to do with a sorting of different other types of identity under partisanship. So we have various different things that become associated to our partisan identity.

And this is something, looking at the data, that people have observed over time that we feel more and more negatively about partisans from the other side. And that is linked to the fact that nowadays maybe what car we drive... if you know what car I drive you, you might know what party I vote for, what sports I watch, what football team I support, or et cetera, et cetera, that all of these things become associated to politics.

And so it's the shift in perspective of understanding polarization to what's been called an oil spill model, where more and more things become drawn into the political conflict rather than as a divergence in specific opinions. And so that's really the shift that I bring to this literature. And then I in turn connect that to this question of the echo chamber and what impact digital media has actually had on our interaction spaces, so with whom we interact.

So those are the three literatures that I bring together. And I come out with this finding that ... Well, there's actually multiple findings I try to speak to. In some ways the paper speaks all of these three literatures, making it a little bit hard to summarize. Yeah, sorry.

Justin Hendrix:

So let's speak just to the echo chamber question to some extent. Because I think listeners to this podcast will no doubt be aware that the model, or the thought around selective exposure, echo chambers, et cetera, has come under significant scrutiny in the last few years. And to some extent, I don't suppose in all cases we can necessarily dismiss it. But what you're saying here is that the evidence suggests that something quite the contrary to echo chambers is what's really at the core of the relationship between digital media and polarization.

Petter Törnberg:

Yeah, exactly. So I think in this case, this paper focuses on platforms such as Twitter and Facebook and so on, where most of the research has found that we don't have so much echo chambers, that there is in fact quite a bit of interaction across the partisan divide. And this is also my research. My previous research has also looked at this and found that we do have a lot of interaction across the partisan divide.

But I think what's interesting there is also this idea of underlying the notion of echo chamber is the assumption of what the interaction consists of. And there is this idea that what happens on social media would be something called rational deliberation, that we have a rational exchange of arguments and become more convinced by each other. And this idea that if we get a lot of arguments on one side and we get more and more convinced towards that side and we get stronger and stronger opinions toward that direction.

So it comes from, has various assumptions that I think-- in part from my own use of social media-- I would question that that is how I experience political discussions. But also in my previous research, looking at the type of interaction that takes place across the partisan divide, we don't actually see that it's characterized so much by a rational argument or an intellectual debate, but it seems more like a conflictual kind of interaction.

So that's what I'm interested in and that this paper, I don't necessarily deny that there are spaces with echo chambers. And actually some of my previous research has focused on looking at, for instance, Stormfront, this neo-Nazi forum, and to some degree that is very clearly an echo chamber. But also in that paper I questioned the idea that is what happens in echo chambers would be a collection of rational arguments in one direction. So the perspective I take, it's that is based on this identity perspective and this idea that we build up separate partisan identities. And what happens on platforms like Twitter is not that we become isolated from the other side.

It's not that we become isolated from the other side, but rather that we get thrown into a conflict which actually accelerates the polarization. We get more and more defined as separate groups, and I mean this links also to this idea of social cohesion, so it's work by Coser and so on, dating way back to the '50s about what it means to have a stable society.

So what does it mean to have a society that's not polarized? And Coser, it doesn't mean that you don't have conflict. Every society contains conflict, every society has differences between different groups. But the question is rather how those conflicts play out, so whether there's crosscutting conflicts or not.

Basically, the way that he tells it is that if you have crosscutting conflicts, then you get a stable web of disagreement. So maybe you and I vote for different parties, but we support the same football team, or we agree on certain issues and then we will sometimes be on the same side. Then it leads to us having a sort of interpersonal understanding, which means that the society as a whole becomes a stable patchwork of conflicts.

Justin Hendrix:

I want to ask you a question about race and how that figures into your thinking on this. When we think about polarization or even affective polarization, I mean, race is a sort of concept that plays an important role in these identity-based concerns, etc. And certainly we know, just as you just said, not all polarization is bad.

The Civil Rights movement in the United States, other very just movements for social justice, racial justice, around the world have been very polarizing in their societies. How do you think about that?

Petter Törnberg:

I think it's a very good point and I think it's interesting, because we tend to look back at the '50s and '60s as less polarized eras or eras of more functional politics to a certain degree. But I think as you raised, at the same time those were periods of race riots and violence in the streets.

It was to a certain degree, the reason that mainstream politics in the US, for instance, wasn't as polarized as it is today, was that a lot of issues and a lot of existential issues weren't led into politics. So to a certain degree, I would say that contemporary politics, it's an improvement that these issues are being raised. Of course, the result is to a certain degree an intensified polarization, because we are discussing to some degree the rights of certain people to be included and to be people and to be given existential rights.

I completely agree that to some degree there is a problem in the discourse around contemporary polarization as being something completely bad or completely problematic or just dreaming back to the politics, the golden age of the politics of the '50s and '60s, because I think that was less polarized because it was based on exclusion. The question is whether it is possible to organize a politics of today where those issues are brought in.

But at the same time, it doesn't lead to a sort of breakdown of the cohesions of society. I think, in that case, looking around the world, I think that there are societies other than the United States that have managed to construct political systems that are more capable of containing societal conflicts than the US is. I'm sitting in Switzerland right now, and I think it's a good example of a society that is capable of containing quite large amounts of diversity and conflict.

Justin Hendrix:

One of the things I've arrived at in my head, and I don't know if it's right, but it feels right to me, is the thought that certainly there are certain types of conflict that very much need to play out and civil rights and social justice, racial justice, are conflicts that normatively, I've chosen sides. I hope most of my listeners have, and we need those conflicts to play out.

But that to some extent, if the entire circumstance of the social discourse is affectively polarized or moving in that divisive way... you use the word 'maelstrom.' You say that digital media contributing to a maelstrom, in which more and more identities, beliefs, cultural preferences have been drawn into all-encompassing societal division.

If that's the case, if that's what our situation is and the media ecosystem is contributing to that, I imagine that that actually hurts the chances that we might be able to have quality and somehow successful dialogue on some of those existential issues.

Petter Törnberg:

I think that this is precisely true. The way I think about it is that the role of the political system should be to channel these conflicts into a more civilized or a more debate-based arena, and to some degree, disarm conflict. I think that's what we see in polarizing societies, like the US, is that it starts having the opposite effect.

Coming from physics, I tend to think of it as positive or negative feedback cycles. But we have ended up in a situation where the political system isn't disarming or calming conflicts or somehow channeling into something constructive, but quite the opposite. It's taking conflicts, like that masks, for instance, that weren't originally somehow intense conflicts, and then it manages to actually transform them into very unconstructive and very dangerous conflict.

I think that that's the type of dynamics that we're seeing. And I do think that social media is part of that landscape, because of its emphasis on engagement based or attention economies. And because of these dynamics, where it's trying to highlight conflict and it's pushing people together from the different parties and highlights the most conflictual and most strongest level of disagreements, then we see it feeding into this system where it's actually making these conflicts worse rather than better.

Justin Hendrix:

You also point to, I mean, there are over a hundred references in the paper, which folks should check out, but one that I spied in particular is a paper called Stewardship of Global Collective Behavior by a variety of researchers, I think almost a couple of dozen researchers, but the lead author is Joe Bak-Coleman, who was then at the University of Washington, now at Columbia.

I also found that paper fascinating for what it suggests about a research agenda that we should have around the role of digital media and society. Why did that paper in particular chime with you? I get the sense that you're interested in complex systems, you're interested in these underlying mathematical relationships. I don't know, what are you thinking there?

Petter Törnberg:

Complex systems is very much my background and where I come from and a lot of my work has actually been more on the philosophy of complex systems and so on. I think the paper, it hits a mark for me in the sense that it combines my interests of how I think about digital media and how I think about society as a complex system. I think that it captures this, the sense that I have that we need to understand social media as an example of a mass interactive system. And often with those systems, it's very hard to predict the outcomes that changes to local rules we'll have. Because you have self-organization. You have these dynamics where small notifications on the local level can lead to very unanticipated macro level outcomes. And so, that is the type of perspective that I bring into trying to study this. And I don't necessarily think that anyone is trying to actually to create this situation. It's not like Twitter wants to create a polarized society that it's rather they're trying to operate, and they can modify rules on a local interaction level. They can modify their algorithms. And they also don't know what outcomes this will have. And so, what that paper resonated with for me was this idea that we need to use computational methods to try to track these really long chains of causality that are producing these unanticipated outcomes.

Justin Hendrix:

So in some ways, the platforms themselves are a massive experiment on society, and we don't really know what the variables are, how they interact, or certainly we no longer have a control.

Petter Törnberg:

Yeah, I think that's exactly it, that we are really experimenting with forces that we do not fully understand. I think Hannah Arendt writes about this very beautifully in The Human Condition about how with the rising capacities of technology, we have attained the power to really change the world, but we don't understand our use of that power. And I think that ... I mean, she spoke of the nuclear bomb, and so on. But I think that it applies equally to the powers that we have in terms of creating communication, and information ecosystems that are transforming the very conditions of our own existence.

Justin Hendrix:

You mentioned that to your mind, social platforms aren't setting out to produce polarization on purpose. But let me just query you in particular on Facebook, which to some extent has a different record in the last couple of years on the question of polarization, I'm sure you're familiar with some of the writings of a senior executive there, Nick Clegg, on this subject with statements that Mark Zuckerberg has made.

Let me just read you something that Nick Clegg wrote in his medium piece a couple years ago. You and the Algorithm: It Takes Two to Tango is the title of it. He has this section on polarization where he says,

"But even if you agree that Facebook’s incentives do not support the deliberate promotion of extreme content, there is nonetheless a widespread perception that political and social polarization, especially in the United States, has grown because of the influence of social media. This has been the subject of swathes of serious academic research in recent years — the results of which are in truth mixed, with many studies suggesting that social media is not the primary driver of polarization after all, and that evidence of the ‘filter bubble’ effect is thin at best."

I point the listener to Clegg's statements. He's made others about the role of social media. There is to some extent this idea that social media is a mirror to society, that the divisions were there before social media existed, which is certainly true, and that it's the evidence that it's exacerbating the situation is somewhat thin. On some level, your research bears out part of Clegg's argument, that the evidence of the filter bubble effect is thin. But would you agree with him, having done this paper, that in fact social media's role in this particular effect in society is negligible or still a question?

Petter Törnberg:

I mean, so there's two things. There's first, the question of whether the longstanding debate of whether social media is causing polarization or not. And to some degree, I think that there's quite some evidence in a lot of senses, and I think that we can all of ... I think most people have this intuitive sense that it is the case. And most researchers on radicalization, and extremism do have the sense that this is the case. But it is a little bit like this old debate about whether television causes violence or not, that we used to will never know how society would be without media. But we cannot really think of society and our media technologies as being separate. They're very much part of the same system. So, in that sense, a conclusive statement about whether it does or does not is not possible to answer completely scientifically.

But I think that there's also an interesting slip there that we often see in these discussions, where Nick raises the fact that there's limited evidence for echo chambers, and then draws from that to the conclusion that social media does not cause polarization, and you often see that argument playing out. And of course, it kind of misses a step because there might be other mechanisms. And indeed, my paper suggests or shows a very different mechanism through which social media might cause polarization, which does fit this evidence in terms of the existence that echo chambers don't exist so much on the mainstream platforms, such as Facebook. And the interesting part there is also that the way that after the 2016 election, the way that Facebook responded to the accusations of Facebook causing polarization has been to try to create more interaction across the partisan divide. And I think it's something that we saw across multiple platforms. And of course, what my paper suggests is that that can have precisely the opposite effect of actually intensifying conflict and intensifying polarization.

Justin Hendrix:

So, one of the things that seems to me to be a consequence of this Identitarian behavior that you describe is in fact misinformation, that as a result of our prosecution of our identity on social media daily, and many people that I talk to literally feel like sometimes they're getting up, and they're suiting up for war on their phone, particularly in intense political moments ahead of a debate ... or sorry, ahead of an election or what have you. And that creates this feedback loop in people's behavior where maybe they are willing to hit retweet on something that is spurious, or that just asks questions, or raises some complaint against the out party in order to, again, prosecute their identity. Do you see that relationship?

Petter Törnberg:

Yeah, for sure. I think that there's a very strong link between polarization and misinformation. This is also an argument that I worked on in previous papers. But coming from this idea that opinions, facts, the entire world of narratives become linked to who we are. So this sense that becomes part of an expression of belonging to a certain group. As a result, the facts and the reality becomes maybe a little bit less important in terms of what opinions we develop, what views of the world we develop. There becomes a sort of sense in where our perspective on the world becomes drawn in to just becoming another expression of our identity. I think fake news is an example of that, and this is also something that I've seen in studying Stormfront, the neo-Nazi forum, where I used natural language processing to look at the language use over time. You see the sense that there's a co-production of an identity, of a sense of we that is taking place through language and through storytelling. Of course, it's a neo-Nazi forum so it becomes extra clear, the sort of madness of it, but I think a similar dynamics I can very much see playing out also in the political mainstream.

I think that there's also the sense that I've had studying both extremists, radicalization, and studying mainstream politics, that the dynamics to certain degrees are becoming more similar. That because radicalization research has long spoken about how people become drawn into a certain identity and certain worldview. They see everything from that angle and that the identity takes over their life, it takes over how they see the world. I think that we see a little bit of that starting to happen also in mainstream politics in certain ways, that the radical politics and mainstream politics start to a little bit obey the same dynamics. I think that that is something that is really quite worrying.

Justin Hendrix:

You write about what you call the crystallization of conflicting identities and the intensification of polarization driven by a process in which sorting begets sorting and polarization begets polarization. So a pretty vicious feedback loop. How do we get out of it? I mean, one of the things that occurs to me in reading this paper and many others that we've discussed on this podcast, but also in other contexts, is that most of the significant interventions that people are aware of or that the companies themselves have invested in that we're aware of are aimed at this rational model. They're about fact checking, they're about provision of right information ahead of elections. They're about labels, they're about other means to intervene at that level. I mean, I don't want to put words in your mouth, but to some extent you're saying maybe that's all well and good, it may have some efficacy, but it's not the underlying issue.

Petter Törnberg:

No, I think that that's very much the case. There have also been quite some studies, as you probably know, showing that pointing out the actual facts is not a very useful strategy in terms of convincing someone who has a more extreme opinion. It's seeking to operate on the sort of rational level and trying to rationally convince someone who is not engaged in a rational conversation because identity is to a certain degree, a different realm. It operates in a different logic, and so I mean, the question of what to do about it. I think coming from a sort of social identity standpoint, there's the context hypothesis, which has been a dominant theory for how to reduce group polarization.

So that's very focused on being brought together under common goals and so on. So the military, for instance, has been historically an institution that was used as an example of something like that. For instance, between Black and white Americans, the Second World War had a very strong effect on reducing intergroup conflict. So the examples that are being raised to reduce the type of group conflict would be something like a few years ago, the example that would most commonly come up is something like a global pandemic, but obviously that was apparently not efficacious in reducing the polarization. Now we unfortunately get to see the effects of a war. So basically, it's not very easy to see a sort of escape from the dynamics from within the system itself because there is a tendency that anything you do to try to address it becomes drawn into the dynamics itself. So, that makes it a very pernicious dynamics that become very difficult to escape.

Justin Hendrix:

I can think of some prominent examples. Look at the, I don't know, in this country, the continued argument over things like the way that the lab leak hypothesis was moderated by the platforms. Or the way that the platforms, in particular Twitter and Facebook, took some action against a New York Post URL concerning an article about Hunter Biden in advance of the 2020 election. Those were actions that were taken on some level because the platforms believed that they were addressing a kind of rational information problem. Yet, those actions themselves are pulled into the polarizing dynamic of the political discourse and have become totems, certainly on the right here, for why the entire social media landscape is an illegitimate playing field.

Petter Törnberg:

Yeah, no. I think it's very much the case. Also, that part of the polarization is this sense that the mainstream institutions have lost legitimacy. That the political institutions have lost legitimacy, and of course then it becomes very hard for those same institutions to come in and try to calm down or address the situations while being perceived as unbiased outsiders to the conflict, which they clearly aren't. I think that it's really quite interesting in the way that the polarization has made to a certain degree visible things that have always been true in the sense that society, that our societal institutions are implicated in our politics.

They are actors, and somehow media institutions have always had a powerful role in shaping societal discourse, but just to some ironic sense that this has become more visible through social media, by social media allowing the interaction and allowing the debate. Which brings this up, that traditional media wouldn't have a debate around its own legitimacy, but now social media does allow us to debate the legitimacy of the media platforms that we are participating in. So I think that that dynamic is something new that I think makes for a very difficult situation.

Justin Hendrix:

A maelstrom indeed, and I assume that you'll continue to study this problem and hopefully we can figure out collectively how to chart a path forward. Thank you very much.

Petter Törnberg:

Thank you. Yeah, this paper is actually part of a book that I'm writing, and as you know, the last chapter always has to make some positive suggestions. So I guess that's what I'm working on right now.

Justin Hendrix:

Thank you, sir.


Justin Hendrix
Justin Hendrix is CEO and Editor of Tech Policy Press, a new nonprofit media venture concerned with the intersection of technology and democracy. Previously, he was Executive Director of NYC Media Lab. He spent over a decade at The Economist in roles including Vice President, Business Development & ...