I would like to stage a fight between two different accounts of the current political landscape—what’s been called the “post-truth” era, the infodemic, the end of democracy, or perhaps most accurately, the total shitshow of the now.

According to one oft-told story, what’s going on is systemic polarization. Our once-peaceful society has been riven into polarized camps. Extremism and political separation are the core problems, and the fix is something like reconnection, intermingling, and friendship across party lines. (The sound of this story is somebody issuing a plea for civility “in these divisive times.”) According to a very different story, what’s going on is propaganda. Certain bad actors are generating false and misleading information for political purposes. To fix it, we need to fight those bad actors.

There are two different accounts of our current political landscape: polarization and propaganda. Which is the best explanation?

Systemic polarization, as it is usually told, is a basically symmetrical story. Polarization arises from a social dynamic that afflicts almost everybody. The social forces at play—social mobility, online media bubbles, algorithmic filtering—are pervasive, and their effect is nearly universal. Like-minded individuals naturally clump together and end up boosting each others’ confidence unreasonably. Conservatives and progressives are approximately as vulnerable and approximately as blameworthy. On the other hand, the propaganda story is usually told asymmetrically: one side is stuck in the propaganda machine, the other side fighting against it. It is certainly possible to tell the propaganda story about both sides, but symmetry isn’t baked into its core.

Which story, or combination of stories, is the best explanation for what’s going on? To make a first stab at an answer, I want to compare two recent books that make particularly compelling and empirically rich cases for their respective stories. In the polarization corner we have political philosopher Robert B. Talisse and his book Overdoing Democracy: Why We Must Put Politics in Its Place (2019), which stands in the long shadow cast by Cass Sunstein’s empirical work on the polarization of like-minded groups. In the propaganda corner we have Network Propaganda: Manipulation, Disinformation, and Radicalization in American Politics (2018) by Yochai Benkler, Robert Faris, and Hal Roberts, a team of researchers whose expertise lies in the empirical study of political and online life, especially the network topology of social media and mainstream news.

Both books take as their starting point the political landscape around the 2016 election of Donald Trump. And in the opening pages of both, you can hear the shock and intellectual disorientation in the writers, as they struggle to come to terms with what, exactly, has happened to American democracy.


The polarization story

Overdoing Democracy offers a sophisticated version of the systemic polarization story, one that takes off from Sunstein’s and Bill Bishop’s pictures of socially and technologically driven polarization. According to Bishop, the last few decades of American life have been marked by a massive geographical self-sorting. It has gotten easier and easier for people to pick up and move. And we mostly like to fit in, so we have ended up clustering—conservatives with other conservatives, liberals with liberals. This self-sorting leaves most of us living and interacting mostly with like-minded people: we constantly hear our own beliefs affirmed and rarely encounter our political opposites. Sunstein adds a technological flourish to the story. Social media, he says, enables even more frictionless self-sorting, which accelerates the process.

According to Talisse, there were once plenty of public spaces where people of different political inclinations mingled and could form some sort of connection. Liberals and conservatives might not actually become close friends, but they could come to see each other as good parents at the Little League game, or as charitable citizens while working together at the soup kitchen. In such contexts, Talisse says, it is easy to form “civic friendship.” We might not become intimates, but at least we can have some respect toward each other as fellow citizens, deserving of an equal role in democratic governance. And we need that respect, because democracy requires that we respect how the vote comes out, even when our side loses.

The polarization story is usually told symmetrically: conservatives and progressives are approximately as vulnerable and approximately as blameworthy.

But, Talisse argues, this natural process of forming civic friendships is threatened by the increasing hyper-saturation of politics. Once we attach political meanings to every single act, we radically accelerate our self-sorting into like-minded groups. When hunting gets coded as “conservative” and cleaning up parks as “liberal”—and so on for an endless array of other dimensions of our lives—then we lose our last spaces for running into and forming some kind of relationship with our political opponents. Even our fashion has become politically coded. Our NPR tote bags and our camo jackets, our trucks and Priuses, all signal a political affiliation. As a result, when we do share a physical space—the city street, the subway—we can still keep ourselves sorted. This sorting happens precisely because we are trying to be good democratic citizens—because we are convinced of the importance of politics, and therefore make everything political. In short, polarization is significantly worsened by our tendency, as Talisse puts it, to “overdo democracy.”

Once sorted into like-minded enclaves, Talisse says, we are subject to a very specific effect. Overdoing Democracy leans heavily here on empirical research regarding group polarization, especially as it has been formulated by Sunstein. (Talisse renames the phenomenon “belief polarization,” but I will stick with the original and standard terminology.) Group polarization “besets individuals who talk only or mainly to others who share their fundamental commitments,” Talisse explains, and leads them to “embrace a more extreme version of their initial opinion.” Gathering in like-mindedness enclaves increases our moral and political certainty. Or, as Sunstein puts it, we are driven to take on more extreme versions of our beliefs. At the same time it becomes easier to demonize the other side.

Our current political rift, Talisse suggests, arises from the profound mutual disrespect between the two sides—which is, in turn, rooted in this basic fracture in the social landscape. The attempt to make everything political actually undermines the goods of politics. The fix is to restore civic friendship, to find our way back to respecting the other side. First, we need to change our view of our rivals. We need to see our political opponents as holding their values sincerely. We need to have “democratic sympathy.” And that involves realizing that group polarization, and other rationality-undermining effects, don’t just affect the other side. We, too, are the products of group polarization. Our own political confidence, too, is significantly irrational and unsupported. Then, once we have repaired our tendency to utterly dismiss the other side, we should engage in non-political cooperative projects with them: picking up litter together, teaching somebody to read at the library, joining a bowling league. We need to engage in parts of life where politics is simply not part of the picture, and actively work to keep politics out of the picture.


The propaganda story

We find a radically different story in Network Propaganda. (This text is open access, available free online and in PDF.) According to Benkler, Faris, and Roberts, a host of factors went into creating the political landscape around the election of Trump, but the dominant factor was propaganda. For these authors, propaganda means the intentional spread of false or misleading information for the sake of political power. And in their analysis, the most influential source of propaganda isn’t the fancy new stuff—bots or Russian troll farms—but the old familiar stuff: propaganda in mainstream news outlets. Whereas the prime movers in Talisse’s story are social mobility and pervasive political saturation, the prime movers in Benkler et al.’s story are Fox News, Breitbart, and their funders and allies among the political elite.

A key element of this account is an effect the authors call the “propaganda feedback loop.” Inside the loop, media outlets stop trying to present truths and to fact-check their fellow outlets. Instead, these outlets are out to confirm their followers’ worldview. And the more time they spend in the loop, the more these followers get used to the experience of constant confirmation and grow intolerant of any challenges to their belief system. At the same time, the loop attracts political elites who are willing to align themselves with the loop’s prevailing belief system, in exchange for a share of those followers. The result is a “self-reinforcing feedback loop that disciplines those who try to step off it with lower attention or votes, and gradually over time increases the costs to everyone of introducing news that is not identity confirming, or challenges the partisan narratives and frames.” Communication becomes more about reinforcing agreement and shared identity than about finding the truth.

The propaganda story is usually told asymmetrically: one side is stuck in the propaganda machine, the other side fighting against it.

For these authors, the propaganda loop is not the inevitable outcome of some systemic social or technological force; it is a specific condition that afflicts some media networks but not others. Much of Network Propaganda is spent documenting the asymmetry of propaganda loops in U.S. politics. At the heart of the analysis is a careful, empirical study of the network topology of the media consumption environment around the 2016 election, including the network structure of Twitter and Facebook activity. The basic structure of the rightwing media ecosystem, the authors say, was completely different from that of the rest of the media ecosystem. The right-wing network—centered around Fox News and Breitbart—exhibited all the features of a propaganda loop. It excluded members that conformed to standard norms of objective journalism. False and misleading claims could (and did) circulate and get amplified, without criticism from anywhere else in the trusted network.

The rest of the media ecosystem exhibited a very different dynamic—what the authors call a “reality check dynamic.” In this setting, media outlets are incentivized to check up on each other; catching another outlet’s mistake counts as “chasing the scoop.” Outlets are thus encouraged to aim for factually accurate reporting and police failures in accuracy. According to the authors, this subset of the media constituted a single, large, interconnected network, which included mainstream, centrist media outlets, traditional liberal outlets, and more radically left-leaning online-native sites, from ABC News, CNN, MSNBC, and the New York Times to the Huffington Post, Daily Kos, Mother Jones, and Occupy Democrats. Despite the relative political diversity of this network, its various members treated themselves as beholden to one another. That is, a fact check from a more left-leaning outlet like Mother Jones would be treated seriously by a politically centrist, mainstream outlet like ABC News, and vice versa. And the typical user of this network, no matter where they were on the political spectrum, treated the whole network as interconnected—reading across the network, and taking seriously fact checks from sources across it.

How do we square this analysis with the fact that reporters and politicians on both sides often lie? The authors report that their preferred neutral fact checker, PolitiFact, recorded approximately the same percent of “mostly false” claims from Republican notable Mitch McConnell (43 percent) and Democratic notable Nancy Pelosi (41 percent), and a similar symmetry between Bill O’Reilly (53 percent) and Rachel Maddow (48 percent). (Notably, the right-wing media ecosystem had much higher heights of falsity—Rush Limbaugh weighed in at 81 percent. Perhaps these heights are enabled by these loop dynamics.)

What matters most, however, is how the network respond to those falsehoods. As Benkler et al. exhaustively document, the right-wing media ecosystem showed no tendency to correct errors or punish false reporting. On the other hand, while the rest of the media ecosystem did deliver false and misleading news stories, it also quickly subjected those stories to fact checking. Errors were quickly discovered; this usually led to retractions, widespread reporting of retractions, firing, and other standard performances journalistic objectivity. These dynamics tended to keep the system, as a whole, significantly more honest. (Some, even on the left, might not be satisfied by these characterizations of mainstream media. Recall, for instance, Amber A’Lee Frost’s lovely screed calling the New York Times a clickbait-driven rag, obsessed with minor elements in the cultural wars, wedded to bourgeois values, and clinging to a moribund political theory. Maybe she is right. But Benkler et al. aren’t claiming that the New York Times lacks bias. Their point is more modest: that the vast media ecosystem that ties together the center, liberals, and far left functioned under basic journalistic norms of fact-checking and evidence, while the right-wing media ecosystem did not.)

For all these reasons, Network Propaganda specifically rejects the framing of “polarization.”

The behavior of the right-wing media ecosystem represents a radicalization of roughly a third of the American media system. . . . To speak of “polarization” is to assume symmetry. . . . No fact emerges more clearly from our analysis of how four million political stories were linked, tweeted, and shared over a three-year period than that there is no symmetry in the architecture and dynamics of communication within the right-wing media ecosystem and outside of it.

The book also offers an argument against the Talisse-style systemic polarization story. If that story is right, the authors say, then we should see similar effects across both sides of the political spectrum, and across different geographical areas and communities. But we don’t. In fact, we see demonstrably asymmetric network dynamics. First, as others have documented, we see asymmetric polarization between Republicans and Democrats: Republicans have moved rightward much more than Democrats have moved leftward. And, as Benkler et al. point out, we see great differences in polarization in different regions, despite similar degrees of social mobility and exposure to technology. And this is best explained in terms of long-term, asymmetric political changes, such as the racial politics of the GOP’s Southern Strategy, rather than in terms of some universal polarizing force.


Two types of polarization

Which of these stories is right? Before we can begin to answer that question, we need to clarify the very different notions of polarization in play.

Talisse stresses the difference between political polarization and group polarization. Political polarization, he says, measures the distance between political opponents. The further apart the various elements—the party planks, the commitments of the member of the party—the greater the political polarization. Group polarization, on the other hand, is a process whereby like-minded enclaves increase the extremity of belief. Political polarization is the fact that two groups are standing far apart; group polarization is a specific story about how the groups got there. In other words, political polarization is a purely descriptive matter, silent about the mechanisms that cause it. Talisse and Sunstein propose a specific causal story that connects these two different notions of polarization. Our present state of political polarization, they claim, is explained by symmetrical group polarization. On the right and the left, groups are caught in the same like-minded bootstrapping effect. But systemic group polarization is only one route to a politically polarized society; Benkler et al.’s account offers an alternate explanation.

Confusing these forms of polarization—political and group—can lead to some serious misreadings of the data. Indeed, the two forms can come apart.

Group polarization can happen when members of any like-minded group increase their confidence in their beliefs—including centrists.

On the one hand, there can be political polarization without symmetrical group polarization. Prewar Nazi Germany was a highly politically polarized environment, but this isn’t best explained as a result of symmetrically irrational group polarization. Perhaps the phenomenon of group polarization helps explain why the Nazis were so Nazi. But what’s the best explanation of the extremely anti-Nazi stance of the German resistance? Plausibly, it’s that the Nazis were evil, morally corrupt—and the German resistance was responding accurately and sensitively to the moral stench. On the other hand, you can have group polarization without political polarization. This is a conceptual point; perhaps a philosopher’s simplified example will make it clear. Imagine a population composed mostly of likeminded centrists, who all believe in the goods of moderation and civility. The group polarization effect could take hold of them, pushing their degree of confidence in their beliefs up beyond their evidence. They might become excessively confident in their moderate beliefs, unwilling to consider the truth of any extreme position. Such group polarization would produce the opposite of political polarization—but that is not always for the best.

These distinctions reveal some common conceptual errors. First, the term “group polarization” is misleading. The phrase conjures an image of a movement toward two (or more) poles of extreme belief, but group polarization can afflict centrists, too. The image of “polarization” conceals the possibility of centrist polarization.

Second, there are two very different senses of “extreme” here. In political polarization, extremity simply denotes distance from the center—toward the far right or the far left. But group polarization can happen when members of any like-minded group increase their confidence in their beliefs, or acquire beliefs with more extreme content. In this latter sense, it is possible for there to be extreme centrists. For example, a centrist might transition from the belief that far-left politics are misguided to the more extreme belief that far-left politics is dangerous—or mere virtue signaling. Speaking of “polarization” invites us to direct all our suspicion toward like-minded enclaves at the extreme ends of the political spectrum, downplaying the danger of like-minded enclaves at the center.

To sum up: political polarization and group polarization aren’t necessarily linked. The fact that society is divided doesn’t mean that it has been subject to the like-minded bootstrapping effect of group polarization; nor, conversely, does the fact that society is united mean that it is free of such bootstrapping. Both the symmetric polarization story and the asymmetric propaganda story offer a mechanism that would explain the current climate of political divisiveness. We cannot infer which is correct simply from observing the divisiveness.


When is polarization irrational?

The choice of explanations here is crucial. Was it symmetrical group polarization, or was it asymmetrical propaganda? Each story makes a very different accusation of irrationality. Benkler et al.’s story targets only the propaganda loop; the systemic polarization story is a discrediting cluster bomb, which accuses almost everybody of irrationality.

As Talisse puts it, we are all drawn to clump together like-minded groups. Once we’re in them, group polarization will increase our confidence—not because we have better reasons or better evidence, but because we are overexposed to supporting voices and underexposed to contrary voices. Everybody we meet agrees with us, so we may take this as a good reason to increase our confidence. And it would be, if we were confronting a real sampling of genuinely independent thinkers. But if we have preselected our community—if we have allowed people in only if they already agree with us—then all that agreement is meaningless. Any of us who spend our time around like-minded folks should seriously question our confidence in any of our shared beliefs.

Speaking of “polarization” trains our suspicion on like-minded enclaves at the poles of the political spectrum, downplaying the danger of like-minded enclaves at the center.

That is a pretty hefty ask. Should we downgrade confidence in our beliefs simply because a lot of our peers share them? We know that group polarization can happen; as Talisse emphasizes, the effect has been well documented in small-groups studies. But is it a consistent and significant force for our own political confidence? And was it the prime mover in the case of the 2016 election? Were we all—left, right, and center—subject to this irrational effect, and should we all discard much of our moral and political confidence? Once we are armed with the diagnosis of group polarization, dismissiveness can come awfully easily. All consensus becomes suspect. It can be tempting to look at any group, marching in confident lockstep, and conclude that its members’ confidence must be the result of irrational group polarization.

But there are other explanations for shared confidence, of course. Take the group of trained medical doctors, who share a high degree of belief in the efficacy of the polio vaccine. Obviously, we shouldn’t dismiss the views of doctors simply because they agree. A group can share confidence in a belief because they are rational thinkers who respond appropriately to the evidence—and the evidence all points in one direction. Uniformity of belief, then, has at least two explanations: it could be irrational groupthink, but it could also be the convergence of clear-sighted people on the truth. What if the group exhibits uniformity across a whole bevy of beliefs? Surely a whole shared belief system must be the result of groupthink? As Talisse warns, group polarization tends to create more “uniform” beliefs in a population. But again: most scientists share beliefs in an incredibly wide variety of beliefs: the roundness of Earth, the efficacy of vaccines, the existence of black holes, the reality of climate change, the explanatory power of evolutionary theory.

Fine, you might say: maybe the convergence of thinkers is well and good, so long as they arrived there independently. But in many cases, it is only after people get together as a group that their beliefs fall into line. Surely this kind of group polarization is problematic?

Again, not necessarily. First, learning a reliable methodology may be an entrance requirement for the group. Doctors often have similar beliefs, because the process of becoming a doctor involves a similar kind of training. So long as they are being trained in a good method, the sheer similarity of their beliefs is not discrediting. Second, enclave boosting can be the result of a healthy exchange of reasons. This too happens in the sciences. One laboratory concludes that Vitamin D prevents cancer based on population-level epidemiological studies; another arrives at the same conclusion from tests on rats; another does it through computer modeling. When these various lab groups gather, they each learn new reasons—good reasons—to confirm the beliefs they already had. And they may have gathered precisely because they had similar interests and recognized in each other good inquisitive methodology. Both their initial like-mindedness, and the ensuing enclave boosting, can be explained in terms of the convergence of good epistemic actors on the truth.

The upshot of these considerations is that the mere observation of shared group confidence does not tell us whether that confidence is unwarranted. Some enclaves may have strikingly similar beliefs because of irrational groupthink, others because they share good reasons. And these two sorts can look quite similar, to an outsider’s hurried glance.

What if there is no exchange of reasons? Consider a purely emotional effect—a raw confidence boost from the support of our fellows. Surely that is irrational? Even then, not necessarily so. It would be irrational if we started at the right level of confidence and the boost pushed us over the edge into over-confidence. But if we were under-confident to begin with—if we were too self-doubting—then that such a boost could help fix our rationality.

The mere observation of shared group confidence does not tell us whether that confidence is unwarranted.

Suppose you have low confidence in your beliefs, and you come to see that this low confidence arises from your life in an unjust social world. You have been soaked in a culture that unfairly disrespects you, perhaps because of your race or gender. It has left you with a habit of excessive self-distrust. As philosopher Karen Jones puts it, self-trust is an emotional disposition; it doesn’t just respond instantly to reasons. It is stubborn, recalcitrant. Baked-in self-distrust doesn’t vanish as soon you realize its irrational origin. You need a long soak in a curative bath—a lengthy exposure to an adequately supportive community—to get the right emotional counterbalance. So you have good reason to consciously seek out a like-minded community, to help the process of repairing your confidence and self-trust. And that emotional boost, when deployed appropriately, is not distorting but reparative. Sometimes, toxin and cure need to come in a similar emotional key. And this, I think, is precisely the idea behind safe spaces.

The blanket condemnation of like-minded enclaves, I think, arises from a background assumption about appropriate confidence. It imagines that people already have, for the most part, the right level of confidence before gathering together in supportive groups, and that the emotional support can only bloat their confidence beyond rationality. (And it would be easy to sustain such a belief, if you happened to live, say, in an enclave of mostly confident, or overconfident, people.) But for people who have been systematically brought to a place of low self-trust, then like-minded support can restore their rationality rather than undermining it.


No free passes from motivated reasoning

Of course, you might think this whole discussion is quite self-serving. Both the Network Propaganda view and the defense of “safe spaces” I have offered are typical lefty views, and I am a typical lefty—so maybe this is all just motivated reasoning. I’m giving just the kind of self-serving argument that people on the left would give to justify their beliefs. And perhaps I am drawn to such an argument precisely because I have already been brainwashed, my whole life spent in a like-minded enclave of lefty academics.

Talisse makes exactly such an accusation. He says that we tend to think group polarization affects the other side, but not us; we tend to “disregard our own vulnerability to the phenomenon.” But this disregard, he says, is itself the result of group polarization. If this view is right, all are guilty of irrational confidence, and we should all do a substantial amount of self-discrediting.

What Talisse misses is that this sort of argument applies equally to all comers. Motivated reasoning isn’t just for extremists and radicals: the worry applies just as well to those who might call for civility, preach for moderation, and disdain extremes. Group polarization can beset any enclave at any place on the political spectrum, and motivated reasoning can affect those who love civility and moderation just as well as it can affect the extremists. The temptation to accept a Talisse-style view of symmetrical group polarization could itself be a result of group polarization—one arising in a body of like-minded centrists who would love to believe that the real problem was in all those irrational, polarized extremists. In fact, Benkler et al. make such an accusation:

As we have repeatedly seen . . . the prominent outlets on the left and center simply do not exhibit a parallel structure, content, or vehement outrage that we observe on the right. These facts are as inconvenient to academics seeking a nonpartisan, neutral diagnosis of what is happening to us as they are to professional journalists who are institutionally committed to describe the game in a nonpartisan way. Both communities have tended to focus on technology, we believe, because if technology is something that happens to all of us, no partisan finger pointing is required. But the facts we observe do not lend themselves to a natural, “both sides at fault” analysis.

This is not to dismiss either position out of hand. The point is that the position of advocating for moderation, civility, and civic friendship does not magically rise above the fray, rendering itself, by its peaceable face, immune to debunking arguments and accusations of motivated reasoning. We can point out that Network Propaganda is comforting to liberals and leftists, but we should also point out that Overdoing Democracy is comforting to centrists—to those wary of radical change, who longs for the civility of a bygone era. Neutrality doesn’t give you a free pass from accusations of motivated reasoning.

There is no particular reason to treat the symmetrical view as the right opening assumption. Civility is not a default.

So what do we do? I think we have to fall back on the evidence. And there is evidence for the reality and causal power of asymmetric propaganda, and less evidence (so far) for the actuality and large-scale causal power of systemic group polarization. Talisse presents an enormous degree of confidence in his conclusion about civic fracture as the original sin of the present discord, but I find this confidence unwarranted by the quality of the presented evidence. Given the current state of the evidence, I am more swayed by Benkler et al.’s data about the deep asymmetry of polarization and the incompatibility of those observations with the systemic polarization. It is notable that Talisse barely discusses propaganda and misinformation. In particular, he provides no reasons to prefer his story instead of the other. This seems particularly striking, since claims about propaganda and misinformation were widespread and highly salient in the 2016 election.

On the evidentiary front, Network Propaganda has one major mark in its favor. Talisse’s account largely relies on small-scale laboratory studies about group polarization. There is no compelling evidence that I can find, in Talisse or his cited sources, that the phenomenon of group polarization is the primary factor at play in causing the large-scale political effects being observed. We are simply supposed to connect the dots between the small-scale studies demonstrating that group polarization does happen and the large-scale conclusion that group polarization is the prime mover for the current discord. And we would have good reason to connect the dots if there were no other good explanations for that discord. But there are: Network Propaganda provides at least one. And Network Propaganda does provide at large-scale evidence of the probable impact of propaganda effects. Of course, this may be because media misinformation is easier to document. So we need to proceed with some significant caution, as we continue to examine the prospects of these two competing accounts.

Nothing I have said here settles the issue. Our political moment is strange and vast and tidal; we should resist the temptation to leap on the nearest comfortable story. We need to gather more evidence. But I want to emphasize one thing: systemic polarization is not a neutral position. There is no particular reason to treat the symmetrical view as the right opening assumption, nor to place the burden of proof on those who seek more asymmetric explanations. The moderate call for peace and friendship does not magically escape from the charge of groupthink. Civility is not a default. And leaping to accept, with total confidence, a story of symmetrical irrationality, without sufficient evidence—especially when there are other competing accounts, themselves with significant evidential support—itself bears the mark of motivated reasoning.

Editors’ Note: This essay is adapted from a forthcoming article in the Journal of Philosophical Research.