What’s so bad about echo chambers?

ABSTRACT

Echo chambers have received widespread attention in recent years, but there is no agreement over whether they are always epistemically bad for us. Some argue they’re inherently epistemically bad, whilst others claim they can be epistemically good. This paper has three aims. First, to bring together recent studies in this debate, taxonomizing different ways of thinking about the epistemic status of echo chambers. Second, to consider and reject several accounts of what makes echo chambers epistemically harmful or not, and then offer an alternative account which builds on some features of existing accounts. In our view, echo chambers – even those that are truth-conducive – are always epistemically harmful because of their reasons-undermining features, e.g. the way that they impede the development of epistemic reasons which can answer and withstand challenges. This account captures important ideas behind several theories in the epistemology of echo chambers, but offers a richer explanation that does not suffer from their limitations.

KEYWORDS:

• Echo chambers
• veritism
• evidence
• social-epistemic structure
• epistemic harm
• climate-change denial

1. Introduction

A popular view about echo chambers is that they are bad for us as believers. Echo chambers have been touted as promoting polarization and extreme beliefs; facilitating confirmation bias; over-inflating self-confidence; and reducing knowledge-sharing (Talisse 2019; Alfano et al. 2020; Sunstein 2017). But are echo chambers bad for us as believers? While the media and social sciences tend to take the view that echo chambers are epistemically harmful, recent work in epistemology suggests that this popular view is mistaken. Echo chambers might not be bad and some might even be good (Lackey 2018). This is because some echo chambers are truth-conducive and protect true beliefs (Elzinga 2020; Lackey 2021; Fantl 2021).

For example, taking in the views of those who affirm climate change whilst disavowing those who deny it can only lead to more true beliefs. From an epistemic point of view, it’s hard to see what could be so bad about echo chambers which produce and protect true beliefs. This, at least, is a position some epistemologists might be attracted to because of the centrality – or indeed, fundamental value – of truth in epistemology.¹ After all, we might think that the point of getting evidence or justification for beliefs is to increase one’s chances of forming or keeping true beliefs (BonJour 1985; Cohen 1984; Graham 2011; Littlejohn 2012) Indeed, some say that the aim of truth is what distinguishes the epistemic from, say, the moral or prudential.² If you already have true beliefs, however, then why risk exposing yourself to misleading information – to the potential loss of that value?

We argue that this line of thinking is mistaken. On our view, echo chambers are always epistemically harmful, including truth-conducive ones. This is because they curtail other epistemic goods besides true belief. All echo chambers, as we’ll argue, fail to provide members with the resources to diversify and strengthen their reasons for their in-group beliefs. This, in rough outline, is what makes even truth-conducive echo chambers still epistemically harmful for their members.

We’ll proceed as follows. Section 2 provides a working account of echo chambers, and explains why they are ordinarily thought to be epistemically problematic. The goal in this section is not to provide a new analysis of echo chambers nor a theory of how to escape them. Instead, we draw on existing theories of what echo chambers are (e.g. Elzinga 2020; Nguyen 2020; Sunstein 2009). What we have to say on the epistemic status of echo chambers should fit with most accounts of the nature of echo chambers. Section 3 explores and criticizes recent ways of defending the view that some echo chambers can be epistemically good (Lackey 2021; Elzinga 2020). Section 4 then explores and criticizes recent ways of defending the view that all echo chambers are epistemically bad (Nguyen 2020; Sunstein, 2017). In section 5, we develop a new view which synthesizes features of the theories from sections 3 and 4, which ultimately argues that, though some echo chambers may have epistemically good features, all are ultimately epistemically problematic. We call our account the Reasons-Undermining View. We argue that this theory provides a more robust account of the epistemic status of echo chambers than any of the views espoused in the recent literature on the epistemology of echo chambers. In section 6, we consider and respond to some objections to our account.

2. What are echo chambers?

An echo chamber is a ‘bounded, enclosed media space that has the potential to both magnify the messages delivered within it and insulate them from rebuttal’ (Jamieson and Capella 2008, 176). Here, we take ‘media space’ to include both offline and online social networks. Echo chambers distribute and reiterate information, where the content of that information is often consonant with the relevant in-group beliefs and other shared in-group attitudes of the people receiving it from within the echo chamber (e.g. that we should not trust Nature Climate Science authors, The Guardian ‘Climate Crisis’ journalists, or some specific politician’s testimony about scientific reports concerning anthropogenic climate change, and so forth). This is the echoing feature of echo chambers. As Robson (2014) puts it, the ‘accepted view of a group is … frequently repeated and reinforced’. In particular, the ‘judgments of the opinion leaders are not merely transmitted but amplified’ (Robson 2014, 2520). Beliefs along with trust, anger, resentment, or even hope can be amplified within echo chambers. For example, anti-vaccine echo chambers might echo their resentment towards vaccine mandates, or their hope that there won’t be vaccine mandates. Echo chambers also restrict information flow in a way that mitigates the distribution of content that is disconsonant with the relevant in-group views.³ This is the blocking feature of echo chambers.⁴ This feature is part and parcel with the fact that echo chambers are ‘enclosed spaces’, whereby ‘dissenting views, if they are present at all, are drowned or ignored’ (Robson 2014, 2520).⁵ As we’ll discuss below, the blocking feature of echo chambers comes in degrees; some echo chambers might block more disconsonant information than others, and in different ways.

Nguyen tells us that: ‘Echo chambers … . work by offering a pre-emptive discredit towards any outside sources’ (2020, 146; our emphasis). This is one way of ‘blocking’ disconsonant information. Indeed, it is rather strong: any outside source is discredited. Importantly, however, rather than blocking content by making it inaccessible – as might happen within certain cults – some sources get in but are pre-emptively discredited because the group already believes that those sources are untrustworthy. In turn, the sources aren’t considered seriously.⁶

However, some researchers think of echo chambers as ‘blocking’ disconsonant information in a more robust sense: leading members to ‘isolate from viewpoints with which they disagree’ (Vaccari et al. 2016; our emphasis), or by outright ‘ignoring the skeptics’ of members’ views (Cossard et al. 2020, 1; our emphasis), which suggests that echo chambers not only moderate members’ engagement with disconsonant information but their access to it. Echo chambers can become spaces whereby members ‘wall themselves off from topics and opinions that they would prefer to avoid’ (Sunstein 2017, 239). The metaphor is suggestive: sometimes disconsonant information is not even considered.

But do all echo chambers block information in such strong ways? As we theorize echo chambers, their blocking function comes in degrees. As Lackey (2021) puts it, echo chambers are social networks in which ‘dissenting voices are either absent or drowned out’ (Lackey 2021, 207). This suggests that disconsonant voices – i.e. views, arguments, sources – are absent or, failing that, ‘drowned out’ – a degree-theoretic notion.⁷ An ideal or perfect echo chamber, as we might call it, will cut access to disconsonant information altogether (think of states like North Korea, or isolated cults), creating the conditions for an echo chamber of only approved views, insulated from criticism. Another way of ‘blocking’ disconsonant information, however, is to use imperfect filtering mechanisms, whether agential, geographical, or algorithmic.

Consider, for example, certain kinds of conspiracy-based climate-change denialist echo chambers. Although ‘all outside sources of evidence have been thoroughly discredited’, these sources of evidence can make it through the filters for the purposes of discrediting, ridiculing, or amplifying distrust (Nguyen 2020, 150). Two agents within the same conspiracy-based climate-change denialist echo chamber might nevertheless have different degrees of access to disconsonant information, based on structural features of their social network as well as different degrees of trust in the in-group and distrust of the out-group. Not every agential ‘node’ in the network has the same access to counter-evidence, for example. This will make most online or digital echo chambers ‘imperfect’ echo chambers. The more deeply embedded the member within their echo chamber, we might think, the more distant the person will become from disconsonant information – whether that’s cashed out in terms of ignoring disconsonant information or a strengthened tendency to distrust it, and so not consider it seriously.

Now that we have a better understanding of the ‘imperfect/perfect’ distinction about the properties of echo chambers, we turn now to the kinds of sources echo chambers might draw upon. The informational sources that people take content from can vary, to include peoples’ testimony; online news sources, print and television media; social media networks; academic journals and websites; podcasts; and books. But in an echo chamber, what is distinctive about this information is that its content is selectively restricted (to some degree) to amplify the pre-existing beliefs of the in-group. In addition, agents may adopt a policy of enhancing the credibility of informational sources that agree with (or are supportive of) in-group beliefs, or deflate the credibility of those that disagree with it (Nguyen 2020). A downstream effect of this way of consuming information is that the beliefs one already holds will then tend to be held more firmly (Nickerson 1998; Stroud 2011).

The formation of echo chambers may well be pervasive, but one informational context where they are thought to commonly operate is in the domain of climate change. Many people only read media and online sources that share the same view as one’s own on whether climate change is real.⁸ Consider Climate Change Denier, who believes that the climate is not changing. She restricts the information she reads on climate change in newspapers, websites, and social media to climate change denialist arguments. She primarily discusses these arguments with people who already agree with her, and distrusts any source critical of climate change denial; so much so that she ignores certain sources altogether (like The Guardian’s ‘climate crisis’ articles, or Nature Climate Change). Contrast this with Climate Change Affirmer, who believes that the climate is changing. She also restricts the information she reads on climate change in newspapers, websites, and social media, but to only climate change affirming arguments. When talking to people on this issue, she primarily discusses it with those who already believe what she does about climate change and distrusts any source critical of anthropogenic climate change; so much so that she ignores certain sources altogether (like Breitbart, or the Heartland Institute’s media outlet).

Both Denier and Affirmer occupy echo chambers. As we have noted, prevailing views assume that echo chambers are bad for us as believers. Intuitively, Denier seems to be in an epistemically bad situation. Note that echo chambers apparently ‘lock us into tribalism’ (Grimes 2017). By pre-emptively discrediting evidence that’s relevant to the epistemic status of their positions, they end up ‘actively alienated’ from contrary sources – they might sometimes hear what the critics say but they don’t engage with them seriously (Nguyen 2020, 147).⁹ Commentators suggest that we should ‘restore a balanced feed’ (see, e.g. Seneca 2020); that social media echo chambers facilitate our natural disposition for myside bias (Avnur 2020), and can lead to confirmation bias (Nickerson, 1998). Major magazines and websites (e.g. Gawq) offer strategies to exit or avoid echo chambers generally. The thought is that there is something bad, from an epistemic point of view, with being in an echo chamber because of the risk of reinforcing myside bias, strengthening polarization, overinflating members’ confidence, and inhibiting ‘knowledge spillovers’, which comes with collective, diverse information sharing (Van Alstyne and Brynjolfsson 2005).

However, are echo chambers always epistemically bad? These views about the badness of echo chambers could hold for Climate Change Denier, but do they hold for Climate Change Affirmer? Would her situation as a believer be improved if she were to consume the kind of information that Denier consumes? Arguably not. This is because the ‘cost of allowing ourselves to be more sensitive to counterevidence and counterarguments is that we will be more easily misled by misleading evidence’ and the thought is that consuming Denier’s information poses such a risk (Fantl 2021, 2). Put generally, we might be attracted to the idea that by consuming information within an echo chamber that echoes only true beliefs and mitigates the flow of critical views of true beliefs – i.e. misleading evidence – we are within a good epistemic ecosystem. As Fantl (2020) puts it, ‘we should want to inhabit truth-conducive echo chambers’ and it seems like Affirmer consumes climate change information from within such an echo chamber (ibid).¹⁰ As such, Affirmer appears to be engaging in good epistemic behavior. In the next section, we will outline the motivation for this view in greater detail and offer reasons for why this view, and modifications of it, are incorrect. We will argue that even truth-conducive echo chambers, such as Affirmer’s, are epistemically bad.

3. Veritism and right inputs and processes

A recent defender of the view that some echo chambers can be epistemically good is Jennifer Lackey. According to Lackey (2018), the ‘real danger [of being in an echo chamber] is not one of structure,’ and by ‘structure’, she means that echo chambers reiterate information consonant with one’s beliefs, and hence reinforce those beliefs. Rather, she says, the problem is one of ‘content’:

When I’m reading an article about the impact of climate change on wildlife, what is the benefit of clicking on a button to reveal the perspective of a climate change denier? Sure, I will be exposed to a different view, but at the expense of … truth (Lackey 2018).

So, since echo chambers are content-neutral, then ‘echo chambers are not the problem’, rather, it is the informational content within them that is problematic. Affirmer affirms pro-climate change beliefs because she’s in an echo chamber which filters for reliable, scientific information about anthropogenic climate change, whereas Denier denies pro-climate change beliefs because she’s in an echo chamber which filters for unreliable, pseudo-scientific and conspiratorial information.

The thought this elicits is that it’s hard to see what’s wrong with Affirmer’s presence in the echo chamber because it protects her true beliefs from being revised as a result of considering misleading denialist voices. If the filtering mechanisms of her echo chamber were altered so that denialist sources were treated as serious alternatives to pro-anthropogenic climate change sources, she would put herself (and others) in the risky situation of forming false beliefs about climate-change. Lackey argues for this as follows:

if I consult one highly reliable media outlet on a regular basis, I’ll not only block out a lot of noise, I’ll also end up forming lots of true beliefs about critical issues. If I add other sources simply to avoid worries about the insulation of beliefs — without any regard for their reliability — I’ll end up out of an echo chamber, but far worse off as a knower. For instance, if I learn about climate change from a reputable environmental scientist, there is only the danger of acquiring false beliefs in also consulting a climate change denier (Lackey 2018).

So, Lackey’s point is that if you’re already in a ‘good’ echo chamber, as we might call it, outside exposure might too easily lead you to revise your true beliefs, which would on her view be epistemically bad.

This account of the epistemology of echo chambers in which some can be ‘good’ and others ‘bad’ is broadly veritistic: echo chambers are good if they result in more true beliefs, and bad if they result in more false beliefs.¹¹ So, Affirmer’s echo chamber is good because it causes her to believe what is true, and Denier’s is bad because it causes her to believe what is false. There are different kinds of veritistic theory that we could advance, but we propose to call this view of Lackey’s the Right Inputs View.¹²

The Right Inputs View takes the goodness or badness of echo chambers to depend on the truth-value of the informational content fed into the network, as well as its outputs (namely, the truth-values of the relevant beliefs of its inhabitants): if the information is true then the echo chamber will result in true beliefs, and hence will be good, but if the information is false then the echo chamber will result in false beliefs, and hence will be bad. On this account, climate change Affirmer is in a good echo chamber because she reads what is true and eschews what is false. In contrast, we might assume that climate change Denier would only read articles supporting the view that anthropogenic activity is not the cause of climate change, and hence would only be reading what is false. So, according to this view, the right inputs are true content, whilst the wrong inputs are false content. An echo chamber is good or bad, on this approach, depending on the truth or falsity of the content echoed within the chamber.

Consider an analogy. We might say that a clothing factory is good if its inputs are of the right kind. For instance, if the materials, dyes and fabrics used to produce the clothes are of high quality. But if these products are of poor quality, then equally, the clothing factory will not be good. Analogously, echo chambers will be good or bad, depending on the truth-value of their informational inputs.

The factory analogy is helpful because it reveals a key limitation with the Right Inputs View: it’s not only the quality of the materials that make a factory good; a factory can still produce poor clothing if the processes within it are sub-standard. What is also good about a clothing factory is the quality of the machines being used, the training level of the staff operating them, and their quality control guidelines. When these processes are high quality, they can increase the likelihood of a good product being produced, and when they’re low quality, they will decrease the likelihood of producing a good quality product, even if the material within the factory is of high quality.¹³

Analogously, in addition to the truth-value of the informational content, there will be many epistemically good- and bad-making processes within echo chambers, which will also determine their epistemic value. For example, one good-making feature of Lackey’s echo chamber is that she is reading articles and watching reports by trustworthy scientists and media outlets. Now, what may be bad about climate change Denier’s echo chamber is that she is reading articles and watching reports by untrustworthy scientists and media outlets. What makes these informational sources trustworthy or untrustworthy is whether they tend to report what is true or false, respectively, just as what is good or bad about the factory is whether its processes tend to produce high- or low-quality products. Hence, on a veritistic view, it is not just the truth-value of their information that makes echo chambers good or bad, but the truth-conduciveness of the processes within the chamber. Call this additional account the Right Processes View.

There are a range of mechanisms that could make echo chambers more or less truth- or falsity-conducive. For instance, according to a recent proposal by Elzinga (2021, 12), two broad kinds of mechanisms might be in play.¹⁴ The first include features of the psychological profile of the agent(s) within the echo chamber, including ‘epistemically vicious cognitive biases and belief-forming practices’, ‘ignorance and meta-ignorance’, or ‘the way collective memories are shaped through conversation’. The second are broadly institutional or social mechanisms, such as ‘hyper-moralized and emotionally charged communication styles’, especially political rhetoric, ‘unstable norms of online communication’, ‘social media and search engine algorithms’, or ‘highly personalized and difficult to monitor political advertisements on the internet’. Any of these mechanisms could make the echo chamber more falsity conducive. And presumably, there are symmetrical processes that can make an echo chamber more truth-conducive, such as stable communication styles, virtuous social media algorithms, and agents within the chamber who possess intellectual virtues and beneficial belief-forming processes.

Whilst the Rights Inputs and Right Processes views are conceptually distinct, Elzinga’s suggestions propose a way of combining them. Consider how authority figures might use propaganda and rhetoric to spread false belief; or consider receiving emotionally charged advertising, which can mislead one about the truth. If we think of the epistemically bad-making features of echo chambers in terms of these types of inputs, we’re leaning on the Right Inputs View. But other mechanisms at play show how features of the chamber itself can lead to the acquisition of true or false beliefs, regardless of the information fed into it. If the agents within the chamber have intellectual virtues or vices, or encourage responsible or irresponsible dialogue and debate with each other, then a chamber that receives accurate information could still produce false beliefs, and conversely, a chamber that receives inaccurate information could produce true beliefs. This would be a kind of Right Processes View.

Can a combination of the Right Inputs and Right Processes View show that some echo chambers are good, whilst others are bad? We think they cannot. This is because even echo chambers with true content and truth-conducive mechanisms have epistemically bad-making features. The problem is, roughly, that they undermine our ability to develop a diverse pool of epistemic reasons for our beliefs, such as reasons which tell against the critic’s reasons. To see this, let’s begin with an example of an echo chamber with false content and falsity-conducive mechanisms, since the problem we will develop is easier to see at first from this perspective.

Consider certain groups who are isolated, both geographically and informationally, from outsiders because of their background religious or political commitments and depend on leaders to tell them about the world. One such case is of Guinevere Turner, who was ‘raised to believe that we were going to live on Venus’ (Turner 2019). She grew up in an insular commune, the ‘Lyman Family’, which closed itself off from outsiders. Children like Guinevere raised in the community ‘had been born into a belief system and simply accepted it, as children do’ (ibid). As a result, there was little informational flow from outside. Outsider information that made its way into the community was filtered by the commune’s charismatic leader, Mel Lyman, who amplified his own false views and sheltered his follower from their critics, making an echo chamber of unusual ‘new age’ beliefs at odds with mainstream and 60’s counter-cultural views.¹⁵

Now, even if the information given to Guinevere and the other children was false, there are ways of showing that her echo chamber still had some epistemically good properties, including believing according to one’s total available evidence, and trust in recognized authorities (as opposed to, e.g. trusting just anyone within the group). For instance, Guinevere was told by the commune leader that they would be taken to Venus by a UFO on January 5th, 1974. As a result, Guinevere came to believe this. This is because Lyman was taken to be the epistemic authority of their community; he was the person considered most trustworthy by her peers and by the adults who had more experience living in the community. Like the others, she listened. Indeed, Guinevere and the others ‘sat in the living room all night, listening for the hum of the UFO’s’, awaiting their anticipated arrival (Turner 2019). This looks rational: if you believe that UFOs will fly over your home, isn’t it rational to expect sounds emitting from them (engine thrusts, etc.)?¹⁶ Moreover, believing that ‘A UFO will take us to Venus’ fit her total available evidence at the time because she had evidence that a UFO would take them to Venus – owing to the leader’s testimony and the fact that other members reciprocated that belief. She also lacked access to defeaters for the belief, since it was formed in an echo chamber which filtered for consonant views and insulated her from opposing views.¹⁷ For this reason, we are hard-pressed to evaluate Guinevere’s belief as irrational. The point is that some of the members of this echo chamber, like Guinevere, aren’t obviously following irrational norms or patterns of thinking. Call this case Commune.

Despite adherence within the commune to the relevant norms, the Right Inputs view would say that Commune is bad because it has false content, and the Right Processes view would say that it’s bad because it has bad processes, such as misplaced trust in an unreliable leader. But there are some other quite significant problems with Commune that these two views fail to identify. To see why Commune has epistemically bad-making features, and how this is, in turn, problematic for even truth-conducive echo chambers with true content, let’s return to the factory analogy. So, for Commune, we would have a factory with bad materials (false content) and poor processes (misplaced trust) for making products, but where the factory produces the best products it can produce with what it has available. The workers (Guinevere and friends) might work hard (adhere to the relevant norms), and could be thoroughly decent and virtuous, but without the right materials or machinery, they’re just stuck making poor products (false beliefs). Now, let’s keep in mind that this (Commune-style) factory does have something good going for it, namely, that the workers are doing the best they can with what they have. But to produce good products (true beliefs) it needs the opportunity to improve the materials and processes in the factory. And it can only do that by receiving better materials and machinery from other factories or people outside itself, and learning better methods for improving its products from those people. Analogously, for Guinevere and others within Commune, they must be able to receive information with true content from people outside of their insular community, and learn the truth about the world from them. That way, they will be able to adopt new beliefs and, in the language of the factory, produce a better product.¹⁸

The distinctively epistemic problem this brings out for echo chambers is that they generally block opportunities to enhance one’s epistemic position. It’s not just that Guinevere has false beliefs, nor that she lacks epistemic reasons for what she believes – ex hypothesis, she had them – but she certainly lacked access to the criticisms, and so reasons which might address the criticisms, or better still: the reasons to reevaluate her own beliefs, due to the exchange of reasons with her critics. This is easy to see in cases like Commune, where the echo chamber prevents one from learning what is true, and also with Denier, where one is prevented by their echo chamber from relinquishing false beliefs, and forming true beliefs instead. But we can also use this idea to show that there are even problems for cases like Affirmer, who holds true beliefs and whose echo chamber involves truth-conducive processes, but who is prevented in various ways from improving her epistemic position. How could this be the case if, by coming out of her echo chamber, she only stands to develop beliefs that are not true, as Lackey (2018) noted? Well, there are a range of ways in which her epistemic position is curtailed and could be substantially improved. She could, for instance, come to learn what reasons others give for their views; why, exactly, their arguments don’t support their false views; why the reasons given in favor of alternative views are misleading; as well as how to respond to criticisms of her own position, or how to identify the flaws in the arguments put forward by others for their incorrect view. In general, by preventing her from improving her epistemic position in these ways – by constitutively blocking her opportunity to diversify her pool of epistemic reasons, whether ‘perfectly’ by filtering out or isolating members from such sources altogether, or else ‘imperfectly’ by mitigating one’s engagement with such sources such as by, e.g. pre-emptively discrediting them – then even Affirmer’s echo chamber has some epistemic disvalue. In broad terms, this is what we call Reasons-Undermining: echo chambers block access to reasons that would allow people to improve their epistemic position.¹⁹ If this is correct, then there is something epistemically bad about even truth-conducive echo chambers with true content, and hence, something missing from both the Right Inputs and Right Processes views.

We will develop this idea – the Reasons-Undermining View – further in section 5. This account will argue that all echo chambers are epistemically problematic. But in order to develop this account, we first need to explore another way of arguing that echo chambers are always epistemically bad – what we call the Structuralist View. As we will see, there are a range of structuralist views, but none of them quite adequately show what is ultimately so problematic about echo chambers.

4. The structural view

In the previous section, we saw that the Right Inputs/Processes Views face the problem of accounting for the apparent epistemic badness of echo chambers that are truth-conducive and have true content. Namely, the fact that all echo chambers impede us from improving our epistemic position in a certain way – what we call reasons-undermining. In this section, we’ll explore accounts that attempt to say that even truth-conducive echo chambers are epistemically bad. However, we will show that the reasons-undermining problem faces those accounts as well.

To introduce this position, let’s return to the factory analogy. Some people think that certain factories are bad because of particular, contingent features of the target factories. Perhaps it produces chemical weapons; or it uses child labor; or it’s an unsafe environment, employing faulty machinery. Notice that these sorts of criticisms don’t take issue with factories as such, but with their implementation. If the factory produced vaccines, using adult workers who earn EU wages, following strict safety regulations, the criticism might lapse. There are good and bad factories, and the bad one’s are bad when their products are bad, their labor practices are bad, or their conditions or equipment are bad.

This kind of criticism of factory work echoes the kind of criticism of bad echo chambers we’ve reviewed thus far. There is another kind of criticism of factories, however, which targets factory work (and other kinds of labor) as such: that there is something problematic about being subject to repetitive labor or labor under external ownership. The key idea is that even if the factory were producing something good, utilizing safe and effective machinery by skilled adults, there would still be a problem with them. One such criticism comes from Marx, who argued that labor under capitalism generally commodifies workers and that factory workers’ creations – their products – are appropriated by the owners (Marx [1845] 2004, 13). This is part of the structure of factory work and for people like Marx it is an essential problem with it.

Consider a similar criticism of echo chambers. According to the Structural View, as we’ll call it, echo chambers are epistemically bad even if they tend to result in true beliefs. Perhaps progressive (liberal, conservative, etc.) views are mostly true and their sources reliable but is it epistemically good simpliciter to be in an echo chamber? Proponents of the Structural View say ‘no’. In turn, echo chambers are epistemically bad because of their social-epistemic structure, independently of the epistemic status of their informational inputs. So, this is a stronger view than the veritistic views we’ve surveyed thus far because it denies that there are epistemically good and bad echo chambers.

What exactly grounds the epistemically bad-making structure of echo chambers? We’ll explore two features: (i) misleading higher-order evidence; and (ii) undermining intellectual diversity via self-silencing. To do this, we’ll focus on recent work by C. Thi Nguyen and Cass Sunstein. As we’ll see later, our own Reasons-Undermining View has affinities with these views. The unifying thought is that these unavoidable structural features are essentially epistemically bad-making, and float free of the echo chamber’s informational content.

4.1. Misleading higher-order evidence

One way of thinking about the epistemic structure of echo chambers is that they necessarily facilitate misleading higher-order evidence. Higher-order evidence is evidence about the epistemic quality of one’s evidence or sources (Kelly 2010). The basic idea is that echo chambers repeat and reinforce member’s beliefs, thereby creating misleading higher-order evidence that the original beliefs have greater support than the evidence permits. Here’s an example of this effect:

Newspapers: S has a stack of copies of the same newspaper, each of which reports that P. S reads the first copy, and believes that P as a result. S uncovers the second newspaper, and reads that P again, and so on, all the while unaware that they’re the same newspapers.

Newspapers is a case in which there is misleading evidence about how much evidence one really has. Reading the first newspaper’s report is some reason to believe that P, but the additional copies shouldn’t increase one’s confidence in P; it’s mere repetition of the same claim, after all (see Wittgenstein 2009, 100; Nguyen 2020, 144). The thought is that paradigmatic cases of echo chambers, like social media echo chambers, have a similar effect. Consider:

Social media: S’s friend S1 shares that P, and judges that P is likely. Their friends in social media groups S2, S3, … ., share that P, and so S’s confidence increases.

Let’s consider how this looks from inside the echo chamber. There are many more people S trusts who also share that P. So, this seems like higher-order evidence for them that P. After all, it had to pass through them before being posted, and so each person’s post is some further evidence that the original is trustworthy. Put another way, many people corroborated that P, as individual eyewitnesses might all attest to some event’s occurrence independently of each other.

But it’s (allegedly) misleading higher-order evidence, since they are just re-sharing the same story. The epistemic problem is clear: members get misleading higher-order evidence that their beliefs are true because each repetition is counted as independent evidence for the original claim, when in fact it merely repeats the original evidence. The effect is that other users are ‘mere conduits for information, rather than sources of information’ (Nguyen 2020, 145).

Crucially, the misleading higher-order evidence-effect is supposed to hold independently of what ‘P’ refers to. Perhaps it is Climate Change Affirmer, enclosed within closely connected Facebook, Reddit, and Discord groups, which repeat and thereby reinforce climate change affirming beliefs and studies, omitting climate change skeptics. Their claims are true, of course; and their posted studies are reliable. The trouble is that while these studies are in fact reliable, mere repetition of the studies shouldn’t increase each member’s confidence that the studies are reliable. But the worry is that’s precisely what happens. Some thinkers articulate this worry in terms of inflated trust. Nguyen, for example, says that echo chamber members facilitate ‘runaway credence levels’ which are ‘built right into the foundations of any echo chamber’ (2020, 150). Call these kinds of worries about echo chambers ‘problematic dependence’ worries.

Our view doesn’t require there to be a symmetry in inflated trust between truth-conducive and falsity-conducive echo chambers, but it also doesn’t deny it. It could be that Affirmer and Denier’s echo chamber members exhibit a rationally impermissible degree of trust, but it could also be that, say, Affirmer is within the bounds of what’s rationally permissible while Denier is not. Importantly, our views about rational degrees of trust might turn on background empirical views about which echo chambers we are considering and their degree of truth-reliability, together with normative views about how to proportion one’s trust in these cases. Our claim is that – independently of our judgments about rational degrees of trust – there would still be something epistemically bad about echo chambers.

At any rate, we should be suspicious of the claim that echo chambers exhibit in-built inflated trust in approved members. Lackey (2021) makes this point compelling. She distinguishes between mere reflectors and evaluators of evidence within echo chambers. Reflectors merely share information, reposting statements or key points from a study, for example. Evaluators, however, first evaluate the information from the shared source, and then repost the information. Evaluators provide additional evidence for others, on top of one’s own evaluation, because their evaluations are independent of each other. As Lackey puts it, if ‘200 people post what they learn in The Guardian … . this has the support of having survived all of these additional sets of beliefs’ (2021, 210).

Echo chambers can contain both reflectors and evaluators. However, an echo chamber cannot contain more evaluators than reflectors; for then it is no longer clear that we are thinking of an echo chamber as opposed to a social network with some tendency to reflect evaluator’s conclusions.²⁰ When we consider the clear case of many more reflectors amplifying the conclusions of the limited evaluators, however, reflectors can be more or less autonomous in their reliance on a source even if they display a high degree of heteronomy in terms of their core beliefs. Reflectors can – consistently with being within an echo chamber – exercise ‘autonomous dependence’, which is a matter of manifesting one’s agency in how they rely on a source (Lackey 2021, 209). Reflectors display autonomous dependence to the extent that:

source beliefs: they have beliefs about the reliability of their shared sources of information, accounting for not only the source of the information, but the content and context of the information (e.g. “A statement made in the context of a political campaign may be regarded with greater suspicion than one offered over coffee”, says Lackey);

monitor: they monitor incoming information for defeaters, and;

responsibility: they take responsibility for their expressions of the relevant views (see Lackey 2021).

How then do echo chambers which contain these sorts of reflectors get around the problematic dependence worries? Return to the case of the 200 people who posted what they learn from The Guardian. Suppose they’re reflectors who exercised autonomous dependence. Lackey’s view is that there is more epistemic support for their views than just whatever the evaluators had. The autonomous dependent reflectors do more than mirror the evaluators in a cascading fashion, which is what’s necessary for the problematic dependence worries. This is because their expression of the shared view was ‘filtered through 200 different doxastic frameworks, which brings along potential differences in reliability assessments, belief acceptance, and defeating conditions’, i.e. for each autonomously dependent reflector in the echo chamber (Lackey 2010, 2021) (our emphasis). Additionally, each one bears responsibility for their expression of the view – it’s a manifestation of their own support. When 200 autonomously dependent reflectors post their support, this is ‘200 additional speakers shouldering the responsibility’ for the view (ibid). The point is that each one ‘vouches’ for its truth.

Even if we thought of echo chamber members as still far too dependent on the same source(s) for there to be any additional support for their expressed views, Lackey says that their trust needs to be ‘entirely blind’ for the problematic dependence worry to work, which might hold of certain echo chambers but not others (or certain agents within those echo chambers) (Lackey 2021, 213). So, our judgments about whether trust is inflated within an echo chamber should turn on the facts about the distribution of the kinds of reflectors within them. For this reason, we should be suspicious of the claim that all echo chambers are epistemically bad because they all feature inflated trust. Granted: inflated trust is bad. But whether echo chambers always feature inflated trust turns on empirical facts about the proportion of certain sorts of reflectors and their influence within their echo chamber. Although an asymmetry in in-group and out-group trust might be endemic to all echo chambers, inflated – i.e. irrational – in-group trust is not.

4.2. Epistemic self-silencing

This takes us to another, potentially problematic structural feature of echo chambers. One popular line of thought is that echo chambers are intrinsically intellectually homogenous, and that this homogeneity is epistemically harmful. Our own view (§5) builds on this idea, but, as we’ll see, diverges in important respects. The criticism is connected to John Stuart Mill’s belief that liberty has cognitive demands; ‘liberty of thought and feeling’ and the epistemic good that flows from the free exchange of beliefs. Sunstein (2017) champions this position. He argues that people should ‘see and hear a wide range of topics and ideas’ but that this is not what happens when we fall into echo chambers (Sunstein 2017, ix). While our own view is Millian, as we’ll see, it diverges from the kind of Millian view we consider here.

To understand why intellectual homogeneity is (apparently) epistemically harmful, we first need to understand ‘informational cascades’ and how they work within echo chambers. The reason why is that echo chambers are thought to unavoidably give rise to information cascades and vice-versa – which in turns yields intellectual homogeneity – because of their echoing and variable blocking functions. Inhabitants who do have exposure to certain arguments or positions which don’t align with in-group beliefs might be considered but not seriously; they are ‘credentially isolated’ from the countervailing considerations. As Nguyen puts it, ‘members are not just cut off, but are actively alienated from any of the usual sources of contrary argument, consideration, or evidence’ (Nguyen 2020, 147). Instead, they mitigate this information and amplify in-group’s beliefs. As we said in §2, echo chambers function by ‘echoing’ these beliefs, and the primary mechanism underlying this echoing-function is informational cascades. Sunstein argues that informational cascades are endemic to echo chambers and the driving force of polarization. An informational cascade occurs when a few people evaluate some evidence and communicate their evaluations to others, whereby (a) the recipient’s beliefs are based entirely on the group’s initial evaluations and (b) they communicate their beliefs to others as well, resulting in a cascade of belief-uptake. Sunstein cites the uptake of beliefs like ‘GMO’s are bad for one’s health’ and ‘vaccines cause autism’, despite the scientific consensus which says otherwise. In these cases, a few influential people’s initial evaluations – the evaluators – of the evidence cascaded, leading to social fragmentation grounded in false beliefs. This is because of the ‘rapid transmission of information within one group but not the other’, i.e. of the presence of mere reflectors (Sunstein 2017, 97). The result is intellectual homogeneity: many people with the same (in this case, false) beliefs.

This raises the question: insofar as Sunstein’s complaint is not that informational cascades are only prudentially harmful – cascades can make true beliefs available to us – wherein lies their epistemic harm? One thought is that informational cascades can easily lead to ‘waves’ of false belief, whereby many group members adopt false beliefs. On this view, cascades are truth-unreliable; cascades can make a view seem much more plausible than it is due to its increasing availability and uptake (Kuran and Sunstein 1999). If the initial evaluators of the evidence made a mistake, this will cascade into the group, via reflectors, leading to lots of people with false beliefs and misinformed inferences (e.g. inferences from false premises).

The natural response here is that cascades can themselves be epistemically good or bad. For example, in the Affirmer case, if we suppose that the initial evaluation of evidence was done by competent climate science experts on reliable data, the resulting cascade will be epistemically good, because people will bracket their own evaluations, aligning their judgments about climate change to fit with the expert’s judgments.²¹ This is likely to be more reliable than if each member evaluated it independently, especially on the assumption that most members won’t be climate scientists, but supporters of climate science. Of course, the same process will occur in Denier’s case, and that cascade will, in turn, be epistemically bad. You will still end up with intellectual homogeneity, but the explanation of why this is epistemically bad remains unclear (since truth-unreliability is off the table).

However, Sunstein’s worry is not only about the epistemic risk posed by initial errors made by evaluators cascading into a truth-unreliable chain. Instead, his worry is that ‘people who are in the cascade do not disclose to their successors and the public the information (or reservations) that they privately hold’ (Sunstein 2017, 45). His worry, then, is that echo chamber inhabitants’ private worries, doubts, or even their own evaluations get side-lined. The result is self-silencing. Hence, the epistemic problem with echo chambers – whether truth-conducive or not – on Sunstein’s view is that they yield self-silencing inhabitants.²²

Granting that self-silencing can be prudentially harmful, why think it is epistemically harmful? One is reminded here of Mill’s famous worry that ‘if any opinion is compelled to silence, that opinion may, for aught we can certainly know, be true.’ (Mill 1863, 101). So, the epistemically bad-making feature of self-silencing is, we might think, that a true belief might be silenced. The worry concerns a deprivation of possible true belief for others.

Notice, however, that in Affirmer’s case, their prior core beliefs are true. Silencing Denier’s challenges in effect protects their true beliefs (cf. Fantl 2021). So, one might think that a challenge to true beliefs about climate change from insiders will also serve to protect their prior true beliefs. After all, if insiders are suspicious of some insider belief, wouldn’t the suspicion likely be grounded in a false belief? If so, shouldn’t we say that self-silencing within a truth-conducive echo chamber is epistemically good?

The short answer is: ‘no’. The reason why is that it depends on the content of what insider’s self-silence. We need to contrast the kinds of direct challenges that Denier raises to Affirmer – e.g. ‘climate change is a hoax’ or ‘climate change doesn’t result from human activity’ – from questions born out of insider’s curiosities, misunderstandings, or interpretative considerations. For example, there is a difference between ignoring the challenge that ‘climate change is a hoax’ from those inspired by the desire to understand for oneself, e.g. why heavy precipitation is likely in some areas whereas draughts are likely in others, or to correct a common misunderstanding, e.g. why certain short-term weather effects are (or are not) evidence of climate change (Lombardi and Sinatra 2012). Self-silencing in these cases affects the promotion of epistemic goods. It does so by affecting one’s ability to understand why one’s prior beliefs are true. It also encourages misunderstanding: self-silencing can lead to misunderstanding, and thereby a failure to comprehend one’s true beliefs. We see this in the empirical literature on people who believe that anthropogenic climate-change occurs, but have false beliefs about why this is. Finally, self-silencing affects knowledge-extension: it impedes the extension of knowledge or rational belief by inference. For example, although insiders might know that anthropogenic climate change occurs, and know that there is a causal relationship between pollution and climate change, they might (and often are) grossly ignorant of the role and severity of certain pollutants (e.g. photochemical smog; radioactive waste disposal) (Lombardi and Sinatra 2012; Gowda et al. 1997).

Self-silencing in these cases impedes understanding-why one’s prior beliefs are true, as well as curtails the extension of one’s knowledge by inference (e.g. one mistakenly infers from an extreme weather case that ‘climate change is getting worse’). By self-silencing, one frustrates their desire for understanding, diminishes the opportunity for other insiders to understand or deepen their understanding, and encourages hasty inferences, thereby impeding knowledge. Although all the insiders in the Climate Change-Affirming echo chamber will believe (truly) that anthropogenic climate change occurs, many will not understand why, exactly, it occurs, or how the broad scientific evidence supports it (e.g. surface temperature measurements with satellites, geological records, weather station measurements, predictive and explanatory models, etc.). Moreover, many will not know why, exactly, certain practices increase global temperatures; or how, exactly, climate change affects environmental systems and biodiversity. Considering insider’s ‘why?’- and ‘how?’-questions provides opportunities for facilitating understanding, promoting knowledge, and strengthening their prior reasons for their beliefs, which is epistemically good, and its deprivation bad.

While this argument is attractive, we think it is limited. The reason why is that it seems like a contingent feature of echo chambers whether they engage in self-silencing. While the evidence indicates that echo chamber inhabitants typically self-silence, this doesn’t mean that it’s an essential feature of them. Insiders might raise their own critical questions about the group’s beliefs with the goal of understanding the support for the beliefs (even if those criticisms are shot down – it depends on the depth of their criticism). Self-silencing of the group’s ‘core beliefs’ might be endemic to all echo chambers but it’s hard to see how self-silencing in general is. However, it is not a contingent feature of echo chambers that they engage in other-silencing. Other-silencing turns the ‘silencing’ in the other direction. Outsiders are not heard, or heard but not taken seriously; they are disparaged, or caricatured. Of course, rationality doesn’t demand that we consider every challenge seriously. Rational belief can survive that.²³ But echo chambers partly function by systematically blocking all outsider challenges from being considered seriously; it exerts a positional influence on what evidence we are exposed to, favoring confirming rather than disconfirming evidence. As we’ll see, this is part of the source of the epistemically-bad making features of all echo chambers. Whenever challenges are let in, they are considered only for non-epistemic purposes, like promoting the prudential good of the group, like solidarity, loyalty, or just plain fun. If self-silencing is epistemically bad for the reasons we’ve seen, building on Sunstein’s work, then other-silencing is also epistemically bad, for similar reasons. The key difference is that while self-silencing is a contingent feature of echo chambers, other-silencing is not. This takes us to our own position.

5. The reasons-undermining view

In the previous section, we looked at recent attempts to argue that all echo chambers are epistemically bad. However, we found that these existing ‘structural’ theories have limitations – there are apparent problems with echo chambers that they do not explain. Those problems fall into the same category as the reasons-undermining problems that we saw, in §3, could not be explained by veritistic accounts. In this section, we offer our own account – the ‘Reasons-undermining View’ – that also argues that all echo chambers are epistemically bad, but which is not limited by the problems we found with earlier structural theories.

According to the Reasons-Undermining View, all echo chambers are epistemically harmful. This is because:

PREEMPTION: all echo chambers have a social-epistemic structure which systematically preempts its members from critically engaging with serious objections.

PREEMPTION holds because of the variable blocking and echoing mechanisms that sustain echo chambers. But what’s so epistemically harmful about Preemption? It’s not, as the existing structural theories suggested, that its members get a problematic kind of higher-order evidence, inflated in-group trust, or that its members engage in self-silencing (when they do).²⁴ Rather, Preemption is epistemically harmful, even for truth-conducive echo chambers, because they preempt the development of new reasons for beliefs (e.g. against novel objections), thereby affecting knowledge promotion (e.g. by blocking opportunities to strengthen one’s total evidence against objections and new critical developments). Knowledge- or understanding-why promotion is also hindered because members are blocked from considering their critic’s arguments, and thereby from understanding why it is that the critics are mistaken (when they are). With blocking in play, the criticisms aren’t reached – or reached but not in the right way, e.g. the criticisms aren’t considered seriously – and so the critic’s arguments aren’t authentically evaluated critically. Echo chambers thus effectively inhibit the expansion and development of epistemic goods which build upon true belief. We call this ‘reasons-undermining’ because echo chambers systematically undermine the development of criticism-responsive or -answering reasons for beliefs. A criticism-answering reason for p is one which responds to criticisms to p, thereby adding to the total support for p. Examples include reasons which explain why a premise in an argument is false or ill-supported, or reasons which explain why a piece of reasoning is misleading or unsound.²⁵

The Reasons-Undermining View appears to be Millian because of its emphasis on the exchange of reasons with critics. But it’s important to distinguish the Reasons-Undermining View from its Millian cousins which have different implications. First, it doesn’t entail that there is an epistemic problem with echo chambers members’ beliefs as such, but it permits such a criticism. For example, it’s consistent with the Reasons-Undermining View that members in the climate-change affirming echo chambers beliefs are not, for example, unjustified, unknown, or epistemically blameworthy for failing to consider criticisms.

Some Millians might say that what’s so bad about PREMPTION is that when one fails to critically engage with objections, this leaves it open whether the agent indeed knows that their view is true. This can be defended in different ways. For instance, some might argue that it’s because being epistemically responsible or even a knower requires ‘taking precautions against their own fallibility’ (Mill 1863, 11), or might appeal to notions like normative defeat, the idea that knowledge requires that there be no unanswered objections that one ought to mitigate.

As we defend it here at least, the Reasons-Undermining View is not committed to these stronger claims. Rather, the Reasons-Undermining View is making a claim about epistemic goods-deprivation which can but need not affect the epistemic status of echo chamber member’s prior beliefs. The epistemic harm of echo chambers for its members is ultimately grounded in epistemic goods-deprivation – like developing criticism-answering reasons for their prior beliefs – as opposed to member’s having unjustified, unknown, or irresponsible beliefs. The Reasons-Undermining View is Millian to the extent that the authentic exchange of reasons is an epistemic value: that without being ‘frequently discussed’ even a true belief is a ‘dead dogma’ (Mill 1863).

This takes us to another point about just how Millian the Reasons-Undermining View is. Mill argued that:

He who knows only his own side of the case, knows little of that. His reasons may be good, and no one may have been able to refute them. But if he is equally unable to refute the reasons on the opposite side; if he does not so much as know what they are, he has no ground for preferring either opinion. (Mill 1863, 67)

What the Reasons-Undermining View implies is not that, e.g. by failing to engage seriously with Denier, Affirmer’s prior reasons for their relevant true beliefs are weakened – i.e. that their justification is lost. That there are criticisms of one’s views does not mean that those criticisms are defeaters – that one must be able to answer those criticism for justification to be retained. Rather, it implies that Affirmer mitigates their opportunity to get specific kinds of good epistemic reasons for retaining their positions – the criticism-responsive or answering reasons – which are sensitive to, and rebuke or undermine their opponent’s views, objections, and arguments. Such reasons supplant one’s support for their belief p. It’s one thing to have a reason, V, which supports believing that p, and quite another to have a reason V* which rebukes or undermines the ∼p reasons; to have reasons which explain why the criticisms go awry.²⁶

More concretely, it is one thing for someone to have good reasons to believe the official explanation about, e.g. 9/11: that it was the result of a planned Al-Qaeda attack, with the airplanes’ impact affecting the Twin Tower’s structural integrity – i.e. belief-supporting epistemic reasons – and quite another to have reasons that rebuke the complex criticisms from deniers who believe that the official explanation is incompatible with material sciences and structural engineering – again, criticism-answering reasons.²⁷ These reasons help to promote new knowledge in the group – knowledge of why the dissent is mistaken – but also to assist with the stability of member’s prior beliefs: the more member’s true beliefs can withstand criticisms, the better positioned members are to retain their true beliefs.²⁸ These are epistemic goods but echo chambers allays their members of them.²⁹

Third, the Reasons-Undermining View does not imply that echo chamber members are obligated to consider dissent, but then – due to mitigating serious counter-evidence consideration via omission- or credibility-filters – mostly fail to discharge their obligations. Assume for the moment that Affirmer is a climate scientist but locked within their climate-change affirming echo chamber. Our claim is not that Affirmer fails in her epistemic obligations, and so is compatible with the idea that scientists (or others) are permitted to ignore certain kinds of criticisms.³⁰ The claim is rather about the epistemic bad-making features of echo chambers: their epistemic disvalue. Although one might not be under an obligation to consider the critic’s objection O, the proponent of the Reasons-Undermining View claims that there is epistemic value in the offering here: understanding the critic’s reasons for their opposed beliefs and the prospect of developing a criticism-answering reason for one’s prior beliefs born out of the sincere mutual exchange of reasons.³¹

Now that we have a better grip on what the Reasons-Undermining View implies, we want to compare it with the previous accounts. Consider climate change Denier’s echo chamber, which echoes false content and blocks true content, and so tends to produce and reinforce false beliefs. What is epistemically bad about this echo chamber, according to the Right Inputs View, is that it leads to false beliefs. Likewise, for the Right Processes View, the problem is that Denier’s echo chamber has falsity-conducive internal processes. But on the Reasons-Undermining View, this is not broad enough. It’s compatible with our view that false belief and unreliable processes is an epistemic harm, but our claim is that there is more wrong with echo chambers than that: again, PREEMPTION results in a relative mitigation of criticism-answering reasons; echo chambers undermine member’s ability to develop them.

Indeed, PREEMPTION and the development of criticism-sensitive or -answering reasons is connected with the fact that echo chambers fail to promote the specific kind of knowledge- or understanding-why that can come from critical engagement with authentic critics: although some members might understand why, e.g. climate change occurs, predictably many will not be able to respond to complex reasoning from skeptics and so be able to explain why their views are true despite those criticisms.³² Indeed, without engaging with the critics seriously, we are liable to mischaracterize and so to misunderstand our critic’s views or at least their reasons for their views (see Catapano, Tormala, and Rucker 2019).³³ Although this scenario is perhaps true for most believers of most complex beliefs, what’s distinctive about echo chambers is that agents are relatively systematically disadvantaged of this kind of knowledge- or understanding-why opportunity, whereas their counterparts outside of echo chambers lack the same degree of systematic social-epistemic disadvantage.

Similar concerns also apply to other echo chambers, like climate change Affirmer, which echo true content and block false content, and so tend to produce and reinforce true beliefs about climate change. Even in these cases, though, their ability to source or develop criticism-sensitive or -answering reasons is undermined. For instance, Affirmer cannot come to understand the kind of positive reasons – Denier’s supporting ‘evidence’ – that she has for her views, nor the kinds of grounds that Denier has for rejecting Affirmer’s evidence, since she is blocked from getting and engaging with those reasons. In turn, it’s hard to see how she could come to know what is problematic with those reasons, and hence why, exactly, their criticisms are incorrect. This helps to pinpoint the deficiency in the earlier veritistic views. As Lackey (2018) emphasizes, there’s no epistemic harm in being within Affirmer’s echo chamber because ‘there is only the danger of acquiring false beliefs in also consulting a climate change denier’, i.e. in engaging with Denier (Lackey 2018). But we should doubt this view because knowing for herself why Denier’s arguments are unsound, for example, would allow her to strengthen her total evidence for her own position in the following ways: she would not only have reasons which supports that Anthropogenic climate-change occurs but reasons which respond to criticisms – reasons which withstand Denier’s criticisms, for example, from the more mundane (e.g. ‘The Artic ice melt is a natural cycle’) to the more complex (e.g. ‘Pacific decadal oscillation is responsible for global temperature variations’) and the reasons for these views – and hence could be improved so as to be amenable to future iterations of Denier’s criticisms.³⁴

These effects can, however, be accounted for in structural terms, but the existing structural theories leave more to be said. For example, Sunstein’s view that echo chambers are harmful because they can lead to self-silencing gives us an insight into why all echo chambers are epistemically bad. Self-silencing is epistemically bad when it inhibits understanding why one’s position is true, but other-silencing inhibits one from developing reasons which undermine the critic’s reasons for their own beliefs. We lack the kind of knowing-why the counter-positions are false – the knowledge of why their arguments misleads – along with subsidiary facts about why their evidence against our own positions are misleading. The blocking mechanism of even truth-conducive echo chambers impede these knowledge- or understanding-why promoting activities by encouraging other-silencing. It’s endemic to echo chambers, truth-conducive or otherwise.

One might be tempted to think other-silencing is a contingent feature of echo chambers. Couldn’t, say, Affirmer’s echo chamber consider Denier’s misleading evidence carefully, and spread the information about why exactly all their arguments are fallacious, why their methods are unreliable, or why exactly their counter-evidence is misleading, so that members’ knowledge and understanding is promoted? No. That’s precisely not what echo chambers do. To the extent that a group of people, united by a core set of shared beliefs, engages in this kind of intellectual activity, it is not an echo chamber; for it is then not a group which engages in blocking – in the systematic mitigation of counter-evidence – even if it engages in echoing.

All structural theories imply that echo chambers mitigate members’ pathways to counter-argument and counter-evidence. Some omit counter-evidence whereas others might include certain counter-evidence but exclude it as part of their total evidence; the source is distrusted or it’s counted as misleading without serious examination or exchange with its proponents. This is because of their structural ‘blocking’ mechanism. However, it is only the Reasons-Undermining View which adequately explains why this is epistemically harmful. It is epistemically harmful ultimately because (and to the extent that) members’ opportunities to develop and so to even authentically use criticism-answering reasons is impeded – and it is impeded because of the structure of the echo chamber. To the extent that many echo chamber inhabitants do engage in expressing their criticism-answering reasons they are here hypothesized as at best conveying identity-signals to their group – of manifesting loyalty, ‘owning’ their critics, or other tribal-like behavior – rather than authentically engaging in the exchange of reasons with their critics.

It’s also important to see that all echo chambers affect members’ epistemic agency in a way that is epistemically harmful. For example, in Affirmer’s echo chamber, the fact that disconsonant information is necessarily blocked implies that her opportunities to improve her own and other members’ reasons for their in-group beliefs is also impeded. This is because one improves the reasons one has to believe that p (an in-group position) if one knows why the reasons against p are mistaken – what we are calling criticism-sensitive or -answering reasons. Echo chambers prevent one from accessing and developing these reasons.

This point is easy to appreciate. Consider that, even in an extreme case, like engaging with Holocaust denialists, whereby one has excellent reasons to believe that the Holocaust occurred, one’s epistemic position is improved by being able to answer the denialist’s challenges. On balance, it’s epistemically better to be the person who believes that the Holocaust occurred and can answer the denialist’s counterarguments than the person who believes that the Holocaust occurred but cannot. In general, the person who has good reasons to believe that p, understands what the criticisms are, and can respond to the salient ∼p arguments with supporting reasons is in an epistemically stronger position than the person who cannot. Of course, this is clearly beneficial, but we now want to explain how this point bears on the badness of echo chambers. What makes all echo chambers epistemically harmful, we will show, is the fact that the development of this epistemic position is systematically curtailed, even for members who are fortunate to be in truth-conducive echo chambers.

So, now that we understand why the Reasons-Undermining View says that even truth-conducive echo chambers have epistemically harmful features, we want to explore exactly why echo chambers are epistemically harmful, since it’s not obvious why deprivation of criticism-answering reasons is epistemically harmful. We’ll elaborate on this below.

Criticism-answering reasons aid the ‘stability’ of one’s true belief (their diachronic persistence) by affecting their rational ‘durability’ (their ability to withstand epistemic pressures, e.g. counter-arguments). This is because the durability of believing that p is improved when one is able to meet challenges to p with reasons. If one can adequately answer challenges to one’s true belief, for example, what keeps one from lapsing into error is not merely pressures from within their social environment – which might be fragile, e.g. dependent on various social media platforms or trust-relations between members – and rather on their ability to defend their beliefs with reasons they understand for themselves. One in effect knows why the dissent is mistaken for oneself – by knowing what the critic’s reasons are, why they mislead, and so why their criticisms go wrong – and not only that the proponents are mistaken, or that their arguments are unsound, and so on. This makes one’s true belief more durable, and thereby more stable. This is a potential epistemically good-making feature of developing criticism-sensitive reasons.

The development and employment of criticism-answering reason is also connected with one’s epistemic agency. It is part of our epistemic agency that we have and can authentically offer reasons to believe what we do – as the reason for which anyone ought to believe what we do. The claim here is not that echo chamber members cannot present reasons for their beliefs, or lack the ability to authentically exchange reasons either, but that they are – relative to the extent to which they’re embedded within an echo chamber – deprived of developing reasons which answer criticisms by way of authentically engaging in the mutual exchange of epistemic reasons with serious critics. This does not mean that developing criticism-responsive reasons just could not occur within an echo chamber. The point is rather that the development method (to the extent that it occurs) is moderated; a method which requires the exercise of our epistemic agency with others. Our epistemic agency involves epistemic activity of a particular sort, namely, the activity of giving and responding to reasons (see Chrisman 2018; Lynch 2013). We are answerable to challenges, and yet we need to know what our critics say – we need to know what their reasons are, why, exactly, they take them to support their positions, and why they go awry – before we can answer their challenges seriously with reasons of our own. It’s an expression of our epistemic agency that we engage in the practice of defending our beliefs from challenges with reasons; or of refining our beliefs in response to such engagements. Echo chambers mitigate this social exercise of our epistemic agency; in extreme cases they effectively rob us of it. Put generally, then, echo chambers deprive members from an otherwise ordinary opportunity to develop and manifest their epistemic agency.³⁵

We now want to comment on how our view interacts with some related positions in the literature. Avnur (2020) has recently argued that the epistemic harm of echo chambers is due to their tendency to initiate myside-biased evaluations of the evidence. The Reasons-Undermining View also helps to see what makes myside-bias epistemically harmful. Myside bias is a matter of getting and evaluating evidence in a way that fits one’s non-truth related goals, like protecting one’s identity. Avnur and Scott-Kakures (2015) distinguish between two types of myside bias influences, positional influence and directional influence. Positional influence affects evidential exposure, whereas directional influence affects evidential processing (Avnur 2020, 580).

Avnur thinks that it’s only evidential processing which can explain why echo chambers are epistemically bad. Put another way, while myside bias-facilitation is the core epistemically bad-making feature of echo chambers, this is because it yields biased evidential processing, i.e. processing which doesn’t fit truth goals.³⁶ And the fact that one processes the evidence this way is typically a defeater for one’s belief. Hence, echo chamber member’s typically have unjustified beliefs because their beliefs are defeated by myside biased evidential processing. We think that the Reasons-Undermining View fits with this in important ways but also diverges.

For starters, it doesn’t imply that echo chamber member’s beliefs are unjustified, but rather that they are not in the market for other kinds of epistemic goods, like understanding-why (of the sort we discussed) along with the support of criticism-answering reasons: reasons which tell against specific challenges rather than only reasons which generically support one’s belief. This is what the right sort of informational exchange and exposure could facilitate but is absent in echo chambers. Second, although the fact that directional influences (like the group’s non-epistemic goals) manipulate member’s belief maintenance in epistemically harmful ways, the ground for this is not (or not always) that it defeats the member’s beliefs, as Avnur claims. The epistemic disvalue of echo chambers is due to features of the structure of echo chambers, but in broader ways than by their ability to defeat member’s beliefs. It’s rather that echo chambers deprive their members of a certain kind of epistemic enhancement. This is an ineliminable bad-making feature of them.

What’s more, this bad-making feature can impact the epistemic quality of one’s beliefs as well by impacting the diversity of one’s total support or epistemic reasons. What we mean is not that echo chambers defeat the justification one has for their relevant in-group beliefs, or prevents them from knowing that those beliefs are true, but from compromising their ability to have a diverse set of epistemic reasons; reasons which not only support their beliefs but answer important challenges to them. Avnur argues that, to the ‘extent that your response to epistemic reasons is compromised, this can serve as a defeater of your justification (if you had any)’ (2020, 586). We think he narrowly misses the target: myside biased inquiry compromises one’s epistemic reasons, but not necessarily because it defeats one’s justification. The compromise is due to the fact that one’s pool of epistemic reasons is so limited.

6. Objections & replies

Here, we want to consider key objections and respond. This will also help to further clarify our view’s commitments.

Objection 1: One might worry that the Reasons-undermining View is incompatible with an alleged feature of echo chambers, what we will call ‘Discrediting’. Nguyen (2020) puts it like this: ‘Members have been prepared to discredit and distrust any outside sources; thus, mere exposure to relevant outside information will have no effect’ (2018, 147). Members practice pre-emptive discrediting of the out-group’s beliefs. But the Reasons-undermining View suggests that what’s wrong with echo chambers is, ultimately, that members don’t engage with outside sources, and so nor do they engage the reasons against their in-group beliefs.

This objection takes two forms. The first form is conceptual. The Reasons-undermining View seems to recommend that members stop discrediting, and thus stop being in an echo chamber. That’s not helpful advice. The second form is epistemological. If we followed the Reasons-undermining View, and engaged with outside sources, it would ‘have no effect’ and thus be epistemically useless.

Against the first form, it’s important to see that the Reasons-undermining View aims to be explanatory of what’s epistemically harmful about echo chambers – including the truth-conducive ones. If discrediting is responsible for member’s inability to consider out-group reasons against their in-group positions, then we know what it is that’s responsible for the epistemic badness of echo chambers – i.e. an intrinsic feature of echo chambers – but not exactly why it (Discrediting) is always epistemically harmful. The Reason-undermining View supplies us with the explanation.

Against the second form, the Reasons-undermining View should not be understood as the view that, by being exposed to the critic’s reasons against insider beliefs, it should alter what one believes (perhaps for the better, e.g. in the case of falsehood-conducive echo chambers). Instead, the claim is that the lack of exposure to the critic’s reasons – again, even within truth-conducive echo chambers – affects the facts about the propriety of one’s total reasons, namely, whether one is in a position to develop reasons which defend their beliefs from critics. To the extent that one’s group is an echo chamber, one will lack access to and thus become less aware of the critic’s reasons against one’s views; one will thus lack the resources to respond to those criticisms. This impedes the group’s knowledge and understanding, namely, their understanding of why the critic’s beliefs are false or unjustified – which they’ll presume anyway, without even grappling with their opponent’s reasons.

Objection 2: This takes us to another challenge. One might think that people in echo chambers know a lot more about their opposition than laypersons outside echo-chambers. If this is right, then their ability to engage with the opposition’s reasons is less problematic than is suggested by the Reasons-undermining view. Rather than think of echo chambered members as reason-deaf, we should think of them as much more sensitive to, and better equipped to evaluate, the counterarguments and reasons than the average layperson. Insofar as the Reason-undermining view predicts otherwise, it looks like a (potentially false) empirical hypothesis about how echo chambers might degrade member’s capacity to handle criticism.

This criticism misunderstands our view. The claim is not that echo chamber members become less sensitive to contrary reasons over time. That is, it does not say that they become less capable of responding to dissent with epistemic reasons. This is a testable empirical hypothesis that we don’t explore here.³⁷ Fortunately, the Reasons-Undermining view as such doesn’t depend on it. What the view says is that echo chambers are epistemically harmful because they systematically block members’ opportunities to strengthen their set of reasons for their beliefs and to develop new reasons, i.e. reasons which specifically tell against (perhaps novel, or developing) criticisms – precisely the criticisms that get systematically blocked. Echo chambers facilitate a kind of epistemic stagnation vis-à-vis the development and strengthening of one’s epistemic reasons for their prior in-group beliefs. By mitigating access to dissenting reasons – a well-documented feature of echo chambers – those within them thereby lose the opportunity to improve their own positions by way of considering the dissenting reasons from outsiders, and crucially how they understand their positions, and so how to counter them with more specific opposing reasons, which is an epistemic shortcoming. It is epistemically good to be someone who, say, understands why, exactly, the critic’s key arguments rest on a mistake – e.g. to understand why their studies’ methodologies or their argumentation is flawed – which furnishes one with reasons which specifically tell against the contrary first-order evidence than to be someone who merely has a reason to believe that the critic is mistaken, or that there is scientific evidence for one’s own view, and that the dissent is misleading. Put very generally, then, the epistemic badness of all echo chambers is grounded in the fact that epistemic self- and group improvement is systematically curtailed, which is an epistemically bad-making feature of echo chambers; it is a kind of needless social-epistemic deprivation.

Objection 3: This relates to a further issue. What makes all echo chambers epistemically bad is that they block opportunities for members to enhance their pool of reasons for their beliefs. Here’s another example to illustrate this. Consider Amy, someone within a climate change affirming echo chamber. Since anthropogenic climate change occurs, and her echo chambers filters reliably for pro-science and climate-change affirming media, she has mostly true beliefs about climate-change. However, Amy is blocked from the specific criticisms levied by the varieties of climate change skeptics. As a result, she loses the opportunity to rebut those criticisms specifically – e.g. to foster new and detailed reasons for her beliefs, ones which tell against those specific criticisms – assisting the durability of her prior beliefs, and strengthening her reasons for her pro-climate change beliefs, in turn. But here’s the thing: isn’t there an opportunity cost for Amy to consider the climate change skeptic’s views? Wouldn’t she be better off reading more of the scientific literature, garnering more detailed support for her prior beliefs, for example, than sifting through climate change skeptic blogs to identify their errors? That’s arguably a much better way to enhance her reasons.

It’s important to see what the Reasons-undermining Views says about managing our epistemic opportunity costs. The point is to see that there is an epistemic asymmetry between people within and outside echo chambers, but who hold the same beliefs. The same person outside echo chambers can engage seriously with the dissent – and so is in a position to improve upon and diversify their reasons by engaging with objections – thereby better positioned to pay the cost of their epistemic opportunity. They might choose against it, however, and ignore the opportunity; they might save their epistemic bucks for others. However, for people within echo chambers, there is a social-epistemic structure which systematically prevents them from engaging with the objections to their views. Put another way, they essentially lack the epistemic opportunity possessed by their non-echo chambered counterpart. A social epistemic opportunity is necessarily missed; they aren’t given as many epistemic opportunities as their non-echo chamber counterparts. That’s bad.

Furthermore, proponents of the Reasons-Undermining View are not saying that, generally, one ought to learn about the views incompatible with one’s own and learn how to respond to them. That’s a view about individual epistemic responsibility. The Reasons-Undermining View is a thesis about the epistemic disvalue of echo chambers. It’s not a normative thesis about what our epistemic obligations are with respect to our critics. We are saying that, even if one wanted to engage criticisms for the purposes of epistemic improvement, echo chambers essentially prevent it – that’s why (in a nutshell) they’re always bad! The thesis that (a) one ought to engage with criticisms of their views and that (b) it is epistemically bad for one to be systematically prevented from engaging with criticisms of their views, are two very different views. The former demands that one actively does something about one’s epistemic position – hence the worry about the opportunity cost – whereas the later evaluates one’s social-epistemic condition, within echo chambers, as an epistemically bad one because of the unique systemic epistemic deprivation in play, viz., one’s lack of access to authentic objections, counterarguments, as well as competing conceptual resources for understanding their own positions.

7. Conclusion

We began by looking at the view that echo chambers can be epistemically good because they protect our true beliefs. We argued, however, that this view is incorrect: although some echo chambers can help one protect their true beliefs, and prevent adopting false beliefs, they have epistemically bad-making features that prevent them from being epistemically good. These include the stifling of access to the positive reasons others have for their own views, even if these reasons are conducive of false belief. By blocking this access, it becomes exceedingly difficult to know what is problematic with those reasons, or why they are incorrect, whereby knowing this for oneself would allow one to strengthen one’s justification for one’s own position by ruling out alternatives, and give one the resources to know how to offer counter-reasons to others with whom one finds oneself discoursing. This is what we called ‘Reasons-undermining’ – an essential feature of echo chambers that promotes epistemic deprivation and inhibits epistemic agency. This view of what is epistemically bad with echo chambers captures the core idea behind several other accounts that also endorse the views that echo chambers are epistemically bad, but offers a richer explanation that does not suffer from the limitations of those accounts.

We want to close with some conjectures about how the Reasons-undermining View could be put to work ‘in the wild’ to precisify our epistemic evaluations of echo chambers. Everyone should acknowledge that some echo chambers have positive veritistic value. Climate change affirming echo chambers are one such example. However, the Reasons-undermining view that we’ve developed here says that even these kinds of echo chambers are not epistemically good; that the positive qualification ‘epistemically good’ is unfitting. Still, we might press the idea further: isn’t there a sense in which some echo chambers are (epistemically) good-in-one-way (retaining true belief and avoiding false belief) but (epistemically) bad-in-another-way (developing criticism-sensitivity or answering reasons)?

With this point in mind, our epistemic evaluations of some echo chambers will be complex. We should acknowledge that it’s epistemically better to be within a climate change affirming echo chamber over a climate change denialist echo chamber, and yet it would be epistemically better still to affirm that there is anthropogenic climate change, to base one’s belief that there is on reliable climate science, and to understand why a core variety of denialist arguments are mistaken – to exercise one’s epistemic agency outside a social-epistemic structure which eschews authentic engagement with critics. Indeed, we can appreciate that even some climate change denialist echo chambers will be epistemically worse than others, depending on the depth of their distrust of climate change affirming sources; some might distrust all climate science because of unshakeable religious beliefs, whereas others might be driven by a firm albeit revisable commitment less connected to their personal or social identity.³⁸

Comparative epistemic evaluations are a natural upshot of the Reasons-undermining View. This is because it recognizes that every echo chamber has the same kind of epistemic drawback vis-à-vis the development of broadly ‘criticism sensitive’ epistemic reasons, but also recognizes that even those echo chambers which have positive veritistic value thereby suffer from these drawbacks. The Lyman Family cult was an echo chamber which protected false belief, but not always irrationally; the Climate Change Affirmer echo chamber protects false belief, and we may suppose it did this rationally, but we can also develop the case so that this is done irrationally, or in ways that prevent knowledge. In each case, their criticism-sensitive epistemic reasons are nonexistent or under-developed. The Reasons-undermining View, then, says that we should qualify them as ‘epistemically harmful’, and yet allows us to see that their relative degrees of truth-protection and rationality also gives us a way to qualify them as ‘partly epistemically good’ – or more accurately, comparatively epistemically better than other echo chambers.³⁹

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was supported by H2020 European Research Council [grant number 851613].

Notes

1 Consider Pritchard: ‘the idea that truth is the fundamental epistemic good was orthodoxy in epistemology. Indeed, this was the kind of claim that was so commonplace that it was almost not worth stating, as to do so would be somewhat superfluous’ (Pritchard 2021, 5515).

2 For example, Lacky says that ‘if … a practice has epistemic value, it would be one that somehow furthers the goal of acquiring true beliefs and avoiding false ones’ (Lackey 2018, 226). Hazlett (2016) notes that ‘epistemic’ is ambiguous technical jargon. The point here is just that inasmuch as there is a distinctively ‘epistemic’ concern or point-of-view, many epistemologists link it to truth, even if the nature or specification of that ‘link’ is open to dispute. See e.g., Cohen (2016) on this point (841).

3 Do echo chambers amplify only beliefs? Some say that echo chambers amplify ‘rumors’ (Choi et al. 2020; Gunn 2021). Other attitudes might be amplified as well. Hanna Gunn, for instance, focuses on the amplification of ‘what is said’, which may include the expression of members’ affective attitudes (Gunn 2021, 194). Consider, for example, a Covid-19 anti-vax conspiracy theory echo chamber. Arguably this kind of group amplifies not only conspiracy beliefs, but also that one should be angry that the government mandated restrictions, or resentful about the fact that ‘elites’ make important political decisions.

4 Some deny what they call the ‘echo chamber effect’. For example, Dubois and Blank (2018) say ‘high-choice environments allow individuals … to consume a variety of media’, i.e., not only media that is consonant with in-group beliefs (730). Some researchers express a softer skepticism about echo chambers, questioning the asymmetry in information flow. Guess et al. (2018) say that ‘people tend to prefer pro-attitudinal information to a greater extent than they avoid counter-attitudinal information’ (5). These concerns might lead one to say that echo chambers either don’t exist, so understood, or simply don’t work by excluding disconsonant information. It’s important to reply to these suggestions. First, the arguments in this paper are consistent with their findings. The arguments don’t depend on the view that echo chambers do not allow its members to consume a variety of media. The argument rather stipulates – based on the work done by, e.g., Sunstein (2009) and Pariser (2011), for example, that there is a mitigated flow of disconsonant information; this doesn’t entail that members cannot access counter-evidence, or even engage with it. Rather, echo chambers function partly by mitigating serious intellectual engagement with counter-evidence. Nguyen’s (2020) analysis of echo chamber also suggests this because echo chambers ‘work by offering a pre-emptive discredit towards any outside sources’ (pg. 146) But why would one engage seriously – i.e., for the purposes of authentically reconsidering what one believes – with evidence one already strongly believes comes from highly untrustworthy sources? Arguably, the purpose would be practical: to express condemnation or to signal in-group loyalty. Second, proponents of Nguyen’s (2020) analysis are likely to reply to skeptics by saying that the social science has insufficiently distinguished ‘filter bubbles’ and ‘online homophilic groups’ from ‘echo chambers’, so that their operationalization of the latter doesn’t track the key features of echo chambers. Third, as Nguyen (2020) argues, there are non-digital or offline echo chambers as well, with isolated cults as a paradigm cases, and surely in these cases it’s hard to see the research about social media behavior is relevant. Fourth, influential research like Dubois and Blank’s (2018) study echo chambers by way of studying online selective-exposure behavior and find the evidence insufficient. They argue that (a) in fact partisan individuals consume a diversity of media – but importantly this research is based on self-reporting survey data – (b) that individuals could be incidentally exposed to disconsonant information, and (c) that the way individuals use some media platforms or sites can vary (e.g., Twitter from Facebook), so that there is variable exposure to disconsonant information. Fortunately, none of this is inconsistent with our arguments. Exposure and consumption are different from serious engagement. Moreover, the effects come in degrees, and our view of the epistemic harms of echo chambers is mindful of the degree to which members are embedded in an echo chamber, with a key difference between ‘perfect’ and ‘imperfect’ echo chambers, as we discuss in the main text. We also note that the view that echo chambers mitigate disconsonant information to some degree is nevertheless the standard way of understanding them. For instance, the key research of Sunstein (2009) and Pariser (2011) focus on restrictions in the flow of information to mitigate that which disagrees with relevant in-group views, along with the selection of information that confirms them. Charitably understood, their work doesn’t presuppose a lack of access to disconsonant information, nor any engagement with it.

5 We take this to be the normal way of understanding echo chambers, for instance, in the work of Sunstein (2009) and Pariser (2011) which focus on restrictions in the flow of information to omit that which disagrees with one’s views, and selection of information that confirms them. In a similar account, Elzinga holds that an echo chamber is ‘a social network where beliefs are robustly and routinely reinforced through the echoing of consonant belief expressions throughout the network’ (2020, 2). This is also implicit in the work on ‘feedback loops’ by (Benkler, Faris, and Roberts 2018). Some social scientists define echo chambers partly in terms of this feature. See, e.g., Cinelli et al. (2021). Empirical research also suggests the restricted flow of disconsonant information. See e.g., Du and Gregory (2016).

6 Battaly (2018) says that ‘In dismissing intellectual options, one engages with them, but doesn’t take them seriously. The engagement is superficial.’ (262). So, the thought is that ‘serious engagement’ implies non-dismissal, and, intuitively, that one treats the received argument, testimony, or information as evidence or reasons that one ought to factor into for the purposes of modulating one’s belief.

7 To use the acoustic metaphor: signals might be drowned out by static, making them barely heard. Or they might be drowned out by excessive audio interference, making it heard but hard to process. Or it might be drowned out entirely, rendering it inaudible.

8 For instance, in an echo chamber of climate change denial, one might watch news channels that deny it (Hmielowski et al. 2014), cluster with likeminded people on social media (Williams et al. 2015), or restrict information from scientists by reading articles from websites such as The Heartland Institute.

9 Importantly, Nguyen thinks that while epistemic bubbles are relatively easy to exit – you just need exposure to the out-group views, etc. – this won’t do for echo chambers because echo chamber’s pre-emptively discredit any information that is critical of the relevant in-group views.

10 Some might find this idea intuitive or initially compelling. Different views might be driving this idea. Consider, for example, epistemic value monism, whereby truth is the fundamental epistemic value. If we think of truth as the fundamental epistemic value, a natural thought is that it’s not worth sacrificing truth for any other epistemic goods, since their value is instrumental value relative to the value of truth. See Pritchard (2014). Others might rely on views about epistemic justification and normativity, like certain epistemic externalist theories – reliabilism, E = K – by urging that the risk of habitually considering misleading counter-evidence will affect the reliability of the processes one uses to form the relevant sorts of beliefs or else affect one’s position to know. Epistemic value pluralists will likely be unmoved by the Affirmer case. They say that while truth is epistemically valuable, it’s not the sole epistemic value. Understanding, epistemic virtue, or wisdom are epistemically valuable and crucially not because their value bottoms out in the value of truth. See Brogaard (2008) and Grimm (2012).

11 See Goldman (1999). The core idea is that epistemic evaluation focuses on the determination of truth, whether it is belief, a belief-forming process, epistemic practices, or methods. Some philosophers understand veritism in a stronger way, so that it is the fundamental epistemic value, in the sense that anything else of epistemic value inherits its value from its relationship to truth. See Pritchard (2014), Pritchard (2021), and Goldman and Olsson (2009).

12 Although the outputs (beliefs of echo chamber inhabitants) matters epistemically for the Right Inputs View, we call the view the Right ‘Inputs’ View because it’s about what kind of information is let in and circulated therein. Lackey says that: ‘restricting our information sources is not objectionable by itself, and it can even have benefits. … . For instance, if I consult one highly reliable media outlet on a regular basis, I’ll not only block out a lot of noise, I’ll also end up forming lots of true beliefs about critical issues.’ In this case, our epistemic evaluations are supposed to track what got into and amplified within the echo chamber – i.e., the truth-reliable media, which is a backward-looking notion – and secondarily the echo chamber inhabitant’s beliefs, namely, their true beliefs, a forward-looking notion. The truth-reliability of input mechanisms and their tendency to produce true beliefs in the echo chamber inhabitants, then, are what proponent of the Right Inputs View are focusing on.

13 One might wonder how we should map our collective epistemic evaluations of echo chambers onto the individuals within it. It’s important to see that the veritistic views are focusing on echo chambers (as collectives, or groups, where applicable) rather than on individuals. Our factory analogy, however, permits some variation. It suggests that individuals within echo chambers can only be as good as their resources and the processes they make use of, but that individuals can of course do comparatively better or worse: some might read more of a study (double-check that a textile is sown) whereas others won’t; some might be more efficient than others at skimming news (using the machines faster), and so forth.

14 In general, Elzinga appears in favor of the approach being explore here. He says that ‘good and bad echo chambers are distinguished by the presence or absence of truth-conducive mechanisms … [an] organized system of parts which tends to yield significantly more true beliefs than false beliefs’ (2020, 11). He claims that this idea of truth-conduciveness is ‘broadly reliabilist’, e.g., Goldman (1979).

15 Those attracted to Nguyen’s analysis of echo chambers and epistemic bubbles might think of the Lyman cult as just an epistemic bubble. However, it’s implausible that the relevant disparity in trust endemic to echo chambers must be explicit – that the leader says ‘don’t trust outsiders’ about the relevant in-group beliefs, for example, as Limbaugh or others might do. Implicitly, many members of the Lyman Family cult distrusted outsider’s views because their own views were so radically different from the mainstream. Mel Lyman told them the exact date of the apocalypse and when a spaceship would pick them, which conflicts with mainstream sources that implicitly deny that those events will occur.

16 Implicit here is the idea that if these downstream expectations are rational, then the further upstream doxastic attitudes, like the belief that UFO’s will pick them up, are rational. The belief is partly what rationalizes the expectation.

17 There’s a question here about normative defeat; defeaters one doesn’t have but should have. See Goldberg (2017). Should Guinevere and the commune have considered that UFO’s don’t exist, such that this consideration defeats their belief? We think that it’s implausible that this is a normative defeater for Guinevere and the other young adults raised in the community. However, it plausibly is a normative defeater for other group members.

18 The Commune case bears some analogy with the New Evil Demon case. Guinevere seems to be rational, but her belief-forming processes are unreliable. For this reason, internalists could argue that the Commune case is one in which members like Guinevere have justified beliefs. Does that mean that internalists should say that some echo chambers are epistemically good as well? As we argue in §5, we think that there are other epistemically bad-making features of all echo chambers – and perforce, cases like Commune – that internalists can recognize. Namely, that once we focus on the structure of echo chambers, as opposed to only key individuals within them, we can understand why they undermine the development and expression of certain kinds of epistemic reasons.

19 This criticism of truth-conducive echo chambers can still be accepted by veritistists, i.e., those who say that truth is the fundamental epistemic value. The reason why is that we might think having more epistemic reasons which favors one’s belief and tell against counter-beliefs is a way of truth-promoting. It is, however, in tension with the kinds of veritistic views on offer in the literature, i.e., the Right-inputs and Right-processes views.

20 Thanks to an anonymous reviewer for pressing us about whether information-sharing social networks which had mostly evaluators would really be echo chambers.

21 This fits with the idea about epistemic authority that we should just defer to the authority if the goal is to enhance our truth-reliability. See Zagzebski (2012).

22 We take Sunstein to be making a point about mostly reflectors, but it can be extended to evaluators. Evaluators, for example, might be bracketing certain worries which they take to be disconsonant with the group’s prior beliefs.

23 Perhaps it demands only that we consider the relevant alternatives to what we believe. See (Battaly 2018).

24 Or, more weakly, these are among the epistemic harms of echo chambers, but there other epistemic harms endemic to them which the Reasons-Undermining View explains. Crucially, the proponent of Reasons-Undermining View doesn’t need to deny that echo chambers result in inflated in-group trust or misleading higher-order evidence for in-group beliefs; they just deny that this is the only epistemic harm.

25 We think criticism-answering reasons are a familiar kind of epistemic reason that we frequently employ in discursive contexts, especially in cases of controversy or debate. They are like defeater-defeaters or defeater-neutralizes. However, since it’s analytic that defeaters remove justification, we don’t think of the reasons which answer criticisms to p as defeater-defeaters for p, since it could be that the criticism fails to remove prior justification for p in the first place. One might think of them as contributing to what some philosophers call ‘dialectical justification’ instead (see Pryor 2005), but we think it’s a more specific notion, since it’s not necessarily a reason that is made explicit to an opponent.

26 As a side note: it’s important to remember that the ‘blocking’ and ‘echoing’ features of echo chambers come in degrees (§2). What this implies for the Reasons-undermining View is that the echo chamber inhabitant’s epistemic opportunity-mitigation comes in degrees as wells: everything else being equal, if Affirmer A is more deeply embedded within the echo chamber than Affirmer A* – i.e., A is closer to occupying a perfect echo chamber than A* – the Reasons-undermining View predicts that A is epistemically worse off than A*.

27 Veber (2021) makes a similar point against no-platforming. Drawing on Mill, Veber argues that even though the Flat-Earther’s clearly have a false belief, engaging with their (sometimes complex) objections give the agent a ‘livelier impression of the truth’, which Veber cashes out as being able to identify exactly where, e.g., the Flat-Earther’s reasoning goes wrong.

28 Cf. Williamson (2000) on the value of knowledge over mere true belief, whereby its greater value is grounded in stability: ‘the probability of your believing p tomorrow is greater conditional on your knowing p today than on your believing p truly today’ (Williamson 2000, 79). Our claim is that the probability of your believing truly that p tomorrow is greater on your possession of criticism-responsive or -answering reasons.

29 Mill says that: ‘Whatever people believe, on subjects on which it is of the first importance to believe rightly, they ought to be able to defend against at least the common objections’ (Mill 2007, 66). Proponents of the Reasons-Undermining View can agree with this, but it’s important to see that it’s ambiguous between a stronger and weaker reading. On the stronger reading, it implies that one is epistemically obligated to be able to defend their belief against common objections, whereas the weaker reading implies only that it’s epistemically bad if one is unable to do it.

30 It will take us too far afield to explore what counts as the right kind of criticisms, i.e., the criticisms one ought to consider. See, however, Battaly (2018), who explores this question in the context of the requirements of open-mindedness, which points in the direction of considering only the relevant alternatives to one’s views. The Reasons-Undermining View focuses just on the epistemic harms of echo chambers, however, independently of the obligations of agents to consider (or not) criticisms of their positions, since it doesn’t depend on deontological claims about member’s obligations, nor whether it would be more virtuous if they were to consider criticisms.

31 Of course, this can sometimes fail. Affirmer might exit their echo chamber and consider Denier’s reasons. But perhaps Affirmer is mislead, or is unable to develop answers to their criticisms. That Affirmer might turn into another Denier is a low-risk scenario, we think, because the empirical evidence suggests that we aren’t receptive to belief-change with actual or simulated critics; see Catapano, Tormala, and Rucker (2019). That Affirmer won’t be able to develop an answer to their criticism, however, just means that the epistemic opportunity didn’t pay-off. This doesn’t mean that being in the echo chamber wouldn’t have been worse: for then she’s constitutively blocked from developing any criticism-answering reasons.

32 Veber (2021) discusses an interesting case of a physicist who engaged with Flat-Earthers and recalls that ‘[s]ome of the thought experiments and experimental puzzles were actually quite hard, not because they were right, but because they were so intricately wrong’ (McIntyre 2019, 694–695; quoted in Veber 2020, 5).

33 See Peters and Nottelmann (2021), §3.1, for discussion. The thought here is that self-generating critic’s arguments or imagining their reasons is incongruent with the critic’s reasons for beliefs. Catapano, Tormala, and Rucker (2019), who they cite, focused on ‘perspective taking’ (imagining the thoughts and feelings of critics), and found that people become less receptive to changing their beliefs (pg. 427). This might be thought to answer the worry that we take an epistemic risk by critically engaging with false views such as by, e.g., revising our true beliefs and so losing knowledge. If we are rather apt to retain our beliefs anyway, the epistemic risk of lost-knowledge isn’t very compelling. But the other epistemic benefits remain, like developing or strengthening criticism-answering reasons.

34 See Cook and Dash (2022) of the Consensus Project and Skeptical Science.com, an online resource which taxonomizes climate-change skeptics’ arguments with rebuttals: <https://skepticalscience.com/>. There is emerging empirical support for the idea that, by exchanging reasons, people depolarize, even on highly partisan issues. See Fishkin et al. (2021).

35 Neither of these points implies that members become less rational in maintaining their prior beliefs. The Reasons-Undermining View is not that we lose or never had whatever epistemic reasons we had with us within echo chambers. It’s rather that echo chambers, by their very nature, prevent member’s reasons for their beliefs from including criticism-answering reasons: reasons which one could take with them to a public discussion, town-hall, or chat room. It’s a kind epistemic deprivation; a deprivation of reason development that can come from social exchange.

36 Interestingly, Avnur thinks that only directional influence is the epistemically harmful feature of echo chambers. One reason is that some internet uses seek diverse information online, and might be regularly exposed to counter-evidence. Of course, this will come in degrees. To the extent that a group’s members actively seek counter-evidence for their shared, core beliefs, this makes them look less like an echo chamber than simply a curious, albeit dogmatic group.

37 This worry can also be turned around: the empirical hypothesis provides another way of making good on the claim that echo chambers are always epistemically bad: if even truth-conducive echo chambers erode member’s epistemic reason sensitivity, this would be strong inductive grounds for thinking that echo chambers are epistemically bad.

38 We want to thank an anonymous referee of Inquiry for their helpful and detailed feedback, especially on this concluding point.

39 Work on this article was made possible by the project EXTREME (Extreme Beliefs: The Epistemology and Ethics of Fundamentalism), which has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation program (Grant agreement No. 851613) and from the Vrije Universiteit Amsterdam, the Netherlands.