Epistemic Bunkers

ABSTRACT One reason that fake news and other objectionable views gain traction is that they often come to us in the form of testimony from those in our immediate social circles – from those we trust. A language around this phenomenon has developed which describes social epistemic structures in terms of ‘epistemic bubbles’ and ‘epistemic echo chambers’. These concepts involve the exclusion of external evidence in various ways. While these concepts help us see the ways that evidence is socially filtered, it doesn’t help us understand the social functions that these structures play, which limits our ability to intervene on them. In this paper, I introduce a new concept – that of the epistemic bunker. This concept helps us better account for a central feature of the phenomenon, which is that exclusionary social epistemic structures are often constructed to offer their members safety, either actual or perceived. Recognising this allows us to develop better strategies to mitigate their negative effects.


Introduction
One reason that fake news and other objectionable views gain traction is that they often come to us in the form of testimony from those in our immediate social circles -from those we trust (Rini 2017, 44). A language around this phenomenon has developed which describes these social epistemic structures as 'epistemic bubbles' and 'epistemic echo chambers'. These concepts involve the exclusion of external evidence in various ways (Nguyen 2020). While these concepts help us see the ways that evidence is socially filtered, it doesn't help us understand the social functions that these structures play, which limits our ability to intervene on them. In this paper, I introduce a new concept -that of the epistemic bunker. This concept helps us better account for a central feature of the phenomenon, which is that exclusionary social epistemic structures are often constructed to offer their members safety, either actual or perceived. Recognising this allows us to develop better strategies to mitigate their negative effects.
The paper will be structured as follows. In Section 2 I describe epistemic echo chambers and epistemic bubbles, and I introduce epistemic bunkers as a new category. In section 3 I provide some justifications for why one might 'bunker down'. And in section 4 I offer some suggestions for how we can make inroads into epistemic bunkers.
Before proceeding, I will make a few clarificatory points to help the discussion along. First, in this paper I am entirely concerned with non-ideal agents in non-ideal circumstances. The kind of people I have in mind are not perfect decision makers. They make mistakes, they overlook pieces of evidence, they misunderstand things, and they are unlikely to have received training in decision theory. However, this doesn't mean that their choices are incomprehensible to outsiders. If given the opportunity to explain themselves, we could see how and why they chose as they did. These agents also operate in non-ideal epistemic environments. The news they receive is incomplete, they are sometimes lied to, and there are external actors who manipulate the information that they receive. Second, I take it that what we believe and who we associate with is mostly entirely our own business and not the sort of thing that should be 'intervened' on by external actors. This only becomes an issue of public concern and policy intervention when those beliefs and associations are connected to activities with harmful consequences to others. However, this paper will be silent on the question of when intervention is appropriate. The focus of this paper is to better understand one kind of exclusionary social epistemic structure -the question of when intervention is appropriate is a separate philosophical concern. On Nguyen's (2020) account, epistemic bubbles are social epistemic structures that incidentally exclude information from outside of our social/political/value circles, but the agent remains responsive to new evidence from outside their bubble if and when they encounter it. This describes the social epistemic landscapes in which most of us live our lives -we talk to certain friends, we read specific newspapers, we watch particular television shows, etc. As such, we miss out on evidence outside of our bubbles, but when we encounter external testimony, we don't immediately discredit it. For example, I was blindsided by the 2016 Trump election because I had been getting all of my American news from the New Yorker; I had been operating in an epistemic bubble. Nevertheless, I believed the news when it was announced.

Echo Chambers, Bubbles, and Bunkers
Epistemic echo chambers are more pernicious; these involve actively discrediting evidence from outside of the structure and holding firm, or doubling down, on pre-existing views when exposed to contrary external evidence (Nguyen 2020). Notably, on Nguyen's account, staying in an epistemic echo chamber for the sake of comfort or security is epistemically blameworthy: For an agent in full possession of a wide range of informational sources, to abandon most of them and place their trust in an echo chamber for, say, an increased sense of comfort and security, is surely some form of epistemic vice. There is some evidence that this may be the case . . . (Nguyen 2020, 154).
In this section I argue that safety can be a key concern in the creation of filtering social epistemic structures. I do this by introducing the idea of epistemic bunkers. I resist the idea that staying in these structures is always indicative of epistemic vice and instead argue that doing so can sometimes be quite understandable, given the hostile environments in which many of us find ourselves. However, I withhold judgement on whether joining or remaining in these structures is blameworthy, given that debates about moral and epistemic responsibility are beyond the scope of this paper.

Introducing Epistemic Bunkers
The idea of an epistemic bunker is inspired by academic work in development studies on 'bunkerization'. 1 This describes the creation of literal fortified and often militarized compounds for international aid workers (members of International Non-Government Organisations, United Nations workers, aid consultants, evaluators, etc.) to live in while working in hostile environments, such as war zones. They are intended to provide safety to international aid workers from attack, kidnapping, disease, and other ills encountered in these environments (Fisher 2017, 101). However, they come with an epistemic cost, which is that bunkerized aid workers do not have much contact with local communities, making them less able to understand their contexts (Autesserre 2014). This can be crippling to intervention efforts. Cartwright and Hardie (2012) have convincingly argued that for policy interventions to succeed, detailed contextual knowledge is required. Bunkerization has the benefit of keeping aid workers physically safe, but with the epistemic cost of losing out on contextual information, which can and does have the practical consequence of failed interventions.
Epistemic bunkers are similar. They are social epistemic structures that are designed to keep their members safe (sometimes emotionally, sometimes physically -more on this in 2.2). They boost the credibility of testimony from internal actors and diminish the credibility of those outside, sometimes blocking external testimony entirely. Like their literal counterparts, epistemic bunkers can be more or less fortified, such that more or less external testimony can make it through. But the epistemic activities within the bunker are not restricted to the passive receipt of information and associated filtering. Rather, the bunker is a secure zone in which members are able to participate in a variety of epistemic activities, such as questioning and offering testimony relevant to the issues of the bunker. So, the individual within the bunker is more likely to accept testimony from fellow members, ask questions of them, and offer their own testimony in response.
Some echo chambers (in Nguyen's sense) are also bunkers, but they need not be. The distinction is that the bunker is focused on creating safety in a hostile environment, and not all echo chambers are constructed to this end. For example, a group of teenage girls might exist in an echo chamber in which they actively discredit the testimony from their parents, not because they feel unsafe, but because their parents are uncool. Or perhaps you are a devoted fan of Taylor Swift -a member of the 'Swifties' -and as such you, along with the other Swifties, actively discredit any negative media about Taylor Swift. Again, the Swifties are not worried about safety, which is why they count as members of an echo chamber, but not an epistemic bunker. Conversely, not all epistemic bunkers are echo chambers. Epistemic bunkers just require that its members preferentially engage in epistemic activities with each other -less fortified bunkers will not fully block external testimony.
Epistemic bunkers can be created with members of one's physical communities, or they can be created virtually as online communities with shared interests and values. Bunkers can be created on a wide range of topics, interests, and commitments. Some are innocuous or even virtuous, others are troubling. In an innocuous example, a 2019 study of vaccine-hesitant mothers (mothers who are uncertain about whether or not to vaccinate their children) in France found that many of the women in the study would trust a medical doctor on issues of vaccination, but only once they had been recommended by female friends or relatives (Peretti-Watel et al. 2019). This is a closed epistemic structure, in which mothers share information about where to receive safe medical care. Nothing about this is especially troubling. The case of vaccine hesitancy also directs us to another feature of the epistemic bunker, which is that it can be 'shallow' or 'deep'. A shallow bunker of vaccine hesitancy might be concerned with value differences between the mother and the medical community ('but is this safe for my child?'), which is the kind of vaccine suspicion described by Goldenberg (2016). Or it can be deep, like vaccine sceptics who also endorse a conspiracy theory about autism and the MMR (Measles, Mumps, Rubella) vaccine (Goldacre 2008). The consequence of the varying depths of bunkers is their relative penetrability. A bunker of vaccine hesitancy based on an unspoken value disagreement might be accessed by having a conversation in which the values are made clear, while a vaccine-hesitant bunker based on an underlying conspiracy would require taking on a broader network of beliefs about the pharmaceutical industry and disease causation.
In contrast to the innocuous example of mothers sharing information with each other about trustworthy medical professionals, Julia Ebner's Going Dark (2020) documents how she went undercover in multiple right-wing extremist groups. In her description of joining a secret online Neo-Nazi chatroom, she talks about how she needed to answer a questionnaire and send in a photograph of a patch of skin on her hand to prove that she was white (Ebner 2020, 12). The process she describes is one of entering an epistemic bunker, and the processes she needed to go through (the questionnaire, the photograph) are examples of the fortifications that have been constructed to keep others out of the discussion space. In this instance, the epistemic bunker is a safe place for Neo-Nazis to have racist conversations. Once she is in the bunker, there are further fortifications in place to exclude outsiders who might have made it through the initial entry procedure. For instance, conversations are often coded and make use of metaphors that are only understandable to those who are already very familiar with Neo-Nazi literature and doctrine. This case is decidedly more troubling than mothers sharing information about trustworthy doctors.
Notably, there is nothing unique about epistemic bunkers in the digital age, except that we can include people further afield. It was possible to have Neo-Nazi gatherings in person before chatrooms were widely available.
Given the variety of ways that people can bunkerise and the variety of topics which can be the focus of these social epistemic structures, it is inadvisable to try to create a list of necessary and sufficient conditions that define epistemic bunkers. Rather, this is more of a family resemblance concept. We can discern some common features of epistemic bunkers. First, they are always motivated by concerns of safety -that is, wanting to create a secure zone in an otherwise hostile environment in which to engage in epistemic activities. Second, like literal bunkers, there needs to be some way of securing entry to the bunker; it isn't a very secure structure if everyone is allowed inside. In the literal militarized compound this might involve security clearance or being a member of the organisation that the compound was created to protect. In the case of the epistemic bunker, this could be that current members invite new members -perhaps women who are already members of a WhatsApp group of vaccine-hesitant mothers invite new mothers to join the group once they have been 'vetted' in the children's playgroup. Or perhaps one needs to send in a photo of one's skin to gain entry to the secret neo-Nazi chatroom. The process of including and excluding members is described as 'boundary maintenance' in the sociological literature (Fetner et al. 2012, 193). Third, literal bunkers have ways of securing the activities of those inside. They have thick concrete walls and sandbags to prevent outsiders from shooting those inside. Epistemic bunkers need to have their own ways of securing internal epistemic activities. These may include only having conversations via secure chat programmes, like Signal, or having secret meetings in the basement of members' homes.
It should also be noted that epistemic bunkers are not the same as the 'safe spaces' described in the sociological literature, even though they share much in common. While safe spaces also arise in response to hostile environments and involve boundary maintenance, they are broader in their scope of activity. For instance, in a study of the safe spaces provided by gay-straight alliances in Canadian and American high schools, the activities in those spaces included organising 'pride proms' and making buttons (Fetner et al. 2012, 202). Epistemic bunkers are focused on epistemic activities: filtering information, asking questions, offering testimony.

Safety in the Epistemic Bunker
What distinguishes epistemic bunkers from other exclusionary social epistemic structures is the focus on safety in hostile environments, whether actual or perceived. The safety provided by the bunker can take on multiple, often overlapping, forms.
Its most obvious form is emotional. Participating in epistemic activities within the bunker can feel safer. Operating in epistemic spaces where one is ignored/shamed/mocked/ridiculed is emotionally tough and being able to have conversations outside of that can offer some emotional reprieve (more on this in section 3.3).
In addition to the emotional costs, engaging in epistemic activities outside of the secure zone can be unsafe in other more practical ways. For instance, Andrew Feinstein, a member of the South African cabinet during the years of AIDS denialism in South African health policy 2 describes how he felt that he could not ask questions or offer oppositional testimony because he was afraid of losing his job (Feinstein 2007, 126). There can also be physical safety concerns associated with operating in hostile epistemic environments -perhaps I don't question the racist ramblings of the man on the night bus because I am afraid that he will punch me in the face.
There are other more indirect ways of achieving safety through bunkerisation. Trusting the testimony of others and acting on the basis of that testimony can have impacts on my literal safety and the safety of those I love. Perhaps I trust the medical testimony of my doctor and vaccinate my child, but in this case my child does suffer negative side effects. More will be said on why it can make sense to add a credibility boost to the testimony of those within my bunker in section 3.1.

The Epistemic Losses of Bunkerisation
Like actual bunkers, safety is gained, but there are epistemic losses, even in those bunkers that aren't themselves objectionable. An obvious potential loss is useful testimony from external agents.
Another negative effect is that of increasing polarization between members of oppositional bunkers, resulting in bunker members adopting more extreme positions than they might have initially intended and decreasing the possibility of constructive engagement across the divide in the future. Adrian Bardon describes this process as follows: [There is a] well-recognized tendency for groups of predominantly like-minded individuals to form more extreme positions as a result of the in-group dynamic. Opinions can feed off each other, as individuals starting out with more nuanced views are influenced by the group consensus; the result is that members of the group tend to collectively move in the direction of the more steadfast and doctrinal way of thinking. After interacting, groups of like-minded individuals will arrive at more extreme, one-sided views about, for example, race, punishment, public policy, and/or the threat posed by out-groups. Group polarization thus can play a part of exacerbating pre-existing ideological differences (Bardon 2020, 25).
Noting the tendencies toward increased polarization also gives us a sense of the life course of bunker membership. One might initially participate in a bunker for innocuous reasons (such as finding information about safe medical care for one's child) and later find oneself participating in a group with substantially more extreme views than those one began with.
Given these epistemic losses, in section 3 I pay closer attention to the gains of bunkerisation.

Bunkering Down
The epistemic environments in which many of us find ourselves are hostile. Death threats are ubiquitous on social media. Vaccine-hesitant mothers report not being able to ask medical practitioners questions about the safety of vaccinations without being dismissed as ignorant vaccine denialists (Larson 2020). The whole literature on epistemic injustice -which is about unjustly excluding individuals and groups from epistemic participation -is at its core about living in hostile epistemic environments (Fricker 2007).
In this section I provide more detail about why it can make sense for individuals to engage in epistemic activity preferentially with those within their epistemic bunkers. Given the wide variety of ways that one's epistemic environments can be hostile, and the wide variety of agents within those environments, these issues might appear in different combinations depending on the context. Or you might be lucky enough to live in a hospitable epistemic environment, in which case none of these conditions will apply to you. All these suggestions are obviously defeasible, and new sociological and psychological research may show that this is wide of the mark.

Values
One reason that it can make sense to boost the credibility of those within your bunker is because of shared values. It is likely that those within your bunker share your values, or at least that their values are more transparent to you than those of external actors. This can increase the credibility of their testimony on a range of issues, from politics through to issues of science and medicine.
Regina Rini argues that it can make sense to follow fellow partisans on normative and political issues. This is because party allegiance can act as a shorthand for values, making their normative testimony more plausible to us (Rini 2017, 52). For instance, if I already know that we share common values on abortion and climate change, I will take your normative opinion on Black Lives Matter protests more seriously -your normative testimony gets a credibility boost. Rini extends this to political testimony, which she argues is value-laden, both because 'political importance' is a valueladen notion and because any reports on someone's political character require that value-laden selection of what is important to comment on. The political testimony of those with whom we share political values thus also gets a credibility boost (Rini 2017, 52-3). More surprisingly, values can make testimony from peers on topics of science and social science that are unrelated to policy more trustworthy too (Furman 2020).
It is well-accepted that the sciences are value-laden. Values play roles in the selection of which topics to study, what methods to use, which theories appear more plausible to scientists and how to communicate findings, amongst other decision points in the activity of doing science (Longino 1990;Douglas 2009;Kitcher 2011). These values are often social and political, and they can be harmful to individuals and groups. For instance, early scientific explanations of the disproportionate prevalence of HIV/AIDS in sub-Saharan Africa relied on racist assumptions about Africans engaging in riskier sexual practices than their European counterparts, thus obscuring the role of social and economic drivers of the epidemic and limiting the scope of appropriate interventions (Stillwaggon 2006;Sawers and Stillwaggon 2010).
An additional layer of value is placed on the science around issues that need to go through policymakers before they make it to the public. For example, current guidance on when HIV positive individuals should start taking ARVs (Anti-Retrovirals, the medication to suppress the HIV virus) is that they should begin as soon as they test positive. This guidance is due to a broader policy commitment to 'treatment as prevention' and may run counter to individuals' own preference to delay treatment (Seckinelgin 2020). Or in an example that might be more familiar to many of us, governmental decisions about how locked down societies should be due to COVID-19 do not immediately follow from the biomedical science, but are balanced against other considerations, such as economic impacts.
The value-ladeness of science and associated policies might not be a problem in situations where the values are clear and not harmful to the agent. However, in situations where the values are obscure, oppositional to the individual's own values, or are harmful to the agent, it can be quite reasonable for the agent to trust those whose values are closer to their own or at least transparent and not harmful to them.

Lived Experience
Often the kind of information we are looking for from our peers isn't abstract knowledge about the world. Rather, we want to know how they have navigated their way through tricky situations so that we can do the same.
We make use of epistemic divisions of labour. None of us can know everything that is worth knowing, so we split the epistemic tasks between us and share the relevant information as and when we need it (Goldberg 2011). This is one rationale for the creation of experts (Kitcher 2011). It is also the rationale for talking to our friends when we encounter new experiences that they have already been through -they have done the experiential 'homework', so to speak. For example, if I was to become pregnant, it would make sense for me to talk to my friends who already have children, to get tips on what to look out for as I navigate my way through the maternal medical structures. Importantly, part of what I am relying on is that my friends are relevantly similar to me and that we share similar values (see section 3.1), so that I am getting information that is applicable to me. This is a good reason to give an epistemic boost to those within my epistemic bunker, especially when the world external to my bunker is perceived to be dangerous. This is the backstory to the classic feminist text, Our Bodies, Ourselves (Boston Women's Health Collective 1973), previously published as Women and Their Bodies in 1970. The medical world was found to be hostile towards women, who were discouraged from asking questions about medical procedures and issues relevant to their own bodies, and so the Boston Women's Health Collective was formed to facilitate women sharing information about their own medical experiences. The book is replete with first-person narratives, because the point is to share information amongst women about how to navigate their way through a hostile medical environment, which precluded constructive epistemic engagement or made it very difficult.

Emotions
Epistemic bunkers are about providing those on the inside with safety. As already mentioned, one way they do this is by providing their members with emotional safety. In an obvious way, it feels bad to be told one is wrong about something and so it is comforting to listen to those with whom one already agrees (Bardon 2020, 23). This is likely the form of comfort that Nguyen dismisses as an epistemic vice (Nguyen 2020, 154). However, the emotional reasons for staying in epistemic bunkers can be more complicated.
The world outside of the bunker can be frightening place. Sometimes that fear is warranted. For instance, the ubiquitous death threats (mentioned at the start of section 3) may be followed by actual violence. Threats can be frightening on their own, but they have additional force because of cases in which they have been followed through. Take, for example, the infamous sorority shootings committed by Elliot Rodger (Manne 2018, 34-41). This indicates that fear of external actors can sometimes be entirely or partially truth-tracking, but it is difficult to tell what kind of bunker one is in from the inside.
That fear is a relevant emotion in bunkerisation can further contribute to the process of fortification. Nussbaum (2018) describes the insulating role of fear as follows: Our narrative of fear tells us that some very bad things can easily happen. Citizens may become indifferent to truth and prefer the comfort of an insulating peer group who can repeat one another's falsehoods. They may become afraid of speaking out, preferring the comfort of a leader who gives them a womb-like feeling of safety. And they may become aggressive against others, blaming them for the pain of fear (Nussbaum 2018, 62).
Nussbaum immediately goes on to discuss the role of anger. Anger can also contribute to the insulating features of the bunker. Further, anger can be an appropriate response to the epistemically hostile environments in which many of us live. Bailey (2018) evocatively describes the feeling of anger that comes with frequent experiences of epistemic injustice: Our anger surfaces quickly pulling us back into our bodies . . . Those of us who live in epistemic twilight zones, in worlds where testimony about our lived experiences is repeatedly silenced, dismissed, distorted or gas lighted are familiar with the ever-present anger these constant erasures trigger (Bailey 2018, 93).
To return to the earlier example of mothers who are dismissed as ignorant denialists when they ask questions about how safe the vaccines are for their children, they might be angry at their hasty dismissal and this can contribute to bunkerisation.
Epistemic bunkers can provide a sense of ease, away from the fear and anger experienced in hostile epistemic environments. This helps us understand why people put the effort into constructing them in the first place.

Routes Out of the Bunker
Recognising that people construct epistemic bunkers for safety helps us better understand what needs to be done to mitigate their negative consequences. We need to pay attention to creating less hostile epistemic environments in which at least some conversations about important issues can take place.
There have been some heart warming tales of people who have done this for themselves. The Guardian's podcast series, 'Today in Focus' (2020), released an episode in November 2020 on 'the Karens'. This tells the story of two women named Karen, Karen Ward and Karen Cotter, in Bloomington, Pennsylvania. They have firmly oppositional politics, one Republican and the other a Democrat, but they have been friends for years and they trust each other enough to have proper conversations about politics. They have now focussed their efforts on improving the political epistemic climate in their town.
While this story is uplifting, not all of us have a Karen with whom can talk to safely about our political issues. It would be much better if there could be something done on a larger institutional level to take the hostility down a notch. In the rest of this section, I will introduce two institutional interventions that address two very different types of epistemic bunkers: 1) community health workers; and 2) citizen assemblies. Overall, given the variety of epistemic bunkers and the diversity of participants in bunkerization, there isn't going to be a one-size-fits-all response that will be good for all of them. This section is meant rather to give the reader a sense of some possible interventions. Good interventions are going to be those that de-escalate hostility. This can either be preventativein that having generally less hostile epistemic environments will lead to fewer bunkers -or they can be interventions on existing bunkers. There are also going to be some bunkers that won't be breached.

Community Health Workers
Philosophers of science have long recognised the epistemic benefits of intermediaries. For example, in standpoint theory, 'insider-outsiders' -those who are socially oppressed, but who must navigate their way through the world of the privileged -are able to see things that others cannot (like the operation of power). This gives them privileged epistemic status on certain topics (Wylie 2003). Intermediaries are epistemically beneficial, not just because of what they can see from their standpoint, but because they can communicate between epistemic communities. They can often access epistemic bunkers in ways that others cannot.
A number of the examples in this paper have been of health-related epistemic bunkers -ranging from vaccine-hesitant mothers to HIV/AIDS treatment regimens. One useful response to healthfocused epistemic bunkers has been community health workers (CHWs). CHWs are community members who receive training on particular health related issues (they don't have formal medical qualifications), receive some payment for their work and act as intermediaries between communities and the health sector. They are sometimes described as 'cultural brokers', because they are taken to understand the workings of the communities (Kok et al. 2015). CHWs are able to successfully gain trust, because they are already insiders in the groups in which they are working and so they aren't perceived to be threatening the way external actors might be.
Having these kind of access points into epistemic bunkers can be transformative. In South Africa, when ARVs became available in the mid-2000s -in the midst of an HIV/AIDS crisis and after a protracted battle to get access to this medication -the testing and treatment system was met with suspicion. There were rumours that those who were administering HIV tests were infecting people with the virus. It initially looked like people weren't going to get tested and receive access to the treatment they desperately needed. However, this ended up being a successful public health intervention and now millions of South Africans receive ARVs (Steinberg 2016, 62). One of the key factors that made a difference in this case were support groups, run by 'adherence counsellors', who received training and salaries from MSF (Médecins Sans Frontières/Doctors without Borders). Other key players in the support groups were confident young HIV-positive women who were members of the support group. Jonny Steinberg, writes this about his observations of the work these young women did in the groups as follows: Perhaps the most distinctive thing about them, aside from their age and their gender, was the combination of esoteric languages they used. Some of it was biomedical. The names of a host of drugs spilled easily from their tongues. They knew off the tops of their heads which combinations of which medicines treated which side effects, which antiretrovirals should not be combined with which antibiotics, at what stage a patient should move from first generation to second-generation treatment, and so forth. Some of the esoteric language they used was both biomedical and at the same time something else (Steinberg 2016, 71).
Having these intermediaries, those who were both members of the community and who had some medical knowledge, meant that access to the epistemic bunker could be gained and the conspiracy theories that hindered the medical intervention could be dismantled.

Citizens Assemblies
Another way of making inroads into epistemic bunkers is to have managed discussion spaces in which individuals can consider issues of politics and public values. One way to do this is through Citizens' Assemblies. Ireland provides us with an instructive case study, especially on charged and divisive topics, such as abortion.
The Irish Citizens' Assembly was established by the government in 2016 and ran until 2018. The practical organisation of this was managed by a market research company, who randomly recruited 99 citizens to participate, and special efforts were made to recruit individuals from diverse backgrounds. Demographic targets for recruitment focused on sex, age, social class and region, the ideas being that no one group would be disproportionately represented and this was underpinned by the theory that better quality deliberation happens in heterogenous groups. They were given five topics to consider: climate policy, the ageing population, abortion, fixed-term parliaments and referenda. The group met over weekends, typically once a month. During this time, they received briefings from experts, including from scientific experts, ethical experts, and testimony from special interest groups and narratives submitted by ordinary citizens affected by the legislation. This was in addition to briefing materials that they received in advance of meetings. Assembly members were able to ask the experts questions. Briefings were followed by discussions. These occurred in small groups (7-8 members) seated around a circular table, which was joined by a trained facilitator. The facilitator was there to keep the conversation on topic and to intervene if the discussion became disrespectful. More time was allocated to topics that were expected to be contested, notably nine and a half days were allocated to abortion compared to one day to fixed-term parliaments (Farrell, Suiter, and Harris 2019;Field 2018).
A noteworthy feature of the Citizens' Assembly was that it actually facilitated discussion on divisive topics. In follow up sample interviews with members of the assembly, they noted that open discussion was able to occur in the managed discussion space. One member responded: The beauty of the whole thing [is] its neutral environment. There's a great level of respect for everybody's opinions: we haven't had any fisticuffs. They're have been strong views expressed at times, but nobody's fallen out over it (Quoted in Farrell, Suiter, and Harris 2019, 118-119).

Another member:
Everyone could make their point . . . There was no shouting [if anyone tried to take over] they were put in their place . . . I felt I could ask anything and didn't feel I would be shouted down by anyone at the table. (Quoted in Farrell, Suiter, and Harris 2019, 119).
A limitation of the Citizens' Assembly is that it is not scalable. Having managed discussion spaces allows some conversations to occur across boundaries that wouldn't have happened otherwise, but it is still a tiny segment of the overall population. Citizens' Assembly meetings were covered in the media, with the intention of improving public confidence in their recommendations to government (Field 2018, 615). However, this isn't the same as going through the experience oneself.

Conclusions
Epistemic bunkers allow us to better understand one of the core features of social epistemic structures in which members preferentially engage in epistemic activities with internal agents over those external to the structure; they may do so to improve the sense of safety of its members. This comes with epistemic losses, such as missing out on information and increased polarisation, edging bunker members into more extreme positions than they would have adopted otherwise. However, we can understand why individuals might bunker down -because of value disagreements, because they want testimony that more closely matches their life circumstances, and because of emotional factors, like fear or anger. There have been practical examples of how to make inroads into epistemic bunkers, such as intermediaries (in the case of Community Health Workers) and managed discussion spaces (such as those created in Citizens' Assemblies). These strategies struggle with scalability, and there are some epistemic bunkers that just cannot be breached. This may be discouraging news, but at least this gives us a better understanding of how these structures operate.