A puzzle of epistemic paternalism

ABSTRACT Since the start of the Covid-19 pandemic in early 2020, conspiracy theories, misinformation, and fake news about the virus have abounded, drastically affecting global health measures to oppose it. In response, different strategies have been proposed to combat such Covid-19 collective irrationalities. One suggested approach has been that of epistemic paternalism – non-consultative interference in agents’ inquiries for their epistemic improvement. While extant literature on epistemic paternalism has mainly discussed whether it is (ever) justified, in this paper, I primarily focus on the potential implementation of widespread epistemically paternalistic policies (such as no-platforming and censorship) and its consequences. I argue that pursuing epistemic paternalism to combat Covid-19 collective irrationalities leads to a hitherto unnoticed puzzle for proponents of epistemic paternalism. Central to the puzzle is the idea those (governments, corporations, social media giants) who actually can (i.e., have the requisite power to) enact widespread epistemically paternalistic policies seem the institutions who are least suited to having such informational control over the populace. Thus, epistemic paternalism appears a sword without a hilt; while it may prove an effective strategy in tackling Covid-19 collective irrationalities, we do not have any way to use it without incurring serious risks.


Introduction
The Covid-19 pandemic seemed to have scarcely begun by the time collective irrationalities surrounding the virus began to spread -seemingly as quick as the virus itself. Whether it was the virus being a deliberately released bioweapon or newly installed 5G towers being responsible for its spread, conspiracy theories and misinformation appeared to be in the prime of their lives. This reached a fever pitch with the announcement of Covid-19 vaccines and the number of people yet to be vaccinated in fairly disparate countries such as the US and some European nations (e.g., Bulgaria, Croatia, the UK) 1 seems to indicate at least some doubt has been sowed in the global community by Covid-19 collective irrationalities. Naturally, as these collective irrationalities have plausibly extended the pandemic and caused the unnecessary deaths of many, researchers have wondered what strategies can be used to effectively tackle such conspiracy theories and misinformation. One such strategy proposed (e.g., Castro et al. (2020), Brown (2021)) is that of epistemic paternalism -that is, nonconsultative interference in agents' inquiries for their epistemic improvement (Ahlstrom-Vij, 2013). If agents can be shielded from such misinformation, the thought goes, then it will not be able to adversely affect them and will protect them from coming to hold false beliefs. In this paper, however, I argue that pursuing a strategy of epistemic paternalism in response to Covid-19 collective irrationalities leads to a puzzle previously unnoticed in the epistemic paternalism literature and by its proponents. Central to the puzzle is the idea that the corporations, social media giants, or governments who actually can enact widespread epistemically paternalistic policies of noplatforming or censorship are those very institutions that we would not want to grant the power of informational control over the populace.
I begin in section 1 by defining my terms. I offer an account of how I'll be understanding Covid-19 collective irrationalities, and adopt the received definition of epistemic paternalism, showing how no-platforming and censorship are indeed instances of the phenomenon. In section 2, I give the puzzle in full. I then justify each of its premises before discussing the significance of discovering the puzzle. In section 3, I consider some objections and responses to the puzzle. First, I examine a rejoinder that it is false that only large institutions can do this widespread epistemically paternalistic work. Second, I discuss whether the removal of misinformation constitutes a bona fide instance of epistemic paternalism. Third, I look at a possible category error in that abuses of no-platforming and censorship no longer constitute epistemic paternalism, dissolving the puzzle. Nevertheless, I find that none of the objections prove particularly problematic and thus, the puzzle stands.

Defining terms
Before I show how pursuing epistemic paternalism in response to Covid-19 collective irrationalities (hereafter, CCIs) could be an effective strategy and how this leads to a puzzle, I'll begin by giving definitions of both CCIs and epistemic paternalism.

Covid-19 collective irrationalities
A paradigmatic example of a CCI arrived early in the pandemic with over one hundred incidents in the UK in April 2020 relating to vandalism and destruction of 5G towers and harassment of telecom personnel. 2 What perpetuated these events was the idea that Covid-19 was spread by 5G towers. The collectiveness comes from the fact that the CCI moved multiple people to action all over the UK. The irrationality comes from the obvious impossibility of a coronavirus to spread through radio waves. In fact, the collective here seem prime examples of conspiracy theorists, and intuitively we might think that CCIs just are conspiracy theories. 3 This is incorrect, however, and indeed not even the received view on collective irrationalities No doubt some CCIs can adequately be considered as conspiracy theories, but some certainly cannot. 4 For instance, an extremely popular CCI around the world is the false claim that the data from the Vaccine Adverse Event Reporting System (VAERS) shows that Covid vaccines are dangerous. This is misleading for a number of reasons; VAERS is easily manipulated due to its entirely open nature (anyone can submit a VAERS report); the data (while potentially looking concerning to the laity) actually shows an extremely low percentage of adverse effects from those vaccinated; and so on. 5 Despite this, there is no conspiracy theory here in that it's not obvious what the "plan" is and public awareness does not seem to have been minimized at all -VAERS' data is entirely open for anyone to look at. So, not all CCIs are conspiracy theories, some are just misinformation. Here, I am taking misinformation to be information intended to mislead agents into developing false beliefs. The misinformation itself can actually be true or at least partially true (although will usually be false), but it is manipulated in some way or another to forward a CCI agenda.
So, in this paper, I will be understanding CCIs as conspiracy theories and/ or misinformation about Covid-19 that affect large groups of people and are irrational. These two definitions will sometimes co-instantiate (in that some conspiracy theories are information intended to mislead agents) but, as illustrated above, these definitions can come apart (some misinformation is not a conspiracy theory, and, as I am using the neutral definition of conspiracy theories, not all conspiracy theories will be misinformation as they are not necessarily irrational).

Epistemic paternalism
I will now turn to epistemic paternalism. A standard way to think of epistemic paternalism is Ahlstrom-Vij's (2013) account. This account has three conditions: A practice is epistemically paternalistic if and only if it interferes with the freedom of inquirers to conduct inquiry in whatever way they see fit(the interference condition) without consulting those interfered with on the issue of whether they should be interfered with in the relevant manner (the non-consultation condition), and moreover interferes -exclusively or not -for the purpose of making those interfered with epistemically better off (the improvement condition). (Ahlstrom-Vij, 2013, 61) The classic example of epistemic paternalism from the seminal work on the subject by Goldman (1991) is of a judge who withholds certain evidence from a jury for the purpose of helping the jury arrive at the correct result. This meets our three conditions; the jury's inquiry is interfered with (they do not have access to all the evidence), they were not consulted on this interference (it was entirely the judge's prerogative), and it is for the jury's epistemic benefit (so they do not come to a false belief and thus make the wrong decision). Epistemic paternalism can come in a variety of different forms, but I will focus on two instances of the phenomenon relevant for my purposes in this paper: the practices of no-platforming 6 and censorship. Peters and Nottelmann (2021) understand no-platforming as, "the practice of denying someone the opportunity to express their opinion at certain venues because of the perceived abhorrent or misguided nature of their view(s)." (p.7231) I broadly adopt this definition, taking "certain venues" here to include social media, television, and conferences. For censorship, I simply understand this as the removal of misinformation from any source, most often on social media. Let's now briefly discuss an example to demonstrate how these are instances of epistemic paternalism: Misinformation Mark. Mark used to run a popular Facebook page called Coronavirus Truth Org. Despite the name, Mark would almost exclusively post articles containing misinformation, conspiracy theories, and fake news about Covid-19. In fact, Mark was an early proponent of the conspiracy theory that 5G towers spread Covid-19. For this, Mark and his page were no-platformed from Facebook and all posts were deleted.
We can now go through and see how this case meets the three epistemic paternalism conditions. First, the interference condition. Agents who wish to inquire as to whether there is any link between 5G and Covid-19 now have their inquiry interfered with in that there is some evidence (Mark's articles) that they no longer have any access to. Second, the non-consultation condition. Presumably, not every single person in the world (if any) inquiring about 5G and Covid-19 were consulted on this censorship, thus meeting this condition as well. Finally, the improvement condition. The removal of the articles was clearly for agents' epistemic improvement in that it prevents them from acquiring misleading evidence, which in turn prevents them from coming to believe false things. Thus, no-platforming and censorship are instances of epistemic paternalism and are the instances that this paper will focus on. Now that I have all my pieces in place, we can turn to the puzzle of epistemic paternalism.

The puzzle
In this section, I begin by giving the puzzle before going through and motivating each of its premises. I then look at the import of the puzzle and its consequences.
First, however, it may seem like something salient to discuss should be whether epistemic paternalism (of this sort or in general) is ever justifiedjust because it makes for a potential response or solution to a bad thing (in this case, CCIs) doesn't inherently justify it. Nonetheless, I am going to set aside this question and assume that it is at least sometimes justified. 7 Instead, my focus is more on the consequences of the actual implementation of the sorts of no-platforming and censorship paternalism mentioned above. While it is a well-discussed topic in the epistemic paternalism literature who is justified in acting epistemically paternalistically, 8 my focus is on the less discussed area of who can (i.e., who has the power to) enact widespread epistemically paternalistic policies (of no-platforming or censorship) which will affect vast numbers of people. The answer to this is a rather short list: institutions such as the government, and corporations such as social media giants (Facebook, YouTube, Twitter) or Google. The puzzle arrives in that it seems like these are the sorts of groups that we absolutely do not want to have the power of no-platforming or censorship. As Goldman writes: Epistemic paternalism on the part of isolated individuals is quite a different matter from paternalism exercised by the state, or any other powerful organ of society. There are historical reasons for being very cautious about state control of information. (Goldman, 1991, 127) Proponents of epistemic paternalism often suggest science communicators as plausible candidates for acting epistemically paternalistically (John, 2018;McKenna, 2020;Medvecky, 2020). John, for instance, talks about a climate scientist making an "[epistemically] effective assertion" over an "honest assertion" (pp.83-84), which is essentially non-consultative interference for an agent's epistemic benefit. Nonetheless, this does not prove problematic for my puzzle despite the focus here on institutions. In fact, my paper complements this extant literature. To explain, suppose a science communicator acts in some epistemically paternalistic way when publishing an article on vaccine safety. The issue I am pointing toward is that this (epistemically good) article could be swamped (or at least surrounded) by misinformation and so has the potential to go unnoticed or have its possible benefits annulled. So, ways to remove this surrounding misinformation (i.e., widespread no-platforming and censorship) could be important. This is where my puzzle comes in; only large institutions of the kind highlighted throughout this paper possess the requisite power to do this widespread noplatforming or censorship -such control is not available to science communicators. However, these large institutions, I argue, are not wellpositioned to act reliably in agents' best epistemic interests. 9 Now, let's lay out the puzzle explicitly: (1) There are pro tanto reasons to implement epistemically paternalistic policies 10 to effectively combat CCIs.
(2) Those who enact epistemically paternalistic policies should be wellpositioned to act reliably in interferees' best epistemic interests.
(3) The only institutions who are plausible candidates for enacting epistemically paternalistic policies which can combat CCIs are institutions like large corporations or the government. (4) These institutions aren't well-positioned to act reliably in interferees' best epistemic interests. (5) Therefore, epistemically paternalistic policies to combat CCIs cannot be enacted. (6) Therefore, CCIs cannot be effectively combatted. 11 (3) is trivially true; a lone individual clearly lacks the requisite power to effectively combat CCIs on their own because they do not have the power to enact widespread epistemically paternalistic policies. 12 In fact, their best strategy would likely be to petition these powerful institutions to do something. (1), (2)., and (4) still require motivation, however, so that's where I'll turn to now.

Premise (1)
As stated in section 1.2, the sort of epistemic paternalism I am interested in here is no-platforming and censorship of certain information. So, for noplatforming the outcomes we would generally see would be purveyors of misinformation (i.e., people who spread CCIs) having their spreading privileges revoked -be this through social media bans, television and conference blacklisting, and so on. Censorship could come in the form of removal of misinformation. This differs from no-platforming in that unknowing or unwilling people could spread CCIs, in which case they're not purveyors in the same way that Misinformation Mark is, rather they are an accidental participant. The result of this epistemic paternalism is that agents in the epistemic environment are very unlikely to even encounter misinformation or conspiracy theories about Covid-19.
So, why might we think this makes for an effective strategy in combatting CCIs? 13 First, we can appeal to some extant literature on the subject. Levy (2019) speaks of the dangers of platforming problematic speakers in terms of the pernicious epistemic effects it can have on the wider epistemic environment: An offer of a platform is a signal that those who issue the invitation consider the person worthy of a respectful hearing. It is a signal that the inviters consider the speaker sufficiently expert, or sufficiently representative of expertise to have an opinion on that topic that should be taken into consideration." (Levy, 2019, pp. -495-496) 14 So, should such a problematic speaker be no-platformed, we remove this worry of conferring legitimacy upon them and their illegitimate opinions (say, for example, that 5G spreads Covid-19). Fantl (2018) discusses a similar idea: [I]f we know that the speakers . . . are uttering falsehoods, then we are prioritizing those other values over the value of truth because we are allowing falsehoods an inroad to the university that they wouldn't otherwise have. (Fantl, 2018, 200) Fantl argues that this sort of epistemic harm 15 is impermissible and thinks that the chief way to prevent this is by no-platforming such falsehood-uttering speakers. Finally, Castro et al. (2020) note the success of an epistemically paternalistic policy from Facebook which involved the demotion of fake news (34-39) and propose that, "The policy, if successful, will protect users from internalizing attitudes that would be inauthentically held." (p.38) In fact, the main concern of most in the literature is whether the epistemic paternalism is justified -the effectiveness or benefits of such strategies are generally taken as a given -but, as I mentioned earlier, I am mainly setting this issue aside.
Second, there is empirical data that suggests epistemic paternalism is an effective strategy in combatting CCIs. Chiou and Tucker (2018) write, "After Facebook's ban on advertising by fake news sites, the sharing of fake news articles on Facebook fell by 75%." (p.1) There seems a very simple diagnosis here: agents didn't see the articles anymore, so they didn't share them. Shen and Rosé (2022) discuss the "quarantining" of two subreddits,/r/ The_Donald and/r/ChapoTrapHouse on Reddit. Quarantining is a process Reddit can use that limits access to particular subreddits without removing them. While this is not as extreme as no-platforming, they found that it did reduce the number of new users and the popularity of posts. A full noplatforming seems certain to be even more effective. Rauchfleisch and Kaiser (2021) write in their study on no-platforming: Our analysis shows that deplatforming is effective in minimizing the reach of disinformation and extreme speech, as alternative platforms that will allow this kind of content cannot mitigate the negative effect of being deplatformed on YouTube.
(p.1) Innes and Innes (2021) look at the no-platforming of David Icke and Kate Shemirani -both of whom had extremely popular online presences prior to their no-platforming. Their research also suggests that no-platforming limits the spread of misinformation (but with the caveat that it does not eliminate it entirely). Jhaver et al. (2021) take a similar line of research, looking at the Twitter no-platforming of Alex Jones, Milo Yiannopoulos, and Owen Benjamin. Again, their research showed no-platforming to be an unqualified success in massively reducing their online impact. Overall, empirical data strongly suggests that no-platforming and censorship of bad actors on social media is effective in reducing their reach and thus reducing the spread of misinformation and conspiracy theories. So, on the whole, I propose we have at least some reason to think that epistemic paternalism could be an effective strategy for dealing with CCIs.

Premise (2)
(2) is a normative claim about how we would want those enacting epistemically paternalistic policies to operate. First, let's get clear on what "best epistemic interests" means. In the epistemic paternalism literature, the "epistemically better off" in the improvement condition is generally understood along veritistic lines (roughly, the idea that true belief is paramount). 16 For my "best epistemic interests" I think understanding this along veritistic lines would be adequate, but I also see no reason why it couldn't be thought of in terms of promoting understanding or developing intellectual virtues. I don't think anything in the puzzle hinges on a specific conception of "best epistemic interests" therefore I am happy to take a broad, pluralistic approach.
Next, let's turn to the "should" that features in premise (2). It's important to note that this is a practical "should" 17 specific to a role, where its denial would imply negligence. Suppose we were talking about rubbish collection and the second premise looked something like: Those who collect rubbish should be well-positioned to collect rubbish reliably.
That is to say, they should possess the means to collect rubbish, be equipped with the ability to do so, have the opportunity to exercise the ability, and so on. If they purport to be occupying the rubbish collector role but fall short in any of these ways, then there is a (role-specific) normative defect here; they would be alleging to occupy a role that requires doing something that they are not well-positioned to do. It is in this normatively narrow, role-specific sense that rubbish collectors should be well-positioned to collect rubbish reliably. This runs analogously with the puzzle at hand here. A precondition for enacting epistemically paternalistic policies is that those who enact them should be well-positioned to act reliably in the interferees' best epistemic interests. If not, then they are failing in their role as paternalistic-policy-enactors.
Being well-positioned to act reliably in an interferee's best epistemic interests is clearly not necessary to act epistemically paternalistically tout court toward someone; imagine a systematically unreliable teacher who sincerely does not care about their students' epistemic status but nevertheless non-consultatively interferes in their inquiry in some way and accidentally improves them epistemically. The key point rather is that these powers of no-platforming and censorship, while plausibly effective at dealing with CCIs, have massive scope to be used (and abused) beyond this. In a sense, this premise acts as a sort of safety net to prevent these abuses of power; we want to ensure that those enacting such policies with these powers at the very least are well-positioned to act reliably in interferees' epistemic interests. The problem, as stated in (4), is that these institutions are not.

Premise (4)
(4) is a descriptive claim about the reality of how these institutions are not well-positioned to act reliably in interferees' best epistemic interests. I suspect that such a claim enjoys significant intuitive support, but we can offer some supporting evidence for it regardless. For instance, a whistleblower report from inside Facebook revealed that they knew that their content algorithm pushed users into further radicalization and promoted the growth of QAnon but did not ameliorate the issue. 18 Or a 2018 study which found that false stories are 70% more likely to be retweeted on Twitter than true ones (Vosoughi et al., 2018). Plausibly, a social media site's main interest is keeping users on the site, and if misinformation and conspiracy theories are what is doing that, then rectifying these issues is unlikely to be a priority and thus they will not reliably act in agents' best epistemic interests. Additionally, these large institutions are companies that are constrained in how they can act. For instance, they have fiscal responsibilities to shareholders. If such fiduciary responsibilities (e.g., keeping users on the site) conflict with interferees' best epistemic interests, then the fiscal responsibilities will win out every time.
Governments don't fare much better. The infamous Brexit bus, plastered with the false claim 19 that the UK sends £350 million to the EU every week and instead this money could be used to fund the NHS is an obvious instance of a government sacrificing its peoples' epistemic health for political gain. Or consider the manipulated evidence used by governments to justify the Iraq War. Empirical evidence on governmental abuses of the public's epistemic health is widespread. And, of course, the two UK governments previously discussed are democratic governments without ultimate control over the flow of information in the epistemic environment. Authoritarian governments that did (or do) have this power of non-consultative interference over the public certainly did not use it for the public's best epistemic interests. For instance, the current propaganda campaign in Russia supporting the war in Ukraine. Or consider the Chinese government's mass cover-up and hiding of the Tiananmen Square Massacre. Democratic or non-democratic, history is littered with governments from all over the world not acting in their constituents' best epistemic interests and so I argue that these institutions are not wellpositioned to act reliably in interferees' best epistemic interests.
So, with the premises justified, the puzzle of epistemic paternalism stands. We have a good strategy to combat CCIs, but no way of using it.

Why does this matter?
What is the import of having identified this puzzle? 20 As a reductio, plausibly the best way to prevent murder is an extremely advanced and intrusive police state where everyone's actions are carefully tracked and monitored, and privacy is a thing of the past. Despite this being an effective strategy to combat murder, we would not want to use it for obvious reasons (breaches of human rights for one), and it does not seem that there's much of a puzzle here nor much interesting to discuss. Is the puzzle I have identified importantly different?
Unsurprisingly, I think yes. No matter how it works out in preventing murder, the police state has inherent problems from the off (such as the human rights abuses I mentioned). The epistemic paternalism I have discussed only has potential problems from potential abuses. Were it guaranteed that the powers of no-platforming and censorship only be used for good (e.g., combatting CCIs) then it seems there would be no worries about implementing epistemically paternalistic policies. The problem -and puzzle -comes from the fact that we can't guarantee that. In fact, we can nearly guarantee the opposite. As I mentioned earlier, I suspect that the puzzle generalizes to all sorts of misinformation, conspiracy theories, and fake news, so should it turn out that epistemic paternalism is indeed the best way to tackle these problems, it's obviously problematic that what goes along with this finding is that it also can never be used -seemingly because of contingent worries.
In fact, I think that when we relate this epistemic paternalism to misinformation and conspiracy theories, the puzzle may be unavoidable. Let's assume that there is some connection between untrustworthy institutions and the growth of conspiracy theories insofar as when our institutions are untrustworthy, the more conspiracy theories will abound. In this scenario, epistemic paternalism is plausibly an effective way of combatting these conspiracy theories, but it cannot be used as the untrustworthy institutions won't be well-positioned to act reliably in interferees' best epistemic interests. In a scenario where our institutions are trustworthy, there would plausibly be less conspiracism, and so epistemic paternalism would probably not be needed anyway. So, whenever epistemically paternalistic policies may be needed, they cannot be used, and if they ever could be used, they're probably not needed.
To briefly summarize, I began in this section by giving the puzzle before motivating the remaining premises. I then considered the import of the puzzle and suggested it may be unavoidable when considered in the context of dealing with CCIs, misinformation, conspiracy theories, and so on.

Responses
I will now look at some potential objections and responses to the puzzle. I will first discuss whether it is the case that only large institutions can do the job of epistemic paternalism. I then turn to whether removing misinformation constitutes an instance of epistemic paternalism. Finally, I will consider a possible category error in that potential abuses of no-platforming and censorship seem to no longer actually be epistemic paternalism. Nonetheless, I will show that none of these objections prove fatal to my puzzle.

Can only institutions do the job of epistemic paternalism?
Perhaps one could argue that this extremely strong and widespread institutional response to CCIs is unnecessary and that instead individuals can do the job of demoting CCIs themselves. Each agent has their own responsibility to combat misinformation and conspiracy theories and prevent their epistemic peers from falling down the rabbit hole of these false beliefs. As a libertarian would perhaps argue that governmental interference is unnecessary in fixing potholes or building bridges, analogous considerations apply to no-platforming and censorship -we can do it ourselves. This also helpfully avoids the legitimate worries about giving powerful institutions informational control. I don't find this a particularly compelling response, however. First, I think that this is essentially how things currently are; governments (sans epistemically paternalistic policies) don't really have much (if any) power to halt the spread of misinformation and conspiracy theories, and corporations, while having almost full autonomy over the running of their own platforms, apply their rules inconsistently and (often) ineffectively. 21 And, as I have mentioned throughout, we are currently in an infodemic; misinformation and conspiracy theories are extremely widespread in the epistemic environment, plausibly indicating the ineffectiveness of individualistic responses to CCIs. Of course, agents should do their best to combat misinformation in any way they can (say, fact checking peers, reporting misinformation on social media, and so on) but I think the present-day situation of widespread conspiracism demonstrates that this is not enough, and some sort of institutional response is necessary.

Is removing misinformation epistemically paternalistic?
Suppose I knew someone was about to lie to you and so before they could speak, I quickly silenced them in some way or another, preventing them from saying their lie. Is this an instance of epistemic paternalism? Plausibly not. Intuitively, this might not be considered interference per se, perhaps it's just an instance of a morally praiseworthy act. The removal of misinformation appears to go analogously: someone's inquiry is interfered with by them not encountering evidence that is, crucially, false. After all, plausibly what is interesting about epistemic paternalism is it is something prima facie epistemically bad (hiding true information from an agent) but it results in something epistemically good (their epistemic benefit). If removing falsehoods does not qualify as epistemic paternalism, then the puzzle obviously dissipates.
This response does not hold much weight, however. For a start, earlier I noted that misinformation isn't always strictly false -it can be true information manipulated in such a way to mislead an agent. So, by the lights of this objection, such instances where this sort of misinformation is censored would qualify as an epistemically paternalistic interference. Our response can also be stronger than this anyway. Recall the interference condition on epistemic paternalism: "it interferes with the freedom of inquirers to conduct inquiry in whatever way they see fit." (Ahlstrom-Vij, 2013, 61) There is clearly no judgment here on whether the interference is a suppression of true or false information. All that matters is that you interfere with agents' inquiries in some way or another such that their autonomy is violated. 22 Someone may want to inquire in such a way that includes all information, and so the removal of false information still qualifies as an interference and thus still qualifies as epistemic paternalism.
A rejoinder to my response here could be a case such as this: your friend habitually violates Grice's maxims 23 and you know this. Imagine someone asks your friend where the nearest toilet is. You know your friend is going to say "it's just round the corner" but neglect to mention that this toilet is closed. You do something to stop them saying anything in the first place. This doesn't seem paternalistic (at least, it seems no more paternalistic than when you stop someone lying) yet in this example what they are going to say is entirely true. 24 This is an interesting case, but it doesn't prove worrisome. I argue that this is unproblematic as it simply doesn't qualify as an epistemically paternalistic act. Recall the improvement condition, "interferes -exclusively or not -for the purpose of making those interfered with epistemically better off." (Ibid., 61) So, a non-consultative interference that isn't for an agent's epistemic improvement is not an epistemically paternalistic act. Applied to this case, our interference is actually making our friend epistemically worse off as they are not learning some piece of knowledge (the location of the bathroom). Ultimately, this is likely for their benefit as it prevents them from having false hope and wasting their time (because the toilet is out of order) but, crucially, this benefit is not an epistemic one. Arguably, this counts as an act of paternalism (you're trying to help the person not waste their time and energy) but not its epistemic cousin.

Possible category error
Let's briefly take a step back and think about the root of the problem. What the puzzle highlights is a general worry we would have if the state or other powerful institutions had these epistemically paternalistic powers of noplatforming and censorship. The worry comes from a concern about potential abuses of these powers. Let's look at an example case: Corrupt government. During the Tovid-25 pandemic, misinformation and conspiracy theories abounded to such a level that it seriously hindered the Schmottish government's response -it adversely affected vaccine uptake, mask wearing, and following restrictions. In response to this, legislation was passed which granted the Schmottish government epistemically paternalistic powers of no-platforming and censorship. This was an unqualified success and reduced misinformation and conspiracy theories to such a level that the government's response was back on track, and they beat the pandemic. Unfortunately, soon after, the Schmottish government became massively corrupt and began using their epistemically paternalistic powers for nefarious purposes -for instance, the silencing of journalists pointing out their corruptness. This meant that they could continue in their corrupt ways without consequence. This sort of case seems the paradigmatic worry behind institutional control of the flow of information. While epistemic paternalism was extremely successful in tackling misinformation and conspiracy theories, the same strategies were reused later to conceal corruption, hide true information from the public, and, importantly, make them epistemically worse off. However, there is a problem here. In the earlier instance when the Schmottish government was using epistemically paternalistic policies to combat Tovid-25 misinformation, this was indeed a bona fide instance of epistemic paternalism; the Schmottish government non-consultatively interfered with agents' inquiries to make them epistemically better off (i.e., so they would not fall prey to misinformation about Tovid-25). In the later instance of the Schmottish government's no-platforming and censorship, however, they were indeed non-consultatively interfering but, crucially, it was not for any agents' epistemic benefit. In fact, it was for the exact opposite reason! So, by this failure to meet the improvement condition, it is no longer even an instance of epistemic paternalism. Thus, if the epistemically paternalistic policies are abused, they are by definition immediately disqualified from even being epistemically paternalistic. The upshot here is that it's not clear what the puzzle is even about anymore -or if there even is a puzzle at all.
There are a few avenues of response here. First, I think any discussion of epistemic paternalism will be (academically) uninteresting if it is built in that for it to qualify as epistemic paternalism it must be good simpliciter. This also seems to go against general intuitions surrounding epistemic paternalism. For instance, the widespread suspicion of epistemic paternalism appears strangely unfounded in the face of this necessary goodness.
For this reason, perhaps an account of epistemic paternalism that only has the interference condition and the non-consultation condition and lacks the improvement condition 25 more closely captures our intuitions about the sorts of epistemic paternalism discussed in this paper. Without the improvement condition, fears about the abuses of epistemically paternalistic policies are no longer a category error and are a salient concern. This may seem like a strange account of epistemic paternalism to endorse considering that paternalism as a whole seems to be primarily about improving agents in one way or another. However, from this discussion above, I think we have prima facie reasons to think that this two-condition account could be appropriate. It makes sense of our fears of epistemic paternalism outlined in Corrupt Government and it also plausibly accounts for the general suspicion of epistemic paternalism.
Even if we don't want to commit to this account however, I think there's a final way of responding to this category error objection. Remember that we're talking about epistemically paternalistic policies here. Although when they overstep they may no longer strictly count as epistemic paternalism, they're still epistemically paternalistic in nature (and description). Indeed, the reason for their inception was in the name of epistemic paternalism. Noplatforming and censorship are epistemically paternalistic when responding to CCIs and, as implemented policies, it is an area worthy of discussion what this might look like when abused even if it would seem to remove them from the specific category of epistemic paternalism (at least as Ahlstrom-Vij conceives of it anyway). The potential problems that could arise from abuses of the epistemically paternalistic policies could not occur if the policies did not exist, and the reason the policies exist is because of epistemic paternalism. Analogously, suppose someone had a terrible drinking problem and it eventually led to them injecting alcohol into themselves. It would be asinine at this point to simply shrug and say, "well, they don't have a drinking problem anymore, they have an injecting problem." The issues are clearly connected even though injecting alcohol is not strictly drinking alcohol -the injecting issue only exists because of the initial drinking problem. Thus, I argue the puzzle is still one of epistemic paternalism irrespective of whether we might think that an abuse no longer strictly qualifies as an instantiation of epistemic paternalism. And, most importantly, the puzzle remains of interest.

Concluding remarks
The question of the best or most effective response to Covid-19 collective irrationalities is not one that can easily be answered. I proposed that a plausible solution is that of widespread epistemic paternalism through the practices of no-platforming and censorship. I argued, however, that pursuing this strategy leads to a previously unnoticed puzzle for those who may support the implementation of epistemically paternalistic policies, which established that epistemic paternalism cannot be used. I considered some responses and objections but found each of them wanting, concluding that the puzzle stands. This makes for a troubling conclusion, however. Conspiracy theories, fake news, and misinformation do not appear to be going away any time soon, and, if my puzzle is correct, then our best weapon against these worrisome issues may prove to be unusable and Covid-19 collective irrationalities (and any new collective irrationalities that will undoubtedly arrive in the future) will continue to be widely propagated among agents in the epistemic environment. 26 Notes 1. Percentages of vaccine uptake in the population: US 67.5%, Bulgaria 30%, Croatia 55.5%, the UK 74.7%. Of course, some of the unvaccinated will not be eligible for some reasons or another, but they do not make up the entirety of the millions who are still not vaccinated. 2. See Satariano and Alba (2020). 3. Here I am understanding "conspiracy" with the neutral definition that is most popular in the literature: "1. There exists or existed some set of agents with a plan; 2. Steps have been taken by the agents to minimize public awareness of what they are up to; 3. Some end is (or was) desired by the agents." (Dentith & Keeley, 2018, 285) A conspiracy theory is just a theory that, "posit[s] the existence of a conspiracy, where the conspiracy is the salient cause of some event." (Dentith, 2016, 577) For a competing account, see Cassam (2019). For an interesting new discussion related to conceptually engineering the term "conspiracy theory", see Napolitano and Reuter (2021). 4. The panic buying of supplies at the beginning of the pandemic is an example of a collective irrationality that epistemically paternalistic policies could have helped to quell but it does not appear to be a conspiracy theory (or even necessarily misinformation). Similar considerations apply to protests against lockdowns and/or Covid-19 in general that happened throughout the pandemic. In those cases, the catalyst may have been conspiracy theories or misinformation but the actual act of the protest is merely a collective irrationality -and, again, something that paternalistic policies could help prevent. (2021) for more on this. 6. Outwith philosophy, the term mainly used is "deplatforming". As far as I can tell, both refer to the same thing, but I will stick with "no-platforming" in this paper. 7. While I am assuming here that epistemic paternalism is at least sometimes justified, my paper can indeed contribute to the debate of epistemic paternalism's justificatory status in general. If the puzzle turns out to be true, then it is possible that powerful organs of society are unjustified in acting in widespread epistemically paternalistic ways. My thanks to an anonymous reviewer for pointing this out. 8. See Goldman (1991) and Croce (2018). Goldman thinks experts are justified to undertake epistemically paternalistic actions. Croce argues against this and proposes epistemic authorities instead. 9. My thanks to an anonymous reviewer for raising this point about science communicators. 10. I am taking "epistemically paternalistic policies" to mean formalized policies of noplatforming and censorship. 11. The focus of the puzzle here is CCIs but I see no reason why this puzzle wouldn't generalize to all brands of misinformation, conspiracy theories, fake news, and so on. 12. This isn't to say that individuals have no power whatsoever. Reporting misinformation online or fact-checking acquaintances who are (inadvertently or not) repeating misinformation could potentially help with combatting CCIs, but these strategies are never going to be as effective as a strong institutional response that affects entire populaces. I address this thought in more detail in section 3.1. 13. This is not to say that epistemic paternalism is the only way of combatting CCIs. For example, as suggested in Rini (2017), third-party fact-checking agencies such as Snopes or FullFact.org could be effective in tackling misinformation. A new metaanalysis (Pennycook & Rand, 2022) does suggest that these sorts of accuracy prompts are indeed effective in reducing misinformation (although other studies disagree, e.g., Thorpe et al. (2021)). Either way, my argument is not predicated on epistemic paternalism being the sole way of combatting misinformation, only that we have reason to think it may be an effective one. 14. One might think the concern here is more about the reputation of the host party as opposed to any altruistic, paternalist concern with the epistemic environment or wellbeing of others. This is an interesting reading of Levy that bears some discussion. Certainly, some acts of no-platforming or censorship could be done out of these more selfish concerns (and no doubt there are empirical examples of these) but if they are, then they simply no longer would qualify as epistemic paternalism as they do not meet the improvement condition (interference for the purpose of making the interferee epistemically better off) and thus do not prove too relevant to my discussion here. This sort of idea, and the potential category errors that could arise therein, is discussed in detail in section 3.3. My thanks to an anonymous reviewer for raising this point about reputational concerns. 15. Fantl also discusses potential "psychological harm" (184-188) and "intrinsic harm" (189)(190)(191)(192)(193)(194)(195)(196)(197) that can be done to agents by platforming problematic speakers. 16. See Pritchard (2013) for a different conception of epistemic value in relation to epistemic paternalism. 17. A different sense of "should" would be whether institutions should try to epistemically improve agents in the environment -i.e., is it their duty to help in this fashion? For general discussion on this sort of idea, see Goldberg (2017). For a more specific (although individualistic) discussion on our duties toward conspiracy theories and misinformation, see Terzian and Corbalan (2021).

See Lyons
18. See Gilbert (2021). 19. See Full Fact Team (2017. 20. While the general problem pointed toward in this paper (can those with the power to change things for the better actually be trusted to do so) is a familiar one, there is import in my identifying of this puzzle. We have seen that epistemic paternalism seems to be an effective strategy in tackling misinformation and conspiracy theories, and, crucially, misinformation and conspiracy theories are endemic in the epistemic environment. Therefore, we have reason to pursue the implementation of epistemically paternalistic policies. The significance of my puzzle is that it points toward a problem that has been thus far overlooked by proponents of epistemic paternalismnamely that, irrespective of how effective epistemically paternalistic policies may be in combatting misinformation and conspiracy theories, such policies cannot be enacted because the institutions who could enact these policies are not well-positioned to act reliably in the agents' best epistemic interests. 21. For instance, many have argued that Donald Trump consistently violated Twitter's content guidelines surrounding misinformation and abuse for years before he was finally no-platformed. 22. For more on this sort of interference and how ubiquitous it actually is, see Medvecky (2020), pp. 82-84. It is important to remember here that not any interference counts as epistemic paternalism as we still need to meet the other two conditions. 23. See Grice (1989), pp. 24-31. Briefly, the four maxims are: quantity -be informative, quality -be truthful, relation -be relevant, and manner -be clear, brief, and orderly. 24. My thanks to an anonymous reviewer for suggesting this case. 25. This is an idea that McKenna (2020) briefly mentions in footnote 7 of his paper. 26. Many thanks to two anonymous reviewers for extremely helpful comments on previous versions of this paper. I would also like to thank Adam Carter and Emma Gordon for all their amazing help and advice. And thanks to Pinelopi Stylianopoulou, Ross Patrizio, and audiences at the University of Glasgow Postgraduate Seminar and at the Scottish Epistemology Early Career Researchers (SEECRs) WiP session for helpful discussion. This paper has received generous funding from the Scottish Graduate School of Arts and Humanities.

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