2. The relationship-modification account of epistemic blame

In what follows, I will present the RMA of epistemic blame and elaborate on the central role it assigns to trust reductions. Let’s start with an initial characterization of epistemic blame. It is commonly regarded as a response to an epistemic failing, not to its practical downstream effects (Boult Reference Boult2024c, p. 31). Epistemic blame is taken to go beyond mere negative evaluation in that it involves a particular ‘sting, force or depth’ (Hieronymi Reference Hieronymi2004, pp. 116–7) and in that the blamer is engaged by his response (Piovarchy Reference Piovarchy2021, p. 794; Boult Reference Boult2020, p. 519; Reference Boult2024c, 19). Finally, some have argued that epistemic blame typically does not involve ‘hot’ reactive attitudes (e.g., indignation or resentment), but rather ‘cool’ responses (Boult Reference Boult2020), such as trust reductions.

The RMA of epistemic blame has been extensively developed by Cameron Boult in his recent book and a series of papers. RMA is inspired by Scanlon’s (Reference Scanlon2008) relationship-modification account of moral blameFootnote ³ and has various moving parts, which I will present in what follows.Footnote ⁴ Here is a first approximation of RMA’s conception of epistemic blame (Boult Reference Boult2024c, p. 66).Footnote ⁵

Let’s unpack. According to RMA, first, one modifies one’s epistemic relationship to $S$ by modifying one’s intentions and expectations toward $S$ . The paradigmatic kind of modification is the reduction of epistemic trust (Boult Reference Boult2020, p. 521, Reference Boult2021a, p. 6, Reference Boult2023, p. 817, Reference Boult2024b, p. 391; Woodard Reference Woodard and Shafer-Landau2024, p. 172; Schmidt Reference Schmidt2024, p. 18; Flores and Woodard Reference Flores and Woodard2023, p. 2557). For instance, if we epistemically blame $S$ by reducing trust in her regarding some topic $t$ , we expect her to be an unreliable informant with regard to $t$ and intend to lend less credence to her $t$ -regarding assertions. Second, a judgment that $S$ is epistemically blameworthy is understood to be a judgment that $S$ has done something that impairs their epistemic relationship with the judging agent. Third, $S$ has impaired an epistemic relationship iff S falls short of the normative ideal of that relationship. Kicking the conceptual can just one step further down the road, the normative ideal of epistemic relationships is thought to be constituted by both general and particular epistemic expectations.Footnote ⁶ Drawing on work from Goldberg (Reference Goldberg2017, Reference Goldberg2018), Boult assumes that we have a general epistemic expectation that others are both epistemically trustworthy and trusting, that is, that they are reliable informants and defer to other reliable informants under appropriate circumstances. There is a corresponding practice of mutual epistemic trust in each other’s testimony, the purpose of which is to facilitate the division of epistemic labor and protect the epistemic commons. Additionally, there are also more particular epistemic expectations – between, for example, teachers and students, doctors and patients, politicians and voters, friends, family, and so on (Boult Reference Boult2023, p. 820, Reference Boult2024c, ch. 6.5) – that are grounded in professional, institutional, or miscellaneous other relationships.

For illustration, let’s apply RMA to our initial case, Fortune. Stephen falls short of the normative ideal of his epistemic relationship with Amari by consulting and believing the fortune-teller concerning financial matters. He violates Amari’s (and general) epistemic expectations by using an unreliable method to form beliefs about the topic in question, namely by trusting the testimony of a fortune-teller. Amari judges that Stephen has thereby impaired his epistemic relationship with her, and she fittingly adjusts her intentions (to not believe him on financial matters) and her expectations toward him (that he will be unreliable on financial matters in the future) to reflect this, thereby reducing epistemic trust. In doing so, according to RMA, Amari epistemically blames Stephen for his epistemic failing.

Epistemic relationship modifications seem to be well-suited to capture the abovementioned characteristics of epistemic blame. First, the modification is made fitting by one’s judgment about the epistemic failing, not its downstream effects.

Second, RMA explains what separates epistemic blame from mere negative evaluation: holding others to normative expectations and altering one’s epistemic relationship if these expectations are frustrated goes beyond giving someone ‘a bad epistemic grade’ (Boult Reference Boult2024c, 83); rather, it is a way of being engaged by one’s response to epistemic failings (ibid., §4.1). Finally, RMA explains the putatively ‘cool’ nature of epistemic blame: as Fortune and other cases illustrate, ‘cool’ -headed trust reductions are often a fitting way to modify one’s epistemic relationships.

RMA’s account of epistemic blame has faced two major challenges from extensional inadequacy. I will briefly introduce these challenges now and consider extant responses in more detail later on. The first challenge is the epistemic analog of a well-known challenge from Smith (Reference Smith, Coates and Tognazzini2013, p. 375) to Scanlon’s account. As Chislenko (Reference Chislenko2020, p. 375) puts the original worry, the relationship-modification account says that

Likewise, if we respond in similarly positively valenced ways to epistemic failings, for example, with curiosity or understanding, and judge these responses to be fitting (i.e., judge that they are fitting responses to someone’s epistemic $\varphi $ -ing falling short of the normative ideal of the respective epistemic relationship), then RMA would count them instances of epistemic blame. In Fortune, for instance, it seems fitting to respond to Stephen with curiosity: why does he believe a fortune-teller, and why only concerning financial matters? Likewise, Boult (Reference Boult2024c, p. 185) discusses a case in which one is concerned about a friend who has been led down a YouTube conspiracy rabbit hole, and where showing understanding can be a fitting response to the friend’s epistemic failing (although Boult argues that this response doesn’t constitute epistemic blame; more on this later). But responding with curiosity or understanding to someone’s epistemic failing is intuitively not a way of epistemically blaming them.

The second challenge has been forcefully raised by Smartt (Reference Smartt2023, p. 1820f) and concerns trust reductions in non-culpable agents (see also Piovarchy Reference Piovarchyforthcoming). Consider the following case from Greco (Reference Greco2021, p. 531):

We judge that the mother has fallen short of general epistemic expectations and fittingly modify our epistemic relationship to her by reducing epistemic trust. Generally speaking, we sometimes seem to fittingly respond to the epistemic failings of non-culpable agents by reducing trust in them. Yet, it is implausible that we thereby blame them. Again, various responses have been given to this challenge. I will consider one of them now, and the others in more detail later on.

The response in question claims that we don’t reduce epistemic trust in non-culpable agents. Both Schmidt (Reference Schmidt2024, §4.2) and Boult (Reference Boult2024b, fn.11) have suggested that we might merely be less inclined to rely on non-culpable agents, without reducing trust in them. This proposal strikes me as a nonstarter: the orthodox view is that trust is reliance ‘plus some extra factor’ (Hawley Reference Hawley2014, p. 5; Goldberg Reference Goldberg and Simon2020).Footnote ⁷ Hence, there can’t be a reduction in one’s willingness to rely without a reduction of one’s willingness to trust. Those who deny this owe us an alternative story about the relation between epistemic trust and reliance.

I will argue that ultimately, both challenges arise because RMA incorrectly conceptualizes trust reductions as paradigmatic instances of epistemic blame. According to RMA, trust reductions are paradigmatic ways of blaming ourselves and others for epistemic failings. In what follows, I will develop a novel account of trust reductions that shows that this view is mistaken.

Let me briefly sketch where I think RMA goes wrong. Proponents of RM think that trust reductions are paradigmatic epistemic blame responses because we often respond with trust reductions to culpable epistemic failings. However, I will argue that the latter is insufficient evidence for thinking that trust reductions (at least those fitting a corresponding judgment of blameworthiness) are paradigmatic manifestations of epistemic blame. Rather, by drawing on resources from legal theory, I will show that there is excellent evidence to think that trust reductions have been misidentified as (broadly conceived) culpability-tracking responses and rather should be understood as distinct responses – non-punitive measures – that can accompany culpability-tracking responses, but are not (partially) constitutive of the latter. My novel account of trust reductions as NoPEMs gets this right and offers a principled way to address RMA’s challenges and assess extant responses.

3. Non-punitive measures

In what follows, I will draw on rich conceptual resources from legal theory and legal codes to elaborate this dual system of responses. The mediate purpose of doing so is to extract characteristics and general application conditions for non-punitive measures. This prepares my subsequent argument that trust reductions exactly match these characteristics and application conditions in the epistemic domain.

Both common law traditions and (particularly) continental law traditions recognize, to various extents, a dual system of responses to legal ‘failings’, that is, unlawful conduct.Footnote ⁸ The first, familiar kind of punitive sanctions (fines, jail time, and so on) serve, depending on your favorite theory of punishment, the purposes of either retribution, restitution, denunciation, harm reduction (e.g., by deterrence), rehabilitation, the strengthening of social values or norm recognition, or, commonly, a pluralist mix of these.Footnote ⁹ Punitive sanctions are responses to culpable unlawful conduct. They impose on the culpable agent negative consequences that are ordinarily considered to be unpleasant (Dressler Reference Dressler2015, §2.02). They thereby express a negative evaluation of the agent and their conduct, where the strength of the negative evaluation is thought to correspond to the severity of punishment.

The second, perhaps less familiar, type of response is non-punitive measures (‘non-punitive sanctions’), which are also responses to unlawful conduct that impose negative consequences. However, non-punitive measures don’t presuppose culpability. They also don’t express a negative evaluation of the agent and their conduct. The sole purpose of non-punitive measures is the individualized (‘special’) prevention of future unlawful conduct by the agent and (sometimes) also the agent’s rehabilitation. For instance, consider a schizophrenic agent who suffers from severe hallucinations and assaults others because of his condition. Responding to his unlawful conduct by committing him to a psychiatric hospital not only aims at preventing further unlawful conduct caused by his condition, but also aims at therapeutically rehabilitating the agent. Importantly, it is sometimes appropriate to respond to unlawful conduct with both punitive sanctions and non-punitive measures. Being a response to unlawful conduct is hence insufficient to classify a response either way. Rather, we also need to look at the respective responses’ features and purposes.

I will argue that epistemologists would benefit from taking notice of the rich and sophisticated dual classification of responses to unlawful conduct in legal theory and codes. Once we do, we will see that classifying trust reductions as epistemic blame responses is erroneous.

Taking inspiration from other normative domains, most notably ethics, to develop accounts of responses to epistemic failings is not a new approach (cf. Boult Reference Boult2021a, §3.1). For instance, Brown (Reference Brown2018, Reference Brown2020), Piovarchy (Reference Piovarchy2021), and Boult (Reference Boult2020, Reference Boult2023, Reference Boult2021b, Reference Boult2024b,Reference Boultc) have drawn on accounts of moral blame from Sher (Reference Sher2005), Vargas (Reference Vargas2013), and Scanlon (Reference Scanlon2008), respectively, to develop accounts of epistemic blame. Why draw on legal theory? One reason is the noteworthy social turn in epistemology and the increasing interest in our epistemic practices. A common thought is that our epistemic assessments are governed by epistemic norms that are grounded in social practices, the purpose of which is to facilitate our division of epistemic labor and protect the epistemic commons.Footnote ¹⁰ In the light of this metaepistemological backdrop, the analogy to the legal domain is only natural. Legal conceptualizations of our responses to unlawful conduct are grounded and refined in long-running social practices. These practices aim, among other things, at facilitating and protecting social commons. Furthermore, legal practices are often explicitly codified to a high degree of systematicity and sophistication, making them particularly powerful conceptual blueprints to draw inspiration from in epistemological theorizing. These broadly Austinian (Reference Austin1957, p. 8) reasons suggest that epistemologists would benefit from taking an interest in legal conceptualizations and classifications of responses to failings.Footnote ¹¹

With the general approach motivated, let me now introduce the notion of a non-punitive measure in more detail. For illustration, I will draw on the particularly rich codification of non-punitive measures in the German criminal code (StGB).Footnote ¹² Presenting them selectively in more detail will serve the aim of extracting their general characteristics and application conditions, which I will then deploy in the epistemic domain in the next section.

Consider again being committed to a psychiatric facility (§63), which is (very roughly) appropriate only if an agent committed an unlawful act, is not (fully) culpable for their conduct due to various psychiatric, drug-related (e.g., addictions) or other cognitive conditions (e.g., intellectual disabilities) and is expected to commit further (severe) crimes because of their condition. The specific purpose of this non-punitive measure is to prevent further unlawful conduct caused by the condition and, if feasible (van Gemmeren Reference van Gemmeren, Erb and Schäfer2020), to improve the condition by therapeutic treatment.

While commitment to a psychiatric facility presupposes the absence of (full) culpability for the committed unlawful conduct, non-punitive measures do not generally presuppose this. Consider, for instance, the not uncontroversial non-punitive measure that is preventative detainment (§66). Depriving an agent of his liberty via preventative detainment is (very roughly) appropriate only if, among other things, the agent has already been sentenced for severe crimes in the past, is sentenced for another severe crime, and, based on an overall evaluation of his character and his crimes, is expected to be a continued danger to society.

Or consider the loss of one’s driver’s license (§69), which can be assigned irrespective of culpability and does not involve a rehabilitative dimension.Footnote ¹³ This non-punitive measure is appropriate for agents who are expected to not safely operate their vehicle because, for instance, they manifest certain excusing conditions (e.g., certain psychiatric or medical conditions) or display a lack of ‘aptitude of character’ (e.g., by engaging in illegal street racing).

While non-punitive measures differ in various ways from each other, some common, general application conditions can be extracted. First, non-punitive measures are responses to unlawful, that is, negatively evaluated, conduct that, second, impose negative consequences ordinarily considered unpleasant. Third, non-punitive measures, in general, can be appropriate irrespective of culpability, that is, they can be applicable in response to both culpable and non-culpable unlawful conduct. Importantly, they can accompany punitive sanctions. Suppose $A$ engages in illegal street racing with their car and gets caught by the police. Not only might $A$ get punitively sanctioned, but they might also be subject to a non-punitive measure: engaging in illegal street racing might reveal that they lack the needed awareness or care for the risk involved in driving, making it appropriate to non-punitively deprive them of their driver’s license. Fourth, there needs to be a (reasonable) expectation of future unlawful conduct, which is grounded in an ‘overall evaluation’ of the agent. Indeed, the initial unlawful conduct has to reveal something about the agent’s character, dispositions, and social environment in virtue of standing in a ‘symptomatic relation’ (Meier Reference Meier2019, p. 299) to the expected future unlawful conduct.

Non-punitive measures are tightly constrained by considerations of proportionality. One such constraint is that the expected unlawful conduct has to be sufficiently likely and sufficiently severe. For instance, preventative detention is a grave interference with one’s rights and liberties; so, it is only an appropriate response if the expected unlawful conduct surpasses particularly high barriers concerning likelihood and severity of the expected unlawful conduct.Footnote ¹⁴ Relatedly, proportionality requires that if the condition that caused or contributed to the unlawful behavior was only temporary (or ceases to persist at some point), then nonpunitive measures are not appropriate (or their continued appropriateness has to be reevaluated and the measure possibly suspended).

Finally, although the primary and characteristic purpose of non-punitive measures is prevention, some non-punitive measures have the subordinate purpose of rehabilitation. Whether this subordinate purpose translates into rehabilitative efforts, and if so, to what extent, depends on the particular measure and the likelihood of successful rehabilitation.Footnote ¹⁵

Summing up, non-punitive measures are (1) responses to unlawful conduct, (2) impose negative consequences, (3) don’t generally presuppose culpability, and (4) are only appropriate if there is a persistent expectation of sufficiently likely and severe future unlawful conduct. (5) They also sometimes involve a rehabilitative dimension.

In the next section, I will present a novel account of trust reductions, according to which trust reductions are NoPEMs. Besides showing that my account is explanatorily powerful, I will argue that, tellingly, the application conditions for trust reductions identified by proponents of RMA straightforwardly match those just extracted for non-punitive measures.

4. Non-punitive epistemic measures

Following Boult (Reference Boult2024c), I will assume a broadly communal metaepistemology as a backdrop for my discussion. Recall that on such a view, the purpose of our assessments and responses to epistemic conduct is, broadly conceived, to facilitate the epistemic division of labor and protect the epistemic commons. We collectively care about epistemic goods being widely available in our epistemic community, and we distribute them via relations of mutual epistemic trust, expecting each other to both be and trust reliable informants.

What role do epistemic trust reductions play in such a view? My proposal is that they should be understood as NoPEMs. They exclusively serve the purpose of protecting the epistemic community and the epistemic commons from unreliable informants. Reducing trust in those who often and expectedly fail epistemically is a form of prevention: it prevents unreliable informants from spreading badly formed beliefs within the community. It is easy to see how: if epistemic trust is what allows information to flow between agents, then depriving unreliable informants of this currency is a way of preventing them from epistemically affecting others and our epistemic goods negatively.

My proposal receives substantial support from the fact that the extracted appropriateness conditions for non-punitive measures also straightforwardly hold for trust reductions. I will demonstrate this in what follows.

(1) Trust reductions are responses to epistemic failings, that is, negatively evaluated epistemic conduct. As mentioned earlier, this includes not just badly formed beliefs, but also instances of bad reasoning or improper epistemic functioning, and so on.

(2) As Fricker (Reference Fricker2007 §2.3, §6) has shown in detail, trust reductions undoubtedly impose negative epistemic consequences on the agents, which are ordinarily considered to be unpleasant. Trust reductions are applicable irrespective of culpability. In Fortune, Stephen is culpable for his believing based on the fortune-telling, and it is appropriate to reduce trust in him. In Senile Mother, the mother non-culpably believes the scammer. While her senility is a good excuse (or, depending on the severity, even an exemption), it is still appropriate to reduce trust in her.

(3) RMA’s characterizations of the requirements for trust reductions to be appropriate read straightforwardly like the expectation requirement of non-punitive measures. Here is a representative example:

It is hard to not read this passage as asking for a ‘symptomatic relation’ between one’s initial epistemic failings and one’s epistemic character and dispositions, where the initial epistemic failing reveals something epistemically bad about one’s epistemic dispositions and character.Footnote ¹⁶ What bad is revealed, in turn, needs to ground a (reasonable) expectation of sufficiently likely future epistemic failings, that is, failings in a wide range of similar circumstances. I agree with Boult that this kind of expectation requirement has to be met for trust reductions to be appropriate. It seems inappropriate to reduce trust in ordinary agents in response to an isolated epistemic failing. That appropriate trust reductions presuppose an expectation of future epistemic failings is explained by the preventative character distinctive of NoPEMs: only given such an expectation is a trust reduction plausibly a proportional way of balancing our collective epistemic interests with the individual’s interests of being regarded as trustworthy.Footnote ¹⁷

Furthermore, proponents of RMA rightly point out that trust reductions are only appropriate as long as the condition that grounds our expectation of future epistemic failings persists. For instance, Boult (Reference Boult2024c, p. 91) considers Barney who unknowingly visits fake barn country and thus a highly deceptive environment. As long as he is driving through barn country, we do well to reduce trust in him on barn-related matters. After Barney leaves fake barn country and enters real barn country again, we are inclined to reinstate our trust in him on these matters.

What is perhaps less obvious is whether there is also a requirement of sufficient severity of future epistemic failings. There are two ways of thinking about this. On the one hand, one might think that this constraint is restricted to the legal domain and has no proper analog constraint in the epistemic domain. Topic-restricted trust reductions are (arguably) less substantial and wide-reaching than interference with one’s rights and liberties by the state.Footnote ¹⁸ On the other hand, one might think that there is an epistemic analog to sufficient severity. Intuitively, we are less stringent in our responses to bagatelle epistemic failings. For instance, we might not bother responding to a false belief that you inattentively formed about an office supplies advert. We are much more stringent if, for instance, agents frequently fail epistemically, if they believe in false conspiracy theories about public health, if they do not lend credence to experts or if they deny the validity of basic inference patterns. It thus strikes me as plausible that the severity of an epistemic failing is a function of the amount of expected false beliefs, of the community’s interest in them, the failing’s practice-undermining capacity, and their epistemic centrality (Quine Reference Quine1953). One interesting consequence of this view is that whether we respond with a trust reduction to an epistemic failing depends not only on the way the belief was formed, but also on its content. To tie this back to earlier observations, perhaps, the appropriateness of NoPEMs is restricted to particular topics. While this idea merits further exploration, doing so is a task for another occasion.

Before identifying rehabilitative aspects of NoPEM, let’s consider a possible worry for the proposed expectation condition, namely that trust reductions are sometimes also ‘backward-looking’ (Flores and Woodard Reference Flores and Woodard2023, p. 2557). For instance, if someone epistemically fails, we might not only reduce trust in beliefs they will form, but also in beliefs they have already formed. How does this square with the idea that trust reductions target future epistemic failings?

I think that trust reductions in already formed beliefs also have a preventative purpose. To see this, note that we not only form new beliefs, but also maintain or revise beliefs. Maintaining beliefs is an ongoing process that regularly takes place, depending on whether we acquire relevant evidence or revisit our beliefs for various other reasons. It is thus a continuing source for future epistemic failings. If Stephen has been believing for some time that he should buy Boeing stocks because they will ‘take off’ very soon, and if he has maintained this belief by dogmatically responding to counterevidence ever since, then a ‘backward-looking’ trust reduction rather aims at preventing the fallout of Stephen’s future maintenance failures.Footnote ¹⁹

(4) Let’s check the final condition and elaborate on the subordinate rehabilitative dimension that is sometimes involved in non-punitive measures. Is there an epistemic analog here? Again, I think so. Trust reductions can be accompanied by responses that are conducive to rehabilitation. Consider Fortune. Amari could appropriately respond to Stephen’s failing with understanding and curiosity (‘Why do you trust a fortune-teller only on these matters, but not on others?’), respond with criticism to cultivate reason-responsiveness (‘Trusting a fortune-teller is irrational!’, cf. Piovarchy Reference Piovarchy2021), provide educational resources (‘Here are some basics on cyclical market trends.’) or encourage him to epistemically atone (Woodard Reference Woodard and Shafer-Landau2024). More paternalistically, we might also offer assistance, such as tutoring or guardianship, or even restrict access to bad sources. If these rehabilitative efforts turn out to be ineffective, they are plausibly discontinued (see the discussion of §63 on p. 7), resulting in purely preventative instances of NoPEMs. ^{Footnote 20,Footnote 21}

Summing up, there is ample reason to think that trust reductions are NoPEMs: They are (1) responses to epistemic failings, (2) impose negative epistemic consequences, (3) don’t generally presuppose culpability, and (4) are only appropriate if there is an expectation of sufficiently likely and (arguably) severe future epistemic failings and only as long as the condition grounding this expectation persists. (5) They also sometimes involve a rehabilitative dimension.Footnote ²²

My account of trust reductions as NoPEMs is explanatorily powerful. It explains why responses to culpable epistemic failings often involve trust reductions: non-punitive measures accompany culpability-tracking responsesFootnote ²³ if the culpable epistemic failing we respond to grounds an expectation of future epistemic failings. The account also explains why we sometimes reduce trust in non-culpable agents: NoPEMs don’t presuppose culpability and can be appropriate responses to both culpable and non-culpable epistemic failings. My account explains why our responses to epistemic failings can include positively valenced responses, like curiosity and understanding: positively valenced responses will sometimes be included in NoPEMs and are conducive to epistemically rehabilitating agents. Finally, the account explains why trust reductions are only appropriate if there is an expectation of future epistemic failings: the purpose of trust reductions is to protect the epistemic commons from future epistemic failings. Only given such an expectation do our collective epistemic interests proportionally outweigh the individual’s epistemic interests of being regarded as someone we can epistemically trust.

5. NoPEMs, RMA, and the challenges

In what follows, I will put my account to work to assess RMA’s challenges and extant responses to them. I will then take stock of the options remaining for RMA and consider alternatives.

Recall that RMA faces two challenges: it counterintuitively counts as epistemic blame positively valenced responses to culpable epistemic failings and trust reductions in non-culpable epistemic agents.

Regarding the first challenge, Boult (Reference Boult2020, p. 528; Reference Boult2024c, p. 104) denies that positively valenced responses fit one’s judgment that the agent in question is blameworthy. He thinks that in making this assessment, we can rely on a ‘shared grasp of the responses we typically associate with blame and relationship impairment’ (my italics). However, the discussion from the last section showed that we cannot. Different kinds of responses, namely culpability-tracking responses (such as epistemic blame) and non-punitive measures, are often run together, namely in cases in which someone’s epistemic failing is both culpable and grounds an expectation of future epistemic failings. Given a sufficiently rich conceptualization of our responses to epistemic failings, typical association is insufficient for a response to be a paradigmatic, constitutive component of epistemic blame.

Another way to address the first challenge is to modify RMA’s account of epistemic blame to simply exclude positively valenced responses (Boult Reference Boult2024c, pp. 72–73):

However, as Boult is acutely aware (ibid.), this modification seems very much ad hoc.

My account provides a principled motivation on RMA’s behalf for the exclusion of positively valenced responses in Epistemic Blame ${\rm{^\prime}}$ . Recall that non-punitive measures sometimes involve rehabilitative efforts. I argued above that positively valenced responses, such as showing understanding and curiosity, are naturally understood as being rehabilitative. To substantiate this further, recent empirical data suggest that being empathetic and receptive to those who, for instance, hold false conspiracy beliefs is rehabilitative: it can be an effective way of gaining their confidence and ultimately reducing their proneness to conspirational thinking.Footnote ²⁴ My account classifies positively valenced responses to epistemic failings as being part of the rehabilitative dimension of NoPEMs.

Interestingly, Boult seems sympathetic to two-pronged responses of this kind: about a case in which a good friend drifted off into a YouTube conspiracy rabbit hole, Boult (Reference Boult2024c, p. 185f, also p. 191) suggests that we both reduce trust in our friend and respond in a positively valenced fashion by ‘warming’ our friendship with him to prevent him from further drifting off. However, Boult thinks that the ‘warming’ is a distinct, non-epistemic response concerned with one’s friendship (ibid., p. 186). My worry with Boult’s response is that it does not generalize, since it is easy to imagine cases in which understanding or curiosity is fitting, but not as a ‘warming’ of the relationship. It is an advantage of my proposal that it offers an epistemically unified account of our responses in these kinds of cases and, more generally, a principled motivation for the exclusion of positively valenced responses in Epistemic Blame ${\rm{^\prime}}$ : positively valenced responses belong to NoPEMs, not epistemic blame.Footnote ²⁵

In response to the second challenge, Boult (Reference Boult2024b, c) has argued that trust reductions in excused agents lack a crucial condition for being manifestations of epistemic blame, namely ‘meaning’. Drawing on Scanlon’s work, Boult suggests that ‘meaning’ is tied to the ‘agent’s intention’ (Boult Reference Boult2024c, p. 65) and is characterized as the ‘significance [epistemic conduct] has for oneself and the other people with whom one stands in various sorts of relationships’ (ibid.). Note that ‘meaning’ can’t just be bad epistemic consequences, since the epistemic failings of excused agents also have bad epistemic consequences for themselves and others. Rather, the ‘meaning’ of a culpable epistemic failing ‘reveals [something epistemically bad] about how the agent regards their epistemic relations with others’ (Boult Reference Boult2024b, p. 394).

A general concern with this proposal is that it seems to make an epistemic failing a mere evidential means for what Boult claims to be the primary difference-maker of blameworthiness, viz., the agent’s regard for their epistemic relations to others. A consequence of this view is that epistemic failings are, in principle, dispensable when it comes to assigning epistemic blame. Suppose, for instance, that there were other ways to find out about whether an agent disregards their epistemic relationships (say, via a brain scanner, via a search of their phone, and so on). If we could learn in these ways that they do, it counterintuitively would seem fitting to respond to their disregard of epistemic relationships with epistemic blame, even if no epistemic failing has occurred. Thus, the epistemic failings are mere evidential means for revealing how agents regard their epistemic relationships which sits uneasily with the natural idea that epistemic blame is primarily a response to epistemic failings.Footnote ²⁶

A related general concern is that Boult’s proposal seems to get the relation between epistemic failings and how one regards one’s epistemic relationships backward. Just like stealing makes you a thief, believing recklessly makes you a reckless believer, who disregards his epistemic relationships by not meeting the expectation of taking due care in believing. In the same vein, if someone is disposed to steal but never manifests this disposition, it seems intuitively incorrect to call him a thief; at most, we might call him a potential thief (Alston Reference Alston, Kiefer and Munitz1970, p. 26; Duff Reference Duff1996, p. 189). The claim that the link between epistemic failings and an agent’s regard for their epistemic relationships is merely evidential does not seem to capture that epistemic failings are plausibly constitutively prior to how failing agents regard their epistemic relationships.Footnote ²⁷

A more particular worry concerns the ‘meaning’-criterion. While I find it quite hard to pin down Boult’s exact view, I think he offers at least two conceptions of ‘meaning’, neither of which yields a plausible criterion. On a first reading, culpable epistemic failings involve bad epistemic intentions. When discussing a case in which an agent deliberately takes a pill that makes them reason hastily and dogmatically, Boult (Reference Boult2024c, p. 80) argues that the agent has (culpably) impaired his epistemic relationships. If, however, he was tricked into taking the pill, he would have a good excuse and would not have impaired his epistemic relationship (ibid.). On this intention reading of ‘meaning’, having epistemically bad intentions when one epistemically fails is thought to be a crucial condition for being epistemically blameworthy. One might think, for instance, that by deliberately taking the pill, one accepts that one will be an unreliable informant afterward, thereby revealing a disregard towards one’s epistemic relationships, impairing them.Footnote ²⁸

The intention reading does not yield a plausible ‘meaning’-criterion. ‘Meaning’ of this kind is not necessary: when we culpably epistemically fail, we rarely will have bad intentions toward our epistemic relationships.Footnote ²⁹ This is because when we form beliefs, we are not typically concerned with our epistemic relationship. At most, we intend to adjust our beliefs to our evidence. When we form non-reflective, ‘animal beliefs’ (cf. Sosa Reference Sosa2009), for example, perceptual beliefs, we may not intend anything at all. But even in cases of epistemic failings where we have intentions that concern our epistemic relationships, it’s not that we intend to (or take ourselves to) impair them. Instead, we will often erroneously think that we are doing well epistemically.Footnote ³⁰

Bad intentions are also not sufficient for blameworthiness. To see this, consider the following case:

Intuitively, Caleb is still excused (or even justified, depending on your view about epistemic justification) and thus not blameworthy for believing falsely, despite his bad intentions. Contrarian is thus a case in which an excusing (or justifying) condition normatively screens off bad intentions. Hence, bad intentions are not sufficient for blameworthiness.

A second reading of ‘meaning’ might be called the attitude and conduct reading: Drawing on Smith (Reference Smith2005), Boult (Reference Boult2024c, p. 111) notes that ‘attitudes and actions that we do not (seem to) have control over can nevertheless express a great deal about how we regard one another’. Likewise, doxastic responses are often involuntary and can reveal a lot about our epistemic character and how we regard our epistemic relationships. For instance, believing the contents of a tabloid newspaper reveals something bad about one’s epistemic character (Boult Reference Boult2024c, p. 112), namely that one is gullible. But doing so after a gullibility pill is slipped into one’s coffee does not.

The resulting ‘meaning’-criterion is not plausible either. ‘Meaning’ in this sense is not necessary for blameworthiness. We sometimes epistemically fail due to slips (Williamson Reference Williamson, Dutant and Dorschforthcoming; Heil Reference Heil2022; cf. Amaya Reference Amaya2011) caused by stupidity, clumsiness, or no discernible reason at all. Suppose you forget to congratulate your friend on her birthday because you made the stupid mistake of failing to transfer the date while reorganizing your calendar. Or suppose you clumsily drop a pile of newspapers, pick out the wrong one, and form a false belief about today’s winning lottery numbers. In both cases, you intuitively are blameworthy for your epistemic failing.Footnote ³¹ While some one-off epistemic failings may be due to lack of care or other defects of character, it is not plausible that they always (or even often) are. Hence, there are cases of culpable epistemic failings that don’t reveal anything about your character or how you regard your epistemic relationships. ^{Footnote 32,Footnote 33}

‘Meaning’ in this second sense is also not sufficient. This can be seen as soon as we look beyond ‘environmental or exogenous’ (Boult Reference Boult2024b, p. 394) excusing condition. Sometimes, excusing conditions are both grounded in one’s epistemic character and affect one’s epistemic relationships by violating legitimate expectations. Consider the following case:

Alfred’s anxiety is overwhelming and thus excuses his epistemic failings. Yet, his epistemic failings reveal something epistemically bad – his unfounded distrust – about how he regards his epistemic relationships to others. Furthermore, Alfred violates the legitimate, default expectations of the guests, his co-workers, and himself to handle these fairly ordinary circumstances well and be trusting and trustworthy concerning the needs communicated to him. Still, even though his epistemic failing reveals something epistemically bad about him and violates epistemic expectations, Alfred is excused.

Consider two possible replies: First, Boult (Reference Boult2023, p. 818; Reference Boult2024b, p. 393) might reply that ‘good epistemic relationships involve agents expecting others to avoid epistemic failings, unless they have a legitimate excuse, or perhaps some overriding reason not to’. On this view, excused epistemic failings could not in principle violate epistemic expectations, because excusing conditions are simply baked into our expectations. However, there is a perfectly reasonable sense in which the default expectation towards Alfred is to do well in the circumstances at issue. That we excuse him for violating these expectations is a ‘concession to human frailty’, as legal theorists sometimes put it, not a reason to believe that we didn’t expect him to do well in the first place.Footnote ³⁴

In the light of these difficulties, I think proponents of RMA would do well to embrace my account of trust reductions and give up on the idea that trust reductions are typical constitutive components of epistemic blame. My account offers a principled response to RMA’s first challenge: positively valenced responses belong to NoPEMs, not to epistemic blame. My account also resolves the second challenge to RMAs: if trust reductions merely accompany epistemic blame responses, but don’t partially constitute them, then no worry about reductions of trust in non-culpable agents arises.

6. Taking stock

Where does this leave RMA? Embracing my account in order to parry the two challenges might seem like a pyrrhic victory for proponents of RMA because without trust reductions, RMA’s account of epistemic blame looks fairly anemic. Proponents of RMA could counter this impression by offering a richer account of epistemic relationship-modifications. They might identify more promising candidates for blame-constituting responses to epistemic failings or claim that the latter are simply varied and contextually determined (Boult Reference Boult2024c, p. 88). In any case, I think proponents of RMA have their work cut out for them.

Those not wedded to RMA might take my argument to be grounds for skepticism or outright rejection of RMA. After all, the latter has partly been motivated by the idea that trust reductions are paradigmatic ways of altering epistemic relationships and thus partially constitutive of epistemic blame. However, if my account is correct, then trust reductions are, in fact, not partially constitutive of epistemic blame.

If we reject RMA, one option would be to start with my account of trust reductions as NoPEMs and enrich it with an account of culpability-tracking responses to epistemic failings. While this is ultimately a project for another paper, let me briefly sketch some options.

First, there are various candidates in the literature on epistemic blame that we might appeal to (see fn. 2). It might indeed be worth revisiting some of these competitors to RMA because one of the main complaints against them is that they have trouble explaining the putative ‘coolness’ of epistemic blame (Boult Reference Boult2020, p. 519f). However, if my account is right, then proponents of RMA might be mistaken in thinking that epistemic blame is particularly ‘cool’, at least if they think so because they mistakenly take ‘cool’ trust reductions to be instances of epistemic blame.

Second, one might think that the notion of blame belongs to the moral and legal domain only (Kauppinen Reference Kauppinen2018, p. 2) and that we should thus look for accounts of culpability-tracking responses that do without the notion of epistemic blame. In this regard, I am sympathetic to Dogramaci’s view (Reference Dogramaci2012, Reference Dogramaci2015) that the purpose of epistemic assessment is to identify reliable informants for deference and to facilitate the communal coordination regarding the use of reliable belief-forming methods.Footnote ³⁵ In response to culpable epistemic failings, say, a belief formed by wishful thinking, we respond with a negative assessment (‘Your belief is irrational!’), thereby expressing disapprobation to influence our audience to not use the employed belief-forming method and to not defer to the thus-formed belief. Dogramaci’s account of culpability-tracking responses would complement my account nicely: trust reductions as NoPEMs would be the right response to agents who can be expected to form false beliefs in the future because they are culpably or nonculpably unreceptive to the force of our epistemic assessments. While I think this sketch is highly promising, I will have to leave filling in the details for another occasion.