Epistemic Value as Attributive Goodness?

Abstract

According to insulationism, a common take on epistemic value, being of epistemic value does not entail being of value simpliciter. In this paper, I explore one version of insulationism which has so far received little attention in the literature. On this view, epistemic value does not entail value simpliciter because it is a form of attributive goodness, that is, being good as a member of a particular kind. While having a significant advantage over some other formulations of insulationism, I argue that the attributive goodness view of epistemic value should be rejected. On the one hand, shifting to a discussion of attributive goodness comes with several contentious commitments, as well as implausible upshots for epistemic axiology. On the other hand, I demonstrate that one can strengthen other forms of insulationism in such a way that the supposed advantage of the attributive goodness view diminishes.

1. Introduction

Over the past decades, epistemologists have turned their attention to the value of the objects of their investigations. This project of epistemic axiology and the corresponding concept of epistemic value have become staple components of debates in epistemology. However, it is unclear precisely what ‘epistemic value’ refers to. There are two metaepistemological positions which scholars have defended concerning this question. First, according to substantivism, ‘epistemic value’ denotes a subset of value simpliciter. According to insulationists, on the other hand, ascribing epistemic value to some object does not entail that said object is valuable simpliciter. In this paper, I explore one variety of insulationism which has not received the appropriate scholarly attention up to this point. According to the attributive goodness view (AGV), ‘epistemic value’ denotes a form of attributive goodness, i.e. being good as an item of a specific kind. While being entertained by at least some authors in the literature, AGV is a clear outlier in the debate about epistemic value.¹ My aim in this paper is to argue against AGV: shifting from standard versions of insulationism to AGV does not yield any advantage while generating a host of further issues.

I start by outlining the standard version of insulationism put forward by Sosa (2007), as well as some prima facie motivation to opt for AGV (Section 2). I argue that on Sosa's simple picture of epistemic value, substantial disagreements about the fundamental epistemic values become inconceivable. Afterwards, I turn towards two sets of challenges one could raise against AGV. First, I discuss the question which goodness-fixing kind the epistemic point of view might be engaged with. After introducing four promising candidates from the literature, I draw up three challenges which either account has to answer (Section 3). Second, one might doubt whether AGV can make sense of relations between epistemic values, which play a major role in epistemic axiology. Since relations between instances of attributive goodness have not been investigated up to this point, they constitute a huge open issue at the heart of AGV (Section 4). Both of these discussions yield prima facie reasons to prefer the standard insulationist picture over AGV. Finally, I return to the issue of substantial disagreements about epistemic value and draw up resources to strengthen Sosa's position in order to avoid the previously identified shortcoming (Section 5). Thereby, I demonstrate that there is no reason to opt for AGV: the standard version of insulationism fares at least as well as AGV.

2. Insulationism and the case for AGV

According to substantivism, epistemic axiology is just a subproject of axiology simpliciter. A subset of the valuable entities are the epistemic goods. Just as any other valuable object, they are fitting objects of our desire. To say that true beliefs are epistemically valuable is to say that it is always fitting to desire true beliefs.² According to insulationists, on the other hand, epistemic axiology can do its work without leaving the domain of epistemology. Thus, evaluations in terms of epistemic value are insulated and do not come with any immediate upshots for axiology simpliciter. Hence, to say that true beliefs are epistemically valuable does not amount to a commitment to their general desirability.³ Insulationists take this non-committal stance to be one of the main reasons speaking in favour of their position: whether or not true beliefs, knowledge, or even understanding are always pro tanto valuable is a contentious assumption.⁴ Yet, or so is the idea, epistemologists are not wrong in assigning epistemic value in broad strokes, for instance, with unrestricted statements like ‘truth is the final epistemic value’ or ‘knowledge is epistemically always more valuable than mere true belief’. On the substantivist's picture, epistemic axiologists might need to be much more cautious and restrict or qualify such statements.

The most prominent appeal to insulationism goes back to Sosa (2007).⁵ According to him, there are many domains of evaluation which are centred around certain domain-specific values. A domain-specific value is not supposed to be any ‘real value’ (2007: 73), but only the subject of one's critical or evaluative judgements. Thus, domain-specific values are grounded in our human proclivity to judge, evaluate, or score things. Epistemic evaluation is the evaluation of some object with regard to the fundamental domain-value of truth (2007: 78). Thus, saying that truth is the fundamental epistemic value does not entail that it is the fitting object of desires or anything like that. Epistemic axiologists are only committed to the idea that truth is part of the epistemologists scoring criteria and the inclusion of other scoring criteria is justified by their axiological relation to truth.⁶ I take it that the most reasonable way to make sense of this form of domain-specific value is in terms of hypothetical value simpliciter: epistemic axiology assumes that every true belief has some special value X and then it uses the tools of value theorists to investigate which other values can be derived from X.⁷

Sosa's appeal to domain-specific values is not the only form of insulationism though. In his work, Sylvan (2018, 2020a) endorses insulationism by a different route, i.e. by opting for AGV. If one treats ‘epistemic value’ as denoting a form of attributive goodness, one thereby endorses a form of insulationism. This is due to the fact that claims about the attributive goodness of some object do not entail that said object has any actual value simpliciter. The value of something as an item of a particular kind does not imply that it is valuable in any other sense. This idea is one of the basic theses in Geach's (1956) seminal work on the attributive usage of adjectives. Saying that ‘Dumbo is a small elephant’ implies that Dumbo is small for an elephant, but not that Dumbo has some property of smallness simpliciter. Thus, a true belief might be epistemically valuable, i.e. valuable as an item of a particular kind K, but it might still fail to exemplify any actual value simpliciter.

Before moving on to the motivation of shifting to AGV, one might ask how it differs from similar, already existing approaches to epistemic value. First, one might suspect that this position is already endorsed by Ridge (2013: 189) in his thesis about the attributive usage of the qualifier ‘epistemically’ in ‘being epistemically good’. Building on Geach's work, Ridge distinguishes two usages of qualifiers in phrases such as ‘is F-ly G’. These phrases either imply that the object is a G, or they do not. In the former case, ‘F-ly’ is used in a predicative fashion and, in the latter case, in an attributive one. Sylvan interprets Ridge's work as endorsing AGV. However, on closer inspection, it is apparent that what Ridge calls an attributive reading of ‘epistemically valuable’ is equivalent to the insulation thesis, and it says nothing about attributive goodness.⁸ Consider, for instance, the phrase ‘being superficially good’. Here, ‘superficially’ works as an attributive modifier a la Ridge. Nevertheless, we do not use this phrase to say that something is good as a superficial item. Therefore, there are different ways to spell out Ridge's position, for instance, by adopting a picture following Sosa.

Second, AGV bears some resemblance to the position Hazlett (2013: 146) calls epistemic essentialism. According to epistemic essentialism, epistemic evaluation is the evaluation of beliefs, qua beliefs, i.e. the evaluation of beliefs according to some belief-specific standard that is somewhat essential to being a belief. At first sight, epistemic essentialism appears to be one particular form of AGV, which identifies the goodness-fixing kind with belief. However, throughout his discussion, Hazlett sticks to using ‘epistemic value’ in terms of goodness simpliciter: epistemic evaluations are appropriate only if epistemic values correspond to some pro tanto goodness. As an insulationist, a proponent of AGV cannot follow along with this characterisation. Nevertheless, some of the problems plaguing epistemic essentialism will also apply to AGV. Thus, Hazlett's scepticism towards essentialism will show up in Section 3. Furthermore, AGV and essentialism seem to share some of their most basic motivation: the general theme of tying epistemic normativity to the nature of the entities under evaluation. This general idea can be identified in a variety of recent work on epistemic normativity. It is, for instance, a recurrent theme in the analysis of epistemic reasons. Both, essentialism and AGV can be regarded as applications of this outlook to the notion of epistemic value.

What is the advantage of AGV over its insulationist predecessors? Sylvan does not provide any explicit motivation for this metaepistemological stance. Nevertheless, we might extract a case in favour of AGV from the way he makes use of it: Sylvan defends veritism, i.e. the position that truth or accuracy is the sole fundamental epistemic value, from the charge of not being able to account for certain other supposed epistemic values like understanding by pointing to AGV. Understanding might be a good thing, but does not add any additional epistemic value since it does not make beliefs good qua beliefs. Therefore, Sylvan argues, veritists do not need to provide an account of the value of understanding in terms of truth. Now, if one is a veritist, this rationale might look appealing.⁹ Thus, conditional on one's other commitments, there might be good abductive reasons to opt for AGV.

However, I argue that there is a better reason to prefer AGV over Sosa's original formulation of insulationism: Sosa's insulationism cannot make sense of substantial disagreements about the fundamental epistemic values. Several epistemic axiologists dispute, for instance, what Sosa appears to take for granted, namely the idea that truth is the sole fundamental epistemic value.¹⁰ However, given that the epistemic point of view is simply concerned with the domain-specific value which is centred around truth, disputing the special place which the epistemic value of truth has to play in one's axiology appears like an idle move. Thus, the most charitable interpretation of these epistemologists is that they have changed the topic and started evaluating from an epistemic* point of view which is concerned with some other domain-specific value. Hence, on Sosa's version of insulationism, substantial disagreements about the fundamental epistemic values are inconceivable. Yet, so the conviction of most participants in these debates, they do not take themselves to be talking past each other: they are concerned with the same kind of evaluation but disagree about the fundamental values at work. Since a metaepistemic account of the nature of epistemic value should take these convictions seriously, Sosa's insulationism is faced with a serious issue.¹¹

AGV is not susceptible to this shortcoming: viewing epistemic axiologists as arguing about the attributive goodness of some kind K makes such substantive disagreements conceivable. If one shifts to an epistemic* point of view, which is either concerned with some other form of evaluation, or the attributive goodness of some kind K*, one has clearly changed the topic of the dispute. However, not every dispute about attributive goodness has to amount to a change in subject matter. Imagine that two chefs are arguing about what makes for a good knife. One might rank a comfortable handle high, while the other defends the importance of a sharp blade. They disagree about the same subject matter and, thus, their disagreement is substantive. Similarly, one might disagree about the attributive goodness of a kind like belief. While authors like McHugh (2012) and Sylvan (2018) argue that what makes beliefs fundamentally good as beliefs is their accuracy, others regard knowledge as the feature which makes beliefs attributively good (Ghijsen et al. 2016; Littlejohn 2018). These authors appear to be engaged in a substantive disagreement.

To conclude, AGV constitutes an alternative to standard forms of insulationism. It has some prima facie reasons going for it since a proponent of AGV can make sense of substantial disagreements about the fundamental epistemic values. While being different in this regard, it is important to note the similarity between attributive goodness and domain-specific value. As I argued above, the most straightforward reading of domain-specific value is in terms of hypothetical value. Now, according to one influential theory of attributive goodness, attributive goodness is nothing but a special form of hypothetical value. This is what makes attributive goodness and goodness simpliciter instances of the same kind – goodness. The attributive goodness of an item X as an item of kind K is the hypothetical value which X has given that one cares about having an item of K which fulfils the K-specific function well.¹² Therefore, AGV and Sosa's domain-specific values have much in common.¹³ The only difference mentioned so far is that AGV does not individuate the epistemic point of view in terms of its fundamental values, but in terms of the characteristic function of a kind. This shift yields the necessary resources to explain substantive disagreements about fundamental epistemic values. In the next two sections, I will work out some further differences, which constitute potential drawbacks of opting for AGV, before turning to a refined version of Sosa's view in the fifth section.

3. What kind are epistemologists talking about?

In this section, I engage with the question which kind's attributive goodness the epistemic point of view is supposed to be engaged with. Provided that most usages of ‘epistemic value’ by epistemologists do not hint at any such kind at all, an answer to this question is far from obvious. I start by proposing a list of plausible candidates for the goodness-fixing kind K. Second, I present three issues that versions of AGV, which endorse one of these Ks, might run into. These concerns provide prima facie justification to prefer certain candidates over others and might generate some general reason to drop AGV in favour of alternatives.

I identify four promising candidates for K which have been envisaged in the literature. First, as mentioned in the previous section, one could follow Sylvan's take on K and identify epistemic evaluation with the evaluation of beliefs qua beliefs.¹⁴ Second, following Sosa's (2007) own lead, one can identify K with certain attempts or performances of our belief formation. Just as an archer's shot can be evaluated as an attempt at hitting its target, so can our cognitive performances. If one understands the evaluation of telic virtue epistemologists in axiological terms, such a version of AGV might be the best representation of their ideas. Third, Alston (2005: 33f.) defines the epistemic point of view as being concerned with the success of our cognitive systems. Thus, one could understand epistemic value as the attributive goodness of cognitive systems. Finally, as it is commonly done in the literature, one could highlight the proximity of the epistemic point of view and the goals of inquiry.¹⁵ In this case, ‘epistemic value’ could refer to the attributive goodness of the result of an inquiry.

Depending on which candidate for K a proponent of AGV adopts, they will face some of the following three issues. I do not regard these concerns as detrimental objections to the respective positions. Yet, avoiding contentious commitments in their regard, is an abductive reason to choose one metaepistemic theory over another. First, as I mentioned in Section 2, Hazlett raises some doubts about what he calls epistemic essentialism about beliefs. Roughly put, Hazlett doubts that there is any unique epistemic normative propriety which is attached to beliefs qua beliefs.¹⁶ For instance, while it seems fair to assume that the kind knife has a characteristic function attached to it, one might doubt that mental states like beliefs come with such a function. Thus, if such functions are at the heart of attributive goodness, beliefs should not be assessable in terms of attributive goodness either. While Hazlett's scepticism is directed at beliefs as a candidate of K, one might wonder whether analogous arguments cannot be found with regard to any of the other three candidates listed above.

Second, even if one is able to circumvent Hazlett's scepticism about a unique normative standard pertaining to our kind K, there is another prominent worry which would dispel K from being a proper basis for attributions of attributive goodness, i.e. being a goodness-fixing kind. According to Thomson (2008: 109–23), not every kind with a unique normative standard attached to it is a goodness-fixing kind. Thus, even if one circumvents Hazlett's general scepticism, one might still be susceptible to Thomson's concern. In her view, there are two forms of such standards: first, there are the kind's standards of correctness or fittingness. These concern the adequacy of an item of the kind with regard to an object or the situation. Belief, for instance, is said to be fitting iff the believed proposition is true. Admiration is fitting iff the admired object is admirable.¹⁷ Second, there are the standards of goodness, which determine whether an item is a good item of its kind. These two sets of standards are not the same and can come apart: trusting a person is fitting iff said person is trustworthy. However, it is far from obvious that such a fitting instance of trust is good as trust (Howard 2018). Or consider assertions: these, one might argue are also fitting iff their object is true. However, an instance of an assertion does not become a better assertion by the mere fact that the asserted proposition turns out to be true (Thomson 2008: 102). Now, Thomson argues that mental states do only come with a standard of fittingness. Therefore, they cannot count as goodness-fixing kinds. If this argument is successful, there is good reason to reject Sylvan's version of AGV and maybe also some of the other alternatives specified above. However, as Thomson herself notes, performances do not fall prey to her reasoning. Thus, the telic virtue epistemologist's take on AGV is not affected by this issue.¹⁸

This brings us to the third and probably most severe problem with AGV – the scope of epistemic value attributions. Epistemologists ascribe ‘epistemic value’ in abundance: computers, glasses, an extra cup of coffee, newspapers, and social networks are all said to have or lack epistemic value, at least in its instrumental variety. However, only items of a kind K can be attributively good qua K. Thus, whatever kind we choose as our basis for AGV, it better be inclusive enough to accommodate all of these different objects. Else, one is forced to restrict epistemologist's usage of ‘epistemic value’ in a more or less drastic fashion, which, I presume, constitutes a pro tanto reason against any metanormative theory of epistemic values. Thus, by the above list of examples alone, there appears to be a good reason to disregard doxastic attitudes, performances, and answers to our questions in inquiry as a reasonable goodness-fixing kind for the epistemic point of view.¹⁹ On either of these views, many epistemologists are confused about the correct application of ‘epistemic value’. The only version of AGV which might have a chance to withstand this objection is the cognitive system view: given a very extended conception of cognition, every object in the above list can be accommodated. However, such an assumption about cognition is controversial and constitutes a contentious commitment. Therefore, even this version of AGV should have some abductive reasons speaking against it.

4. Attributive value relations

In this section, I discuss a glaring issue for AGV as it stands up to this point. Epistemic axiologists make frequent use of concepts from axiology simpliciter, and as far as the state of the literature in value theory goes, nobody has tried to translate these ideas to the domain of attributive goodness yet. Epistemic values are said to be final or instrumental, fundamental or derivative, and, in order for AGV to have any merit as a metaepistemological position, it has to make sense of these categorisations. In the following, I provide some first steps towards closing this gap in the literature, i.e. I propose a reasonable extension of the final–instrumental and fundamental–derivative distinctions to different instances of attributive goodness. While I demonstrate that AGV can make sense of relations between epistemic value in general, I point out that several theses of epistemic axiologists become quite contentious when reconstructed in this framework. I start with instrumental value derivations (4.1) before turning towards non-instrumental ones (4.2).

4.1. Instrumental value derivations

The finality or fundamentality of a value concerns its relation to other values. Hence, as a first step, one has to make sense of the idea that there is more than one form of attributive goodness attached to some item of a kind. If evaluations of attributive goodness are just one-dimensional and only sort items in terms of better or worse items of a kind, then these relational-concepts appear to be out of place. Here is how I conceive of the required plurality of attributive goodness: attributive value, just as value simpliciter, comes in a contributory, or pro tanto fashion in that we can identify aspects of items which make them attributively good. For instance, a knife might be attributively good in one respect, for instance, being sharp and attributively bad in another aspect, for instance, being fragile. Similarly, a belief might be attributively good in one aspect, for instance, being accurate and attributively bad in another aspect, for instance, lacking justification. With this I do not want to say that such an aspect of an item X which makes X good as a K is itself attributively good as a K. Rather, the aspect makes its bearer X pro tanto attributively good. Now, the overall or pro toto attributive goodness of such an item is the result of weighing the pro tanto attributive goodness and badness, which is based on the different aspects of said item, against each other. However, besides comparing the relative weight of these contributions, we can also investigate their relation. Just as in the case of goodness simpliciter, we can ask whether some aspect of an item makes it attributively good in virtue of some other attributively good-making aspect of said item.²⁰

Given these resources, one can now ask what the different axiological concepts should look like when applied to the relations between attributively good aspects of an item. The following definitions of instrumental and final goodness appear to be straightforward translations of these concepts to the case of attributive goodness:

Instrumental Attributive Goodness

An item X of kind K is pro tanto instrumentally attributively good as a K iff there is an aspect A₁ of X which makes X pro tanto attributive good as a K in virtue of promoting some other aspect A₂ of X which makes X pro tanto attributively good as a K.

Final Attributive Goodness

An item X of kind K is pro tanto finally attributively good as a K iff there is an aspect A₁ of X which makes X pro tanto attributive good as a K, but not in virtue of promoting any other aspect A₂ of X which makes X pro tanto attributively good as a K.

In order to elucidate these definitions, I will use the following example: the functional kind of being a sweeper in the sport of curling. A good sweeper is an athlete who sweeps well. Put differently, sweeping well is an aspect that makes a sweeper good as a sweeper. This aspect of pro tanto attributive goodness appears to be final. There is no other aspect of X in virtue of whose promotion the above pro tanto attributive goodness holds. Contrast this with another possible aspect of X: having a strong arm. Having a strong arm is also an aspect that makes X pro tanto good as a sweeper. However, if it were not for the pro tanto attributive goodness associated with the very function at the core of the functional kind sweeper, as well as the contribution which a strong arm provides in this regard, the strong arm would not count as a positive sweeping-attribute of the athlete. Hence, it is only in virtue of promoting one aspect of X that makes X pro tanto attributively good, namely sweeping well, that another aspect makes X pro tanto attributively good, namely having a strong arm. Furthermore, a strong arm might make X attributively good as a sweeper even though X does not end up having a great sweeping ability. There might be other circumstances or pro tanto attributively bad-making properties due to which the athlete does not instantiate much of the final attributive goodness itself. Yet, one would still say that the strong arm makes for an attributively good-making feature since it promotes or contributes to the aspect of X which makes X finally attributively good.

Unfortunately, such a straightforward take on attributive value relations becomes problematic as soon as one turns to epistemic value. Let us assume for the moment that Sylvan's version of AGV is correct, in that epistemic evaluations are evaluations of beliefs qua beliefs. Furthermore, assume the commonly accepted stance that true beliefs are epistemically valuable and that justification is (inter alia) instrumentally valuable as a means towards gaining true beliefs. This instrumental value of justification is supposed to yield an explanation of why some false beliefs are still epistemically valuable. However, given the abovementioned account of instrumental attributive goodness, this cannot be correct: if a belief X is false, there is no aspect of X in virtue of whose promotion being justified would make X pro tanto attributively good as a belief.

In order to circumvent this issue, a proponent of AGV would have to endorse a different translation of instrumentality and finality to attributive goodness. Instead of understanding promotion in terms of de facto contribution, one would have to appeal to a counterfactual notion of promotion.

Instrumental Attributive Goodness #2

An item X of kind K is pro tanto instrumentally attributively good as a K iff there is an aspect A₁ of X which makes X pro tanto attributive good as a K in virtue of promoting some other possible aspect A₂ of X which would make X pro tanto attributively good as a K.

Final Attributive Goodness #2

An item X of kind K is pro tanto finally attributively good as a K iff there is an aspect A₁ of X which makes X pro tanto attributive good as a K, but not in virtue of promoting any other possible aspect A₂ of X which would make X pro tanto attributively good as a K.

Given that justification makes it more likely that a belief is true, one can argue that being justified is a pro tanto attributively good-making aspect of a belief even if said belief turns out to be false. The fact that truth is a possible aspect of the belief which would make it attributively good as a belief suffices to make justification an attributive good-making feature as well.²¹

However, while such a counterfactual concept of instrumental value derivations is commonly used in axiology simpliciter, it is quite contentious when applied to attributive goodness.²² Imagine you are at a scrapyard looking for metals to forge a good knife. You have the choice between a handful of iron and a handful of copper. Neither the handful of iron nor the handful of copper can cut anything. Yet, a knife forged out of iron will be a better knife than a knife forged out of copper. Does it follow that the handful of iron is already better as a knife than the handful of copper? Such a usage of attributive goodness talk strikes me as idiosyncratic. The most plausible extension of instrumental value derivations to attributive goodness seems to be limited to relations of de facto promotion or contribution. Therefore, a proponent of AGV faces a dilemma: either they have to appeal to a very odd usage of attributive goodness claims or they have to deny certain widespread ideas of epistemic axiology.

4.2. Non-instrumental value derivations

Going back to Sylvan's (2018) own work, epistemic axiologists have also started to appeal to non-instrumental value derivations: there are derivative forms of epistemic value which are not valuable in virtue of promoting some other value but due to some other in virtue of relationship. Thus, not every final epistemic value needs to be a completely non-derivative or fundamental one. The main motivation for this turn to non-instrumental value derivations is the so-called swamping problem for veritism.²³ How can knowledge be any more valuable than a mere true belief if all the additional valuable aspects of knowledge are only valuable as a means towards the end of truth? If the end of truth has been reached, those instrumental values should not add anything to the overall value of the state in question. To put it in a more principled manner, instrumental values do not accumulate with the final values from which they are derived; the instrumental value is swamped by the final value. However, most epistemic axiologists endorse the view that knowledge is more valuable than true beliefs. Thus, one is either forced to give up veritism and assign an independent fundamental value to knowledge, or one has to identify a non-instrumental way in which the value of knowledge can be derived from the value of true beliefs. I start this subsection by showing that the swamping problem does also occur in AGV. Afterwards, I translate Sylvan's principle of non-instrumental value derivation to AGV. Finally, I argue that, when it comes to attributive goodness, Sylvan's manoeuvre is highly contentious.

Given AGV, the same rationale for the swamping problem for veritism seems to occur. For the purpose of clarification, I will again use Sylvan's version of AGV in terms of the attributive goodness of beliefs. On AGV, veritism amounts to the claim that being true is the only aspect of a belief which makes a belief fundamentally attributively good as a belief. Now assume that any additional attributive goodness generated by a belief's amounting to knowledge is only valuable in virtue of promoting another attributively good-making aspect of the belief – being true. Additionally, it appears very plausible that the same principle I described above in terms of goodness simpliciter holds with regard to attributive goodness: the pro tanto instrumental attributive goodness associated with an aspect A₁ of an item X does not accumulate with the pro tanto attributive goodness created by an aspect A₂ in virtue of whose promotion A₁ is an attributive good-making feature. To repurpose the example from before, a strong arm does not make the athlete any better as a sweeper besides its contribution to their sweeping ability. The pro tanto attributive goodness generated by the sweeping ability swamps the pro tanto instrumental attributive goodness based on the strong arm. Similarly, the pro tanto attributive goodness associated with being true swamps the additional pro tanto instrumental attributive goodness grounded in the belief's amounting to knowledge. Therefore, one can reconstruct the swamping problem in AGV.

In order to defend veritism against the swamping problem, Sylvan alludes to ideas of Hurka (2001). Consider the following principle:

Hurka Principle

When V is a non-instrumental value, proper ways of valuing V have some derivative non-instrumental value. (Sylvan 2018: 383)

Having a fitting pro-attitude towards valuable things is itself valuable. Yet, or so the underlying idea, these attitudes are not only valuable as a means towards promoting the amount of the value they are directed at. Thus, we have identified a form of non-instrumental value derivation: the value of these fitting pro-attitudes is derivative of the value of their object. Yet, this value is not purely instrumental. Now, we can translate the Hurka Principle to AGV in the following fashion:

Hurka Principle (Attributive Goodness)

When a possible aspect A of an item X of kind K would make X a pro tanto non-instrumentally attributively good K, proper ways of valuing A and their manifestations, insofar as they are aspects of X, make X pro tanto derivatively, yet non-instrumentally, attributively good.²⁴

One of the main contributions of Sylvan's paper is his way of reconstructing the additional epistemic value of justification, knowledge and other epistemically valuable states in terms of them manifesting proper ways of valuing true beliefs. If this part of his rationale is successful, one has identified a way in which amounting to knowledge is a final attributively good-making feature of a belief which is not subject to the swamping principle described above: since its additional value is non-instrumentally derived, the attributive goodness of a belief which amounts to knowledge adds to the attributive goodness of a belief which is merely true.

I presume that the Hurka Principle for attributive goodness is susceptible to the same worry as the counterfactual approach to instrumental value derivations in the previous subsection.²⁵ However, even if one endorses this version of the Hurka Principle, I argue that there is still a good case to be made for a form of swamping problem which nonetheless plagues veritism on AGV. This has to do with the fact that the swamping principle alluded to above strikes me as much broader when applied to attributive goodness: it is not only final attributive goodness which swamps instrumental attributive goodness derived from it, but fundamental attributive goodness which swamps any form of attributive goodness derived from it, even if it is derived in a non-instrumental fashion. Consider the following idea: the attributive goodness of an item X of kind K is in two ways linked to X performing some K-specific function well. First, the aspect of X which makes it fundamentally attributively good as a K is its ability to perform the K-specific function well. This thesis was implicit in much of the foregoing discussions. Second, the overall attributive goodness of X as a K is the degree to which X performs the K-specific function well. A knife X is attributively better as a knife than some other knife Y iff X performs the knife-specific function better than Y. At least when it comes to the overall degree of goodness in question, evaluations in terms of attributive goodness exemplify the form of one-dimensionality I mentioned in the outlines of this section. If these two points are on the right track, we derive at the following conclusion: the only aspects which matter for the overall attributive goodness are the fundamentally attributive good-making features. While we can talk about aspects of items which make them pro tanto attributively good in a derived manner, these instances of pro tanto attributive goodness do not alter the overall attributive goodness of said item. Therefore, despite non-instrumental value derivations, the envisaged additional attributive goodness of a belief's amounting to knowledge is swamped by the fundamental attributive goodness based on its truth. Hence, the swamping problem for veritism is even harder to solve on AGV than it is on other approaches to epistemic value.

To conclude, it is possible to make sense of relations between values within an AGV framework. Thus, the general worry raised against AGV is out of place. Yet, the previous discussion has brought two troublesome aspects of these value relations to light. First, on AGV, instrumental value derivations appear to be restricted to de facto contribution rather than counterfactual promotion. This is in stark contrast to the way in which many epistemic axiologists make use of instrumental epistemic value. Second, AGV cannot make sense of the primary motivation of turning towards non-instrumental value derivations in epistemic axiology. Since the attributive swamping problem appears to be broader than its simpliciter version, non-instrumental value derivations are no solution to it. Therefore, rather than lending additional support to veritism as outlined in Section 2, Sylvan's version of AGV seems to create severe problems for it. I take it that both problematic aspects generate an additional abductive reason against AGV.

5. How to disagree about fundamental epistemic values

In the second section of this paper, I motivated the shift to AGV by an apparent shortcoming of Sosa's formulation of insulationism. If we determine the domain of epistemic evaluation by fixing the fundamental epistemic values, any diverging opinion concerning the fundamental epistemic values appears to be a shift in topic. AGV is able to circumvent this issue by fixing the domain of epistemic evaluation via a particular goodness-fixing kind with which epistemologists are engaged. In this section, I present three alternatives to AGV, which, on the one hand, stick to Sosa's main idea that ‘epistemic value’ denotes a domain-specific value. On the other hand, these theories provide different ways to fix the domain of epistemic evaluation, which, in turn, makes substantive disagreements about the fundamental epistemic values conceivable.

First, one might opt for a ‘light’ version of AGV. Following Sosa, epistemic evaluation is still an evaluation of hypothetical values. Which values are the fundamental ones of the epistemic domain, however, is determined by the fundamental attributive goodness of some kind. I think that Sosa's ‘world of coffee’ is to be understood along the lines of this light version of AGV (2007: 73). The fundamental value of our domain of coffee is determined by whatever features make coffee attributively good as coffee (being delicious, aromatic, etc.). The remaining part of the coffee point of view is concerned with hypothetical value: given that good coffee is valuable, one investigates what other states of affairs become valuable in virtue of it. Thus, even objects which are not coffee themselves (baristas, coffee machines, etc.) can be evaluated in terms of their coffee value. In an analogue fashion, Sylvan might water down his appeal to AGV: epistemic value is not the attributive goodness of beliefs, but a form of domain-specific value. The domain of the epistemic, in turn, is determined by the attributive goodness of beliefs. Such a view on the fundamental values of a domain of evaluation comes with the very same merits as AGV itself. However, it is not susceptible to the issues in Section 4, as well as the scope objection from Section 3.

Second, and along very similar lines, one might argue that what determines the fundamental value of some domains of evaluations is the fittingness condition of some attitudes.²⁶ Just as before, we understand the epistemic point of view to be engaged with a form of hypothetical value. Yet, what determines the fundamental values of the epistemic domain is whatever makes some attitude fitting. Since there can be substantial disagreements as to what makes an attitude fitting, this fittingness-centred approach exemplifies the very same advantages as AGV. Just as AGV-light, it also avoids the problems from Section 4, as well as the scope objection. Furthermore, such a theory does not appeal to any goodness-fixing kind. Thus, it is not susceptible to the Thomsonian challenge in Section 3.

Finally, one might also opt for a bottom-up approach, which is envisaged by Bjelde (2021). Epistemic axiology is seen as the project of ordering or systematising the different evaluations of epistemologists. In doing epistemology, we categorise things in terms of justification, knowledge, understanding, etc. All of these are positive features on the epistemologist's score card. In doing epistemic axiology, we try to find out whether this plurality of evaluations can be unified and put in axiological relations to each other. A disagreement about the fundamental epistemic values is simply a dispute about whether the importance of some evaluation, for instance, in terms of knowledge, can be understood in virtue of its axiological relation to some other evaluation, for instance, in terms of truth. If there is no such evaluation in virtue of which all the other elements on the epistemologist's score card can be accounted for, then one should opt for a form of pluralism concerning the fundamental epistemic values. Thus, the domain of the epistemic is fixed by the plurality of all kinds of epistemic evaluations and not by the fundamental epistemic value. Hence, on this bottom-up view, substantial disagreements about the fundamental epistemic values are clearly possible.²⁷ Furthermore, such a view appears to be immune to any of the objections to AGV.

To conclude, there are several variants of Sosa's basic insulationist position which come with the very same advantages as AGV. Nevertheless, they generate significantly less issues. Thus, any insulationist seems to have good reason to adopt one of these variants rather than AGV.

6. Conclusion

The goal of this paper was to investigate an underexplored position in the debate about the nature of epistemic value. I demonstrated that, on closer inspection, an AGV variety of insulationsm is no improvement on more standard versions of insulationism. If one supplies Sosa's general outlook on epistemic values with an independent account of how the domain of the epistemic is individuated, there is no advantage in shifting to AGV. Furthermore, AGV is the object of several shortcomings and has to make a lot of contentious commitments in order to be a viable account of epistemic value. The basic choice of which goodness-fixing kind epistemologists are talking about appears to force one into contentious commitments and the nature of relations among instances of attributive value generates some implausible upshots. Therefore, one should not view ‘epistemic value’ as denoting a form of attributive goodness.

The aims of this paper were quite modest. I just argued that AGV fares worse than other versions of insulationism and, thus, can be dismissed in favour of one of them. Yet, this is not to make an all-out recommendation in favour of these versions of insulationism. After all, the debate between substantivists and insulationists is still on open issue.²⁸