Evidentialism and the Problem of Basic Competence

1. Evidentialism and Competence

Evidentialists form a fairly heterogeneous group, and I cannot discuss every view falling under that heading. For our purposes, we can simplify the terrain a bit and think of all evidentialists as committed to (at least) the following principle:

E: An agent A’s evidence at a time, t, determines which inferences are justified for A at t.

As stated, this principle requires two points of clarification. First, E is meant to be a principle of propositional (or ex ante) justification, rather than a principle of doxastic (ex post) justification.

According to a standard way of understanding that distinction, in order for A to be doxastically justified in adopting doxastic attitude D, A must (i) be propositionally justified in adopting D, and A must (ii) adopt D on the basis of that which propositionally justifies it. Notice that condition (ii) of doxastic justification requires that an agent properly base her doxastic attitudes on her evidence, which already requires more of her than merely having that evidence. I will simply proceed as if the standard way of drawing the propositional-doxastic distinction is correct, in which case E is much more attractive and plausible as a principle of propositional—rather than doxastic—justification. In less technical terms, E concerns which inferences an agent would be justified in making, whether or not she makes them.

Second, this evidentialist principle is fairly restrictive in one sense: E is meant to apply only to cases of inferential justification, not to cases of non-inferential justification. An evidentialist principle more general than E might concern the relationship between one’s evidence and one’s total doxastic state, but this more general principle would expose itself to problems specific to debates around the nature of non-inferential justification.⁴ (Some of the attitudes in one’s total doxastic state, perhaps one’s simple perceptual beliefs, will be non-inferentially justified.) It is enough for our purposes to note that E, restricted as it is to cases of inferential propositional justification, has some initial plausibility.

With those qualifications laid out, the target of this paper is someone who accepts E, to whom I’ll refer as an “inferential evidentialist”. The general complaint I want to lodge against inferential evidentialists is that they are forced to accept an implausible account of an agent’s inferential competences. The idea, laid out more fully below, is fairly straightforward: if there are cases in which possessed inferential competence makes a difference to which inferences an agent would be justified in making, that competence is just more evidence. That follows from a more general commitment of inferential evidentialism: anything that makes a difference to which inferences an agent would be justified in making is just more evidence, since evidence is the only thing that makes such a difference.

Before I launch into a criticism of this view, I want to spend some time fleshing out the kind of explanation it provides. To that end, I’ll discuss an example that is meant only to be illustrative, not critical.

Birdwatching

Ed is an expert ornithologist; among other things, he knows how to recognize the species of a bird by the sounds of its songs. Neil is a novice birdwatcher; while he is not totally incompetent, Ed is his superior across most of the relevant dimensions. Suppose that Ed and Neil each hear a certain birdsong out in the brush, one which both men recognize as sounding like the song of a predatory bird. While Neil is quick to infer that the two hear a predatory bird, Ed is not. Instead, Ed’s makes a more modest inference: that the two either hear a predatory bird or hear an avian mimic.

Birdwatching is a scenario in which an expert and a novice share a body of evidence but arrive at different conclusions on its basis. Moreover, the expert and novice are each rational to arrive at different conclusions.

Just in case that characterization of the cases raised any eyebrows, here’s a fairly intuitive and (I hope) innocent way to think of what’s going on. If Neil is simply unaware of the possibility of avian mimics in the area, or if he had never considered such things as avian mimics, it would not make sense for him to infer, on the basis of hearing such a birdsong, that it might have come from an avian mimic. Since it would make no sense for him to consider the possibility of avian mimics, it makes no sense to fault him for ignoring it; most of us reject a conception of rationality that requires agents be sensitive to the evidence that there is, rather than to the evidence that they have.⁵ In contrast, part of what makes Ed the expert of the two is that he has acquired, through years of training, a certain competence in distinguishing bird species by their song, one which is fairly sensitive to the possibility of avian mimics. It’s natural to think of what’s going on here as a situation in which Ed’s competence makes a difference to which inferences he’s justified in making. In particular, Ed’s competence justifies him in making a more cautious inference than Neil, in virtue of making him sensitive to a possibility of which Neil is unaware.⁶

It’s worth emphasizing that Birdwatching is not a recherche thought experiment meant for the only the most hardened epistemological intuitions; instead, it is meant to be a stylization of a totally ordinary and familiar situation, one that any plausible epistemology of inference should be able to explain.⁷ And, fortunately, inferential evidentialists can easily explain it.

In rough outline, the inferential evidentialist explanation of birdwatching and related cases goes like this. Differences in competence can give rise to differences in propositional inferential justification, but two agents cannot differ in competence while having exactly the same evidence. Instead, the expert and novice share only some evidence; in the case of Neil and Ed, their shared evidence might be something like that the birdsong sounds thus-and-so. But the expert has a great deal more evidence, perhaps only in the background, that the novice lacks. Ed, for example, plausibly knows that if a birdsong sounds thus-and-so, it comes from either a predatory bird or avian mimic, while Neil does not. It is the expert’s possessing certain background evidence that the novice lacks that explains why their inferences could rationally diverge.

Here is what I hope to draw out of the discussion so far: the story that inferential evidentialists must tell about competence is a reductive one. If there are cases in which competence makes a difference to which inferences an agent would be justified in making, competence is just more evidence. That follows from a more general commitment of inferential evidentialism: anything that makes a difference to which inferences an agent would be justified in making is just more evidence, since evidence is the only thing that makes such a difference.

Given that inferential evidentialists have to accept a reductive account of competence in terms of evidence, one might reasonably wonder: if competence just is evidence, which evidence is it? Presumably, the most plausible thing to say on behalf of inferential evidentialists is that this competence-constituting evidence must in some way link an agent’s first-order evidence to various hypotheses. Call this the “linking principle”:

Linking principle: For an agent to be justified in inferring H from first-order evidence E, given her competence, her competence-constituting evidence must link E to H.

The linking principle is meant to be somewhat of a precisification of, and somewhat of a substantive constraint on, the reductive view of competence. Of course, the language of “linking” leaves room for ambiguity. As I explain in the next two sections, how we think of the “linking” characteristic of competence-constituting evidence will depend on the version of evidentialism in question. But as a rough and ready pass, competence-constituting evidence could perform this linking function if we thought of it as consisting of conditionals, the antecedent of which is a body of first-order evidence, and the consequent of which is a hypothesis supported by that evidence. (We will consider other ways to flesh out “linking” besides this one.) And if we accept that competence consists in nothing more than having these conditionals as evidence, we could straightforwardly explain how a difference in competence between two agents could result in divergent rational inferences from shared first-order evidence. For instance, in Birdwatching, Ed is justified in making a different, more cautious inference than Neil because Ed’s competence-constituting evidence (including, perhaps among many other things, the conditional if a birdsong sounds thus-and-so, it comes from either a predatory bird or avian mimic) provides a link to a different space of hypotheses.

I want to briefly comment on potential objections from two directions. First, some readers might think that inferential competence, whatever it is, is only relevant to questions of doxastic justification. Presumably, the motivation for this position stems from the idea that manifesting inferential competence is a way of making properly based inferences, and that rings true in my ears. But we should not be tempted to slide from the idea that manifesting inferential competence is exclusively relevant to questions of doxastic justification to the idea that possessing inferential competence is exclusively relevant to questions of doxastic justification.⁸ In fact, in the Birdwatching case above, we have already shown how possessing inferential competence is relevant to propositional justification: inferential competence is sometimes constituted by relevant background evidence. As such, the problem with this objection is that it goes too far; it blocks my criticisms at the cost of undermining the very plausible story that evidentialists already tell about cases like Birdwatching. A working assumption of this paper is that there is some initial plausibility to the idea that inferential competence performs a kind of theoretical double-duty: its manifestation is relevant to doxastic justification, while its possession is relevant to propositional justification. The interesting question, then, is not whether but how inferential competence is relevant.

Second, some readers might be inclined to explain the relationship between inferential competence and inferential justification by appeal solely to features of non-inferential justification, not by way of the linking principle. Authors like Siegel (2010) and Chudnoff (2020), for instance, have defended positions according to which experts and novices sometimes differ in what they are non-inferentially justified in believing by differing in the contents of their experiences, and these differences in non-inferential justification “bubble up” to give rise to differences in inferential justification. Somewhat roughly put, experts sometimes “just see” things differently than non-experts, and this difference explains why experts and non-experts can differ in which inferences they would be justified in making. For instance, there may be cases where a radiologist “just sees” a compound fracture in the x-ray, whereas the patient might “just see” a broken bone. In such a situation, the radiologist would, having different non-inferentially justified perceptual beliefs, much more plausibly be in a position to settle the question of whether surgical intervention would be required.

But even if we accept these sorts of “cognitive penetration”-inspired views of non-inferentially justified perceptual belief and its connection to inferential justification, there is a residual question—the one that we are concerned with here—about what further differences might explain why sometimes experts and non-experts who really do share a body of evidence differ in what they could justifiably infer from that evidence. It strikes me as implausible that all interesting questions about inferential justification bottom out in a theory of non-inferential justification, and to deny this much seems to simply beg the question. And anyway, it is my hope that exploring these issues will supplement existing views about the relationship between expertise and non-inferential justification, rather than compete with them.

Having addressed some early concerns, let’s briefly take stock. I have presented a core inferential evidentialist thesis, E, and I showed how a commitment to E brings with it a reductive view of inferential competence in terms of evidence. I then offered the linking principle as a plausible way to characterize which evidence constitutes one’s competence, according to the reductive view. I have acknowledged a few reasons to doubt how I’ve set up the problem: perhaps one is skeptical about the role of competence in propositional inferential justification, or perhaps one thinks that all interesting questions about propositional inferential justification bottom out in appeals to non-inferential justification. Both of these concerns overreach.

In the next section, I’ll look at cases that are difficult for the reductive view of competence to explain. The difficulty, I suggest, is due to the fact that the reductive view is only suited to explain acquired competences, and not all competences are acquired.

2. Propositionalism and Difficult Cases

There are two different ways of developing evidentialism: one can take a propositionalist or a non-propositionalist view of evidence. Both of these views, I will argue, have troubles adequately explaining the role of competence in inferential justification. In this section, I focus on propositionalist versions of evidentialism.

According to “propositionalism”, the following claims are true (in addition to E):

• Evidence is, by its nature, propositional.

• Having evidence is a matter of standing in some special epistemic relation to the propositions which constitute one’s evidence.

The first commitment distinguishes propositionalism from non-propositionalism, which I consider in the next section. The second commitment is meant to cast a wide net across a few battlelines in epistemology over the nature of the special epistemic relation; one might think that in order to have a proposition P as evidence, an agent must know that P, rationally believe that P, or merely believe that P. I wish to remain neutral on any of these particular characterizations of propositionalism.

Qualifications aside, here is what propositionalist versions of inferential evidentialism might look like: consider Fumerton’s (1995) “Principle of Inferential Justification” (“PIJ”), according to which, for an agent to be propositionally justified in inferring H on the basis of E, the agent must (1) be justified in believing E, and (2) be justified in believing that H makes E probable (also see Hasan 2013; Foley 1993: e.g., ch. 3, §2).⁹ A PIJ-style view satisfies the linking principle articulated in the last section; evidence of the form if E, then H is probable is well-suited to “link” E and H.

On a view like Fumerton’s, inferential competence consists in nothing but justified beliefs in certain conditionals, the antecedent of which is a body of evidence, and the consequent of which is a hypothesis. This sort of view is equipped to explain Birdwatching and similar cases: Ed is justified in making a more cautious inference than Neil because Ed justifiably believes, perhaps only in the background, that if a birdsong sounds thus-and-so, it comes from either a predatory bird or avian mimic, while Neil does not.

I admit that very many cases of justified inference can be handled within this sort of framework; very many cases of expert inference can be explained by pointing out background commitments, justifiably held, that are characteristically present in experts but absent in novices. But, as I hinted at the end of Section 1, this framework is only suited to explain rationally acquired expertise in inference, and not all expertise is rationally acquired.

To flesh out this claim, start with the idea that some competences are “basic”; they are deployed in the rational acquisition of other (non-basic) competences without themselves needing to be rationally acquired. Short of offering a general account of what distinguishes the basic from the non-basic competences, there are certain clear candidates for bona fide basic competence: competence with reasoning by enumerative induction, representativeness, and inference to the best explanation all have some claim to count as basic, if any inferential competence does. For instance, one does not learn how to use enumerative induction by employing other rational capacities; one’s capacity for induction is a kind of starting point for rational learning.

Of course, if the reader disagrees with me on the particular candidates for basic competence, I encourage them to read the cases with their preferred candidates in mind. If, instead, the reader is suspicious about the existence of basic competences, I hope to show by the end of this section that the cost of denying their existence is rather steep.

To get to that point, it will help to consider a few more cases, ones which focus on (putatively) basic inferential competences. Each case presented below involves an agent getting evidence by observation that the first 99 balls pulled from an opaque urn known to contain 100 balls are all black. Call this evidence “O”. Antecedently ignorant of the colors of any of the balls in the urn, each agent considers whether the color of the 100^th ball is black. Call the proposition that the 100^th ball is black “B”. Here, then, are the cases:

Case 1: Larry gets evidence O, and then, taking some time to reflect on his observations, justifiably believes that if O, then probably B. Being fully aware of the colors and quantity of the balls so far observed, Larry’s confidence that B increases; suppose he began inquiry by suspending judgment on B, and he now believes that B.

Case 2: Moe gets evidence O, but he has also been presented strong but misleading evidence by his most trusted colleagues that it is not the case that if O, then probably B, and he believes the reports of his colleagues.¹⁰ Being fully aware of the colors and quantity of the balls so far observed, Moe’s confidence that B nevertheless increases; suppose he began inquiry by suspending judgment on B, and he now believes that B.

In case 1, Larry is meant to be a paragon of rationality for propositionalists; he satisfies PIJ and so would be all-things-considered rational to infer B from O. Moreover, Larry seems to possess whatever background evidence constitutes competence with enumerative induction: in this case, his competence would consist in his having as evidence conditionals of the form if O, then probably B. Let’s grant this for the sake of argument.¹¹

In case 2, Moe fares very badly by the lights of propositionalism. Not only does Moe infer B from O despite his (misleading) linking evidence, it seems that he does not have the right kind of linking evidence to count as fully competent with enumerative induction in the first place. (We can imagine, in Moe’s case, that the right kind of linking evidence has been defeated by his colleagues’ testimony.) Moe is, at worst, both irrational for failing to respect his linking evidence and less-than-fully competent with enumerative induction, or, at best, neither rational nor irrational because totally incompetent. This strikes me as entirely the wrong result; not only is Moe not incompetent with enumerative induction, he does something remarkable.

To claim that Moe does something remarkable is not to claim that Moe is all-things-considered rational or fully epistemically justified to believe that B, at least if that is taken to mean that Moe is believing as an ideal epistemic agent would believe. Among other things, ideal epistemic agents are perfectly coherent, and Moe is plainly not. But here is something to say in favor of Moe’s belief: his first-order evidence in fact supports it, and he comes to believe it by disregarding higher-order evidence that is in fact misleading. At the very least, Moe’s inference does not wear its irrationality on its sleeve; it is not a paradigm case of all-things-considered irrationality.

To get a sense of how Moe’s inference might be afforded some positive epistemic status, compare Moe’s predicament to that of Huckleberry Finn: Huck is in a position of deciding whether to turn Jim in to slave catchers, and he feels deeply conflicted about doing so. On the one hand, Huck seems to believe that morality requires that he turn Jim in. After all, Jim is a fugitive slave, and Huck has been raised to host a range of negative beliefs about slaves in general, and fugitive slaves in particular. On the other hand, Huck cannot bring himself to do it. Jim is also Huck’s friend. Something drives Huck to free Jim, but it is certainly not Huck’s “better judgment” about what morality requires of him. On pain of classifying Huck as a moral monster, or a merely accidental do-gooder, many authors in the moral responsibility literature have attempted to accommodate “inadvertent moral virtue” into their accounts of morally worthy action. Perhaps, as Arpaly (2002) and Arpaly and Schroeder (2013) have suggested, the fact that Huck is moved by some deep-seated, intrinsic concern for the right-making features of action (in this case, Jim’s humanity) explains why Huck is virtuous, albeit inadvertently.

Put in those terms, Moe’s predicament looks very much like an instance of a related phenomenon, one that is sometimes called “inadvertent epistemic virtue”.¹² Cases of inadvertent epistemic virtue are instances of rationally permissible akrasia, a particular kind of inner-conflict between one’s first- and higher-order beliefs. Brian Weatherson (2019) discusses a case of inadvertent epistemic virtue in which a testimonial skeptic, Aki, comes to believe something on the basis of her friend’s testimony. (Testimonial skepticism is the view that one cannot gain knowledge by testimony.) Aki, hearing her friend testify that the Tigers won the night before, comes to believe that the Tigers won the night before, and in so doing ignores her belief in testimonial skepticism. (This belief is firmly held in the seminar room, we can imagine.) Weatherson is inclined to treat Aki as a “paragon of rationality” (2019: 171). He defends that claim on the grounds that Aki’s first-order evidence evidentially screens-off her judgment about what that evidence supports (see Weatherson 2019: ch. 11). But we don’t have to go as far as defending Aki as a paragon of rationality by appeal to evidential screening-off to recognize some positive epistemic status in Aki’s belief, any more than we have to defend Huck as a paragon of morality to recognize some positive moral status in his freeing Jim.

Moe, like Aki, infers that B despite his “better judgment”—which in this case provides him with evidence linking his first-order observations to various hypotheses—about what counts as evidence for what. Of course, someone might rightly insist that Moe would be “epistemically improved” or “closer to a rational ideal” if he were not merely inadvertently epistemically virtuous. Granted, not all cases of akrasia are rationally permissible, and perhaps no case of akrasia is rationally ideal, but there is a lot of daylight between accepting that coherence between first- and higher-order evidence is a rational ideal, on the one hand, and accepting that less-than-fully-ideal rational agents can sometimes make rational inferences despite this kind of incoherence, on the other.¹³

The point here is not to defend any particular account of inadvertent epistemic virtue, nor to defend the implication of any such account that there are cases of rationally permissible akrasia. I accept that implication, and others have defended it extensively.¹⁴ The point here is much more modest: Moe’s belief that B seems to have a positive epistemic status, perhaps short of ideal rationality, one that is shared with Aki’s testimonial belief, and one that is analogous to the positive moral status of Huck’s freeing Jim, an action which itself might fall short of ideal morality. Propositionalism seems to be insensitive to these finer details of epistemic assessment. If Moe is irrational to infer B and less-than-fully competent with enumerative induction, it’s hard to see what positive epistemic status an inference to from O to B could have for him. There seems to be nothing about Moe that would even prima facie justify an inference from O to B.

What about the idea that Moe is totally incompetent? If so, this would surely undermine my claim that he is prima facie rational to infer B from O.

One consideration that brings out the implausibility of this idea is that, if we take seriously the idea that competence consists in having certain linking evidence, we will hold competence hostage to arbitrarily higher-order defeaters. This proposal has severe costs; to see why this is so, compare Moe’s case to Curly’s:

Case 3: Curly gets evidence O, and he, like Larry, has come to justifiably believe that if O, then probably B. Curly has, however, been presented with strong but misleading evidence by his most trusted colleagues that it is not the case that if (O and if O, then probably B), then probably B, and he believes the reports of his colleagues. Being fully aware of the colors and quantity of the balls so far observed, Curly’s confidence that B nevertheless increases; suppose he began inquiry by suspending judgment on B, and he now believes that B.

On one way of reading Curly’s case, he has the same first- and second-order evidence as Larry, the paragon of rationality for propositionalists. In that case, Curly would, like Larry, satisfy a PIJ-style constraint, and his inference would be all-things-considered rational. But this would render Curly’s third-order evidence doubly normatively irrelevant; that third-order evidence would neither affect which inferences would be rational for Curly to make, given his first-order evidence, nor would it constitute Curly’s competence with enumerative induction.

More plausibly, I think, a propositionalist would insist that Curly’s second-order linking evidence is defeated by his third-order linking evidence. What they would require in so insisting is a kind of “mesh” between higher-order linking evidence.¹⁵ Unless one’s second-order linking evidence appropriately meshes with one’s third-order (and so on) linking evidence, one is not genuinely competent in the first place, or so says the propositionalist. But if this is the line one takes, then it is not clear that one can stop at any finite level of higher-order evidence; competence would, in some sense, consist of an infinity of “higher-order” evidence linking one’s “lower-order” evidence to some hypothesis. It seems plainly wrong that any justified inferences (for things like us) are justified by an infinity of further, higher-order beliefs about what our evidence supports.

Compare this position to Achilles’ predicament in “What the Tortoise said to Achilles” (Carroll 1895). There, the Tortoise asks Achilles to say what must be added to someone who accepts some proposition A, and who accepts that A logically entails B, but who fails to recognize that they must thereby accept B (or reject a premise). Achilles goes through a process of appealing to ever-higher-order premises, linking A, A implies B, and B; the absurdity of this response is apparent. And the dialogue is often taken to show, at the very least, that to follow an inference rule (in this case modus ponens) is not to infer by way of an additional (perhaps suppressed) premise, on pain of regress. The moral of Carroll’s dialogue generalizes; for any inference rule one can create a Carroll-style regress to motivate the thought that to infer competently is not merely to infer by way of an additional premise. The contrast between Larry, Moe, and Curly is merely a case in point.¹⁶

These sorts of regress considerations, I hope, present a bit of a defense of the cogency of basic competence, even to those who disagree with me about the particular candidates I suggested (enumerative induction, representativeness reasoning, and inference to the best explanation). Unless one appeals to basic inferential competence somewhere, one is left with a view according to which putatively basic competences consist of an infinitely large mesh of ever-higher-order linking evidence. Even if, as I’ve granted in Section 1, there are plenty of everyday cases in which a particular inferential competence may consist in nothing more than the possession of certain linking evidence, this concession should not be taken to indicate that, in general, to be competent is to possess the right sort of evidence; it should certainly not lead us to think that, rather implausibly, that inferential competence consists of evidence “all the way down.”

Despite its problems, the initial appeal of propositionalism might derive from the fact that following one’s evidence is an ideal of rationality, and someone who ignores any of her evidence falls short of this ideal, whether or not the evidence she ignores turns out to have been misleading. Moe, insofar as he ignores some of his higher-order evidence, falls short of this ideal. Similar remarks apply to Aki, and perhaps also to Huckleberry Finn. But I have tried to stress that it is one thing to fall short of a rational ideal, and it is quite another to be incompetent with a rule of inference. These problems suggest that being basically inferentially competent does not consist having evidence that satisfies the linking principle, at least if evidence is construed propositionally.

In the next section, I argue that non-propositionalist versions of inferential evidentialism have the virtue of avoiding both problems for propositionalism. Non-propositionalism does, however, invite a different sort of problem, one which should push us to look for alternatives to inferential evidentialism.

3. Non-Propositionalism and Idle Explanations

The arguments in Section 2 attacked a propositionalist version of inferential evidentialism. Someone who espoused that sort of view would endorse three claims: E; that evidence is, by its nature, propositional; and that having evidence is a matter of standing in some special epistemic relation to the propositions that constitute one’s evidence. In contrast, non-propositionalist versions of inferential evidentialism (see, e.g., Conee & Feldman 1985; 2001; and Plantinga 1983) can be characterized by their acceptance of E, plus the following:

• An agent’s evidence at a time consists of her total mental state at that time.

Non-propositionalists have a strictly more expansive notion of ‘evidence’ than propositionalists. For any bit of evidence, P, a propositionalist would countenance an agent A as having, they would do so only because A stood in a special epistemic relation to P. Standing in such a relation to P is plausibly a mental property of A. Consequently, the non-propositionalist would countenance the agent’s standing in such a relation to P as part of her evidence. So, whatever propositions a propositionalist would countenance as part of A’s evidence, it seems a non-propositionalist would countenance a corresponding bit of evidence that is A’s attitude towards those propositions. But the non-propositionalist would also countenance further, non-propositional mental states (mental dispositions, skills, bare sensations without propositional content, and so on) as part of an agent’s evidence.

For non-propositionalist versions of inferential evidentialism, basic competence is still evidence, but it is not evidence that is determined entirely by the propositional mental states of the agent. For instance, competence with enumerative induction may consist, on such a view, in nothing over and above a cluster of mental dispositions to make certain inductively supported inferences. While the stimulus and manifestation conditions of such dispositions may implicate certain propositional mental states of the agent, the disposition itself is not a propositional mental state of the agent.

Non-propositionalist versions of inferential evidentialism avoid the two problems I raised for propositionalism because they are not forced to accept that inferential competence consists in propositional mental states, but they nevertheless retain the idea that inferential competence consists in having certain evidence (by occupying certain mental states). For one thing, there is no Carroll-style regress threatening a view that appeals to non-propositional evidence; non-propositional mental dispositions are not the kinds of things that could serve as premises in inference, so there is no vicious regress of ever-higher-order premises looming. Moreover, non-propositionalists do not obviously face the problem of denying inadvertent epistemic virtue. Non-propositional linking evidence is not the agent’s “better judgment” against which they can act in cases of inadvertent virtue, since judgments (explicit or not) are propositional mental states.

These considerations suggest that non-propositionalism is better suited to explain the relationship between competence and inferential justification than the view considered in the last section. But here I want to stress that the relative advantage non-propositionalism enjoys should not distract us from recognizing that that view has a very different sort of problem. In particular, non-propositional linking evidence does not perform what might be naturally thought of as any of the characteristic normative functions of evidence. In short, the sort of non-propositional linking evidence in question does not make the right kind of normative difference to count as evidence at all.

Of course, non-propositional linking evidence makes some normative difference on these views. For instance, if an agent lacks the relevant competence-constitutive evidence, then she is incompetent, and so she is not propositionally justified at all in making inferences that would be the manifestation of such a competence. Were Larry incompetent with the rule of enumerative induction, for instance, he would lack propositional justification for believing that B, given O. But competence with enumerative induction does not itself raise the probability of inductive inferences; rather, it is what explains why his first-order observations O raise (rather than lower) the probability of various inductively supported hypotheses. Competence looks rather like a mere enabling condition on justified inference, normatively on a par with other, more mundane enabling conditions, like there being oxygen flowing to the agent’s brain—these are conditions that must be satisfied in order for an inference to be justified, but not because they bear directly on the justification of the inference.

Perhaps in and of itself, the claim that basic competence is an enabling condition on justified inference is not objectionable; it may be that the first-order evidence, all on its own, supports one inference over another, but it is only in the presence of basic competence that that inference is, in some sense, ‘available’ to the agent. Of course, if non-propositional linking evidence is a normatively relevant enabling condition, we are owed an account of what this normative relevance comes to, one that does not simply amount to special pleading.¹⁷ Absent that account, it would appear that the non-propositionalist avoids the problems of propositionalism only at the cost of positing “idle” evidence.¹⁸

It is worth taking some time to emphasize what is objectionable about appealing to idle evidence. Importantly, appeals to idle evidence flout a plausible meta-epistemological view about the nature of evidence, one according to which evidence just is that which performs certain characteristic normative functions. It is widely accepted that evidence is that which makes a difference to the degree to which an agent is justified in adopting certain attitudes towards various propositions,¹⁹ it is that which makes a difference to the resilience of one’s degree of belief (as when one gets evidence supporting what one already knows), and it is that which could serve as a basis for justified belief.²⁰

This meta-epistemological claim is meant to be as neutral as possible. It is neutral with respect to the nature of justification more generally, with respect to whether the explanatorily fundamental doxastic states are binary or admit of degrees, with respect to the relationship between binary and degreed doxastic states, whether evidentialism is correct, etc. It is, to put it plainly, just meant to encode a truism. That “idle evidence” fails to satisfy any of these difference-making conditions characteristic of evidence would seem to undermine its explanatory power as evidence. Again, idle evidence does not seem to make the right kind of difference to count as evidence at all. Absent independent motivation, one wonders what, beyond an unshakeable commitment to E, would make this option look attractive.

To sum up: the last two sections have motivated the thought that inferential evidentialists face serious problems in accounting for basic competence, whatever their favored account of evidence. Either evidence is propositional, or it is not. If it is, then inferential evidentialists must either deny the phenomenon of inadvertent epistemic virtue or accept a vicious regress; if it isn’t, they draw normative distinctions that seem to mark no normative difference. Both routes lead to the idea that the normative contribution of basic competence is not captured by the linking principle. In light of the problems outlined in this and the last section, one should look for alternatives to inferential evidentialism.

Before I present my own view, however, it is instructive to compare the position criticized here to others that bear a family resemblance to it. First, take mentalism, the view according to which any difference in the epistemic status of a doxastic state between two agents supervenes on a mental difference between them (e.g., see Conee & Feldman 2001). (For all mentalism says, these mental differences may or may not be evidential differences.) Mentalism does not have a problem accommodating my claim that basic competence makes a non-evidential difference to justification, since mentalism does not claim that evidence is the only thing that makes a difference to justification; mental dispositions, skills, and bare sensations without propositional content, perhaps among other things, can be normatively relevant without being evidentially relevant. On these sorts of views, it is possible for two agents to share a body of (first-order) evidence but differ in non-evidential mental states, and as a result differ in inferential propositional justification. This is how someone broadly sympathetic to mentalism would explain how expertise could be normatively relevant to inferential justification in a distinctly non-evidential way.

Next, take various forms of Bayesian confirmation theory. These views treat evidential support as a three-place relation between a subject’s evidence (typically a set of propositions or sentences expressing them), a hypothesis, and that subject’s conditional credences. Conditional credences are not evidence—not even “higher-order” evidence. Instead, they encode something like an agent’s evidential standards, what she takes to be evidence for what.²¹ On these sorts of views, it is possible for two agents to share a body of (first-order) evidence but differ in evidential standards, and as result differ in inferential propositional justification. This is how someone broadly sympathetic to Bayesian confirmation theory would explain how expertise could be normatively relevant to inferential justification in a distinctly non-evidential way.

Both mentalism and Bayesian confirmation theory have the theoretical resources to respect what seems to be a fundamental epistemological distinction between the mental states that function so as to provide an agent with a basis for justified inference and the mental states that function so as to guide how one bases one’s inferences in a way that counts as justified. It is a serious mark against non-propositionalist version of inferential evidentialism that it runs these two things together.

In the next section, I lay out my own view, “inferential dogmatism”, so-called because it is structurally very similar to various forms of dogmatism defended in the epistemology of perception. After presenting a sketch of inferential dogmatism, I try to differentiate it from standard forms of Bayesian confirmation theory while situating it among other forms of mentalism.

4. An Alternative to Inferential Evidentialism

I have motivated the idea that being basically inferentially competent is not simply a matter of having certain evidence. We should reject the reductive account of basic competence in terms of evidence, and the more general reduction of being a competent reasoner in terms of having certain reasons.

This idea is captured by the following constraint on any plausible epistemology of inference:

BC: The normative contribution that basic inferential competence makes to what an agent is rationally permitted to infer is not an evidential contribution.

This constraint is purely negative; it says what the normative contribution of basic competence cannot be. It does not say what basic competence is, nor does it say what the normative contribution of basic competence is, if not an evidential one.²²

This section has two aims. The first is to sketch and defend a positive proposal, “inferential dogmatism”. This will go some way towards explaining what the normative contribution of basic inferential competence comes to. The second is to situate inferential dogmatism within the range of views that plausibly satisfy BC. For instance, In the last section I indicated that mentalism and certain forms of Bayesian confirmation theory seem to satisfy BC, at least to the extent that those views permit non-evidential factors to make a difference to inferential justification. But, depending on how we think of the nature of the conditional credences to which Bayesian confirmation theorists appeal, inferential dogmatism will have a better explanation of how cases of inadvertent epistemic virtue are both possible and prima facie rational.

5. Conclusion

Inferential evidentialism is a popular and attractive view that explains what I’ve called “basic” epistemic competence, which provides “basic” inferential justification, in terms of derivative or mediate justification by way of background evidence linking one’s first-order evidence to what it is evidence for. I raised a number of problems for two versions of this view—for propositionalist versions, that they faced a regress of premises, and that they failed to countenance inadvertent epistemic virtue; and for non-propositionalist versions, that they only avoided those problems at the cost of positing idle evidence.

These problems suggested that the normative contribution of basic competence is not an evidential one; to be competent with a rule of inference is not, in general, simply to possess certain linking evidence. I then sketched a dogmatist view of basic inferential justification according to which one’s basic competences give immediate, defeasible justification for certain inferences. While some authors have been attracted to dogmatic views for cases of non-inferential justification, dogmatic views have been largely ignored in cases of inferential justification. This under-explored option, inferential dogmatism, avoids the objections I raised to inferential evidentialism in its various forms. And ultimately, this position vindicates the natural thought that being a competent reasoner is not simply a matter of having certain reasons to believe this or that, even if a good deal of (derivatively) competent reasoning can be adequately explained in those terms.