Part II - Resistance to Evidence and Epistemic Proper Function

Chapter 6 Resistance to Evidence as Epistemic Malfunction

This part of the book develops a full positive epistemology: an account of the epistemic normativity of evidence resistance, in conjunction with novel accounts of epistemic oughts, evidence, defeat, and permissible suspension. This first chapter argues that resistance to evidence is an instance of epistemic malfunction. It first puts forth a normative picture according to which the epistemic function of our cognitive systems is generating knowledge, and epistemic norms drop right out of this function. Second, it shows how this picture accommodates epistemic obligations, which, in turn, explain the normative failure instantiated in cases of resistance to evidence. According to this view, cognitive systems that fail to take up easily available evidence and defeat instantiate input-level malfunctioning. Input-level malfunctioning is a common phenomenon in traits the proper functioning of which is input dependent, such as our respiratory systems. Since our cognitive systems, I argue, are systems the proper functioning of which is input dependent, we should expect the failure at stake in resistance cases.

6.1 Epistemic Oughts

Let us start with the elephant in the room: the epistemic impermissibility of evidence resistance implies that there are such things as epistemic oughts to govern our practices of belief forming, updating, and maintaining. As such, any epistemology that is able to predict epistemic impermissibility in resistance cases will be an epistemology that is able to incorporate epistemic oughts.

This is not trivial. It is not trivial, first and foremost, methodologically: until very recently, normative work in epistemology has, for the most part, been negative, in that it has concerned itself with restricting what we are permitted to, for example, believe, assert, or use as a premise in reasoning. Investigations into epistemic oughts are thin on the ground.¹

It is also not trivial normatively: permissions are not easily turned into obligations (henceforth I am following the literature in deontic logic² in using ‘oughts’ and ‘obligations’ interchangeably: nothing hinges on it – I am employing a notion of obligation that merely maps onto an ought). Just because I’m permitted to believe things about objects located towards the periphery of my visual field, it does not follow that I ought to do so – after all, a lot of things are happening in my visual field; I cannot possibly be expected to form beliefs about all of them. Furthermore – and going back to methodological difficulties – some permissions endorsed by traditional epistemological frameworks simply don’t speak at all towards the kind of obligation I’m breaching when I’m evidence resistant. We have seen this problem surface clearly with permissions to believe in virtue of knowing: this permission is silent when it comes to obligations to update. Similarly, think of classical process reliabilism: even if we grant its champions that we are permitted to believe the outputs of our reliable belief-formation processes, the view remains silent on the norms I’m breaking when my processes fail to deliver outputs that they should deliver.

What we need is an epistemological framework that has the resources to be naturally extended to incorporate obligations to update.

Another, more well-known difficulty has to do with the nature of these epistemic oughts: most people think voluntarism about belief is false. Even bold voluntarists would likely accept that voluntarism is false about the vast majority of doxastic phenomena. Notably, people have worried about non-voluntarism being incompatible with epistemic oughts for belief: if ought implies can, the thought goes, and since I cannot believe at will, I cannot be subject to norms obliging me to do so either.

Now, a lot of ink has been spilled on the implausibility of an unrestricted ‘ought-implies-can’ principle for normativity in general, and on putting forth more or less successful restrictions thereof (e.g. see Ryan 2003 for excellent work on this topic). One thing that has become clear in recent years in epistemology is that, while we don’t want an individual-based ‘ought-implies-can’ to constrain our normativity – after all, we don’t want the sexist cogniser to come out as justified in his sexist beliefs just because he can’t believe otherwise due to his sexism – we do want general facts about limitations pertaining to adult human cognitive architecture to play a role in restricting our normative claims. Norms governing ideal epistemic agents are just not very informative, nor is there a straightforward way to go from ideal to non-ideal epistemological theorising: for instance, it’s not clear what the obligations of an ideal agent who knows that they know for all of their pieces of knowledge imply for people like me, who can hardly come anywhere close to such achievements (nor do we particularly want to).

Keeping all of this in mind, several views in the literature attempt to offer accounts of epistemic ought that bypass the voluntarism objection; I will not run through all of them in great detail here. However, one thing that transpires from even a brief sketch of the literature is that the endeavour of incorporating epistemic oughts in one’s epistemology is faced by a strength dilemma: make the source of the ought too thick, normatively, and it seems incompatible with non-voluntarism. Make it too thin and it seems to fail to capture intuitions of ‘normative oomph’ when it comes to the epistemic (i.e. that epistemic norms are, in a significant way, ‘heavier’ than mere conventions). To see the dilemma, let’s look at a couple of classic proposals.

According to Richard Feldman, epistemic obligations are what he calls ‘role oughts’:

There are oughts that result from one’s playing a certain role or having a certain position. Teachers ought to explain things clearly. Parents ought to take care of their kids […]. Incompetent teachers, incapable parents […] may be unable to do what they ought to do. Similarly, I’d say, forming beliefs is something people do. That is, we form beliefs in response to our experiences of the world. Anyone engaged in this activity ought to do it right. In my view, what they ought to do is to follow their evidence (rather than their wishes or fears). I suggest that epistemic oughts are of this sort – they describe the right way to play a certain role.

(Feldman 2000, 677)

Feldman’s account is the paradigmatic case of a theory that is affected by the lack of ‘normative oomph’ worry: after all, role oughts can – and often are – in an important sense uninteresting, or even bad oughts. Role oughts generated by the role of mafia boss, for instance, seem too normatively thin to constitute the right kind of model for epistemic normativity.

A similar worry affects Hilary Kornblith’s view: according to Kornblith, a desire for truth, together with the corresponding hypothetical imperatives it generates, is all we need for explaining epistemic oughts. That’s all there is to epistemic normativity: not much more ‘normative oomph’ than this is needed to explain the data we need to explain.

I share Kornblith’s naturalism as well as, to a large extent, his ‘normative oomph’ scepticism. I think, however, that a desire-based picture remains unsatisfactory on precisely the desiderata Kornblith and I both endorse: naturalistic friendliness. To see why, note that desire is also, plausibly, normatively constrained: there are such things as more and less fitting desires. In turn, hypothetical imperatives are only as thick, normatively, as the desire triggering them is fitting. Ideally, we want to explain this datum in a naturalistically friendly fashion: we want to be able to predict – in a naturalistic normative framework – why some norms are thicker than others, and we want, ideally, to get the result that the epistemic is on the thicker side of things than, again, norms generated by mafia-boss-characteristic desires. In other words, we should be able to predict that nature cares more about the epistemic than about the mafia. Kornblith’s response is that, as opposed to other desires, a desire for truth is pretty much implied by a desire for anything else, since whatever else one might be trying to achieve, truth will be needed in order to make the right decisions to the aim of getting it. This instrumental way of looking at things doesn’t help much with a mafia boss-type issue, though: after all, on this instrumental picture, a desire for truth will be just as fitting as the desires it is instrumental to fulfilling; in this, the view fails to beef up the normative oomph-ness of truth-generated norms.

Finally, Matthew Chrisman ventures to explain belief oughts in a non-voluntarist-friendly fashion by pointing out a classic distinction between norms governing actions, or ‘ought-to-dos’ – which may be subject to some variety of the ought-implies-can principle – and ‘rules of criticism’, governing states, or ‘ought-to-bes’, which, Chrisman thinks, are not thus constrained. Here is Chrisman:

In developing an account of robustly normative claims about what someone ought to believe, my strategy is to treat these as adverting to a species of state norms. For instance, the claim, ‘You ought to believe you are reading this text right now’ could be understood to be an instance of the general form, ‘S ought to be in doxastic attitude A towards proposition p under conditions C’. Then, the crucial observation is that some such normative claims seem to be true, just like ‘Clock chimes ought to be disposed to strike on the quarter hour’, ‘The beds ought to be made by 8am every morning’, ‘A child ought to be able to tie their shoes by age seven’ […]. Yet these true normative claims don’t presuppose that their subjects be capable of voluntarily following the rule.

(Chrisman 2008, 362)

The worry for Chrisman’s account is, once more, one having to do with ‘normative oomph’: state norms are clearly not always normatively thick – indeed, several of the examples given are paradigmatic examples of social standards of correctness. Is there any interestingly thick normative sense in which beds that are unmade by 8 a.m. are defective? Not really. If so, we need to find extra normative resources to account for the epistemic.

The general lesson to be learnt, I think, is that we need a model for epistemic normativity that, at the same time, circumvents voluntarism worries and is normatively thick enough to account for the intuitively non-conventional nature of the epistemic. Otherwise, if the model put forth is too normatively thin, the suspicion will be that it only circumvents voluntarist worries in virtue of its normative thinness – and thus that an extensionally more adequate, normatively thicker incarnation thereof will fail to do so.

I think that the view we are after should be naturalistic, exhibit prior normative plausibility, and be generalisable to other normative domains but also, at the same time, have enough ‘normative oomph’ to explain the intuitive categoricity of epistemic normative constraints. In the next sections, I will develop a view that purports to meet all of these desiderata: on this account, epistemic normativity is etiological functionalist normativity.

6.2 Functions and Norms

In traits, artefacts, and practices alike, functions generate norms.³ There is such a thing as a properly functioning heart, a properly functioning can opener, and a proper way to make coffee. If that is so, when we are interested in a particular type of norm governing a particular type of trait, it is helpful to first identify its function.⁴

On the etiological theory of functions,⁵ functions turn on histories that explain why the item exists or operates in the way it does. Take my heart; plausibly, tokens of the type pumped blood in my ancestors. This was beneficial for my ancestors’ survival, which explains why tokens of the type ‘heart’ continue to exist. As a result, my heart acquired the etiological function (henceforth also e-function) of pumping blood. Acquiring an etiological function is a success story: traits, artefacts, and actions get etiological functions of a particular type by producing the relevant type of benefit. My heart acquired a biological etiological function by generating biological benefit. Through a positive-feedback mechanism – the heart pumped blood, which kept the organism alive, which, in turn, ensured the continuous existence of the heart – our hearts acquired the etiological function of pumping blood.

Importantly, while aetiology does require some history of beneficial effects, it does not require an awful lot of it; what it all amounts to is explaining the existence/continuous existence of a trait through a longer or shorter history of positive feedback:

Functions arise from consequence etiologies, etiologies that explain why something exists or continues to exist in terms of its consequences, because of a feedback mechanism that takes consequences as input and causes or sustains the item as output.

(Graham 2014, 35)

Functions can be of different sorts: there are biological functions, aesthetic functions, social functions, etc. In contrast to the Graham/Millikan view, my account takes functions to be typed by the corresponding benefit. As such, if a trait produces a benefit of type B in a system, the function thereby acquired will be a function of type B. The heart’s function to pump blood is a biological function in virtue of the fact that the produced benefit is also biological (i.e. survival). The function of art is an aesthetic function in virtue of the fact that the produced benefit is an aesthetic benefit. Now, of course, aesthetic benefit might, and often will, also result in biological benefit. This, however, in no way renders the function at stake a biological function. What is important to keep in mind is that the benefit that is essential to aesthetic function acquisition is the aesthetic one. The fact that biological benefit is also associated with the latter is a mere contingent matter of fact. Here is, then, the full etiological account to be employed here:

• E-function: A token of type T has the e-function of type B of producing effect E in system S iff (1) tokens of T produced E in the past, (2) producing E resulted in benefit of type B in S/S’s ancestors, and (3) producing E’s having B-benefitted S’s ancestors contributes to the explanation of why T exists in S.

Note that etiological functions are successes. They explain the continuous existence of the trait that bears them because that is so. The etiological economic function of knife-producing economic systems is not just that of producing knives; it’s to produce good, sharp knives. To see this, note that the positive-feedback loop that is presupposed by etiological function acquisition – the trait produces the effect and the effect benefits the system and thereby contributes to the explanation of the continuous existence of the trait – presupposes a history of success. What contributes to the explanation of the continuous existence of knife-producing economic systems is their producing good, sharp knives. Blunt ones would plausibly not have done the trick.

Just as the economic function of knife-producing economic systems is to produce good, sharp knives, the epistemic function of our belief-forming systems is to produce good beliefs. Mere belief, then, is a failure on the part of our cognitive system to fulfil its epistemic function just as blunt knives are failures on the part of knife producers to fulfil their economic function.

Note that functions will also come with associated norms: these prescribe the right way to proceed in order to reliably enough⁶ fulfil the function in question under normal conditions. Because its function contributes to the explanation of its very existence, the trait in question ought to perform in a way that is associated with likely enough function fulfilment. Your heart will be properly functioning when it functions in the way that reliably enough delivers function fulfilment under normal conditions: it will beat at a rate between 40 and 100 beats/minute, which, under normal conditions (i.e. conditions similar to those present at the moment of function acquisition), reliably leads to pumping blood in your circulatory system.

In my view, generating knowledge is the function of our epistemic practice of inquiry; in turn, cognitive processes, in virtue of their being the central mechanisms engaged in our practice of inquiry, inherit this function of generating knowledge. Further, norms governing moves in inquiry – such as beliefs, suspensions, withholdings, credences, assertions, or pieces of reasoning – will drop out of this knowledge-generating function.

Moves in practices generally aim – either directly or indirectly – to fulfil the function of the practice. The difference between direct and indirect aiming lies with achievability: for some moves in practices, the function of the practice is only indirectly rather than directly reachable. Cardiological consults are moves in the practice of medicine, and they aim directly at fulfilling the goal of the practice of medicine: curing diseases. In turn, some moves aim at this final practice goal only indirectly while aiming directly at intermediate goals: performing electrocardiogram (ECG) tests aims (directly) at informing the cardiologist as to how well the patient is doing, which, in turn, aims directly at their diagnosing the patient correctly and, further, at curing their disease. In this, ECG tests aim indirectly at the function of the practice of medicine: curing diseases. They aim at making progress towards it. Similarly, baking cakes is a move in the practice of cooking, and it aims directly at fulfilling the function of the practice: producing tasty, nourishing food. My getting the flour off the shelf aims indirectly at the general function of the practice by aiming directly at adding flour to the cake mix. It aims at making progress towards producing tasty, nourishing food. And so on.

Similarly, on the epistemic normative picture I favour, moves in the practice of inquiry – that is, all epistemically significant states and actions – will be governed by norms borne out by this central knowledge function of the practice. Moves in inquiry, that is, will aim either directly (plausibly: beliefs, assertions, reasonings) or indirectly (credences, suspensions, withholdings) at the aim of the practice of inquiry. The difference, once more, will lie with goal achievability: beliefs, assertions, and conclusions of reasonings can be knowledgeable in a way in which things like credences, suspensions, and withholdings cannot. Belief formation aims directly at the aim of the practice of inquiry (knowledge), whereas, at the same time, credence, withholding, and suspension aim at knowledge indirectly: they are transitional attitudes, in the sense in which these are attitudes held en route to knowledge but are not in the running for knowledge.

Just like the biological functions of our hearts generate etiological biological norms, the epistemic functions of our cognitive systems generate etiological epistemic norms governing our belief formation.

In previous work (e.g. Simion 2019a, 2021a), I have argued extensively that knowledge is the etiological function of our cognitive systems. This makes my epistemological approach, in a crucial sense, a knowledge-first epistemological approach: epistemic normative constraints, on this view, drop out of the knowledge function of our cognitive systems.

A few things by way of clarification are in order: in the more than twenty years since Timothy Williamson’s (2000) seminal work putting forth the first defence of a full knowledge-first epistemological framework, the knowledge-first research programme has generated an impressive amount of high-quality work and very promising results across epistemological subfields (e.g. epistemic justification, defeat, evidence, epistemic normativity, social epistemology, know-how, the nature and normativity of inquiry) and also at the intersection of epistemology with philosophy of language (e.g. the nature of speech acts, contextualism), mind (e.g. the nature of mental states), and moral and political philosophy (e.g. blame, trust, responsibility, political discourse).

Against this backdrop, however, this book zooms in only on what I take to be the core normative claims of the knowledge-first programme: that knowledge is the central epistemic value, and that thereby central normative notions in epistemology – such as justification, evidence, and defeat – are to be analysed in terms of knowledge. This view, in contrast to its extant knowledge-first competition, analyses epistemic normative categories in terms of knowledge without requiring further theoretical commitments (e.g. to the non-analysability of knowledge or to knowledge being a mental state in its own right). These questions remain open.

The key claim that I endorse is that generating knowledge is the function of our cognitive systems. To see why this is plausible, very briefly, note that knowledge meets the conditions for an e-function: it is plausible that it has been generated by our cognitive systems and those of our ancestors in the past (after all, knowledge is readily available in our environment), that this benefitted our organisms (e.g. by informing us about the presence of predators and the whereabouts of food), and that the fact that knowledge benefitted us in this way contributes to the explanation of why cognitive systems continue to exist in individuals like us. So generating knowledge is at the very least one function of our cognitive processes. Is it the main function? Note that what determines the proper level of generality for main function individuation is the T-value of the relevant T-function. The main biological function of the heart, for instance, maps onto its most valuable biological contribution: its main function is not ‘pumping blood and making a ticking sound’, but neither is it merely ‘pumping’, for instance. Plausibly, that is because if the heart pumps orange juice in our circulatory system, that’s not very valuable for our survival. I submit that knowledge is more valuable than any lesser epistemic standing⁷: that much is very widely accepted in the literature. It is easy to see that, if I am right about main function individuation, the distinctive value of knowledge thesis together with e-function deliver the result that the main epistemic function of our cognitive processes is generating knowledge.

Functions provide us with a straightforward way to identify the norms governing a particular trait or activity: first, if we are interested in identifying a norm of type T of the relevant trait, we start of by identifying its type T function. Once that is done, we look at the way in which it is reliably fulfilled under normal conditions. That will give us its conditions for proper function and, correspondingly, the content of the norm of type T we were looking for (Simion 2019a).

In what follows, I employ the functionalist machinery in investigating the epistemic normativity of belief. Before we move on, however, let us go back to the strength dilemma generated by voluntarist worry and the question of ‘normative oomph’: the good news is that, clearly, norms of proper function do not imply any voluntarist claim: just like my heart is governed by norms of proper function about blood pumping, my cognitive capacities are governed by norms of proper function about belief forming.

Is functionalist normativity going to generate a normatively thick enough model for epistemic normativity? Let’s go back to norms and practices in the domain of mafia. Here is a worry one might have: mafia practices may well continue to exist because they achieve the corresponding ‘values’ internal to the domain of mafia. If so, on a functionalist picture, we can get norms out of these functions: norms that regulate proper ways of going about achieving the domain-specific values that the domain of mafia is organised around. But do the resultant ‘norms’ have any normative oomph at all?

One would think that the case of epistemic norms is different. For instance, the fact that S said that p and that S is very reliable on p-related matters seems like a reason to believe that p has normative force – I can’t just set it aside in the way I can set aside the orders from my crime boss (e.g. Fricker 2007). These cases case feel different in a way that needs to be explained.

On functionalism, indeed, it can be the case that, for example, efficient killing is a domain-specific value in the domain of mafia, which, in turn, generates corresponding (domain-internal) functionalist norms. It’s a completely different question, however, if the domain itself is, for example, valuable simpliciter to begin with – and I take it that the domain of mafia is not. If so, the normative force of its norms will be restricted to the domain of mafia.

I take it to be empirically plausible that doing well epistemically is, at least for the most part, good for us biologically. If so, the domain of the epistemic, in contrast to the domain of mafia, will itself be valuable; if so, its internal functional norms will bear ‘normative oomph’ in a way in which norms of the domain of mafia do not. Importantly, though, I can afford to stay neutral on the extent to which the epistemic is good or bad: for ‘normative oomph’, I just need the fairly weak claim that it is generally good for us. This is important because it gives my view the flexibility to adapt to empirical results that purport to show that, at times, epistemic unreliability co-varies with biological benefits.

Compatibly, though, and plausibly, the domain of epistemology is valuable for our survival in a way that generates thick normative constraints. My functionalism thus does have the capacity to distinguish different kinds and strengths of normative force. On my view, epistemic norms have (1) domain-bound normative force, in that they promote knowledge, which is the value around which the domain is organised, and (2) non-domain-bound normative force, in that ‘the epistemic’ is a domain that is (empirically plausibly) valuable for our survival.⁸

6.3 Resistance to Evidence as Epistemic Malfunction

On my account, the main function of our belief-formation systems is to generate knowledge. In turn, they are properly functioning just in case they work in a way that is normally conducive to generating knowledge. When that happens, the beliefs they generate are justified.

I dub my view of justification ‘knowledge-first functionalism’: the account is functionalist in that it follows Millikan (1984), Burge (2010), and Graham (2012) in taking the epistemic normativity of belief to drop out of the epistemic function of our cognitive capacities. It is knowledge-first epistemological in that, unlike traditional, truth-first functionalism, it unpacks the function at stake in terms of knowledge. Here is a more precise formulation of the view:

• Knowledge-first functionalism (KFF): A belief is prima facie justified if and only if it is generated by a properly functioning cognitive capacity that has the etiological function of generating knowledge.

On this knowledge-centric picture, good belief is knowledgeable belief, while justified belief – belief that is permissible by the epistemic norm of belief – is belief generated by a properly functioning cognitive capacity that has the etiological function of generating knowledge. The standards for proper functioning are thus natural normative standards, and they are constitutively associated with promoting knowledgeable beliefs.

So far, we have been solely talking in moderate normative terms: we have taken justification of belief to supervene on epistemic permissibility and, in turn, epistemic permissibility to have to do with the proper function of our knowledge-generating belief-formation systems. How does resistance to evidence fit in this picture? After all, an epistemic subject being resistant to evidence seems to be a matter of obligation breach rather than permissibility breach: one is resistant to evidence insofar as there is evidence one should take up but one fails to do so.

I would like to propose an account of epistemic obligation according to which what all of subjects in Cases 1–7 from Chapter 1 have in common, epistemically, is that they are in breach of their epistemic ought to believe in virtue of resistance to available evidence. Here it is:

• The ought to believe (OTB): A subject S has an epistemic⁹ obligation to form a belief that p if there is sufficient and undefeated evidence for S supporting p.

Once more, importantly: when I talk of obligation, I am following the literature in deontic logic¹⁰ in using oughts and obligations interchangeably (i.e. I am employing a notion of obligation that merely maps onto an ought). Now, note that OTB, together with a moderate evidentialist assumption that one’s belief that p is epistemically justified only if there is sufficient and undefeated evidence for S supporting p, straightforwardly implies that epistemic justification is epistemic obligation (and, more generally, that justifiers are obligations). One might wonder, at this stage: is that right? After all, the vast majority of the literature assumes that epistemic justification is mere epistemic permission.

A few things about this: first, I take it that what resistance cases show is that this assumption was wrong all along. Mere epistemic permissions cannot, in virtue of their weak normative force, explain why the main characters in these cases (epistemically) ought to take up some evidence that they fail to take up.

Second, as I’m about to argue, there are in-principle theoretical reasons for which we should be suspicious of the thought that epistemic justification is mere epistemic permission. Here it goes: note that defeaters are obligations – when our justification is defeated, by definition, it is impermissible to ignore defeat and hold on to the corresponding doxastic attitude. Since it is impermissible to ignore defeaters, it follows that they constitute epistemic obligations (since it is always permissible to ignore mere permissions). Note, also, that there is such a thing as merely partial defeat: these are garden variety cases in which the epistemic agent needs to weigh their evidence in favour of p against their evidence against p. If Mary tells me that the train comes at 8 a.m. and you tell me that the train comes at 7 a.m., and Mary and you are, for all I know, equally reliable testifiers, my justification to believe that the train comes at 8 a.m. is partially defeated – I should lower my confidence in this being the case, but I don’t have sufficient epistemic support to move to outright believing that the train doesn’t come at 8 a.m. Similarly, I don’t have enough epistemic support to believe or disbelieve what you said either: it is impermissible both to form an outright belief that the train comes at 7 a.m. and to form an outright belief that it does not. Justifiers and defeaters can outweigh each other.

However, if defeaters constitute epistemic obligations, and if defeaters and justifiers can outweigh each other, it follows that justifiers constitute epistemic obligations as well: otherwise, if they constituted mere permissions, they would be normatively inert against defeaters, since permissions are normatively inert against obligations. As such, it seems as though the mere possibility of partial defeat implies that justification maps on to epistemic obligation. Here is the argument unpacked for the reader’s convenience:

1. (1) Defeaters affect what one is justified to believe.

2. (2) If defeaters affect what one is justified to believe, then it is epistemically impermissible to fail to adjust one’s doxastic attitudes in light of defeaters.

3. (3) It is epistemically impermissible to fail to adjust one’s doxastic attitudes in light of defeaters (from (1) and (2)).

4. (4) If it is epistemically impermissible to fail to adjust one’s doxastic attitude in light of a normative consideration x, then x constitutes an epistemic obligation.

5. (5) Defeaters constitute epistemic obligations (from (3) and (4)).

6. (6) Defeat can be partial defeat.

7. (7) If defeat can be partial defeat, then justifiers can carry normative weight against defeaters.

8. (8) Justifiers can carry normative weight against defeaters (from (6) and (7)).

9. (9) Justifiers constitute either epistemic permissions or epistemic obligations.

10. (10) Permissions cannot carry normative weight against obligations.

11. (11) Epistemic justifiers constitute obligations (from (5), (8), (9), and (10)).

What is the source of epistemic obligation? What grounds the epistemic OTB, in my view, is proper epistemic functioning. Pieces of evidence are pro tanto, prima facie justification-makers: they are the proper inputs to our processes of belief formation, and when we have enough evidence and the processes in question are otherwise properly functioning, the resulting belief is epistemically justified. In turn, when our belief-formation processes either fail to take up justifiers that they could have easily taken up or they take them up but fail to output the corresponding belief, they are malfunctioning:

• Resistance to evidence as epistemic malfunction (REEM): A subject S’s belief-formation capacity C is malfunctioning epistemically if there is sufficient evidence supporting p that is easily available to be taken up via C and C fails to output a belief that p.

The proper function of belief-formation capacities, then, on my view, is input dependent: failing to take up the right inputs – whether this occurs by taking up the wrong inputs or by failing to take up the right inputs – is an instance of malfunctioning.

One illuminating analogy here is the proper functioning of the lungs: as opposed to functional traits whose proper function is not input dependent (e.g. hearts), what it is for our lungs to function properly is, partly, for them to take up the right amount of the right stuff (i.e. oxygen) from the environment. Lungs that fail to do so, in environments where oxygen is easily available, are improperly functioning – whether they fail via taking up carbon dioxide or by just failing to take up easily available oxygen.

In contrast, hearts can take up and circulate whatever liquid they are fed, while, at the same time, remaining properly functional. Take your heart and place it in a vat with orange juice: insofar as it continues to pump at the normal rate, your heart is normally functioning – in spite of the fact that now it’s pumping orange juice rather than blood. The proper functioning of the heart is not input dependent.

Our cognitive systems do not work like hearts, they work like our respiratory systems; inputs matter for proper function. Our belief-formation capacities can’t take up wishes and form beliefs based on them whilst at the same time remaining properly functional. A cognitive system that processes wishes into beliefs is malfunctioning. If this is so, it follows that the proper functioning of our cognitive systems is input dependent.

Similarly, then, just like in the case of our respiratory systems, we should expect our cognitive systems and belief-formation capacities to malfunction in at least two input-dependent ways: via taking up the wrong kind of inputs (e.g. wishes), but also, and crucially for my purposes here, via failing to take up easily available good inputs (i.e. easily available evidence).

It is important to note that empirical results also overwhelmingly confirm the hypothesis that the proper functioning of our cognitive systems is input dependent (i.e. that our cognitive systems are malfunctioning if they fail to respond to environmental stimuli). Our levels of neuroplasticity¹¹ – the brain’s disposition for neuron-level changes in response to the environment – predict the brain’s capacity to take up information from the environment: when a cognitive system displays abnormally low levels of structural neuroplasticity,¹² learning in response to novel stimuli from the environment fails to occur at a normal rate. In turn, abnormally low levels of neuroplasticity, generating low responses to environmental stimuli, predict improper cognitive functioning. But if this is so, the proper functioning of our cognitive systems is input dependent: one way in which they can malfunction is by failing to respond to easily available environmental stimuli.

We now have a straightforward explanation of what goes wrong in cases of resistance to evidence: it is an epistemic incarnation of input-level malfunction. Our cognitive systems fail to take up easily available evidence from the environment. Just like respiratory failure is an instance of our respiratory systems failing to take up normal levels of oxygen from the environment, resistance to evidence is an instance of our cognitive systems failing to respond normally to stimuli from the environment.

6.4 Conclusion

The function of our cognitive systems is to generate knowledge, and justification turns on this function: we are justified to believe just in case our cognitive systems work in the way in which they generate knowledge reliably under normal conditions. This is not the full story, though: the proper functioning of our cognitive systems is not limited to the fair treatment of evidence that we pick up from the world – it extends to picking up the relevant evidence when easily available. When our cognitive systems fail to do so, they are malfunctioning and are in breach of epistemic norms.

In the next chapter, I further unpack REEM. I will not take a stance on what the sufficiency threshold stands for. Views will differ on this, and they will also differ on what fixes the threshold in question – whether it’s a purely epistemic affair or whether practical and moral considerations may play a role.¹³ My focus from here onwards will be on how to understand evidence, defeat, and permissible suspension in order to make good on REEM and, in turn, on OTB and the resistance intuition.

Chapter 7 Evidence as Knowledge Indicators

This chapter puts forth a novel view of evidence in terms of knowledge indicators, and it shows that it is superior to its competition in that it can account for the epistemic impermissibility of resistance cases, as well as for the effect that resistance to evidence has on doxastic justification. Very roughly, knowledge indicators are facts that enhance closeness to knowledge: a fact e is evidence for S that p is the case if and only if S is in a position to know e and e increases the evidential probability that p for S.

7.1 Knowledge Indicators

In the previous chapter, I have argued that evidence resistance is an instance of input-level epistemic malfunctioning of our cognitive systems. Input-level malfunctioning is a common phenomenon in traits the proper functioning of which is input dependent, such as our respiratory systems. Since our cognitive systems, I have argued, are systems the proper functioning of which is input dependent, we should expect the failure at stake in resistance cases. I have also argued that, since pieces of evidence are pro tanto, prima facie justification-makers, they are the proper inputs to our processes of belief formation. When we have enough evidence and our belief-formation cognitive capacities are otherwise properly functioning, the resulting belief is epistemically justified. In turn, when our belief-formation capacities either fail to take up justification-makers that they could have easily taken up or they take them up but fail to output the relevant belief, they are malfunctioning.

The question that this chapter purports to answer is: how should we understand one’s evidence such that we predict its normative impact on our properly functioning cognitive systems? Or, in other words, how should we understand one’s evidence such that our account thereof predicts the epistemic impermissibility of resistance to evidence for cognitive systems that have generating knowledge as their epistemic function?

To lay my cards right on the table, the answer I will offer will make use of the notion of a knowledge indicator: on my view, evidence consists of knowledge indicators, which enhance closeness to knowledge by enhancing evidential probability. In turn, for any system S with a function F, since S ought to fulfil F, it is plausible that S ought to enhance closeness to F fulfilment. If so, our cognitive systems should take up pieces of evidence because they enhance closeness to function fulfilment (i.e. they enhance evidential probability and thereby closeness to knowledge).

Here is, in more detail, how I think about these things: evidence consists of facts. They can be facts about the world around us or mere facts about a subject’s psychology. My having a perception as of a table in front of me is a psychological fact; it (pro tanto, prima facie) supports the belief that there is a table in front of me. So does the fact that there is a table in plain view in front of me.

In my view, evidence consists of facts that are knowledge indicators: facts that one is in a position to know and that increase one’s evidential probability (i.e. the probability on one’s total body of evidence) of p being the case. The fact that I see that there is a table in front of me is a piece of evidence for me that there is a table in front of me. It is a knowledge indicator, in that it raises the probability on my evidence that there is a table in front of me, and I’m in a position to know it.

Not just any psychological facts will constitute evidence that there is a table in front of me: my having a perception as of a table will fit the bill in virtue of having the relevant indicator property. Perceptions are knowledge indicators; the fact that I have a perception as of p is a fact that I am in a position to know, and that increases my evidential probability that p is the case. The fact that I wish that there was a table in front of me will not fit the bill, even if, unbeknownst to me, my table wishes are strongly correlated with the presence of tables: wishes are not knowledge indicators, for they don’t raise my evidential probability of p being the case (although they may, of course, raise the objective probability thereof). For the same reason, mere beliefs, as opposed to justified and knowledgeable beliefs, will not be evidence material; they lack the relevant indicator property.

Here is the view in full:

• Evidence as knowledge indicators: A fact e is evidence for one for a proposition p if and only if one is in a position to know e and one’s evidential probability that p is the case conditional on e is higher than one’s unconditional evidential probability that p is the case.

Or, slightly more formally, and where P stands for the probability on one’s total body of evidence:

• Evidence as knowledge indicators: A fact e is evidence for p for S iff S is in a position to know e and P(p/e) > P(p).

Let’s unpack the view further. What is it for me to be in a position to know e? Plausibly, a certain availability relation needs to be instantiated. On my view, availability has little to do with the limits of my skull. Evidence may consist of facts ‘in the head’ or facts in the world. Some facts – whether they are in the head or in the world, it does not matter – are available to me; they are, as it were, ‘at hand’ in my (internal or external) epistemic environment. Some – whether in the head (think of justified implicit beliefs, for instance) or in the world, it does not matter – are not thus available to me.

Here are, for starters, some paradigmatic cases that illustrate what I’m talking about: if there is a table right in front of me, but I’m not paying attention to it, there is evidence for me that there is a table in front of me. If, unbeknownst to me, you put a new table in the other room, the fact that you put it there is not available to me: it is not evidence for me. Similarly, if I have some mental state that is so deeply buried in my psychology that I can’t access it, it is not evidence for me.

As a first approximation, my notion of availability will track a ‘can’ for an average cogniser of the sort exemplified (e.g. with the relevant kind of cognitive architecture, social and physical limitations, etc.).

Here is some theory about this. First, there are qualitative limitations on availability: we are cognitively limited creatures. There are of types information that we just cannot access or process: the fact that there is a table in front of me is something that I can easily enough access. Your secret decision to put the table in the other room is not something I can easily access. There are also types of support relations that we cannot process: the fact that your car is in the driveway is evidence for me that you’re home. But it’s not evidence for my three-year-old son, Max, to believe that you’re home. Max belongs to a variety of epistemic agents that are not sophisticated enough to process¹ the support relation into a belief that you are home. Evidence is not available to you if the kind of epistemic agent that you are cannot access or process the particular variety thereof at stake (henceforth also qualitative availability).

There are also quantitative limitations on my information accessing and processing. The fact that there’s a table somewhere towards the periphery of my visual field – in contrast of it being right in front of me, in plain view – is not something I can easily process: I lack the power to process everything in my visual field: it is too much information (henceforth also quantitative availability). My cognitive limitations make it such that the facts available to me are only a subset of what is going on in my visual field. More on this later.

The ‘can’ at stake here will be further restricted by features of the social and physical environment: we are supposed to read the newspaper on the table in front of us, but not the letter under the doormat. That’s because we can’t read everything, and our social environment is such that written testimony is more likely to be present in the newspaper on the table than under the doormat (henceforth also environmental availability).

In sum, for a fact to be such that I am in a position to know it, it needs to be at hand for me in my epistemic environment: at hand qualitatively (it needs to be the type of thing a creature like me can access and process), quantitatively (it needs to belong to the quantitatively limited subset of facts that a creature like me can access and process at one particular time), and environmentally (it needs to be easily available in my – internal or external – epistemic environment; i.e. in my mind or in my physical and social surroundings).

I take this availability relation to have to do with a fact being within the easy reach of my knowledge-generating cognitive capacities. A fact e being such that I am in a position to know it has to do with my having a properly functioning knowledge-generating cognitive capacity that can take up e:

• Being in a position to know (BPK): S is in a position to know a fact e if S has a cognitive capacity with the function of generating knowledge that can (qualitatively, quantitatively, and environmentally) easily uptake e in cognisers of S’s type.

A few crucial clarifications about this account: first, note that BPK is a sufficiency claim. It is not necessary that e is available to me in order for me to be in a position to know e: I can also come to know e via taking up facts that increase my probability for e.

Second, note that BPK is a restricted ought-implies-can: agent obligations imply capacities in the kind of cogniser that they are. This opens the account to a mild generality problem, of course: how to individuate the relevant type of cogniser? Stable, constitutive features will matter: cognitive architecture, inherent social and physical limitations. Fleeting, contingent features will not (i.e. mere cognitive ‘furniture’): biases, previously held beliefs, wishes, among others. The advantage of the view is that, in restricting ‘ought implies can’ to types of cognisers, the account will predict that biased cognisers are in breach of their epistemic obligations: they may be unable to, for example, believe women because of bias, but cognisers with their cognitive architecture can, and therefore they should too.

Third, it is important to distinguish between being in a position to know and being in a position to come to know²: I am in a position to know that there is a computer in front of me; I am not in a position to know what is happening in the other room. I am, however, in a position to come to know the latter. Roughly, then, the distinction will, once more, have to do with epistemic availability: if all that needs to happen for me to come to know e is that my relevant cognitive capacities take up e and process it accordingly, then I am in a position to know e. If more needs to be the case – I need to open my eyes, or turn around, or go to the other room, or give you a call – I am in a position to come to know e but not in a position to know it. For now, I have not made any claim about the epistemic import of being in a position to come to know. Compatibly, being in a position to come to know might also, in some cases, deliver epistemic oughts: some cases of normative defeat and failure of evidence gathering are cases in point (e.g. see Lackey 2008, Goldberg 2016, 2017) See the next chapter for a discussion of this phenomenon.

Finally, and crucially, note that quantitative limitations on being in a position to know will make it so that I can only take up a limited number of the e₁, e₂, e3 … e_n facts that lie within reach with my knowledge-generating capacities. What facts go in my body of evidence in these cases? Which are the ones I am in a position to know, and which are the ones I am merely in a position to come to know (by changing focus, etc.)? On the account defended, in these cases, I will shoulder an epistemic obligation to take up a subset of e₁, e₂, e₃ … e_n that is as large as my quantitative take-up limitations. Therefore, my body of evidence will only include the relevant subset that a creature with my cognitive architecture can (quantitatively) take up at one time. When looking straight at my computer, my visual field is populated with very numerous facts, such that taking them all up exceeds my quantitative take-up limitations. I am only under an obligation to take up a quantitatively manageable subset of facts.

The crucial question that arises is: which is the set that takes normative primacy and thereby delivers my body of evidence? Availability rankings will deliver the relevant set, on my view: the most easily available subset of facts that I can take up delivers the set of evidence I have. In the case of visual perception, for instance, these are the facts located right in front of me, in the centre of my visual field, which are the brightest, clearest, etc. – in general, those facts that are most easily available to the cognitive capacities of a creature like me.

Tim Williamson (in conversation) worries that there will be cases in which too many facts (too many for my quantitative limitations) will have the same availability ranking. I see the worry (although I suspect it can be alleviated for most cases by our relation to space, time, complexity, brightness, etc.). Maybe the easiest case to imagine along these lines is the case of very simple arithmetical truths. In these cases, other normative constraints will have to decide the relevant set: I will have an all-things-considered obligation to attend to a particular range of simple arithmetical truths, and, among these, the most easily available will constitute my evidence, in virtue of them delivering the corresponding epistemic obligation to take them up.³

With the account fully unpacked, let’s move on to checking how it fares on accommodating the resistance data.

7.2 Evidence and the Impermissibility of Resistance

Here are, first and foremost, a few theoretical virtues of this view of evidence. First, it is naturalistically friendly, in that it situates the epistemic normativity of epistemic oughts to believe within an etiological functionalist picture of normativity: epistemic oughts to believe have to do with the proper function of our cognitive capacities, just like biological oughts to take up oxygen have to do with the proper function of our respiratory systems.

Second, the view enjoys high extensional adequacy. In line with intuition, it predicts that there is evidence for the Gettierised victim that there is a sheep in the field: the fact that they have a perception as of a sheep is a fact that they are in a position to know and that raises their evidential probability that there is a sheep in the field.

Also, there is evidence for the (recently envatted)⁴ brain in the vat (BIV) for p: ‘there is a tree in front of me’ when they have a perceptual experience as of a tree, since that is a fact that they are in a position to know and that raises their evidential probability that there is a tree in front of them.

There is no evidence for Norman the clairvoyant that the President is in New York: clairvoyant experiences are not evidential probability raisers when one is ignorant of the reliability of clairvoyance.

Finally, and most importantly for our purposes, it is easy to see that, when plugged into REEM, this view of evidence delivers the straightforward resistance intuition and thus explains that subjects in Cases 1–7 from Chapter 1 are in breach of their obligation to believe for failing to take up available evidence. Recall REEM:

• Resistance to evidence as epistemic malfunction (REEM): A subject S’s belief-formation capacity C is malfunctioning epistemically if there is sufficient evidence supporting p that is easily available to be taken up via C and C fails to output a belief that p.

Anna’s testimony in Case 1; media testimony, Dump’s statements, etc., in Case 2; the scientific testimony in Case 3; the perceptual experience as of a table in Case 4; the partner’s behavioural changes in Case 5; the fact that the Black students raise their hands in Case 6; and the incriminating fingerprints, etc., in Case 7 all constitute facts that are indicators of knowledge in virtue of being evidential probability enhancers that the subjects in these cases are in a position to know. These indicators of knowledge are easily available to creatures such as our protagonists: the subjects in Case 1–7 are members of a type of cogniser that hosts cognitive capacities with the function of generating knowledge that can easily take up these facts. Since they fail to do so, their cognitive capacities are malfunctioning, just like their lungs would be were they to be disinclined to take up the right amount of easily available oxygen. The account predicts that these subjects are all exhibiting resistance to evidence (by REEM) and are in breach of their obligation to believe (by OTB).

To see just how efficacious a view like mine is in accounting for evidence resistance and obligations to update, it will be useful to compare my account to E = K once more. In Knowledge and Its Limits, Williamson considers an account of evidence in terms of being in a position to know, and he dismisses it based on the following rationale:

[…] suppose that I am in a position to know any one of the propositions p₁, …, p_n without being in a position to know all of them; there is a limit to how many things I can attend to at once. Suppose that in fact I know p₁ and do not know p₂, …, p_n. According to E = K, my evidence includes only p₁; according to the critic, it includes p₁, …, p_n. Let q be a proposition which is highly probable given p ₁, …, p_n together, but highly improbable given any proper subset of them; the rest of my evidence is irrelevant to q. According to E = K, q is highly improbable on my evidence. According to the critic, q is highly probable on my evidence. E = K gives the more plausible verdict, because the high probability of q depends on an evidence set to which as a whole I have no access.

(Williamson 2000, 189)

Two things about this: first, note that, in virtue of the quantitative limitations that my account imposes on being in a position to know, the view does not suffer from the problem Williamson points to here. Indeed, given that there is a limit to how many things I can attend to at once, it is only the most available subset that I can attend to that is part of my body of evidence.

Even more importantly, I submit that once we put flesh on the bones of Williamson’s case, my view, and not E = K, gives the intuitively right prediction. Here it goes:

FRIENDLY DETECTIVE 2: It’s highly probable that John killed the victim given that (p₁) John is a butler, (p₂) John is a very nice guy with an impeccable record, and (p₃) the only butler who’s a very nice guy with an impeccable record was seen stabbing the victim. Friendly Detective is told p₁, p₂, and p₃ but can’t get himself to believe p₃ because of wishful thinking, and he believes John didn’t do it based on p₁ and p₂.

FRIENDLY DETECTIVE 2 is an instance of Williamson’s case. It is easy to see, however, that it is E = K that delivers the counterintuitive result here: according to E = K, the detective is justified to believe John didn’t do it. My view disagrees, and it scores on extensional adequacy.

Going back to the high societal stakes of evidence resistance: crucially, real-world, high-stakes cases of climate change denial and vaccine scepticism will sometimes be diagnosed by this account of evidence as evidence resistance. This will happen in cases of cognisers who have easily available evidence that climate change is happening and that vaccines are safe but fail to take it up and update their beliefs accordingly. It is compatible with this account, however, that this is not always the case: not all evidence rejection is evidence resistance. Sometimes, cognisers inhabit an epistemic environment heavily polluted with misleading evidence against the reliability of scientific testimony and public policy: if reliable testifiers in one’s community testify that not-p: ‘climate change is not happening’, and one has every reason to trust them (say, because they have an exceptional track record of reliability as testifiers – although they get it wrong on this particular occasion), it can happen that one justifiably rejects evidence for p due to being in a position to know ‘heavier’ (albeit misleading) evidence against p. Note, however, that these cases – cases of justified evidence rejection – will be fairly specific cases epistemically that, while they may happen in fairly isolated communities, the more one has access to evidence for p, the less justified their evidence rejection will be.

Now, all of this tells us that the account put forth is extensionally adequate: the view gets the resistance cases right. That is an important theoretical virtue of the view, and, as we have seen, it singles it out in the epistemic normative landscape.

That being said, extensional adequacy is not explanatory adequacy: even if thinking of evidence in terms of evidential probability increasers that one is in a position to know delivers the result that there is evidence for the subjects in Cases 1–7 that they fail to take up, the question as to why they should have done so remains open. One task remains, then, for the theorist of evidence resistance: explaining the normative force exercised by available evidence on our properly functioning cognitive systems. Or, in other words, explaining why, given the account of evidence proposed, it is epistemically impermissible for cognitive systems that have generating knowledge as their epistemic function not to take up easily available evidence.

Here it goes: some evidence I take up with my belief-formation machinery, whereas some I fail to take up, although I should. What grounds this ‘should’, in my view, is proper epistemic functioning.⁵ Because they are knowledge indicators, pieces of evidence are justification-makers: they are the proper inputs to our processes of belief formation that have generating knowledge as their function, and when we have enough thereof, and the processes in question are properly functioning in all other ways, the resulting belief is epistemically justified

Since evidence for S that p, on my account, consists of facts that enhance closeness to knowledge that p for S by enhancing S’s evidential probability for p, our cognitive systems are malfunctioning if they fail to take up easily available evidence, in virtue of thereby failing to take up opportunities for enhancing closeness to knowledge. Since for any system S with a function F, S should fulfil F, and it is plausible that S should enhance closeness to F fulfilment, and since the function of our cognitive systems is to generate knowledge, our cognitive system should take up enhancers of closeness to knowledge. Our cognitive systems should take up pieces of evidence because they enhance closeness to function fulfilment (i.e. they enhance evidential probability and thereby closeness to knowledge).

In turn, when our belief-formation capacities either fail to take up knowledge indicators that they could have easily taken up or they take them up but fail to output the corresponding belief, they are malfunctioning. A subject S’s belief-formation capacity C is malfunctioning epistemically if S has sufficient evidence supporting p that is available to be taken up via C and C fails to output a belief that p.

7.3 Infallibilism: Evidence and Knowledge

Before moving on, I would like to address an important worry that has been put forth in recent literature for views of evidence like the one defended in this chapter (i.e. knowledge-centric views of evidence).

Most contemporary epistemologists are fallibilists: they think that you can know a proposition p, even if your evidence does not entail that p. In recent work, Jessica Brown (2018) offers a thorough defence of fallibilism against knowledge-centric views of evidence, or what I will dub ‘new infallibilism’. More specifically, her central aim is to show that epistemologists who also want to be non-sceptics and want to endorse a non-shifty view of knowledge attributions should be fallibilists rather than new infallibilists. To this end, Brown argues that there is reason to think that fallibilism compares favourably with new infallibilism when it comes to evidence and evidential support. Perhaps most importantly, Brown identifies and takes issue with three key commitments of the new infallibilist’s view of evidence, to wit:

• The factivity of evidence: If p is part of one’s evidence, then p is true.

• The sufficiency of knowledge for evidence: If one knows that p, then p is part of one’s evidence.

• The sufficiency of knowledge for self-support: If one knows that p, then p is evidence for p.

Brown argues against all three of these claims. Since fallibilists can avoid these commitments, the thought goes, fallibilism scores points against new infallibilism.

The account of evidence I defended in this chapter implies all of the claims above. As such, if Brown is right, my account is in trouble, alongside its E = K Williamsonian cousin.

However, I think that there are ways to be an infallibilist that survive Brown’s excellent arguments. Thus, in what follows, I will explore ways in which new infallibilism can resist both Brown’s case against infallibilism and her fallibilist response to at least some of the data points that have been thought to favour the new infallibilism.

Let’s start by looking at Brown’s argument against the sufficiency of knowledge for evidence (i.e. the claim that if one knows that p, then p is part of one’s evidence). Brown’s key idea is to appeal to citable evidence. She points out that one cannot felicitously cite p when queried about one’s evidence for p, not even if one knows that p (Brown 2018, 49–50). But given that knowledge is sufficient for evidence, it is hard to see why this should be the case.

Note, however, that fallibilists, too, will need an account of when p is part of one’s evidence. I can think of a few options here: if p is justified for one/if one believes that p/if one justifiably believes that p, then p is part of one’s evidence. Crucially, since knowledge entails justified belief, their view entails the sufficiency of knowledge for evidence, no matter which of these options the fallibilist goes for. This means that in cases in which one knows that p, it is equally hard for fallibilists to explain why one cannot cite p when queried about one’s evidence for p. In this way, there is no reason to think that new infallibilism is at a disadvantage here.

Let’s move on to another of the claims above: the sufficiency of knowledge for self-support (i.e. that if one knows that p, then p is evidence for p). Why think that new infallibilists are committed to this claim in the first place? Here is Brown:

To see why the infallibilist should embrace the Sufficiency of knowledge for self-support, consider […] knowledge by testimony, inference to the best explanation and enumerative induction. It’s hard to see how one has evidence for what’s known in these ways which entails what’s known without allowing that if one knows that p, then p is part of one’s evidence for p. […] So, it seems that embracing the Sufficiency of knowledge for self-support is the best way for the infallibilist to avoid scepticism.

(Brown 2018, 43)

I agree that it may be hard for fallibilists to see how one can have the evidence for what is known here unless one subscribes to the sufficiency of knowledge for self-support. However, the same is not true of new infallibilists. Note that, according to new infallibilism, what one’s evidence is will turn on worldly states (e.g. on the friendliness of the epistemic environment one finds oneself in). For instance, what is one’s evidence for the claim that there is a barn before one may vary depending on whether one is in Normal Barn County or in Fake Barn County. But once this point is properly appreciated, there is little reason to think that testimony, inference to the best explanation, and enumerative induction pose a particularly difficult problem. While data from testimony, inference to the best explanation, and enumerative induction may not entail what is known, they may do so when conjoined with a sufficiently friendly epistemic environment.

This leaves the factivity of evidence (i.e. p is part of one’s evidence only if p is true). Brown relies on a familiar line of objection to this claim. Here is Brown:

As is well-known, this conception of evidence [which combines the factivity of evidence with the sufficiency of knowledge for evidence] is open to the objection that it holds that certain pairs of subjects who are intuitively equally justified in some claim (e.g. a person and her BIV twin), are not equally justified.

(Brown 2018, 22)

Brown considers a response on behalf of new infallibilists in terms of blamelessness.⁶ They key idea is that while BIVs don’t believe justifiably, they are nonetheless blameless for their beliefs. At the same time, there is empirical evidence that suggests that we are prone to mistaking cases of unjustified but blameless belief for cases of justified belief, which is why intuition leads us astray in these cases.

According to Brown, this move remains unsuccessful. Her strategy is to look at a number of ways of analysing what blamelessness amounts to and to argue that none of these ways will do the trick for new infallibilists.

Note, though, that while it is true that the particular infallibilists (e.g. Williamson, Littlejohn) that Brown discusses have historically held a view that equates justification and knowledge, this is optional to new infallibilisms. There has been a surge of views in the literature that explain justified belief in terms of knowledge without identifying justified belief and knowledge (e.g. Bird 2007, Ichikawa 2014, Miracchi 2015, Kelp 2018, Schellenberg 2018, Simion 2019a). Champions of these views have argued at great length that these views can allow for agents in bad cases (e.g. BIVs) to be justified. If so, they can successfully explain the intuition at issue here. Crucially, the view of justification defended here is precisely one such view: on this account, BIVs believe justifiably insofar as they employ properly functioning cognitive capacities with the function of generating knowledge – which, by stipulation in the justification-intuition-triggering cases (paradigmatically, of recently envatted BIVs) they do. At the same time, and crucially, this view of justification is entirely compatible with new infallibilism. After all, what is key to new infallibilism is a view about the relation between knowledge and one’s evidence.

7.4 Conclusion

On the account defended here, one’s evidence consists in facts that one is in a position to know and that increase one’s evidential probability that something is the case. In turn, being in a position to know has to do with the variety of cogniser at stake: should one be the kind of cogniser that hosts cognitive processes that are able to pick up the relevant facts from the world, the facts at stake will belong in one’s body of evidence.

Chapter 8 Defeaters as Ignorance Indicators

This chapter puts forth and defends a novel view of defeat, and it shows that it is superior to its competition in that it can account for the epistemic impermissibility of defeat resistance cases and normative defeat cases, as well as for the effect ignored defeat has on doxastic justification. On this account, defeaters are ignorance indicators¹: facts that one is in a position to know and that reduce one’s evidential probability that p. Furthermore, I also put forth a novel account of the normativity at work in cases of normative defeat and negligent inquiry and evidence gathering.

8.1 The Nature and Theoretical Importance of Defeat

The notion of defeat is central to epistemology, practical reasoning, and ethics. Within epistemology, it is standardly assumed that a subject who knows that p or justifiably believes that p can lose this knowledge or justified belief by acquiring a so-called defeater, whether this is evidence that not-p, evidence that the process which produced their belief is unreliable, or evidence that they have probably misevaluated their evidence. Within ethics and practical reasoning, it is widely accepted that a subject may initially have a reason to do something, although this reason is later defeated by their acquisition of further information.

Investigations into the nature and normativity of defeat come with high theoretical stakes. The notion of defeat has been central to a wide range of different philosophical debates, including, but not limited to:

1. (1) The nature of justification and knowledge: since knowledge and justification are taken by many to be defeasible, the extent to which one account or another of the nature of knowledge/justification is able to account for/accommodate defeat constitutes an important ground for assessing its theoretical credentials.

2. (2) Internalism versus externalism: several epistemologists worry that epistemic externalism has a hard time accommodating psychological defeat; at the same time, conversely, if justification supervenes on mental states alone, as per internalism, it seems mysterious that it could be defeated by normative defeaters lying outside of the cogniser’s ken.

3. (3) Epistemic norms and reasons: for accommodating the phenomenon of defeat, debates on epistemic norms and epistemic reasons owe us, at a minimum, an account of epistemic normative overriding, as well as an account of reasons against belief.

4. (4) Evidence and higher-order evidence: since evidence is widely taken to be defeasible, a plausible account of evidence should come with a corresponding plausible account of defeat. For instance, one important desideratum on any such account is that it explains the defeating power of higher-order evidence, namely of evidence that one’s first-order beliefs are the output of a flawed process.²

5. (5) Closure and transmission: one popular solution to alleged failures of closure principles for knowledge and transmission principles for warrant is known as ‘the defeat solution’. Roughly, the thought goes, closure and transmission hold prima facie, and the intuitions of failure are to be explained in terms of psychological defeat. This solution, of course, hangs on the assumption that there is such a thing to begin with (i.e. that psychological defeat is a genuine epistemic category).

6. (6) Disagreement: one way to characterise the debate between conciliatory and steadfast views of disagreement is as centred around the question: can the testimony of one’s peer carry defeating power? Steadfastism answers ‘no’, conciliationism answers ‘yes’. The correct account of the nature of defeat can help settle the issue.

7. (7) Reductionism versus anti-reductionism about testimony: say that a suspect for murder S tells you that she did not do it. According to both of the main views in the epistemology of testimony, you are not justified to believe S. According to reductionism, that’s because you always need positive, non-testimonial reasons to believe what you are being told. In contrast, according to anti-reductionism, you are prima facie justified to believe S, but your justification is defeated. The correct account of the nature of defeat will likely go a long way towards settling this issue.

Given these high theoretical stakes, investigations into the nature and normativity of defeat carry significant philosophical weight. Unfortunately, not many systematic, full accounts of defeat have been put forth on the market (although see Brown and Simion (2021) for the first full volume on defeat and Kelp (2023) for the first book-length treatment). In what follows, I will look at the classical accounts of defeat on the market – one internalist evidentialist, coming from John Pollock, and one externalist reliabilist, championed by Alvin Goldman – to help situate my account in the extant landscape.³

8.1.1 Traditional Evidentialism About Defeat

The first and what is now considered the classic view on the nature of defeat in epistemology is due to Pollock (1986). According to this view, d is a defeater of e’s support for p for S if and only if (1) e is a reason to believe p for S and (2) e&d is not a reason to believe p for S (henceforth ‘Pollock’s view’).

The account has a lot going for it: it nicely promises to cut across normative domains in virtue of being framed in terms of reasons; after all, epistemologists hardly enjoy an exclusivity on reasons. With Pollock’s view in play, it is easy to see how we could generalise it to cover different targets (e.g. actions) and types (e.g. moral, prudential) of normativity. Second, Pollock’s view makes good on the intuitive thought that defeaters are actualisers of the possibility of a positive normative status to be overridden or undercut; what the view says, in a nutshell, is that defeaters are the kind of things that render a permissible belief impermissible.

For our purposes here, there are two important limitations to Pollock’s view: first, it does not account for partial defeat – nor is it trivial to see how it could be extended to do so. Since many cases of defeat resistance will be cases of partial defeat, the view will not deliver the needed theoretical resources for the data I am trying to explain. Second, the account remains silent on the nature of reasons and, most importantly, on what it is for something to be a reason for S to believe: however, in order to understand the impermissibility of defeat resistance, it is crucial to understand both of these things. We need to know what reasons are and which reasons are reasons for S to believe, since this is essential to understanding the impermissibility of S’s resistance to defeat.

One way to spell out Pollock’s view that suggests itself, given his evidentialist leanings, is a traditional, seemings-based recipe: on this account, reasons for S to believe are S’s relevant seemings. Of course, as we have seen in Chapter 2, a view like this will get us into trouble with resistance cases rather rapidly: on the necessity direction, recall only the very sexist George, who zones out whenever a woman speaks to him. This guy doesn’t host any relevant seemings – intuitively, however, his beliefs are defeated by women’s testimony. Against the sufficiency direction, notably, cases of cognitive penetration will create trouble for a seemings-based defeat account (e.g. see Lyons 2011): the fact that it seems to me – due to sexist bias – that women don’t know what they’re talking about is not enough to defeat my justification to believe their testimony.

8.1.2 Defeaters as Reliable Processes

Reliabilist theories of justification have been extremely popular in the last three decades and come in a variety of forms, but the gist of the view is that a belief is justified if and only if it is formed via a (normally) reliable procedure or ability. Reliabilism is a theory of prima facie justification. As such, in line with normative theories in general, it needs a theory of defeat in order to hold water. The standard reliabilist account of defeat comes from Alvin Goldman:

The alternative reliable process account (ARP): S’s belief is defeated iff there are reliable (or conditionally reliable) belief-forming processes available to S such that, if S had used those processes in addition to the process actually used, S wouldn’t have held the belief in question.

(Goldman 1979)

One can see how ARP is an elegant reliabilist translation of the Pollockian thought that defeat is the kind of normative entity that, when taken in conjunction with the extant epistemic support for the relevant belief, fails to render it justified.

Bob Beddor (2015) is the locus classicus for criticism of ARP; if Beddor is right, ARP is both too weak and too strong. Against ARP’s sufficiency direction, Beddor offers the following case:

Thinking About Unger: Harry sees a tree in front of him at t. Consequently, he comes to believe the proposition TREE: ⟨There is a tree in front of me⟩ at t. Now, Harry happens to be very good at forming beliefs about what Peter Unger’s 1975 time-slice would advise one to believe in any situation. Call this cognitive process his Unger Predictor […]. What’s more, […] whenever it occurs to Harry that Unger would advise him (Harry) to suspend judgement about p, this causes Harry to […] suspend judgement about p. So if Harry had used his Unger Predictor, he would have come to […] suspend judgement regarding TREE.

(Beddor 2015, 152)

What this cases shows is that ARP is too weak normatively: contra ARP, for my belief that p to be defeated, it is not enough that I would change my mind about p in a counterfactual world due to employing some reliable process. Just because I would change my mind in world W, it does not follow that I should change my mind in world W: defeat is a normative notion.

More importantly for our purposes here, however, Beddor’s case against ARP’s necessity direction is, indeed, a paradigmatic case of evidence resistance. Here it is:

• Job Opening: Masha tells Clarence that her department will have a job opening in the fall. Clarence believes Masha; assuming that Masha is usually reliable, Clarence’s belief counts as prima facie justified. Sometime later, Clarence speaks with the head of Masha’s department, Victor, who informs him that the job search was cancelled due to budget constraints. Now suppose that Clarence harbours a deep-seated hatred of Victor that causes him to disbelieve everything that Victor says; what’s more, no amount of rational reflection would rid Clarence of this inveterate distrust. Consequently, he continues to believe that there will be a job opening in the autumn.

This case shows that ARP is also too strong: just because, in all counterfactual words, I would irrationally and stubbornly hold on to my belief, it does not follow that I should do so. Once again, ARP is not normative enough to do the job it is supposed to do. Going back to our purposes here, Job Opening is a paradigmatic case of testimonial evidence resistance: it shows that APR’s normative weakness results in difficulties for accounting for resistance to defeat.

Along similar lines, recall the case of Professor Racist:

• Case #6. Misdirected Attention: Professor Racist is teaching college-level maths. He believes people of colour are less intelligent than white people. As a result, whenever he asks a question, his attention automatically goes to the white students, such that he doesn’t even notice the Black students who raise their hands. As a result, he believes Black students are not very active in class.

To see the problem that this cases poses for ARP, let us again ramp up the epistemically problematic aspects of the case and stipulate that Professor Racist is not only racist but also dogmatic (indeed, this assumption should not be very hard to make, since these two epistemic vices tend to be encountered together): even if he had seen the Black students raising their hands, he would have still strongly believed that they’re not very active in class. If so, ARP predicts that there is no defeat at stake in this case: after all, there is now no alternative reliable process that is such that, had Professor Racist used it, he would have abandoned his belief that Black students are not very active in class. Not only is ARP then not predicting defeat in this case, but it predicts that Professor Racist is epistemically better off in this version of the case than in the original version. Since being racist and dogmatic is epistemically worse than being only racist, ARP remains unsatisfactory.⁴

Note, finally, that we can make the same stipulation of dogmatism in all of the other resistance cases as well, with the same unsatisfactory result. As such, process reliabilists still owe us an explanation of what is going wrong in cases of defeat resistance.

The account I will develop next, somewhat unsurprisingly, complements the account of evidence as knowledge indicators developed in the previous chapter. On the view I will defend, conversely, defeaters are ignorance indicators. It is interesting, I think, to consider the view against its historical evidentialist and reliabilist predecessors: like the evidentialist, my account takes evidence and defeaters to have normative strength independently of whether they are being taken up via particular types of processes or abilities. Like the reliabilist, however, the account bottoms out in processes and abilities: a fact only constitutes a defeater if one is in a position to take it up via one’s cognitive capacities.

8.2 Defeaters as Ignorance Indicators

In my view, defeaters are indicators of ignorance: they are facts that one is in a position to know and that lower one’s evidential probability that p is the case:

• Defeaters as ignorance indicators: A fact d is a defeater for S’s evidence e for p iff S is in an position to know d and S’s evidential probability that p conditional on e&d is lower than S’s evidential probability that p conditional on e.

Or, slightly more formally:

• Defeaters as ignorance indicators: A fact d is a defeater for S’s evidence e for p iff S is in a position to know d, and P(p/e&d) < P(p/e).

Recall also that, on my view, S is in a position to know a fact e if S has a cognitive capacity with the function of generating knowledge that can (qualitatively, quantitatively, and environmentally) easily uptake e in cognisers of S’s type. It is easy to see that the view of defeat defended here nicely predicts that the justification of some occurrent beliefs hosted by the characters in several of resistance cases is defeated by the presence of ignorance indicators. Take Bill, Dump’s supporter: since the information coming from several sources around him is such that he is in a position to know it, and it lowers the evidential probability that Dump is a good president, Bill should lower his confidence that Dump is a good president. Similarly, Mary should lower her confidence that her husband is just making friends, and Professor Racist should not believe, as he does, that Black students are not very active in class.

In my view, rebutting and undercutting defeaters share one and the same central epistemic normative property: they are evidential probability decreasers. What differs is the mechanism by which they achieve this effect: rebutters lower one’s evidential probability for p by raising one’s evidential probability for not-p. In contrast, undercutters reduce the degree of confirmation that a particular piece of evidence e confers on p (see also Kotzen (2019) for a detailed formal treatment along these lines). This comes in stark contrast to literature that gives different treatment to first- and higher-order evidence or rebutting and undercutting defeat. I think mine is the right result, and we should, all else equal, prefer this unified treatment on grounds having to do with theoretical adequacy. Here are some quick reasons why scepticism about the defeating power of higher-order evidence does not work: one’s evidence comes with a having relation and an evidential support relation (also known as degree of confirmation: how much a piece of evidence probabilifies p). Plausibly, one’s confidence in p should match the degree of confirmation that one’s evidence offers to p. Higher-order evidence/undercutting defeat works via raising/lowering the degree of confirmation that first-order evidence provides to p. Now, here is a case of higher-order evidence that increases the degree of confirmation of the first-order evidence: I believe p based on my neighbour George’s testimony (alternatively, I have, e.g., 0.8 credence that p). Mary tells me that George is the top expert in the world on the matter. Her testimony is evidence that q: ‘George’s testimony gives very high support to p’. In probabilistic terms: if George’s original testimony probabilifies p to x, Mary’s testimony translates roughly as ‘George’s testimony probabilifies p to y & y > x’ (how high y is will depend on Mary’s epistemic credentials). So now that Mary has spoken, I am in a position to know something along the lines of r: ‘I have a 0.8 credence that p based on George’s testimony and the probability that p conditional on George’s testimony is 0.9’. Intuitively, I should revise to 0.9.

Moving on to a corresponding case of undercutting defeat: I believe p (indeed, I know p) based on my neighbour George’s testimony (alternatively, I host, e.g., 0.8 credence that p). Mary tells me that George is a well-known liar on p-issues. Her testimony is evidence that q: ‘George’s testimony gives lower support to p’. George’s original testimony probabilifies p to x; Mary’s testimony, just like before, amounts to: ‘George’s testimony probabilifies p to y & y < x’ (it will depend on Mary’s epistemic credentials how low y is). Just like before, I am in a position to know that r: ‘I have a 0.8 credence that p based on George’s testimony and the probability that p conditional on George’s testimony is 0.4’. Intuitively, I should revise to 0.4. Scepticism about the defeating power of higher-order evidence is wrong, and defeat affords unified treatment.

Going back to our central cases of evidence resistance – again, crucially, real-world, high-stakes cases of climate change denial and vaccine scepticism will sometimes be diagnosed by this account of evidence and defeat as evidence resistance: this will happen in cases of cognisers who have easily available evidence that climate change is happening and that vaccines are safe but fail to take it up and update their beliefs accordingly.

As previously shown, however, it is compatible with this account, however, that this is not always the case: not all evidence rejection is evidence resistance. Sometimes, cognisers inhabit an epistemic environment heavily polluted with misleading defeat: if reliable testifiers in one’s community testify against p: ‘climate change is happening’, and one has every reason to trust them (say, because they have an exceptional track record of reliability as testifiers – although they get it wrong on this particular occasion), it can happen that one justifiably rejects evidence for p due to being in a position to know ‘heavier’ defeaters (i.e. evidential probability decreasers). Note, however, that these cases – cases of justified evidence rejection in virtue of misleading defeat – will be fairly specific cases epistemically: for example, cases in which the cogniser has more reliable (although misleading) testimony that not-p than evidence that p, or cases in which the cogniser has overwhelming undercutting defeaters (e.g. based on reliable although misleading testimony) for the source of p. While this may happen in fairly isolated communities, the more one has access to evidence for p, the less justified their evidence rejection will be.

Here is a question that arises for the account put forth: why think that there is such a thing as an ought governing our belief formation to take up defeat? After all, on the account proposed, defeaters are ignorance indicators (i.e. facts that lower evidential probability for one). Why should a system with the function of generating knowledge be under an obligation⁵ to take up ignorance indicators, which, by stipulation, decrease closeness to knowledge?

Two things about this: first, note that defeaters are not merely ignorance indicators for p, but also either knowledge indicators for not-p (rebutting defeaters) or knowledge indicators for ‘piece of evidence e does not confirm/offers weaker confirmation for p’ (undercutters). This, in turn, affords them the same normative explanation as that offered for garden variety evidence: defeaters exercise normative pressure on our cognitive systems in virtue of them being knowledge indicators.

Second, and furthermore, I think that there is more to the normative pressure of defeat than this merely general evidential normative pressure; in particular, I think that the normative pressure of defeaters for my belief that p can also be explained in a p-centric fashion. Here it goes: consider, for starters, a case in which I know that p at t₁, and (misleading) evidence that not-p becomes available to me at t₂. In a situation like this, crucially, whether I take up the defeating evidence or not, my knowledge that p is defeated (short of my knowledge constituting defeat defeat): I am now left, at best (depending on the relevant evidential weights), with a somewhat – but not fully – justified true belief that p. My full belief has thereby been rendered impermissible: I now hold a doxastic attitude that is stronger than what the evidence affords. Note, also, that this is so even if what I started with is not knowledge, but rather, for example, a justified credence of 0.6. Now that defeat is available in my environment, whether I take it up or not, my 0.6 credence is no longer permissible, since my level of justification has been lowered by defeat.⁶ What is it that my cognitive system ought to do now that it’s hosting an impermissible doxastic attitude? The answer that suggests itself is: abandon it, either altogether (if defeat is full defeat) or in favour of the weaker attitude that remains supported by the evidence.

Note, though, that abandoning the (now) impermissible doxastic attitude in conjunction with not taking up the relevant defeaters seems irrational. If I, at the same time, (mistakenly) continue to take my evidence to support a 0.6 credence and adjust to a 0.5 credence, something has gone amiss, rationality-wise. In sum, it would seem as though, if I don’t take up the relevant defeat from the environment, I am faced with a normative dilemma: either I hold on to an impermissible doxastic attitude that is no longer supported by my evidence or I hold a novel doxastic attitude that enjoys propositional justification but not doxastic justification, for lack of proper basing. Grasping neither of these horns is knowledge conducive. If so, it would seem, the only path left for knowledge conduciveness is taking up the relevant defeaters and adjusting my doxastic attitude in light of them.⁷

8.3 External Defeat and Norms of Evidence Gathering

Consider a doctor, X, who believes that p but missed a recent development in the field that (q =) the data that have been taken to support p don’t really do so. X does not justifiably believe that p: q is a defeater for this belief, which undermines X’s justification for believing p. Now contrast X with a layperson, Y, who had been told by their doctor that p and still believes that p. Despite the fact that there has been a recent development in the field of medicine, Y’s belief that p continues to be justified. In particular, q does not undermine Y’s justification for their belief that p. Since the central difference between X and Y is that X occupies a certain social role (i.e. they are a doctor), there is reason to think that social roles can be sources of defeat via giving subjects reasons to inquire (Goldberg 2018).

Note, first, that so far we have looked only at epistemic functions in individual agents. And while epistemic functions may arise in individual agents, they also arise in broader social systems. It is precisely this idea that will be of central importance in what follows.⁸

To begin with, I take social systems to be systems that feature multiple agents who are connected to one another in at least some ways. The social roles we are interested in are properties of agents in social systems. Being a doctor, teacher, parent, etc., is a property of an agent in a social system.⁹

One interesting feature of social roles is that many of them have constitutive functional properties in that what it is to be an X (doctor, teacher, baker, fireperson, etc.) is to have the particular function in question (to treat ill people, to teach people stuff, to make baked goods, to put out fires, etc.). To see this, note that we cannot even fully understand the roles in question without understanding their functions: to fully understand what a doctor is, you have to understand what the function of a doctor is (i.e. to treat ill people). It is easy enough to see that the same is true of a whole host of other social roles, including teachers, bakers, firepeople, among many others.

Given that many social roles have constitutive functional properties, the prospects for an analysis of the epistemic norms constituting these social roles already look bright. The question that we need to consider is: are the functions constitutive of these social roles generating any constitutive epistemic norms of proper functioning? If the answer to this question is yes, then the route to an account of epistemic norms constitutive of social roles is a going to be a short one. By the same token, an account of how social roles may lead to defeat may come into view.

To get an idea of how this might be, let’s take another look at the case of the doctor. Note that having an up-to-date understanding of their field is part of the normal functioning of doctors in the social system that we occupy. More specifically, it is part of such normal functioning that doctors engage in inquiries into recent developments in the field, as a result of which they maintain an up-to-date understanding of the field and thereby know how to treat people. In fact, that doctors maintain an up-to-date understanding is a key element in the feedback loop that explains the continued existence of this important social role in the social system that we occupy¹⁰: doctors’ understanding informs their treatment practices, and the fact that it is kept up to date enhances their success rate of these treatments, which in turn explains why the social role of doctor continues to exist in our social system. But since maintaining an up-to-date understanding of the field is part of the normal functioning of doctors in our social system, we get the by now familiar normative import. It is thereby part of such proper functioning. This, in turn, means that we get a norm that doctors violate if they fail to maintain an up-to-date understanding of their field and, by the same token, if they fail to engage in the inquiries needed to do so. And, of course, it is easy enough to see that the same holds, mutatis mutandis, for many other social roles, including teachers, lawyers, academics, and so on. In turn, since the norms in question are generated by the constitutive functions of these social roles, they will be constitutive norms.

It turns out, then, that we can explain normative defeat as a breach of a constitutive norm, sourced in the constitutive function of these social roles. Since our doctor X is a practicing doctor, they occupy the social role of doctor. As a result, they violate an epistemic norm associated with proper functioning for this role when they fail to maintain an up-to-date understanding of their field (e.g. by missing the research that indicates that q). In this way, it is epistemically proper for them to believe that q. And since q is a reason against believing that p, we get the desired result that X’s justification for believing p is defeated.

How does this account of normative defeat map onto my general account of defeaters? Recall the proposed view: a fact d is a defeater for S’s evidence e for p iff S is in an position to know d and S’s evidential probability that p conditional on e&d is lower than S’s evidential probability that p conditional on e. Recall also that, on my account, being in a position to know is spelled out in terms of availability, as restricted by the type of cogniser instantiated. What I want to suggest is that, in cases like that above, the social role individuates the type of cogniser at stake and the availability conditions follow the corresponding constitutive epistemic oughts: since cognisers like our doctor should be aware of recent developments in their field, the account will predict that these are available to them in the relevant sense – of course, with reasonable qualitative, quantitative, and environmental restrictions.

But won’t this account suffer from a problem parallel to Goldberg’s view? Recall that Goldberg wanted to explain the evidence one should have had in terms of social expectations. Recall, also, that we said that epistemically illegitimate (albeit reliable) social expectations make problems for Goldberg’s account: if epistemic normativity is encroached by social normativity, reliable but epistemically problematic social expectations cannot be further explained in epistemic normative terms.

Won’t my account have the same problem, in virtue of appealing to social roles? Can’t there be social roles that are functionally constituted by norms that are bad, epistemically? Consider, for instance, the social role ‘judge’ in a judicial system where discrimination based on race is written into the laws of the land: isn’t my account going to deliver the result that judges shouldn’t update based on the testimony of, for example, Black testifiers?

It will not. To see this, note that one important advantage that my account has over Goldberg’s is that epistemic normativity is not encroached upon by social normativity: the epistemic remains an independent normative domain with its own independent evaluative structure. On my view, some genuine epistemic norms – associated with promoting epistemic values, such as knowledge – constitute social roles. Compatibly, norms constituting social roles that are bad epistemically, in that they conflict with norms sourced in the proper functioning of our cognitive system – such as ‘don’t believe Black testifiers!’ – are not epistemic norms: they are mere (bad) social norms with epistemic content.

8.4 Conclusion

This chapter developed an account of defeat that builds nicely on the account of evidence developed in the previous chapter. On this view, defeaters are ignorance indicators: they are facts that one is in a position to know and that decrease one’s evidential probability. What differs is the mechanism by which they achieve this effect: rebutters lower one’s evidential probability for p by raising one’s evidential probability for not-p. In contrast, undercutters reduce the degree of confirmation that a particular piece of evidence e confers on p. Finally, the chapter developed a novel, functionalist account of normative defeat and the impermissibility of negligence in evidence gathering.

Chapter 9 Inquiry and Permissible Suspension

This chapter develops an account of permissible suspension that builds on the views of justification, evidence, and defeat defended in the previous chapters. The view is superior to extant competitors in that it successfully predicts epistemic normative failure in cases of suspension generated by evidence and defeat resistance. On this view, doxastically justified suspension is suspension generated by properly functioning knowledge-generating processes. In turn, properly functioning knowledge-generating processes uptake knowledge and ignorance indicators.

9.1 Suspension and the Knowledge Function

I have argued that generating knowledge is the function of our cognitive processes, and that the norms governing moves in inquiry – such as beliefs, suspensions, withholdings, credences, assertions, or pieces of reasoning – will drop out of this function.

Moves in the practice of inquiry – that is, all epistemically significant states and actions – aim either directly (plausibly: beliefs, assertions, reasonings) or indirectly (credences, suspensions, withholdings) at the aim of the practice of inquiry. The difference will lie with goal achievability: since beliefs, assertions, and conclusions of reasonings can be knowledgeable, in a way in which things like credences, suspensions, and withholdings cannot, belief formation aims directly at fulfilling the function of the practice (generating knowledge), while, at the same time, credence, withholding, and suspension aim at knowledge indirectly – they are transitional attitudes, in the sense in which these are attitudes held en route to knowledge but that are not in the running for knowledge.

On my view, such transitional attitudes¹ aim at getting us closer to knowledge: they aim directly at adjusting one’s doxastic states to the available evidence, which, in turn, ultimately aims at the aim of inquiry – knowledge generation. In what follows, I put more flesh on the bones on this general thought.

Importantly, topics in the epistemology of credence, or degrees of belief, deserve a book-length treatment of their own, so I will not touch on this here.² This book restricts itself to full belief and suspension and how evidence and defeat resistance affect the permissibility thereof. In what follows, I offer a sketch of how the knowledge function of inquiry will generate norms for suspensions.

9.2 Justified Suspension

Suspension was, for the longest time, not very hot in epistemology: historically (from Descartes to Clifford, from internalist evidentialists to reliabilists and knowledge-first externalists), people worried mostly about the risks and sins involved in believing without justification, and they ignored the risks of failing to believe when one has plenty of evidence. Recent social and political difficulties sourced in science denialism brought the normativity of suspension to centre stage.

One might think that an account of suspension is straightforwardly predicted by the view of evidence defended here. In particular, one might expect that something like the following principle is correct:

• Suspension–evidence link (SEL): A subject S is justifiedly suspended on p iff S has equally weighty evidence for and against p.

There are, however, many problems with SEL. First, there is a purely terminological problem: most often, the epistemic permissibility of suspending on p is taken to be synonymous with the epistemic permissibility of not forming a full belief that p. In this sense of suspension – which only affects full belief – SEL is false on the necessity direction: one need not have equally weighty support for p and not-p to permissibly withhold full belief in p. Any level of epistemic support short of full propositional justification for p will be sufficient for permissibly withholding full belief that p.

If so, we need to distinguish between (1) what makes full belief permissibly suspended – which is compatible with being permissibly more confident that p than that not-p – and (2) outright permissible neutrality³: taking a fully neutral attitude with regard to p (e.g. suspending belief and forming a 0.5 credence, or withholding (not forming) full belief and holding a 0.5 credence, or withholding both belief and credence). To this effect, for ease of recognition, I will refer to the former variety as permissible belief suspension and to the latter as permissible neutrality. On this picture, we get two different permissibility principles. Here they are:

• Full belief suspension–evidence link (SEL): Suspending belief in p is epistemically permissible for S iff S does not have enough evidential support for a justified belief that p.

• Neutrality–evidence link (NEL): A subject S is justifiedly neutral on p iff S has equally weighty evidence for and against p.

SEL tells us when a subject S can permissibly suspend on p – although they may well be epistemically normatively constrained to form a fairly high credence that p. In contrast, NEL normates tout court neutrality: epistemically permissibly taking a neutral doxastic attitude with regard to p.

SEL and NEL seem fairly plausible at first glance, and they also give us a nice way to think about the nature of reasons to withhold/suspend and their relation to reasons to believe/against believing: on this account, a subject S has a reason to suspend just in case S does not have sufficient reason to fully believe, and, in turn, S has a reason to be neutral on p just in case S has equally weighty reasons for and against p.

Unfortunately, things aren’t as easy as this: suspension and neutrality, just like any other doxastic attitude, afford two types of justification: propositional and doxastic (see also Lord and Sylvan 2021, 2022). In particular, the two will come apart in cases in which S will have sufficient evidence to suspend on p/be neutral on p but will nevertheless fail to do so epistemically permissibly.

The reason why this can happen is improper uptake and evidence handling: one can have evidence for/against p that one fails to uptake/update on or improperly uptakes/processes/updates on, which will result in a lack of doxastic justification for suspension/neutrality just as it results in a lack of doxastic justification for belief. To see how this can be the case, it is easy to imagine cases in which S’s evidence is as per SEL/NEL, but S’s suspension/neutrality either is not based on this evidence (but, say, on wishful thinking) or is based on this evidence in the wrong way (e.g. the evidence supports suspending/neutrality inductively, but S takes it to do so deductively).

If this is so, SEL and NEL are false. What we need are more fine-grained principles that distinguish between these varieties of justification for withholding/suspending. Here it goes, for propositional justification:

• Propositionally justified full belief suspension: Suspending belief on p is propositionally justified for S iff S does not have enough evidential support for a justified belief that p.

• Propositionally justified neutrality: A subject S’s neutrality on p is propositionally justified iff S has equally weighty evidence for and against p.

De facto, then, propositionally justified neutrality will occur in cases in which the relevant evidential probability is at 0.5. In this, the view is evidence-based but not evidentialist (i.e. not evidence-first), since evidence is further unpacked in terms of facts that can be taken up by cognitive processes hosted by the relevant type of cogniser.

An account of doxastically justified neutrality falls outside of the scope of this book, since it will rest on the correct account of credence justification.⁴ How about doxastically justified suspension? It should not come as a surprise to the reader, at this stage, that in my view this will be, once more, a matter of proper functioning. Here it goes:

• Doxastically justified suspending: S’s suspension on p is doxastically justified iff formed via a properly functioning belief-forming capacity that has the function of generating knowledge.

What is the relationship between doxastically justified suspensions and the evidence for which they are held? Once more, pieces of evidence are pro tanto, prima facie justification-makers: they are inputs to the process of belief formation, and when the latter has the function of generating knowledge and is properly functioning, the resulting doxastic attitude is epistemically justified. When evidence is sufficient for full belief, a properly functioning belief-formation capacity with the function of generating knowledge will generate a full belief. When it is not enough, a properly functioning belief-formation capacity with the function of generating knowledge will generate a suspension.

This view of permissible suspension will deal well with the cases of impermissible suspension that made trouble for Sosa’s virtue-theoretic view: George the sexist, for instance, will not be permissibly suspended on where Glasgow Central is, since he has undefeated evidence (Anna’s testimony) that it is to the right. More precisely, since Anna’s testimony raises his evidential probability that Glasgow Central is to the right and no other facts lower it, George should (at least) be more confident that Glasgow Central is to the right than that it is not.

Why should we believe that this account is the metaphysically correct account of suspension (i.e. why should we think that knowledge-generating belief-formation processes are the ones in charge of generating suspensions)? There are a few reasons for this. First, recall the normative picture defended here: I take it that generating knowledge is the function of our epistemic practice of inquiry, and that norms governing moves in inquiry – such as beliefs, suspensions, assertions, or pieces of reasoning – will drop out of this function. This normative picture fits snugly with a picture in which the cognitive capacities in charge of generating knowledge will be the same ones responsible for generating withholdings and suspensions when enough support for knowledge is not available: these processes will seek to form a belief if and only if the belief in question is knowledgeable (Sosa 2021). This function, in turn, will translate into them generating knowledgeable beliefs whenever knowledge is available, but also, as Ernie Sosa puts it, into forbearing when knowledge is not available. In this, as predicted, the normativity of suspension drops right out of the knowledge-generating function of our inquiring practice and of our cognitive systems.

9.3 Suspension and the Normativity of Inquiry

Before moving on, I would like to address a worry that the view of suspension – and, correspondingly, the account of the ought to believe – put forth here is too demanding, in that it would seem as though it asks of us to believe too many things: after all, it would seem as though, at all times, we are both in a position to know a very high number of facts from our immediate environment⁵ and in a position to inquire into a variety of questions.

To the contrary, as I’m about to argue, an important theoretical advantage of the account of suspension proposed here is that, while being able to account for the epistemic impermissibility intuition in cases of resistance to evidence, it also nicely explains the permissibility of ignoring a multitude of facts in our environment to the aim of focusing on issues that we care (or that we should care) about inquiring into. If this is right, the account is just as strong and just as permissive as we want it to be.

To get this into clearer view, consider a puzzle about the normativity of inquiry notably put forth by Jane Friedman:

The Chrysler Building

Say, for instance, that I want to know how many windows the Chrysler Building in Manhattan has. I decide that the best way to figure this out is to head down there myself and do a count. To do my counting, I set up outside of Grand Central Station. Say it takes me an hour of focused work to get the count done and figure out how many windows that building has. During that hour there are many other ways I could make epistemic gains. There is obviously a huge amount of facts around me that I can come to know.

(Friedman 2020, 503)

Here is the puzzle: if some inquiry norms (i.e. norms of gathering evidence, or zetetic norms) are epistemic norms, as the account defended here predicts, then it might looks as though the following paradigmatic zetetic norm is an epistemic norm:

• ZIP: If one wants to figure out [the answer to a question] Q, then one ought to take the necessary means to figuring out Q.

At the same time, the following is an epistemic norm par excellence:

• Kp: If one is in a position to know a proposition, p, then one is permitted to come to know that p.

But, Friedman argues, it would seem as though in the Chrysler Building case, ZIP and Kp come into conflict; after all, as soon as I focus on Q (counting the windows), I am no longer able to pay attention to the myriad of other things happening around me. Because of this, there will be very many things happening around me that I am in a position to know but that I will, as a matter of fact, fail to know. I will, thereby, register a huge amount of epistemic loss. Since events like the one described in the Chrysler Building case are ubiquitous – whenever we inquire into a specific question, we seem to ignore many unrelated facts – it seems to follow that the epistemic domain is peppered with normative conflict and, indeed, failure. Since, according to Friedman, it is implausible that this might be so, one of ZIP or Kp has to go.

A few things about this: first and foremost, as currently stated, ZIP and Kp do not come into normative conflict – after all, in the current formulation, Kp is a permission, whereas ZIP is an obligation. Permissions and obligations cannot come into normative conflict, in that their normative strength cannot pull in two different directions: it is always permissible by the lights of both norms to do whatever the obligation requires.

That being said, on a view like mine, which incorporates justifiers as epistemic obligations, we can – and, indeed, Friedman herself does so later in her paper – reformulate Kp as an obligation and thereby get a revamped version of Friedman’s puzzle:

• ZIP: If one wants to figure out Q, then one ought to take the necessary means to figuring out Q.

• Kp*: If one is in a position to know a proposition, p, then one ought to come to know that p.

While many epistemological accounts will not accept Kp* – indeed, as we have just seen, the vast majority of the literature we have looked at has difficulties accommodating epistemic oughts – and thus will not make the proper target of Friedman’s puzzle, that is not the case with the account defended here: OTB, together with a plausible assumption that, at least most of the time, when there is sufficient evidence for one to come to believe that p, then one is in a position to know that p,⁶ imply Kp*. As such, for now, it would seem as though my account owes Friedman an explanation of what is going on in Chrysler Building-type cases.

Before moving on, though, I want to take one last look at Friedman’s puzzle, only this time focusing on ZIP. Note that, as stated, ZIP is a desire-conditional ought: given the scope of the deontic operator, the obligation only arises upon the desire to inquire being present. However, inquiry norms proper (i.e. norms constituting our practice of inquiry) are not plausibly desire conditional (Kelp 2021). Indeed, constitutive norms never are: think about games. Once you’ve engaged in playing chess, and short of ceasing to do so, it is not up to your desires anymore if you are allowed to move the bishop diagonally or not: it’s a categorical rule of the game. Similarly, what is desire conditional is entering the zetetic domain to begin with, rather than being subject to its constitutive norms once already engaged in inquiry. In order to see this, it will be helpful to distinguish between zetetic norms (i.e. norms constituting inquiry) and norms about inquiry (i.e. norms regulating when one should take on inquiry in a particular domain).⁷ An example of the latter is the norm ‘if you want to be a biologist, go study biology’. Clearly, this is not a zetetic norm, although it is a norm about when one should inquire into a specific domain. In contrast, consider: ‘biologists should know the latest findings in their field’. This, arguably, is a zetetic norm proper: now that one has engaged in biological inquiry, one is under epistemic normative pressure to take up the latest evidence in the field. Let’s restrict ZIP accordingly and outline the final revamped Friedman Puzzle:

• ZIP*: If you engage in an inquiry aimed at figuring out Q, then you ought to take the necessary means to figuring out Q.

• Kp*: If one is in a position to know a proposition, p, then one ought to come to know that p.

The final revamped Friedman Puzzle is, indeed, on the face of it, a puzzle for a view like mine, which takes epistemic justification to be epistemic obligation to believe. After all, it would seem as though, in the Chrysler Building case, there is a lot of evidence lying around about all of things happening around Central Station that I completely ignore. For instance, just as I count the windows on the Chrysler Building, there is a man with a green hat exiting the station. Clearly, the thought would go, given that the man is walking in plain view, I am in a position to know that he’s exiting the station (p). If so, by Kp, I ought to come to know that he’s exiting the station. However, at the same time, since I’ve engaged in counting the windows on the Chrysler Building, it seems as though now I am subject to an obligation to come to know the number of windows on the Chrysler Building. Since I can’t do both at the same time, the thought would go, I’m faced by an inescapable normative conflict.

I believe that many views endorsing epistemic oughts to believe will face this problem (for more about normative conflicts and epistemic dilemmas, see Chapter 10); at the same time, as I’m about to argue, my account does not. Indeed, an important theoretical advantage of my account of evidence, defeat, and suspension is precisely that it not only accommodates intuitive epistemic obligations, but it also, conversely, nicely explains the permissibility of ignoring a multitude of facts in our environment to the aim of focusing on issues that we are inquiring into.

In a nutshell, the reason why my account escapes Friedman’s puzzle is that, on my view, evidence, defeat, and permissible suspension are unpacked in terms of a notion of being in a position to know that predicts that I am not in a position to know that the man with the green hat left the station, nor any other such detail about what is going on at Central Station, at the time when I am counting the windows on the Chrysler Building. Recall the account:

• Being in a position to know: S is in a position to know a fact e iff S has a cognitive capacity with the function of generating knowledge that can (qualitatively, quantitatively, and environmentally) easily uptake e in cognisers of S’s type.

Recall, also, the rationale for the quantitative restriction on easy uptake: there are quantitative limitations on my information accessing and processing – the fact that there’s a table somewhere towards the periphery of my visual field (in contrast of it being right in front of me, in plain view) is not something I can easily process. I lack the power to process everything in my visual field – it’s just too much information.

Quantitative limitations on being in a position to know will make it so that I can only take up a limited number of the e₁, e₂, e₃ … e_n facts that lie within reach with my knowledge-generating capacities. On the account defended, I only shoulder an epistemic obligation to take up a subset of e₁, e₂, e₃ … e_n that is as large as my quantitative uptake limitations. Availability rankings will deliver the relevant set, on my view: the most easily available subset of facts that I can take up is the one that I ought to take up. Crucially, also, note that quantitative limitations on being in a position to know imply the denial of conjunction introduction for being in a position to know: being in a position to know p, q, r, and s individually does not imply being in a position to know p&q&r&s.

If all of this is the case, and given that, by stipulation, I am not able to pay attention to everything that’s going on at the train station while I’m engaged in counting the windows on the Chrysler Building, it follows that, as soon as I will have started counting, I am not in a position to know what is going on at the station anymore. I am not in a position to know that there are eighty-nine windows and a man with a green hat exited the station.

In turn, since, on my account, epistemic obligations are grounded in being in a position to know, I am also under no obligation to form any beliefs about what is going on at the station after I started my inquiry. As soon as I’m subject to ZIP* – because I will have engaged in my inquiry into the question of how many windows there are on the Chrysler Building – I am no longer subject to Kp, because I am not in a position to know what is happening at the station. Therefore, I am at no point subject to ZIP* and Kp* at the same time, and thereby neither to a ZIP*–Kp* normative conflict. My account escapes the Friedman Puzzle.

One might think this is a bit fast.⁸ Of course, Sophie of Sophie’s Choice is also not faced by a dilemma anymore once she has already chosen the twin to save (at t₂). The interesting normative conflict, though, happens at t₁, when she needs to make the choice and she’s faced with a normative dilemma. Similarly, one might think, the interesting normative version of the Friedman Puzzle concerns t₁, when one is supposed to choose between which epistemic obligation to fulfil: that of inquiring into the number of windows or that of forming beliefs about the man in the green hat. Or so the thought would go.

A few things about this. First, note that there is a difference between Sophie’s Choice and the Friedman Puzzle (revamped): Sophie is subject to two unconditional obligations – to save her twins, respectively. In contrast, ZIP* is an obligation only conditional upon already engaging in the relevant inquiry. At t₁, therefore (i.e. before engaging in the relevant inquiry), I am under no epistemic obligation to count the windows: my only epistemic obligation concerns forming beliefs about what’s going on around me (i.e. at the station). At t₂, a practical norm along the lines of ZIP simpliciter (i.e. sourced in my peculiar desire to find out how many windows there are on the Chrysler Building) overrides this epistemic obligation and makes it permissible for me to direct my attention towards the Chrysler Building and start my inquiry into the number of windows on the Chrysler Building. As soon as that occurs, I am no longer under an obligation to form beliefs about what’s going on at the station because I am no longer in a position to know what is going on at the station. Since, on my account, at no time am I under the normative pressure of both ZIP* and Kp* in Friedman’s case, my account does not face Friedman’s puzzle.

9.4 Conclusion

One should suspend if and only if one does so via a properly functioning cognitive process that has the function of generating knowledge and that, in virtue of it being properly functioning, takes up one’s available evidence and defeat. An important theoretical advantage of the account of suspension proposed here is that, while being able to account for the epistemic impermissibility intuition in cases of resistance to evidence, it also nicely explains the permissibility of ignoring a multitude of facts in our environment to the aim of focusing on issues that we care about inquiring into. If this is right, the account is just as demanding and just as permissive as we want it to be.