Noninferential Antiskepticism and the Problem of Easy Knowledge

Abstract

How should epistemologists respond to skepticism about knowledge of the external world? Michael Bergmann advocates noninferential antiskepticism. The thought is that, to reply to a skeptical argument, we should start with premises that do not require inference. I argue that Bergmann’s reasoning runs into the problem of easy knowledge and propose an alternative inferential antiskepticism. This view faces the problem of vicious circularity. I agree that, if we go down the inferential path, a certain type of circularity is unavoidable. I deny, however, that this type of circularity is vicious.

1. Introduction

For epistemologists, perhaps the most formidable task is the refutation of skeptical arguments.¹ How do you know that you are not a brain in a vat (a BIV)? Whatever evidence you might want to bring to bear against the BIV hypothesis, it does not entail that you are not a BIV. At best, it shows that it’s highly improbable that you are a BIV. But can that be shown without getting stuck with one of the three pitfalls the Pyrrhonian skeptics deemed inescapable: choosing a dogmatic starting point, slipping into an infinite regress, or getting entangled in vicious circularity?²

In his excellent Radical Skepticism and Epistemic Intuition, Michael Bergmann offers an admirably deep and systematic response to skeptical arguments. The recipe he recommends—noninferential antiskepticism—is to choose starting points that do not require inference. Critics will argue that this response succumbs to the first of the three aforementioned Pyrrhonian pitfalls: the vice of choosing a dogmatic starting point. I count myself among these critics. I’m going to argue that Bergmann’s noninferential antiskepticism falls victim to what is known as the Problem of Easy Knowledge. What, however, is the alternative? As I see it, the alternative is inferential antiskepticism. Its downside is obvious. If we attempt to infer the falsehood of skeptical alternatives, either an infinite regress or vicious circularity seems inescapable. I agree that, if we go down the inferential path, circularity is unavoidable. I will deny, however, that it is vicious.

2. Two Types of Theories of Justification

Bergmann’s noninferential antiskepticism view has three ingredients:

1. (i) he denies that justification is subject to any higher-order requirements;

2. (ii) he advocates Chisholmian particularism;

3. (iii) he appeals to epistemic intuitions.

We can appreciate the significance of the first of these by comparing two different types of theories about justification: higher-order (HO) and lower-order (LO) theories. HO- theories endorse what we could call the Principle of Meta-Justification: all justification requires higher order or meta-justification. LO theories reject this requirement.

How are we to understand the claim that justification requires metajustification? To answer this question, it is helpful to work with the framework of Michael Huemer’s phenomenal conservatism (PC) as an example of noninferential antiskepticism. According to PC, your beliefs about your physical environment are justified by perceptual seemings, your beliefs about what you did in the past by memory seemings, and your beliefs about a priori matters by intellectual seemings. Seemings are mental states with propositional content, and they represent this content assertively.³ Suppose there is a table before you and, noticing the table’s color, you believe that table is red. In a typical case like that, your belief is justified. PC says that your justification for this belief comes from a visual seeming: the table looks red to you. The question is whether having such a seeming is sufficient for having justification for believing that the table is red.⁴ According to PC, it is, provided you do not have a defeater:

(Sp + you do not have a defeater) → You have justification for believing that p.⁵

This is an example of a LO theory. It says that the table’s looking red to you, provided you have no reasons to distrust this visual seeming, is sufficient for giving you justification for believing that the table is red. It might be argued, however, that having an undefeated visual seeming of redness is not sufficient. In addition to the seeming, you need evidence indicating that the seeming is trustworthy or reliable. This view, credentialism, as I have called it elsewhere, asserts the following:

(Sp + you have undefeated evidence that Sp is reliable) → You have justification for believing that p.⁶

Credentialism is an example of a HO theory. It says that Sp (the table’s looking red to you) is lower order justification, and your evidence for taking Sp to be reliable is meta-justification: justification for believing that, under the circumstances, Sp makes it highly probable that the table is indeed red. The key idea of credentialism is that a visual seeming that p is not justification for you for believing that p unless you have evidence that provides the seeming with proper epistemic credentials: evidence indicating that the seeming is trustworthy. What might such evidence consist of? That is, of course, a crucial question. I’ll get back to it in due time. For now, I’ll just say the following. It consists of memories that the seeming in question—in the case at hand, a visual seeming of redness under normal conditions—has in the past rarely misled you. In short, the meta-justification comes from memory seeming to the effect that a visual seeming of the kind in question has a good track record.

Huemer’s PC is an internalist LO theory. Reliabilism and proper functionalism are externalist LO theories. Bergmann’s preferred theory is proper functionalism. However, in Radical Skepticism and Epistemic Intuition, he takes an ecumenical approach, which I consider one of the book’s virtues. Since PC qualifies no less as a LO-theory than reliabilism and proper functionalism do, it is a view that falls under the umbrella of noninferential antiskepticism and thus, at least as far as the businesses of rejecting skepticism is concerned, meets with Bergmann’s approval.

3. Metajustification: The Good and the Bad

Why think there is no justification without meta-justification? Three reasons come immediately to mind. First, if you have evidence that a particular seeming, Sp, is reliable, then you have assurance that Sp points in a trustworthy manner to the truth of p. Second, since a perceptual seeming that p does not entail the truth of p, there is a skepticism-motivating evidential gap between our beliefs about the world and the way we perceptually experience the world. Meta-justification bridges this gap. Third, meta-justification puts you in a good position when the skeptic, as Peter Klein would put it, “goes meta” and asks you why you a seeming that p is supposed to be a reason for believing p.⁷

At the same time, however, the Principle of Meta-Justification is obviously highly problematic because it is grist on the skeptic’s mill. Using it as a premise, the skeptic can advance the following argument:

The Skeptic’s underdetermination argument⁸.

1 J → meta-J

2 ~◇ meta-J

C ~◇J.

The skeptic’s rationale for the first premise is the aforementioned evidential gap between perceptual experience and your beliefs about the physical world: perceptual seemings do not entail the truth of their content. How to bridge the gap? There is, it would seem, only one way: to show that your perceptual seeming are reliable. But now the second premise kicks in. Is metajustfication possible? Can we show that our perceptual seemings are reliable. The skeptic says ‘no’, on the ground that showing this is impossible for two reasons. First, there is no a priori path to establishing the reliability of perceptual seemings. Second, there is no viable inductive path from past correlations between various types of perceptual experiences, for any such path runs into fatal regress and circularity problems.⁹

4. Two Responses: Noninferential and Inferential

Bergmann finds the skeptic’s rationale for the second premise compelling. His view is, therefore, that theories of justification must be LO theories. They must reject the Principle of Meta-Justification. That, he thinks, is the only way to avoid skepticism. And, as mentioned before, Bergmann is in this regard ecumenical: internalist theories like Huemer’s PC and externalist theories like reliabilism and proper functionalism—all of them LO theories, have equally good antiskeptical credentials.

I, however, agree with the skeptic about the first premise. My view is that we must reject the second premise. I accept the first premise for two reasons: first, the aforementioned good things metajustification has to offer; second because denying it runs into the Problem of Easy Knowledge, which I will turn to soon. To many epistemologists, denying the second premise seems hopeless. I believe, however, that a good case can be made that the kind of circularity meta-justification induces is in fact benign.

5. Mediate and Immediate Justification

Before proceeding, a clarification is in order. Exactly how are we to understand the distinction between inferential and noninferential justification? There are two meanings of “inferential” to be distinguished. Taking inference to be a mental event, the first meaning is psychological:

A belief, B, is noninferentially justified ↔ B is not inferred from another belief.

B is inferentially justified ↔ B is inferred from another belief.

The second meaning is epistemic, having to do with the structure of justification:

A belief is noninferentially justified ↔ its justification is immediate.

A belief is inferentially justified ↔ its justification is mediate.

The distinction between mediate and immediate justification has been helpfully highlighted by James Pryor (2000, 532f). Let us use as an example a perceptual belief about an ordinary object like “There is a red table before me.”

Your justification for this belief is mediate: it comes from a perceptual seeming (the table looks red to you) and metajustification for taking this seeming to be reliable.

Your justification for this belief is immediate: it comes solely from a visual seeming. It does not come from any meta-justification, that is, it does not come from having justification for any additional proposition, for example the proposition that your visual seeming is reliable, or that the table is not white with red light shining at it.

It’s important to see that a belief can be psychologically noninferential even if its justification is mediated. How? Suppose credentialism is true. Your belief that the table is red is justified by a visual seeming and meta-justification for believing the seeming is reliable. If so, the justification for your belief is mediated. But that does not mean that your belief is psychologically inferential. It might not, and typically would not, be inferred from any additional beliefs about the reliability of your visual seeming and the nonobtaining of any skeptical alternatives. Typically, your belief would be cognitively spontaneous: not inferred from anything. However, it’s being psychologically noninferential leaves it completely open whether your total justification for the belief does or does not include meta-justification.

I assume that, when Bergmann advocates noninferential antiskepticism, what he has in mind is the epistemic, not the psychological, sense of inference. According to him, I take it, perceptual seemings are sources of immediate justification. That is what HO-theories deny. According to them, all perceptual justification is mediated.

Next, I will turn to the second ingredient of Bergmann’s view: Chisholmian particularism.

6. The Problem of the Criterion

The prominence of the problem must be attributed Roderick Chisholm’s influential writings on what he called the ancient “problem of the wheel.”¹⁰ I think the problem is seen best as that of how to respond to the following skeptical argument:

EC You can know the extent of knowledge → you already know the criteria of knowledge.

CE You know the criteria of knowledge → you already know the extent of knowledge.

NP (EC + CE) → you can know neither the extent nor the criteria of knowledge.

Therefore:

BN You can know neither the extent nor the criteria of knowledge.¹¹

What are the criteria of knowledge? Ideally, the conditions that are individually necessary and jointly sufficient for knowledge. On this understanding, the first premise is not very plausible. For example, for you to know that you know that you have hands, you do not have to have a correct analysis of knowledge. According to a more easy-going approach, to determine whether you know in particular cases, you merely need to have a basic grasp of the link between knowledge, truth, and justification. If you do not even realize that knowledge requires true belief and having good reasons or evidence for what you believe, it looks like you are not in a position to make any judgments at all about whether, in a particular case, you have knowledge. On such a relaxed understanding of “criteria of knowledge,” the first premise is plausible.

The second premise is plausible, too. Some criteria of knowledge are too demanding. For example, knowledge does not require truth-entailing justification. Other criteria are not demanding enough. For example, knowledge requires more than just having a true belief. But, if you are entirely clueless about what you know and what you do not, how can you tell which criteria are correct? So, the CE premise is plausible as well.

The problem is that, when conjoined, EC and CE constitute the “wheel,” a vicious circle: to have A, you need to have B first, and to have B, you need to have A first. If that is the way A and B are connected, then you cannot have either one. That’s the rationale for the NP premise. The upshot is that the skeptical outcome, BN, seems inescapable.

There are four ways of responding to the problem. The first is to accept the skeptical conclusion. That does not have a lot of appeal. The second is Chisholm’s preference: particularism. According to this response, CE is true, but EC is false. The alternative that Chisholm rejects is what he calls “methodism”: EC is true and CE is false. I myself think the best response is to accept both EC and CE but to deny NP. For this move to be plausible, it would have to be shown that the conjunction of EC and CE actually does not create a vicious circle. I happen to think it does not.

7. Benign Mutual Dependence

NP is plausible if we understand the dependence relation signified by “→” as a temporal sequence. It is then a diachronic relation: one thing must come in time before the other. If you can have A only if you acquire B first, and if you can have B only if you acquire A first, then you cannot have either one. However, NP is false if we understand the relation as epistemic dependence at a time. It is then a synchronic relation: the two things in question must obtain at the same time. The claim would merely be that you cannot have A without also having B, and that you cannot have B without also having A. That kind of relation does not preclude having both at the same time.

How can this idea be applied to the issue at hand: knowing both the extent and the criteria of knowledge? The key, I suggest, is concept acquisition, or what we might call cognitive evolution. ¹² I cannot offer an in-depth development of this idea here. Just conveying the basic thought must suffice. What I have in mind is that acquiring the concept of knowledge involves learning—at least in a rudimentary way—the criteria of knowledge. Learning the concept and its application conditions go hand in hand. The upshot is that the EC-CE circularity is not vicious. We need not, therefore, take on board either particularism or methodism. Since either one of these theories comes with considerable pain, that would seem to be a good thing.¹³

8. Particularism and Methodism Rejected

If we accept the outcome that the EC-CE circularity is not vicious, we need not sign on to either particularism or methodism. We can, as I think we should, reject both of these responses. Particularism does not look plausible because at least some principles can reasonably be taken as starting points. For example:

The Truth Condition: Kp → Tp.

Justification Closure: [Jp + J(p → q)] → Jq.

Conservatism: Seemings provide justification unless they are defeated.¹⁴

Methodism, on the other hand, is problematic because some epistemic intuitions can reasonably be taken as starting points. For example:

It seems to you that you know that you exist, that you have hands, that other people exist, that you existed yesterday, etc.

According to G. E. Moore, epistemic intuitions of this kind—an intellectual conviction that we won’t in fact know the things just mentioned—are more trustworthy than the skeptic’s intuitions about principles. I agree with Bergmann that Moore is right about that, and thus agree with the third ingredient of Bergmann’s noninferential antiskepticism: it is legitimate to appeal to epistemic intuitions. I would add, however, that accepting Moore’s antiskeptical move does not commit one to accepting particularism. The appeal to epistemic intuitions also paves the way towards treating as starting points epistemic principles like the ones I just mentioned. Therefore, we should reject both methodism and particularism: the former because it can reasonably appeal to epistemic intuition about particular cases, the latter because we can also reasonably appeal to epistemic intuitions about axiomatic epistemic principles.¹⁵

9. The Problem of Easy Knowledge

In his influential paper “Basic Knowledge and the Problem of Easy Knowledge,” Stewart Cohen sets forth a line of reasoning that amounts to a serious objection to LO-theories (Cohen 2002. See also Markie 2005, Pryor 2004, and Cohen 2005). According to Cohen, LO theories make the acquisition of certain kinds of knowledge too easy. The first kind is knowledge of the falsehood of local or global skeptical alternatives; the second is knowing that a particular belief source is reliable. Here, I will focus only on the first of these. The LO theories that Cohen targets are externalist reliabilism and internalist theories like Huemer’s PC or James Pryor’s dogmatism.¹⁶ The hallmark of these theories is that they reject high-order requirements. The specific higher-order requirement Cohen has in mind is:

KR A potential knowledge source, S, can yield knowledge for you → you know that S is reliable.

Although I think the KR principle, if suitably interpreted, is defensible, I will focus on a parallel principle about perceptual justification:¹⁷

JR_P Perception is a justification source for you → you have evidence that perception is reliable.

Theories that deny JR_P run into the Problem of Easy Justification: they make it too easy for you to acquire justification for denying local or global skeptical alternatives. To see why, let us use the framework of Huemer’s PC.

Let us stipulate the following: You see a red table before you under perfectly normal conditions of observation. You have no reason whatever to think you are hallucinating, or that red lights are shining at the table. You are justified in believing, and indeed you know, that the table is red. Your justification comes from a visual seeming: the table looks red to you. You are also aware of the following obvious entailment: if this is a red table, it is not a white table with red light shining at it. Now let us ask a Chisholmian Socratic question: What gives you justification for believing that the table you see is not a white table with red light shining at it?

10. The Problematic Argument

Suppose you respond to the question with the following argument:

(1) This is a red table.

(2) (1) → ~(It is a white table with red light shining at it.)

Therefore:

C ~ (It is a white table with red light shining at it.)

Given our stipulations, the argument is sound, and you are justified in believing its premises. Hence, by justification closure, it follows that you are justified in believing the conclusion. What, then, could possibly be wrong with responding to the Socratic question by giving this argument?

What’s wrong with the argument is that it begs the question! Question-begging arguments never track how you actually acquire justification for their conclusions. What gives you justification for denying the skeptical alternative is not the visual seeming that’s your justification for the first premise combined with the a priori intuition that’s your justification for the second premise. Instead, it’s independent evidence: you see that there are no red lights shining at the table. The problem for LO theories like reliabilism and Huemer’s PC is that this criticism is not available to them. It’s unclear on what grounds they can reject the Problematic Argument as question begging.

11. Question Begging and Epistemic Circularity

What is it for an argument to beg the question? Here is a plausible answer: it is for the argument to be epistemically circular. An argument is epistemically circular just in case it has at least one premise for which you cannot have justification unless you already have justification for believing the conclusion. Let us apply this thought to the Problematic Argument.

HO theorists can say that the problematic argument is infected with epistemic circularity because having justification for the first premise is impossible without already having justification for the conclusion. That is, for the visual seeming to give you justification for believing that the table is red, you must already have justification for taking the visual seeming to be reliable, which you have only if you also have justification for denying the skeptical alternative. LO-theorists, since they reject JR_P, are not in a position to say this. It looks like they cannot explain what’s wrong with the Problematic Argument.

12. The Problem Generalized

The red table example is just one of many. There are lots of local and global skeptical alternatives, for example:

• - This women looks like Judy, but she is in fact her identical twin Trudy.

• - These animals look like zebras, but they are in fact cleverly painted mules.

• - What’s in this glass looks like red wine, but it’s in fact red-colored water.

• - It looks like I have hands but I am in fact a handless brain in a vat.

For each, there is a version of the Problematic Argument that deduces the falsehood of the skeptical alternative from two things: immediate justification for the proposition the skeptical alternative targets, and immediate and obvious justification for the relevant entailment.¹⁸ In each case, the Problematic Argument seems defective. HO theorists can explain its defect in terms of epistemic circularity. In each case, the argument’s first premise cannot be known without having already justification for denying the alternative. Since LO theories deny this requirement, it is unclear on what grounds they can reject the Problematic Arguments as defective.

13. Bergmann’s Solution to the Problem of Easy Knowledge

Bergmann argues that the Problematic Arguments are not always bad. He distinguishes between justification-precluding and justification-permitting situations (2021, 182–86). Returning to our initial example of the red table, suppose you have reason to doubt that the table is really red. Perhaps you have reason to believe that something is wrong with your color vision, or that there are hidden red lights. Bergmann would say that that’s a justification-precluding situation. You cannot acquire justification for rejecting the alternative on the basis of the Problematic Argument because, given your doubts, you do not have justification for the argument’s first premise. Now, suppose you have no reason to doubt that the table is red. As far as you can tell, the conditions of observation are just fine. Mike would say that that’s a justification-permitting situation. You can reasonably reject the alternative on the basis of the Problematic Argument because now you have justification for the argument’s first premise.

This move gives us the right result for a wide range of cases. Suppose that, walking across campus, you see a student from your class. You say “Hi Judy.” You then remember that Judy told you she has a look-alike identical twin-sister, Trudy. What might justify you in believing that she is not Trudy? Surely not that she looks like Judy. Likewise, if you have reason to doubt that the animals in the pen are really zebras because you suspect they might be cleverly painted mules, then it would be bad to argue “I know they are not cleverly painted mules because I know they are zebras.”

To sum up, Bergmann’s solution is that the Problematic Arguments are not always bad. They are bad only when you are in a justification-precluding situation.

14. Bergmann’s Solution Rejected

I do not think Bergmann’s solution works. About the red table case, I have stipulated what Bergmann would call justification–permitting circumstances. You have no reason to doubt that the table is red. You are justified in believing, and you know, that the table is red. Nevertheless, the argument is bad: incapable of generating justification for the conclusion. What gives you justification for denying the alternative is not the Problematic Argument. Rather, for the visual seeming to give you justification for the first premise, you must already have justification for thinking that the alternative does not obtain. Again, what gives you justification for denying the skeptical alternative, even in the stipulated felicitous circumstances, is independent evidence.

Consider a case of an identical twin alternative. Suppose I’m giving a talk and I see Mike Bergmann in the audience. I’m justified in believing that the person in the audience who looks like Mike Bergmann is not another person who looks exactly like Bergmann because he is Mike Bergmann’s identical twin, a twin who also happens to have an interest in epistemology. This is a justification-permitting situation: I have no reason at all to doubt that the person looking like Mike Bergmann is in fact Mike Bergmann. But what justifies me in rejecting the twin alternative is not that this person is Mike Bergmann because he looks like Mike Bergmann. Instead, it’s independent evidence: I’ve known Mike Bergmann for a long time, but he never said anything about having an identical twin with an interest in epistemology.

So, I do not buy Bergmann’s solution. I think the kind of epistemic circularity the problematic argument displays is always bad.¹⁹ It’s badness cannot be explained without appealing to the Principle of Meta-Justification. Since LO-theories reject this principle, they cannot explain what’s bad about problematic arguments for the denial of skeptical arguments.

15. The Regress Worry Dismissed

Let us review the dialectical situation. There are serious problems for both types of theories. LO theories run into the problem of easy knowledge. HO theories face regress and circularity worries. I will quickly dismiss the former and then focus on the circularity issue. So, does the Principle of Meta-Justification trigger an infinite regress? Well, if the principle required beliefs about the reliability of our belief-sources, then it certainly would. However, as I have articulated the principle it does not require higher-order beliefs. It merely requires possessing evidence—in the form of suitable memory seemings—that your belief-sources are reliable. If you have such evidence, then, on reflection, you are in a position to form beliefs about the reliability of your faculties. The principle does not, however, require that you reflect and form such beliefs.²⁰

16. Bergmann’s “You Can’t Get Started” Argument

According to Bergmann’s noninferential antiskepticism, “unverified sources can produce justified beliefs” (2021, 175). Bergmann’s reason for this claim is that, if the claim was false, we could not get started in forming any justified beliefs. If the claim was false, he says, “we couldn’t form our first justified belief unless we had already formed a previous justified belief, which would mean that justified belief is impossible.”²¹ This is a serious objection.

As I see it, the problem here is analogous to the Problem of the Criterion. It’s the Pyrrhonian Problem of the Wheel. As before, we can present the problem in the form of a skeptical argument:

Perceptual Reliability and the Problem of the Wheel.

JR_P Perception is a justification-source for you → you have evidence of perceptual reliability.

RJ_P You have evidence of perceptual reliability → perception is a justification-source for you.

NP_P (JR_P + RJ_P) → Perception cannot be a justification-source for you and you cannot have evidence of perceptual reliability.

Therefore:

BD_P Perception cannot be a justification-source for you and you cannot have evidence of perceptual reliability.

JR_P is the perceptual justification counterpart to Cohen’s KR principle. The RJ_P premise asserts dependence in the opposite direction: since there is no a priori path to establishing the reliability of perception, you can acquire justification for taking perception to be reliable only if perception is, to begin with, a source of justification for you. The NP_P premise asserts that the first two premises constitute a vicious circle. If they are true, neither of the desired epistemic goods is possible. Unless we can reject one of these three premises, we must accept radical skepticism.

As with the Problem of the Criterion, I want to reject the NP premise. The solution to the problem is to insist that the first two premises, if plausible, assert merely a synchronic dependence relation:

JR_P Perception is a justification-source for you at t → you have, at t, evidence that perception is reliable.

RJ_P You have, at t, evidence that perception is reliable → perception is, at t, a justification-source for you.

If the diachronic interpretation of JR_P-RJ_P premises were true, there would be no way out of the circle.²² But there is no reason to agree with the skeptic that the two premises must be understood in this way. If we articulate them as displayed above, then two conditions can be satisfied simultaneously. Next, let us consider an example of a benign synchronic dependence relation: a case in which two things obviously exist although they depend on each other in such a way that one could not exist without the other one.

17. Benign Mutual Dependence

The protagonists of my example are the clownfish and the sea anemone. With its toxic tentacles, the sea anemone protects the clownfish from predators. In return, the clownfish nourishes the sea anemone with its droppings. The corresponding skeptical circularity argument goes as follows:

FA The clownfish exists → the sea anemone exists.

AF The sea anemone exists → the clownfish exists.

NP FA and AF are both true → neither one can exist.

Therefore:

BN Neither one can exist.²³

Obviously, the conclusion is false. Both kinds of creatures do exist. But the premises are true: the clownfish and the sea anemone depend on each other. So, the culprit must be NP. It’s a case of benign mutual dependence. The upshot is this: the fact that two things depend on each other does not entail that they cannot exist. The same lesson applies to perceptual justification and evidence of perceptual reliability. From the premise that they depend on each other, it does not follow that neither one is possible.

18. How We Get Started: Cognitive Evolution

How to acquire evidence of perceptual reliability without getting stuck in the wheel? That’s a puzzle. My solution to the puzzle is the first part of my response to Bergmann’s “You Can’t Get Started” Argument. The solution is to see that the mutual dependence between perceptual justification and evidence of perceptual reliability is compatible with having both. But seeing that we can have both does not explain how we can have both. Providing that explanation is the second part of my response.

How, then, do we get started acquiring evidence of perceptual reliability? Consider again the clownfish and the sea anemone. How do they get started? The answer is: through biological evolution. Over thousands of years, these creatures evolved in tandem, ending up in a situation of mutual dependence. A parallel answer suggests itself for the other question: How can we get started acquiring evidence of perceptual reliability? We get going through a process we could call “cognitive evolution.” Consider infants: when they start observing the world around them, they more or less immediately start acquiring track-record evidence of perceptual reliability. Things that can be seen can be touched, heard, tasted, and smelled. In this way, perceptual seemings of different kinds confirm each other, thereby establishing a track record of perceptual success. Perceptual justification and evidence of reliability evolve in this way simultaneously. As the developing human gets older, the web of mutual confirmations begins to include introspection, memory, and reason. Upon reaching full cognitive development, an evidential web has evolved that justifies humans in taking there faculties to be reliable.²⁴

19. Conclusion and Outlook

Bergmann suggests that, in theory choice, philosophers rely on their strongest intuitions (2021, 125). I agree. When we consider LO and HO theories, there are supporting intuitions on both sides of the issue. It’s an intuitively attractive move to reject the principle of metajustification on the ground that it triggers a regress and circularity problem. However, my strongest intuitions push me in the direction of endorsing the principle. It seems to me that, if properly construed, the principle has a lot to offer and triggers neither an infinite regress nor vicious circularity.

However, the devil is, as always, in the details. I have argued that the Problem of Easy Knowledge counts against LO theories because the problematic argument is epistemically circular, and epistemic circularity is always bad. At the same time, though, it’s obvious that HO-theories are unavoidably tied up with source circularity. It’s impossible to acquire evidence of perceptual reliability, in the way I have suggested, without using perception. But now a crucial question arises. Can we, in the way I am envisioning, rely on our faculties to acquire evidence of their reliability without relying on epistemically circular reasoning? Perhaps source circularity inevitably turns into epistemic circularity. In that case, HO theorists would have to distinguish between two kinds of epistemic circularity: good and bad. The idea would then be that the problematic argument is epistemically circular in a bad way, whereas source-circular track-record arguments in support of perceptual reliability are epistemically circular in a good way. Alternatively, perhaps a case can be made that track-record reasoning in support of reliability, while source circular, is not an instance of epistemic circularity. These are difficult questions for further reflection.²⁵