From Synthese, 55 (1983), 73-95. Reprinted in William P. Alston, Epistemic Justification: Essays in the Theory of Knowledge, 1989.

What's Wrong with Immediate Knowledge?

William P. Alston

In this essay I will consider what seem to me the most interesting current arguments for the impossibility of immediate knowledge. I shall conclude that they all fail to foreclose that possibility. I shall not explicitly argue that the possibility is realized, though it will become clear in the course of my argument where I think that obvious examples are to be found.

Attacks on immediate knowledge are nothing new. They were a staple of nineteenth-century absolute idealism1 and were prominent also in its American offshoot, pragmatism.2 But after a hiatus from roughly 1920 to 1950, these attacks have been resumed in English-speaking philosophy, with the revival of pragmatist and holistic ways of thinking in such philosophers as Quine, Sellars, Rorty, and Davidson. I feel that the time is ripe for a critical review of these arguments in their most recent guises. Before starting on that I should make it explicit that my rejection of these arguments does not imply that I consider everything in recent pragmatism, holism, and coherence theories to be unsound.


Let me specify at the outset in what sense I will be defending the possibility of immediate knowledge, since the term is by no means unambiguous. The rough idea is that whereas mediate knowledge depends for its status as knowledge on other knowledge, immediate knowledge does not. Mediate knowledge is, immediate knowledge is not, mediated by other knowledge. To make this more precise we will have to dig down into the concept of knowledge, and that takes us into highly controversial territory. If we could suppose that knowledge is true justified belief, plus some fourth requirement to avoid Gettier-type counterexamples, we could make the distinction between mediate and immediate knowledge hang on the distinction between mediate and immediate justification, which could then be explained as follows.

  1. S is mediately justified in believing that p -- S is justified in believing that p by virtue of some relation this belief has to some other justified belief(s) of S.
  2. S is immediately justified in believing that p -- S is justified in believing that p by virtue of something other than some relation this belief has to some other justified belief(s) of S.
However, some contemporary epistemologists think that what converts true belief into knowledge is reliability rather than justification, where a "reliable" true belief is one that has originated, and/or is sustained, in a way that is generally reliable, that will generally produce true rather than false beliefs.3 To further compound the confusion, some reliability theorists take reliability to be, or to be an adequate criterion for, justification.4 In this essay I want to avoid these controversies so as to focus on the issues raised by the arguments I will be examining. I can do this by leaving open just exactly what it is that plays the role in the concept of knowledge that many contemporary theorists assign to justification. I shall coin a neutral term, 'epistemization', for the function performed by whatever fills this role. That is, an "epistemizer" will be what converts true belief into knowledge, perhaps subject to some further condition for avoiding Gettier counterexamples. Justification and reliability will be two leading candidates for the role of epistemizer (or the same candidate, depending on how 'justification' is explained).5 We can then distinguish between mediate and immediate epistemization in the same terms we used above for distinguishing mediate and immediate justification.
  1. S's belief that p is mediately epistemized -- S's belief that p is epistemized by some relation this belief has to some other epistemized belief(s) of S.
  2. S's belief that p is immediately epistemized -- S's belief that p is epistemized by something other than some relation this belief has to some other epistemized belief(s) of S.
Putative mediate epistemizers include (a) having adequate evidence for the belief in question and (b) the belief in question having been arrived at by inference in a way that will generally produce true beliefs. Immediate epistemization is a wastebasket category. It embraces any form of epistemization that does not involve relations to other epistemized beliefs of the same subject. Hence the range of conceivable immediate epistemizers is much wider. Popular candidates include (a) immediate experience of what the belief is about, (b) for certain special cases, simply the truth of the belief, or the fact that it is believed or understood, (c) facts about the origin of the belief, for instance, the fact that a certain perceptual belief arose from normal perceptual processes.

Plausible candidates for immediate knowledge include one's knowledge of the simplest logical and mathematical truths: 'No proposition is both true and false', '2 + 3 = 5' -- and one's knowledge of one's own current states of consciousness: 'I feel relieved', 'I am thinking about next summer's vacation'. In both sorts of cases it seems implausible to suppose that one knows the item in question only by virtue of knowing or being justified in believing something else, on which the first knowledge is based. Requests for evidence or reasons for one's first-person current conscious state attributions are clearly out of place. "What do you mean, what reason do I have for supposing that I feel relieved? I just do, that's all."6 Again, although '2 + 3 = 5' can be derived from other propositions (as can 'I feel relieved', for that matter), one normally feels no need to do so or to be able to do so, in order to know it to be the case. It seems that we can see that 2 + 3 = 5, just by considering that proposition itself. A simple perceptual belief, for example, that there is a tree in front of me, or if you prefer, that I see a tree in front of me, is a more controversial case. A normal adult could provide a reason if pressed: "It looks like a tree" or "I am having the kind of experience I would have if I were seeing a tree". But it seems that a being too unsophisticated to come up with any such reasons could still have perceptual knowledge that there is a tree in front of him just by virtue of forming that belief by normal perceptual processes in normal circumstances,

I should make it explicit that what I am going to be defending in this paper is what we may call "wholly immediate knowledge". Recently it has been pointed out by several writers that one might think of certain beliefs as justified partly immediately and partly mediately, in such a way that the belief has justification sufficient for knowledge only by combining the two sources.7 Thus it might be that a perceptual belief is justified to some extent just by being formed by normal perceptual processes in normal circumstances, but that this is not sufficient for knowledge (even given truth and whatever may be required over and above justification and truth). In addition, the belief would have to "cohere" with other things one knows, or it would have to be supported by reasons for supposing that the conditions, of perception are normal. In that instance we might speak of a case of perceptual knowledge as "partly immediate" since part of what epistemizes the belief is something other than its relation to other justified beliefs of the same subject. This is an interesting suggestion and worthy of careful examination, but in this paper I shall restrict myself to the question of the possibility of wholly immediate knowledge.

The question of the possibility of immediate knowledge is frequently assimilated to the question of the viability of foundationalism, but the questions are distinct. Foundationalism is a theory of the structure of knowledge. It holds, to put it briefly, that all mediate epistemization ultimately rests on immediately epistemized beliefs. Trace back a chain of mediate epistemization and you will eventually reach an immediately epistemized belief. Clearly foundationalism entails the possibility of immediate epistemization, but not vice versa. One could recognize that some beliefs are immediately epistemized but deny that mediate epistemization always rests on such beliefs, as foundationalism maintains. I will not be discussing the contentions of one or another version of foundationalism, other than the possibility of immediate epistemization.

Much of the attack on immediate knowledge has focused on some particular putative immediate epistemizer. The concept of immediate awareness, has been extensively criticized, in absolute idealism, in pragmatism, and in more recent writings.8 The notion of a belief's being "self-justified" has come in for a good deal of attack.9 Such opponents often assume that disposing of their chosen target will amount to the elimination of immediate knowledge. But even where such arguments succeed in unmasking a particular alleged epistemizer, they fail in their more ambitious task, because of the indefinite plurality of possible immediate epistemizers. Even if there is something radically wrong with the concept of an immediate experience of a particular or of a fact, there is still the claim that some beliefs are self-warranted, the claim that some beliefs are epistemized by a reliable noninferential origin, and so on. One could set out to discredit all the immediate epistemizers that have actually been put forward, one by one. But at best such a procedure would fail to show that all possibilities have been eliminated.

In this paper I am going to confine myself to arguments that are directed against any sort of immediate epistemization and immediate knowledge. In keeping with this restriction I shall even forgo considering an important argument to the effect that wherever an immediate justification for a belief is defeasible we can be (sufficiently) justified in the belief only if we are justified in believing that no defeating circumstances obtain.10 Since there are putative immediate justifications that do not seem to have this prima facie character, for example, my justification for supposing that I feel tired now, or for supposing that 2 + 3 = 5, this argument, even if successful, would not rule out all immediate knowledge.


As a preliminary to examining the arguments I take most seriously, I shall dispose of some tempting but misdirected arguments that turn out to hit some other target instead.

(1) First I will briefly note that some theorists seem to suppose that the beliefs involved in immediate knowledge must be infallible, incorrigible. or indubitable,11 and hence that by showing that none of our beliefs enjoy those immunities, one will have shown that there can be no immediate knowledge. At least opponents of "foundational" or "basic" beliefs, which must be immediately epistemized to fill that role, have often supposed that such beliefs must enjoy such immunities.12 But a moment's reflection will assure us that there is nothing in the concept of immediate epistemization, any more than in the concept of mediate epistemization, that limits its application to beliefs that cannot (in some significant sense) be mistaken, refuted, or reasonably doubted.

(2) I have a sense that it is a rather widely shared view that a belief can be immediately epistemized only if it in no way depends on other knowledge of the same subject, only if it could be held without the subjects knowing anything else; though I must confess to some difficulty in finding this explicitly affirmed in print.13 In any event, if that were a condition of immediate knowledge, it would be a serious liability, for there are powerful reasons for denying the possibility of knowledge that is isolated to that extent. Speaking with absolute generality, it is plausible to hold that I can't know something of the form 'x is P' without having general knowledge as to what it is for something to be P. And getting down to standard putative cases of immediate knowledge, it is a widely held view that I can't have knowledge only of my own conscious states. Such knowledge, and hence any particular instance of such knowledge, presupposes that I know something about the ways in which states of consciousness are manifested in publicly observable behavior and demeanor. And as for '2 + 3 = 5' and the like, it is very plausible to hold that one could not have knowledge of a particular arithmetical truth without knowing at least some significant part of a larger arithmetical system. If one tried to teach a child that 2 + 3 = 5 while keeping him ignorant of, for instance, '1 + 1 = 2', he would fail miserably. Of course, these contentions can be, and have been, controverted. But since I will be arguing that they are, in any event, irrelevant to the issue of immediate knowledge, I need not defend them. It is enough that they have been held with some show of reason.

I want to deny that the cases of dependence just cited are incompatible with the existence of immediate knowledge. How can this be? Well, it all depends on the sort of dependence involved. Immediate knowledge requires independence of other knowledge, so far as the epistemization of belief is concerned. Immediate knowledge is knowledge in which the belief involved is not epistemized by a relation to other knowledge or epistemized belief of the same subject. But in the above cases what is alleged is that the very existence of the belief depends on other knowledge. Unless I know what it is to be P, I can't so much as form the belief that x is P, for I lack the concept of P. Unless I know something about outward criteria of conscious states, I cannot so much as form the belief that I feel tired, for I lack the concept of feeling tired. Unless I know something about the rest of the number system, I cannot so much as form the belief that 2 + 3 = 5, for I lack the requisite concepts. But all this says nothing as to what epistemizes the belief, once formed, and it is on this that the classification into immediate or mediate depends. The question of what epistemizes a belief only arises once the belief is formed. That question presupposes the existence of the belief and hence presupposes any necessary conditions of that existence. It is then a further question whether the belief is epistemized and, if so, by what. Hence it is a further question whether that epistemization is mediate or immediate. To suppose that the conditions for forming the belief are themselves conditions of epistemization, and hence determinative of the choice between mediate and immediate, is to confuse levels of questioning. It would be like arguing that since a necessary condition of my making a request (orally) is that I have vocal chords, part of what justifies me in making that request is that I have vocal chords. The existence of immediate knowledge is quite compatible with a thoroughgoing coherence theory of concepts, according to which one could not have a single.concept without having a whole system of concepts, and even with the further view that the possession of a system of concepts requires having various pieces of knowledge involving those concepts.14

(3) It is very plausible to suppose that any belief, however it arose, can be evaluated for truth, justification, or rationality by reference to reasons or evidence. However I came to believe that 2 + 3 = 5 or that there is a tree in front of me, or even that I feel tired, it is possible, for me or for someone else, to look for reasons for supposing that it is true or false. And sometimes such reasons can be found. There is even some plausibility in holding that it is always, in principle, possible to find such reasons. But whether or not the latter claim is correct, it will at least follow that any belief is subject to assessment in terms of reasons or evidence. And it has been thought that this is incompatible with supposing that any belief is immediately epistemized. But again this is just a confusion. To say that a belief is immediately epistemized is not to imply that it could not also be mediately epistemized, even at the same time. It is only to say that there is an epistemization, not involving other knowledge or epistemized belief of the same subject, that is sufficient for knowledge.15 Epistemic overdetermination is just as possible as the causal variety. Just as the existence of one set of causally sufficient conditions does not rule out the possibility of another set, so the existence of one (mediate) epistemization is quite compatible with the existence of another (immediate) one.


Now I turn to the criticisms I will take more seriously. They all involve what we may call the "Level Ascent" argument. According to this argument, when we consider any putative bit of immediate knowledge, we find that the belief involved really depends for its epistemization on some higher level reasons that have to do with its epistemic status, with the reliability of its mode of formation, or with what it is that is supposed to epistemize the belief. In recent decades the Level Ascent argument has been prominent in the writings of Wilfrid Sellars, and I shall first look at its Sellarsian form.16 It may be doubted that Sellars can be counted among the foes of immediate knowledge, for he is wont to present his position as a sort of synthesis of foundationalism and coherentism.17 But as we shall see, the foundationalist ingredient in the brew does not include any recognition of full-blooded immediate knowledge.

The earliest explicit rejection of immediate knowledge known to me in Sellars' works comes in an oft-quoted section of "Empiricism and the Philosophy of Mind" (EPM), first published in 1956. Having disposed, to his satisfaction, of the view that the "authority" of observational reports stems from their correctly formulating the content of non-propositional awarenesses that are "self-authenticating". Sellars goes on to consider what alternative there might be. He begins with the following possibility.

An overt or covert token of 'This is green' in the presence of a green item . . . expresses observational knowledge if and only if it is a manifestation of a tendency to produce overt or covert tokens of 'This is green' -- given a certain set -- if and only if a green object is being looked at in standard conditions, (p. 167)
This is what has since come to be known as a reliability account of observational knowledge. What makes this a case of knowledge is that the belief (or in this case the statement) stems from a habit that can be relied on to produce true beliefs (statements). This would be one form of the view that such knowledge is immediate knowledge, for the specified necessary and sufficient condition does not require the subject to have other knowledge or justified belief. But Sellars does not accept this account. It "won't do as it stands" (p. 167). Although the "authority" of the report stems from "the fact that one can infer the presence of a green object from the fact that someone makes this report" (p. 167), that is, from the fact that the report was a manifestation of a reliable tendency, still "to be the expression of knowledge, a report must not only have authority, this authority must in some sense be recognized by the person whose report it is" (p. 168). In other words, "no tokening by S now of 'This is green' is to count as 'expressing observational knowledge' unless it is also correct to say of S that he now knows the appropriate fact of the form X is a reliable symptom of Y, namely that. . . utterances of 'This is green' are reliable indicators of the presence of green objects in standard conditions of perception . . ." (p. 169). In still other terms, Jones does not know that this is green unless he is able to take the formation of his statement (belief) in these circumstances as a reason for supposing that a green object is present (p. 168). Since what is required for knowing that this is green (over and above true belief, that is to say, what is required for epistemization) includes Jones's having certain specific pieces of knowledge and the ability to use them to support the proposition in question, Sellars is clearly denying that observational knowledge is or can be immediate knowledge, as that term was explained above. His reason for denying it clearly falls under our Level Ascent rubric. One's belief counts as knowledge only if one knows something about the epistemic status of that belief, viz., that it counts as a reliable sign of the fact believed. And, equally clearly, this move could be used against any claim to immediate knowledge.

The exposition in EPM leaves things insufficiently explicit in at least two respects. The first and less serious has to do with the way in which the view is supposed to give something to the foundationalist. In an oft-quoted passage, Sellars writes:

There is clearly some point to the picture of human knowledge as resting on a level of propositions -- observation reports -- which do not rest on other propositions in the same way as other propositions rest on them. On the other hand, I do wish to insist that the metaphor of 'foundation' is misleading in that it keeps us from seeing that if there is a logical dimension in which other empirical propositions rest on observation reports, there is another logical dimension in which the latter rest on the former, (p. 170)
The discussion in EPM, summarized in the previous paragraph, makes clear the way in which Sellars thinks that observation reports rest on other propositions, but not the way in which he thinks that they do not (i.e., the way in which others rest on them). The second and more serious respect is that no adequate support is given for the position. The author just lays it down that "to be the expression of knowledge, a report must not only have authority, this authority must in some sense be recognized by the person whose report it is (p. 168).18

Are other writings of Sellars more explicit in these two respects? The most systematic presentation of Sellars' general epistemology known to me is the third of the Matchette lectures, given in 1971 at the University of Texas and published under the general title of "The Structure of Knowledge" (SK) in Action, Knowledge and Reality: Critical Studies in Honor of Wilfrid Sellars, ed. H. N. Castaneda (Indianapolis: Bobbs-Merrill, 1975). There we shall find that though the first lack is filled, the second is not.19

In the third of these lectures, entitled "Epistemic Principles", he makes two distinctions between observation reports and, for example, the generalizations that are traditionally thought to be based on them. First (a point that was at least implicit in EPM), the former differs from the latter in being "non-inferential" in the sense that they are not, typically, arrived at on the basis of inference of any sort. They are formed "spontaneously" (pp. 324, 342). But this is not a difference in epistemic status, at least not according to Sellars' lights. It does not constitute a way in which observation reports "do not rest on other propositions in the same way as other propositions rest on them". Sellars spells out the distinctively epistemic difference as follows. The way in which other propositions rest on observation reports is given by the following schema:

I have good reasons, all things considered, for believing p;
So, p;
So, I have good reasons, all things considered, for believing q. (p. 335)
Here we are justified in believing q because it "can be correctly inferred, inductively or deductively, from other beliefs which we are justified in holding" (p. 336). But the way in which an observation report is justified is given by the following schema:
I just thought-out-loud 'Lo! Here is a red apple'
(no countervailing conditions obtain);
So, there is good reason to believe that there is a red apple in front of me.

Notice that although the justification of the belief that there is a red apple in front of (Jones) is an inferential justification, it has the peculiar character that its essential premise asserts the occurrence of the very same belief in a specific context. It is this fact which gives the appearance that such beliefs are self-justifying and hence gives the justification the appearance of being non-inferential, (p. 342)

Thus the respect in which an observation report does not rest on other justification of other beliefs on the same level. The beliefs that Jones must be justified in believing, in order that he be justified in believing B (that there is a red apple in front of him), are beliefs about B, that it occurred in certain circumstances that satisfy certain conditions. In fn. 12 on p. 342 Sellars refers to a passage in his essay "Phenomenalism", in which he says that the kind of credibility generated for B by the above schema is a "trans-level credibility" (Science, Perception, and Reality, p. 88).20

Thus the thesis that observation reports do not rest on other propositions, as Sellars understands that thesis, does not imply that they express immediate knowledge, as we have explained that notion. Sellars remains committed to the thesis that I know that there is a red apple before me only if I know the relevant facts about what gives my utterance its "authority".

But what about some reason for accepting this position? Here SK is less satisfactory, though the hints are broader than in EPM. For one thing, Sellars talks as if it is central to the concept of justification that it involves having reasons for the justificandum.

Presumably, to be justified in believing something is to have good reasons for believing it, as contrasted with its contradictory, (p. 332)

Is it not possible to construe 'I know that-p' as essentially equivalent to 'p, and I have reasons good enough to support a guarantee. . . '? (p. 333)

Against this background, the question:
If knowledge is justified true belief, how can there be such a thing as self-evident knowledge? And if there is no such thing as self-evident knowledge, how can any true belief be, in the relevant sense, justified? (p. 332)
ultimately on a distinction between two ways in which there can be, and one can have, good reasons for believing that-p. (p. 334)
That distinction is the one between same-level and trans-level reasons that we have just been discussing.

Well, if it is essential to the epistemic justification of a belief that the I believer have adequate reason for her belief, then there can be no immediate justification, and, if justification is necessary for knowledge, no immediate knowledge. But unless that claim is itself defended in some way, it is too close to the question at issue to advance the discussion. It is very close indeed; the principle of justification through reasons alone is precisely what the partisan of immediate knowledge is denying. For to have reasons for a belief is to have other knowledge or justified belief that supports the belief in question. And immediate justification is justification for which that is not required.

We may find something far enough back to advance the discussion, by considering the way in which Sellars hints that all justification is higher level in character. It always consists of showing, or of the capacity to show, that one's belief is justified, or reasonable, or that one has adequate reasons for it. Note that the two schemata of justification that were cited on p. 68 have as their conclusion not the proposition the justification of the belief in which is in question, but rather a higher level proposition to the effect that the subject has good reasons, or that there are good reasons, to believe the lower level proposition. And of the second schema Sellars says, "Like all justification arguments, it is a higher-order thinking" (SK, p. 342). One could wish the author to be more explicit, but this does suggest that. Sellars is thinking of epistemic justification in general as consisting of, or requiring, the capacity of the subject to produce adequate reasons for supposing that it is reasonable to believe the proposition justified.

If this is the case, then justification does require adequate reasons, for I couldn't have the capacity to produce adequate reasons without there being such reasons to produce. But why should we suppose that this is required for epistemic justification? We frequently take ourselves to know things with respect to which we have no such capacity. I often suppose myself to know that my wife is upset about something, where I would be hard pressed to specify how I can tell, that is, hard pressed to specify what makes it reasonable for me to believe this. The same goes for much of our supposed knowledge about history, geography, and physical regularities. In the face of all this, why should we accept the thesis that justification essentially involves the capacity to demonstrate reasonableness?

It is tempting to suppose that Sellars has fallen victim to the pervasive confusion between the activity of justifying a belief -- showing the belief to be reasonable, credible, or justified -- and a belief's being justified, where this is some kind of epistemic state or condition of the believer vis-a-vis that belief, rather than something he is or might be doing?' There are enough locutions that are ambiguous between these two to provide a spawning ground for the confusion. (The belief is justified.' 'What does it take to justify the belief?') One who has fallen into the confusion will realize, of course, that we can't require S to have actually gone through the activity of justifying B in order to be justified in accepting B. But if still in the toils of the confusion, he is likely to take it as obvious that at least S must be capable of justifying B in order to be justified in accepting B.22

But perhaps Sellars' higher level slant on justification has a more respectable origin. Perhaps he is simply exhibiting the widespread tendency of epistemologists to think of knowledge as the exclusive possession of critically reflective subjects, where being "critically reflective" essentially involves the tendency to ask, and the capacity to answer, questions as to what it is that justifies one's beliefs or makes them reasonable. If one has to be that kind of subject in order to have knowledge, then knowledge does require what Sellars says it does. But it seems clear that none of us satisfy that antecedent condition with respect to all our beliefs, and that many human subjects, and all lower animals, satisfy it with respect to few or none of their beliefs. An examination of the epistemic status of one's beliefs is a highly sophisticated exercise that presupposes a massive foundation of less rarefied cognitive achievements. Presumably epistemology is not limited to understanding the condition of philosophers and other choice spirits who have achieved a considerable ability in making explicit what it takes to render one or another sort of belief rational. It is, more generally and more basically, an attempt to understand the nature and conditions of such cognitive achievements as getting accurate information about the immediate environment through perception, one's awareness of what one is thinking or feeling at the moment, and one's recollection of what happened to one in the past. If terms like 'knowledge' are confined to the cognitive achievements of critically reflective subjects, we shall have to find a new term for the territory in its full extent.

The above should not be taken to imply, nor does it imply, that reflective knowledge of one's knowledge and of the epistemic status of one's beliefs is not valuable; nor does it imply that there are not important goals for the attainment of which it is necessary. It would seem to be required for answering skepticism, for being fully self-conscious about one's cognitive situation, and, more generally, for doing epistemology, an activity I am scarcely in a position to brand as pointless. But all this is quite compatible with the point just urged that one can genuinely have propositional knowledge without being capable, and especially without being fully capable, of a reflective assessment of that knowledge. We must not confuse epistemology with its own subject matter.


In his essay "Can Empirical Knowledge Have a Foundation?"23 Laurence Bonjour mounts an argument against immediate knowledge that displays many of the features of Sellars' attack. Let's consider whether Bonjour does any better by way of providing support for the crucial contentions of that attack.

Bonjour is concerned to show the impossibility of "basic beliefs", beliefs that are justified otherwise than by other justified beliefs, what we have been calling "immediately justified beliefs". The central argument runs as follows.

If basic beliefs are to provide a secure foundation for empirical knowledge, if inference from them is to be the sole basis for the justification of other empirical beliefs, then that feature, whatever it may be, in virtue of which a belief qualifies as basic must also constitute a good reason for thinking that the belief is true.24 If we let 'φ' represent this feature, then for a belief B to qualify as basic in an acceptable foundationist account, the premises of the following argument must themselves be at least justified;
(i) Belief B has feature φ.
(ii) Beliefs having feature φ are highly likely to be true.
Therefore, B is highly likely to be true.
. . . And if we now assume, reasonably enough, that for B to be justified for a particular person (at a particular time) it is necessary, not merely that a justification for B exist in the abstract, but that the person in question be in cognitive possession of the justification, we get the result that B is not basic after all since its justification depends on that of at least one other empirical belief, (pp. 5-6)

It is clear that this argument passes my test for a general argument against immediate knowledge. The argument is quite indifferent as to what the feature φ is. It could be "formulating the content of an immediate awareness" or "being a true self-presenting proposition" or "being formed by a reliable perceptual process" or what-you-will, and the argument will be just as strong, or just as weak.

As already indicated, I am not at all disposed to quarrel with the claim that premises (i) and (ii) must be true whenever B is immediately (or mediately) justified (and hence that the conclusion must be true as well since it is a valid argument). To admit so much is no more than to agree that any justifying feature must be "truth-conducive". But this is perfectly compatible with the existence of immediate knowledge. The premise "B is justified by virtue of having feature φ, which is truth-conducive" has no tendency to support "B is justified by the fact that the subject has adequate reasons for it". It is the further requirement that is the clinker: "For B to be justified for a particular person (at a particular time) it is necessary, not merely that a justification for B exist in the abstract, but that the person in question be in cognitive possession of it." In other words, in order that I be justified in accepting B, I must know, or be justified in believing, the premises of the above argument. And why should we suppose that? Again, unless some significant grounds are adduced, our opponent of immediate knowledge has done nothing more impressive than to affirm the contradictory.

Now Bonjour, like Sellars, roundly affirms that justification, in general, requires possession of adequate reasons by the subject (pp. 5, 7). And so, as in Sellars, when confronted with a putatively basic belief, we are driven to higher level reasons. But, again, this by itself is to repeat the position rather than to defend it. In Bonjour's article there is rather more ground than in Sellars for suspecting a confusion between justifying a belief and being justified in a belief. After enunciating "the traditional conception of knowledge as adequately justified true belief" he writes: "Now the most natural way to justify a belief is by producing a justificatory argument. . . ." The obvious suggestion is that "justified" in the conditions for knowledge means "having been the target of a successful activity of justifying", rather than, for instance, "it's being all right for the subject to hold it". He backs out of this in the next paragraph when he writes, "a person for whom a belief is inferentially justified need not have explicitly rehearsed the justificatory argument in question to others or even to himself", but he feels he is still left with the requirement that "the inference be available, to him if the belief is called into question by others or by himself. . . and that the availability of inference be, in the final analysis, his reason for holding the belief" (p. 2). And three pages later, after opining that "the very idea of an epistemically basic empirical belief is extremely paradoxical", he supports the opinion by writing: "For on what basis is such a belief to be justified, once appeal to further empirical beliefs is ruled out?" (p. 5; emphasis mine; see also "the first paragraph of p. 8).

However, Bonjour also has a way of defending the demand for reasons that is different from anything in Sellars, and we ought to consider that. In spelling out the concept of justification that is involved in his argument he writes:

Knowledge requires epistemic justification and the distinguishing characteristic of this particular species of justification is, I submit, its essential or internal relationship to the cognitive goal of truth. Cognitive doings are epistemically justified, on this conception, only if and to the extent that they are aimed at this goal -- which means roughly that one accepts all and only beliefs which one has good reason to think are true. To accept a belief in the absence of such a reason, however appealing or even mandatory such acceptance might be from other standpoints, is to neglect the pursuit of truth; such acceptance is, one might say, epistemically irresponsible. My contention is that the idea of being epistemically responsible is the core of the concept of epistemic justification, (p. 5)
Some of the transitions in this line of thought are unconvincing as they stand. Accepting "all and only beliefs which one has good reason to think are true" is by no means the same thing as aiming at the goal of truth, even if we modify the former to "accepting all and only beliefs that one takes oneself to have good reasons to think are true". To suppose it is obvious that they come to the same thing is to assume the anti-immediacy thesis that is at issue. But what I want to focus on at the moment is the support given this transition by what follows -- the conception of justification as epistemic responsibility.

To think of epistemic justification as amounting to epistemic responsibility is to treat the former as a normative concept, one that belongs to a circle of concepts that includes duty, obligation, blame, reproach, right, and wrong. Bonjour is thinking of being justified in believing that p as either having done one's epistemic duty in so believing, or as not having violated any epistemic duty in so believing. If we want to keep epistemic justification in line with other species of the genus, we will have to opt for the latter. What I am justified in doing is not always something I have an obligation to do, but it is always something that I am permitted to do, something the doing of which does not violate any obligations. To say that I am justified in taking a taxi to the airport (and charging it to my expense account) is not to imply that I have a duty to take a taxi, rather than a bus; it is only to imply that I am allowed to do so, that doing so does not violate any regulations. So let's say that, on a normative construal, S's being justified in believing that p amounts to S's not violating any epistemic obligation in believing that p.A

This pushes the question back to "Why should we suppose that one who believes that p without having adequate reason for supposing p to be true is violating any intellectual obligation?" If I have acquired a propensity to form perceptual beliefs in circumstances favorable to their truth, why suppose that I am violating some epistemic obligation by manifesting that propensity, where I don't have any good reason for supposing that the circumstances are propitious? Why wouldn't an acceptable set of epistemic norms permit me to form beliefs in that way? So far as I can see, Bonjour would have to reply as follows.

To be responsible in my doxastic decisions I have to make them in the light of the reasons available to me, for that is all I have to go on. Therefore what is required of me as a seeker after truth, as a cognitive subject, is that I decide between believing that p and refraining from that belief on the basis of whatever relevant reasons are available to me. To make the decision on any other basis or in any other way would be to flout my intellectual obligations. It would be "epistemically irresponsible"25

If this is the way the wind blows, then it shows, first of all, that Bonjour is assuming that obligations and the like attach directly to believing and refraining from believing, and hence that he is assuming believing and the reverse to be under voluntary control. "Ought implies can." He is assuming that, with respect to each candidate for belief, the subject has a choice as to whether or not to believe it. This voluntaristic version of a normative conception of justification can be contrasted with an nonvoluntaristic version according to which belief is not, either in general or ever, under voluntary control, and intellectual obligations attach rather to the various things people can do (voluntarily) to affect their belief-forming process.B Second, even granted the voluntarism, Bonjour's demand for reasons would not be supported by a severely objectivist version, on which a believing's being in accord with my obligations is simply a matter of whether that believing is in fact in violation of any obligation, whatever I believe, know, or justifiably believe about the matter. If one of my obligations is to refrain from a perceptual belief if the conditions of perception are abnormal, then whether I violated that obligation in believing that p would be a matter of whether, in fact, the conditions were abnormal, not on whether I believed, knew, or justifiably believed that the conditions are abnormal. On that version justification hangs on the way things are, rather than on what reasons I have that bear on the question. To squeeze a universal demand for reasons out of the concept of justification. Bonjour will have to be using a more subjective version of a voluntaristic normative conception, according to which one has satisfied one's obligations in a belief iff one knows or is justified in believing that the objective requirements have been satisfied. On that reading it will be the case that one is proceeding as one ought in believing that p only if one has adequate reason for supposing that q, where q amounts to whatever is required by the relevant (objective) epistemic obligations.C

Thus we have found one not disreputable ground for the universal demand for reasons. But however respectable, the subjective-voluntaristic-normative conception of justification is not immune from criticism, especially as regards the claim that justification in this sense is a necessary condition of knowledge. I myself am disinclined to allow that justification on any normative conception is necessary for knowledge. The reason for this is as follows. Normative conceptions like obligation and reproach apply only to beings that are capable of governing their conduct in accordance with norms, principles, or rules. It is for lack of this capacity that we refrain from using such concepts in application to very small children and lower animals. But surely these creatures are not devoid of knowledge. Both infants and dogs acquire knowledge about their immediate physical environment through perception. If Bonjour denies this last claim we have an opposition quite similar to the earlier opposition between Sellars and myself as to whether subjects should be credited with knowledge only to the extent that they are capable of critical reflection on the epistemic status of their beliefs.

But even if we employ some sort of normative conception of justification, there are strong objections to a voluntaristic version thereof. It seems clear that belief is not, in general, under direct voluntary control. When I seem to myself to see a truck coming down the street, or when I am in any of the innumerable situations, perceptual and otherwise, where it seems obvious to me that something is or is not the case, I do not have the capacity to believe or refrain from believing at will, as I choose. If in the above situation I were to set myself to retrain from believing that a truck is coming down the street, perhaps in order to prove to myself that I can, I wouldn't know how to begin. I wouldn't know what button to push. (Of course, I can undertake a regimen that is designed to gradually wean myself away from reliance on the senses; but even if I should succeed in this, that is a different story. There are many things not themselves under direct voluntary control that I can affect by what I do, e.g., my health and my wealth.) Whether I can ever believe at will is a matter I will not go into. However that question is resolved, it is clear that belief is not always, or even generally, a matter of choice. Hence a conception of justification that presupposes voluntary control of belief cannot be applied to belief in general. On that construal, justification cannot be a general requirement for knowledge.26

Thus in Bonjour, as in Sellars, the contention that putatively immediate knowledge really rests on higher level reasons itself rests on a foundation of sand.27,28


A. For an extended discussion of normative conceptions of epistemic justification (there termed "deontological"), see Essay 5.

B. Again, see Essay 5, Sections VI and VII, for details.

C. See Essay 4, Section III, where various "modes" of epistemic obligation (and other sorts of obligation and justification) are distinguished. There the one under consideration at this point is called a "cognitive" mode.


1 'Bernard Bosanquet, Logic or the Morphology of Knowledge (London: Oxford University Press, 1911), bk. 2, chap. 9. F. H. Bradley, Essays on Truth and Reality (Oxford: Clarendon Press, 1914), chap. 8. Bradley, The Principles of Logic, 2d ed. (London: Oxford University Press, 1922), Terminal Essay II. Brand Blanshard, The Nature of Thought (London: George Allen & Unwin, 1939), chap. 25-28.

2 C. S. Peirce, "Questions Concerning Certain Faculties Claimed for Man," in Collected Papers, ed. C. Hartshorne and P. Weiss (Cambridge: Harvard University Press, 1934), vol. 5. John Dewey, Logic: The Theory of Inquiry (New York: Henry Holt, 1938), chap. 8.

3 D. M. Armstrong, Belief, Truth, and Knowledge (London: Cambridge University Press, 1973)> chap. 12-15. Alvin I. Goldman, "What Is Justified Belief?", in Justification and Knowledge, ed. G. S. Pappas (Dordrecht: D. Reidel, 1979). Marshall Swain, Reasons and Knowledge (Ithaca: Cornell University Press, 1981).

4 Goldman, "What Is Justified Belief?"; Swain, Reasons and Knowledge.

5 I will continue to use the term 'justification' when discussing epistemologists who think of knowledge in those terms. I shall use 'epistemization' when I am striving for maximum generality.

6 The inappropriateness of the request for reasons here has moved some to deny that this is a case of knowledge. That move, I believe, would have to be defended with the same arguments we shall be criticizing in the body of the paper. Since these arguments are directed against the possibility of immediate knowledge, they can be used either to discard the immediacy and keep the knowledge, or to discard the knowledge and keep the immediacy.

7 See, e.g., Roderick Firth, "Coherence, Certainty, and Epistemic Priority," Journal of Philosophy, 61 (1964) 545-77. This should not be confused with prima facie immediate justification, where the justification, when it comes off, is wholly immediate, but where the justification could be "pyerridgn'' or "defeated" if conditions are not propitious.(John Pollock, Knowledge and Justification [Princeton: Princeton University Press, 1974], chap. 2; Roderick M. Chisholm, Theory of Knowledge, 2d ed. [Englewood Cliffs, N.J.: Prentice-Hall, 1977], chap. 4.) Thus one might take the perceptual belief that there is a tree in front of one to be prima facie justified merely by one's having a certain visual experience; then if conditions are abnormal in a certain way that justification is "overridden". Here it is not required for justification that one have one or more other justified beliefs related in a certain way to the target belief.

8 For an influential recent attack see Wilfrid Sellars, "Empiricism and the Philosophy of Mind," in Science, Perception, and Reality (New York: Humanities Press, 1963).

9 See, e.g., Bruce Aune, Knowledge, Mind, and Nature (New York: Random House, 1967); Keith Lehrer, Knowledge (London: Oxford University Press, 1974); F. L. Will, Induction and Justification (Ithaca: Cornell University Press, 1974).

10 See, e.g., Georges Dicker, Perceptual Knowledge (Dordrecht: D. Reidel, 1980), chap. 1.

11 For the distinction between these terms, see Essay 10.

12 See, e.g., Aune, Knowledge, Mind, and Nature, chap. 2 and Will, Induction and Justification, chap. 7; and for a response see Essay 2 in this volume.

13 In Will, Induction and Justification, p. 203, there is a passage that might be interpreted in this way.

14 See Firth, "Coherence, Certainty, and Epistemic Priority."

15 Note that (IV) does not read: "S's belief that p is immediately epistemized -- S's belief is epistemized only by something other than some relation this belief has to some other epistemized belief(s) of S." The 'only' was omitted specifically to allow for the possibility that the belief might also (contemporaneously) be mediately epistemized. A parallel point holds for (III).

16 No doubt, Sellars' best known sally in this arena is his attack on "giveness" and the idea that foundational beliefs are justified by virtue of formulating what is given in a nonpropositional cognitive act. But because of my limitation to general arguments against immediate knowledge, I will not be discussing that aspect of his polemic.

For different reactions to the Level Ascent argument see Ernest Sosa, "The Raft and the Pyramid," Midwest Studies in Philosophy 5 (1980), where it is called the "Doxastic Ascent" argument, and R. G. Meyers, "Sellars' Rejection of Foundations," Philosophical Studies, 39 (1981), 61-78.

17 In speaking of Firth, "Coherence, Certainty, and Epistemic Priority," he refers to "one aspect of his enterprise, which is, as I would put it, to reconcile as far as possible the claims of those who stress warrantedness grounded in explanatory coherence (among whom I count myself) with the claims of those who stress the non-inferential warrantedness of certain empirical statements (among whom I also count myself)". "More on Givenness and Explanatory Coherence," in Justification and Knowledge, ed. G. S. Pappas (Dordrecht: D. Reidel, 1979), p. 174.

18 To be sure, Sellars prefaces this remark with "For we have seen that. . .", but it is not clear to me just where in the essay he supposes it to have been seen. Perhaps he was thinking of this passage: "Statements pertaining to this level, in order to 'express knowledge' must not only be made, but, so to speak, must be worthy of being made, credible, that is, in the sense of worthy of credence. Furthermore, and this is a crucial point, they must be made in a way which involves this credibility. For where there is no connection between the making of a statement and its authority, the assertion may express conviction, but it can scarcely be said to express knowledge" (p. 164). If this is intended to be an argument for the crucial claim quoted above from p. 168, then I will have to retract my statement that Sellars "just lays it down". But if this be support, it is quite inadequate to the task. I will agree that knowledge requires some connection between "the making of a statement and its authority", i.e., in this case, between the making of a statement and the fact that it was made in circumstances in which it is likely to be true. A merely accidental concatenation of the two would be a case in which it was just a matter of luck that the statement was true, and being right by accident is not knowledge. But, and this is the crucial point, Sellars' candidate for the connection is not the only possibility. Sellars thinks that if there is to be a "connection", it will have to be a relatively sophisticated one in second intention; it will have to be that the speaker makes her statement in recognition that the circumstances are propitious for its truth. But there is a humbler candidate, the one that is already built into the initial suggestion that Sellars thinks we must go beyond, viz., that the statement "is a manifestation of a tendency to produce overt or covert tokens of 'This is green' -- given a certain set -- if and only if a green object is being looked at in standard conditions" (p. 167). That is, the mere fact that the particular utterance is a manifestation of a general tendency to make such utterances only in truth-conducive circumstances is itself a "connection between the statement and its authority" that removes the case from the class of lucky guesses or accidental hits; and this is true whether or not the speaker knows that the circumstances are propitious. What we need from Sellars is a reason for thinking that this simpler "connection" is not enough, and that the higher-level-knowledge connection is required for knowledge of the lower level proposition.

19 In another prime source for Sellars' general epistemology, "More on Givenness and Explanatory Coherence," he assumes that it is reasonable to accept introspective, perceptual, and memory (IPM) judgments only because it is reasonable to accept the higher level judgment that IPM judgments are generally true (pp. 177, 178, 180). But in that article the focus is on what it takes to be justified in those higher level judgments, and as a result the claim about what it takes to be justified in IPM judgments is not even discussed, much less adequately supported.

20 In scrutinizing the above schemata one may be struck by the fact that in the second schema, unlike the first, the premises make no reference to the justification of any other beliefs of the subject. And from this one may infer that Sellars supposes that the justification of observational beliefs depends in no way on the justification of other beliefs. But this is not Sellars' position. He is committed to holding both that an observational belief can be correctly inferred from the premises of the appropriate schema of the second sort, and that such a belief is justified only if the believer knows, or is justified in believing, those premises. "(See, e.g., p. 342.)

21 "The essential point is that in characterizing an episode or state as that of knowing, we are not giving an empirical description of that episode or state; we are placing it in the logical space of reasons, of justifying and being able to justify what one says." (EPM, p. 169)

22 Although I am perhaps too much given to seeing instances of another confusion, a level confusion between, e.g., being justified in believing that p and being justified in believing that one is justified in believing that p (Essays 2 and 6), I can hardly find Sellars guilty of this charge, in view of what we have already noted to be his clear recognition of the distinction. It is worthy of note, though, that if one did fail to make the distinction, as many epistemologists do, this could easily lead one to the Level Ascent argument. For, clearly, in order to be justified in the higher level belief that one is justified in the lower level belief, one must have reasons that have to do with the epistemic status of that lower level belief. And so if one fails to distinguish the two justifications, one will automatically take it that such reasons are required for being justified in the lower level belief. It may be that such a confusion plays a role in Bonjour's position, to be discussed below, but I will not pursue that possibility.

23 American Philosophical Quarterly, 15 (1978), 1-13.

24 Bonjour supports this claim, cogently in my opinion, as follows. ", . . knowledge requires epistemic justification, and the distinguishing characteristic of this particular species of justification is, I submit, its essential or internal relationship to the cognitive goal of truth. . . . A corollary of this conception of epistemic justification is that a satisfactory defense of a particular standard of epistemic justification must consist in showing it to be truth-conducive, i.e., in showing that accepting beliefs in accordance with its dictates is likely to lead to truth (and more likely than any proposed alternative)" (p. 5).

25 In considering the reliabilist position that S knows that p provided S has a true belief that p that was formed in a reliable manner (whether or not S knows it to be reliable), Bonjour writes: "But P himself has no reason at all for thinking that B is likely to be true. From his perspective, it is an accident that the belief is true. And thus his acceptance of B is no more rational or responsible from an epistemic standpoint than would be the acceptance of a subjectively similar belief for which the external relation in question failed to obtain" (p. 8).

26 For the concepts of justification mentioned here, as well as others, see Essays 4 and 5.

27 The considerations of Sections III and IV can also be used against the position Michael Williams takes in his book Groundless Belief (New Haven: Yale University Press, 1977). I have not explicitly discussed his arguments in the body of the paper, for he does not squarely oppose the possibility of any immediate knowledge as does Sellars and Bonjour. He recognizes the possibility that, e.g., perceptual beliefs might be justified just by virtue of having been reliably formed, even if the subject knows nothing about that (p. 69). But he holds that if that is the whole story, such beliefs do not meet the foundationalist's requirements since a "potential infinite regress of justification" has not been closed off (p. 69). This is because empirical facts will have to be produced to justify the supposition that the perceptual beliefs in question were reliably produced. "To say that there is an empirical presumption in favor of beliefs of a certain kind being true is to trace the prima-facie credibility of these beliefs to further general facts and thus to lead ourselves back into the very regress from which intrinsically credible beliefs are supposed to liberate us" (p. 76; see also pp. 158-61). Of course, if someone makes the higher level statement that certain perceptual beliefs are reliably produced and therefore credible, he will need reasons for that statement and he will not be at the terminus of a regress of justification. But that does not imply that the perceptual believer in question needs reasons to be justified in holding his first-level perceptual beliefs, and hence it doesn't imply that he is not at the terminus of a regress of justification. (See Essay 1) Williams does nothing to support that claim, and if he were to support it with considerations of the sort deployed by Sellars and Bonjour, the same responses would be in order.

28 This paper has profited greatly from comments by Robert Audi and Jonathan Bennett.