I. Other Persons

Consider our beliefs about other persons. Consider, in particular, beliefs ascribing specifically personal states to them: being appeared to redly, or believing that Moscow, Idaho, is smaller than its Russian namesake, or intending to go to graduate school, or being confident about the outcome of a biopsy. My general questions are, Do these beliefs have warrant for us, and if so, How do they get it? As a matter of fact I am less interested in the first question than the second; like the rest of us, I take it for granted that they do indeed have warrant for us, warrant that is sometimes sufficient for knowledge. So the question is, How does this warrant acquisition work? How do our beliefs ascribing mental states to others acquire warrant? There are substantially three answers lurking in the contemporary woods: that they get warrant by virtue of (1) an analogical argument, (2) being or being like scientific theories or hypotheses for each of us, and (3) Wittgensteinian criteria. I shall consider each of these in turn, arguing that none is correct and suggesting an answer that is both closer to the truth and more in line with the general theory of warrant I propose.

A. The Analogical Position

1. Mental State Ascribing Beliefs as Basic

Now the traditional answer in the dominant epistemological tradition going back to Descartes and Locke is this: warrant acquisition works via some kind of analogical argument. Roughly speaking, I note correlations between my own bodily behavior (broadly construed) and my own inner or mental states; I therefore suppose that these correlations hold for others as well, thus ascribing mental states to them; this conclusion therefore has for me the warrant enjoyed by any proposition that is the conclusion of an inductive argument.

Thus John Stuart Mill:

By what evidence do I know, or by what considerations am I led to believe, that there exist other sentient creatures; that the walking and speaking figures which I see and hear, have sensations and thoughts, or in other words, possess Minds? … I conclude it from certain things, which my experience of my own states of feeling proves to me to be marks of it… I am conscious in myself of a series of facts connected by an uniform sequence, of which the beginning is modifications of my body, the middle is feelings, and the end is outward demeanour. In the case of other human beings I have the evidence of my senses for the first and last links of the series, but not for the intermediate link… Experience, therefore, obliges me to conclude that there must be an intermediate link.¹

Mill apparently believes that as a matter of fact human beings (or at any rate Mill himself) are “led to believe” in others by analogical arguments. Here he might have done well to heed an anticipatory scoff by Thomas Reid:

No man thinks of asking himself what reason he has to believe that his neighbour is a living creature. He would be not a little surprised if another person should ask him so absurd a question: and perhaps could not give any reason which would not equally prove a watch or a puppet to be a living creature. But, though you should satisfy him of the weakness of the reasons he gives for his belief, you cannot make him in the least doubtful. This belief stands upon another foundation than that of reasoning and therefore, whether a man can give good reasons for it or not, it is not in his power to shake it off.²

Surely Reid is right: surely none of us actually comes to these beliefs in this sort of way. I look at Paul and say to myself “Oh, oh, he's furious again—what have I done this time?” thus forming the belief that he is furious again. Do I form this belief by way of a quick but tacit induction, an application of an analogical argument from premises involving the proposition that he looks a certain way, and when I look that way I am ordinarily furious? Clearly not. First (in analogy with the perceptual case [see pp. 93ff.]), it seems that I don't ordinarily form any belief (any explicit belief, anyway) at all as to how Paul is looking: I move directly to the view that he is furious. Perhaps I could form such a belief: but typically I don't. Obviously I don't form a belief consisting in a qualitative³ description of how Paul looks and sounds (“his brow is knit; his eyes are narrowed to slits; his mouth is wide open; loud noises of such and such timber and pitch emanate therefrom”). And even if I did form such a belief, it would be far too crude to play the role of a premise in a decent analogical argument: any such description would fail to distinguish the way he looks from a thousand other ways that do not warrant the belief that he is furious. Of course, I could form what we could call for want of a better term a haecceitistic or quidditative⁴ belief: I take a quick look at you; you look out of sorts; I look again, quickly but carefully, and form a belief I express to myself by saying, “Look out, she's got that look again” (where what I believe is not a qualitative proposition describing a certain look, but instead the sort of proposition that gets expressed by a sentence containing a demonstrative term); but in the typical case I don't do so.

Furthermore, I ascribe to others a wide variety of mental states, making fine and subtle discriminations between rather similar states, often ascribing to them states I have seldom if ever experienced myself; how could I do this on the basis of simple analogical reasoning from correlations between behaviour and mental states in my own case? As a matter of fact, much of the relevant behavior is such that I can't observe it in my own case: facial expression, for example, is extremely important,⁵ and I typically can't observe what sort of facial expressions I am presenting to the world. Of course we have mirrors: but our ancestors, prior to the advent of mirrors, no doubt sometimes knew that someone else was angry or in pain. And we ourselves form these beliefs without adverting to mirrors; who among us carries one with him, or (when in the grip of strong emotion) remembers to consult it in order to establish correlations between his mental states and his facial expressions?

Accordingly, even if we do form beliefs about the mental states of others on the basis of what we know or see about their bodies and behavior, we clearly don't do so by way of an ordinary inductive inference from what we have observed in the past about the connection between such bodily states and such mental states in our own case. We don't form such beliefs on the basis of an analogical argument. None of us remembers ever having done that; small children apparently form beliefs about the mental states of their parents long before they come to the age at which they make inductive inferences. The capacity for this sort of belief formation is not something one gains by inductive learning but is part of our native and original cognitive equipment.⁶ Of course, it is always open to someone to claim that the inferences in question are there, all right, but carried on at a subconscious level—just as the Freudian can claim, your indignant protests to the contrary notwithstanding, that you have always hated your father for preventing you from enjoying your mother's sexual favors. There is about as much plausibility in the one suggestion as in the other; but in any event I am willing to scale my claim down to the claim that we don't form beliefs about others on the basis of conscious inferences from conscious beliefs about their bodies and behavior. If our mental-state-ascribing judgements are not basic, if we do form beliefs ascribing mental states to others on the basis of beliefs about their bodily states, then at any rate it is not by way of ordinary inductive reasoning; there will be some other connection between the beliefs ascribing behavior and the beliefs ascribing mental states.

2. Analogical Arguments Examined

Here, no doubt, the analogical arguer will remind us that the genetic issue isn't the issue. The question is not, Are these beliefs typically taken the basic way? That is as may be; the question is whether these beliefs have warrant; and if they do, how do they get it? And the analogical position is that these beliefs have warrant because analogical arguments are available to us, even if in fact hardly anyone takes the trouble to form beliefs about others on the basis of such arguments. The analogical argument for other minds is not nowadays discussed with the fervor it was a couple of decades back. It is widely accepted (and accepted by first-rate epistemologists⁷), however, and, as we shall see in the next section, the sort of position that seems to have assumed its mantle is no improvement; so suppose we briefly look into its credentials.⁸ It seems immediately apparent, first, that the strength of our convictions about the mental states of others is entirely out of proportion to the strength of the available analogical arguments. We all firmly believe, for example, that childbirth is often extremely painful, whether we have borne children or not; and the strength of our conviction is nowhere nearly matched by the strength of the relevant analogical arguments.

But do those analogical arguments have any appreciable degree of strength at all? A version of the argument may perhaps be stated initially as follows. First, I note a correlation between certain states of my body and certain of my mental states; then I argue as follows:

(1) Whenever this body (that is, my body) is in state S, there is a person whose body it is and who is (for example) angry.

I then observe that

(2) B over there (a human body distinct from mine) is in state S and conclude that

(3) B is a body of some person who is angry.

Since I am not angry, I conclude there is a person whose body is B, who is distinct from me, and who is angry.

Now the initial problem with this argument, of course, is that it appears to have an appallingly small sample class: its only member is my own body. According to a popular response, however, this objection is hasty:

The objection that one is generalizing from a single instance can perhaps be countered by maintaining that it is not a matter of extending to all other persons a conclusion which has been found to hold for only one, but rather of proceeding from the fact that certain properties have been found to be conjoined in various circumstances. So the question that I put is not: Am I justified in assuming that what I have found to be true only of myself is also true of others? but: Having found that in various circumstances the possession of certain properties is united with the possession of a certain feeling, does this union continue to obtain when the circumstances are still further varied? The basis of the argument is broadened by absorbing the difference of persons into the difference of the situation in which the psycho-physical connections are supposed to hold.⁹

But this reply doesn't really help; if anything, it exacerbates matters. The chief problem with the argument thus revised is that there are parallel arguments of the same apparent strength for the denials of many of our most cherished beliefs about other minds.

We can see the problem as follows. As an analogical arguer, I don't initially know much about the general connections between bodies, behavior, and mental states. I don't know such things, for example, as that I feel pain only in this body, that no one else feels pain there, and that pain in bodies other than my own need not be connected with my behavior; these are the sorts of things I am supposed to learn from an analogical argument. The premises of a proper analogical argument, therefore, will be limited to what I know from introspection about my own mental states, what I know by perception about bodily states on the part of my body and other bodies, certain necessary truths, and what I can deduce from propositions of those sorts. As for such propositions as when a person feels pain, her body will ordinarily display pain behavior or each person ordinarily has just one body, these must be established by way of analogical arguments. Now it is obvious that I cannot determine by introspection or observation that a certain area of your body—your arm, for example—is free from pain, is such that no sentient creature feels a pain in it. The best I can do is note that I feel no pain there; but that does not show that no one does. And the same really goes for my own arm. As Wittgenstein quite properly observed, it is possible (in the broadly logical sense) that someone else should feel pain in my arm. Of course, we don't ordinarily think that others do feel pain in my arm, but that is precisely the sort of contingent truth that (on the analogical position) must be established by way of analogical arguments. I therefore cannot determine by introspection that no one feels a pain in my arm: as in the case of your arm, all I can do, along these lines, is to determine that I don't feel a pain there.

The unfortunate fact, however, is that there is an analogical argument for the conclusion that every pain is in my body! For every pain such that I have determined (by introspection and perception), whether or not it was in my body, was in fact in my body. (I have never felt a pain in any other body.) I therefore conclude, and conclude quite properly, that probably every pain is in my body. You might object that as a matter of fact we don't think it possible that I should feel pain anywhere but in my body, so that the fact that I don't feel pain elsewhere doesn't prove much. No doubt you are right; on the analogical position, however, this fact is one we must learn by virtue of analogical argument; we therefore cannot use it to impugn this use of an analogical argument. Further, there is an analogical argument, for me, for the conclusion that every case of pain is accompanied by pain behavior on the part of my body (whenever there is pain, I engage in pain behavior): for (supposing, anyway, that I am a demonstrative sort) every case of pain such that I have determined whether it was accompanied by pain in my body was thus accompanied. I also have an argument for the proposition that whenever my body displays pain behavior, I feel pain; putting these two together, I conclude that whenever any sentient being feels pain, I feel pain—a nasty state of affairs indeed. I therefore have analogical arguments for such conclusions as that every pain is felt in my body and is accompanied by pain behavior on the part of my body, as well as for the doleful conclusion that whenever any sentient being feels pain, so do I. Of course, there will be similar arguments for such other mental states as anger, belief, and the like.

True, there will also be analogical arguments for the denials of these conclusions: but that just points up the weakness of the analogical position. Thus, for example, I have observed (in my own case) that pain behavior on the part of a body is accompanied by pain in that body: I note that B (a body other than my own) is displaying pain behavior; I conclude that this behavior is accompanied by pain in B (that is, that some sentient being is feeling a pain in B); since I feel no pain there, I conclude that sometimes there are pains that I don't feel. The problem is that there are analogical arguments both for the views we ordinarily hold about the general connections between mental and bodily states and other persons and their mental states, and for the denials of these views. The problem is that if the analogical position supports those ordinary views, it also supports their denials; hence it does not support those views. More irenically and more moderately, it does not support those views nearly strongly enough to justify the strength of our convictions with respect to them.

B. Beliefs about Others as Scientific Hypotheses

At present the analogical position is less popular than a sort of generalization of it: the view that our beliefs about other persons and their mental states are, or are like, scientific hypotheses for each of us. Each of us believes a host of propositions about other persons and about the connections between mental states, bodies, and behavior; and the idea is that these beliefs acquire the warrant they have by way of being, for each of us, elements of scientific theories. Each of us observes bodily behavior on the part of his own body and that of others; each of us also observes some of his own mental states.¹⁰ Then, remarkably enough, each of us forms the conjecture (perhaps by way of a sort of inference to the best explanation) that perhaps there are other persons with mental states more or less like his own.

Now as an account of how we actually come to or form our beliefs about others, this is no more plausible than the claim that we come to such beliefs by way of analogical arguments. A child's belief, with respect to his mother, that she has thoughts and feelings, is no more a scientific hypothesis, for him, than the belief that he himself has arms or legs; in each case we come to the belief in question in the basic way, not by way of a tenuous inference to the best explanation or as a sort of clever abductive conjecture. A much more plausible view is that we are constructed (‘hardwired’ to use the current buzz word) in such a way that these beliefs naturally arise upon the sort of stimuli (being spoken to in ‘motherese’, for example) to which a child is normally exposed. But perhaps here, as in the case of the analogical position, the claim is not a broadly scientific claim about the genesis of these beliefs, but an epistemological claim about their warrant; and the claim is that they get the warrant they have by way of abductive inference, by way of being, for each of us, part of such a scientific theory.

But this can't be right—not, at any rate, if we know, as we think we do, that there are other persons who hold beliefs, are depressed, angry, exultant, who suffer pains, make plans, hope for the best, believe in God and a thousand other things. For clearly any such inference to the best explanation would be extremely tenuous. There are plenty of other explanatory hypotheses equally simple or simpler: for example, that the only minds or selves are mine and that of a Cartesian demon who, for reasons of his own, gives me these strong inclinations to believe in the existence of others. Alternatively, the only mind is mine; but by way of wishful thinking or projection or some other mechanism I come to believe that there are many other centers of consciousness like myself. These suggestions seem utterly fantastic; only a madman would believe them. But that is not because they are enormously inferior as scientific hypotheses to the views we do in fact hold; they are not. It is instead because, again, we are so constructed that the natural view, for us, is that there are others like ourselves. This view, then, is also mistaken; if our beliefs about others acquired the warrant they have by way of this abductive inference, they would have little warrant indeed.

C. Criteria

But of course there is no real reason to assume that if we do have knowledge of other minds, it must be by way of the traditional analogical argument or else by way of a sort of scientific inference to the best explanation. (Such an assumption is no more warranted than the corresponding assumption that if we have knowledge of ordinary physical objects—tables, chairs, houses, mountains—it must be by way of some sort of inference from propositions about our own mental states to these conclusions.) There are other possibilities: for example, there is the Wittgensteinian suggestion that “inward processes stand in need of outward criteria.”

Wittgenstein's notion of criterion has not wanted for critical scrutiny over the past three or four decades; it remains uncomfortably murky. I don't here have the space to examine the notion with the requisite care¹¹or recapitulate the critical scrutiny; let us simply note its most prominent features. “If Mill has no criterion for the existence of feelings other than his own,” said Norman Malcolm, “then in that sense he does not understand the sentence ‘That human figure has feelings.’” On the other hand, “If he had a criterion he could apply it, establishing with certainty that this or that human figure does or does not have feelings (for the only plausible criterion would lie in behavior and circumstances that are open to view) and there would be no call to resort to tenuous analogical reasoning that yields at best a probability.”¹² Here the suggestion is that (1) if behavior of a certain sort (call it pain behavior) is indeed a criterion for pain, then my observing pain behavior on the part of a person S tends to make evident for me the proposition that S is in pain. (Tends to make this evident—this tendency could be overridden by other circumstances.) (2) Although pain behavior on the part of S tends to make evident for me the proposition that S is in pain, this is not by way of analogical or inductive reasoning on my part. (3) My grasping that pain behavior is in this way a criterion for pain is a necessary condition of my understanding the sentence ‘That human figure has feelings’; more generally, it is only because of this criterial connection, whatever exactly it is, that we are able to grasp the idea of others' having mental states, and can ascribe those mental states to them.

In explaining the “sense of ‘criterion’ which has become current through the influence of Wittgenstein's late work,” Sydney Shoemaker suggested something similar:

If so and so's being the case is a criterion for the truth of a judgment of identity, the assertion that it is evidence in favor of the truth of the judgment is necessarily (logically) rather than contingently (empirically) true. We know that it is evidence, not by having observed correlations and discovered empirical generalizations, but by understanding the concept of a φ … The search for the criteria for the truth of a judgment is the search for necessarily true propositions asserting that the existence of certain phenomena or states of affairs is evidence of the truth of that judgment.¹³

Here the idea, I think, is as follows: where we come to learn via induction or abduction that As are associated with Bs, then we quite properly take the presence of an A as evidence that a B is present. We take smoke as evidence of fire; or to use an example of Reid's, on the basis of past experience we take a certain crunching gravelly sound to be evidence that a coach is coming up the drive. But of course it would be wrong to take smoke as evidence of fire if (ignoring testimony for the moment) we had not observed this connection between smoke and fire. We might say, here, that it is only contingent that smoke is evidence of fire; it is evidence only because (as we have discovered) there is a correlation (a correlation that is itself contingent) between the presence of the one and the presence of the other; in the absence of such a correlation the one would not be evidence for the other. Where the connection between the phenomena is criterial, however, things go differently: the criterion is evidence for that of which it is a criterion, whether or not we have discovered any such correlations. Indeed, in the typical case, there will not be any way of discovering such a correlation without using the criterion. So, says Shoemaker, that the criterion is evidence for the relevant proposition is a necessary truth rather than a contingent truth.¹⁴

Malcolm and Shoemaker (and Albritton and Strawson) apparently concur on the following. (1) For many mental states S, there are behavior-and-circumstances B that together constitute criteria: good (but defeasible) evidence for the mental states in question, so that in many cases we ascribe S to someone, and have warrant for so doing, on the basis of our awareness of B. (2) This evidence does not proceed by way of our observing a correlation between B and S. More generally, the warrant these ascriptions have for us is not by way of induction or abduction or analogical argument. Even if it is not possible to learn of the existence of these correlations without making use of these criteria, these ascriptions of mental states to others still have warrant for us.

So far Malcolm et al. seem to be quite correct; there are criteria or something like them, and the connection between them and the mental states for which they are evidence is not merely inductive or theoretical. They go on to claim, however, that the connection between B and S must somehow be logical. It must somehow be a necessary truth that B is evidence for S, that is, that someone who is aware of B has evidence for S; that is, that someone who is aware of B and forms the relevant belief about S (and has no undefeated defeaters for that belief) is such that the belief in question has warrant for him. And here there is a most interesting confluence between the Wittgensteinian criteriology rampant thirty years ago and post-classical Chisholmianism (see chapter 3 of my Warrant: The Current Debate): the former is best seen as a special case of the latter. The essence of post-classical Chisholmianism is the suggestion that warrant is to be explained in terms of the fact that certain epistemic pairs (that is, pairs whose first member is a belief B and whose second member is an evidence-base, a maximal psychological property diminished with respect to B¹⁵) simply have more by way of intrinsic value than others; and the greater this intrinsic value, the greater the degree of warrant enjoyed by B for a person whose total evidence is the second member of the pair. Chisholm (and Brentano) hold, plausibly enough, that these relations of greater and less value among the pairs will be necessary: <B_i, P_x> has more intrinsic value than <B_j P_y> will be necessarily true if true at all. The criteriological suggestion can be seen as a special case of this post-classical Chisholmian position. Put initially and crudely, the criteriological position goes as follows: consider those epistemic pairs whose evidence bases include awareness of pain behavior (and no undefeated defeaters for the ascription of pain) and where the first member of the pair is a mental-state-ascribing belief S: such epistemic pairs have a great deal of intrinsic value—so much that S has a good deal of warrant. And because this is really a claim about the relative intrinsic values of various evidence pairs, it is a necessary truth.

But isn't this a dubious contention? Why suppose it necessary that someone who is aware of pain behavior has evidence for the relevant ascription of pain? Perhaps those who make this claim are thinking along the following lines. Suppose we deny that the warrant-inducing connection between the behaviorial state of affairs B that serves as the criterion and the mental state ascribing belief S is inductive or analogical (or abductive); that is, suppose we deny that there is a good inductive or analogical or abductive argument for S from B; but suppose we also hold that we do believe S on the basis of B (and that when we do so, our mental state ascribing belief S has warrant). How can this be unless there is a logically necessary connection between B and S? If the connection is not inductive or analogical or abductive, what other possibilities are there?

Whatever the reason, however, the claim seems wrong. It is in fact true of us human beings that beliefs ascribing mental states to others have warrant for us, and don't have it simply by way of analogical or inductive or abductive evidence; it is in fact true that there are mental state ascriptions and behavior-cum-circumstance pairs such that the latter constitute evidence, for us, for the former. That is, I may have warrant for the ascription of a given mental state S by way of being aware of a certain behavior-cum-circumstance B even if it isn't possible to make a good (noncircular) analogical or inductive or abductive argument from B to S. But it isn't a necessary truth that B and S are correlated. We can see this as follows. First, quite different sorts of behavior could have been correlated with pain (or with anger or fear). When afflicted with pain, human beings cry out, or moan, or whine, or stoically grit their teeth; but no doubt we could have been so constructed that we would instead smile, or do a little dance, or stand on our heads. So suppose first that these correlations had in fact been quite different. Now of course properly functioning human beings find themselves inclined, when aware of B, to make the S ascription; but suppose further that we had been so constructed that (when functioning properly) we did not find ourselves inclined to make the S ascription upon being aware of B. Then, surely (under those conditions), it would not have been the case that someone who was aware of B had evidence for S. So it isn't necessary that anyone who is aware of B has evidence for S.

D. Other Minds and Warrant

But if the connection between B and S isn't either inductive or deductive, what is it? If neither the Millian analogical position nor the abductive scientific hypothesis position nor the Wittgensteinian criterial position is right, what is left? What other possibilities are there? From the present perspective on the nature of warrant, the answer is simplicity it self. The answer, first, is just that a human being whose appropriate cognitive faculties are functioning properly and who is aware of B will find herself making the S ascription (in the absence of defeaters). There is nothing in the least unjustified about such ascriptions; not is there anything strange, odd, nonstandard about making them quite independently of any analogical or inductive or abductive arguments. Indeed, the pathology is on the other foot: it is the person who believes in others only on the basis of analogical arguments (and believes with a strength that matches the strength of those arguments) who is weird or nonstandard. It is part of the human design plan to make these ascriptions under those circumstances. It is part of the human design plan, in fact, to make such ascriptions with very considerable firmness; I may believe very firmly indeed that someone who has just suffered a shoulder dislocation and is holding his shoulder and moaning in that characteristic way is in pain. So if the part of the design plan governing these processes is successfully aimed at truth, then ascriptions of mental states to others will often have high warrant for us; if they are also true, they will constitute knowledge. This is so despite the fact that the epistemic connection between pain behavior and pain is neither inductive nor abductive nor deductive.

This is not to say, of course, that analogy plays no role at all in my knowing that someone else is in pain, or wishes he were home watching television, or is thinking about Vienna, or means rabbit by ‘gavagai’ rather than rabbit stage or undetached rabbit part or whatever.¹⁶ Obviously we aren't born with the full complement of our highly subtle and sophisticated belief-forming procedures about the minds of others. The belief-forming procedures of an adult human being arise from a native complement of belief-forming mechanisms that (1) change and mature with time and (2) are modified by way of experience and learning—including, of course, learning by way of analogy with my own case. Precisely how this works—just what our inborn belief-forming mechanisms here are like, precisely how they are modified by maturation and by experience and learning, precisely what role is played by nature as opposed to nurture—these matters (fortunately enough) are not topics for this study. I mean only to argue that we can (and do) have warrant for beliefs ascribing mental states to others even if there aren't any good inductive or abductive arguments from premises of the sort to which the analogical arguer must confine himself to conclusions ascribing mental states to others.

A lurking skeptic may ask: isn't it possible that we should all be deceived here? Isn't it possible that I should have the very experience I do have and there be no other minds, or perhaps only one other mind, or many other minds, but not at all the sorts of minds I think there are? Couldn't I be a brain in a vat, the only survivor of an Alpha Centaurian intergalactic attack, kept alive by my captors for scientific purposes, and completely deceived about what sorts of other minds there are? Might I not be the victim of a Cartesian evil demon who delights in deception? Perhaps she and I are the only persons in the universe; she takes malicious pleasure in seeing what an elaborate and absurdly mistaken set of beliefs she can induce in me, how articulate and detailed a ‘world’ she can create for me; she delights in my naive simplicity, my thoughtless acquiescence in my natural belief impulses; and perhaps she sometimes adds a certain subtle piquancy by inducing in me the merest, most tenuous suspicion that in fact I am such a victim.

There is no need to declare these things impossible; indeed, they seem perfectly possible (in the broadly logical sense), if a bit farfetched. But nothing much follows. In particular it does not follow that no warrant is enjoyed by my beliefs ascribing mental states to other persons or my belief that there are many other human beings. Nor does it follow that I am unjustified in these beliefs, or that there is something irrational in my holding them. All that follows is that I don't have the sort of certainty Descartes sought. I don't have the sort of certainty (if indeed there could be any such thing) in which I can simply see (in a way that is somehow beyond the possibility of doubt or mistake) that things could not be otherwise than thus and so. In the present case, of course, my beliefs can be false, despite my best efforts. But that doesn't mean I don't have knowledge of them; for while knowledge requires psychological certainty—or at any rate a high degree of belief—it is not the case that knowledge requires Cartesian certainty.

II. Testimony

A. Testimony Characterized

Another important source of belief is testimony, or teaching, or (to use Reid's name for the mechanism whereby we come to assimilate testimony) credulity:

The wise author of nature hath planted in the human mind a propensity to rely upon human testimony before we can give a reason for doing so. This, indeed, puts our judgment almost entirely in the power of those who are about us in the first period of life; but this is necessary both to our preservation and to our improvement. If children were so framed as to pay no regard to testimony or authority, they must, in the literal sense, perish for lack of knowledge.

I believed by instinct whatever they [my “parents and tutors”] told me, long before I had the idea of a lie, or a thought of the possibility of their deceiving me. Afterwards, upon reflection, I found they had acted like fair and honest people, who wished me well. I found that, if I had not believed what they told me, before I could give a reason for my belief, I had to this day been little better than a changeling. And although this natural credulity hath sometimes occasioned my being imposed upon by deceivers, yet it hath been of infinite advantage to me upon the whole; therefore, I consider it as another good gift of Nature.¹⁷

Reid's points here are worth emphasizing. First, the importance of credulity, testimony, to our entire intellectual enterprise is seldom sufficiently recognized. Testimony is the source of an enormously large proportion of our most important beliefs; it is testimony and learning from others that makes possible intellectual achievement and culture; testimony is the very foundation of civilization. The Enlightenment looked down its rationalistic nose at testimony and tradition, comparing them invidiously with science; but, without learning by testimony, clearly, science would be impossible. Newton stood on the shoulders of giants; indeed, every scientist must stand on the testimonial shoulders of others. Nearly all of what we know of the history of humanity or the structure of the universe we know by virtue of testimony; but it is also by virtue of testimony that I know such homelier items as what my name is and that I live in Indiana. You visit Armidale: you believe that it is indeed Armidale you are in, and that Armidale is in New South Wales. I have never visited Armidale and indeed have never ventured beyond the borders of Königsberg; but you rely upon testimony for your knowledge of those items as much as I do. You are also dependent upon testimony for your knowledge that New South Wales is in Australia (a fact you perhaps learned from a map or encyclopedia) and that there is such a nation as Australia.

Sigmund Freud, that Enlightenment figure born out of due time, offers an account of religious belief that, oddly enough, includes testimony as a special case: “Religious ideas are teachings and assertions about facts and conditions of external (or internal) reality which tell one something one has not discovered for oneself and which lay claim to one's belief.” (Obviously testimony involves “teachings and assertions about facts and conditions of external [or internal] reality which tell one something one has not discovered for oneself and which lay claim to one's belief.”) He immediately goes on to contradict this account of ‘religious ideas’ by claiming that what distinguishes religious ideas from testimony is that what you learn by way of testimony you can always check or verify for yourself, thus finding out whether what you were told is true.¹⁸

But surely this is Enlightenment optimism run amuck. Can I really discover, in a way independent of testimony, that in the fifth century B.C. there was a war between the Athenians and Spartans? Can I discover in this way that Plato was a philosopher? Or that the woman I take to be my mother really was? Or that I was given the name I think I was? Or that there is such a country as Australia? Indeed, the mayor of Armidale himself depends upon testimony for his knowledge that it is Armidale of which he is the mayor; and though a lifelong resident of Australia, he too depends upon testimony for his knowledge that Australia is the continent of which Armidale is a tiny part. You say: perhaps he just thinks to himself: “Armidale is a part of ——,” where the blank is to be filled by his own name of the land he sees around him, land on which Armidale is obviously to be found. But if —— is his name for Australia and is bestowed or introduced by way of the description ‘the land around here’ or ‘the land I now see’, the proposition he expresses by ‘Armidale is a part of ——’ is not the one we express by ‘Armidale is in Australia’. To express the same or an equivalent proposition, his sentence would have to contain a name of Australia; and it is not easy to acquire a name of Australia on one's own. He might try to name Australia by picking it out with a definite description: ‘the continent of which where I stand is a part’ or ‘the country to which this land belongs’; but of course it is only by testimony that he knows there is such a continent or country, or indeed any continents or countries at all.

We are therefore dependent upon testimony for most of what we know. Further, it is likely that most of our beliefs are such that the very possibility of our forming them is dependent upon testimony. For if there were no such thing as testimony, as a source of belief, then, in all likelihood, there would be nothing but the most rudimentary sorts of language. I don't mean to endorse Wittgenstein's enigmatic suggestions to the effect that it is impossible (in something like the broadly logical sense) that any person have a private language; that is as may be. (And the way it may be, I think, is at best inconclusive.) But it seems likely, as a matter of contingent fact, that language and testimony are mutually dependent phenomena in such a way that apart from testimony, there would be no language. And without the resources conferred by language we should have been unable to form any but a small proportion of the beliefs we do in fact hold.

Second, Reid is surely right in thinking that the beliefs we form by way of credulity or testimony are typically held in the basic way, not by way of inductive or abductive evidence from other things I believe. I am five years old; my father tells me that Australia is a large country and occupies an entire continent all by itself. I don't say to myself, “My father says thus and so; most of the time when I have checked what he says has turned out to be true; so probably this is; so probably Australia is a very large country that occupies an entire continent by itself.” I could reason in this way and in certain specialized circumstances we do reason that way. But typically we don't. Typically we just believe what we are told, and believe it in the basic way. Of course, as Reid says, we learn to modify, qualify, modulate our native tendency to believe what others tell us: we believe certain people on certain topics but are skeptical of others on others. I believe you when you tell me about your summer vacation, but not when you tout on television the marvelous virtues of the deodorant you have been hired to sell. We learn not to form beliefs about a domestic quarrel until we have heard from both parties; we learn to mistrust pronouncements of campaigning politicians, lawyers arguing a case, and people with a strong financial interest in our believing what they tell us. Here, as in other cases, there is a complex and subtle interaction among sources of belief.

I say I could reason in the inductive way to what testimony testifies to; but of course I could not have reasoned thus in coming to the first beliefs I held on the basis of testimony. As Reid says, “if I had not believed what they told me, before I could give a reason for my belief, I had to this day been little better than a changeling.” Perhaps the occasional great genius—a Leibniz, say, or an Augustine—could do better; but even they could do only a little better. Someone might say: “True, as a matter of contingent psychological and historical fact I can't learn what we learn by testimony by relying on induction and analogical reasoning; but now that I have reached the years of discretion with the help of others, I can reconstruct my noetic structure in such a way that I can retain all my knowledge but free it from reliance upon testimony. For example, I note that when I say ‘That is a house’ I express a proposition and express the proposition that is a house. So probably the same goes for others; they too express propositions when they utter sentences, and they too express the proposition that is a house when they utter the sentence ‘That is a house’. But this is feckless; here the problems that bedevil the analogical argument for other minds return in spades. First, it is doubtful that I can know that my sentence expresses the proposition it does without already knowing something about how others use it. Second, supposing I could, there will be the same problem here as with the other conclusions the analogical arguer means to argue for: there will be similar analogical arguments for such dismal conclusions as that every sentence that has been used to express a proposition has been uttered or written by me, that anyone who asserts any proposition uses my body to do so, and so on.

So the warrant furnished by testimony isn't and couldn't be furnished by induction, analogy, and abduction. Testimony is an independent source of warrant for me; testimonial evidence is a basic sort of evidence for me. Is it, as Richard Swinburne thinks, necessarily evidence?¹⁹ That is, is it a necessary truth that if I know that someone else has told me thus and so, then I have evidence (defeasible evidence) for that proposition, am such that (in the absence of undefeated defeaters) the proposition has at least some warrant for me? I don't think so. It is part of our design plan to learn from testimony; your telling me thus and so in fact gives me evidence for thus and so; but that this is so is not necessarily true. We, or to beg no questions, creatures like us, could have been so constructed that testimony would not furnish evidence at all. For example, we could have been constructed by a whimsical creator according to the following plan: (a) testimony is mostly false; and (b) we have an inclination to believe the denials of what we are told by others, and when we function properly, we do so. Then someone who believed P on the basis of testimony to P would be such that P has little or no warrant for him.

Or is this possible? Donald Davidson thinks it is “impossible correctly to hold that anyone could be mostly wrong about how things are.”²⁰ He argues that in trying to understand the utterances of another human being, we must make the assumption that most of what she says is true; otherwise we will have no way at all of beginning the task of trying to understand her. Of course this doesn't show that most human beliefs are true; at best it shows something much weaker: that in order to understand someone, I must make the assumption that most of what she says is true. And (as Davidson asks) “couldn't it happen that speaker and interpreter understand each other on the basis of shared but erroneous beliefs?” The answer is this “couldn't be the rule”:

For imagine for a moment an interpreter who is omniscient about any sentence in his (potentially) unlimited repertoire. The omniscient interpreter, using the same method as the fallible interpreter, finds the fallible speaker largely consistent and correct. By his own standards, of course, but since these are objectively correct, the fallible speaker is seen to be largely correct and consistent by objective standards. We may also, if we want, let the omniscient interpreter turn his attention to the fallible interpreter of the fallible speaker. It turns out that the fallible interpreter can be wrong about some things, but not in general; and so he cannot share universal error with the agent he is interpreting. Once we agree to the general method of interpretation I have sketched, it becomes impossible correctly to hold that anyone could be mostly wrong about how things are.

Richards Foley and Fumerton point out that this argument has its problems: the premise appears to be

(1) If there were an omniscient interpreter using Davidson's methods of interpretation, he would believe that most of what 5 believes is true,

but the conclusion appears to be

(2) Most of what S believes is true;

which does not follow from the premise.²¹ (Of course, it would follow if, as the ontological argument concludes, it is a necessary truth that there is an omniscient being.) Foley and Fumerton conjecture (very plausibly, in my estimation) that Davidson is relying upon a further but unspoken premise:

(3) Any proposition that is such that any omniscient being would believe it, is true.

They then point out that (3) is surely true if indeed there is an omniscient being, but need not be true if there isn't. Here they are right. Indeed, more can be said; as they point out, (3) is necessary for the proposition that there is an omniscient being, but in fact it is also sufficient. Clearly (and necessarily) one proposition any omniscient being worth its salt would believe is that there is an omniscient being. This conclusion won't much disturb a theist, but there will be those (and possibly Davidson is among them) who may find it unsettling.²²

No doubt there are better ways of arguing for the conclusion that it must be the case that most human beliefs are true; if so, I leave them to others. Reid makes a much more modest and more plausible claim. There is in us, he says,

a propensity to speak truth, and to use the signs of language so as to convey our real sentiments. This principle has a powerful operation, even in the greatest liars; for where they lie once, they speak truth a hundred times. Truth is always uppermost, and is the natural issue of the mind. It requires no art or training, no inducement or temptation, but only that we yield to a natural impulse. Lying, on the contrary, is doing violence to our nature; and is never practised, even by the worst men, without some temptation. Speaking truth is like using our natural food, which we would do from appetite, though it answered no end; but lying is like taking physic, which is nauseous to the taste, and which no man takes but for some end which he cannot otherwise attain.²³

Reid perhaps overlooks the fact that a habitual liar may come no longer to be able to distinguish truth from falsehood or to tell whether he speaks the truth or a lie; our natural tendency to tell the truth can be smothered by sufficient determination and practice. But surely there is in us a tendency of the sort of which Reid speaks. Equally surely, its presence is an essential part of the design plan for the human cognitive situation; were it not present, credulity or testimony could not properly perform its function of producing in us beliefs that are for the most part true. So when you tell me what your name is or that you are thirsty, then (if things are going according to the design plan) you believe that you are thirsty, you will indeed be thirsty, you will be telling me this in the hope and expectation that I will believe you, and I will in fact believe you. Here it is of the first importance, once more, to see that the design plan for our cognitive system does not involve us merely as individuals; our cognitive systems are designed to work together in a certain way. The human design plan is oriented toward a certain kind of cognitive environment: the sort of cognitive environment in which our faculties originally arose, whether by the hand of God or of evolution (or both). But from the point of view of the individual person, other people are part of the cognitive environment; the design plan does not cover my cognitive faculties in isolation from yours or yours from mine: as it applies to my faculties it presupposes that you and your faculties will function and react in certain ways.²⁴ So the design problem set God or evolution was less like that of designing a rocket ship for travel through as yet uncrowded interstellar space than that of designing an automobile for use in Mexico City, say, or Bangkok—except, of course, that the reciprocal reaction and interrelation is vastly subtler and more intimate in the case of our cognitive faculties than with automobiles, no matter how nasty the traffic.

On the present account of warrant, therefore, if you tell me what your name is or that you have just returned from Australia or that you own a Ford, I may thus acquire evidence for this belief and it may have warrant for me: it may be produced by my faculties functioning properly in an appropriate environment, with the segment of the design plan covering this belief acquisition successfully aimed at truth. Testimonial evidence is indeed evidence; and if I get enough and strong enough testimonial evidence for a given fact—for example, that there was such a thing as the American Civil War, or that London, England, is larger than London, Ontario—the belief in question may have enough warrant to constitute knowledge.

B. Testimony and Gettier Problems

But doesn't this interrelation and interdependence make trouble for my account? Being of a whimsical turn of mind, my parents intend to teach me mainly falsehoods about the geography of Scotland; fortunately for me, however, they suffer from a rare but well-defined malady that causes them to believe the denials of what they were taught about British geography, so that most of what they believe on that head is false. Although they wind up teaching me mainly truths about Scottish geography, these are not truths I know. You come into my office, show me the bill of sale and title made out to you for a new Ford, explain to me in confident and convincing tones that you now own a new Ford, take me for a ride in what you say is your new Ford, and all the rest; I form the belief, on the basis of you testimony, that you own a new Ford. The sad fact of the matter is that you don't believe for a moment that you own a new Ford; you are misinforming me as part of an initiation into the Elks Club. Unbeknownst to you, however, you uncle has just died and left you a brand new Ford. Then my faculties may be working perfectly in an acceptable environment in producing in me a true belief that I hold very firmly; but I surely don't know that you own a Ford.

In these two cases there is intent to deceive on the part of the testifier: but of course that isn't essential to these cases. Consider the young tribesman whose elders fill him with wildly false beliefs about the stars—the stars, they say, are pinholes in a giant canvas pulled over the sky every night to enable us to get a good night's sleep—with a few true beliefs tossed in (for example, that the stars are not made of wood). These true beliefs do not constitute knowledge for him. In these case we may suppose that the cognitive faculties of the testifiee are functioning properly in the very sort of environment for which they were designed, and the segments of the design plan governing the formation of those beliefs on the part of the testifiee are in fact aimed at truth (rather than psychological comfort or survival of a disease or the possibility of loyalty and friendship); so why don't the beliefs formed have warrant? Why can't they have enough warrant for knowledge?²⁵

What we have here are semi-Gettier problems.²⁶ (Semi because what Gettier problems really show is that justification is not sufficient for knowledge.) But just as the interrelatedness and interactions of our cognitive systems seems to provide the problem, so it provides the solution. The cognitive design plan, as I argued, applies not only to the individual cognizer and her faculties, but also to the whole cognitive situation in which she finds herself, including the cognitive functioning of other persons. In these case the testifiee's beliefs lack warrant, not because of malfunction in his own cognitive system, but because of lack of warrant elsewhere in the chain leading to the belief he forms on the basis of testimony. The last item in this sort of epistemic chain has warrant only if the preceding items are formed in a way that accords with the design plan for the whole chain.

There is a sort of recursive structure here. Consider a chain of length two: you communicate to me a belief B by way of testimony. If B arises in you by way of cognitive malfunction, then B has little or no warrant for you and little or no warrant for me (although as I shall explain, I may be entirely justified in accepting it). Similarly, if B arises in you by way of proper function of cognitive processes not aimed at truth, then the belief has little or no warrant for either of us. (You are certain you will survive this disease; this belief arises in you not because of a sober and reasoned calculation of the odds but because of the operation of the Optimistic Overrider (see p. 42); this belief has little warrant for you, and if you communicate it to me by way of testimony, it will have little for me.) Again, a belief may arise in you by way of perceptual illusion: I have come to North Dakota for the first time; the road ahead looks wet; I announce that it has recently rained a half mile up the road; you (who are otherwise occupied and don't look) believe me. Under these conditions my belief arises by way of perceptual illusion and hence by way of cognitive tradeoff (pp. 38ff.); thus it has little or no warrant for me; and the same then holds for your belief acquired from me by way of testimony. In the chain of two members, therefore, the testifiee's belief has warrant only if the testifier's belief does.

Here we have the base case for a fanciful application of mathematical induction: and clearly enough, for a chain of greater length, the n + 1st member will have warrant only if the nth member does. But that means then, that the last item in such a chain has warrant only if the preceding items do. In some varieties of natural deduction logics, there is a form of inference sometimes called ‘Repetition’: from A to infer A. Testimony, we might say, is a kind of interpersonal Repetition; and clearly there will be no warrant for the second occurrence of the belief unless there is for the first. It is also true, I think, (although I won't argue it here) that in the simplest sorts of cases the testifiee's belief has no more warrant than the testifier's. (I won't stop to explore what happens when I believe A on the basis of the testimony of several people, for whom A has widely different degrees of warrant, or where some tell me A and others not-A, or where some tell me not-A and others merely express doubts about A, and so on.) This exemplifies an important principle: to put it technically, testimonial warrant, like water, rises no higher than its source. (Alternatively: a testimonial chain is no stronger than its weakest link.)

Now we can return to the semi-Gettier cases. If, by virtue of cognitive malfunction, you form some belief or other and communicate this belief to me by way of testimony, then the belief has for me no more warrant than it has for you, despite the fact that there is nothing wrong with my cognitive faculties. In the case of the young tribesman, we may suppose that somewhere back along the cognitive chain this belief arose in such a way that it had little by way of warrant. Perhaps it arose out of sheer guess work, or perhaps it began its career in an imaginative bedtime story and its youthful auditor mistook it for intended sober truth, passing it on as such. As it was passed on, perhaps it was believed ever more strongly, gradually assuming the status of established and unquestioned lore. But if it didn't originally have much by way of warrant, then it doesn't in the case of the young tribesman.

The Gettier cases involving lying and deception are slightly different and require slightly different treatment. Here the testifier doesn't accept the belief involved at all, but nonetheless testifies to it, the testifiee thus acquiring it. But obviously here too the belief in question (if it has no other source of warrant) will have little or no warrant; the belief in question has no warrant for the testifier, for the testifier does not so much as accept that belief. When the testimonial situation is going according to design plan, the testifier testifies to what he believes is true. The purpose of our cognitive nature is to furnish us with appropriate truths; one module or aspect of the design plan involves our learning from others, from our parents when we are young and from various appropriate others throughout our lives. But of course this plan will be successful only if those from whom we learn teach us truths.²⁷ And in the typical case, they will teach us truths only if they intend to do so, an intention sadly lacking in cases of deception.

The design plan for the whole situation, therefore, is such that things are proceeding according to the plan only if those from whom I learn by way of credulity intend to teach me the truth—and we must add, are not themselves deceived in some wholesale way incompatible with the design plan of the whole cognitive situation. The lack of warrant, however, is not (or need not be) by way of malfunction in the testimonial chain. When I lie to you, my cognitive faculties may be functioning properly (that is, there need be no cognitive malfunction)—just as the strangler's hands display no dysfunction in being used for that evil purpose. But then how shall we account for the lack of warrant? It is not a matter of malfunction; nor is it like the sort of case where the segment of the design plan governing the relevant modules of the system is aimed at something other than truth. Perhaps we must say something like this (at least for the simplest cases of lying): although the natural aim or purpose of our faculties is to produce true beliefs, certain aspects of the cognitive situation are under our control; some of our cognitive faculties are such that we can aim them in a different direction, use them for a different purpose. (In the same way, I can use my car as a battering ram, or an anchor, or a large paperweight.) When this happens, when the testifier employs the relevant segments of the cognitive system for, say, deceit or subterfuge, then on that occasion the testifier's intentions override the natural purpose of the cognitive modules in question. On that occasion their use is not (as it ordinarily is) aimed at the production of true beliefs but at something else—in this case, indeed, the production of false beliefs. And hence the belief formed by the testifiee has little or no warrant for him.²⁸ If it happens that when she lies, the testifier nonetheless unawares and unintentionally speaks the truth, the testifiee still does not know. Richard Foley points out that here there is something like a corrective to Cartesian individualism in epistemology; Cartesianism is skewed from the start by its focus on the solitary knower. Justification may be a solitary matter; it may be such that I can achieve it without reliance upon others; whether I am justified depends solely upon my own efforts. The same is not true for warrant.

But doesn't it seem a bit peculiar to say that the warrant of your present belief (when it is acquired by way of testimony) may depend upon the degree of warrant that belief had for someone else, perhaps someone long dead, someone such that you have no way at all of discovering anything about his noetic structure? Can the degree of warrant your belief has depend in this way upon something by now wholly inaccessible to you? But where, exactly, is the problem? The warrant a belief has for you, after all, depends upon whether your faculties are functioning properly, a condition that may be equally inaccessible to you. A belief may fail to have warrant for you by failing to meet a condition of warrant, even if there is no way in which you could discover that that condition is not met. By virtue of cognitive malfunction, I believe I am Napoleon; this belief has little warrant for me, even though there may be no way in which I could find out that I am subject to cognitive malfunction.

So a belief on the part of the testifiee has warrant only if that belief has warrant for the testifier. There are further subtleties, some of which have been brought to our attention by Gilbert Harman. What if you acquire a belief by way of testimony but fail to note, later on, that the relevant experts have given it up? Does it still have warrant for you? Suppose you are a simple believer in Darwinian evolution. Suppose the experts become doubtful about Darwinian evolution, due to reflection on the spotty character of the fossil record. (It presents so few intermediate forms as to make it unlikely that anything much like Darwinian evolution occurred.) Suppose these doubts get communicated to nearly everyone around you, but by some fluke you continue to believe, somehow not hearing or heeding the doubts of the skeptics. Suppose, as it turns out, Darwinian evolution is indeed the truth. Did you know all along? Or was the warrant that belief held for you reduced by the fact that others (mistakenly) doubted? Here we must say, I think, that the answer is not clear; this is one of those penumbral borderline areas.

Where you lack warrant because of conditions in a part of the epistemic chain far removed from you, it is clear, of course, that you may nonetheless have justification. You may be perfectly within all of your rights; you may have been flouting no duties whatever; you may have been doing your level best to achieve the truth. (Given the semideontological tang of ordinary uses of ‘warrant’, we could put it paradoxically like this: you are entirely warranted in the belief in question, even though the belief has little warrant for you.) You may therefore have justification: you may have much more. For example, you may have post-classical Chisholmian justification: that is, the relation between your purely psychological properties and the belief in question may have as high a degree as you please of intrinsic value. More generally, you may be such that everything is going splendidly with respect to everything that is cognitively accessible to you in the way in which, according to the internalist, what confers warrant must be cognitively accessible.²⁹ Still more generally, everything involving your own cognitive faculties and your use of them may be going properly. You may also be entirely rational and that in a fourfold sense; (a) in the sense that contrasts with someone's irrationally believing that he is Napoleon or that his head is a gourd or made of glass, (b) in the sense of Foley rationality, so that upon sufficient reflection you would think that forming this belief in these circumstances is a good way to achieve your epistemic goals, (c) in the sense that forming this belief in this way is following the dictates of reason, and (d) in the deontological sense of being entirely within your intellectual rights. All of this is compatible with the belief's having little or no warrant for you.

Testimony or credulity, therefore, is a crucially important part of our noetic arsenal; it is the foundation of culture and civilization. I conclude by pointing out two ways in which it is nonetheless a second-class citizen of the epistemic republic. First, testimony is ordinarily parasitic on other sources of belief so far as warrant goes. We have already seen a special case of this phenomenon: if you tell me what you think is false, then I don't know it even if you are mistaken and I meet the other conditions of knowledge. More generally, if you tell me something and I believe it on you say-so, I have warrant for it only if you do. To take an example of Steve Wykstra's: most of us who believe in quantum mechanics do so on the say-so of others; most of us have no independent evidence (independent of testimony) for the results of, say, double slit experiments. I may be entirely justified in believing as I do, on the basis of testimony, and in some cases (though perhaps not in this case) when I believe what I read in science textbooks I know what I come thus to believe. But I wouldn't have this knowledge if there weren't others in the neighborhood (that is, in the cognitive chain) who had nontestimonial evidence for the fact in question. As Wykstra says, if no one has nontestimonial evidence for the claim in question, then the whole epistemic community is in “big doxastic trouble.” (And the kind of trouble is this: if no one has nontestimonial evidence for the facts in question, then none of our beliefs on this head has warrant, even though we are both justified in forming the beliefs and such that our faculties are functioning properly.)

In the typical case, therefore, if I know something by testimony, then someone else must have known that proposition in some other way. Of course, this condition isn't always met. Perhaps you and I and many others together map the coast of Australia: then I know by nontestimonial means that this bit has this shape; you know similarly that that bit has that shape, and so on for the rest of the members of our crew; we all know what shape the whole continent has, even though none of us has nontestimonial knowledge of the fact that it has that shape. The principle has to be state more carefully to be correct; I leave this as a homework problem.

Second, in many situations, while testimony does indeed provide warrant, there is a cognitively superior way. I learn by way of testimony that first-order logic is complete, or that the continuum hypothesis is independent of ordinary set theory; I may thus come to know these things. I do even better, however, if I come to see these truths for myself, by understanding an appropriate argument, let's say.³⁰ You tell me that So and So was at the corner of Fifth and Broadway at midnight last night; I then have warrant for this belief, but not as much warrant as if I had seen him there and then myself. An eyewitness report carries more weight than a report from someone to whom the eyewitness told what he saw. Thus Sigmund Freud:

I was already a man of mature years when I stood for the first time on the hill of the acropolis in Athens, between the temple ruins, looking out over the blue sea. A feeling of astonishment mingled with my joy. It seemed to say: “So it really is true, just as we learnt at school!” How shallow and weak must have been the belief I then acquired in the real truth of what I heard, if I could be so astonished now!³¹

Testimonial evidence is indeed evidence; it is not always the evidence of choice.