Epistemology of Silence - Oxford Scholarship Jump to ContentJump to Main Navigation
Social Epistemology$

Adrian Haddock, Alan Millar, and Duncan Pritchard

Print publication date: 2010

Print ISBN-13: 9780199577477

Published to Oxford Scholarship Online: January 2011

DOI: 10.1093/acprof:oso/9780199577477.001.0001

Show Summary Details
Page of

PRINTED FROM OXFORD SCHOLARSHIP ONLINE (www.oxfordscholarship.com). (c) Copyright Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a monograph in OSO for personal use (for details see http://www.oxfordscholarship.com/page/privacy-policy). Subscriber: null; date: 28 August 2016

The Epistemology of Silence

The Epistemology of Silence

(p.243) 12 The Epistemology of Silence
Social Epistemology

Sanford C. Goldberg (Contributor Webpage)

Oxford University Press

Abstract and Keywords

Epistemic disagreement is disagreement over epistemic principles, or principles concerning the reliability and extent of our epistemic methods. This chapter argues that disagreement over this sort raises a new problem distinct from skepticism. Like some skeptical arguments, the problem of epistemic disagreement is rooted in part in the issue of epistemic circularity. But it is not a problem about whether we in fact have knowledge or are justified in our opinions. It is about rationally resolving explicit disagreement over the reliability of our most basic methods for forming beliefs. The bulk of this chapter is concerned with getting clear on the problem and its nature. Finally, the chapter sketches a solution.

Keywords:   epistemic circularity, epistemic disagreement, justified belief, knowledge, rational resolution, reliability, skepticism


Consider the following two sorts of case.

CASE 1. In the course of a conversation, the issue comes up whether Jones regularly wore a bright red suit. You think about it come to the belief that, no, he did not. You reason that if he had regularly worn such a suit you would have seen him in it, and you can recall nothing of the sort.

CASE 2. In the course of a conversation, the issue comes up whether weapons of mass destruction were ever found in Iraq after the second Gulf War. You think about it and come to the belief that no such weapons have been found to date. You reason that if they had been found you would have heard about it by now, and you have heard nothing of the sort.

In this chapter I aim to compare and contrast these two sorts of belief based on the ‘silence’ of a trusted source.

Let us begin with the most salient features of the comparison. In both cases two salient points emerge immediately. First, the subject does not have any relevant confirming memory (of having seen Jones wearing such a suit; of having been informed that WMDs were found in Iraq). And second, the subject assumes that if the relevant proposition (that Jones regularly wore the suit; that WMDs were found in Iraq) were true, she would have obtained the relevant information (through perception or testimony) and would have retained that information in memory.

There are other, perhaps less salient similarities between the cases. For one, in both cases the proposition believed is a negation: Jones didn't regularly wear such a suit, WMDs have not been found to date. For another, in both cases the corresponding affirmative proposition is one regarding whose truth you might well have assigned (or have been disposed to assign) some low probability prior to reflecting on whether you had seen or heard about such a thing. For example, if Jones is not known for sartorial flamboyance, the hypothesis that he regularly wore a bright red suit would get a low prior probability (even before you reflected (p.244) on whether you had ever seen him in one, etc.). Or if you are deeply suspicious about Bush's motives in starting the second Iraq war, you might well have doubted the existence of WMDs in Iraq, even before reflecting on whether you had ever come across a report of such. In this light, the fact of ‘silence’ in the sources that are relied upon should be seen as bearing on a proposition regarding whose truth you already have other (indirect) evidence.

So far we have been attending to the similarities between the cases, but there are differences too. The most obvious one concerns the identity of the information source being relied upon. In CASE 1 you are relying upon yourself, whereas in CASE 2 you are relying in addition on someone (or ones) other than yourself. In the latter case your ability to identify the relevant source or sources may be quite limited—you may only know that some source or other would have found out about the discovery, and would have reported it—or it may be relatively informed—you may know the identity of the CNN reporter on whom you rely for news from the Middle East, and in fact may know quite a bit about the CNN research team as a whole. But this difference between CASE 1 and CASE 2 may not be that significant. Granted that it is only in situations like CASE 2 that the question can meaningfully arise regarding the identity of the source on whom you are relying, even so, in many cases of this type there may be no question in your mind regarding whom you are relying upon (you know you are relying upon the reporters of the New York Times for your news). What is more, while it is true that in CASE 2 there can be variability in how much the subject might know about (the reliability1 of) the relied‐upon source (you might fail to know much about the Times' methods of gathering and disseminating that information), this might well be a similarity with CASE 1, as you may fail to know very much about how you gathered and stored the relevant information about Jones' clothes. To be sure, you will probably know enough in CASE 1 to say that your methods involve perception and memory; but at this level of generality the cases might well remain similar, since you might say in CASE 2 that the New York Times gathers its information by placing journalists in the field, and then disseminates the information through the publication of the daily paper (in hard copy or on the web). This leaves open precisely how much you know about (the reliability of) these processes you are designating as ‘perception’ and ‘memory’, ‘journalistic practices', and ‘publication of the daily paper’.

How should an epistemic assessment of these beliefs accommodate the similarities and differences between the individualistic case (CASE 1) and the social case (CASE 2)? In what follows I will argue that in both cases the beliefs themselves involve the same potential sources of unreliability, (p.245) and depend for their reliability on the goodness of the ‘coverage’ that is provided.


My ultimate aim in this chapter is to characterize the epistemic significance of the ‘silence’ of a relied‐upon source, whether that source is oneself (CASE 1) or another/others (CASE 2). To appreciate the epistemic significance of silence, however, it will help to begin with the notion of epistemic reliance on an information source, as this bears on cases in which an information source, far from being silent, offers a ‘report’. Once again we can distinguish the case of self‐reliance—for example, the case of having a memory impression as of it being the case that p—and cases in which the epistemic reliance is on another subject as information source—the familiar case of testimony. After examining these cases, we will then return to the cases where the source in question is silent.


Take the self‐reliance (memory) case first. What sense does it make to say that in cases of memory one relies on oneself qua information source? Confronted with an apparent recollection whose content is p, one is relying on oneself in two ways. One is relying on oneself, first, as the source of the information that p: here one relies on the reliability of the processes through which one originally acquired the belief that p. In addition, one is also relying on oneself as responsible for the reliable transmission of a previously acquired belief‐content: here one relies on one's memory to have succeeded in transmitting—retrieving and making available for one's present use—the very content one previously acquired, such that one wouldn't currently have the memory impression as of p unless one acquired that very information, p, at some earlier time. Hence we can distinguish reliance on oneself qua source of information,2 from reliance on oneself qua transmitter (or preserver) of information. (p.246)

Parallel distinctions can be made in the interpersonal case, where one's reliance is on an information source other than oneself. Here too we can distinguish two dimensions of reliance: on the source, and on the process of content preservation (or content transmission).3 With respect to the latter sort of reliance, the parallel rests on the idea that we can regard proferred testimony, just as we regard apparent recollection, as serving the role of transmitting or preserving previously acquired information. Just as it is clear that memory (qua would‐be preservative process) can fail to preserve the content of a previously acquired belief, in a way that does not impugn the reliability of the belief on acquisition, so too it is clear that the process by which a journalist's report is publicized (qua would‐be preservative process) can distort the report, in a way that need not reflect poorly on the reliability of the original report.

So far I have been arguing that there are two dimensions along which we can develop the parallel between the sort of epistemic reliance in play in memorial belief and that which is in play in testimonial belief. In particular, both involve (i) an original source of information (ii) whose contents are presently transmitted (made available) to the subject. This distinction between (i) and (ii) will bear a good deal of weight in what follows.

In the case of testimony the distinction is clear enough: one's source for the information in question is another person, and that person might or might not choose to provide her audience with the information in question, and might or might not be reliable in the transmission. In the memory case the distinction between (i) and (ii) may not be so clear, but it is present. Perhaps everyone will agree that self‐reliance in memory cases includes reliance on oneself for information preservation (or what Burge (1993) calls ‘content preservation’). In recollecting that p, the information that p is retrieved and (through the memory impression) made available for the subject's present purposes. But self‐reliance in cases of memorially sustained belief goes beyond this, to include reliance on oneself qua source as well. That is, one is not merely depending on oneself to have retrieved information that one in fact had acquired earlier (one doesn't want one's memory just making things up); one is also depending on oneself, or on one's past self, to have reliably acquired that information right at the outset (beliefs acquired through wishful thinking, even when successfully preserved in memory, are none the more reliable for having been so preserved).4

I have been arguing that in both memorial belief and in testimonial belief, we can factor the epistemic reliance involved into reliance on (i) the source of information and (ii) the transmission process whereby information acquired at some earlier time is presently made available to the subject. In connection with (p.247) (ii), however, there would appear to be several differences between the memory case and the testimony case.

One such difference concerns the link between the message and its recipient. Clearly, the link between the two can be severed in testimony cases: the source(s) on whom one relies might well make a report without one's coming across that report.5 In memory cases, however, the link would not appear to be severable: it makes little sense to say that one's memory system has issued an apparent recollection which one oneself has failed to ‘come across’. (One might try to make sense of this in terms of different parts of the mind/brain failing to communicate; but as this is a matter for empirical theories of memory, not for epistemology, I will not pursue this idea here.) However, this difference, though real, need not impugn the parallel I have been developing between memory and testimony. Even if it is true that there are no ‘memory transmissions' that fail to be apprehended by the subject herself, whereas there are ‘testimonial transmissions' that can fail to be apprehended by her (as when she is not near the testifying source at the time of the testimony), still the distinction between source and transmitter remains in the memory case. This is seen in a case in which one oneself has information stored in memory, yet fails to be able to access it at some later time. Such cases are naturally thought to be the analogue of the case in which a subject reliably believes that p but does not express this belief to her audience.

A second, related difference between the nature of information transmission in memory and testimony cases concerns intermediaries in the process of transmission. Whereas few of us get our news ‘straight from the horse's mouth’, and so stand in need of intermediaries in the chain of communication that links our source's testimony to us, there are no such intermediaries in the case of memory: nothing comes between you and your apparent recollections. Again, however, this difference does not impugn the relevance of the distinction between (i) and (ii) for memory cases and testimony cases alike. Rather, it merely suggests that the process of transmission in testimony cases can be significantly more involved than it is in memory cases.

Finally, a third apparent difference along the dimension of information transmission concerns the need for message comprehension. Whereas testimony, if it is to be useful to a subject, must be comprehended by her, it is by no means obvious that there is any intra‐personal analogue of comprehension in cases of apparent recollection. The issue is somewhat complicated. Suppose (for example) that at least some memories are stored via sentential (or sentence‐like) (p.248) vehicles. And suppose that for a public‐language speaking subject the sentential vehicle is not a sentence of mentalese but rather a sentence from one's public language—the sentence of the language one would use to express the content of the recalled belief. In that case we might think that these ‘stored sentences' require comprehension just as much as do pieces of testimony. Indeed, Burge 1999 appears to suggest a variation on this theme:

Comprehending standing, conceptual aspects of one's own thought and idiolect is itself, as a matter of psychological and sociological fact, normally dependent on having comprehended thoughts (one's own) that were shaped and expressed by the words of others. (Burge 1999: 243)

(What is interesting here is Burge's use of ‘comprehension’ and its cognates, in connection with one's own thoughts.) The matter is vexed, however, and I would not want to put too much weight on this aspect of the alleged analogy between memory and testimony. But once again it is worth pointing out that even if memory is disanalogous from testimony on this score, in that memory does not involve any ‘comprehension’ of a message, this need not undermine the claim that the epistemic reliance involved in both memory and testimony can be factored into reliance on (i) and on (ii). All that it would suggest is that the process of transmission differs between the two cases.

In sum, even if all of these alleged differences between memory and testimony are conceded to be actual differences, such a concession would only establish that the process of content transmission is more complicated (and allows for more distinctive kinds of error possibility) in the case of testimony. These differences in question are not sufficiently great to call into question the parallel between memory and testimony, nor are they sufficiently great to undermine the claim that in both types of case epistemic reliance can be factored into (i) and (ii). What is more, we might even think that whatever differences there are between testimony and memory in the above respects are to be expected, given the differences between intra‐personal and interpersonal processes of information transmission. For it would seem that all three of the recently‐noted differences flow from the difference between a system's retrieval of information that is already stored within the system itself, and a system's reception of information proffered by some source other than the system itself. The latter sort of case involves the possibilities of ‘missing’ the message altogether, of the utility of communication chains involving intermediaries, and of the need to ‘decipher’ and ‘re‐encode’ the message once it is received. These differences would thus appear to point to two distinct species (intra‐personal and interpersonal) of a single underlying genus (information transmission).


So far we have been considering epistemic reliance in cases in which a subject receives a ‘transmission’ from an information source. But what should be said (p.249) of epistemic reliance in cases in which a relied‐upon source is silent? These would appear to be importantly unlike the standard cases in the epistemology of memorial or testimonial belief: arguably, the lack of a memory impression, or of a report, should not itself be thought of as something that the source in question ‘presents'.

This initial difference would appear to make an epistemic difference. For while the fact that you have an apparent recollection to the effect that p, or that you have observed a report to the effect that p, might (together with whatever justifies you in accepting the ‘presentation‐as‐true’) ground a justification for your sustaining or acquiring the belief that p, it is not clear how the fact that you have no apparent recollection to the effect that p, or that you have not observed a report to the effect that p, might (together with other things) ground a justification for your acquiring a belief in ∼p. After all, a popular (though not universally endorsed) claim among people working on testimony is that, when successful, testimony affects a transmission of epistemic status, from the testimony, to the hearer (or the hearer's belief). Since no such transmission is affected in cases in which there is no testimony (or other sort of report) in the first place, it is not immediately clear how to account for the epistemology of silence.6

The issue before us, then, is this: how (if at all) does the epistemic status of a hearer's belief depend on the epistemic perspective of those on whom she is epistemically relying, in those cases in which her sources are silent? This issue can be helpfully framed by comparing two subjunctive conditionals, one that (I will suggest) lies at the heart of the epistemology of testimony (and memory), and the other that lies at the heart of the epistemology of silence (whether the relied‐upon information source is oneself or another).7

Take first the testimony and memory cases. In these cases the relevant source does (purport to) transmit information. In order to have a single expression covering the sort of ‘transmission’ involved in both, it will be helpful to follow Burge (1993) and use ‘presentation‐as‐true’ and its cognates as the general term covering the positive deliverances of a source. (The idea, then, is that both memory‐impressions and attestations/reports are species of the single genus ‘presentation‐as‐true’.) Here the conditional relevant to assessing the epistemic status of a subject's belief, formed through accepting the presentation‐as‐true in question, is what we might call the presentation‐to‐truth conditional:

  1. (1) If source φ were to present‐as‐true that p, then it would be the case that p.

(As stated, (1) is a bit awkward. More intuitively, the claim is this: φ would present‐as‐true that p only if it were the case that p.) If the case is one of memory, then φ is oneself, and the presentation‐as‐true takes the form of a (p.250) seeming recollection (or memory impression); whereas if the case is one of testimony, then φ is the (news) source on whom (or on which) one relies for information of this kind, and the presentation‐as‐true takes the form of a piece of testimony.

The role of (1) in the epistemology of memorial or testimonial belief is best seen as a claim exploited by the epistemologist in determining the reliability of a given presentation‐as‐true. In saying that (1) is a claim exploited by the epistemologist, I mean that it is not something that needs to be entertained by the subject herself in the course of memorial or testimonial belief‐fixation. It is arguable that the truth of (1), together with a hearer's counterfactual sensitivity to indications that (1) is false, suffice to render a (memorial or) testimonial belief justified. The subject who acquires such a belief need not draw any inferences from premises like (1), nor even so much as have (1) ‘in mind’ in the process of memorial or testimonial belief‐fixation.8

Let us move on to the silence cases.9 These differ both in the subjunctive conditional involved, and in the role that the relevant subjunctive conditional plays in the epistemology of silence. Begin with the conditional involved. Whereas the relevant conditional in cases involving a presentation‐as‐true is the presentation‐to‐truth conditional, (1), in cases of silence, the relevant conditional is one that we might call the truth‐to‐presentation conditional:

  1. (2) If it were the case that p, then some source would present‐as‐true that p.

(2) captures the idea that there is relevant ‘coverage’ in the domain in question: all relevant truths will be presented as such.10 This is what lies behind the reasoning one exhibits when, on the basis of not recalling having seen Jones wear a bright red suit, one concludes that he did not regularly wear any such suit; or when, on the basis of not recalling having come across any report of the discovery of WMDs in Iraq, one concludes that there has been no such discovery to date. Of course, it is not (2), so much as one of (2)’s implications, that is used in our reasoning in such cases. The relevant implication is this:

  1. (2*) If no source has presented‐as‐true that p, then ∼p.

(p.251) In what follows my claim will be that what I am calling the ‘epistemology of silence’ is best explored by examining the epistemic status of (2*), and in particular, by exploring the epistemic grounds one may have for endorsing a conditional of this sort.

Not only do silence cases differ from the presentation cases in the conditional that is involved in epistemic assessment; the role that the conditional plays in the epistemology of the resulting belief differs as well. In particular, whereas (1) is a claim used by the epistemologist during assessment (and need not be exploited by the subject who comes to acquire the belief in question), the relevant instance of (2*) itself would appear to figure in the very process by which a silence‐supported belief is formed by the subject—at least if that process is to be a reliable one. To see this, consider the difference between having evidence for 〈p〉, and lacking evidence for 〈∼p〉.11 Without introducing further qualifications, it would be false to say that a lack of evidence for 〈∼p〉 is evidence for 〈p〉. So insofar as one's belief in 〈p〉 is acquired through one's recognition of a lack of relevant evidence for 〈∼p〉, the epistemic goodness of one's belief in 〈p〉 would appear to depend on how reliable an indication of 〈p〉 it is, when one lacks evidence for 〈∼p〉.

What is more, it would seem that without background beliefs to justify the hypothesis that the lack of evidence for 〈∼p〉 is a reliable indication that p, a subject who formed her belief in 〈p〉 merely on the grounds of recognizing the evidential lack for 〈∼p〉 will have formed an unjustified belief. This is so whether one's theory of justification is reliabilist or more ‘internalist’ and reasons‐based. If one is a reliabilist, then the belief in question is unjustified because unreliably formed; whereas if one is more internalist, then the belief is unjustified since grounded in an inference from ignorance (which is a fallacious inference‐type).

Under what conditions, then, might a silence‐supported belief enjoy a doxastic justification? To some extent, our answer depends on our favoured theory of justification. If one is a reliabilist, reliability in the formation of such a belief would appear to require having inferred the belief in 〈p〉 in a conditionally reliable way from a reliably‐formed belief in the relevant instance of (2*). If one regards justification as a matter of good reasons, one will regard silence‐supported beliefs as justified only if the subject has good reasons for regarding the relevant state of ignorance as indicative of the truth of what she believes. But whatever one's favoured theory of justification, we can now see that the acquisition of a silence‐supported belief that ∼p is best seen as a kind of inference from ignorance; and so the role of (2*) in this process is that of the key premise in the inference from which the subject herself draws the conclusion that ∼p. Insofar as the subject formed her belief that ∼p in this way, this belief is justified only if she is justified in believing the relevant instance of (2*). We have reduced (p.252) the epistemology of silence to the question, What does it take to be justified in believing the relevant instance of (2*)? I propose to return to address this matter below.


Let us first ask after the truth‐likeliness of a belief formed on the basis of the silence of a relied‐upon source of information: how likely is it that a belief of this sort will be true? And let us deal first with the case where the relied‐upon source is oneself, and the ‘silence’ is that of one's lacking any relevant memory (CASE 1). As anticipated above (in our discussion of the sources of possible error in these beliefs), two main factors are relevant to the truth‐likeliness of these beliefs: (a) the coveragereliability of oneself qua information source, and (b) the preservation‐reliability of one's memory.

Coverage‐reliability supervenes on considerations bearing on the subjunctive conditional corresponding to (2) above, as follows12:

  1. (S‐M) If it were the case that p, subject S would have found out that p.

The idea is that one can rely on oneself for coverage of the relevant facts in the domain in question. Let us say that a 〈p〉‐world w enjoys ‘coverage by subject S’ with respect to 〈p〉at time t when in w S finds out that p at some point at or before t. And call a set of 〈p〉‐worlds ‘clean at t relative to S’ when all of the worlds in the set are such that each enjoys coverage by S at t regarding 〈p〉, i.e. none of them are such that by t S does not find out that p. Then we can characterize a source's coverage‐reliability in a rough but intuitive way in terms of nearby 〈p〉‐worlds. The broader the circle of worlds (with the actual world at centre) that are clean at t relative to the source, the more coverage‐reliable that source is regarding 〈p〉. Of course the temporal dimension is also relevant here: if one's source finds out that p long after the time by which the subject needed to know whether p, then such knowledge, even when it is presented to the subject by the source, will be of little or no help. So we might want to restrict t to the latest time at which possession of the knowledge in question by the subject would satisfy the relevant practical and theoretical aims bound up with her interest in 〈p〉. A source who is coverage‐reliable in this sense will be such that for most or all of the propositions π in some domain of interest to the subject, were π true the source would know this, and she would present this knowledge ‘on time’ to help the subject in whatever aims were bound up with the subject's interests in (p.253) the matter. Clearly, to the extent that a source is not coverage‐reliable, to that extent the source's failure to discover that p is a less reliable indication that ∼p. To just this extent, the beliefs one forms on the basis of the ‘silence’ of that source will not enjoy a high truth‐likeliness quotient—with the result that it will be increasingly a matter of luck if one's current ‘silence’‐based belief turns out to be true.

Of course, the truth‐likeliness of one's ‘silence’‐based belief also depends on how well one's memory preserves (the contents of) one's beliefs. The point can be put in terms of another conditional:

(P‐M) If (i) at some earlier point S acquired the belief that p, (ii) in the interim S acquires no relevant information bearing on whether p, then if presently S were to be queried regarding whether p, S would have the memory impression that p.

A subject's memory may fail to be good in this regard in at least two distinct ways. There is ‘straight forgetfulness’, in which (although at some earlier point S had acquired the belief that p) memory fails to retain that belief. And there are failures of ‘content preservation’, where (although at some earlier point S had acquired the belief that p) what is preserved in connection with that belief is a content distinct from 〈p〉.13 Either way, as one's memory fails to preserve one's original beliefs (rendering (P‐M) false), one increases the risk of error in a ‘silence’‐based belief. This is easiest to see in cases of straight forgetfulness. Here it should be clear that the more straightforwardly forgetful a subject is, the less her having no recollection of 〈p〉 (when queried) correlates with the falsity of 〈p〉 (= the truth of 〈∼p〉). Arguably, a similar point holds in cases involving the failure of content preservation. Here it can happen that the subject (when queried) has no recollection of it being the case that p, under conditions in which she originally acquired the belief that p: her memory shifted the content of her original belief from 〈p〉 to (logically distinct) 〈q〉, and her current memory‐sustained belief that q failed to trigger the recollection that p.14 Once (p.254) again, a subject whose memory exhibits this sort of failure is a subject facing an increased risk of falsity in her silence‐supported beliefs.

It is clear that these same points hold in the interpersonal case, when the subject relies on the coverage‐reliability of some source other than herself (CASE 2). Here coverage‐reliability depends on the interpersonal analogue of the subjunctive conditional (S‐M), as in the following:

(S‐T) If it were the case that p, one of the sources on whom S relies would have found out that p.15

Suppose our subject forms the silence‐supported belief that ∼p, under conditions in which (S‐T) is false. Then her silence‐supported belief is exposed to an increased risk of falsity: even assuming that her failure to recall having been told that p accurately reflects that her sources did not in fact report or publicize that p, her silence‐supported belief that ∼p is not properly sensitive, since the failure of her sources to report/publicize that p reflects a failure on their part to have discovered that p (as when the fact that p has been successfully covered up by the relevant authorities).16 Alternatively, the truth‐likeliness of one's silence‐supported belief is decreased under conditions involving unreliability in the transmission of the message from source to subject. In these cases the following subjunctive conditional (corresponding to (P‐M) in the memory case) would be false:

(P‐T) If the source(s) on whom S relies had acquired the information that p, they would have reported or publicized that p, in such a way that S would have come across the transmission (or something that stood in a reliable communication‐relation to that transmission, and which preserved its content).

Even on the assumption that one's source(s) is/are coverage‐reliable in the sense of (S‐T), S's silence‐supported beliefs (formed through relying on said source(s)) will have an increase in the likelihood of falsity if said source(s) doesn't/don't reliably transmit the information in question. For in that case, even assuming one's failure to recall having been told that p accurately reflects that one's sources did not in fact report or publicize that p, even so, one's silence‐supported belief that ∼p is not properly sensitive, since the failure of one's sources to report/publicize that p reflects a failure on their part to have reported what they had, in fact, discovered.

In the light of the foregoing considerations about the reliability of silence‐supported belief, we can now explicitly address the question regarding the conditions on the doxastic justification of such beliefs. Insofar as the subject (p.255) formed her belief that p by inference from the relevant instance of (2*), her belief is justified only if she is justified in believing that the relevant instance of (2*) holds. This does not require the truth of that instance of (2*), so much as it requires whatever is necessary and sufficient for justified belief in (what is expressed by) the relevant instance of (2*). Assuming a reliabilist theory of justification, this will require that the process or method through which a hearer arrived at her belief in the relevant instance of (2*) is globally reliable.

Under what conditions would one count as globally reliable in the formation of the belief that

  1. (2*) If no source has presented‐as‐true that p, then ∼p?

Although it might be too much to ask for necessary and sufficient conditions on the reliable formation of such a belief, perhaps we can be satisfied here with sufficient conditions. A subject S reliably forms a belief in (2*) so long as S's own relevant expectations for the news are well calibrated to the sorts of news reports that are prevalent in her community.17 Other accounts of justification, of course, may demand other things of her in order for her belief in (2*) to be justified. But whatever one's favoured theory of justification, the point remains: whereas our subject cannot acquire silence‐supported knowledge unless (2*) is true, doxastic justification in these cases does not require the truth of (2*) so much as it requires that the subject's belief in (2*) be justified.18 Insofar as testimonial belief does not require that the subject have any belief in the presentation‐to‐truth conditional (1) (let alone that such a belief be justified),19 testimonial belief and silence‐supported belief receive different epistemological treatments in this respect.


Parallels between testimony and memory have been profitably explored elsewhere.20 It is popular to assimilate the two by treating one on the model of the other, for example, by thinking of memory as a matter of one's ‘past’ self testifying to one's ‘present’ self. Though I have touched on this, the plausibility of such an assimilation has not been defended here. My claim, rather, has been that there is another dimension to this assimilation—one not often described by those who liken memory and testimony—and that this dimension is one that brings (p.256) out in a particularly vivid way another form taken by our epistemic reliance on our social peers. In this connection I have been stressing the similarities between beliefs in 〈∼p〉 formed through noting that one does not remember having observed otherwise, and beliefs in 〈∼p〉 formed through noting that one does not remember having heard otherwise. With this comparison in mind, we might wonder whether the epistemology of silence‐supported belief, in those cases involving our reliance on sources other than ourselves, is fruitfully thought of as the epistemology of ‘socially extended memory’.

I begin with the least controversial of the claims that can be made on this score (which is not to say that the claim is uncontroversial!). Epistemology of silence cases involving reliance on sources other than oneself illustrate the following hypothesis in social epistemology: there are occasions on which the epistemic assessment of (the cognitive states of) a given individual will be incomplete without taking into account the dispositions and cognitive states of the individuals' social peers. The illustrating cases are those in which the members of the individual's community play a role akin to that of a perceptual and memory system: they acquire information from the world through perception, they store this information, and they transmit this information in such a way as to enable its subsequent retrieval by the subject. It is clear that cases in which the subject's peers play these roles are importantly different, and in many ways, from cases in which only her own perceptual and memory systems are in play. Even so, for at least some of a subject's beliefs, if we are to give an accurate assessment of the epistemic status of these beliefs, then the (perception‐ and memory‐like) roles played by the subject's peers must be taken into account, in a way that is analogous to the way that we take into account the performance of her own perceptual and memory systems, in cases where she relies on these. I call this ‘the Thesis.’

If we endorse a broadly reliabilist approach to epistemology, then the Thesis is true.21

Suppose that you are not particularly attentive to the way people dress, and that you do not find anything noteworthy in what others would regard as outrageous attire. So, had it been true that Jones regularly wore a bright red suit, even so you would not have attended to (or registered) this fact. Were you then to form the belief that Jones did not regularly wear a bright red suit, on the basis of the fact that you can recall having seen no such thing, your belief would not be reliable. On the contrary, the lack of coverage‐reliability of your perceptual system in the relevant domain translates into a lack of reliability in the belief you form through reliance on that coverage‐reliability.

Consider now the case involving epistemic reliance on the coverage provided by one's community. The less coverage‐reliable are the individuals on whom one is relying (relative to the domain of facts in question), the less reliable will be (p.257) any silence‐supported belief formed through reliance on the coverage‐reliability of these peers. Suppose that neither the New York Times nor CNN nor any of the other media sources on whom you rely for the news is particularly attentive to what is going on in Iraq.22 We can suppose, contrary to the facts, that without publicizing this fact the media have left Iraq after deciding not to put any resources into reporting from there any more, and after deciding not to get any reports from others (such as the Associated Press) who remain there. So, had it been true that WMDs were found in Iraq last week, even so they would not have attended to (or registered) this fact. Were you then to form the belief that WMDs have not been found to date in Iraq, on the basis of the fact that you cannot recall having come across any report of such a finding, your belief would not be reliable. On the contrary, the lack of coverage‐reliability of your sources in the relevant domain translates into a lack of reliability in the belief you form through reliance on those sources for coverage.

The point common to both cases would appear patent: the epistemic significance of ‘silence’—the epistemic significance of the fact that you recall no source having presented 〈∼p〉 as true—is in direct proportion to the quality of coverage by, and transmission through, the source on whom you are relying. This is as true when that source is someone other than yourself—as in the case of coverage by members of news organizations, where 〈∼p〉 (were it presented at all) would be presented via testimony—as when it is you yourself—that is, in memory cases, where 〈∼p〉 (were it presented at all) would be presented through a memory impression. As we saw above, in both sorts of case the main threat of unreliability in the silence‐supported belief can be decomposed into the threats of (a) unreliability in the coverage (= failure to satisfy the truth‐to‐presentation conditional, (2)), and (b) unreliability in the transmission of the source's information (= failure to satisfy one or both of the ‘transmission’ conditionals (P‐M) and (P‐T)). Admittedly, in cases involving reliance on the coverage provided by someone other than one oneself, cases are complicated by the fact that, in addition to one's source having to satisfy the transmission conditional (P‐T), one must currently be reliable in one's judgement that one did not encounter any such report in the past. But this only suggests that the case of relying on another, in addition to having the same features as cases involving reliance on oneself, has other features besides. This should come as no surprise: self‐trust appears to be a basic form of trust.23

Might we then make the stronger claim, that silence‐supported beliefs formed through reliance on the coverage‐reliability of others is akin to a case of ‘extended (p.258) memory’? In defence of this idea, we might try comparing the role of others as repositories of information with the role of what might be called personal mental prosthetics (diaries and notebooks).24 The core contention of Clark and Chalmers (1998) is that a diary could count as a case of ‘extended memory’ so long as (a) it stores information that one has acquired and information about plans for the future that one has put into it, and (b) this information stored there is readily accessible to the subject.25 Those philosophers convinced by the case from Clark and Chalmers (1998) might try seeing the role other people play in a subject's acquisition of a silence‐supported belief as that of a repository of information for the subject, a sort of ‘social’ version of the ‘extended memory’ hypothesis.

Unfortunately, the ‘extended memory’ proposal runs into difficulties with respect to both (a) and (b). It runs into difficulties with respect to (a), in that other people are not repositories for information that one oneself has put in those repositories.26 Of course it might be replied that the assumption, that memory storage is storage of information that one oneself has put into the relevant repository, is question‐begging against the proposal of a socially extended form of memory. Perhaps a better way of thinking of a subject's memorial storage is as a repository of information that she can exploit in the course of her belief‐fixation and belief‐revision. So long as information is stored in a repository in such a way that the information is sufficiently accessible to her, and in particular is accessible as needed for belief‐formation and belief‐revision, one's peers might constitute such a repository (at least by the standards of Clark and Chalmers 1998). Or so it might be argued.

But this brings us to the problems that the ‘extended memory’ proposal faces regarding (b): it would seem that the information in the minds of those on whom one is relying for coverage is not sufficiently accessible to one. In this respect it is worth bearing in mind that there is one important feature of the silence case involving reliance on others for coverage: the subject does not have access to the information stored in others' minds, but instead merely notes that she (the subject) cannot recall having received a relevant report from any of them. The result is that if we regard them as storing information for the subject, this is information that will forever be inaccessible to her (unless she can read their minds or they testify to her). It thus seems that we do not properly regard information stored in others' minds as relevantly accessible to the subject herself. (p.259)

The considerations just advanced suggest that the case has not been made for regarding our reliance on others for coverage on the model of a ‘socially extended memory’. Still, we must find some way to acknowledge the role that others play in the epistemology of silence‐supported beliefs. If it is not by way of assimilating their role into that of an ‘extended’ sort of memory, how should we characterize this role?

Here is a proposal that I favour. (I offer this in the spirit of a suggestion that is worth developing further.) We begin by noting that, in cases where a subject is assuming the coverage of others, reliable silence‐supported belief is not the cognitive achievement of any single individual. We then go on to hold that the relevant social facts that underwrite the reliability of a given silence‐supported belief—the social facts that ensure that the subject has reliable coverage in the case at hand—are all‐important background conditions on the formation of silence‐supported belief. In particular, it is against these conditions that we assess the epistemic features of silence‐supported beliefs. There is in this way a kind of division of epistemic labour. The subject herself must be justified in believing—must have reliably acquired a belief—in the relevant instance of (2*). This is a matter of her news expectations being properly calibrated to the relevant news‐dissemination practices in her community. But whether or not she is properly calibrated in this way, whatever the prevalent news‐dissemination practices in her community happen to be, it is against the background of such practices that we assess the reliability of her belief in (2*), as well as the reliability of her silence‐supported belief (inferred from her belief in (2*)). This sort of view thus regards the cognitive achievement of silence‐supported knowledge and justified belief (in cases where the relevant coverage is assumed to be provided by others) to be an achievement that is made possible by the existence of certain social arrangements, even as it denies that we should regard those arrangements as constituting anything like a ‘socially distributed memory system’.

It is worth noting that the proposal to regard social factors in this way is thoroughly in keeping with the spirit of reliabilist epistemology. After all, reliabilists have long assessed the reliability of a belief‐forming process against what Hilary Kornblith calls ‘the natural environment in which [the process itself] operate[s]’. Here is Kornblith's extended comment, made in connection with social factors:

Just as we wish to know whether our natural inferential tendencies are likely to give us an accurate or a distorted picture of the world, we also need to know whether our social institutions and practices are helping to inform us or to misinform us. And just as we need to examine our perceptual and inferential equipment against the background of the natural environment in which they operate, we also need to investigate these mechanisms against the background of the social environment in which they operate. Such investigations are straightforward extensions of the naturalistic project in epistemology.

(Kornblith 1994b: 97, italics added)

(p.260) Since Kornblith makes this point in a paper which is otherwise advocating a ‘conservative extension of naturalistic epistemology’ (1994b: 94, italics added), it seems that the present proposal should be relatively uncontroversial to anyone who is already on board with reliabilist epistemology.

No doubt, there is great variation in social practices and institutions worldwide, and even between local communities there can be relevant differences. But this fact of social variation is no objection to the proposal to regard the relevant social conditions as background conditions on silence‐supported belief. For one thing, social variation itself is a normal feature of our larger world‐environment. For another, it is a normal part of socialization within a community that one becomes sensitive to prevailing social arrangements. Included in this socialization process is the process of learning what counts around here as newsworthy, what sorts of news one can expect (and where and when one can expect this news), and so forth.27 Acknowledging all of this merely forces us to acknowledge that the background conditions that obtain in any given case depend on the community in play.

In short: it may be too much to regard silence‐supported belief, where one is relying on others for coverage, as involving a kind of ‘socially extended memory’. At the same time, we can recognize the all‐important role that a subject's peers play in ensuring the reliability of her silence‐supported belief in another way: we can think of the relevant social practices as part of the background conditions on the process(es) used in the formation of silence‐supported belief.


The epistemic significance of the silence of a relied‐upon source of information is in many respects the same, whether that source is oneself (as in memory cases) or members of one's social community (as in cases of coverage‐reliance on a news organization). To say this is to encourage the parallel between memory and testimony, in ways that go beyond the standard parallels. It is also to encourage novel thinking about testimony itself. In addition to relying on what others tell us, we also rely on what they do not: their silence can back our claims to know, and in any case brings to the fore our reliance on them for coverage. I submit that the phenomena of coverage and silence‐supported belief go alongside (p.261) that involving actual testimony, as manifesting distinct varieties of our epistemic reliance on our social peers.28


(1) Below I will note that the relevant reliability is not the reporting‐reliability of the source (as when most of the sources' reports would be true), so much as it is the coverage‐reliability of the source (as when most of the relevant truths would be reported).

(2) What should be said of memory's preserving what was originally a testimonial belief? I think these should be treated as compound cases, where one's epistemic reliance is on another subject as information source, but where the preservation of the information is a joint effort including both the testifier and the subject. The preservative role of the testifier is this: she must preserve a content she originally acquired, and perform a speech act that enables her audience to quickly and easily recover the content of the information itself. That the testifier has a role in content preservation here should not obscure that the subject herself also has such a role: the subject is relying on her own reliable apprehension of the testimony that confronted her. As argued in Goldberg (2007b), this reliability can be analysed as involving (i) reliability in discerning the force and content of the speech act (reliable comprehension), and (ii) reliability in discriminating reliable from unreliable testimony (reliable filtering).

(3) It is to Burge (1993) that I owe my use of ‘content preservation’, as covering both memory (or at least what Burge calls ‘preservative’ memory) and testimony.

(4) Indeed, it is precisely this point that motivates Goldman's well‐known (1986) distinction between belief‐dependent and belief‐independent processes.

(5) This point, asserting the possibility of failing to come across testimony proffered by a source we (standardly) rely on, is insignificant from the perspective of the usual issues that come up in the epistemology of testimony. For the reliability of a source's testimony is independent of how likely it is that her intended audience will come across that testimony. However, if her intended audience is relying on her for coverage, the likelihood that this audience will fail to come across her testimony is all‐important. See Goldberg (forthcoming a) for further discussion.

(6) Thanks to Jesper Kallestrup for this suggestion.

(7) I introduced these in Goldberg (forthcoming a).

(8) I am assuming something like a reliabilist picture here. See Goldberg (2007b) for an extended defence of this picture.

(9) Where ϕ is some source on which subject S relies throughout interval i for coverage in domain D, we can say that ϕ is ‘silent’ regarding p (some proposition in D) throughout i when throughout i ϕ has not issued any report that bears directly on the question whether p. (To be sure, in the standard cases it is not the actual silence of a source, so much as it is the subject's belief in that silence, that will be relevant to an assessment of subject's beliefs formed on the basis of the (alleged) silence of a given source.) We have here the core of the issue before us: when you believe that Jones did not regularly wear any bright red suit, or that to date WMDs have not been found in Iraq, what is the epistemic bearing on these beliefs of the (alleged) ‘silence’ of a source on whom (or which) you relied for reliable coverage?

(10) The restriction to ‘relevant’ truths makes this realistic. After all, even as venerable a daily paper as the New York Times only promises to publish ‘All of the News That's Fit to Print’.

(11) In saying this I do not mean that testimony should be treated as evidence; that would be a mistake for reasons developed in Goldberg (2006).

(12) My nomenclature for claims is as follows: claims will be designated as ‘(X‐Y)’, where the term substituting for ‘X’ will tell which sort of reliance is in question (‘S' for reliance on a Source, ‘P’ for reliance on the part of the system that Preserves contents acquired by the source), and the letter substituting for ‘Y’ will tell what type of case it is (‘M’ for Memory case, ‘T’ for Testimony case).

(13) To a first approximation, these cases might be analysed as follows. A belief is acquired at t; this acquisition causes a modification in one's memory system (the belief is put in memory; perhaps there is a process involving transmission first to short‐term memory, then to long‐term memory); and this modification in the memory system is causally responsible for the subsequent sustainment of a belief with a content distinct from the original belief. Presumably, the sustained content is related to the original content in such a way as not to elicit any suspicion of a content‐shift on the part of the subject herself. (It is noteworthy that some content externalists have theories of the semantics of memory on which content‐shifts of this sort are to be expected under certain conditions. See Ludlow (1995), Gibbons (1996), and Goldberg (2007c) for a discussion.)

(14) While such a possibility is clear in the abstract, it would appear to be a substantial empirical question whether, and if so to what extent, human memory is susceptible to such failures. Suppose 〈p〉 is a proposition that is originally the content of a belief we acquired at some earlier time, but where memory shifts the content of this belief, so that on recollection we have the memory impression that q (where 〈q〉 and 〈p〉 are logically independent of one another). Then, when at some later time one considers whether p, it may be that, despite the fact that one presently has no recollection of 〈p〉, one still has the ‘sense’ that it would be wrong to conclude that ∼p on this basis. So even if we do suffer here from a failure of content‐preservation, our so suffering would not incline us to form the corresponding ‘silence’‐based belief.

(15) Here I note that (S‐T) does not identify any particular source; it speaks only of ‘one of the sources on whom I rely’.

(16) In this respect the motto of the New York Times is noteworthy for its coverage‐related ambition: ‘All of the News That's Fit to Print’.

(17) See Goldberg (forthcoming b) for a full presentation of this idea.

(18) Of course, a justified silence‐supported belief would also require S to be sensitive to the likelihood that S herself would come across any relevant report were one to be made. See Goldberg (forthcoming a).

(19) See Goldberg (2007a: ch. 5) for details.

(20) See e.g. Burge (1993), Dummett (1994), Christensen and Kornblith (1997), Lipton (1998), Edwards (2003), Lackey (2005), and Goldberg and Henderson (2006).

(21) I develop this claim at length in Goldberg (forthcoming b).

(22) To make the case do the job I want it to do, we have to imagine that any relevant facts would only be reported by the news sources on which you rely. No doubt, this assumption is unrealistic. But cases are easily imaginable where such an assumption would be realistic: suppose the topic is one where there are only few people interested in the topic, and only one news source that covers such a topic, etc. I will disregard the unrealistic nature of my example in what follows.

(23) This is a point made in Lehrer (1997) as well as in Foley (2006).

(24) Here, and in the suggestions that follow, I am indebted to Alan Millar for very helpful comments.

(25) I should perhaps acknowledge that, while I find the hypothesis of extended cognition to be an intriguing one, I am dubious about its truth. I offer the present argument to those who do not share my scepticism on this score.

(26) That is, other people are not such repositories unless the subject herself testified to them, and they go on to store the subject's information. This is not the sort of case we are envisaging here.

(27) Indeed, so far as I can tell, the epistemological issues surrounding the notion of newsworthiness have received no attention whatsoever from epistemologists (as opposed to folks in journalism schools). I regard this as a lacuna in the epistemology literature. I hope to address this at some future time.

(28) I would like to thank Frank Döring for suggesting to me the interest (for social epistemology) of cases like the ones considered here—where you reason that if such‐and‐such were true, you would have heard about it by now. I would also like to thank the members in the audience at a conference on Social Epistemology at Stirling University, where I presented a paper on a similar theme: Jessica Brown, Igor Douven, Miranda Fricker, Alvin Goldman, Peter Graham, Klemens Kappell, Jon Kvanvig, Jennifer Lackey, Peter Lipton, Nenad Miscevic, Alan Millar, Erik Olsson, Duncan Pritchard, Ernie Sosa, and Finn Spicer, for particular comments on that paper. Special thanks to Jasper Kallestrup, who served as my commentator at that conference, and whose suggestions inform the present chapter in a variety of ways; and to Alan Millar, for extensive comments on an earlier draft of this chapter.