Author Information: Leonie Smith, University of St. Andrews, firstname.lastname@example.org
Smith, Leonie. “Challenges and Suggestions for a Social Account of Testimonial Sensitivity.” Social Epistemology Review and Reply Collective 5, no. 6 (2016): 18-26.
The SERRC thanks the contributors and participants—especially William Tuckwell—at the Tartu Graduate Conference in Social Epistemology, at the University of Tartu on 26-27 March 2016, for allowing us to publish selected papers. We will bring these papers, and subsequent replies, together in a special issue of the SERRC.
Image credit: Christine McIntosh, via flickr
Recent work on epistemic injustice has re-ignited the importance of sensitivity-analyses of knowledge, via the possibility of audiences being insufficiently sensitive to the testimonial credibility of prejudiced-against speakers. The focus of this work has quite rightly been on demonstrating the epistemic harms to potential testifiers when sensitivity fails to apply. However there is clearly a corollary impact on audiences, and their ability to achieve knowledge from testimonial sources, when they fail to apply appropriate conditions of sensitivity with regard to the testimony of others. The notion of being appropriately-sensitive, implicitly assumes that we can make sense of what these sensitivity conditions on testimonial-knowledge formation might be.
In this paper, I therefore go back to basics, and put the claim that sensitivity is a necessary condition for knowledge under pressure, by considering its applicability with regard to testimonially-formed beliefs. Building on, and departing from, Goldberg, I positively draw out how we might understand the required sensitivity as a social interaction between speaker and hearer in testimonial cases, in sections one to four. In doing so however, I identify a concern in section five which places the whole notion of testimonial sensitivity in potential jeopardy. I find an apparently paradoxical inverse relationship between better-differentiating methods that fulfil sensitivity conditions, and being able to have sensitive beliefs with regard to specific instances of testimony. After examining potential resolutions in section six, I conclude that only a focus on minimally realising the problem, rather than “solving” it, will enable us to retain sensitivity as a necessary condition for testimonial-knowledge.
(1) Sensitivity and Knowledge
As a response to the problem of luckily true belief, sensitivity analyses of knowledge require that our beliefs externally “track the truth.” Nozick proposes two conditions for the sensitivity principle:
(1) “If p were false then S would not believe that p, via method M,” and;
(2) “If p were true, then S would believe that p, via method M,” where “M” distinguishes that the same method has been used to generate the belief, as would be used in the counterfactual case.
Typical criticisms of sensitivity as a necessary condition for knowledge are that it breaks the closure principle; it allows for “abominable conjunctions,” where I both know that I have hands but not that I am not a brain in a vat without hands; it attributes knowledge in situations which are not knowledge (Kripke’s famous reliable red barn’s in fake barn county); and that it does not coincide with our intuitions of the existence of knowledge in certain non-sensitive cases.
Arguably however, none of these objections successfully demonstrates that sensitivity is not a necessary condition for knowledge. Non-exhaustively, plausible responses to these challenges include that: The closure principle might be reclaimed by a different interpretation of method individuation, or by sophisticated contextualist analyses; the abominable conjunctions are acceptable, in light of sensitivity’s ability to “get the cases right”; we could add additional necessary and sufficient conditions to “false knowledge” cases, or better individuate method; and that there may be answers to the problem of fallibility raised by Sosa (I will return to this example later). Plausibly, some combination of these responses would prevent a knock-down argument against sensitivity’s necessity for knowledge, even if it did not form an entirely sufficient set with “true belief.”
An area of concern which seems less exhausted in the literature however, is that of the “method M.” This inclusion is necessary to prevent counterfactuals in which the truth of the antecedent affects the method by which I come to know it. But in raising “method,” “…we leave large questions open about how to individuate methods… identify which method is at work, and so on.” This becomes particularly difficult to resolve in the case of testimonially-based beliefs; those beliefs at the heart of cases of epistemic injustice. In discussing methods for forming truth-tracking beliefs about necessary truths, Nozick asserts that; “Mathematical truths raise no special questions about learning from others”. This brief aside notwithstanding, I suggest that Nozick has not considered the specific challenges that “learning from others”—forming belief through testimony—poses for the sensitivity principle.
(2) Testimonial Method: General vs Individual
It is highly debatable whether there could be an “epistemically significant fact of the matter regarding the general reliability of testimony.” We might broadly construe testimony as relating to speech acts, written documentation, word of mouth etc., where all that they have in common is the result that: “…a hearer form a given belief on the basis of the content of a speaker’s testimony.” This broadness should, I suggest, make us wary of taking on a catch-all interpretation of “method,” that is independent of facts about the testimonial source itself. Goldberg seeks to motivate that concern; that “method” cannot be only “testimony,” broadly construed.
Assume a generic interpretation of “method” as “via testimony,” M(T-GEN). According to the Nozickian conditions, S therefore knows that p through testimony iff, if p were false, S would not believe that p by M(T-GEN), and iff, if p were true, then S would believe that p, via M(T-GEN). Goldberg applies a few tweaks to cater for the impact of other potential testifiers and cases in which S forms no belief regarding p, but broadly M(T-GEN) does not require any real ad hoc adaptation of what “method” might mean in the sensitivity conditional for testimony, versus in the case of perception. However, there is a problem with M(T-GEN). Consider ROOM:
S enters a room in which, unbeknownst to S, each member is an expert liar who actively seeks to deceive others, with the exception of person T, a highly reliable, sincere and competent testifier. S just so happens to encounter T, who informs her that p, which is in fact true. From this, S comes to believe that p.
The closest counterfactual situation may be one in which p is true, but S encounters a liar, L. In this case: L informs S that NOTp; S believes that NOTp; and S’ belief is incorrect. Thus in ROOM above, S does not know p (in agreement with our intuitions), as her belief fails the second condition of sensitivity: “If p were true, then S would believe that p, via method M,” when her method is understood to be M(T-GEN). This appears to vindicate the individuation of “method” in testimony cases as M(T-GEN). However, the closest counterfactual situation may instead be one in which NOTp is true, and S continues to encounter T. In this case, T informs S that NOTp, S believes that NOTp, and her belief is correct. If M(T-GEN) is her method, her belief now passes both the conditions for sensitivity, entailing that S does know p in ROOM, but this appears to be the intuitively wrong result. This suggests that M(T-GEN) is not an appropriate candidate for “M” in cases of testimony; it should be adapted or replaced.
Goldberg takes the latter option. Building on work by Graham, he proposes that the reliability of T (or L) appears essential to whether or not S’ testimonially-based belief counts as knowledge, not S’ sensitivity to the truth via M(T-GEN). Graham presents two cases of the boy who cried wolf: REFORMED and ARBITRARY. In REFORMED: The boy T who falsely cried wolf in the past, sees an actual wolf, and is shocked into reforming his character so that he will henceforth only report “wolf” when there is an actual wolf present. He cries “wolf” to S, who now has the true belief that there is a wolf. In ARBITRARY: L just enjoys making exciting reports based on a coin toss, whether there is a wolf present or not. He sees a wolf, and cries, “wolf” to S, because his coin came up heads, not because there was actually a wolf present. S now has the true belief that there is a wolf.
Graham wants to say that in REFORMED, S has knowledge, but that in ARBITRARY, S does not have knowledge, despite the fact that in both cases S formed the same true belief. The lesson, Graham proposes, is that what matters to whether or not S’ belief is knowledge is, “the disposition underlying the boy’s report and whether it will underwrite the truth of the subjunctive.” Goldberg now takes this on. M(T-GEN) fails to individuate between these cases. But a new principle M(T-IND), considers method at the individual level—by what method did the testifier (X) come by their belief? Goldberg fleshes this out:
… the relevant testimonial method involves a specification not only of the particular speaker herself, but also of the very processes that eventuated in the proffering of the testimony itself.
He suggests that the correct thing to do is to therefore abandon M(T-GEN) in favour of M(T-IND) as the criteria for sensitivity in testimony-based belief-formation.
(3) A Problem with this Account
I agree that Graham and Goldberg’s instincts to demonstrate how M(T-GEN) alone fails to distinguish cases of testimonial knowledge, and to include the dispositions, or general reliability, of X in a sensitivity method for testimony are correct. I believe however that the knowledge-intuitions they rely on are incorrect, and that in this instance, this leads to a premature and incomplete conclusion. “Method” should be individuated in a way that captures both the role of X, and of the hearer, S, in belief-formation.
Consider, PRE-REFORMED, identical to REFORMED other than the cause of the boy’s reformation:
T* hears a story about a neighbouring village, in which the villagers fail to believe a girl who falsely cried wolf previously, and are all killed. T* realises the error of his ways, and reforms before ever seeing a wolf. As such, T* would not have told S that p (there is a wolf), had there been no wolf on this occasion.
This, I suggest, is closer to being a case in which S has knowledge, than in REFORMED. In PRE-REFORMED, if p were false, T* would not report that p, and S would not believe that p. Temporarily ignoring “method,” it seems intuitively true that S knows that p in this case. In REFORMED, if p were false, T would report that p, and S would falsely believe that p. Goldberg wants to say that the method has changed (X is not disposed to tell the truth unless p is true), and so S has knowledge in REFORMED.
But it simply seems wrong to allow that whether or not S has knowledge has nothing to do with the reliability of X’s dispositions in the most obvious counterfactual case (that in which no wolf appears). And the reliability of X’s dispositions, is something that S is required to judge in forming her beliefs; it is part of the “method” used that S’ attitude towards X be relevant. To accept that REFORMED is not knowledge, but attempt to retain M(T-IND), Goldberg could alternatively claim that what is relevant is not only X’s disposition, but when and how X formed the disposition, but I suggest that this takes us into increasingly ad hoc attempts to define method in terms which intuitively lack an importantly relevant fact: S’ attitude towards X as a testifier.
(4) Testimonial Method: Hearer and Speaker
“Method” then, is not M(T-GEN), which excludes sensitivity to the disposition of X, but nor is it M(T-IND), in which method is individuated purely according to X’s general reliability or dispositions. Rather, “method” involves S’ sensitivity to X’s dispositional reliability; both parties’ attitudes are relevant, as both parties play a role in producing knowledge from testimony. Call this method, M(T-SX). Via this method, S’ belief is sensitive to X’s dispositional reliability in PRE-REFORMED and thus counts as knowledge, which seems correct. In REFORMED, if p had been false, S would still have believed that p, under M(T-SX), and so S lacks knowledge in the true-belief case, as I have argued. And S would continue not to have knowledge in ARBITRARY, under M(T-SX), which also seems correct.
What backs this interpretation of method up, as a non-ad-hoc solution? Firstly, Nozick’s advice is that method must be the most reliable (relevant to the belief) available to the agent. And that “…any method experientially the same… ‘from the inside,’ will count as the same method.” But in REFORMED and ARBITRARY, if we accept M(T-IND), S’ method presumably feels identical “from the inside,” and a difference is not available to her. We do not have to believe that S’ method is entirely transparent to her, but it is valid to ask in what way S has sensitive beliefs, if the beliefs she would actually form under the counterfactual, are irrelevant.
But secondly, we can motivate this further, by considering another unique aspect of testimonial belief. The sensitivity condition contains a second proposition (2): That if p were true, I would believe that p via method M. When we attempt to individuate method purely with regard to T’s dispositions and actions, we fail to see the active role of S in recognising testimonial-sources. The reality of testimonial-belief is that we often have little knowledge of the source. As such, “a judgement of credibility must reflect some kind of social generalisation about the epistemic trustworthiness … of people of the speaker’s social type.”
To prevent prejudicial attitudes blocking belief-formation, there is a demand on the hearer to develop, “a rational sensitivity,” which Fricker cashes out in terms of epistemic virtues. Without this, “… a hearer may be so constituted as to prevent the epistemic properties of a speaker’s belief from being transmitted to her.” For condition (2) to hold then, S’ method must place demands on S, with regard to T. Of the options considered, only M(T-SX) allows for this, whilst avoiding the problem cases Goldberg and Graham introduced.
I have argued then that the method used in sensitivity in testimonially-based beliefs ought to reflect the interaction of S with X, in belief-formation. We may individuate method as “via the testimony of X, formed in an epistemically virtuous way, responsive to the credibility of X,” and in doing so, plausibly find that sensitivity is a necessary condition for knowledge in the case of testimony.
(5) A Problem: The Otherwise-Reliable Liar
In producing this interpretation of method, one which essentially individuates to increase reliability of method, I believe that we also introduce a problem for testimonial assessment however. I take Sosa’s broader objection to sensitivity as a springboard: That there is potential for “incredibly rare” events to entail that belief may not be sensitive to truth, in cases where we would want to claim that our evidential beliefs were constitutive of knowledge. In CHUTE, S’ belief that p (the rubbish will reach the basement), when I release it down the chute appears to be a case of knowing that p. But if it were snagged on the way down, (¬p), S would still believe that p. S’ belief is therefore insensitive to the truth of p, and according to sensitivity, ought not to count as knowledge, despite our intuition otherwise. Others might alternatively cash this out as a problem for all internally individuated methods.
We could respond generally to Sosa. Perhaps rare occurrences are not what happens in close possible worlds. Alternatively, if it were a common occurrence, then (i) S would be unlikely to believe that p in this actual world, or (ii) we could reasonably bite the bullet and say that S does not have knowledge. Sosa’s criticism perhaps only plays with our intuitions about method, which can be improved by sharpening up our definitions and assessment of it. But these responses do not stretch to all beliefs based on the testimony of others.
Adapting Comesana, imagine a scenario, HALLOWEEN 2. Judy has the job of directing people to Andy’s Halloween party. Judy is someone I have no reason not to trust. She has no objective predisposition to lie generally, I know that she’s a friend of my good friend Andy, and she herself wants Andy’s party to be a great success as she will be attending. Judy will be dressing up as a traffic-warden once she arrives at the party. I see Judy and she truthfully directs me to the party. However, unbeknownst to me, Judy has had a bad day, and although not premeditated, and not in character for her, had she met anyone dressed as a traffic-warden, she would have given them false directions to prevent them upstaging her. I considered dressing as a traffic-warden, but couldn’t find my outfit, and dressed as a penguin instead. I take Judy’s directions, form a true belief, and get to the party.
In a close possible world then: My belief about the location of the party would have been false; I would have relied upon the same highly reliable method, M(T-SX), to form my incorrect belief; and I would not know the location of the party. Yet, it seems that I do know where the party is in this world, despite the fact that if the directions were false, I would still have believed they were true. This generalises: A person X who has a disposition to tell the truth, features in S’ reliable method as a source of information. But suppose that X has reason to tell S, that p, just on this occasion, regardless of the factivity of p. Then S’ belief is “counterfactually insensitive to the truth.”
This makes testimony’s sensitivity problem more uniquely damaging than is first apparent. To understand why: In the case of perception, if S has truth-tracking beliefs via a highly reliable method, then arguably only sceptical scenarios and / or extreme bad luck (CHUTE) would make her belief insensitive. S’ beliefs become more sensitive to the truth of p via method M, when M is increasingly reliable, in the closest possible worlds.
In the case of testimonial foundations however, the opposite is true. Testimonial sources are exactly the kind where in ordinary situations, the counterfactual might easily have been the case. X’s truth-telling likelihood does not supervene on X’s knowing, or even fallibly but sensitively believing the truth himself, but on all of the factors affecting X’s psychology at the time of uttering testimony. More bluntly: people lie. As such, unlike in CHUTE, these types of events are not rare occurrences, and do occur in close possible worlds. But paradoxically, as S’ method includes nurturing epistemically virtuous beliefs, based in part on the disposition and behaviour of X, the more reliable X is in the general case, the less S will question those beliefs in any token instance, and the more likely that “if p were false, S would still believe that p.” S’ belief in fact becomes less sensitive in any given instance, the more sensitive her general belief-forming method actually is.
(6) Possible Resolutions
I will end by considering just three possible ways in which we might try to resolve the problem, highlighting the kind of considerations that would be needed in any solution.
First, can we argue that there has been a change of method, in HALLOWEEN2? Taking a cue from Graham, the idea might be that we ought to add “that T would not mislead me on this occasion” into the method used because, “grounds that p are those that establish the fact that p or guarantee the truth that p.” As such, the cases where T would mislead involve a different method. However asking for a “guarantee” of counterfactual factivity in our method is dangerously close to saying that what we need is not fulfilled by the idea of generally truth-tracking evidence—that T has never lied previously, that we are not living in a totalitarian dictatorship with state-controlled media, that T is disposed to tell the truth etc.—but only by knowledge of whether in fact T would tell the truth if p were false in this specific instance. This relies on an assessment of what T would do that is potentially inaccessible even to T, and certainly inaccessible to S. We track via method, “otherwise only God could satisfy the condition that if p were true one would believe it.” By micro-individuating, we seem to have created a definition of method that could only be fulfilled by some perfect observer, and we deny the validity of sensitivity at all.
Second, we might consider whether testimonial issues are closer to the problem of how beliefs about necessary truths might be sensitive to knowledge, in that “they can’t be,” at least with regard to condition (1). As such, might we similarly abandon condition (1) and retain condition (2)? It is hard to see this as anything other than an extremely ad hoc solution here however, and one which undermines the purpose of introducing truth-tracking at all. In the case of necessary truths, we remove condition (1) because the antecedent is necessarily false. The same is not true for testimony, and this should also be dismissed.
Third, a final more radical approach is to hit the problem head-on, and deny it. Non-reductionist interpretations of testimony by which we are entitled to trust testimony sui generis seem suspect. But we might be more optimistic regarding the possibility of some kind of appeal to a general testimonial tendency towards individual factivity. This may involve, for example, default positions from a state of nature, or “commonsense psychological concepts.” I will not comment here on how plausible these types of solution might be, but I suggest that if we are to save sensitivity as a necessary condition for testimonial knowledge, whilst retaining the insight that method requires some kind of two-way relationship between both hearer and speaker, it could only be by undermining the scale of the concern through one of these methods. We need to find a way to turn testimony into a belief-source that has no greater potential to be untrustworthy via reliable methods than any other source; there is no obvious “fix” to the sensitivity conditions, once the problem has actually been admitted, because all fixes to enhance general sensitivity, only serve to exacerbate the issue.
It is not new to highlight that justification for testimonial knowledge is a contested area. But I hope to have shown that (i) testimonial knowledge entails methods which involve both speaker and hearer, but that (ii) principles which seek to improve the reliability of our testimonially-based beliefs, by their nature, reduce our sensitivity to cases in which T simply misleads on any occasion; the problem of the otherwise-reliable liar. Short of full epistemic access to the content of other minds, the more we take care to be sensitive in our belief-forming method, the less sensitive belief can be in any individual case. As such, it seems likely that making coherent sense of sensitivity as being necessary for knowledge will require a pragmatic response which provides assurance that the problem of the otherwise-reliable token-instance liar does not matter, or is realistically only minimally realised; not that it can be “solved.” Only if we can find one, is sensitivity plausibly a necessary condition for testimonial knowledge.
Adams, Fred and Murray Clarke. “Resurrecting the Tracking Theories.” Australasian Journal of Philosophy 83, no. 2 (2005): 207-221.
Austin, J.L. How to Do Things with Words. Cambridge: Harvard University Press, 1962.
Becker, Kelly. “Methods and How to Individuate Them.” In The Sensitivity Principle in Epistemology, edited by Kelly Becker and Tim Black, 81-98. Cambridge University Press, 2012.
Craig, Edward. Knowledge and the State of Nature: An Essay in Conceptual Synthesis. Oxford University Press, 1999.
Comesana, Juan. “Unsafe Knowledge.” Synthese 146, no. 3 (2005): 395-404.
DeRose, Keith. “Solving the Sceptical Problem.” The Philosophical Review 104, no. 1 (1995): 1-52.
Dotson, Kristie. “Tracking Epistemic Violence, Tracking Practices of Silencing.” Hypatia 26, no. 2 (2011): 236-57.
Fricker, Elizabeth. “Against Gullibility.” In Knowing From Words, edited by Bimal K. Matilal and Arindam Chakrabarti, 125-161. Kluwer, 1994.
Fricker, Miranda. Epistemic Injustice. Oxford University Press, 2007.
Goldberg, Sanford. “Sensitivity from Others.” In The Sensitivity Principle in Epistemology, edited by Kelly Becker and Tim Black, 43-65. Cambridge University Press, 2012.
Graham, Peter J. “Transferring Knowledge.” Noûs 34:1 (2000): 131-152.
Ichikawa, Jonathan. “Quantifiers, Knowledge and Counterfactuals.” Philosophy and Phenomenological Research 82, no. 2 (2011): 287-313.
Lackey, Jennifer. “Knowing from Testimony.” Philosophy Compass 1, no. 5 (2006): 432-448.
Leite, Adam. “Skepticism, Sensitivity, and Closure, or Why the Closure Principle is Irrelevant to External World Skepticism.” Croatian Journal of Philosophy 4, no. 12 (2004): 335-350.
Nozick, Robert. Philosophical Explanations. Belknap Press of Harvard University Press, 1981.
Pritchard, Duncan. “Sensitivity, Safety, and Anti-Luck Epistemology.” In The Oxford Handbook of Skepticism, edited by John Greco, 437-455. Oxford University Press, 2011.
Sosa, Ernest. “How to Defeat Opposition to Moore.” Noûs Vol. 33, Supplement: Philosophical Perspectives, 13, Epistemology (1999): 141-153.
Williamson, Timothy. Knowledge and Its Limits. Oxford University Press, 2000.
 Miranda Fricker 2007; Kristie Dotson 2011.
 Sanford Goldberg 2012.
 Robert Nozick 1981, 179.
 Keith DeRose 1995, 28.
 Ernest Sosa 1999, 145-46.
 Duncan Pritchard 2011.
 Jonathan Ichikawa 2011.
 Particularly given that we have no agreed response to the failure of closure in e.g., paradigm lottery cases either (Kelly Becker 2012, 85).
 Fred Adams and Murray Clarke 2005.
 Nozick 1981, 184.
 He follows this up with an interesting footnote discussion about restrictions on knowledge originating in transmission from insensitive sources (Nozick 1981, 187).
 Jennifer Lackey 2006, 440.
 Lackey 2006, 433.
 Goldberg 2012, 44.
 Goldberg 2012, 45-50.
 Goldberg 2012, 47.
 Goldberg adds other scenarios for the counterfactual, to demonstrate the scale of the problem (48).
 If your intuitions are that in fact S does know p in ROOM, then the earlier counterfactual involving L produces the “wrong” outcome. Either way, we have a conflict for M(T-GEN) as “method.”
 Peter J. Graham 2000.
 The cases are Graham’s, the titles are Goldberg’s.
 Graham 2000, 141-42.
 Graham 2000, 142.
 Goldberg 2012, 56.
 Goldberg’s result now provides the “wrong” result in the case of ROOM. He chooses to determine that we either do in fact have knowledge in ROOM, or that we might add defeaters to save M(T-IND) (2012, 63-64). As I do not intend to save M(T-IND), I will put this concern, and considerations, aside.
 T could not have had the disposition to tell the truth if not p, as seeing the wolf triggered his transformation. The situation would of course be different for S’ second instance of deciding whether that p, as T would be in the same dispositional state regardless of whether p, having been scared by the wolf already.
 Nozick 1981, 265.
 Nozick 1981, 184.
 It could be externally verified, or some form of hybrid interpretation of Nozick (Becker 2012, 91-94).
 Fricker 2007, 32.
 Fricker 2007, 69.
 This faculty also presents bias in the other direction; that of the compulsive believer with respect to X (Lackey 2006, 436).
 Sosa 1999, 145-46.
 Timothy Williamson 2000, 156; Pritchard 2011, 52-53.
 Adam Leite 2004, 340.
 E.g., Becker: “Perhaps S knows only that it is very likely that the bag will drop… in the closest possible worlds where it is not even likely that the bag will drop, I do not believe that it is likely, given my actual belief-forming method, for example, believing that it is likely that the bag will drop because chutes normally work and I have no evidence to the contrary, because things would have been very different in that world, and I would have noticed it” (Becker 2012, 87, footnote).
 Juan Comesana 2005, 397.
 Lackey 2006, 437.
 Graham 2000, 132.
 Adams and Clarke 2005, 214-15.
 This is of course, a variant on Williamson’s idea. If “method” is a “way of believing” (Williamson, 2000, 153), then we cannot incorporate “always only knowing how X would act in cases where p were false” into that phrase in a meaningful way. It seems to stretch the plausible externalism of sensitivity beyond any sense of meaning to say that “method” should include counterfactual truths (even if we can make sense of what this might mean).
 Nozick 1981, 186.
 Edward Craig 1999.
 Elizabeth Fricker 1994, 156.