1. INTRODUCTION
In Relying on Others, Sanford Goldberg considers interpersonal epistemic reliance – and testimony in particular – in relation to the general framework of process reliabilism (as introduced in Goldman Reference Goldman and Pappas1979, Reference Goldman1986). Process reliabilism has been applied to social epistemology in various contexts. But Goldberg provides an original angle by connecting the social epistemology of testimony to a particular long-standing problem for reliabilists: the problem of how to individuate cognitive processes for epistemic purposes. The general connection of these issues and Goldberg's specific take on it deserve attention.
Towards the end of the introduction, Goldberg states the three main aims of the book:
(i) to assuage any remaining concerns people might be having regarding the legitimacy of the project of social epistemology; (ii) to develop a reliabilist program for how at least some such research might proceed; (iii) to make clear that some epistemic relevance of the social goes beyond the status of knowledge itself. (9Footnote 1)
While Relying on Others largely succeeds in contributing to at least some readings of (i) and (iii), I will argue that it is less successful with regard to (ii). Moreover, since (ii) may reasonably be regarded as the most central and controversial of Goldberg's three aims, I will focus on it.
Goldberg characterizes the view that he seeks to replace, Orthodox Reliabilism, as the conjunction of Process Reliabilism and what he labels the doctrine of Process Individualism (PI):
PI For every subject S, all of the cognitive processes implicated in the formation or sustainment of S's beliefs are cognitive processes that take place within S's own mind/brain. (44)
Goldberg does not provide a full account of process-individuation. For example, he sets aside the generality problem (82). Rather, he argues for an alternative to Orthodox Reliabilism that consists in replacing PI with the extendedness hypothesis:
The testimony itself, along with the cognitive processes implicated in the production of that testimony, are more appropriately regarded as part of the testimonial belief-forming process itself. (79)
In chapters 1–4, Goldberg argues for the extendedness hypothesis, and in chapter 5 he defends it against various objections. In chapter 6, the focus shifts to an interesting phenomenon that Goldberg labels ‘coverage reliability of one's community’ in which ‘reflecting on whether p, one comes to believe that ~p on the grounds that if p were true one would have heard about it by now’ (154). In chapter 7, Goldberg concludes by considering the extendedness hypothesis as a reliabilist framework for social epistemology.
Space does not allow for even a superficial discussion of all the arguments, cases, distinctions and theses that fill the pages of Relying on Others. So, I will be very selective and discuss only some of Goldberg's main aims, assumptions and arguments.Footnote 2
The book's central dispute can be illustrated with a twin case that contrasts ‘good’ and ‘bad’ cases (14–15). In the good case, a hearer H generates a true testimonial belief that p on the basis of the testimony of a competent and sincere speaker, S. In the bad case, a twin-hearer, H*, generates a true testimonial belief that p by accepting the testimony of a speaker, S*, who seeks to mislead H* but accidentally expresses a true proposition.
Goldberg concludes, plausibly, that H acquires testimonial knowledge and that H* does not. The central dispute, however, concerns Goldberg's diagnosis, according to which H* fails to know because H* is not justified.Footnote 3 According to Goldberg, this is because the epistemically relevant process extends to include S*'s cognitive processes involved in generating the testimony. However, Goldberg recognizes that Orthodox Reliabilists may argue that H*'s belief is as justified as H's belief but that the case is a Gettier-style case in which H*'s process is locally unreliable (Goldman Reference Goldman1986). Consequently, much of Relying on Others is devoted to arguing that the extendedness account is superior to the Orthodox Reliabilist account. Below I argue that some of Goldberg's central arguments are unlikely to convert Orthodox Reliabilists.
2. AN ALLEGED ANALOGY BETWEEN TESTIMONY AND MEMORY
In the above-mentioned ‘bad case’, H* fails to know. This much is common ground between Goldberg and his opponents. But Goldberg's diagnosis, that this is because H*'s extended belief-generating process is unreliable, is controversial. This diagnosis, in turn, requires that the processes involved in generating testimonial belief are epistemically similar to those involved in memory and inference.
Goldberg sensibly considers this putative analogy in terms of Goldman's distinction between belief-dependent and belief-independent processes (Goldman Reference Goldman and Pappas1979, Reference Goldman1986). A central common denominator for memory and inference is that they are belief-dependent processes. In each case, the relevant input is a belief. This is epistemically important. Simplified a bit, Goldberg argues that unless the belief that serves as input for memory is itself generated by a reliable process, the memory-generated belief will not amount to knowledge – even if the memory process is impeccably reliable. So, in the case of memory (and inference), the epistemically relevant process extends to include the processes that provide the input.Footnote 4 So, Goldberg's similar account for testimony requires an analogy between testimony and memory in this regard.
However, Goldberg correctly notes that the inputs of testimonial processes are not beliefs (68–71). But the conclusion he draws from this insight is not that testimony is not belief-dependent. Rather, Goldberg concludes that what is needed is ‘a more liberal understanding of the notion of a belief-dependent process – one that allows, for example, another's testimony to serve as an input’ (71–2).
At this point, an Orthodox Reliabilist may reasonably worry that Goldberg is, in effect, tailoring the notion of belief-dependence, eventually labeled quasi-belief-dependence, to his purposes. Moreover, I found Goldberg's liberalization of belief-dependence to quasi-belief-dependence to reinforce this worry. Goldberg claims that
the core idea behind a belief-dependent process is that of a process the reliability of whose outputs depends on the reliability of its inputs. This, after all, is the idea behind the GIGO (‘Garbage In, Garbage Out’) principle that motivates the reliabilist's talk of belief-dependence. (72)
It may not be accurate that GIGO is the idea that motivates Orthodox Reliabilists' talk of belief-dependence. The fact that belief-dependent processes are attributable to the individual is, at least for some Orthodox Reliabilists, another motivation. But Goldberg takes the GIGO-consideration to motivate three individually necessary and jointly sufficient conditions on quasi-belief-dependence:
First, such a process must be a cognitive process the reliability of whose outputs are a function of the reliability of its inputs. This, in turn, requires that the process should have inputs, and that these inputs be assessable in terms of their reliability. And, in order to be strict about what it is for an input to be assessable in terms of its reliability, we will insist that an input satisfies this condition only if it (the input) is itself the output of a cognitive process (process-type) whose reliability can be assessed in its turn. (72)
Goldberg claims that the three conditions are sufficient for quasi-belief-dependence by noting that ‘the conditions are demanding enough to ensure that any process that satisfies them has some reliability-assessable input’ (73). Goldberg then argues that testimony satisfies the three conditions. This last step is plausible. But the worry persists that this is largely because Goldberg's three conditions jointly set forth an excessively inclusive sufficient condition on quasi-belief-dependence.
To see this, note that if perceptual beliefs are generated on the basis of perceptual representations that are the reliability-assessable outputs of cognitive processes, certain perceptual processes qualify as quasi-belief-dependent.Footnote 5 This result would seem to be in tension with Goldberg's attempt to repudiate ‘the perceptual model of testimonial belief’ (77 et passim). Furthermore, this putative overgeneralization worry augments the concern that the notion of quasi-belief-dependence is overly inclusive. So, I think that Orthodox Reliabilists may reasonably eschew the notion of quasi-belief-dependence as ad hoc and too liberal.
Note that Orthodox Reliabilists may avoid Goldberg's liberalization and uphold the view that belief-dependent processes such as memory and inference extend, for the purposes of reliability assessment, to the individual's cognitive processes that produce their input-belief. As mentioned, a partial rationale for this view is that the processes that produce the input-belief are attributable to the agent. Goldberg's liberalization of belief-dependence to quasi-belief-dependence does not respect this difference. It strains the analogy between testimony and memory in this regard. So, given their antecedent commitments, Orthodox Reliabilists have little reason to adopt the analogy.
Of course, a central project of Relying on Others is to challenge those antecedent commitments concerning the epistemic significance of attributing a process to the individual. But the analogy between memory and testimony was supposed to contribute to this project. To do so it requires some independent motivation and this is what I have questioned.
This is not to suggest that Goldberg fails to recognize that memory is intrapersonal whereas testimony is interpersonal (see e.g. 76). But he does not convincingly address a central related asymmetry. If someone forms a belief unreliably and later (re)generates it by memory,Footnote 6 an unreliable cognitive process is ascribable to her. In contrast, no unreliable cognitive process may be ascribed to the hearer in the ‘bad case’ in which her testimonial belief is generated on the basis of a testimony that is indiscernibly unreliable (or not-quite-good-enough for knowledge).
3. NOT QUITE GOOD ENOUGH
The parenthetical remark in the preceding sentence is inserted to address Goldberg's best candidate case contra Orthodox Reliabilism. The case concerns a testimonial belief that is not knowledge, although it is, according to Orthodox Reliabilism, ungettieredFootnote 7 and reliable (98). In this case, the testifier is sincere and his true belief is ‘formed through a process that, though formed in a way that was not epistemically irresponsible, was not quite reliable enough to count as knowledge (and not for having been gettiered)’ (97).
Goldberg labels this property ‘not quite good enough’Footnote 8 and gives an example:
a case in which the testimony expresses a would-be recognition-based belief formed through a momentary glance at what appeared to the subject to be a familiar object, but where the object in question was slightly obscured to the subject, at a bit of a distance from her, under non-ideal lighting conditions. (98)
Let's call the recipient of this not-quite-good-enough testimony ‘H**’ and the speaker ‘S**.’ Goldberg claims that postulating a Gettier-style condition to motivate the verdict that H** lacks knowledge ‘is both unnecessary and without independent support’ (103).
Goldberg moreover stipulates that the testifiers in H**'s social environment are generally reliable and that S** also generally ‘testifies in highly reliable fashion’ (103). So it is only with regards to p that S** gives not-quite-good-enough testimony. Given these stipulations, Goldberg claims that it is implausible to regard H**'s testimonial belief as gettiered due to local unreliability (104).
Although this is Goldberg's best candidate case against Orthodox Reliabilism, several responses are available. Recall that Orthodox Reliabilists argue that indiscernibly unreliable testimony may undermine knowledge without affecting justification. So, it is unclear why Orthodox Reliabilists cannot uphold a uniform diagnosis of indiscernibly not-quite-good-enough testimony. This general response does not appear to hinge on a particular classification of the not-quite-good-enough aspect of the testifier's belief-generating process.
However, more specific Orthodox Reliabilist responses may involve distinctions within the notion of local reliability, which is – as Goldberg notes (12 n. 2) – a difficult notion that has been debated since its introduction (McGinn Reference McGinn1984; Goldman Reference Goldman1986). Consider, for illustration, a tripartite taxonomy between global, local, and what I will here call belief-specific reliability – where the latter concerns the token-process involved in the generation of the relevant belief-token.Footnote 9 The taxonomy may be illustrated by the Fake Barn Scenario (Goldman Reference Goldman1976). Here the subject's process is globally reliable, not locally reliable but, as a matter of luck, belief-specifically reliable. The local unreliability explains why the subject lacks knowledge, the global reliability explains why the subject's belief is justified and the belief-specific reliability explains why it is true. However, Orthodox Reliabilism can allow that belief-specific unreliability can undermine knowledge. To see this, consider Chisholm's case of seeing a rock visually indiscernible from a sheep on a field where a sheep happens to be hiding (Chisholm Reference Chisholm1977). Here it is the belief-specific, rather than the local or global, unreliability of the subject's belief-generating process that explains why she is not a knower.
Given these distinctions, the twist in Goldberg's tale is that the testifier's process is globally and locally reliable but that the testifier's belief-specific reliability is not-quite-good-enough for knowledge. So, Goldberg's opponent may cite the not-quite-good-enough belief-specific reliability as the relevant factor. This approach is an extension of the Orthodox Reliabilist view that global reliability determines the justificatory status of the relevant belief although local and belief-specific unreliability can undermine knowledge without affecting justification.
The central additional assumption is that, just as belief-specific unreliability can undermine knowledge, so can not-quite-good-enough belief-specific reliability. This assumption is partly motivated by the idea that H** generates the testimonial belief by a process, attributable to H**, in a way that can be characterized as follows. The process is reliable relative to general features of H**'s social environment, but it is, on the particular application, too lucky (relative to the degree of belief-specific reliability) that the belief is true.
Further work is required to determine whether the putative distinction between local and belief-specific reliability is tenable. I re-emphasize that I have primarily hinted at such a putative distinction to explore a specific response strategy. But, as mentioned, an Orthodox Reliabilist account does not require this particular distinction. Independently of how the not-quite-good-enough testimonial belief is classified, the core of the response remains: just as an indiscernible but unreliable testifier can defeat knowledge (without affecting justification), so can an indiscernible but not-quite-reliable-enough testifier. This approach offers the Orthodox Reliabilist a far more uniform account of twin cases than Goldberg's approach. Hence Orthodox Reliabilists can account for Goldberg's key case by extending their favored account rather than by adopting the extendedness hypothesis. So it appears that Goldberg's best candidate case for the extendedness hypothesis is not quite good enough.
4. A DOPPELGÄNGER OBJECTION
In section 5, Goldberg responds to a number of objections. Since he relies heavily on twin cases throughout Relying on Others, I select for discussion his response to Schmitt's doppelgänger objection:Footnote 10
[A] doctor performs a thorough test to determine whether you have strep throat and tells you that it's certain that you do; you then believe that you have strep throat on the basis of the doctor's testimony. Now modify the case in such a way that it merely appears to you that the test is thorough; in reality the doctor is irresponsibly guessing and saying that the result is certain. Alternatively, modify the case even further, so that he is faking its performance. We might even imagine a case in which the doctor appears to be performing the test, but in fact you are just hallucinating the whole thing: there is no test, or even a doctor. On [the ‘extendedness’ hypothesis], your belief in either of these modified cases fails to be justified. But intuitively your belief is (at least in the first modified case and perhaps in the second as well) as justified as in the paradigmatic case. (148, citing Schmitt Reference Schmitt, Greco and Sosa1999: 370)
As Goldberg remarks, the cases sketched by Schmitt are importantly distinct. So, I will focus on the juxtaposition between the ‘good case’ in which all goes well and the first modification of it, which Goldberg labels the ‘not-so-good case’. Schmitt and Goldberg leave the general social environmental conditions unspecified.Footnote 11 But I will work on the interpretation that testimonies from doctors about medical matters are generally very reliable in the relevant environment.
Schmitt's verdict that the hearer in the good case and his doppelgänger in the not-so-good case are equally justified may be supported in various ways. Goldberg begins his criticism of Schmitt's verdict by expressing puzzlement that Schmitt has ‘different reactions to these cases’ since ‘the only thing backing the “equally justified” verdict (so far as I can tell) is the fact that all of these scenarios are indistinguishable-to-you’ (149).
I disagree with Goldberg's complaint about Schmitt's asymmetric treatment of the cases.Footnote 12 But I will focus on Goldberg's remark that the indiscernible-to-you consideration is the only thing backing the ‘equally justified’ verdict. In particular, I will argue that this claim underestimates more promising Orthodox Reliabilist rationales. I cannot, due to space limitations, assess all of Goldberg's discussion. But by juxtaposing two central aspects of it, I will try to indicate why I think it will be unlikely to sway Orthodox Reliabilists.
Goldberg provides a ‘programmatic reason for the justification-reliabilist to reject the “equally justified” verdict that generates Schmitt's objection’ (151). Roughly, the programmatic worry is that the ‘equally justified’ verdict cannot ‘honor the link between truth and justification’ (151). The programmatic worry is connected to more specific worries, and one of them concerns a strand of motivation for the ‘equally justified’ verdict: ‘One thing that might be used to support this verdict is the claim that two hearers who are equally epistemically responsible in their consumption of testimony are equally well justified’ (152).
To be sure, it would be an inadequate reliabilist response to solely appeal to sameness of epistemic responsibility. But epistemic responsibility is far from the sole factor that Orthodox Reliabilists would appeal to. While Goldberg recognizes this, I was puzzled that he did not, at this point, do more to address the distinctively social externalist aspect of his opponent's account. For example, the Orthodox Reliabilist will claim that the processes involved in accepting a doctor's testimony about a medical matter are globally reliable. The fact that the particular doctor in the not-so-good case is unreliable does not change the global reliability of the relevant processes.
So, Schmitt's verdict is consistent with a social externalist version of Orthodox Reliabilism that does honor the link between truth and justification. After all, the processes are evaluated with regard to truth-conduciveness. They are globally reliable relative to the general social environment that both hearers are embedded in. The account is social externalist in two senses. First, the hearer's process involved in accepting the doctor's testimony is globally reliable relative to general external social facts. Second, the hearer need not have cognitive access to these facts or the global reliability of the process. I have only sketched this motivation in a broad, generic manner.Footnote 13 But it should be clear that it goes beyond responsibility considerations and the associated ‘indistinguishable-to-you’ consideration. Consequently, Goldberg's defense of the extendedness hypothesis against the doppelgänger objection is unlikely to convert Orthodox Reliabilists.
5. CONCLUSION
I have assessed some of Goldberg's central arguments for the extendedness hypothesis and found them wanting. But, of course, this assessment does not amount to a defense of Orthodox Reliabilism or a positive account of process-individuation. Moreover, the discussion of Goldberg's case for the extendedness hypothesis is far from comprehensive. So, the sketched reservations are correspondingly inconclusive. The issue of process-individuation is notoriously hard, and Goldberg deserves credit for connecting it to central issues in social epistemology. This focus provides a fruitful context within which reliabilists may reconsider process-individuation.
I regret having to set aside other aspects of Relying on Others, such as the interesting phenomenon of epistemic coverage in chapter 6, in order to focus on Goldberg's main aim of providing a novel reliabilist framework. Despite my reservations about Goldberg's arguments for this framework, Relying on Others is interesting due to its creative exposure of important phenomena in social epistemology.
ACKNOWLEDGEMENTS
Thanks to Jessica Brown, Julie Brummer, Tyler Burge, Alvin Goldman, Jennifer Lackey, Klemens Kappel, Nikolaj Nottelmann, Asbjørn Steglich-Petersen, and, especially, Anna-Sara Malmgren. My central debt, however, is to Sanford Goldberg for detailed comments. I have benefited from these comments and from numerous other conversations and correspondences with Sandy since the middle of grad-school.
