Commentary on Karyn Freedman, “Testimony and Risk: The Dependence Account,” Rebecca Kukla

Author Information: Rebecca Kukla, Georgetown University,

Kukla, Rebecca. “Commentary on Karyn Freedman, ‘Testimony and Risk: The Dependence Account’” Social Epistemology Review and Reply Collective 3, no. 11 (2014): 46-52.

The PDF of the article gives specific page numbers. Shortlink:

Please refer to:

 Bars_Black_Yellow Image credit: Mike Kniec, via flickr

In “Testimony and Epistemic Risk: The Dependence Account,” Karyn Freedman argues that in the case of testimonial knowledge, “Justification is an interest-relative relation” (4). Specifically, the more ‘epistemic risk’ an agent takes on in believing a report that p, the more evidence she needs in order for the belief to be justified, where her epistemic risk depends on how much it matters, given her interests, values, and needs, if she is wrong. The less an agent has at stake in something being true, the lower the evidence bar for justification. Thus “Justification depends on evidence and how much evidence is needed, in each case, depends on the interests of the hearer” (14). Indeed, Freedman argues that all beliefs work this way, whether or not they are testimonially derived; testimonial evidence just offers more opportunities for epistemic risk than usual.

Freedman’s position and arguments bear a close resemblance to two other recent sets of discussions: the “pragmatic encroachment” literature in epistemology, which she cites in passing, and the “inductive risk” literature in philosophy of science, which she does not mention. Writers in both camps have argued that whether it is reasonable to accept a proposition or a hypothesis depends not just on pure, internal features of the evidence available, but also on the values and interests of the epistemic agent, and in particular on how bad various kinds of mistakes would be for that agent.

Discussions of pragmatic encroachment, such as Fantl and McGrath (2002) and Sripada and Stanley (2012) for instance, often use examples such as one adopted from DeRose (1992): You are passing the bank on a Friday with your paycheck, and you decide to skip the Friday lines and deposit the check on Saturday. You say to your spouse, “It’s OK, I know that the bank is open on Saturday mornings.” But now he reminds you that you absolutely have to get the check deposited before Monday, when an important check needs to clear, and he says, “Do you really know that the bank is open on Saturday?” Realizing that hours change and so forth, you retract your original knowledge claim. The point here is that whether you can properly claim to know something depends crucially what the stakes are in being wrong: once you realize that being wrong about the Saturday hours could have a serious negative effect, you also realize that your evidentiary and justificatory standards need to be higher than you thought.

Theorists of “inductive risk” defend a structurally identical point with respect to hypothesis acceptance in science. Their key insight is that there is no internal, evidentiary answer to how much evidence is the right amount to warrant acceptance of a hypothesis; rather, we set our bar depending (again) on the gravity of different kinds of errors—we need to balance the practical risk of accidentally accepting a false hypothesis against the practical risk of accidentally rejecting a true hypothesis. The correct balance of these ‘inductive risks’—both of which are unavoidable every time we draw an uncertain inductive inference based on empirical evidence—can only be settled with reference to the values and interests at stake. In Richard Rudner’s classic example and formulation:

If the hypothesis under consideration were to the effect that a toxic ingredient of a drug was not present in lethal quantity, we would require a relatively high degree of confirmation or confidence before accepting the hypothesis – for the consequences of making a mistake here are exceedingly grave by our moral standards. On the other hand, if say, our hypothesis stated that, on the basis of a sample, a certain lot of machine stamped belt buckles was not defective, the degree of confidence we should require would be relatively not so high. How sure we need to be before we accept a hypothesis will depend on how serious a mistake would be. (Rudner 1953, 2)

Given that such interest-relativity accounts are already quite live in epistemology and in philosophy of science, how is Freedman’s own position distinct? She puts aside the pragmatic encroachment literature quite quickly. Her claim is that while its contributors only care about the practical difference that getting a proposition p right or wrong would make, she thinks the interest-relativity of justification extends to cases where we have merely an ‘emotional investment’ in p, because p is relevant to our happiness or well-being, regardless of whether the truth of p has practical import for us. She gives the example of caring about whether her favorite philosopher, Santayana, was an anti-Semite; this might matter emotionally to her, even if it has no practical import. She argues that in this case, her emotional investment would still impact her standards for accepting a testimonial report of his anti-Semitism.

I am not convinced that adding emotional investments to the roster of interests that impact justificatory standards makes Freedman’s position as neatly distinctive as she thinks. Indeed, the dualism between emotional and practical investments seems pretty fragile and surface-level. On the one hand, even a fairly minimal set of behaviorist commitments would seem to suggest that all emotional investments come with at least some practical upshots; it’s hard to imagine an emotional investment that has literally no practical impact on behavior and decision-making. If I find out that my (formerly!) favorite philosopher is an anti-Semite, this would presumably affect my behavior: how likely I am to announce to my students that he is my unqualified favorite philosopher, for example, or whether I will display his book cover on my office wall, along with micro-behaviors such as whether I smile when I see his name or twist my hands anxiously when reading my starry-eyed earlier papers about him. On the other hand, it is hard to imagine the truth of a proposition having practical import for me unless it is accompanied by some emotional charge or relevance to my well-being, broadly construed. If not, it’s hard to see how it could have any motivating force in any particular direction.

Hence it is hard to imagine that a pragmatic encroachment theorist would object to Freedman extending the interest-relativity of justification to propositions in which we have less directly practical stakes. Freedman’s paper has lots to offer by way of helpful juicy examples, an excellent discussion of the relationship between testimonial and perceptual evidence, and a nice analytic clarification and negotiation of the debate between the ‘credulists’ and the ‘reductionists.’ But all-in I don’t think her central move is an unfamiliar one.

Let me return to the inductive risk point again, as I think it can be used to both highlight and enhance Freedman’s main point. Freedman and the pragmatic encroachment theorists focus only on the epistemic risk of believing something false. Inductive risk theorists, on the other hand, focus on both ‘type one’ errors, or false positives (accepting a hypothesis or believing a proposition that is actually false), and ‘type two’ errors, or false negatives (rejecting a hypothesis or withholding belief in a proposition that is actually true). We incur epistemic risk not just from accepting things too hastily, but also from over-caution. Consider Rudner’s belt buckle example: if a manufacturer accepts that a batch of buckles is not defective too hastily—that is, if she makes a type one error—she risks damaging her business image and having to deal with pesky returns. But if she is overcautious, tossing out a whole batch of basically sound buckles—thereby making a type two error—she wastes money and resources. Or consider De Rose’s bank example. The cost of making a type one error in this case is, by hypothesis, fairly steep: The check that really needs to clear on Monday will not clear, because the bank will be closed on Saturday when they show up. But a type two error here also has negative consequences: Not trusting that the bank will be open tomorrow will result in a less convenient stop and a rushed and less pleasant Friday night. Testimonial examples work the same way. Whenever the truth of an uncertain proposition matters at all, regardless of the source of our evidence for it, there will typically be risks incurred both by accepting it and by being cautious about acceptance.

The inductive risk theorists’ central argument is twofold: First, every decision about whether to accept a hypothesis or proposition on the basis of uncertain empirical evidence involves a judgment about how to balance competing type one and type two risks. As we raise our evidence bar, we lower the chance of false positives, and correspondingly raise the chance of false negatives, and vice versa. This balancing decision is unavoidable in any uncertain empirical judgment. Second, there is no standard inherent in the evidence itself about how to balance inductive risks. As Freedman puts it, “Evidence, on its own, can never tell us how much evidence is needed to support a given proposition” (9). It is our values and interests that settle how risky it is to us to be wrong in each of the two ways, and thus that necessarily determine how to balance inductive risks. And although Freedman doesn’t make the point explicit, one cannot even in principle decide to just err on the side of caution and set a high bar for evidence just to be ‘safe,’ because a high evidence bar always raises the risk of a false negative, and sometimes those are more harmful than the false positive would be.

This basic point about the need to balance inductive risks applies to testimonial evidence as well. Imagine that my colleague, who is only as modestly organized as philosophy professors are wont to be, tells me that the campus is closed tomorrow for an obscure federal holiday. Imagine also that I was planning on going into campus tomorrow. Perhaps I was planning on going in just to get a bit of writing done amidst some mild social stimulation, but could perfectly well work at home instead. Here, the cost of a false positive – of believing her and staying home, even though the campus was in fact open – is quite low, whereas the cost of a false negative—disbelieving her and trekking to campus just to find it closed—is more substantial. Hence my evidence bar for accepting her testimony will be quite low. But if I am scheduled to teach tomorrow, the cost of a false positive (believing her and missing my own class) would be quite high compared to the cost of a false negative (wasting time going to campus unnecessarily); hence my evidence bar will be higher in that case. In the first case I’d be rational to just accept her word at face value. In the second, I would certainly want to confirm by checking the website, asking others, or quizzing her further.

None of this is to undermine Freedman’s central point. Indeed, quite the opposite. My point is that our values and interests will affect our appropriate bar for accepting testimonial evidence, which is a matter of trading off risks rather than being lax or stringent about minimizing them. I do think, however, that once we accept this conclusion, things get problematically messy in ways that Freedman does not acknowledge.

Here is what I see as a really thorny problem: Let us accept, as I think we should, that Freedman is right about the interest-relativity of justification, with respect to testimony and other sources of belief. In the case of testimony, then, I have to judge whether S’s word that p suffices to justify my belief that p, and whether it suffices will depend in part upon my interests. But in turn, S will have necessarily drawn on her own interests in order to set her own evidence bar for p. And her interests are unlikely to be just like mine. So my judgment that her word is trustworthy needs to be a judgment not just that she’s a reliable and sincere epistemic agent in this case, but also that she is one whose interests are sufficiently similar to mine in relevant ways that ensure that I can allow her evidentiary standards to serve as mine.

This problem wouldn’t be so bad if we could, in each testimonial encounter, know enough about the interests of the testifier to make some educated guesses about her evidentiary standards. But the problem iterates! As Freedman rightly emphasizes, testimony is not a one-off affair. When someone reports something to me, that report is embedded within a vast network of past testimonial reports from long chains of others, perceptual evidence, and other sources of belief, all inextricably intertwined. Following Sellars in particular, Freedman argues convincingly throughout that we move through the world supported by a vast and tightly intertwined network of mutually dependent perceptually and testimonially derived beliefs. Everything we see and hear shows up against the background of this tight web. So asking about the evidence or justification for one particular belief treated in isolation is misleading. As she puts it, “In most cases, the sort of evidence we seek out to justify beliefs based on testimony is not merely one single, isolated telling, but rather a vast informational resource … thick with sources that are both perceptually and testimonially-laden” (15). This means that in the case of testimonial evidence, our standards of justification have to be incredibly complex. It’s not enough we set our own evidence bar in accordance with our interests. We also have to build in standards for how to tolerate the epistemic risk involved in counting on other people’s interest-driven evidence bars, reaching back effectively ad infinitum. But this threatens to seem intractable, given that the origins of chains (or better, messy webs) of testimony are opaque to us, typically. On an interest-relative dependency account, then, justification becomes a deeply complicated problem involving meta-judgments about standards. The layers of uncertainty and epistemic risk iterate, with no end in sight.

This is not just a problem infecting everyday, informal knowledge. A vast amount of contemporary science is what I have elsewhere called ‘radically collaborative’—it is produced by teams of dozens or hundreds of investigators and other epistemic laborers, trained in multiple disciplines, and spread out over multiple sites that are often continents away from one another. Indeed, such interdisciplinary collaborative research is highly prized by granting agencies and top journals. Furthermore, it is often essential to addressing research questions that cross disciplinary boundaries and require simultaneous data collection in multiple locations, such as much research in pharmacology, public health, and climate science. In such collaborative endeavors, each researcher sees only a small part of the overall research project and has to accept the testimony of other researchers, often while knowing very little about how their claims were generated or about the methodological practices of their disciplines.

The point is, in such cases, everyone involved has a rich variety of stakes and interests in the truth of various propositions. The various claims being traded about affect people’s careers, funding, and relationships with their teammates, advisors, and lab assistants; they also have investments in various research programs that go beyond pure love of the truth, if there is such a thing. All these participants are calibrating their evidence bars using their distinctive and complicated sets of interests, and all of them are collaborating in a giant network of testimony, trading around claims upon which others build. In this circumstance, it is unclear what it means for any beliefs arising out of the project to be justified, as each one embeds a giant iterative network of standards built upon standards, with no one in a position to assess the legitimacy of each step.[1] If we wish to base policy decisions, future research projects, and the like upon well-justified scientific claims, then this is a serious practical problem.

Several philosophers of science have argued that the kind of inductive risk balancing that goes into hypothesis acceptance or rejection also shows up within perception itself. Most notably, Heather Douglas (2000) looked at how experts working within different studies classified slides showing biopsies of possible rat tumors. Using the very same slides, coders working for industry-funded studies were much less likely to classify borderline cases as tumorous than were coders working for government-funded studies. In other words, coders’ balance of inductive risks shaped their perceptions of the slides.

Remember that Freedman argues that perceptual and testimonial beliefs are inextricably intertwined. Consider what happens when we put together that point with Douglas’s point about the role of interests in perception. I would suggest that if all this is right, we can no longer separate purely epistemic concerns or reasons of any kind from other sorts of concerns or reasons (such as practical or moral ones). There is no epistemic step we can take or consideration we can marshal that is not infected by interests in ways that are effectively impossibly complicated to disentangle. No ‘purification’ is possible at any level.

I think that Freedman should embrace this fundamental impurity, but at times she stops short of doing so. She frequently distinguishes between ‘prudential and moral’ reasons and ‘epistemic’ reasons. For instance, in the course of discussing an example in which she is particularly concerned to make sure that a certain flight is on schedule, she writes, “Again, I do not think there is any epistemic obligation on me to investigate the matter further, but rather that without further investigation, my belief that p is unjustified; a more appropriate doxastic attitude for me, in this case, is suspension of belief” (12-13). But surely having an appropriate doxastic attitude, or harboring an unjustified belief, is of epistemic concern![2] More generally, it seems that on Freedman’s view, taken to its proper conclusion, epistemic and prudential reasons are so thoroughly intertwined that any distinction between them is unsupportable, at least in such stark form.


DeRose, Keith. “Contextualism and Knowledge Attributions.” Philosophy and   Phenomenological Research 52, no. 4 (1992): 913-929.

Douglas, Heather. “Inductive Risk and Values in Science.” Philosophy of Science 67, no. 4 (2000): 559-579.

Fantl, Jeremy and Matthew McGrath. “Evidence, Pragmatics, and Justification.” The Philosophical Review 111, no. 1 (2002): 67-94.

Freedman, Karyn L. “Testimony and Epistemic Risk: The Dependence Account.” Social Epistemology 3 March 2014, published online. doi: 10.1080/02691728.2014.884183.

Huebner, Bryce, Rebecca Kukla, and Eric Winsberg. “Making Authors.” In Scientific Collaboration and Collective Knowledge, edited by Thomas Boyer, Conor Mayo-Wilson, and Michael Weisberg. Oxford University Press. Forthcoming.

Kukla, Rebecca. “’Author TBD’: Radical Collaboration in Contemporary Biomedical Research.” Philosophy of Science 79, no. 5 (2012): 845-858.

Rudner, Richard. “The Scientist Qua Scientist Makes Value Judgments.” Philosophy of Science 20, no. 1 (1953): 1-6.

Sripada, Chandra S. and Jason Stanley. “Empirical Tests of Interest-Relative Invariantism.” Episteme 9, no. 1 (2012): 3-26.

Winsberg, Eric, Bryce Huebner, and Rebeeca Kukla. “Accountability and Values in Radically Collaborative Research.” Studies in the History and Philosophy of Science Part A. 46 (2014): 16-23.

Authors Note: My thoughts in this piece owe a huge debt to my many discussions and collaborations with Bryce Huebner and Eric Winsberg

[1] For an in-depth discussion of the epistemological problems that attend such radically collaborative science, see Huebner, Winsberg, and Kukla (forthcoming); Kukla (2012); and Winsberg, Kukla, and Huebner (2014).

[2] Indeed she pushes in the opposite direction a few pages later: “One [of two relevant questions] is regarding the epistemic status of a belief based on testimony, specifically, what is required for that belief to be justified” (16, my emphasis). This quotation, contrary to the previous one, seems to make justification a distinctively epistemic concern.

Categories: Critical Replies

Tags: , , , , ,

4 replies

  1. This is a useful development of an already excellent article. See my comments on Freedman’s reply.


  1. Group Accountability Versus Justified Belief: A Reply to Kukla, Karyn L. Freedman « Social Epistemology Review and Reply Collective
  2. A Further Look at Standards of Justification, Rebecca Kukla « Social Epistemology Review and Reply Collective
  3. Interest-Relativity and Testimony, Jeremy Fantl « Social Epistemology Review and Reply Collective

Leave a Reply