3.1 Second-order epistemic trustworthiness

For Daukas, one’s trustworthiness as a testifier depends not only on sincerity and accuracy, but also on one’s second-order assessment of one’s own knowledge: judging the salience and limits of what we might know in a particular context. If I hold forth on a topic I know relatively little about, I behave in an untrustworthy way even if I am accurate and sincere. To avoid this, Daukas recommends a self-reflexive approach to epistemic trustworthiness, which asks experts to responsibly characterize and steward their own epistemic resources. On this account, being trustworthy requires not only that experts make some reliable assessment about the limits of their own knowledge, but also that they have reason to think they are relatively well positioned to complete the epistemic task in question. ^{Footnote 2} This is a second-order dimension of epistemic trustworthiness: experts should attend not only to what they know but also to whether they have good reasons to think that they know it.

I endorse this account, and interpret it as demanding much more with respect to the knowledge we need to be trustworthy than some values-matching approaches seem to require. This is because in order to know whether our knowledge is well suited to a task, we need to have reasons to think we know what the task is. This is both prior to, and broader than, matching the values of scientists and the public(s) with respect to individual methodological choices within the scientific process. It requires that scientists learn about the values of their audience in ways largely neglected by values-matching approaches. ^{Footnote 3} Specifically, trustworthy experts may need to play an active role in negotiating value judgments about the aims of inquiry and in helping to characterize the values and aims of audiences for science. This is partly because some audiences may not be well positioned to know what these values or aims could be, and partly because scientists have not earned a monopoly on possible aims.

3.2 Asymmetric epistemic dependence

Epistemic asymmetries may impact agents’ abilities to characterize their values with respect to the aims of science. This is especially likely when expertise is relevant to knowing what our aims could be. To see how this might work, consider two examples.

In the first case, a cyclist enters a bicycle store and asks to buy the fastest bicycle in stock. The proprietor asks the cyclist what sort of riding he hopes to do, and he explains that he is new to cycling but hoping to do some long rides on dirt roads, bike packing, and so forth. Naturally, the proprietor would be happy to sell him a racing bicycle, but she suggests he consider a couple of alternative bikes that might better suit his purposes: perhaps one with larger tires and a more comfortable geometry for long days on the trail. She would have been perfectly within her rights to sell him the racing bicycle, but to do so would have exploited his relative epistemic disadvantage. Taking advantage of this epistemic asymmetry would be untrustworthy: Here I agree with Pielke (Reference Pielke2007) and Almassi (Reference Almassi2012).

In the second case, a city resident asks an epidemiologist to explain why her neighborhood has higher rates of COVID-19 mortality than another neighborhood. Specifically, she asks what people in the other neighborhood are doing differently that protects them from severe disease. Like the bike shop proprietor, the epidemiologist could simply give her a true answer to her original question: Perhaps they have higher compliance with masking recommendations. But to do so would obfuscate other explanations that the epidemiologist could offer instead, such as the fact that the differences in the distribution of relevant comorbidities among neighborhoods is partly the product of environmental racism. Simply answering the resident’s question would align with the resident’s aims, but it would withhold information about other possible aims that she might prefer, such as assigning responsibility to industry or ineffectual regulators. This kind of obfuscatory move is a common feature of neoliberal discourse about health (Valdez Reference Valdez2021).

Although it might seem like scientists are less crassly incentivized than bicycle dealers to “sell” their expertise, feminist science studies scholars have shown that opportunistic and naive advertisements of the social salience of scientific pursuits are all too common (e.g., Roberts Reference Roberts and Matheson2018). Along these lines, it is untrustworthy for the epidemiologist to take the resident’s original question for granted for the same reasons that it is untrustworthy for the bike shop proprietor to sell the cyclist the racing bike: It exploits her epistemic disadvantage with respect to characterizing her aims. Nothing about this means that, at the end of the day, the cyclist cannot choose a bicycle or that the resident cannot reasonably prefer to change her own behavior and leave the other issues to environmental activists. It simply reveals how insensitivity to aims characterization might render an expert intuitively untrustworthy, despite values-matching (see Haybron and Alexandrova Reference Haybron, Alexandrova, Coons and Weber2013). Epistemic trustworthiness demands that we have good justification for our moral epistemology as well as our scientific knowledge, and this means that experts should attend to whether their audience is in a position to characterize their aims and values along these lines.

3.3 Values-matching: Reprise

This has two consequences for our understanding of trustworthiness. First, it shows that experts cannot take nonexpert aims and values for granted. Second, it means that values-matching efforts to align audience values with those of scientists cannot fully discharge the demands of trustworthiness. To see why this is the case, consider two strategies for dealing with audience values: appeals to best interests and deliberative polling.

Best interest approaches appeal to epistemic paternalism (Kelsall Reference Kelsall2021). These approaches discount what agents want to know or do in favor of what some allegedly well-positioned expert thinks they should want to know or do, or what is allegedly in their “best interests” (see, e.g., John Reference John2018). Deliberative polling approaches seek to manage diffuse and conflicting audience aims by using the process of deliberation to confer political legitimacy on the outcome of a value characterization process, often a choice among preset options (Fishkin Reference Fishkin2021). Alexandrova (Reference Alexandrova2018) and Schroeder (Reference Schroeder2019) both endorse something like this.

Both of these approaches to learning audience values can be criticized independently: Best interest approaches beg the question of audience aims and are politically and epistemically illegitimate (Kelsall Reference Kelsall2021); deliberative polling approaches can be compromised by nonideal epistemic situations such as testimonial smothering and silencing (Rolin Reference Rolin2020 aptly cites Dotson Reference Dotson2011 on this point).Footnote ⁴ I mean to press a distinct, unified worry about these approaches, which is that they may neglect the challenge of negotiating audience values, either by assuming audiences are (or are not) well positioned to characterize values, or by preemptively restricting values to a set of options curated by experts. In contrast to paternalists, who would disregard audience values, I am arguing that experts should pay more attention to audience values, that they should cooperate in articulating these values, and that this cooperation may improve both experts’ and nonexperts’ epistemic situations regarding value judgments.

What I hope to have shown here is the following. An expert’s epistemic trustworthiness depends on her ability to assess the relative fit of her knowledge to the task at hand. To do this, she needs to understand the task. Sometimes she can learn this from her audience. But sometimes they are not well positioned to explore the values relevant to characterizing the aims of inquiry. In such cases, it would be easy for her to present them with a set of options. But to do so could inadvertently (in the best case) or maliciously (in the worst case) “sneak controversial values in through inattention” as Alexandrova (Reference Alexandrova2018, 433) puts it, or simply fail to meet their needs. Their values may call for some bespoke epistemic goods. ^{Footnote 5} And maybe our expert can provide them! But to think her knowledge is well suited to the task, she may not merely assume that this is the case. She must earn the warrant for this inference, and this can be tricky business. ^{Footnote 6}

4.1 The cooperative account

Cooperative experts should prioritize audience voice over merely “consulting” an audience or offering them a fixed set of candidates. By this I mean the sense in which Anderson (Reference Anderson1993) distinguishes voice, as a way of capturing the heterogeneity of ways in which we might value, from choice, a limited form of cardinal value ordering: we should not expect to meaningfully characterize what an audience might need or value simply by asking them to choose from a set of prefabricated options. Second, an expert should attend to the social conditions shaping a conversation or a process of negotiating aims. As Rolin (Reference Rolin2020) argues, contexts of asymmetric epistemic dependence are often characterized by power relations that can prevent agents from exercising their voice. Warrant for believing that aims are well characterized depends on the extent to which audience voices can be expressed. Third, experts should attend to the epistemic situation of an audience with respect to their values. Sometimes audiences may be far better positioned to characterize relevant values than scientific experts might be: they may be more familiar with a problem or better equipped to formulate a research question (Harding Reference Harding, Alcoff and Potter1993; McHugh Reference McHugh2015). In other cases, experts may be able to inform values characterization by offering more information about possible aims and methods. Either way, this aspect of the cooperative approach will allow experts to improve their knowledge about audience values.

A cooperative account goes beyond making a process politically legitimate. If an expert wants good reason to think she knows the values of her audience, then she must account for the fact that both the values of her audience and her own understanding of those values are likely to change over the course of a conversation. The real decision to be made here is when to stop characterizing values and pursue a line of inquiry. Here I am indebted to Brown (Reference Brown2020, 152), who argues that because facts are relevant to values and vice versa, “value judgments are, or ought to be, a form of inquiry, sharing a common structure with scientific inquiry,” and that we must decide when these inquiries are adequate.

Part of the motivation for our expert to engage in value negotiation, rather than mere polling or consultation, is that it is not clear at the outset whether their audience is well positioned to characterize their values, socially or epistemically. If this is the case, experts should expect audience values to change as their epistemic situation changes. Sophisticated values-matching advocates (e.g., Alexandrova, Rolin, and Schroeder), and more general accounts of the aims of science that emphasize deliberation, are well aware of this. But importantly for the second-order considerations of trustworthiness, the expert should also expect her own epistemic situation to change in conversation with her audience. She knows more about what they know and want after each exchange, and so her own understanding of the task at hand should evolve as well.

Presenting an audience with options, as many values-matching accounts would do, allows scientists to retain control of, and responsibility for, the possibility space of scientific inquiry. But scientists are notorious for their failures of imagination, both moral and epistemological, in this regard (Stanford Reference Stanford2006; Brown Reference Brown2020). A cooperative account avoids this by inviting the audience to expand or reject the possibility space as characterized by scientists and to participate in what Brown (Reference Brown2020) calls the “moral imagination” of science, or to refuse research altogether (Tuck and Yang Reference Tuck, Wayne Yang, Paris and Winn2018; Liboiron Reference Liboiron2021). This is concordant with the view of value articulation in Irzik and Kurtulmus’s (Reference Irzik and Kurtulmus2021) amendments to Kitcher’s (Reference Kitcher2011) well-ordered science, though not restricted to it.

Although the cooperative approach is demanding, it is already reflected in a number of existing models for inquiry. For instance, scientific/intellectual movements (SIMs), community-based participatory research (CBPR), and participatory action research (PAR) can all involve iterative refinement of the aims of inquiry in collaboration between scientific experts and communities affected by their research (Brydon-Miller et al. Reference Brydon-Miller, Greenwood and Maguire2003; Rolin Reference Rolin2016, Reference Rolin2020; Brown Reference Brown2020). These approaches have been endorsed by philosophers of science on broadly moral and political grounds, and by feminist philosophers on epistemic grounds (Wylie Reference Wylie, Padovani, Richardson and Tsou2015; Tekin Reference Tekin2022). Here, I am arguing that they are also valuable for moral epistemological reasons, and that these reasons are relevant to epistemic trustworthiness. To illustrate this, consider an example from health disparities research.

4.2 The preconception stress and resiliency pathways model

The preconception stress and resiliency pathways model (PSRP) is a framework for research designed by the NIH Community Child Health Network (CCHN). The problems at stake in the PSRP model are racial disparities in pregnancy outcomes, such as prematurity, infant mortality, and birthweight (Ramey et al. Reference Ramey, Peter Shafer, DeClerque, Lanzi, Madeleine Shalowitz and Raju2015). CCHN researchers used “community-academic partnerships” to develop a framework for studying these disparities. This involved sustaining long-term collaborative relationships between academics from “social, behavioral, and biomedical sciences” and “community representatives” from several research sites (ibid., 709). The academic researchers claim that community members were treated as coinvestigators (ibid.).

Many aspects of this project seem to embody a cooperative approach. The project began with a two-year “getting acquainted phase,” in which coinvestigators shared expertise and interests with one another, followed by iterative refinement of the conceptual framework for inquiry (ibid., 709–10). When disagreement arose about the direction of inquiry, each research site deliberated and received a single vote; this deliberation extended beyond the original framing of the aims of inquiry into choices that arose within the research process (ibid.). Ramey et al. report that the outcomes of this process were a framework for guiding inquiry, specific causal hypotheses, and a prospective study that tested two of these hypotheses. This framework stands out because it emphasizes paternal, as well as maternal, contributions to health; acknowledges community resources rather than focusing exclusively on stress; and deemphasizes specific mechanisms and biochemical pathways (ibid.).

The development of the PSRP model is imperfect. (We might ask: How do academic researchers ensure accountability to community co-investigators? Can community members refuse research if they are out voted? How does the model increase surveillance during the “preconception” period?) But it is an active, iterative approach to characterizing the aims of inquiry. It is at least nominally sensitive to “community” needs and interests, and is responsive to these interests not just at discrete stages of inquiry but also before, while, and after the model is implemented. All else being equal, this means that academic coinvestigators should have better reason to believe that the aims of community coinvestigators are well characterized. As Scheman (Reference Scheman, Tuana and Morgen2001, 44) reminds us, trustworthiness is not an all-or-nothing attribute, but rather “a rolling horizon we move toward” and that admits of degrees. The iterative, collaborative elements of the PSRP process improve the degree of trustworthiness of the academic coparticipants.

Negotiating audience aims often calls for a process of collaboratively characterizing values and tailoring inquiry to these goals. To be trustworthy, experts need reasons to believe they understand the aims of their audience, and that their audience is well positioned to articulate these aims. “Well positioned” here can involve access to information about what science might have to offer, but it can also invoke the social conditions that make it possible for an audience to express or negotiate their aims. In many cases, this will require scientists to take an active role in characterizing and negotiating aims, and this is best captured by a collaborative approach. My point is that SIMS, CBPR, PAR, and other such approaches are motivated not only by our intrinsic desire to institute collaborative research projects but also because they provide the epistemic warrant for scientists to know they are engaging with nonexpert audiences in a trustworthy way.