1. Introduction
In a recent article in this journal, Kevin Elliott and Dan McKaughan (Reference Elliott and McKaughan2014) argue that it can be permissible for scientists to accept a less epistemically adequate alternative, even when an epistemically better one is available. As they put it, “what we are claiming is that, in some cases, one can have good reason to accept a theory (i.e., take it as a basis for practical reasoning and action) that is recognizably inferior to a rival from an epistemic point of view” (11).Footnote 1 For convenience, I label this claim AEI, for accept an epistemically inferior option. Elliott and McKaughan defend AEI on the grounds that branches of science often have goals apart from advancing knowledge, such as developing safe and effective treatments for disease, protecting public health, facilitating environmental regulation, or promoting economic growth. Moreover, they suggest that there is no reason why epistemic considerations should always be predominant: “Scientists use models and theories to represent the world for specific purposes, and if they can serve those purposes best by sacrificing some epistemic features for the sake of nonepistemic ones, it is entirely legitimate for them to do so” (7). Elliott and McKaughan illustrate these claims with two case studies, one concerning toxicity assessments of chemicals and another involving ecological assessments for wetland banking. They also suggest two restrictions for how nonepistemic values should influence scientific research, namely, that their role be explicitly stated and that they promote the aims of inquiry (15).
One objection to AEI is that it permits epistemically compromised science designed to promote an aim of inquiry, such as getting a drug approved by the FDA or protection from a product liability lawsuit. In fact, Elliott and McKaughan concede that this is the case: “Suppose, for example, that a think tank working for the petroleum industry performed a study and was explicit about the fact that the study conclusions were based on simplifications designed to advance their practical goal of underestimating the risks associated with climate change. While these influences of nonepistemic values seem problematic, they might satisfy the two principles elaborated in this section [that the role of nonepistemic values be explicit and advance the goals of inquiry]” (Reference Elliott and McKaughan2014, 17). They answer this objection as follows: “Our response is that one cannot capture the reasoning exemplified by the cases in this article while relegating nonepistemic values to a secondary role” (17). In other words, Elliott and McKaughan claim that their case studies are definitive examples of AEI. And that claim coupled with a naturalized approach to philosophy of science (15) leads to the conclusion that AEI should be adopted.
In this discussion note, I question this argument by challenging Elliott and McKaughan’s claim that their case studies are examples of AEI. An argument that a case is an instance of AEI requires a premise stating sufficient conditions for an option to be inferior from an epistemic perspective. Elliott and McKaughan’s interpretation of their case studies as examples of AEI depends on two such premises. One asserts that, when the results of a slower-but-more-reliable method are pending, it is better from an epistemic perspective to suspend judgment than to accept the results of a quicker-but-less-reliable method. The other premise claims that simpler, less detailed models are epistemically inferior to more fine grained and complex ones. I argue that both of these premises are highly problematic.
2. Time Sensitivity and Ease of Use
An example of AEI would be a case in which scientists accept a claim, or perhaps methodology, that is epistemically inferior to an available alternative in order to advance a nonepistemic aim of inquiry (Elliott and McKaughan Reference Elliott and McKaughan2014, 11). It may be helpful to begin with a simple example. Suppose that two conflicting assessments, A and B, of the toxicity of chemical X have been produced. Assessment A says that X is relatively harmless while B says that X is a likely carcinogen. Moreover, suppose that A was produced by a method that is highly biased in the direction of underestimating the toxicity of X. In contrast, assessment B was generated by a much more reliable method. Deciding to accept A anyhow—say, because the aim of the inquiry is to protect the manufacturer of X from lawsuits—is a prima facie illustration of AEI.
Cases of AEI, then, exhibit the following pattern: A and B are mutually exclusive and A is epistemically inferior to B, but A is accepted. Showing that a specific example is indeed a case of AEI, therefore, requires establishing the epistemic inferiority of the accepted option. And that in turn requires a premise about what, from a strictly epistemic perspective, is inferior. In the examples above, that premise would be something like:
(E) If A and B are mutually exclusive claims and the evidence favors B over A, then A should not be accepted (i.e., accepting A is an epistemically inferior option).
Establishing that a particular example is a genuine case of AEI, then, can often be difficult, because doing so may depend on debatable premises about epistemology. For instance, although E seems fairly obvious, it might be questioned from the perspective of social epistemology. Even if the evidence favors B over A, it might be better that some scientists accept A anyhow and actively pursue it as a basis for their research. For instance, Kitcher (Reference Kitcher1993) argues that it is beneficial that a few scientists pursue hypotheses that are less well supported by current evidence than rivals, since the currently available evidence might be misleading in some respects, and the currently disfavored hypothesis might ultimately be vindicated. Thus, heterodoxy can function as an epistemic precaution against the scientific community jumping to the wrong conclusion. Of course, there might be good reasons why such an argument would not excuse behaviors such as those in the example above. But for the present purposes, the point is that it may be difficult to establish the epistemic premises required for showing that an example is a genuine instance of AEI. This difficulty, I argue, is especially acute in the cases described by Elliott and McKaughan.
Elliott and McKaughan describe two cases that they claim illustrate AEI: expedited risk assessments and rapid assessments for wetland banking (Reference Elliott and McKaughan2014, 7–13). In the first case study, Cranor (Reference Cranor1993) advocates expedited but somewhat more error-prone methods for assessing the toxicity of chemicals, while the second concerns models used to decide whether a wetland restored in one place compensates for another destroyed somewhere else. Both illustrate what I will call time sensitivity, wherein it may be better for practical or social reasons to accept the results of a quicker-but-less-reliable method rather than wait for a slower-but-more-reliable one. In both instances, there is a pressing interest to draw inferences in a timely manner: the protection of public health in the first and the economic interest of not unduly delaying construction projects in the second. In addition, Elliott and McKaughan take the second case to illustrate how considerations of “ease of use” can legitimately override epistemic considerations (Reference Elliott and McKaughan2014, 11–12).
But are these examples in which scientists chose an epistemically inferior option; that is, are they real examples of AEI? Let us consider the two types of issues at play here more carefully, beginning with time sensitivity. Suppose one is confronted with a choice of whether to accept results from a quicker-but-less-reliable method or to suspend judgment until a slower-but-more-reliable procedure is completed. Is accepting the quicker result inferior from a strictly epistemic point of view? Elliott and McKaughan appear to assume that it is. They write:
Carl Cranor (Reference Cranor1993) has argued that the social costs of relying on current risk-assessment procedures (which are fairly accurate but very slow) are generally greater than they would be if regulatory agencies relied on less accurate but much quicker methodologies for assessing risks. Thus, his analysis shows that when scientists are accepting a methodological approach or a model for the purposes of guiding regulatory policy in ways that minimize social costs, they sometimes have to sacrifice epistemic values such as accuracy for the sake of nonepistemic values such as the ability to generate rapid conclusions. (Elliott and McKaughan Reference Elliott and McKaughan2014, 8)
But it is far from clear that a real example of AEI is present in this example.
To see this, note that Cranor does not recommend that, if both the results from quick-but-less-reliable and slow-but-more-reliable methods are in hand, one should accept the results from the quicker method. Rather, the question is whether one should wait for the results of the slow-but-more-reliable process (i.e., refrain from accepting or rejecting any claim on the topic until the results from the more reliable method are in) or accept the results of the quick-but-less-reliable method in the meantime. Thus, Elliott and McKaughan’s reasoning appears to require an implicit epistemic premise of the following sort:
(E1) It is epistemically better to wait for results generated by a more reliable method if one exists.
Given E1, the case of expedited toxicity assessments would indeed be an example of AEI. That is, Cranor (Reference Cranor1993) recommends that toxicity assessments derived from quicker-but-somewhat-less-reliable methods be accepted while results from slow-but-more-reliable approaches are pending. Thus, if E1 were correct, Cranor’s recommended option would be inferior from a purely epistemic perspective.
However, E1 is problematic because it overlooks the epistemic costs of extended suspension of judgment (see Steel Reference Steel2015, 182–84; cf. Steel Reference Steel2010). To illustrate, suppose that the most epistemically reliable method will not produce any results for at least 1,000 years. Then a community that refuses to accept the results of less reliable methods may be at a distinct epistemic disadvantage because they remain ignorant while the knowledge of other communities that adopt less strict epistemic standards can advance by building on prior, albeit tentative, achievements. Indeed, E1 threatens to entail the absurd result that scientists should never accept any claim. This consequence follows when, for each method m, there is another m* that is slower but more reliable, for instance, because it waits for more research to be completed. Such a situation is quite plausible in the case of toxicity assessment, since it is always possible for further research to refine scientific knowledge of the toxic effects of a chemical. In such circumstances, E1 is unreasonable, and some attainable standard of evidence must be chosen as sufficient for accepting claims. Moreover, epistemic considerations alone might not determine what this standard should be.Footnote 2 In fact, this is arguably the situation in the case of Cranor’s expedited toxicity assessments (see Steel Reference Steel2010, 26–28). That scientists decided to accept the result of a quicker-but-less-reliable method rather than to wait for a slower-but-more-reliable one, therefore, is not sufficient to show that an epistemically inferior option was chosen.
Let us turn, then, to “ease of use.” This phrase could mean a number of different things depending on the context. However, in Elliott and McKaughan’s example of rapid assessments for wetland banking, it seems to refer to science operating within a regulatory setting that requires information in a specific simplified format that is usable within the regulatory agency’s framework of norms and rules. To see the idea, consider a developer whose proposed project will destroy a wetland in one site and who wishes to purchase credits from a restored wetland bank to compensate and thereby satisfy regulatory requirements. Models exist, then, to score the ecological services provided by the two wetlands in question. The scores, then, are a means for regulators to certify that the restored wetland offsets the loss of the wetland incurred by the development project.
This example illustrates that uptake of scientific information within an institutional setting, such as environmental regulation, often requires simplifications and standardizations that would not occur in research science. But does this show that an epistemically inferior option has been selected for reasons of expediency in such cases? Elliott and McKaughan suggest that it does. Regarding the wetland-banking example, they write: “models for characterizing wetlands in the banking context differ from those that one would employ if one were aiming for the most ecologically sophisticated characterization. These differences arise in large part because bankers need assessments to be authoritative, cheap, and quick” (Reference Elliott and McKaughan2014, 12). Let us unpack the reasoning here. Elliott and McKaughan note that ecological models used for wetland-banking assessment purposes are simplified in certain ways and, presumably, simpler than some other models used in ecology. Furthermore, these simplifications are motivated by the desire for “authoritative, cheap, and quick” assessments. From these premises, they conclude that these practical concerns have resulted in the acceptance of an epistemically inferior option. But this reasoning is invalid as it stands, because the premises do not entail that selecting the simpler model was an inferior decision from a purely epistemic perspective. Elliott and McKaughan’s reasoning thus requires a premise like:
(E2) It is bad from an epistemic perspective to select a simpler, less detailed model over one that is more complex and more detailed.
In the wetland example, then, regulators choose a simpler representation when they could use a more complex one instead, which is an inferior epistemic choice if E2 is correct.
However, E2 is unreasonable because more complex, more detailed models are not always better from an epistemic point of view. Indeed, it is often thought that Ockham’s razor demands, on epistemic grounds, that scientists accept hypotheses and models that are no more complex than necessary to account for the phenomena (cf. Forster and Sober Reference Forster and Sober1994; Kelly Reference Kelly2007; Sober Reference Sober2015). Moreover, whether a simpler or more complex hypothesis is preferable from a strictly epistemic standpoint can depend on the epistemic purpose to which that hypothesis is put. For example, if the aim is a general explanation of gravitational phenomena, then Einstein’s General Theory of Relativity is epistemically superior to Newtonian Mechanics. But if the aim is to predict the projectile motions of stones near the surface of the earth, then Newtonian Mechanics may be just as good as General Relativity from an epistemic standpoint. Moreover, the epistemic aims of regulators in the wetland-banking example are likely to be more similar to the second of these two cases. In contrast to academic ecological researchers, regulators are driven by empirical concerns about specific cases rather than an interest in general explanatory principles. And for all Elliott and McKaughan have said, a simpler model might be epistemically just as good as a more sophisticated alternative for such purposes.
In addition, like E1, E2 threatens to entail that scientists should not accept any hypothesis or model because it is always possible to include just a bit more detail, nuance, or complexity. Thus, it is necessary to make choices about how complex a model should be. And as in the case of time sensitivity, epistemic factors alone might not uniquely determine how such decisions should be made. Consequently, that factors relating to institutional expediency have influenced decisions about how detailed or simple a model should be does not suffice to show that an epistemically inferior option has been selected.
3. Conclusion
Elliott and McKaughan, then, have not shown that their case studies are genuine examples of AEI. More generally and more significantly, they have failed to show the same regarding similar cases involving time sensitivity or ease of use. Finally, the focus here on whether Elliott and McKaughan’s cases are truly examples of AEI should not be taken as a tacit endorsement of their naturalized approach to philosophy of science. To the contrary, that naturalist stance cries out for clarification and defense. To do the work that Elliott and McKaughan need, naturalism must entail that normative philosophical proposals about an area of science should not conflict with its standard and widespread practices. But it is unclear why such an approach should be adopted, since it is possible that practices in an area of science might be problematic in some widespread and systematic way.
