When relying on someone whom we take to be an epistemic authority, should we believe preemptively what she does or not? According to Linda Zagzebski,Footnote ¹ the answer to this question is yes: beliefs formed on the basis of what an authority believes can be epistemically rational, where “[w]hat is essential to authority is that it is a normative power that generates reasons for others to do or to believe something preemptively” (Zagzebski Reference Zagzebski2012: 102). Modeling her conception of authority on Joseph Raz's, Zagzebski holds that a preemptive reason is “a reason that replaces other reasons the subject has” (Zagzebski Reference Zagzebski2012: 102). What this means is that a subject should not treat the reasons provided by an authority as evidence to be weighed against, or aggregated with, other relevant evidence that she might have. Rather, she should let the authority “stand in for [her] in [her] attempt to get the truth in that domain, and to adopt his belief” without deliberation (Zagzebski Reference Zagzebski2012: 105).

On Zagzebski's view, reliance on an authority can be justified in one of two different ways: by a subject conscientiously judging either (i) that she is more likely to form a true belief and avoid a false belief, or (ii) that she is more likely to form a belief that survives her conscientious reflection, if she believes what an authority believes than if she tries to figure out what to believe herself.Footnote ² Conscientious reflection is “[u]sing our faculties to the best of our ability in order to get the truth” (Zagzebski Reference Zagzebski2012: 48). This is not an externalist notion, where one can strive to be as conscientious as possible but still fall radically short. It is doing the best that one can epistemically, where this is grounded in a natural trust that Zagzebski argues we all have in our own faculties – a trust that cannot be supported with a non-circular defense of the reliability of these faculties.

The most powerful support for this view comes from what we might call the Track Record Argument: if the belief of an authority is taken, not as providing a preemptive reason for belief, but as simply one piece of evidence to be weighed with one's other evidence, then it is said that one's record in getting at the truth will be worsened. Since acquiring true belief is a key epistemic goal, preemptively trusting an authority is epistemically justified. Joseph Raz puts this point as follows:

In addition to endorsing this argument from Raz, Zagzebski discusses empirical studies that show not only that preemption leads to better results, but also that rats are better at preempting than humans are. For instance, suppose that a green light has flashed 75 percent of the time and one's goal is to predict whether a red or a green light will flash. Humans proportion their verdicts to the probabilities and thus predict green 75 percent of the time, while rats always choose the option that appears 75 percent of the time – thus always predicting green. The result is that rats end up right 75 percent of the time while humans end up worse off. According to Raz and Zagzebski, this shows not only that we ought to preemptively trust an authority if our goal is true belief, but also that “[o]nce we identify … an expert, we should follow that person's … advice all of the time” (Zagzebski Reference Zagzebski2012: 115, note 11).

Sarah Wright takes issue with this conclusion, arguing that the considerations about accuracy emphasized in the Track Record Argument fail to support the general claim that the reasons provided by an epistemic authority should be preemptive. According to Wright, the clearest case that tells against preemption is one in which an epistemic superior and I both believe that p. Regarding this sort of case, she writes:

There are at least three principles or premises at work in Wright's argument here. In particular, where you are my epistemic superior, and you and I both believe that p:

The thought underlying all three of these principles is that, when there is agreement regarding a proposition, it is always better to add my reasons to those acquired through my reliance on an epistemic superior than it is to preempt. This is because, unlike with preemption, my reasons will be in agreement with those that I get from my superior's belief, my level of accuracy will be at least as high as hers, and the more reasons I have for my belief that p, the better. This “adding reasons” approach, then, is offered as a general policy to be followed in all cases of agreement with an epistemic superior.

Each of the three principles supporting this approach, however, is subject to at least two different kinds of problems. The first we might call the bad initial evidence problem: my evidence for believing that p might be epistemically bad in the first place, and so adding the evidence I get from relying on my epistemic superior to mine will be worse than the reasons that I get from her belief wholly replacing mine. For instance, suppose that I am a sexist and my initial basis for believing that a new cancer-related enzyme has been identified is the combination of my beliefs that the male scientists in my lab testified to this and that only men are reliable sources in matters of scientific inquiry. Suppose further, however, that you are a woman and I make a singular exception in your case, regarding you as an epistemic authority on scientific matters, not on the basis of any biases about the sexes, but because of my reliably formed beliefs about your knowledge and competence. You also believe that a new cancer-related enzyme has been identified, and I recognize both that you hold this belief and that you are an epistemic superior regarding this matter.

Now, let's consider this case in relation to the three principles above, beginning with Agreement: the mere fact that I believe that a new cancer-related enzyme has been identified because you believe this in no way ensures that our reasons here are “in agreement.” Indeed, as we see in this case, my prior reasons for holding a belief might be quite different from, and significantly worse than, those that I get from your belief. Given this, Accuracy is also false, since combining these reasons for believing will render my p-related track record of accuracy worse than yours in the long run. In particular, having my belief that only men are reliable sources in matters of scientific inquiry interact with the reasons that I get from your belief that a new cancer-related enzyme has been identified will surely result in my acquiring fewer true beliefs, and more false beliefs, than if I completely deferred to you. For instance, suppose you believe that another woman scientist discovered a new type of cancer treatment, and I add my sexist beliefs with those I get from you relating to this matter. In such a case, I will most likely end up not accepting what you believe because it is outweighed by my prior biases about the sexes, thereby resulting in my having an inferior track record of accuracy. Finally, as should be clear, more evidence on behalf of a belief is not always better evidence, as is seen in the case above: preemption would make me better off epistemically, as it would force me to toss out bad reasons, replacing them with only good ones.

The second worry about the “adding reasons” approach is what we might call the conflicting evidence problem: even if my initial evidence for believing that p is not bad epistemically, combining it with the evidence I get from relying on my epistemic superior might result in a body of evidence that is worse off. Suppose, for instance, that everything is the same in the case above, except that my initial basis for believing that a new cancer-related enzyme has been identified – i.e., the combination of my beliefs that the male scientists in my lab testified to this and that only men are reliable sources in matters of scientific inquiry – is epistemically good. This need not mean that my belief about men, for instance, is true but, rather, that it has been reliably produced, or is well-supported by the available evidence, and so on. Now, even though I would be adding together two sets of reasons that are independently good here, there is no reason to assume that their combination will also be so. Indeed, combining my belief that only men are reliable sources in matters of scientific inquiry with my beliefs that you are a woman and an epistemic superior on scientific questions will result in an incoherent set of beliefs. Such an incoherent set would not only leave me confused about what to believe moving forward, it would also impact my track record of accuracy in this domain. I may, for instance, find myself frequently withholding belief due to my confusion, or giving more weight to the part of my evidence that is misleading than to that which is not. Once again, then, all three principles are shown to be false: my reasons are not in agreement with those that I get from your belief, my level of accuracy will not be at least as high as yours, and it is not the case that the more reasons I have for my belief, the better. As a general approach to responding to all cases of agreement with an epistemic authority or superior, then, this “adding reasons” approach is unsuccessful.

It is worth emphasizing that these points are not minor quibbles. If we're considering cases where we regard someone as our epistemic superior on a particular question, chances are that my prior reasons for belief might also be inferior or deficient relative to those provided by the superior. Given this, it is not at all clear that, as a general policy, simply adding reasons up is the right approach. Indeed, it is precisely this point that lies at the heart of the Track Record Argument: if I recognize that someone is better than I am at getting at the truth in a given area, wouldn't it be better epistemically for me to just wholly defer to her rather than aggregating her input with my own views on the topic? As John Hardwig writes, “because the layman is the epistemic inferior of the expert (in matters in which the expert is expert), rationality sometimes consists in refusing to think for oneself … appeals to epistemic authority are [thus] essentially ingredient in much of our knowledge” (Hardwig Reference Hardwig1985: 336). To make this point vivid, suppose that I recognize that I am unreliable at identifying instances of sexism and I also know that my colleague is excellent at doing so. If I take my colleague's reliable belief that someone's comment was sexist as simply a piece of evidence to be combined with all of my unreliable beliefs on the matter, then I would surely be worse off than if I just wholly deferred to her. This is due not only to my unreliable beliefs possibly outweighing my colleague's for me, but also because the very beliefs and processes that I am using in the weighing process might themselves be riddled with sexism. Paulina Sliwa makes this point when she writes, “if [an] agent is worried about her moral judgment being biased, there is no guarantee that her further deliberation won't be biased or overly impressed with her self-interested reasons as well. And once she starts reflecting further, she opens the way to succumbing to temptation and rationalizing the testimony away … Moral advice is so useful precisely because it's a means to put an end to one's reflections about what the right thing to do is” (Sliwa Reference Sliwa2012: 183).

In addition to arguing that the case in which an epistemic superior and I both believe that p clearly tells against preemption, Wright also clams that the case in which there is disagreement – where my epistemic superior believes that p and I believe that not-p – clearly tells in favor of preemption. She writes:

While Wright disagrees with Zagzebski that we ought to always believe preemptively what an epistemic authority does, she here suggests this as a general policy in cases of disagreement with such an authority. For it is precisely when there is such a difference in belief that one risks ending up with a worse track record than one's authority.

However, there are two different problems with this claim. To see this, recall that Zagzebski argues not only that considerations about accuracy require preemption, but also that preemption is required all of the time in relation to an authority – otherwise one will end up like the humans rather than the rats in the empirical studies. But both the preemption and the “all of the time” requirements seem misguided. Let's begin with the former: suppose that it is granted that in order to avoid worsening one's track record for getting the truth, one should always follow the beliefs of an authority. This does not show that this should be done preemptively. For humans would do just as well as rats if they accepted the testimony of an authority all of the time, but without the authority-based reasons replacing their other reasons. Instead, the authority-based reasons might simply be regarded as so epistemically powerful that they always outweigh the other reasons. The result would be the same: both the rats and the humans end up being right 75 percent of the time.

Regarding the latter, it is not the case that in order to avoid worsening one's track record for getting the truth, one should always follow the advice of an authority. Here are some alternative polices that would have even better epistemic results: follow the advice of an authority, except when one is certain that the authority is wrong; follow the advice of an authority, except when one knows that the authority is wrong; follow the advice of an authority, except when what the authority says is highly doubtful. If humans adopted any of these policies, they would end up faring better than the rats and, moreover, following them relies directly on not screening off the normative force of background evidence, as preemption requires. For instance, something would strike one as highly doubtful only against the background of one's other relevant information. The upshot of these considerations, then, is that there is simply no connection between the goal of true beliefs and either preemption or the requirement to always follow the advice of an authority, even in cases of disagreement.Footnote ⁴

So, as a general policy – even relative to certain kinds of cases, such as those involving agreement or disagreement – we should neither always add nor replace our reasons with those provided by an authority.