2. Epistemic Spillovers

In economics, spillovers (also called “externalities”) are the effects (usually negative) of some economic transaction on third parties. From the perspective of social welfare, these effects are important to take into account. If negative spillover effects are strong enough, we might say there is a significant “market failure”—that is, the inefficient allocation of resources via market processes, given a particular set of property rights.

To illustrate, trade is usually a benefit to both parties—so long as they are informed, not coerced, and so forth. Thus, consider the buying of a laptop computer. Here, money is exchanged for the computer, and by each party’s lights, there is a net benefit. The buyer sees the new computer as more valuable to her than, say, the 1000 dollars. And the seller sees the 1000 dollars as more valuable to them than the computer. So, both parties can be said to win here. However, this is only part of the picture from the perspective of social welfare. For, suppose that manufacturing the computer involves creating harmful chemical byproducts. And suppose the manufacturer dumps these into a nearby river. Now, there are negative spillover effects. Among other things, this dumping may be harmful to fishermen who will face reduced stocks, to people who might rely on the river’s water for household purposes, and of course to the environment in general, which many of us want to preserve. Effective policymaking thus involves mitigating such externalities, for example, by placing an appropriate “Pigouvian” tax on pollution, banning the use of especially harmful chemicals, and so on.Footnote ⁶

In a similar vein, I argue, there can be epistemic spillover effects.Footnote ⁷ Here is a stylized example to bring this out. Mustard believes that Wadsworth killed Green. Let us suppose that Mustard’s belief is justified—he is appropriately responding to the total evidence he has. Furthermore, let us even say Mustard knows that Wadsworth committed the murder—Wadsworth did, in fact, do it. Now, suppose that here are the relevant pieces of evidence that Mustard possesses, as far as this case is concerned: (1) Wadsworth’s fingerprints were on the murder weapon; (2) Wadsworth had a strong motive to kill Green—Green had recently discovered Wadsworth’s insurance fraud scheme and was about to go to the police; (3) Wadsworth has no alibi; (4) Scarlett was seen leaving the mansion the evening that Green was killed; and finally, (5) Scarlett and Green had been married and had gone through a contentious divorce.

This above evidence is mixed, as evidence often is. Some of it points to Wadsworth, some of it to Scarlett. However, let us stipulate that an appropriate weighting of the evidence suggests that Wadsworth did it, not Scarlett. Now, Mustard is trying to convince Peacock that Wadsworth did it. Mustard knows that Peacock knows (5), and he thinks this prejudices Peacock unfairly against Scarlett. From Mustard’s perspective, Peacock is prone to give (5) more weight than it deserves. Because of this, Mustard, in trying to convince Peacock that Wadsworth did it, intentionally withholds (4), which Peacock does not know. On the other hand, he tells her about (1), (2), and (3), and thereby succeeds in convincing Peacock that Wadsworth did, in fact, do it. In withholding evidence for the sake of Peacock’s own epistemic good, Mustard acts like Goldman’s judge from Section 1.

So far so good—Mustard has gotten Peacock to believe a true proposition, namely that Wadsworth killed Green. From a certain perspective, this is an epistemic upgrade for Peacock. Though, we may quibble about the details. Perhaps due to the omission of (4), she lacks the full picture, and maybe she does not achieve a full understanding with respect to how and why the crime occurred. However, maybe this is okay. I lack anywhere close to full understanding of why and how climate change is occurring, but I take myself to know that it is occurring. And this latter knowledge is an epistemic upgrade (and indeed, a significant one) relative to my being agnostic on the issue or believing that climate change is not occurring.

However, let us add now to the above case. There are further pieces of evidence that Peacock is privy to, which are not on Mustard’s radar: (6) Scarlett works for Big Insurance and (7) Scarlett has significant gambling debt that she needs to pay off soon. Now, (6) and (7), let us suppose, are not by themselves, enough to point to conspiracy on Scarlett’s behalf with Wadsworth’s insurance fraud. However, when (4) is added to the picture, especially given the timing of Green’s death, it strongly suggests that Scarlett is complicit in the insurance fraud. Had Mustard not withheld (4) from Peacock, she would have formed an important true belief: namely that Scarlett is in on Wadsworth’s insurance fraud scheme.

Thus, Mustard’s paternalistic intervention—that is, the withholding of a piece of evidence that he thinks will mislead Peacock regarding the murder—has a negative epistemic spillover. It has the effect of preventing Peacock from forming a significant true belief constituting knowledge. When viewed as a whole, it is not obvious that Peacock gets an epistemic upgrade due to Mustard’s paternalism. Depending on whether and to what extent Peacock might have overweighted (5), she may well, on the whole, have suffered an epistemic downgrade.

Suppose that Peacock would have overweighted (5), taken in conjunction with (4), to such an extent that she would have believed Scarlett killed Green, not Wadsworth. In this case, she comes to a false belief regarding the murder, but also comes to know that Scarlett was in on the insurance scam. How to balance these against each other for an overall assessment depends on our epistemic axiology—how we weigh the good of true beliefs against the bad of false ones. It also depends on which beliefs we consider as more “important,” from an epistemic perspective. (Presumably, my having a true belief about climate change is more important, even in a purely epistemic sense, than my having a true belief about how many blades of grass there are in my backyard.) But suppose that Peacock would have overweighted (5), but not to such an extent that she would have been led to believe Scarlett killed Green—she would have perhaps suspended judgment or believed it was likely Wadsworth but with less confidence than appropriate. Here, it seems that Mustard’s paternalistic intervention plausibly leads to an epistemic downgrade for Peacock.

Note here that this is an epistemic reason to avoid epistemic paternalism. It is epistemic because the intervention has yielded an outcome that has downsides with regard to Peacock’s epistemic good. That said, it does not follow that the intervention is not epistemically justified, all things considered.Footnote ⁸ For it may be thought that Peacock’s coming to the true belief about the murder is more important than her coming to the true belief about the insurance fraud. The point is, however, that the significant possibility of an epistemic spillover provides pro tanto epistemic reason to avoid paternalism. In Section 3, I will present a moral reason to avoid epistemic paternalism.Footnote ⁹

Before proceeding, I want to clarify one important difference between spillovers in the traditional economic sense and the sense I mean here. Spillovers in the economic sense are costs (or benefits) that accrue to third parties. You, as a buyer, and the seller of a computer both benefit from the transaction, but a third party—say, the fisherman on the nearby river—faces a cost due to the factory’s pollution. In the sort of case, I am interested in, the (epistemic) cost does not accrue, in the first instance, to another party but rather with respect to an agent’s doxastic attitudes with respect to another proposition (or propositions). Although of course, this is not to say such epistemic spillovers might not affect other individuals. Perhaps Peacock would have told others about Scarlett’s complicity with Wadsworth’s insurance fraud, and this would have led to an epistemic upgrade for them.

In this sense, epistemically paternalistic interferences can affect the rest of the community as well—indeed, as Clifford (Reference Clifford1877), p. 292) emphasized, perhaps “no one man’s belief is in any case a private matter which concerns himself alone.” From a social epistemic perspective, our epistemic fate is bound up with that of others—thus, an epistemic spillover that affects one person is unlikely to be sequestered there, as it were. Nonetheless, on the way I am defining an epistemic spillover here, the key point is it need not affect a third party but might instead have a negative effect on some other doxastic state(s) of the individual on whose behalf the paternalistic intervention is conducted.

3. What Does it Take to Do Paternalism Well?

The above case brings out the fact that a new piece of evidence can have complex interactions with one’s background total evidence. Thus, blocking a piece of evidence may have unforeseen epistemic consequences regarding some other proposition Q, even if it leads to an epistemic upgrade with respect to P—which is the focus of the paternalist’s intervention. Indeed, there can also be cases where the same evidence supports P relative to my total background evidence whereas it supports not-P relative to yours (Kelly, Reference Kelly2008a). Insofar as disagreements are driven by differences in background evidence in this way, successful paternalistic intervention is likely to be difficult (Davies, Reference Davies2022).

So, what does it take to do epistemic paternalism well—so that it avoids or minimizes negative epistemic spillovers? First, of course, the paternalist must have the right intentions and an adequate conception of what constitutes the epistemic good of others. She must also be reasonably good at evaluating the total evidence she has, as it bears on the target proposition(s) regarding which she intends to interfere with others’ inquiry. Thus, if I am bad at evaluating evidence regarding economic data and policy, say, I should not engage in epistemic paternalism with respect to this area. Likewise, if Mustard is not good at analyzing evidence pertaining to criminal activity, he should not act paternalistically toward Peacock. What this suggests is that for someone to be justified in being epistemically paternalistic, they must have an adequate level of expertise about the subject matter pertaining to the target proposition(s).

However, crucially, this is not enough. The paternalist must also have an adequate sense of the total evidence possessed by the target individual(s). This is what goes wrong in the case of the preceding section. Even if Mustard is an expert crime-solver, he does not have adequate access to Peacock’s total evidence. This makes it hard to carry out the paternalistic intervention without unintended and unforeseen spillover effects. That is because pieces of evidence can have complex interactions with one another.

The problem would be less severe (though it still would not disappear) to the extent that evidence worked in a uniform, predictable way. Thus, suppose that as a general matter, each piece of evidence had equal weight, and either counted for, against, or was neutral with respect to any target proposition. Weighing evidence would then look like adding unit weights on either side of a scale and seeing where the balance tips. Here, it might sometimes be the case that the paternalist has so much evidence on one side of the scale that he may reasonably block inquiry, especially if he has good reason to think that the target individual is prone to errors of reasoning. Even here though, the paternalist must have a fairly good sense of the epistemic landscape surrounding the target proposition and be able to anticipate that significant negative spillovers would not occur.

There are, then, two broad sets of challenges that face the epistemic paternalist. First, he must have adequate expertise regarding the domain of the target proposition(s) but also regarding the domains where spillover effects might occur. Second, he must have an adequate sense of the target individual’s total evidence so that he can be reasonably sure that negative epistemic spillovers will likely not occur. In the next two sections, I offer some reasons to think these conditions will often be difficult to meet.

Before proceeding, though, I want to highlight an important potential objection and explore how it might be addressed. In general, in the economic case, spillovers can be positive as well as negative. A beautiful building or mural, for example, has positive effects on others besides the individuals who may be parties to the relevant transaction. So, when it comes to epistemic spillovers, why not think that there can be positive spillovers of this kind? And further, it may be thought, if we are not assuming an autonomy-based default presumption against paternalism, then the justification for any particular instance of epistemic paternalism might depend on a sort of epistemic cost–benefit analysis. And there may be nothing useful to say in general about which way these analyses will tend to point.

In response, I want to note two things. First, to the extent that we think people are generally reliable in recognizing the import of new evidence, we should expect the spillovers to be negative. Note that this need not mean people are always reliable in this way with respect to every question. Indeed, this is the necessary assumption for getting the desirability of epistemic paternalism off the ground. However, the thought here is that people will be, for the most part, good enough, in general, at telling where the evidence points. Thus, in the case above, even if Peacock is prone to overweighting (5), she is good enough at processing new evidence that she can see (1)–(7) as pointing to Scarlett’s being in on the insurance fraud.

On the face of it, denying the above general claim seems to invite an unrealistic and unduly pessimistic view of our epistemic capabilities. In general, most people can see, for example, that the fact that the streets are wet is evidence that it rained, the fact that X is leading Y by 20 points in the polls is evidence that X will win the election, the fact that someone is crying is evidence they are upset, and so on. If we were so terrible at seeing where the evidence points, we would not have made it this far as a species. In this vein, Mercier (Reference Mercier2020) has argued that we are reasonably vigilant when it comes to acquiring and incorporating new information in deciding what to believe. Of course, we have several well-documented biases and make use of rough and ready heuristics (Kahneman, Reference Kahneman2011), but even these can be thought of as “optimally irrational” (Page, Reference Page2022)—that is, as trying to make the best use of limited cognitive resources. An analogy with our visual system might be apt here. We know that we are prone to various optical illusions. However, it does not follow that our visual capabilities are not, for the most part, highly reliable in our typical environment. The upshot then is that insofar as we are generally good at seeing where new evidence points, the spillover effects of epistemic paternalism are likely to be negative.

I want to note one limit of this argument, however. If the paternalist has high enough expertise with respect to some domain D and if it can be reasonably predicted that the spillover effects will only affect claims within D, then epistemic paternalism might be carried out without significant risk of negative spillovers. However, when it comes to many policy-relevant matters, the spillovers might not be able to be “contained” within a specific domain of expertise in this way. I will say more about this in Section 5.

There is a second, moral reason to be especially concerned with negative epistemic spillovers, and this has to do with responsibility for the resulting harms. Suppose I withhold some evidence from you and, as a result, you (rationally) choose a course of action φ rather than ψ. But were you to have the evidence I withheld, the right course of action would have been to ψ. It seems that I am responsible, at least partially, for the harms that result from your φ-ing rather than ψ-ing. To go back to the earlier example, because Mustard holds (4) from Peacock, she does not have enough to go on to think that Scarlett was in on the insurance fraud. Suppose that as a result, she does not bring this up with the authorities, who are led to convict the innocent Mrs. White, given the evidence they have. Mustard seems at least partially responsible for this resulting injustice.

This asymmetry seems to mirror how we intuitively treat cases of negative spillovers. If someone paints a beautiful mural but as a result pollutes a nearby lake with harmful paint chemicals, they seem to bear the responsibility for the negative spillover in a way that is not canceled out by the positive effects of the mural. That is, they do not get off the hook simply by pointing to the mural. Indeed, this asymmetry seems to be reflected in our various legal and regulatory systems surrounding environmental regulation. Moreover, in this case, it is plausible that if a course of action carries a significant risk of such negative spillovers, that provides a strong reason against avoiding that action. For example, if the paint may or may not be harmful to the lake’s ecosystem, the precautionary principle suggests that the paint should not be dumped there. Analogously, it is plausible that if there is a significant risk of epistemic spillovers, especially with regard to practically relevant propositions, that provides a strong reason against epistemic paternalism.

4. The “Fog of Debate”

Military commanders refer to the “fog of war” to describe uncertainty regarding the opposing side’s position. Especially prior to modern communications and imaging techniques, it was very difficult to know about the enemy’s movements, their weaponry, the terrain, and so on. Recently, Ballantyne (Reference Ballantyne2021) has argued that efforts to persuade someone else of our position on an issue face similar challenges—we often operate within the context of a “fog of debate.” For Ballantyne, there are four “clarity conditions,” which may or may not hold within an argumentative context—to the extent they do not hold, we operate within a fog of debate. These conditions (Ballantyne, Reference Ballantyne2021, 94) are as follows:

Ballantyne’s focus here is the context of argumentative persuasion, where we attempt to persuade someone of a position we hold, and they do not. The fog of debate can make our attempts to persuade fall flat or even backfire. For example, to the extent that Standpoint fails to hold, we might attribute to our opponents a more extreme (or silly) position than what they, in fact, believe. As a result, we might end up “straw manning” the opposition, which may lead them to (perhaps rightly) take us less seriously going forward. Feedback can also fail to hold for many reasons and frustrate our attempts to persuade. For instance, sometimes norms of politeness and decorum might lead people to display faux agreement and we may end up thinking we have given a persuasive argument when, in fact, we have not.

Epistemic paternalism faces similar challenges. As I have been framing it, successful intervention requires the paternalist to have an adequate grasp of the total evidence that the target individual possesses. Standpoint and Force are related to this condition. As far as Standpoint is concerned, accurately estimating our audience’s position on an issue means understanding not only their doxastic attitude toward the target proposition, but also the reasons why they believe what they do. Force also depends on the total background evidence the audience has. Indeed, on some accounts of rational polarization, two individuals may move apart on their attitudes toward some proposition P (so that one becomes more confident in P and the other becomes more confident in not-P) even after encountering the same mixed evidence.Footnote ¹⁰ The important point for our purposes is that paternalistic interventions also operate from behind a fog—even assuming no irrationality on the part of the paternalist, she must be able to discern the target individual’s relevant total evidence.

Empirical evidence on this issue, particularly when it comes to political disagreement, is not encouraging news for paternalists. It turns out that partisans are often mistaken about what the other side believes—and particularly, they tend to attribute to the other side more extreme positions than they actually hold. For example, in the US context, partisans tend to value democratic principles (such as free and fair elections) but significantly underestimate the extent to which out-party members value those principles (Pasek et al., Reference Pasek, Ankori-Karlinsky, Levy-Vene and Moore-Berg2022)—potentially driving democratic backsliding (Braley et al., Reference Braley, Lenz, Adjodah, Rahnama and Pentland2023). Further, a range of studies brings out the prevalence of “false polarization”—that is, a misperception of the distance between in-party and out-party members on various policy issues such as taxation, affirmative action, and so on (Blatz & Mercier, Reference Blatz and Mercier2017; Levendusky & Malhotra, Reference Levendusky and Malhotra2016; Sherman et al., Reference Sherman, Nelson and Ross2003). Interestingly, individuals who are more politically active and more involved in trying to persuade others are more likely to misperceive what their political opponents think (Westfall, Van Boven, Chambers, & Judd, Reference Westfall, Van Boven, Chambers and Judd2015). Thus, perhaps ironically, would-be epistemic paternalists, at least when it comes to politically charged topics, are less likely to have an accurate sense of the target individual’s total evidence, in the way that is required to do paternalism well.

5. Expertise and the Problem of Containment

Successful epistemic paternalism requires expertise. Goldman is keen to emphasize this in his original article. He says, “To justify any particular instance of such paternalism, involving a particular controller, we must have grounds for taking that agent to be an expert” (Goldman, Reference Goldman1991, p. 128). For Goldman, the problem then becomes primarily one of identifying the relevant experts. Although there can be some difficult cases where experts disagree (Goldman, Reference Goldman2001), in many cases finding the relevant experts will be straightforward enough, he argues.

On a very simple model, it would seem the task of doing epistemic paternalism well proceeds as follows. There is some proposition P that experts agree is true but that a significant proportion of the public believes is false, or they are agnostic about it. If these latter beliefs are recalcitrant enough, epistemic paternalism is justified. This will involve restraining the inquiry of those who hold such beliefs. This interference may involve a range of things, but one main method is blocking evidence via various means of “communication control,” as Goldman puts it. But who is to be charged with this control? The answer is simple enough: whoever constitutes the group of experts with respect to the subject matter that P belongs to. If P is a historical claim, then the relevant group is historians; if P is a climatological claim, then the group is climate scientists, and so on.Footnote ¹¹

The problem though is that fields of genuine expertise are narrow. This is particularly true in a time like ours, where the sciences have developed a high degree of division of labor so that expertise requires (hyper)specialization (cf. Hardwig, Reference Hardwig1991). Thus, genuine experts are experts with respect to some highly specialized domains.

On the other hand, many issues of practical policy relevance do not fall within the somewhat artificial bounds of disciplinary demarcation. How best to tackle climate change? This question is essentially multidisciplinary—the considerations relevant to this question fall within the purview of economics, other social sciences, chemistry, physics, geology, and so on. Similar points hold regarding a host of other policy questions—pandemic responses, geopolitical relations, immigration policy, and so on.

This puts the would-be paternalist in a dilemma. On the one hand, a disciplinary expert can, we may reasonably assume, get the target individual to come to an accurate view about some proposition P, which falls within that expert’s purview. But precisely because paternalism involves interfering with the individual’s inquiry—by withholding evidence, misrepresenting the level of scientific certainty, strategic oversimplification, censorship, and so on—it poses a significant risk of epistemic spillovers with respect to complex policy questions.

The structure of this problem is analogous to the case of Mustard and Peacock from Section 2. Suppose there is a piece of evidence E that is relevant to whether P. Suppose that E is weak evidence against P, which is a true proposition that the expert knows. However, the expert worries that the target individual will overweight E and this will cause her to believe not-P, or perhaps suspend judgment. (Or more weakly, it might cause her to have more doubts about P than is warranted.) As far as the target individual’s epistemic position with respect to P is concerned, the paternalistic intervention (i.e., the withholding of E) yields an upgrade.

However, given that complex policy questions draw on evidence from separate fields of inquiry, E is relevant to the question whether Q. Suppose Q is part of a subject matter that is not within the paternalistic expert’s purview. Because E is relevant to whether Q, the target individual now forms an inaccurate view about Q, depending on just how much evidential import E has in this domain. We can call this the problem of containment—paternalistic interventions with respect to one domain risk epistemic spillovers into other domains.

Let me use a (highly simplified) example to illustrate. (The reader may feel free to use another case to fix ideas, if they do not think this example is plausible.) Suppose a person is skeptical about the severity and cause of climate change. Assume this person is on the conservative side of the political spectrum, and in the background is worried that climate activism is a ruse to set more regulations, get more state control over the economy, and so forth. An effective persuasive strategy with respect to this person might be to highlight the possibility of new “green jobs,” innovation, increased exports, and so forth, that climate legislation might bring about (cf. Kahan et al., Reference Kahan, Jenkins-Smith and Braman2011).Footnote ¹² This might make her more inclined to accept the proposition that climate change is severe and caused by human activity.

However, suppose the reality is that climate regulation is bound to include trade-offs. That is, we might have to forgo some by way of economic growth and current standard of living to ensure a safer planet for future generations. Depending on the persuasive strategy we take, this person might come to have a false view of this question—that is, she might come to think there is no tradeoff, or that the tradeoff is small. So, we get the person coming to believe one true proposition and one false proposition. Her coming to believe this false proposition is an epistemic spillover. Now it is not obvious whether the total change in her attitudes is indeed an epistemic upgrade.Footnote ¹³

Efforts to deal with the problem of containment bring us to the other horn of the dilemma. For the expert to withhold (or suppress) E while being reasonably confident that it will not result in negative spillovers, she needs to make judgments about how E affects inquiry into other domains. But in doing so, the expert must depart from the boundaries of her subject matter expertise—in other words, she must engage in what Ballantyne (Reference Ballantyne2019) has called epistemic trespassing.

Trespassing of this kind is problematic for both epistemic and ethical reasons. The epistemic problem is that the expert is making calls with respect to domains where she lacks the relevant expertise. This displays the epistemic vice of hubris. Even leading experts within a particular domain may simply lack the tools and background knowledge required to properly evaluate claims within other, even nearby, domains—as the case of Linus Pauling vividly brings out. Pauling was an expert if there ever was one, having won the Nobel Prize in chemistry and being celebrated as one of the top scientists of the twentieth century. However, he also claimed that high doses of Vitamin C can be a cure for various ailments, including cancer, which turned out to be unsupported by the evidence. In making these strong claims about Vitamin C, we might say he trespassed, by stepping outside the bounds of his expertise—from chemistry to medical science.Footnote ¹⁴

As DiPaolo (Reference DiPaolo2022) has pointed out, there are ethical problems with epistemic trespassing as well. When an expert addresses a novice, the latter typically has to surrender their epistemic autonomy. Thus, when my doctor prescribes a medicine to me, the appropriate response from my end is usually to take it for granted that the drug is appropriate, rather than trying to figure things out myself. However, this sort of surrendering of autonomy puts novices in a vulnerable position—outsourcing our beliefs in this way opens us up to a range of epistemic and practical harms. When experts pose as such in areas in which they lack disciplinary expertise, they are neglectfully abusing their authority.

The way out of this dilemma is to appeal to experts who can span a range of fields. Thus, perhaps there are genuine experts on public policy or geopolitics, or economics simpliciter rather than a particular subfield of research in these areas, like game theory or market microstructure. If there are such experts, then perhaps they can legitimately engage in epistemic paternalism without the risk of spillovers or trespassing.

However, the prospects for such expertise are dim. Putative general experts of this kind—or “pundits”—have a poor track record at prediction. Notably, Tetlock’s (Reference Tetlock2005) extensive studies on the topic revealed that experts in geopolitics, for example, had no better success at prediction than lay people, or even dart-throwing chimpanzees. Or consider the financial crisis of 2008. People like Alan Greenspan or Ben Bernanke—experts on economics if there are any—failed to see it coming (Applebaum, Reference Applebaum2019). The common wisdom had been that there was no “housing bubble” in the US market. Tetlock’s work further finds that the people who were better at predicting such types of events typically were more provisional about their predictions and were loath to give yes or no answers (Tetlock & Gardner, Reference Tetlock and Gardner2015). In other words, better predictors are relatively more intellectually humble and open-minded.

What all this seems to suggest is the following. When it comes to policy, the evidential landscape is very complex and it is very hard for putative experts to predict what will happen. The best we can do is to make provisional claims and be open to new evidence as it comes along. Engaging in epistemic paternalism cuts against these points for two chief reasons. First, the gathering of evidence requires inquiry and interference with this inquiry is likely to frustrate the aim of gathering good evidence—especially in cases where such evidence might be discovered by novices or dissenting experts.Footnote ¹⁵ Second, epistemic paternalism is less likely to be justified the less certain one is about the target proposition. If I have no clue whether it will rain in 7 days, presumably I ought to be less disposed to interfere with others’ inquiry about this proposition. However, as mentioned above, the individuals who were best able to make relatively accurate predictions about the future, in Tetlock’s studies, were precisely those who were less certain. The overconfident individuals, who would presumably be more prone to thinking epistemic paternalism would be justified, were the ones with the bad track record. So, when it comes to such complex policy matters, those who would be most disposed toward epistemic paternalism are ironically perhaps the least qualified to do so.

6. Analogs to Old-Fashioned Paternalism and Some Limits

As individuals, we have different tastes, predilections, desires, and values. Mill (Reference Mill and Gray1859), p. 66) says in Chapter 3 of On Liberty, that “Human nature is not a machine to be built after a model, and set to do exactly the work prescribed for it, but a tree, which requires to grow and develop itself on all sides, according to the tendency of the inward forces which make it a living thing.” For him, this makes paternalism unwarranted, as different ways of life are best for different people and, moreover, individuals themselves are (for the most part) in the best epistemic position to determine what is best for them.

The epistemic analog to this is that we all have different sets of total evidence. We have different life experiences, have read different things, talked with different people, and so on. And what matters for rational belief is one’s total evidence, not a proper subset of it (Kelly, Reference Kelly2008b). Furthermore, pieces of evidence have complex interactions with one another. The more distinct our sets of total evidence are and the more complex the interactions between various pieces of evidence are, the less I can be sure what it is rational for you to believe once a new piece of evidence comes in. This is what goes wrong in the case of Mustard from Section 2: because Mustard is not privy to Peacock’s total evidence, his paternalistic intervention has a negative spillover.

I do not want to overstate the import of this point. Old-fashioned paternalism is often justified, in particular when it comes to children. Thus, it is justifiable to give your 4-year-old broccoli even if he wants to eat candy at the moment. Analogously, in specific cases, it might be appropriate for a mathematics teacher to interfere with her students’ inquiry in getting them to come to understand long division. However, old-fashioned paternalism is harder to justify when it comes to adults—and likewise, epistemic paternalism is difficult to justify when it comes to adults, especially in cases where the evidential landscape is complex and the evidence relevant to the proposition in question does not fall neatly within a domain of expertise.

Before concluding, I want to briefly address the case of judges blocking certain pieces of evidence in court, which has been central to defenses of epistemic paternalism (Ahlstrom-Vij, Reference Ahlstrom-Vij2013; Goldman, Reference Goldman1991). Judges can make certain pieces of evidence, for instance, about the defendant’s purported character or their past crimes, inadmissible in court. Part of the stated rationale for this is that the jury might be unduly prejudiced by such evidence. They might overweight it in their deliberations. If these rules and their justifications are appropriate, then epistemic paternalism seems to be baked into our institutions in an important way.

I want to note two points about these sorts of cases that make them unique. First, when it comes to what kinds of evidence are admissible in court, what is relevant is not only the epistemic position of a particular jury, but also the sorts of downstream incentives that might be created. For example, most obviously, if evidence obtained illegally, without the appropriate warrants, is admissible in court that creates all sorts of perverse incentives for law enforcement and the rest of the population. Similarly, we might worry about evidence of past criminal activity and character traits, and what kinds of perverse downstream incentives would be created if such evidence were admissible in court.

Second, and more importantly, the worry of epistemic spillovers in such cases is of minimal normative importance. What is at hand in a criminal trial is a special context where it is of utmost importance to determine whether the defendant is guilty beyond reasonable doubt. If there are epistemic spillovers—and there might well be—they are normatively insignificant when compared to the matter at hand. If the jury comes to an inaccurate belief about the defendant’s character, say, that is much less significant as compared to coming to an accurate decision about guilt. Furthermore, paternalism is restricted to a specific deliberative context—the jury presumably can inquire as they please once the trial is over.

Part of what makes epistemic paternalism about policy matters different is that the spillovers may not be insignificant. And such paternalism, if it is to succeed, needs to be ongoing. If the goal is to get the broader public to have an accurate view about some proposition, and policymakers determine that epistemic paternalism is warranted, then their interference with the public’s inquiry must be constant. Or, if the target proposition pertains to some “emergency situation,” then the interference must last at least till the emergency is over. (This also raises the prospect of the paternalistic policy being revealed after the emergency, which would presumably erode trust in the society’s epistemic institutions.)

Epistemic paternalism, then, is a blunt tool for policymakers. It might be useful here to compare blunt medical interventions. Certain drugs or procedures may cure an ailment but cause many negative side effects. Ideally, what medical professionals ought to do is to cure the ailment while minimizing such side effects. If there is a rash on someone’s leg, the proper course of action is not amputation but rather some effective ointment. In the epistemic case, we might explore whether there are less blunt tools for expert intervention.

One possibility that has been recently discussed by Begby (Reference Begby2021) is that of evidential preemption. Suppose there is some piece of evidence E that is likely to mislead a novice about whether P because the novice is prone to overweighting E. An expert may preempt this issue by communicating to the novice that E has already been taken into account by the relevant research. So, rather than paternalism, this sort of preemption constitutes guidance, and presumably avoids the problems with spillovers I have been discussing here. It also reduces the risk of a loss of trust in expert deliberation and communication. At any rate, if there are less blunt tools of this form that can be used effectively, the arguments presented here suggest that we should do so.