Rational irrationality: A conceptual defense

Paulson’s critique: The toxin puzzle analogy

Paulson’s critique of rational irrationality centers on an interesting analogy he draws between the voting scenario and the so called “toxin puzzle” introduced by Gregory Kavka in an influential article with the same title (Kavka, 1983). In Kavka’s original puzzle, an eccentric billionaire offers you a million US dollars if you can form the intention to drink a non-lethal but sickening toxin the next day. Crucially, you need only form the intention, not actually drink the toxin, to receive the reward. Your intention will be detected by a perfect “mind-reading” device at midnight, at which point you’ll receive the reward. The puzzle emerges because it seems impossible to genuinely form the intention to drink the toxin while knowing that once the money is deposited, you’ll have no reason to follow through with drinking it.

Kavka’s puzzle is rooted in a conflict between reasons to intend to perform an act and reasons to perform the act. Paulson suggests this tension is present in the voting rational irrationality scenario, where a conflict exists between reasons to believe a policy p is best and reasons to enact P. I will refer to the voting scenario as VOTING throughout this paper.

For the purposes of his argument, Paulson discusses this analogy through a variation of the Kavka’s toxin case presented by Hieronymi (2009). In this scenario, a scientist offers a minor financial incentive ($100) to form the intention to jump out of a third-story window. The scientist, equipped with a reliable intention-detector, will reward the agent solely for forming the intention, regardless of whether they ultimately follow through with the action. Given that merely intending to jump carries no inherent risk, the agent seemingly has sufficient reason to form the intention. However, actually jumping remains irrational due to the physical harm involved which clearly outweighs the financial reward. I will refer to Hieronymi’s scenario as JUMP throughout this paper.

How then should Hieronymi’s JUMP be “solved” from the standpoint of the normatively approved way of proceeding? One proposal, raised by Niko Kolodny, is that it is irrational to form the intention to jump because doing so significantly increases the probability one will actually jump. ² Kolodny’s straightforward resolution eliminates the supposed tension between reasons for intending and reasons for acting by suggesting the two are simply linked through the causal impact of intentions on actions.

This solution to the puzzle would work well for Caplan’s model because it ties the rationality (or lack thereof) of forming the intention/belief to its probabilistic impact on practical consequences. This, in turn, introduces a major disanalogy between JUMP and VOTING. In JUMP, as Kolodny suggests, forming the intention to jump plausibly increases the probability of the agent jumping from the third story window. In VOTING meanwhile, forming the belief that policy P is best all-things-considered does not (in the normal case) change the probability of P being enacted.

Kolodny’s resolution of JUMP fails, according to Paulson, because it fails to appreciate that forming the intention to jump to collect $100 is to settle for oneself the matter of whether to jump and is therefore subject to rational criticism independently of the consequences that follow. As Paulson puts it, the reasons for intending to jump are “transparent” to the reasons for jumping, “intending to jump and jumping don’t receive separate cost-benefit analyses” (2024: 11).

Paulson’s main thesis is that VOTING should be analyzed in the same way as JUMP. The reasons for forming the belief that policy P is all-things-considered best are transparent to the reasons for enacting P. As in the JUMP scenario, Paulson contends that the probability that the belief/intention to enact P will affect the enactment of P is irrelevant as far as rational criticism of the belief/intention is concerned. If forming the intention to jump in JUMP is deemed irrational so is forming the belief in P in VOTING, regardless of practical considerations.

A final piece of Paulson’s critique of Caplan’s model is his response to the issue of tacit awareness of costs/benefits. As Paulson rightly notes, Caplan’s position is not that voters first figure out that policy P is bad but then decide to support it anyway for various rewards. Rather, the idea is that voters are tacitly sensitive to the costs and benefits of continued inquiry, leading them not to scrutinize their cherished political beliefs. This differentiates VOTING from the toxin case or JUMP, where the agent is explicitly aware of all the relevant details and knows drinking the toxin/jumping is detrimental.

Paulson responds to this potential rejoinder by emphasizing that Caplan’s model is meant not merely to describe the psychological processes of voters but to normatively validate their behavior. In Paulson’s terminology, rational irrationality is proposed as a rationalizing explanation of voter behavior. This, in turn, means that an ideally rational agent would proceed in exactly the way voters do. This means, finally, that an ideally rational agent could explicitly go through the considerations voters are tacitly aware of and reach the same decision. Thus, appeals to tacit awareness fail to vindicate rational irrationality, according to Paulson.

Paulson’s critique in context: Novelty and significance

Caplan’s theory of rational irrationality has generated substantial scholarly response since its publication. The critical literature can be broadly divided into two categories: empirical challenges that question whether the model accurately describes voter behavior, and conceptual challenges that question the internal coherence of the model itself. Paulson’s critique falls squarely within the latter category, but offers distinctive insights that advance the theoretical debate in important ways.

While empirical critiques have been offered by scholars like Landemore (2012) and Mackie (2012) who question Caplan’s assumptions about voter knowledge and electoral impact respectively, I will focus exclusively on the conceptual debate around rational irrationality, as this is where Paulson makes his contribution. Among the theoretical challenges to Caplan’s model, two are particularly noteworthy: Bennett and Friedman’s (2008) argument that rational irrationality and the earlier rational ignorance model are conceptually incoherent, and Lomasky’s (2008) claim that Caplan’s model is built on a failure to properly distinguish practical from theoretical rationality. Through examining these critiques and how Paulson’s argument relates to them, I will suggest that Paulson’s toxin puzzle analogy offers a novel framework that deepens our understanding of the philosophical issues at stake in rational irrationality.

In their contribution to the Critical Review symposium on The Myth of the Rational Voter, Bennett and Friedman (2008) offer an extensive critique of the conceptual underpinnings of Caplan’s model. They argue is not only that the empirical case for rational irrationality is unconvincing—though they say that as well—but, like Paulson, that the very idea of rational irrationality is conceptually flawed.

As Bennett and Friedman rightly observe, rational irrationality is an extension of the earlier rational ignorance model. As they see things, the rational ignorance model already is incoherent. Caplan’s extension only makes matters worse. Let us then briefly review Bennett and Friedman’s objection to rational ignorance and then turn to their treatment of rational irrationality.

Bennett and Friedman’s conceptual critique of rational ignorance rests on a rejection of the basic framework of applying rational choice to information, i.e., information economics. As they put it:

“Information economics reduces ignorance to the known cost of searching for knowable ‘information’—as if the searchers knew what they were looking for, where it was to be found, and its value. None of that would be possible, however, unless the searchers already knew the very things of which they are supposed to be ‘ignorant’ ex ante.” (Bennett and Friedman, 2008: 204)

As Caplan notes in his reply to Bennett and Friedman, there is no reason to think we must know everything about something to estimate the value of learning more about it (Caplan, 2008: 378). To borrow one of Caplan’s examples, a student considering whether to pursue a PhD in philosophy can estimate the value of doing so in a variety of ways. She could look at common outcomes for philosophy PhDs in the labor market. She could chat with some recently minted PhDs to get a sense of the experience. The idea that no calculus is possible without having obtained the PhD is—as Caplan rightly notes—absurd.

As mentioned, however, Bennett and Friedman argue that even if Downs’s model were viable, Caplan’s extension of it remains incoherent. The reason is that, as they interpret it, Caplan’s model requires that voters hold policy beliefs they believe to be false. Caplan’s reply to this objection hinges on the idea that (a) people can and do in fact hold contradictory beliefs; and (b) when a person holds contradictory beliefs, it is plausible to expect that the side of the contradiction consciously held will be responsive to incentives.

Here we can already flag an area in which Paulson’s critique of rational irrationality takes a similar path to earlier objections but offers something more. Paulson, as we saw, also puts pressure on the notion of holding contradictory beliefs. However, Paulson takes an important step beyond the exchange between Caplan and Bennett and Friedman by emphasizing the normative significance of reflective stability and its relation to the ideally rational agent. In doing so, Paulson offers a potential rejoinder to Caplan on behalf of Bennett and Friedman which I will address later on in this paper.

Bennett and Friedman’s central conceptual critique of rational irrationality is that Caplan’s model, if it is not incoherent, is simply a roundabout and misleading way to affirm their preferred explanation of voter error, namely that voters make inadvertent mistakes due to their ignorance, what Friedman in later work calls ‘radical’ ignorance (Friedman, 2019). Voters are ignorant of both politically relevant information and the causal inefficacy of their vote. Hence, their errors are not borne of rational calculus but of simple inadvertent ignorance.

In his reply to this objection, Caplan posits that there is a psychological spectrum between, on the one hand, a fully conscious calculus to indulge in erroneous views and, on the other, complete inadvertent error. Here, too, Paulson’s contribution to the dialectic is helpful. In drawing a parallel between voting and the toxin case, Paulson provides us an illuminating angle from which to think about the relationship between beliefs/intentions and rationality that speaks directly to this exchange between Bennett and Friedman and Caplan and the distinction between inadvertent error and rational irrationality.

In his review of The Myth of the Rational Voter, Lomasky (2008) presents a somewhat less polemical (compared to Bennett and Friedman) but I think more philosophically probing critique of rational irrationality.

Lomasky’s most fundamental critique claims Caplan fails to distinguish between theoretical and practical rationality. As Lomasky uses these terms, theoretical rationality refers to “the agent’s relation to beliefs”, practical rationality to her relation to actions. As Lomasky points out, the distinction is an important one because while the two forms of rationality share similarities “insofar as each involves proper functioning of mental faculties”, they differ significantly “in their products”. In framing the issue of voting in terms of theoretical and practical rationality, Lomasky highlights the fact that pursuing theoretical rationality carries practical opportunity costs. From this it follows that “one who is practically rational will not give priority above all else to belief cultivation” (Lomasky, 2008: 473).

Lomasky brings up the theoretical/practical distinction and the opportunity costs tradeoffs to press Caplan on his attempt to distinguish between rational ignorance and rational irrationality. Caplan, according to Lomasky, misdiagnoses what appears to be a systematic bias as irrationality when it may simply reflect a practically rational decision not to invest scarce resources in an area with little payoff. Lomasky argues that this would suggest voters are rationally ignorant, not irrational. This, in turn, would help vindicate the expressive model of voting, favored by Lomasky, which he articulated in his earlier work with Brennan (Brennan and Lomasky, 1993).

The dialectic between Lomasky and Caplan is less about the empirics of voter behavior and more about how we should conceptualize and think about this behavior from the standpoint of rational choice theory. Both theorists agree that voters gain utility from expressing (in part by voting) political views that are incorrect (from the perspective of relevant experts). Both agree that voters pursue this course of action as a result of the lack of practical consequences to how they think or vote politically. The conceptual question is whether it is appropriate to use the term “irrational” to describe this behavior.

Though Paulson does not engage explicitly with this exchange between Caplan and Lomasky, his arguments are here as well helpful for deepening the dialectic. In particular, as I will lay out in the following subsection, Paulson’s discussion can be framed in terms of Lomasky’s distinction between practical rationality and what I’ll refer to as epistemic rationality. In my interpretation of Paulson, his argument offers a particular answer to the question of how to resolve conflicts between practical and epistemic rationality in the normatively approved way. If I am right, there is value to engaging with Paulson’s argument which may help shed more light on some issues not sufficiently explored in previous discourse.

Before turning to a deeper analysis of the normative structure of Paulson’s critique, it is worth briefly addressing an objection raised by a reviewer of an earlier version of this paper. According to this objection, Paulson’s analogy rests on a category error: It conflates individual with social cost. In Caplan’s model, the individual cost of holding irrational political beliefs is effectively zero, since a single vote never influences electoral outcomes. The problem with political irrationality lies in the social costs it generates when many individuals behave this way. By contrast, in the toxin or JUMP scenarios, the individual directly bears a significant cost from the intended action. Paulson’s analogy is said to fail because it invokes socially distributed harm in the voting case while equating it to individually borne harm in the toxin case.

While this critique raises an important issue, I do not believe it accurately captures the structure of Paulson’s argument. Paulson does not compare voting for a harmful policy to drinking the toxin. Rather, his analogy is between intending to enact a harmful policy and intending to drink the toxin. The puzzle he raises concerns the coherence of forming beliefs or intentions that commit one to outcomes recognized as harmful from the standpoint of one’s own utility function. One might object to this analogy on feasibility grounds, since individual voters do not in fact have the power to enact policies. But that is a point about agency, not about cost, and it is a line of response I develop in the next section.

Alternatively, one might attempt to defend the category error critique by maintaining that, even if the belief concerns the enactment of a harmful policy, the personal cost to the individual voter remains negligible because the effects are socially distributed. For this defense to hold, however, two auxiliary assumptions must be accepted: first, that the policy’s consequences do not meaningfully affect the voter’s own welfare, and second, that voters care only about their own outcomes. Both assumptions are empirically questionable. Policies such as tariffs, for instance, can directly affect prices, employment, and economic stability in ways that do touch individual voters, even if the impact is broadly dispersed. And even those not directly harmed may reasonably care about wider societal consequences, such as inflation or recession.

Practical and epistemic rationality: A framework for understanding Paulson’s critique

Having presented Paulson’s toxin puzzle analogy and situated it within the literature on rational irrationality, I now want to examine his argument through the lens of the relationship between practical and epistemic rationality. This framework, highlighted in Lomasky’s review of Caplan’s book, provides a powerful tool for understanding what makes Paulson’s critique distinctive and illuminating. I will understand the practical-epistemic distinction to refer to utility maximization on the practical side and arriving at justified beliefs on the epistemic side.

The natural way to think about the relation between practical and epistemic rationality is to examine cases where these two standards potentially conflict. Before exploring such conflicts, we should note that some may deny their possibility entirely, arguing that epistemic rationality is involuntary—since one doesn’t choose what to believe, there can be no meaningful clash between utility maximization and epistemic standards. However, this view sidesteps rather than resolves the philosophical puzzles around rational irrationality. While much more can and has been said about this possibility, it will not be pertinent for my purposes. My task is to respond to Paulson, who accepts that practical and epistemic rationality can meaningfully come into conflict. How Paulson thinks about these conflicts will become clear as we proceed.

To develop this lens of thinking through practical versus epistemic rationality, it’s important to consider a third case Paulson introduces, which I will call DISEASE. In this scenario, a person has a serious illness from which most patients do not recover, but their odds of recovery increase significantly if they believe they will recover. Unlike JUMP and VOTING, which Paulson argues are structurally similar, DISEASE represents, according to Paulson, a scenario where it might be rational to prioritize practical concerns (improved health outcomes) over epistemic ones (forming beliefs based strictly on the evidence).

In all three cases (JUMP, VOTING, DISEASE), what seems optimal from the standpoint of utility maximization seems to conflict with what is required in terms of holding epistemically justified beliefs. Speaking in broad terms, three normative positions can be identified for addressing these conflicts. The epistemic-first position says that epistemic rationality is always prior to practical considerations. Hence, the normatively approved way to resolve the three conflicts is to believe what is warranted by the evidence regardless of the practical consequences.

The concept of rational irrationality has no normative validity according to the epistemic-first view because it involves violating the normative priority of epistemically sound reasoning. While one could defend this position, Paulson, as we can already observe, does not. He purports to debunk Caplan’s view without resorting to what might be referred to as epistemic extremism, as I discuss below.

The inverse position toward these conflicts is the practical-first position. On this view, utility maximization is always prior to epistemic justification, thus the normatively approved way of resolving these conflicts is to hold whatever belief maximizes one’s utility, regardless of its epistemic merit. The concept of rational rationality applies to all cases of conflict according to the practical-first position. In all three cases, holding the epistemically dubious belief is normatively justified so long as it maximizes utility.

Finally, we have the contextualist position according to which the primacy of epistemic or practical rationality varies based on the context. In some contexts, epistemic rationality has greater normative force; in other contexts, practical rationality is more authoritative.

We now have three basic views on the relation between epistemic and practical rationality, and three specific cases in which the standards conflict. We can use these parameters to locate and understand Paulson’s view and argument. Let us then reiterate Paulson’s verdict on the three cases. Paulson thinks it is rational to prioritize the practical over the epistemic in DISEASE but that it is not rational to do so in the cases of JUMP and VOTING. ³

Paulson’s position is therefore a contextualist view on the relation between practical and epistemic rationality. It is dialectically important to observe that Paulson is not arguing for the principled epistemic-first position according to which rational irrationality is never normatively valid. Arguments in defense of this view could certainly be made. However, since my aim is to address Paulson’s arguments, I set the epistemic-first view and the arguments for it to one side.

Paulson’s argument against Caplan and his contextualist view has the following overall structure. As mentioned, Paulson thinks rational irrationality is intuitively sensible in the case of DISEASE. He holds, as we saw, that with a bit of reflection we can see that rational irrationality is normatively invalid in JUMP. The pertinent question thus becomes: Is VOTING similar more to DISEASE or to JUMP? As noted above, Paulson argues that VOTING is analogous to JUMP and hence that rational irrationality is likewise invalid in the case of voting. ⁴

To substantiate his categorization of the three cases, Paulson offers a demarcation principle by which to categorize cases of conflict between practical and epistemic rationality. The key feature, according to Paulson, is whether the belief in question is directed toward action. The reason it is irrational to believe one should jump out of the window (to collect the payoff) is because that belief is equivalent to a practical commitment to jump out of the window. Being committed to jumping out the window while knowing doing so is suboptimal is an irrational state to be in, regardless of its effects. Conversely, rational irrationality is valid in DISEASE because the belief in question is not linked to action.

Paulson argues that the relevant belief in the case of voting is an action-oriented belief. Therefore, VOTING should be analyzed in the same way as JUMP. Hence, rational irrationality is invalid in the case of voting.

Before I turn to address Paulson’s argument, let me clarify the dialectical approach I will be taking. One way of meeting Paulson’s argument is to defend the uncompromising practical-first position. That view, as mentioned, holds that rational irrationality is valid in all cases. The burden of this dialectic path is to show that rational irrationality is valid in JUMP (or the toxin case), since that it the paradigmatic case for the invalidity of rational irrationality according to Paulson.

Paulson’s reading of Caplan is that Caplan endorses the practical-first view and that he is therefore committed to defending rational irrationality in all cases, including JUMP. ⁵ My purposes do not require arguing over how Caplan should be read. Though I think there are things to be said for the practical-first position (as there are for the epistemic-first) I will accept Paulson’s contextualist view which judges rational irrationality to be valid in DISEASE and invalid in JUMP. I will argue that, contrary to Paulson’s claims, VOTING should be understood as analogous to DISEASE rather than JUMP. Hence, rational irrationality is a normatively valid explanation of voter behavior/epistemics.

Feasibility of the action

In examining the relationship between belief and practical commitment in JUMP, we must first consider what makes this link particularly plausible in this case. One potential explanation is that the belief in question directly concerns an action, thereby creating a conceptual link to a practical commitment. This perspective might initially seem to support Paulson’s categorization, placing VOTING alongside JUMP and distinguishing both from DISEASE. After all, jumping out of a window and implementing public policy are actions, while believing in one’s recovery from a disease is not inherently action oriented.

However, before looking deeper into this action-oriented belief hypothesis, I wish to introduce another element that could explain the strong link between belief and action in JUMP: the feasibility of the action to which the belief is directed. By feasibility, I refer to acts that are realistically doable for the agent. This concept of feasibility extends beyond mere logical possibility or physical possibility; it includes actions that fall within the scope of what the agent can reasonably imagine herself doing.

The notion of feasibility provides an alternative lens through which we can analyze the relationship between belief and action in our three cases. It allows us to consider not just whether a belief is about an action, but also whether the action in question is one that the agent could realistically envision undertaking. This perspective may offer insights into why JUMP and VOTING might be categorized differently, despite both involving beliefs about actions.

The presentation of JUMP naturally leads us to assume that jumping out of the window is clearly feasible for the agent in the sense we have defined. This implicit assumption of feasibility may play a crucial role in shaping our intuitions about the case. To examine the importance of this feasibility variable, we can construct modified versions of JUMP where the act of jumping is not feasible for the agent. Let’s call these modified scenarios JUMP*.

Consider the following examples of JUMP*:

(1) The scientist offers the agent $100 to jump from a window at the summit of Mount Everest.

In these JUMP* scenarios, the belief is still directed toward an act of jumping from a window, just as in the original JUMP case. However, the critical difference is that the act in question is now outside the agent’s feasibility set — it’s not something the agent can realistically perform.

When we consider these JUMP* cases, the link Paulson draws between belief and practical commitment becomes tenuous. The agent’s belief that they should jump out of the window no longer seems to have the same implications for action as it does in the original JUMP scenario. In fact, one could argue that these JUMP* cases bear a closer resemblance to DISEASE than to JUMP. This similarity to DISEASE suggests that rational irrationality could potentially be applied validly to JUMP* scenarios.

Having examined the role of feasibility in JUMP and JUMP* scenarios, we must now consider whether VOTING aligns more closely with JUMP or JUMP* in terms of the feasibility of the target act from the agent’s perspective.

To properly analyze VOTING in the context of rational irrationality, it’s important to clarify the precise nature of the beliefs and actions in question. A possible misunderstanding is that the irrationality in voting behavior relates to beliefs about the voting process itself or about specific candidates. However, this is not where Caplan’s concept of rational irrationality applies.

The irrationality Caplan identifies is not in voters’ understanding of the electoral process or their decision-making about how to cast their ballots. In fact, Caplan’s model posits that voters are rational in their comprehension of and response to the voting process and its associated incentives. Voters understand (explicitly or implicitly), for instance, that their individual votes are extremely unlikely to determine an election’s outcome.

Rather, the epistemic irrationality that Caplan’s model addresses lies in voters’ process of forming beliefs about public policy itself. These are beliefs about how society works, what economic policies will lead to what outcomes, what foreign policy approaches will be most effective, and so on. The “target act” corresponding to these beliefs is not the act of voting, but the implementation of these policies at a societal level.

With this clarification, we can more accurately assess the feasibility aspect of VOTING. The relevant action—the implementation of public policy—is clearly not within the direct control of individual voters in the way that jumping out of a window is in JUMP. While voters can take related actions like casting ballots or engaging in political discourse, the actual enactment of public policy is removed from their sphere of influence.

This distinction is crucial because it places VOTING closer to JUMP* scenarios than to JUMP in terms of feasibility. Just as an agent in JUMP* cannot feasibly perform the requested jump, individual voters cannot implement the public policies they believe in. This separation between belief and action capability is a key factor in understanding why rational irrationality may be applicable to voting behavior, despite Paulson’s arguments to the contrary.

These considerations about collective action reinforce the feasibility distinction between JUMP and VOTING. While JUMP involves a straightforward physical action within the agent’s direct control, VOTING involves beliefs about collective actions that are fundamentally mediated by complex social and political processes. This mediation affects not just the probability of implementation, but the very nature of the rational commitments created by such beliefs.

Before moving on, I must address Paulson’s brief response to the distinction between individual and collective action. In a single paragraph before his conclusion, he argues that the collective nature of policy implementation does not undermine his critique of rational irrationality. He writes: “Consider the following variation of the toxin puzzle: the scientist offers you $100 to believe that it is best-all-things-considered for us to enact a policy that will result in our defenestration. I urge that this is problematic in much the same way as the original toxin puzzle” (Paulson, 2024: 13).

This cursory treatment understates crucial differences between direct individual commitments and beliefs about collective action. While Paulson is right that beliefs about what “we should do” carry some normative force, the nature of this commitment differs fundamentally from beliefs about what “I should do.” The institutional mediation of political action through democratic processes (or other kinds of political processes) creates both practical and conceptual distance between belief and commitment that is not present in immediate physical actions. The difficulty of coordinating group action, the multiple intervening steps between individual belief and collective implementation, and the diffuse responsibility for outcomes all affect how beliefs about collective action relate to practical commitments.

Consider an analogy: Believing “we should build more housing” carries different practical implications than believing “I should build a house.” The latter creates immediate action-oriented commitments, while the former’s practical entailments are necessarily mediated through complex social and political processes. Paulson’s variation on the toxin puzzle elides this distinction by choosing an example (group defenestration) that collapses back into coordinated individual action rather than engaging with the distinctive features of political belief and action.

Having established the distinction between VOTING and JUMP in terms of feasibility, we ought now to address a potential counterargument Paulson could raise. Paulson might contend that this disparity in feasibility merely restates the issue of causal probability. The claim that enacting policy is infeasible, while jumping is feasible, could be seen as another way of saying that the probability of the former is far lower than the latter. Paulson might argue that this disparity in probability is only relevant if we accept Kolodny’s resolution of JUMP, which he explicitly rejects. Therefore, he might claim, the disanalogy in feasibility fails to undermine the argument against rational irrationality.

While seemingly plausible, this rejoinder leads to an odd conclusion. If we accept that the link between beliefs/intentions and practical commitments in no way depends on feasibility or causal probability, it follows that beliefs about any logically possible act would constitute a practical commitment in the relevant sense. This seems counterintuitive and potentially problematic.

To further explore this issue, let us consider two political scenarios that test our intuitions about the role of feasibility in the political case. First, imagine a subject forming beliefs about policies a hypothetical global government should implement. This government has never existed and likely never will. If feasibility and causal probability are truly irrelevant, this case of global government policy beliefs should admit of the same analysis Paulson applies to rational irrationality in more typical voting scenarios.

For those who might find the global government case too fanciful, let us consider a much more realistic political scenario: policy beliefs under complete dictatorship. Imagine an intelligent inhabitant of North Korea. In this case, the consequences of being epistemically rational about politics might well be imprisonment and death. At the same time, the ability to enact the policies in question is non-existent. Paulson’s position seems to imply that North Koreans who adopt the “right” political beliefs, that is, those approved by the party, are not acting in a fully rational way, because rational irrationality is supposedly incoherent.

The North Korea case highlights an important theoretical issues not addressed in Paulson’s discussion, namely the role of democracy and voting in the coherence or lack thereof of rational irrationality. This is closely related to the feasibility dimension we have been discussing. The question is whether the link Paulson draws between policy beliefs and action is dependent on the political system in place.

Paulson could respond to this in one of two ways. He could maintain that beliefs about public policy are inherently related to practical commitments, and thus the concept of rational irrationality is equally incoherent in a totalitarian dictatorship as in a liberal democracy. This response requires accepting the counterintuitive implications for the scenarios we have described, denying that they are meaningfully different from the jumping case from the perspective of rationality. Given the stark differences between these cases, this does not seem like a promising strategy.

Alternatively, Paulson could qualify his objection to rational irrationality by arguing that it is incoherent in the case of democracy but not necessarily in non-democratic contexts. This could explain the intuition that rational irrationality is not incoherent in the North Korea case. In other words, a North Korean subject who is epistemically irrational about policies might not be making any errors from the standpoint of rationality.

This line of response would require Paulson to specify the difference that renders rational irrationality coherent in the North Korean case but incoherent in the democracy case. One way of articulating this difference is to say that in a democracy, policy belief is linked to action because one has a practical path to affect policy, namely voting. In the North Korean case, no such path exists, which explains the lack of connection between political beliefs and practical commitments.

However, we must dig deeper and ask what this practical path from policy beliefs to action means or amounts to. One could point to the brute act of voting itself and claim it links policy belief to action. But this does not work for Paulson’s purposes because the believed incoherence he argues for is not between one’s belief and vote but between one’s belief and judgment about actually implementing the policy. Therefore, if voting makes the difference between the North Korea and democracy cases, it is not by virtue of the act of voting itself but rather because voting conceptually ties together one’s policy beliefs and the implementation of public policy.

Yet, seeing voting as an intermediary between policy belief and policy implementation only makes sense in probabilistic terms. To illustrate this point, consider a scenario where, instead of voting, a person conveys their political preferences by mentioning them in their prayer to God. This option is open to North Koreans as well. Would we say that now that the policy belief has turned to action (praying), it constitutes a practical commitment and the rest of Paulson’s analysis follows? Here, of course, we are assuming God does not exist, listen, or act on prayers, so there is no causal relation between the person’s prayer and the implementation of public policy.

Maintaining the North Korea/democracy distinction therefore hinges on the causal power of voting in a democracy. But this now begs the question of magnitude. Are we to say that voting links belief and practical commitments regardless of the probability of one’s vote influencing policy? This is the position Paulson needs to defend dialectically, given that he acknowledges the infinitesimal probability a vote will affect public policy.

Yet maintaining that any probabilistic magnitude renders policy belief and practical implementation connected leads to counterintuitive conclusions. It leads, for example, to the conclusion that a North Korean who writes their policy beliefs on paper that they proceed to shred and place in the trash is as practically committed to the enactment of the policy as a democratic voter in a 50-50 race. Why? Because there is a probability greater than zero that someone will reassemble their shredded paper and give it to the Supreme Leader who will proceed to enact the policy because the subject wrote it down. The probability of this scenario might be lower than the probability that a single vote will decide an election, but that does not matter given the supposed irrelevance of probabilistic magnitude.

The belief formation process

The second crucial distinction between JUMP and VOTING lies in the relationship between apparent and actual truth, and consequently, in the epistemic process required to form well-founded beliefs in each case. In JUMP, appearance and reality are perfectly aligned—the consequences of jumping from a window are immediately apparent to any competent reasoner through basic experience and common sense. No careful study or critical reflection is required to understand that jumping would lead to injury. When someone forms the belief that they should jump to collect the $100 reward, they are explicitly contradicting what they already know with, in practical terms, complete certainty.

DISEASE presents a fundamentally different epistemic structure. Here, the relationship between appearance and reality is more complex and often unclear. While statistics might suggest a poor prognosis, the actual outcome for any individual case involves numerous variables and uncertainties. Forming beliefs about one’s medical prospects requires navigating complex probabilistic information, and even then, considerable uncertainty remains. When someone in DISEASE believes in their recovery despite unfavorable odds, they are not contradicting clear and obvious truth in the way that the JUMP scenario demands.

VOTING, I argue, actually presents an even more challenging epistemic landscape, where surface appearances are often systematically misleading. Consider international trade policy: The intuitive view that protecting domestic industries through trade restrictions will preserve jobs appears self-evident. Yet careful economic analysis reveals this intuition to be false—free trade typically benefits both parties in the aggregate despite local displacements. This gap between appearance and reality is not incidental but fundamental to many policy issues. The history of economic thought illustrates this point clearly: It took the revolutionary insights of thinkers like Adam Smith and David Ricardo to demonstrate why seemingly obvious mercantilist policies were fundamentally misguided.

Similar examples abound in policy discourse. Rent control appears to be a straightforward solution to housing affordability, yet more reflective analysis reveals its counterproductive effects on housing supply and quality. Minimum wage laws seem like an obvious way to help low-income workers, but understanding their full employment effects requires careful consideration of business responses and labor market dynamics. In each case, reaching well-founded beliefs requires overcoming initial intuitions through sustained critical reflection.

This epistemic structure has important implications for rational irrationality. When Caplan describes voters as rationally irrational, he is not suggesting they believe things that directly contradict what they clearly know (as in JUMP). Rather, he is arguing that they rationally choose not to invest in the demanding process of critical reflection and bias correction that would be necessary to arrive at well-founded policy beliefs. The rationality of this choice stems not from weighing known truth against practical benefit (as in JUMP), but from weighing the costs of rigorous epistemic investigation against its benefits.

This distinction helps explain why rational irrationality appears more coherent in VOTING than in JUMP. In JUMP, the agent would need to actively contradict clear and present knowledge. In VOTING, the agent simply continues to hold intuitive but incorrect beliefs by choosing not to engage in the difficult process of critical examination. This makes VOTING more analogous to DISEASE, where beliefs can diverge from best available evidence without requiring explicit contradiction of known truth.