On the Epistemic Value of Reflection

1.1 What is Reflection?

To a first approximation, reflection is a capacity to double-check our unreflective responses to the world, including beliefs and intentions. Reflection—as I understand it—refers to a process of second-order reasoning wherein agents seek to confirm or revise their unreflective responses in light of normative considerations about the reasons for or against those responses.

For example, when I make the perceptual judgment that there is a cat in front of me, I might ask myself, “Is my perceptual experience a good reason to believe that there is a cat in front of me? I am, after all, in the museum of illusions.” Consequently, I might arrive at the second-order belief that my first-order judgment about the existence of the cat is unjustified. Reflection is thus a higher-order mental phenomenon. Following Pettit (2007: 498–500), there are two distinct kinds of higher-order reasoning: one could reason meta-attitudinally or meta-propositionally.³ The former involves reasoning about one’s own first-order attitudes, as in the museum of illusions example. The latter involves reasoning about evidential support relations, as when I ask myself the question, “Is p a good reason to believe q?” Such reasoning concerns relations between the propositional contents of mental states rather than the mental states themselves.

Like me, Kornblith (2012: 1, 28, 43, 109) understands reflection as a process of “second-order scrutiny” guided by normative considerations, and not just any kind of careful thinking. For instance, he writes:

We are capable of reflecting on our beliefs and desires, stopping to assess them, and stopping to question the wisdom and advisability of further belief or action. When we reflect, and when we engage in reflective evaluation of our first-order states, it seems that we bring to bear certain normative standards on our beliefs and actions in a way that other animals could not possibly do. (2012: 109)

As evidenced by his discussion of various cases, Kornblith (2012: 20, 89, 142) acknowledges not only meta-attitudinal reasoning, but also meta-propositional reasoning as a genuine form of reflection. Consider one such case:

Jury. Suppose that I am serving on a jury in which someone is charged with murder. Imagine as well that I don’t simply react to the evidence presented. Instead, I stop to reflect. I self-consciously consider whether the evidence presented supports a guilty verdict. (2012: 89)

In Jury, the agent does not reflect on their own first-order beliefs about the matter. Instead, the agent reflects on whether the evidence presented to them supports a particular proposition, namely that the accused is guilty. The agent’s reflection is meta-propositional, not meta-attitudinal.

Now, it is crucial to note that on this definition of reflection, it is inessential whether an agent reflects by herself or does so in collaboration with others. For example, thinking through a logic quiz can be an instance of reflection, whether one does so alone or jointly with other quizzers. And one’s reflection could be meta-attitudinal (reasoning about our own beliefs about the quiz) or meta-propositional (reasoning about the validity of steps in a proof).⁴ Social reflection is collaborative in nature, requiring the joint effort of multiple agents. Consequently, the quizzers’ reflection is an instance of social reflection only if they work through the logic quiz collaboratively.

Philosophers influenced by Descartes have tended to hold a narrow conception of reflection on which reflection just means meta-attitudinal reasoning restricted to the individual case. This conception is captured in the familiar picture of the lone Cartesian thinker who takes a step back from their inner life and asks whether the first-order attitudes they have are worth having (Frankfurt 1971; Korsgaard 1996). Even if it is the case that Kornblith’s arguments undermine only this narrow conception of reflection, it would be a mistake to think that the narrow conception was his intended target all along. For Kornblith’s (2012: 89, 109) conception of reflection is broader than the Cartesian one.

As it turns out, then, Kornblith fails to recognize the implications of the broad conception of reflection he endorses. More specifically, he overlooks the fact that meta-attitudinal and meta-propositional reasoning can be done socially as well as individually. As we shall see, this distinction is crucial in rebutting his critique of the epistemic value of reflection.

1.2 Why is Reflection Epistemically Valuable?

Philosophers diverge on what reflection can achieve, and the epistemic value it generates. Philosophers from the Harvard-Pittsburgh tradition cash out the epistemic value of reflection in terms of the basing relation that is necessary for knowledge or justified belief (Korsgaard 1996; Moran 2001). They hold that in order for an agent to believe that p on the basis that q, (i) the agent must have the higher-order reflective capacity to represent that q is a good reason for p, and (ii) that such a capacity is relevant to explaining why the agent holds the belief that p. Epistemic basing is a necessary condition on justified or rationally held belief: if one has a justified belief that p, then one believes p on the basis of good reasons. These philosophers think that without reflection we cannot have justified belief. They differ on how exactly reflection manages to establish basing relations. Some, like Moran (2001) and Leite (2004), argue that sincere avowals—such as “I really think that p because q”—can directly establish basing relations. Since knowledge requires justified belief, these philosophers are committed to saying that we cannot have knowledge without reflection either. Both knowledge and justified belief are said to have epistemic value. For many epistemologists, knowledge has epistemic value par excellence. Some epistemologists cash out the epistemic value of justified or rationally held belief without reference to knowledge (Wedgwood 2017). In the Harvard-Pittsburgh tradition, reflection is epistemically valuable to the extent that it generates knowledge or justified belief.

This approach has been accused of over-intellectualizing the basing relation, and along with it, epistemic statuses like rational belief and knowledge (Kornblith 2012: ch. 2). Opponents argue that some pre-reflective and non-reflective agents (human infants and animals respectively) can have both rational belief and knowledge. When a squirrel forms the belief that “here is an acorn,” for example, it has responded rationally to its perceptual evidence, which in turn gives it reason for belief. And so the squirrel can come to know that “here is an acorn.” The squirrel displays sensitivity to reasons, which is necessary for rational belief and knowledge. The capacity for reflection is not required for either epistemic status.

To avoid the over-intellectualizing problem, some epistemologists have argued that reflection is a necessary condition on epistemic responsibility, as opposed to justified belief or knowledge (Burge 1996; Smithies 2015). The potential to reflect on the justification of one’s beliefs opens up the agent’s management of one’s doxastic life to public criticism. Such an agent is a fair target for epistemic (as opposed to moral or pragmatic) praise or blame. In holding someone epistemically responsible for their beliefs, one makes a demand on them to comply with certain normative standards. This practice has two presuppositions. First, that the target can understand our demand. Second, that the target can bring this understanding to bear on causally regulating their own beliefs.

There are various accounts of the epistemic value of holding beliefs in an epistemically responsible way. One strategy is to say that beliefs that are held in an epistemically responsible way tend to be more reliable: epistemic responsibility is truth-conducive (Smithies 2015). The thought is that first-order beliefs that pass the test of reflective scrutiny are more likely to be true.

Whatever one thinks about epistemic responsibility and why it is epistemically valuable, several contemporary epistemologists will agree with the following minimal commitment: even if reflection is not required for epistemic responsibility, it often leads to true belief. In slogan form: reflection is a tolerably reliable method of belief formation and maintenance. Indeed, something like this minimal commitment might be required to explain philosophers’ enthusiasm for promoting critical thinking courses in university and high school settings.

To sum up, contemporary epistemologists have variously thought that reflection is necessary for (a) rational belief, (b) knowledge, or (c) epistemic responsibility. Those who disagree that reflection is required for (a) and (b), but hold that reflection is required for (c) cash out its value in terms of epistemic reliability, or in some other way. And pretty much every epistemologist will agree that forming beliefs via reflection, or maintaining them by putting them under reflective scrutiny, sometimes increases their reliability.

2.1 The Empirical Evidence for Anti-Introspection

Studies from empirical psychology provide evidence for the following phenomena:

Opacity: We often form and maintain our beliefs due to influences that we are unaware of; influences that are opaque to first-person introspection.

Confabulation: Verbal expressions of one’s reasons for belief, however sincere, are often confabulatory: the agent “generates” a normative story to justify their beliefs, such that this story has little to do with why they actually hold the relevant beliefs.

Opacity and Confabulation jointly constitute evidence for Anti-Introspection. We begin with Opacity. Although some of the empirical research I shall discuss below is explicitly only about subliminal influences on action, we can extend the moral of these empirical studies to judgment and belief, since patterns of behavior are often associated with particular judgments and beliefs.

A study conducted by Nisbett and Wilson (1977) revealed that subjects’ judgments about the quality of stockings depended significantly on the position of the stockings in the store: in an array of stockings, the stockings placed on the right end of the array were chosen four times more than the stockings placed on the left end. When asked about their reasons for their choices, subjects could only cite features like knit, weave, workmanship etc. The actual basis of their judgment, namely the relative position of the stockings, was inaccessible to introspection. Some other prominent examples of arbitrary, subliminal influences on belief and behavior involve pictures of watching eyes (Haley & Fessler 2005), pronouns (Gardner et al. 1999) and the order of names in ballots (Webber et al. 2014).

Kahneman and Tversky’s (1982) discussion of the “Asian Disease Problem” suggests that moral judgment can be affected by “ethically arbitrary factors” like how a hypothetical moral problem is verbally framed. Presented with a hypothetical epidemic scenario, different responses to proposed interventions were elicited from participants, depending upon how the scenario was described. When the expected outcome of the intervention is framed in terms of survival, people have risk-averse judgments; when the same outcome is framed in terms of expected mortality, people have risk-seeking judgments. Eskine et al. (2011) and Kelly (2011) have shown that emotions like disgust can affect moral judgment. The framing of a moral problem and certain emotional reactions are obviously wrong kinds of reasons for moral judgment/belief. In addition, people rarely, if ever, are disposed to cite such factors as reasons for belief.

Last, there is a vast literature on implicit bias or implicit cognition.⁶ Implicit bias involves the pre-reflective attribution of negative qualities to members of social out-groups, e.g. by race, age, gender, sexuality, weight, or religion. Even if such bias is best modelled using conceptual categories other than belief, the effects of implicit bias appear at the level of both action and belief. Consider implicit racial bias. A study conducted by Bertrand and Mullianathan (2003) showed that fabricated résumés sent in response to employment advertisements in Boston and Chicago newspapers elicited fifty percent more interview callbacks if they contained “white sounding” names as compared to “African American sounding” names. This is clearly a case of racial bias affecting action, but there may be associated beliefs here too, i.e., beliefs accompanying racist actions, like “Jamal does not seem to be a guy who is fit for this job.”

Now, it is not as if subjects in these studies will stay quiet or remain baffled about their reasons for action or belief. On the contrary, subjects typically produce very confident answers when prompted. Confabulation documents a specific instance of a more general phenomenon which goes by the same name: confabulation, understood generally, concerns sincere, confident but erroneous reporting about various aspects of oneself, including one’s reasons for action or belief. According to Doris (2015: 82), prominent researchers like Nisbett and Wilson (1977: 233), Gazzaniga (2000: 1316–1321), Hirstein (2005: ch. 1), and Carruthers (2009: 126–127) all assert that confabulation, understood generally, is not restricted to patients with pathological disorders, but is common among healthy people.

A study done by Estabrooks (1957: 86–87) showed that subjects always found socially acceptable ways to justify or excuse their behavior when “incomprehensible behavior” was induced into them via hypnosis. Participants in a study conducted by Maier (1931) were asked to solve a problem. Those who had difficulty finding a solution were given a hint. Even when the hint was helpful, participants did not credit it for their success. Instead, they said the solution “dawned” on them, or they credited their past courses of study. A study done by Latané and Darley (1970) on the so-called “Group Effect” demonstrates the tenacity with which people hold onto their confabulations. “Group Effect” refers to the inversely proportional relationship between a tendency in individuals to extend help and the number of bystanders present. Despite being challenged, participants in the study denied that bystanders had anything to do with their behavior.

For the sake of argument, I shall grant my opponents’ assessment of the empirical evidence in favor of Opacity and Confabulation. Therefore, by extension, I shall grant Anti-Introspection. I recognize that some philosophers and scientists may push back here. Philosophers have an important role to play in clarifying the concepts and methods used in empirical research. Sometimes, the upshot is that some strand of empirical research is orthogonal to a long-standing philosophical discussion or debate. When some empirical research is relevant to a philosophical debate, philosophers can point out hasty generalizations based on that research. Scientists might have their own concerns: measurement problems, replicability, data dredging, publication bias, and more specific concerns regarding the methods and assumptions of individual studies. I shall keep all such skepticism at bay. To block The Introspection Argument, my strategy is instead to reject the Introspection Requirement.

2.2 Rejecting the Introspection Requirement

On a popular way of understanding epistemic basing, the reasons for which one holds a belief are already in place before one reflects on the normative credentials of one’s belief. Kornblith (2012: 21–22) maintains that reflection can be reliable only if we have reliable introspective access to basing relations, i.e., the causal history of one’s belief.

Consider a case that might help motivate the Introspection Requirement. Bad Detective believes that the butler committed the murder. Before submitting a report of his investigation to the police, he asks himself, “Am I right in believing the butler is the murderer? Do I really have good reasons to think that?” He introspects on his own reasons for belief. Not long afterwards, he tells himself that he has enough circumstantial evidence to make the case that the butler did it. As it turns out, his false belief that the butler did it is actually based on racial bias: Bad Detective has a deep-seated hatred of people who belong to the same race as the butler. He does not have reliable introspective access to the actual basis for his belief, and when he considers what reasons he has for his belief, he confabulates. While reflecting, Bad Detective is unable to see the proper force of the evidence he has because his first-order belief about the butler is maintained due to racial bias. As a result, Bad Detective forms the false second-order belief that his first-order belief about the butler is well-proportioned to the evidence. His reflection failed to improve his epistemic condition because he could not accurately identify and correct his own bias.

But do we really need reliable introspective access to basing relations to improve the accuracy of our first-order beliefs via reflection? I think not. Reliable introspective access to basing relations is unnecessary for the project of epistemic amelioration. The causal history of my belief constitutes my backward-looking reason for belief. But reflection can proceed in an entirely forward-looking manner. To improve my epistemic condition, I do not need to reflect on what my own reasons are for believing something. Instead, I can pay attention to my evidence and re-examine it carefully. The evidence I have at my disposal gives me forward-looking reasons for belief. I need to know what my evidence supports and respond to it accordingly. In doing so, I may get closer to the truth.

To get a sense of how this might work, consider a variant of the case discussed earlier. Good Detective believes that the butler committed the murder. Before submitting a report of his investigation to the police, he asks himself, “Did the butler really commit the murder?” He does not explicitly think about what his own reasons are for believing that the butler did the murder. (Let us suppose that like Bad Detective, Good Detective does not have reliable introspective access to the reasons for his belief. Furthermore, let us suppose that the Good Detective would engage in confabulation if he were to seriously reflect on his own reasons for belief). Instead, Good Detective considers afresh the complicated body of evidence in his possession. After a careful re-examination of the evidence, Good Detective realizes that he has been mistaken all along: the evidence does not point to the butler, it points to the businessman. The businessman had deliberately planted some evidence on the crime scene which was supposed to mislead investigators into thinking that his butler committed the crime. Having figured this all out, Good Detective changes his mind. He acquires the true belief that the businessman committed the murder. And he does this without making any commitments about the basis on which he formerly believed that the butler committed the murder. Good Detective knows what his evidence supports, and he responds to it accordingly. He has improved his epistemic condition. In this way, reflection can proceed in an entirely forward-looking manner.⁷

I take it that the Good Detective case is a kind of possibility proof: forward-looking reflection can succeed even if backward-looking reflection has failed. In other words, even if one does not have reliable introspective access to basing relations, one can still successfully carry out the project of ameliorating one’s first-order beliefs via reflection.

Now, my opponent could concede this point but wonder how often we are in the good case as opposed to being in the bad case. To reject The Introspection Argument, however, the burden is not on me to establish the positive claim that forward-looking reflection succeeds often enough to count as “reliable” even when backward-looking reflection has failed. That said, let me put forward an optimistic, though tentative suggestion. Harman (1986: 41–42) rightly observed that many of our beliefs are formed on the basis of evidence that we have now forgotten. Some of these beliefs are false. If one is to improve upon these false beliefs via reflection, then one must begin by considering afresh the evidence one currently has at one’s disposal. Thus, to the extent that we succeed in improving—via reflection—the accuracy of beliefs whose bases we have now forgotten, we do it via forward-looking reflection, not backward-looking reflection.

Still, one might have positive reasons for thinking that we are more often in the bad case rather than the good case. One might think that we are biased in all kinds of ways. Plausibly, bias negatively impacts our ability to properly see the force of the evidence, and ultimately leads us to resist epistemically fruitful belief revision. This is what happened with Bad Detective.

But are we always, or for the most part, biased? Can we not overcome our biases? These are interesting empirical questions, but they express distinct worries related to The Efficacy Argument. That argument attacks the claim that our reflection is causally efficacious in the ways we want it to be, which includes overcoming wishful-thinking and bias. I shall critically evaluate The Efficacy Argument in section 3.

For now, let me point out that The Introspection Argument neither states nor entails the claim that we are always, or for the most part, biased. Opacity and Confabulation do not—individually or jointly—entail that one will remain unmoved after re-evaluating one’s evidence. Nor is it clear that Opacity and Confabulation increase the likelihood that one is biased. The Introspection Argument provides fuel for skepticism regarding the reliability of backward-looking reflection—i.e., reflection that incorporates introspection on basing relations. It provides no grounds for skepticism regarding the reliability of reasoning in general. For if we were hopelessly biased, all reasoning—either reflective or unreflective—would be unreliable. Surely, that is not what The Introspection Argument is supposed to show.

The Introspection Requirement stands unmotivated. As illustrated in the Good Detective case, we can take steps towards epistemic amelioration so long as we know what our evidence supports. Having reliable access to basing relations is a superfluous requirement on the reliability of reflection. Perhaps the Introspection Requirement as formulated is too strong to be plausible. Might there be room for a weaker version of the introspection requirement?

The answer to that question depends on our view of evidence. On a familiar view of evidence, my evidence is constituted only by my mental states. On such a view, even on the forward-looking picture of reflection, I would need introspective access to my mental states. For I can know what my evidence supports only if I have access to my evidence. At this point, one might wonder whether the empirical research that Kornblith and Doris cite in support of Opacity and Confabulation undermines the kind of introspective access we need to get forward-looking reflection going. It does not. The relevant empirical research only undermines backward-looking introspection: that is, reliable access to basing relations. Subliminal influences on belief formation and our impressive ability to cook up normative stories to justify our beliefs speak to the fact that we are often in the dark about our reasons for belief. Since one’s total evidence is not exhausted by the reasons on which one’s beliefs are based, lacking reliable access to basing relations does not entail that we lack reliable access to our evidence simpliciter. In the Good Detective case, the agent may not have access to the evidence that moved him to believe that ‘the butler committed the murder.’ But that is no reason to think that he does not have access to other bits of evidence. Thus, even if one could motivate an introspection requirement on the reliability of reflection, we have not been given a plausible argument to think that we would fail to meet such a requirement.

On an alternative view of evidence, one’s evidence is not constituted by internal mental states, but by facts or states of affairs. On this view, Good Detective’s revised belief that ‘the businessman committed the murder’ is justified not on the basis of the beliefs he holds about the businessman’s motives and behavior, but on the basis of facts like the businessman planted misleading evidence. Of course, on such a view, reflection can proceed in a forward-looking manner without the need for any kind of introspective access at all.

Whether or not there is room for some introspection requirement on reflection, the Introspection Requirement as stated is too strong to be plausible. If my argument so far is on point, we have earned the right to reject The Introspection Argument on the grounds that it imposes a superfluous requirement on the reliability of reflection.

This brings us to The Efficacy Argument. This argument concerns whether reflection is causally efficacious i.e., whether it leads us to change our minds for the better. Empirical research—which I have not yet discussed—seems to suggest that reflection is unlikely to prompt a change of mind. The Efficacy Argument allows that cases like the Good Detective case are possible but maintains that they are somewhat anomalous. It maintains that reflection, for the most part, fails to increase the accuracy of our first-order beliefs.

3.1 The Empirical Evidence for Anti-Efficacy

Studies from empirical psychology provide evidence not just for Opacity and Confabulation, but also for the following phenomenon:

Epiphenomenality: When we reflect on the normative adequacy of the reasons for our beliefs, our higher-order deliberation is often either:

(a) causally inert: it does not influence why we continue holding onto a belief (Inertness), or

(b) it causally influences our first-order beliefs in the wrong kind of way (Counterproductivity).

Note that Opacity and Confabulation do not individually or jointly entail Epiphenomenality. It is one thing to say that people’s higher-order deliberation is often disconnected from their actual reasons for belief or action, and quite another to say that such higher-order deliberation does not causally alter one’s first-order mental states after the fact. Epiphenomenality is supported by studies on motivated cognition, self-enhancement, and confirmation bias.

Motivated cognition and self-enhancement, as the names suggest, involve a tendency in individuals to think highly of their own credentials (epistemic or otherwise). Strictly speaking, motivated cognition refers to believing what one wants to believe. However, almost everyone wants to believe better of themselves. People find ingenious ways of making their doxastic and conative lives “look good.”⁸ Such factors strongly dispose people in preserving the first-order beliefs they have already formed. Thus, the evidence one has at one’s own disposal is taken at face value or selectively scrutinized. Evidence in favor of one’s beliefs is better remembered. New evidence supporting one’s beliefs is easily accepted, whereas opposing evidence is more easily rejected. Moreover, the process of searching for new evidence is skewed in such a way that the results are likely to favor what one already believes.⁹

Kornblith’s (2012: 25–26) overall assessment of this research is as follows. We have little reason to hope that episodes of reflection alter, let alone improve, our epistemic condition. Reflection suffers from Inertness: episodes of higher-order deliberation on one’s reasons for belief are little more than exercises in “self-congratulation.” Instead of resulting in epistemically fruitful belief revision, they produce an agent who has a false sense of security about the accuracy of their first-order beliefs. If anything, the agent becomes more entrenched and dogmatic after subjecting their beliefs to critical scrutiny. Thus, reflection is also susceptible to Counterproductivity: even when reflection exerts causal influence on one’s first-order beliefs, it often does so in the wrong kind of way.

For the sake of argument, I shall grant Kornblith’s assessment of the empirical research, but with one important qualification. The empirical research—even if it is shown to be solid—only impugns the reliability of individual reflection. However, for The Efficacy Argument to go through, Anti-Efficacy must be true for reflection in general.

In what follows, I will argue that the case for Anti-Efficacy has not been made for reflection in general. To be clear, I will not be arguing for the positive claim that social reflection is reliable. My goal is to rebut the negative claim that social reflection is bound to be unreliable for much the same reasons that putatively undermine the reliability of individual reflection. Since my goal is not to provide a positive argument for the reliability of social reflection, I will only be providing a provisional defense of the comparative claim that social reflection does better than individual reflection in several respects, rather than the non-comparative claim that social reflection does well enough overall to count as reliable.

I will argue that—under certain conditions—social reflection is not plagued by Inertness or Counterproductivity. In other words, it is not self-congratulatory, it does not make us more intransigent, and it does not lull us into a false sense of epistemic security.

3.2 Resisting Anti-Efficacy in the Context of Social Reflection

To see why worries about the putative Epiphenomenality of individual reflection do not generalize to social reflection, we must focus on certain unique features of social reflection.

First, social reflection often involves dealing with challenges to our beliefs in a way that individual reflection does not. Second, challenges to our beliefs encountered socially are harder to push under the rug, so to speak. Consequently, when faced with such challenges—as opposed to challenges encountered individually—one is more likely to change one’s mind. Third, due to what I will call the critical function of social reflection, one’s higher-order deliberation on normative reasons for belief is far from “self-congratulatory.” Instead, one is confronted with powerful objections and arguments against one’s beliefs. Fourth, under a range of circumstances, one can harness the critical function of social reflection to bring about a change of mind for the better.

The first two features speak against the Inertness of social reflection, whereas the last two raise doubts about its Counterproductivity. Let us consider each pair of features in turn.

3.2.1 Resisting Inertness

The evidence for Epiphenomenality—which I have granted for the sake of argument—suggests that we do not often reflect individually on the evidence we have for our beliefs. If we are strongly predisposed to preserve the first-order beliefs we have already formed, then it is hardly surprising that we do not investigate the accuracy of our first-order beliefs by critically reflecting on the evidence we have for them. Furthermore, even when we do manage to reflect on the evidence we have for our beliefs, we rarely disagree with ourselves. More often than not, one’s individual reflection regarding the normative question of whether one should believe p will be in harmony with one’s first-order belief that p—even if one entertains objections or considers contrary evidence, one will, due to confirmation bias and the like, eventually end up “rationalizing” what one already believes (Evans & Wason 1974).

So far, I have mentioned contingent limitations of individual reflection. There are normative limitations as well. It is widely assumed that rationality demands that an agent’s first-order beliefs be consistent with her higher-order beliefs. An agent who believes that p, and that “I do not really have good evidence for my belief that p” is being epistemically irrational.¹⁰ Relatedly, there is rational pressure on us to not disagree with ourselves. An agent who believes the conjunction “p but after carefully reflecting on my evidence, ~p,” is epistemically irrational. The norms of rationality therefore prohibit disagreeing with ourselves.

It seems, therefore, that when we form false first-order beliefs, we dig ourselves into a hole too deep for individual reflection to succeed in correcting our false beliefs. Whether or not this really is the case with individual reflection, this need not be the case with social reflection.

For starters, interpersonal disagreement is more common than intrapersonal disagreement. When I assert my belief that p, my interlocuter might challenge my assertion and I will be invited to consider whether p, or why p. When I publicly—as opposed to privately—assert my belief that p, I am more likely to encounter disagreement. This fact is both normatively and empirically significant.

Disagreement with others puts rational pressure on me to reconsider whether I have responded appropriately to my evidence. The extent of rational pressure to reconsider my response may vary, depending on the nature of the disagreement. If my interlocutor and I have the same evidence, then one of us has responded wrongly to the evidence, and it would be dogmatic to discount the possibility that I am the guilty party. If my interlocutor has different evidence, then one of us has better evidence, and it would be dogmatic to discount the possibility that mine is inferior. The mere fact of disagreement raises the possibility that I might be wrong. In a social context where the truth of p is being challenged, there is some rational pressure on me to reconsider whether p. In this way, public disagreement is normatively significant.

Public disagreement is also empirically significant. It often triggers social reflection. When agents disagree about some matter, they often pursue their disagreement collaboratively, by jointly pursuing the question, “What does the evidence give us good reason to believe?” More importantly, social reflection that stems from interpersonal disagreement is more likely to result in a change of mind, in comparison with individual reflection.¹¹ As Mercier (2020: 47–50) points out, with respect to “non-sensitive” cases of disagreement (e.g., arithmetic, geography etc.), interpersonal disagreement usually results in a change of mind. With “sensitive” (or emotionally fraught) topics like religion, politics, or morality, we sometimes witness the so-called “backfire effect”: presented with contrary opinions or arguments from interlocutors, individuals become more entrenched in their views (Nyhan & Reifer 2010). As we shall see later, this need not rule out the prospects of social reflection in increasing the reliability of our beliefs regarding “sensitive” topics.

To recap so far: (1) we are more likely to encounter disagreement when we reflect socially as opposed to reflecting individually, (2) disagreement puts rational pressure on us to reassess our responses to our evidence, (3) encountering disagreement publicly is more causally efficacious than intrapersonal disagreement, and (4) likely effects of encountering disagreement publicly include a change of mind. So far, I have said nothing to establish that social reflection increases the reliability of our first-order beliefs. When we reflect socially, are we more likely to change our minds for the better? If so, how? This brings us to Counterproductivity. Part of the answer to those questions lies in the critical function of social reflection, which speaks to the potential of social reflection in generating a better quality of higher-order deliberation on normative reasons for belief.

3.2.2 Resisting Counterproductivity

For the most part, we are lousy at coming up with good arguments against our own beliefs. However, we are much better when it comes to finding good objections to, and producing solid arguments against, others’ beliefs. This is what I shall call the critical function of social reflection. Psychology tells us that we are more critical of others’ professed beliefs, and that we are quicker and better at undermining them.

A study done by Kuhn et al. (1994) asked mock jurors to give their verdict on a particular case. Most mock jurors, when presented with an alternative verdict, were able to produce a counterargument to it. Another study done by Shaw (1996) showed that all the participants in the study could quickly come up with counterarguments against a claim which was not theirs. In a study done by Trouche et al. (2016) demonstrating the “selective laziness” of human reasoning, researchers attempted to trick participants into thinking that their own arguments for a claim were someone else’s. Over half the participants who fell for the trick rejected the argument (which was in fact theirs). Furthermore, participants were more likely to reject invalid arguments, which suggests that they were better able to tell valid from invalid arguments when they perceived the arguments as not theirs. Resnick et al. (1993) created groups of three participants who disagreed over an issue. In such an “argumentative setting,” what they observed was that participants demonstrated skill in reasoning, including the ability to understand argument structure, to build complex arguments of their own, and to identify and undermine premises of others’ arguments. The key point here is that the critical function of our reasoning faculties is enhanced in a social, argumentative setting.¹²

In the game of giving and asking for reasons, the listener remains “epistemically vigilant”: More often than not, instead of immediately accepting the speaker’s assertion as true, the listener asks clarificatory questions, and offers pushback (Sperber et al. 2010). Speakers are typically aware of this, and so they tailor-make their arguments to convince their listeners, say, by appealing to the listener’s own background beliefs. As Mercier and Sperber (2017: 222–237) have shown elsewhere, listeners’ “epistemic vigilance” acts as quality control viz-a-vis the speakers’ claims and arguments. The speaker’s “myside bias” or motivated reasoning is mitigated: although the speaker retains a strong tendency to justify her own beliefs, she is forced to produce contextually persuasive arguments for them. As Mercier and Sperber (2011: 60) put it, this speaker-listener dynamic, among other things, makes communication (on the whole) more reliable and hence more advantageous.

Thus, in addition to factors (1)–(4) discussed earlier, (5) the critical function of social reflection, and (6) the dynamic between speakers and listeners in the game of giving and asking for reasons, all speak in favor of the potential of social reflection in generating a better quality of higher-order deliberation on normative reasons for belief. Furthermore, even if listeners are on guard, disagreement often results in a lowering of confidence in their views, which sometimes culminates in a change of mind.

Now, a crucial question is: under what conditions do individuals harness the potential of social reflection to successfully ameliorate their own epistemic condition? In response, I will discuss three conditions that favor epistemic amelioration in the context of social reflection: group settings, viewpoint diversity, and the formal division of cognitive labor.

3.3 Summary and Upshot

To sum up: empirical evidence suggests that social reflection—under certain conditions—is not beset with the same distortions that plague individual reflection. Due to interpersonal disagreement, which often spurs social reflection, an individual is more likely to change their mind. The critical function of social reflection is enhanced in group settings such as public debates, discussions where cognitively diverse individuals try to solve problems, public fora with moderate ideological diversity, or via the peer-review process. In such social contexts, one’s chances of changing one’s mind for the better increase. Thus, in a range of conditions, social reflection fares better than individual reflection in increasing the accuracy of our first-order beliefs. And so, the case for the unreliability of individual reflection does not generalize to social reflection.

Now, one could still wonder about the extent to which social reflection is carried out under the kinds of favorable conditions I mentioned. This remains an important empirical issue in adjudicating the overall reliability of social reflection—a task which I have not undertaken in this paper for two reasons. First, the extent to which social reflection is carried out under favorable conditions is an issue that cannot be adjudicated without a careful evaluation of the distinct macro-level distortions faced by social reflection. While I briefly discuss these distortions in the conclusion, such an evaluation is beyond the scope of this paper. Second, given my argumentative aims, this evaluation is unnecessary. To defend the reliability of reflection against Kornblith’s (2012) critique, as opposed to the ambitious task of providing a positive argument for the overall reliability of reflection, it is enough to show that the case against the reliability of individual reflection does not generalize to social reflection. My discussion on the unique features of social reflection suffices for that purpose. Furthermore, it gives us prima facie grounds for optimism regarding the reliability of social reflection.

I therefore reject The Efficacy Argument on the grounds that we lack sufficient evidence to believe that Anti-Efficacy is true for reflection in general.