According to Hilary Kornblith, we philosophers have “assigned a great deal more value to reflection than it really deserves” because we have had a “terribly inaccurate view” of the nature of reflection and its powers. In On Reflection (Reference Kornblith2012), he argues for these claims with respect to the appeal to reflection in connection with knowledge, reasoning, free will and normativity and, then, in a brief closing chapter, provides an empirically informed and avowedly “demystified” account of the nature of reflection. The book is clearly and straightforwardly written and it succeeds both in refuting many of its targets and in bringing unity to apparent diversity. In spite of these virtues, it suffers from a lack of clarity regarding the exact nature of its unifying subject. This lack of clarity raises doubts about the general significance of its subject to philosophical inquiry and about whether all of the many topics discussed are really related in significant ways.

On Kornblith's accounting, the same two difficulties appear each time some reflective second-order belief or desire is alleged to be required to solve a philosophical problem with first-order beliefs or desires. First, the problem which gave rise to the imposition of the second-order state requirement arises again with respect to the second-order state and, by the same reasoning, gives rise to a third-order state requirement, which, in turn, gives rise to a similar problem. An infinite regress arises, showing that the problem cannot be solved in that manner. Second, there are broadly empirical reasons for thinking that reflection is not, even at the initial stage, really up to the task which it is invoked to perform. In each of his first four chapters, Kornblith makes the case that both problems afflict the appeal to higher-order beliefs and desires in the aforementioned four domains of philosophical concern.

Kornblith's first chapter is addressed to knowledge and other epistemological notions and is a descendent of his discussion in his Knowledge and Its Place in Nature (Reference Kornblith2002). It begins with a consideration of the suggestion that a necessary condition on justified belief that p is that one reflect critically on one's belief that p. In Kornblith's hands, this requirement to reflect on one's beliefs is construed a requirement that one have another (second-order) belief regarding the status of the initial belief. However, the second-order belief cannot, it seems, contribute to the justification of the first-order one unless it is itself justified and by the putative necessary condition on all justified belief, such a second-order belief cannot be justified unless one has a third-order belief regarding the second-order one. An infinite regress of levels ensues and, says Kornblith, the result is that no belief is ever justified.

This argument (and each of the structurally analogous regress arguments considered in subsequent chapters) is successful only if the sort of reflection allegedly required can occur only if one forms a higher-order state (in this case, belief) about the target state and if the higher-order state at issue is itself open to the same sorts of worries. A proponent of the critical reflection requirement might hold that the duty to reflect need not produce (though it can) beliefs regarding the epistemic virtues of one's first-order belief. Instead, it requires that one consider whether one's belief has sufficient evidential support. Such consideration may require only that one occupy various occurrent non-belief states such as states of direct acquaintance or rational intuition. If so, then the duty to reflect on one's beliefs will not yield an infinite regress. Still, it remains true that if one must actually have reflected on a belief for it to be justified, then precious few of our beliefs are in fact justified.

Kornblith considers the attempt to avoid the regress produced by the higher-order belief requirement by requiring higher-order belief for a certain kind of positive epistemic status but not for another kind of positive epistemic status. Taking Ernest Sosa's distinction between reflective knowledge and animal knowledge as an exemplar of this kind of approach, Kornblith argues that while such a distinction is well-defined, there is no significant difference of kind between reflective knowledge and animal knowledge.

The suggestion that beliefs on which we have reflected are more likely to be true than beliefs on which we have not is subjected to particular criticism by Kornblith. The evidence adduced against this suggestion will be familiar to those who have followed Kornblith's work in recent decades. It consists of accounts of studies from cognitive and social psychology which, Kornblith claims, show that we are often quite ignorant of the causes of our beliefs in cases in which they are unreliably caused and, what is worse, that we are inclined to rationalizations when we reflect on why we hold those beliefs. Where we lack good grounds, reflection yields confabulation and the false conviction that we have good grounds. So, the idea that reflecting on our beliefs is generally likely to yield epistemic improvements is, according to Kornblith, false. Instead, “we are engaged in a process which, in a very wide range of cases, makes us more confident that we are right than before we began, but a process … which is almost useless for improving our reliability” (24).

While Kornblith is most focused on our ability to reflectively determine what caused various of our beliefs or judgments, many of us are more concerned on whether we have good evidence or grounds for our beliefs. It seems far more plausible that reflection is able to detect the presence or absence of evidence than the unconscious causes of our opinions. Indeed, it seems plausible that reflection – reflection more serious and sustained than that evinced by the experimental subjects Kornblith discusses – on our grounds for the beliefs at issue might result in our correctly determining that we lack such grounds and, ideally, in our giving up the beliefs at issue. Moreover, one might think that the evidence summarized by Kornblith, while showing that such reflection is difficult, surely does not show that we are generally bad at determining, given sufficient attention and time, whether we have grounds for our beliefs.Footnote ¹

Kornblith is equally critical regarding a number of suggested alternative epistemological roles for reflection. According to the first, reflection is uniquely indispensable or foundational as only reflection can properly assess the merits of beliefs or belief sources. According to the second, the traditional significance of skeptical concerns in epistemology can only be explained by the significance of reflectively screened belief. The third maintains that reflection allows for epistemic agency or autonomy in a way which other modes of belief fixation do not.

On the first suggestion, reflection plays a foundational and ineliminable role in the assessment of our beliefs. Even if we use other means of belief-fixation and even if we might sometimes evaluate reflection by appeal to such other means of arriving at belief, the merits of our other belief sources and their relevance for assessing reflection can only be properly assessed by reflection. Kornblith disagrees and argues that it is possible to assess the epistemic merits of reflection without using reflection in the process. Considering psychologists who investigate experimentally the reliability of reflection (and, presumably, naturalistic philosophers seeking to rely upon their work) he argues that they need not use reflection in their assessment. In the course of his discussion, he notes that the psychologists might well think “at some length about the data” they have produced and about “what would explain the data.” While one might be inclined to think of such explicit theorizing about the evidential support of some proposition as an instance of reflection, Kornblith is quite clear that this (admittedly more standard) sense of “reflection” has “nothing to do with the kind of reflection under discussion” in his book, because “the process of reflection under discussion [in the book] involves thinking about one's own first-order mental states in a first-personal way” (28, emphases added). While the notion of a “first-personal way” could, I think, use substantially more unpacking than it receives from Kornblith,Footnote ² it seems clear that, as construed by Kornblith, reflection is either the process of introspecting or, on an even broader construal of “thinking,” one's being in mental states which have as their content propositions regarding one's own mental states (considered as one's own mental states).Footnote ³

This is one of the most explicit passages in which Kornblith characterizes the central unifying target of his discussion and so we ought to pause to note that it yields a narrower target for his critique than many readers will have initially supposed and this narrower target correspondingly diminishes the significance of his critique. This account of the nature of reflection seems to render false Kornblith's claim that “philosophy itself is a reflective activity” (1). While philosophy is a reflective activity in the thicker commonsensical sense of “reflection” discussed above, the vast majority of philosophical inquiry is not reflective in Kornblith's proprietary sense at all as it does not essentially involve introspection or second-order states. Philosophical investigation of knowledge and justification, the metaphysics of persistence, the nature of distributive justice, the proper account of confirmation, and the correct decision theory, for example, don't seem to involve introspection or beliefs about one's own beliefs at all, though they seem paradigmatically philosophical enterprises.

Moreover, even in the case of self-conscious consideration of the probity of what one presently believes or desires, reflection, so construed, may play an extremely minimal role. For example, if I, introspecting that I believe p, go on to consider whether or not other propositions (which I happen to believe but where the claim that I believe them is not the content of any mental state of mine) serve as good reasons for p, then the only use of reflection in Kornblith's sense consists in my initial introspective awareness of p.Footnote ⁴

Finally, for reasons just outlined, a minimal but indispensable role for reflection is guaranteed for our hypothetical psychologist or naturalistic epistemologist whenever she considers the question of whether or not her own belief in the unreliability of reflection (or, indeed, any belief she has) is justified or evidentially supported. Indeed, while Kornblith may be correct that one need not use reflection oneself in the psychological investigation of reflection, the results of such empirical investigations will be able to serve an ameliorative role only if the investigator is able to adjust her own opinions in light of such results. While it may be possible in some broad sense for such results to affect one's beliefs without one's using reflection, it seems quite unlikely to occur absent reflection.

The final section of Kornblith's first chapter considers a related way of ensuring the epistemological significance of reflection, construed as noted above. A variety of internalist epistemologists are internalists because they regard the fundamental epistemological question, “What ought I to believe?” as properly answered only from the first-person perspective. Many such philosophers who focus on the first-person pronoun will regard reflection – i.e. introspection – as essential to justification and knowledge because they will, like Bonjour, require that “what is appealed to for justification must be internal to the individual's first-person cognitive perspective” (Reference Bonjour2001, 54). As Kornblith notes, the suitably “internal” items here are her conscious mental states and her rational intuitions because these items yield justification which is allegedly direct or unproblematic. The envisioned internalist, then, holds that all belief must be justified ultimately by one's conscious mental states and one's rational intuitions. Kornblith, again, rejects the claim that there is anything epistemologically unique about such an epistemic basis. One can, he avers, “raise skeptical worries about features of one's mental life just as easily as one can raise worries about our knowledge of the physical world” (38) and so the introspective beliefs at issue are not “unproblematic in any way that matters to epistemology” (39).Footnote ⁵

This repudiation of a broadly Cartesian perspective on epistemology is, I believe, too quick. First, the Cartesian perspective isn't focused simply on “our mental life” but on our current conscious mental states. Even if, contrary to Descartes, there are mental states and mental processes which aren't conscious, the Cartesian claim that we have a distinctive sort of access to conscious mental states remains plausible. Second, the psychological results to which Kornblith earlier appeals seem to feature people unaware of the causes of their conscious mental states or people making retrospective mistakes about which conscious states they previously occupied. These are not domains to which the Cartesian maintains we have a special access. Third, and most significant, it isn't at all clear that one can raise the same sorts of skeptical worries about our beliefs regarding our current conscious mental states as one can about our beliefs about the contingent external world. Merely noting that one can form false judgments regarding one's contemporaneous conscious mental states does not show that one can be mistaken in every such judgment nor does it show that one cannot stand in cognitive relations to such states more epistemologically intimate than one can stand in to non-conscious contingent states of affairs. So, even if it is the case that our first-person grasp of our conscious mental lives is not utterly unproblematic, it may well be significantly less problematic than our grasp of the contingent external world.

I won't pretend that these venerable and foundational issues can be settled here.Footnote ⁶ And, in fairness to Kornblith, his broad synthetic aims in the book require that many issues receive only a cursory treatment. Still, while the connections Kornblith draws between the disparate topics considered are illuminating, I wager that my dissatisfaction about the inconclusive nature of the discussion here is apt to be felt by others at various points when reading the book. A book which covers a lot of ground in a variety of specialized areas is unlikely to provide a detailed and satisfactory discussion for specialists in each area, especially those inclined to defend the views which come under criticism.

Kornblith's second chapter is addressed to views on which second-order beliefs about one's mental states are a necessary condition for reasoning. What is it for one to form a belief that p on the basis of reasons?Footnote ⁷ This question is often answered by appealing to the existence of a causal relation between the belief that p and whatever belief (or other psychological state) which serves as its reason. On such views, “if I am to reason from one belief to another … then the first belief must have played some causal role in producing (or sustaining) the second belief” (42). Such a causal connection, however, is not enough, as the problem of deviant causal chains shows. Kornblith considers an additional condition, beyond the causal one, which holds that to be engaged in genuine reasoning one must also believe that the first belief provides one with a good reason for believing the second. As such second-order beliefs have their “origins in reflection,” a second-order state likely produced by introspection is alleged required to distinguish between mere causal transitions between mental states and genuine reasoning.

A simple version of this view leads, Kornblith argues, to an infinite regress. If one appeals to a second-order belief that one's belief that p is a good reason to believe q in order to distinguish reasoning from a mere arational causal process, then it would seem that the higher-level belief must also play a similarly suitable role in the reasoning and this, in turn, seems to require that one must have a third-order belief about one's second-order belief in order that the second-order belief is, along with belief that p, a good reason for believing that q. An infinite regress again seems to ensue.Footnote ⁸

One way around this problem, which Kornblith finds suggested in the work of Sydney Shoemaker, holds that the higher-level beliefs and desires allegedly required for reasoning supervene on an agent's having the requisite first-order states and having sufficient rationality and conceptual capacities. Kornblith argues compellingly that such a view conflates the possession of capacities with their exercise and, regardless, that this won't stop the regress because even our mere conceptual capacities do not extend to include concepts of beliefs of arbitrarily high order. So, there will be some level at which, even on Shoemaker's account of the metaphysics of higher-level belief, there is no requisite higher-level belief.

The remainder of the chapter considers a view (which Kornblith attributes to Brandom, Davidson, Haugeland, McDowell, Sellars and Williams) on which a creature cannot have any beliefs at all without having higher-order beliefs or, at least, the current capacity for such beliefs. Kornblith argues that this view has great difficulty accommodating the empirical evidence that non-human animals have states which seem to be beliefs and the empirical evidence that human cognitive development involves the acquisition of first-order beliefs prior to the capacity for higher-order beliefs.

Chapter 3 is concerned with views on which reflection is a requirement for some sort of significant freedom or agency. Again, Kornblith argues that no such requirement is sound. The most famous instance of such a view is due to Harry Frankfurt. Frankfurt held that freedom of the will required second-order desires or volitions – desires that one's first-order desires be effective in action. One has free will, on this account, when one acts on a desire which one reflectively endorses – a desire which one desires to have or act upon. As Kornblith argues compellingly, this account also leads to an infinite regress. According to Frankfurt, if a desire is not suitably endorsed by a higher-order desire, then actions to which it contributes will not be actions of a free will. Such a requirement seems to apply equally to a second-order desire and require a third-order desire that one's second-order desire that one's first order desire be effective in action. Here, again, a first-order state is judged insufficient for some important property or attribute absent coupling with a second-order state of the same kind which has the first order state as its object. And, again, it seems clear that the second-order state, being of the same fundamental sort as the first-order state which gave rise to the problem in the first place, is incapable of solving the problem.

The bulk of Chapter Three is taken up with murky questions surrounding distinctively epistemic agency. As some have noted, we sometimes reflect on what we ought to believe and deliberate about what is best justified. In such situations, we seem to ourselves to be playing an active role in belief formation and, in some sense, deciding what to believe. Kornblith holds that “the notion of epistemic agency … deserves more scrutiny than it has thus far received” (88) and he subjects the notion to searching scrutiny. He grants, for the sake of argument, that there are indeed actions one might undertake during epistemic deliberation. For example, one might decide to think about a topic, to consult various sources of evidence, or to reevaluate one's views. He points out, however, that such activities seem in principle no different from the sorts of activity or action which accompany beliefs usually thought paradigms of passively formed beliefs. For example, I might actively turn my head to look in a direction prior to forming the belief that there is an animal in the bushes, but such deliberate action doesn't make the resulting perceptual belief a belief which I decided to have. Similarly, even in the fully reflective and self-conscious case, it seems that the ultimate belief formed is not distinctively a product of epistemic agency. So, if the worry is that we are somehow too passive in ordinary belief formation, the problem seems not evaded by reflection.

Chapter Four is addressed to the large topic of normativity. Kornblith, by way of initially motivating the thought that reflection is importantly connected to normativity, suggests that a “realist view about normative demands seems implausible” because such norms do not seem to be part of the “furniture of the world.” If normative requirements are not part of the natural world, they might be thought to have a source internal to us in our ability to reflect on the extent to which we have reasons for action and belief. Perhaps, that is, “reflection is … to be credited with the creation of norms which bind us” (109), a view Kornblith takes to be endorsed and defended by Christine Korsgaard.

As he construes Korsgaard, her view is that “what it is for a belief or action to meet appropriate normative standards is nothing more nor less than that it survive reflective scrutiny” (110), a kind of scrutiny that involves, in her later words, “reasoning all the way back” to a “completely undeniable, satisfying and unconditional answer” to the normative question at issue. Again, such a role for reflection is held by Kornblith to produce an infinite regress as any level of scrutiny is, in principle, open to further reflective scrutiny and so there cannot be a completely undeniable and unconditional answer. Moreover, Kornblith suggests that we cannot simply identify proper normative standards with those accepted at the end of reflective inquiry as such reflective inquiry can, in principle, lead us awry.

While I am in complete sympathy with Kornblith's criticisms of Korsgaard's constructivist views,Footnote ⁹ the notion of reflection at issue in Korsgaard's project is not clearly Kornblith's, as he explicitly characterized it in Chapter One. It is true that, on Korsgaard's way of putting her project, it is motivated by one's noting that one has a belief or desire and such noting is surely the product of introspection. However, whether or not a mental state is a reason for action or belief seems not determinable by introspection. If so, then the use of “reflection” in much of Korsgaard's discussion is perhaps misleading.

I worry, as well, that Kornblith's own grasp on the target of his criticism is here again shown to be less firm than one might desire. Noting that empirical inquiry seems prima facie unlikely to produce justification for believing normative truths, he considers the view that we need to “turn away from empirical methods and engage, instead, in reflection” and suggests that this is the thought behind “the idea that normative claims can be known a priori” (120). However, it is implausible that the a priori should be held to include the introspective and utterly implausible that the notion of the a priori encompasses no more than the introspective. It seems here that Kornblith is, contrary to his earlier explicit stipulation, here deploying the thicker notion of reflection on which it includes rational intuition. Additionally, in a particularly puzzling passage concerned with Gilbert Harman's (Reference Harman1977) well-known concerns about thought experiments in moral theory (120–121), Kornblith suggests that the judgment one arrives at when considering a thought experiment regarding some hypothetical scenario is the product of reflection. For reasons noted earlier, this seems clearly incorrect, if “reflection” has the narrow meaning of introspection. While one might introspect that one has the intuition that it is impermissible to involuntarily harvest the organs of a healthy person to save five others whose lives are threatened, the intuition is not the product of introspection, having as its content a claim about the moral status of a particular sort of action rather than a claim about one's own mental states. These puzzling passages make little sense on Kornblith's proprietary sense of “reflection” and seem to be motivated by a more inclusive sense on which reflection includes intuition.

In his final chapter, Kornblith turns to an experimentally informed attempt to provide a realistic view of the powers of reflection. This realistic account of reflection views it through the lens of the increasingly popular “dual-process” account of cognition. The dual-process account posits a division between two different systems of cognition. System 1 processes are typically “fast, parallel, unavailable to introspection, automatic, found in non-human species” and low in variation between individuals. System 2 processes are typically “slow, serial, present to consciousness, subject to control,” found only in humans and higher in interpersonal variation. Kornblith is quite clear that we cannot “identify the processes of System 2 with the kind of reflection which is the subject matter of” his book but he holds that it is nonetheless “clearly relevant” to the issues addressed in his first four chapters.

Kornblith's central claim is that our mistaken tendency to “see first-order processes in mechanistic terms, while viewing the process of reflection as somehow different” is enabled by a tendency to view “first-order mental goings-on in mechanistic terms – to view them from third-person point of view – but to view the processes involved in reflection, instead from the first-person perspective” (154). This, in turn, gives rise to the tendency, repeatedly noted in earlier chapters, to regard first-order states as the product of brute, impersonal mechanistic causes and to the felt need for something more in order to yield epistemic responsibility and justification, real reasoning, autonomy, freedom, and genuine reasons. The central suggestion of the chapter – that the tendency to think reflection is suitable to tasks which non-reflective processes are not is the product of a perspective shift – however, seems untethered to the empirical details of dual-process accounts of cognition. The primary connection is that the perspective shifting is enabled by the fact that our first-person perspective yields an extremely limited grasp of the System 1 processes undergirding and interacting with our System 2 processes.

In sum, Kornblith is, in my judgment, unquestionably successful in showing that there is, in his memorable phrase, “no second-order magic” and that, contrary to a large number of influential philosophers, no serious philosophical problem for first-order belief and desire can be solved merely by coupling them with higher-order beliefs and desires. Moreover, in spite of the reservations that will likely remain for specialists in the various areas he discusses, his discussion does significantly illuminate their interconnections. He is, I have suggested, less successful in arguing that introspection is not philosophically significant, especially for epistemology. Moreover, in spite of some equivocation in Kornblith's writing as to what he intends as his target, reflection, in the broad sense which philosophers value most, includes rational intuition and the value of that faculty remains untouched by Kornblith's arguments.Footnote ¹⁰