Jonathan Haidt's Moral Foundations Theory is an influential scientific account of morality incorporating psychological, developmental, and evolutionary perspectives. The theory proposes that morality is built upon five innate “foundations,” each of which is believed to have been selected for during human evolution and, subsequently, tuned-up by learning during development. We argue here that although some general elements of Haidt's theory are plausible, many other important aspects of his account are seriously flawed. First, innateness and modularity figure centrally in Haidt's account, but terminological and conceptual problems foster confusion and ambiguities. Second, both the theory's proposed number of moral foundations and its taxonomy of the moral domain appear contrived, ignoring equally good candidate foundations and the possibility of substantial intergroup differences in the foundations' contents. Third, the mechanisms (viz., modules) and categorical distinctions (viz., between foundations) proposed by the theory are not consilient with discoveries in contemporary neuroscience concerning the organization, functioning, and development of the brain. In light of these difficulties, we suggest that Haidt's theory is inadequate as a scientific account of morality. Nevertheless, the theory's weaknesses are instructive, and hence, criticism may be useful to psychologists, neuroscientists, and philosophers attempting to advance theories of morality, as well as to researchers wishing to invoke concepts such as innateness and modularity more generally.
Morality permeates human existence, playing a role in a vast array of choices and evaluations, both public and private, momentous and trifling. Given morality's central place in human social life, it is important for scientists to address its origin and features. Although the scientific study of morality is still in its infancy, research is underway on topics including the role of emotion in moral judgment (e.g., Greene, Nystrom, Engell, Darley, & Cohen, 2004; Greene, Sommerville, Nystrom, Darley, & Cohen, 2001; Haidt, 2001), the neural bases of punitive behavior (e.g., de Quervain et al., 2004) and cooperative behavior (e.g., Rilling, Sanfey, Aronson, Nystrom, & Cohen, 2004; Rilling et al., 2002), and the evolutionary roots of morality (e.g., Bowles, 2008; Casebeer, 2003; Gintis, Bowles, Boyd, & Fehr, 2003).1
Still needed, however, is a comprehensive explanatory framework within which to situate these findings and numerous others concerning phenomena such as moral development and inter- and intragroup moral differences. Does such a unifying framework currently exist? Jonathan Haidt thinks so. He, along with collaborators including Craig Joseph, Jesse Graham, and Brian Nosek, has proposed a theory, the Moral Foundations Theory (MFT), according to which morality rests upon five innate “foundations” (Haidt & Joseph, 2007).2 MFT's influence extends beyond academia, having received favorable coverage in popular media outlets (e.g., Brooks, 2009; Shermer, 2009; Wade, 2007).
Although some aspects of MFT are likely broadly correct, our analysis shows that most of its defining features are seriously flawed. First, the terminological and conceptual apparatus on which MFT relies—particularly as it relates to innateness and modularity—is rife with confusion and ambiguities that leave it either highly implausible or explanatorily toothless (see Terminological & Conceptual Confusions section). Second, the theory's taxonomy of the moral domain appears contrived, sidelining equally good candidate foundations and neglecting substantial intergroup differences in the foundations' contents (see Difficulties with the Five-Foundations Taxonomy section). Third, the mechanisms (viz. informationally encapsulated, domain specific modules) appealed to by MFT do not comport well with discoveries in contemporary neuroscience concerning the organization, functioning, and development of the brain (see Neurobiology and MFT section). In light of these difficulties, we suggest that Haidt's account is currently inadequate as a scientific theory of morality.
AN OVERVIEW OF MFT
MFT's essential idea is that morality is made up of five3 sets of foundations, each set corresponding to an adaptive challenge faced by early humans and their nonhuman primate ancestors. The five foundations, followed by their respective adaptive challenges, are:
harm/care—protect and care for young, vulnerable, or injured kin
fairness/reciprocity—reap benefits of dyadic cooperation with nonkin
ingroup/loyalty—reap benefits of group cooperation
authority/respect—negotiate hierarchy, defer selectively
purity/sanctity—avoid microbes and parasites4
In what sense, specifically, might humans be “prepared” to acquire and deploy moralities corresponding to the five foundations? Here, Haidt invokes the notion of mental modularity, originally articulated by Fodor (1983). In line with its conception of innateness, MFT's modules are taken to be dependent on environmental, and not just genetic, factors. Moreover, the definition of modularity is a loosely inclusive one; drawing on the work of Sperber (1994, 2005), Haidt says that a putative module needs only to be “modular to some interesting degree” (Haidt & Joseph, 2007, p. 380, quoting Sperber, 1994 5) to be counted as such.
More specifically, the modules that “prepare” humans to acquire certain moral concerns are what Haidt, again following Sperber (2005), calls “learning modules.” These correspond to each of the five foundations and facilitate the learning of specific responses to patterns of events in the social world.6 Continuing his utilization of Sperber's “teeming modularity” framework, Haidt contends that the outputs of these learning modules are, themselves, a plethora of modules, each of which is dedicated to producing some situation-specific moral intuition falling within one of the five foundations (Haidt & Joseph, 2007). The modules so produced are “like little bits of input–output programming connecting the perception of a pattern in the social world (often a virtue or vice) to an evaluation and in many cases a specific moral emotion (e.g., anger, contempt, admiration)” (Haidt & Joseph, 2007, pp. 379–380).7 (We will henceforth call these second-order modules as a convenient way of distinguishing them from learning modules.)
Second-order modules produce moral intuitions—flashes of approval, disapproval, or other emotions upon detecting some pattern in the social world (Haidt & Joseph, 2007, p. 379). These intuitions, in turn, are what actually drive our moral judgments, beliefs, actions, and the like. This is an important point of divergence between MFT and views of morality historically accepted in philosophy (e.g., Kant, 1785/1998) and psychology (e.g., Kohlberg, 1981, 1984), which have tended to focus on reason as the sole driver of moral judgment.8 On Haidt's view, by contrast, reasoning plays a mostly post hoc role, coming into play only after (affective) intuitions produce a moral judgment and serving to justify that judgment to others (see also Haidt, 2001).
A second way in which MFT differs from many other accounts of morality is in the breadth of the moral domain that it posits. Certain other theories (e.g., Turiel, 1983) seem to restrict morality to the harm/care and fairness/reciprocity foundations. This, Haidt notes, accounts reasonably well for the moralities of “modern” (i.e., western, secular, liberal) cultures. However, Haidt and Graham (2009) argue, on the basis of cross-cultural data (e.g., Shweder, Much, Mahapatra, & Park, 1997; Haidt, Koller, & Dias, 1993), that these theories give a very incomplete picture of morality as manifested in “traditional” cultures—a category which includes many nonwestern cultures as well as western religious conservatives. The moralities of these groups do include concern for the first two foundations, but they also include a high degree of concern for the other three—ingroup/loyalty, authority/respect, and purity/sanctity.
This approach, Haidt suggests, provides an elegant developmental explanation of moral difference: Although all people are “prepared” to develop concern for all five foundations (i.e., they are genetically endowed with learning modules for all five foundations), the environment in which they are raised may result in second-order modules developing for only some of these domains (at least to any substantial degree). Thus, a person raised by liberal parents may develop second-order modules concerned principally with harm/care and fairness/reciprocity, whereas her counterpart raised by conservative parents may develop second-order modules concerned with all five foundations to a similar (high) degree. In this way, MFT purports to be able to explain both moral similarity and difference between individuals and cultures.
We have aimed to capture the essence of Haidt's theory as best we can, although of course, many details have been glossed over for reasons of brevity. Despite our criticisms of many pillars of MFT in what follows, we wish to pause briefly to acknowledge a number of its genuine strengths. One strength is MFT's recognition of emotion and intuition as important contributors to moral behavior and judgment, an aspect of the theory that is consistent with a long tradition stretching from the 18th-century philosophy of David Hume (1739/1740/2008; for discussion, see Blackburn, 2008) to present-day cognitive science (see note 7). Another is the theory's rejection of the common philosophical (and, sometimes, psychological) project of searching for a single, unified principle or psychological mechanism that, by itself, accounts for all of morality, a project that has thus far proved something of a quagmire.9 A further strength is MFT's emphasis on the need to examine cross-cultural data to see whether the moral domain has a different composition outside of the western cultures in which most scientific research on morality has been conducted. Finally, we commend Haidt's recognition of the need to provide a developmental and evolutionary account of morality, a project that is too often overlooked by researchers focusing on the proximate mechanisms of morality as manifested in adults.10
Despite these strengths, we believe that crucial aspects of MFT are implausible or, at a minimum, in need of substantial rethinking. These difficulties with the theory are the subject of the remainder of this article.
TERMINOLOGICAL AND CONCEPTUAL CONFUSIONS
Innateness is a notoriously treacherous concept that is used in many different ways by many different researchers.11 Recent advances in embryology and molecular biology have rendered dubious any neat division of traits into those that are innate and those that are acquired through experience by making it clear that behavior, in general, is the outcome of interactions between genes, brain, and behavior/experience (Flint, Greenspan, & Kendler, 2010; Richerson & Boyd, 2005; Greenspan, 2001). Even so, as Richerson and Boyd (2005, p. 9) point out, one can reject the simplistic nature–nurture dichotomy but still acknowledge that “different traits do vary in how sensitive they are to environmental influences.” Hence, to avoid mere hand-waving, innateness claims have to provide evidence that the traits they target tend to the “insensitive-to-environmental-influences” end of the spectrum, and, for adaptationist accounts, that these traits were selected for in the course of human evolution. Thus, they have to tread a fine line: On the one hand, too strong a notion of innateness is likely not to be applicable to cognitive and behavioral traits of any complexity, morality included, because so much learning is involved. On the other, too weak a notion may apply to far too many cognitive and behavioral traits to be useful for explanation, prediction, and categorization.
Although Haidt generally seems to recognize the traps in the unduly strong version, he is less sensitive to those at the weak end of the spectrum. The result is that his version of nativism excludes very little, being so loose that it applies to nearly any trait displayed in human behavior. This promiscuous applicability compromises its power to categorize traits or to generate detailed predictions about their evolutionary and developmental origins, or cognitive mechanisms. One of the few clear predictions generated by the theory is that some moral norms will be more easily learned than others; this prediction is central to Haidt's definition of innateness as “preparation for” or “organiz[ation] in advance of experience.” Such definitions, however, risk triviality due to the absence of details concerning what, precisely, this evolved preparedness or pre-experiential organization amounts to. Some things most humans can easily learn are not plausibly innate in any meaningful sense, such as how to ride a bicycle, tie a reef knot, use a spoon, paddle a boat, or use a phone. Nor is the ease or speed of acquisition always a reliable marker of innateness in Haidt's sense. Children, for example, can frequently learn to open drawers and untie shoelaces before they can walk, something the neurobiological evidence shows they are neurally prepared to do.12
This problem could, to some extent, be avoided if Haidt supplied some specifics of how—at the level of cognitive psychology, developmental psychology, neuroscience, and so on—humans are “prepared for” the acquisition of some moral norms but not others. Unfortunately, he provides nothing of the sort. Consequently, dubbing a given trait “prepared for” amounts to little more than a restatement of the behavioral data, providing little basis for nontrivial empirical evaluation of the claim.
Here, Haidt might reasonably protest that he does provide a more specific account of the sort of preparedness at work: learning modules. If there is a learning module for traits or abilities falling within some domain, these traits or abilities are innate (in the sense of being organized/prepared for). Without such a module, the trait or ability is not innate. But how do we know when a trait emerges from a learning module and when it does not?
As with innateness, MFT's invocation of the concept of modularity is murky, raising the question of whether it is truly substantive, or merely conveniently vague and context-morphable. Had Haidt made a strong modularity claim along the lines of those made by certain evolutionary psychologists (e.g., Cosmides & Tooby, 1994, 1997), it would at least qualify as substantive (in some minimal sense). Regrettably, however, it also would almost certainly be incorrect, as such claims are neurobiologically very dubious, however tempting they may seem at a psychological level.13 Alternatively, if a weaker notion of modularity is used, there arises the question of whether the notion actually explains the phenomenon in some deeper, more informative way, or whether it is merely a vague, “black-box” restatement of the behavioral data, lacking computational, neurobiological, or other details. This latter difficulty is the principal one afflicting MFT.
What, precisely, are the criteria Haidt associates with modularity? So far, as we can tell, there are two. The first is domain specificity. In the case of learning modules, the domains in question are the five foundations. The modules respond to and facilitate learning of moral norms falling within these foundational domains. In the case of second-order modules, they are particular patterns in the social world (Haidt & Joseph, 2007, pp. 379–380), such as seeing a fellow shopper cut in line or a hunter club a seal. The second criterion is informational encapsulation. This criterion, unlike the first, is not made entirely explicit in some of Haidt's published work, but Haidt (personal communication, May 5, 2010) explains that it was, in fact, what first drew him to modularity theories. The link was his earlier work on the phenomenon of “moral dumbfounding,” in which individuals cling to their intuitive moral judgments despite being unable adequately to justify them. The individuals, moreover, may continue to endorse their initial judgments even when shown that the justification they attempted to offer (e.g., saying that there is some harm involved) is explicitly ruled out by the terms of the scenario they are judging (see Haidt, 2001; Haidt et al., 1993). For Haidt, this perseveration of certain moral judgments, despite new or confounding information, implied that the processes producing the judgments are informationally encapsulated (to some degree) and, therefore, that the judgments are the product of mental modules.
We suggest that these criteria are inadequate to ground an informative, explanatorily useful notion of modularity. Beginning with Haidt's version of domain specificity, the problem, as we analyze it, is that the modules of MFT appear to be domain-specific only in the vacuous sense that whatever set of input–output (i.e., stimulus–behavior) patterns happens to be manifested is said to be the result of modular processing and to constitute the domain to which that module is dedicated. This, of course, robs the view of any distinguishing power because any cognitive process or trait qualifies as domain-specific on this trivial description. This is particularly—but not solely—a problem for second-order modules, which seem to be little more than a way of designating somewhat arbitrarily chosen (and, perhaps, arbitrarily fine-grained) stimulus–behavior patterns without shedding any light on the underlying processes' computational workings or other features. Second-order modules are thus examples of “black-box” psychology par excellence, giving the illusion of explanation and understanding but, in fact, providing very little of either. Conceivably, neurobiological data might shore up the idea, but as we show below (Neurobiology and MFT section), this seems unlikely.
Learning modules face similar difficulties. Although Haidt is undoubtedly correct that certain sorts of moral norms are more easily learned than others, the inference to domain-specific learning modules for these easily learned norms is completely unsupported by the available evidence. The mere commonness of moral norms corresponding to the five foundations and the ease with which some adaptationist account can be concocted for this commonness do not sanction the move to something as concrete as domain-specific, genetically endowed learning modules.14 Myriad other psychological explanations, with far greater convergent support from neurobiology and other fields, could account for the phenomena. Broadly speaking, these explanations involve the combined effect of numerous nondedicated processes and their interactions, a possibility we examine in greater detail in the Neurobiology and Domain Specificity section below.
How does informational encapsulation fare? Lack of evidence and lurking triviality are again the principal difficulties. Triviality comes in when Haidt declines the clear-cut type of informational encapsulation of other theorists (e.g., Fodor, 1983), preferring instead “partial” or “to-some-degree” encapsulation. Unfortunately, nearly any higher cognitive process can lay as much claim to such “partial encapsulation” as moral dumbfounding, the phenomenon Haidt takes to be paradigmatic of encapsulation in the moral domain. Psychological research—to say nothing of everyday experience—provides a host of examples of situations in which people are insensitive to or distort novel or potentially threatening information.15 A cuckolded husband may maintain the comforting belief that his wife is faithful even in the face of overwhelming evidence to the contrary. A loving mother may refuse to believe her wonderful son misbehaves in school. An experimental subject may persist in believing that she is adept at a certain task even after being told that the feedback on which she based this belief was entirely bogus (e.g., Ross, Lepper, & Hubbard, 1975). And, to take an example involving science, a researcher may cling to a preferred theory even after powerful arguments and evidence have accumulated against it. For example, Joseph Priestley, the brilliant English chemist, espoused to his death the theory that combustion and rusting involved release of phlogiston, rejecting the increasingly well-confirmed oxidation theory of Lavoisier. More recently, Fred Hoyle, despite mounting evidence against him, clung to the steady-state theory of the origins of the universe. Following Haidt, one might call this “scientific dumbfounding,” and propose an encapsulated module to explain it.
Perseveration in these cases does not, of course, necessarily mean that the beliefs are the product of modules, for there are probably motivational, affective, and other nonmodular factors that are more than sufficient to account for the phenomenon. Ironically for Haidt, an explanation invoking such factors actually counts against encapsulation, insofar as it suggests robust links between affective processes, on the one hand, and those involving factual beliefs, on the other.16
Setting aside the above difficulties concerning innateness and modularity, might Haidt's five foundations still be a productive way to understand morality? In the next section, we examine the five-foundations taxonomy and its rationale.
DIFFICULTIES WITH THE FIVE-FOUNDATIONS TAXONOMY
One shortcoming of the five-foundations taxonomy is that other basic moral values exhibited by humans across various cultures have as much—or as little—call to be included as do Haidt's favored five.17 Two strong contenders for this role are industry and modesty.18
Industry is highly moralized in many societies—including modern, secular western ones, and groups as varied as the Inuit, Scots, Japanese, Haida, Finns, and Dene, to name a few. It is manifested in a strong “work ethic,” a repudiation of laziness, and a disapproval of various other forms of shirking and free-riding. This value on industry seems to exist above and beyond any potentially harmful effects sloth might have on individual, community, or purity/sanctity considerations, and on top of any connections hard work may have to hierarchical roles (e.g., working because a superior directs one to, providing for lower-status individuals). Fairness/reciprocity seems to be the only current foundation into which one might try to slot industry. This, however, fails given that hard work seems to be valued and laziness scorned even when a lack of industry would not compromise one's ability to meet obligations incurred in the course of cooperative or reciprocal interactions with other individuals.
As for modesty, many societies have subtle norms proscribing overtly calling attention to one's achievements, status, wealth, and so forth. Bragging behavior is frowned upon except in a limited range of situations. For example, in western societies such as the United States, a child may boast to her parents about some achievement in school or on a sports team, but doing the same around her friends would likely result in her being shunned and labeled a braggart. As with industry, finding a niche for modesty in the five current foundations is difficult. Authority/respect is the only one into which modesty might plausibly be slotted, but at present, this foundation deals rather specifically with behaviors related to hierarchy. Although behaviors related to modesty/immodesty could be described, at least in part, in terms of their role in helping one navigate the social hierarchy, modesty seems, like industry, to be valued even in situations where such hierarchical concerns are not present.
Content of the Foundations
The converse of omitting some basic moral values is plugging in too many, and here too, Haidt's taxonomy looks worryingly ad hoc rather than principled. For example, the ingroup/loyalty and purity/sanctity foundations may, for all we can be sure, merely be extensions of harm concerns to entities other than individual persons—for instance, to a supraindividual entity, the community, the welfare of which is taken to be more than the mere sum of its individual members' welfares. Purity/sanctity, likewise, may concern perceived harms to supernatural entities not acknowledged by secular morality: deities, souls, and the like. Violations of purity/sanctity norms may also take the form of using something—one's body or a sacred site, for example—in a way that does not accord with the intentions/wishes of deities, thereby harming and defiling it. Placating the gods or spirits or what have you to prevent natural disasters and diseases, and blaming the failure to do this when these or other calamities occur, is entirely common, suggesting a further way in which purity/sanctity may be linked to harm.
Also infected by the ad hoc problem is the list of “characteristic emotions” paired with each foundation. For example, Haidt lists anger as a characteristic emotion of the fairness/reciprocity foundation (see, e.g., Haidt & Joseph, 2007, p. 382, Table 19.1). Yet anger can just as naturally be deployed in the service of many of the other foundations, and perhaps all of them. Most obviously, it is a crucial and ubiquitous aspect of defensive responses of the sort associated with the harm/care foundation, although Haidt does not list it as such. For example, females of many mammalian species, humans included, become impressively hostile toward predators and conspecifics who threaten their young. Anger is also routinely triggered by events associated with the other putative foundations, such as a failure of conspecifics to show proper deference or being “dissed” by a peer or subordinate (authority/respect), malicious actions that undermine the group (ingroup/loyalty),19 or the flouting of religious norms (purity/sanctity), not to mention various allegedly nonfoundational violations, such as bragging, lying, and shirking. Other “characteristic” emotions, such as guilt and disgust, likewise seem to have substantially broader moral expression than their “characteristic” association with particular foundations claimed by MFT would imply.
These criticisms point to yet another murky issue: What counts as an instance of a behavior falling within a given foundational category? This difficulty is particularly acute for the purity/sanctity foundation. Even within the rather coarse category of “traditional” societies, there is huge variability in religious practices, with many religions not invoking a Deity or a set of gods as such, but rather assorted spirits, ancestral ghosts, or merely beliefs in the efficacy of sacrifices and other rituals. This diversity of content is important because it bears on Haidt's claim that a signal advantage of MFT is that it allows one to understand differences in the moral profiles of “traditional” and “modern” groups in terms of the differential development of the five foundations. Fair enough, MFT does, at first, appear to provide a natural explanation for such differences (see Haidt & Graham, 2009; Haidt & Joseph, 2004, 2007) and to have experimental evidence supporting this contention (see, e.g., Graham, Haidt, & Nosek, 2009). A closer look suggests, however, that this apparent achievement for MFT is, in fact, no such thing, and its seeming otherwise is simply due to the fact that Haidt has gerrymandered the content of the three “traditional” foundations (ingroup/loyalty, authority/respect, and purity/sanctity) in such a way as to make this outcome inevitable. Our contention is this: By and large, liberals also have significant moral concerns that arguably fall within these three categories; the contents or targets of these concerns just happen not to be the same as those of conservatives.
As an illustration, consider again the purity/sanctity foundation. Haidt argues that the adaptive challenge underlying this foundation—and its “characteristic” emotion, disgust—is the need to avoid microbes and parasites in one's food. This system for evaluating and rejecting potentially illness-causing food, he suggests, was then exapted for “social evaluation and rejection” (Haidt & Joseph, 2007, p. 384). Although one may quibble with the details, this is a reasonable enough hypothesis about the origin of disgust's role in morality. However, when Haidt specifies the sort of contemporary moral concerns associated with the purity/sanctity system, he focuses on the “sanctity” aspect of the putative foundation, very narrowly interpreted. Consequently, his examples, and the resulting conception of the foundation, end up being confined to those that some religious individuals consider important, such as “keeping religious objects set apart from pollutants and profane objects [and] overcoming carnal desires and treating the body as a temple… [as well as] virtues such as chastity and temperance, and vices such as lust and intemperance” (Haidt & Joseph, 2007, p. 384).
This neglects the possibility that liberals also make substantial use of the purity/sanctity system. For example, at present, many liberals are extremely concerned about the environment. These concerns are often highly moralized, with members of groups such as Greenpeace and Earth First being every bit as sanctimonious about damage to old growth forests, the clearing of tropical rainforests for palm plantations, and the pollution of watersheds as conservatives allegedly are about issues such as promiscuity, homosexuality, and the defilement of religious objects. Their concerns about these issues may be voiced through use of the language of purity, defilement, holiness, and sanctity that is every bit as sincere as that of religious conservatives. To claim that their behavior is not “true” purity/sanctity behavior simply because they do not believe in a Personal Deity would require some fancy, and almost certainly ad hoc, maneuvering.
Similar points can be made about liberals who care deeply about “animal rights.” They may refuse to wear fur or leather, not consume meat or even any animal products at all, or eat only organic food—and feel disgust toward people and companies that do consume or produce these things. Activists may also accuse those who do not share their views of wanton immorality through public protests and campaigns. Just as certain conservative groups may rally at a state capitol to protest the legalization of same-sex marriage, members of People for the Ethical Treatment of Animals may gather outside a dog show dressed in Ku Klux Klan regalia to decry the show's supposed efforts to create a “master race” of purebred dogs (see Farris, 2009).
In light of this, it seems likely that if subjects in studies such as that of Graham et al. (2009) were given questions pertaining to a broader range of purity/sanctity concerns—including not only things like chastity and sin but also respecting the environment and animal rights—the gap between conservatives' and liberals' concern with this foundation might well disappear entirely.20
Gradations between “Modern” and “Traditional”
Haidt often touts his theory's seeming ability to capture a contrast between the moralities of “modern” (or “liberal”) and “traditional” (or “conservative”) groups. He claims that the former's morality consists principally of the harm/care and fairness/reciprocity foundations, whereas the latter's involves substantial development of these two foundations plus the three others, ingroup/loyalty, authority/respect, and purity/sanctity. This supposed discovery has been excitedly picked up on by the popular press (e.g., Shermer, 2009).
We are more skeptical, however, as examples abound of individuals and groups that do not fit neatly into the categories of “modern” or “traditional.”21 Many twentieth- and twenty-first-century Swedes, Danes, Germans, and French, for example, have not only the expected “modern” moral concerns but also concerns for social welfare and collective/national identity that the “modern” foundations alone seem insufficient to account for. And what of individuals who, in the United States, identify themselves as “libertarians” or as “socially liberal and economically conservative”? For libertarians, harm/care and fairness/reciprocity values are evident, but given the differences between the moral judgments, voting patterns, and so forth of these individuals and American liberals—whom Haidt also takes to have predominantly harm/care and fairness/reciprocity concerns—there must be other factors that set the two groups' moralities apart.
Can MFT account for the existence of these groups in a systematic, evidentially powerful way? First appearances are promising, as MFT's approach of positing separate learning modules for each of the five foundations seems to provide a convenient way of accounting for the possibility of such independent development.22
We suggest, however, that the moral values of groups that are not paradigmatically “modern” or “traditional” are not so smoothly explained by MFT. It seems a stretch, for instance, to reduce a western European's concern for social welfare to the development of the ingroup/loyalty foundation in the same way that Haidt (e.g., Haidt & Joseph, 2007, p. 383; Graham et al., 2009) sometimes suggests that an American conservative's nationalism might be so reduced. Indeed, after the horrors of two world wars, the nations of western Europe are extremely wary of anything that smacks of nationalism or militarism. Similarly, if one accepts Haidt's contention that American liberals have only very limited development of the three “traditional” foundations, it is not clear how their foundational repertoire differs from that of libertarians. Nor are the differences between these groups explicable in terms of the relative development of the two “modern” foundations; both seem to have very high concern for both harm/care and fairness/reciprocity.
These difficulties call attention to a methodological flaw hitherto unremarked in our discussion: the limitations of the dataset Haidt has relied upon in constructing his theory. He has, in effect, selected two points on a broad and varied moral landscape—“modern”/“liberal” and “traditional”/“conservative” morality—and tailored his theory to account for them. If he had selected two (or more) different points on which to base his theory—for instance, western Europeans and American libertarians—the set of foundations that he came up with might have been quite different. At a minimum, the contents associated with each foundation may have differed substantially.23 Given this, MFT may be open to the same charge of myopia that Haidt has leveled at other researchers' views (see, e.g., Haidt & Graham, 2009; Haidt & Joseph, 2007, Section 2).
NEUROBIOLOGY AND MFT
Why Neurobiology Matters
Progress in neurobiology, developmental psychology, and genetics in recent decades means that innateness hypotheses are now expected to be supported by, or at least consilient with, evidence from these fields. Long gone are the days when conjectures about universality and selective advantages would suffice. Supporting evidence is manifestly required for strong forms of nativism that take complex cognitive traits to be completely genetically specified, requiring little environmental input for their proper development. However, evidence of this sort is also needed for weaker forms of nativism such as Haidt's “preparedness,” all the more because such data could firm up awkward vagueness in behavioral criteria.
Our point is not that anyone utilizing nativist concepts in their accounts of morality must also commit themselves to an identification of moral kinds with neurobiological kinds. Like Haidt, we are skeptical that any such neat identity relationship will be found. But acknowledging this inevitable messiness does not absolve researchers invoking innateness from the responsibility of having a theory that is consilient with what is known in neurobiology. To belabor the obvious, if a theory posits certain phenomena as owed to evolution by natural selection, this implies that the phenomena have at least some basis in our evolved biology. And in the case of cognition, behavior, and learning, the primary biological locus of these functions is the brain.24 Thus, if there are Haidt-type mental modules, we should expect the organization of the brain to reflect or, at an absolute minimum, be consistent with such modularity.
Neurobiology and Informational Encapsulation
Let us start with informational encapsulation. In the mammalian brain, the neuroanatomical rule is “loopy” (re-entrant, recurrent, etc.) architecture, not informationally encapsulated, feedforward modules (Logothetis, 2008). The cortex also has a “small world” architecture, meaning that local connections are dense and long-range connections are sparse, but everything is easily accessible to everything else in a few synaptic steps (Bullmore & Sporns, 2009; Buzsáki, 2006). This pattern is typical even in primary visual cortex (V1), the first area of cortex to receive inputs from the retina (via the lateral geniculate nucleus, or LGN), where more than 80% of the synaptic contacts on V1 neurons come not from the LGN, but from elsewhere, in particular, from higher levels in the visual system, other regions of the thalamus, and so forth. Additionally, every area of the cortex is “loopily” connected to the thalamus (Sherman, 2005). One region of the thalamus—the intralaminar nuclei—projects reciprocally to every part of the cortex, save V1. These data call into question the neuroanatomical plausibility of informational encapsulation.
Physiologically, informational encapsulation is undermined by the fact that cortical tissue regularly, not rarely, exhibits spontaneous activity, that is, measured neuronal spiking not occasioned by external stimuli or attention to a task. As Ringach (2009) points out, experiments using multielectrode techniques, as well as those using voltage-sensitive dyes, have shown that spontaneous activity of single cells is structured in space (coherent across millimeters of cortical tissue) and at various time scales. This activity can be related to hallucinations during sensory deprivation tests and to subsequent task performance. Super, van der Togt, Spekreijse, and Lamme (2003) showed that higher levels of spontaneous activity in V1 prior to presentation of a difficult task increased the monkey's successful performance of the task, a result which challenges the traditional hunch that spontaneous activity of single cells is just “noise.” Imaging techniques have revealed increases in the level of activity seen in the so-called default network during the nontask state (e.g., Buckner, Andrews-Hanna, & Schacter, 2008; Gusnard & Raichle, 2001). The fact that nontrivial amounts of energy are used by spontaneous activity of this sort provides further reason to think that it is not just neuronal “idling” or noise.
Even those processes in so-called early perception, advertised by Fodor (1983) as the “obvious” best candidates for modularity, fail to satisfy the requirement of informational encapsulation. Cells as early in visual processing as the LGN—not even yet in cortex—are sensitive not only to visual input but also to task demands and motor planning (Casagrande, Sáry, Royal, & Ruiz, 2005). This is owed to input from other regions, including the brainstem, that feeds into the LGN. At the single-cell level, Paradiso, MacEvoy, Huang, and Blau (2005) showed that there is extra-receptive field modulatory input to cells in V1. Niell and Stryker (2010), in a landmark study, demonstrated that single neurons in V1 double their firing rate in response to the very same stimulus depending on whether the motor system is engaged or not (i.e., on whether the animal is walking in a wheel). A similar result was found in Drosophila: Neural responses to precisely the same visual stimulus vary as a function of whether the fly is flying or not (Maimon, Straw, & Dickinson, 2010). These data challenge the idea that cells processing early visual information are shielded from activity in distant parts of the brain, suggesting instead that the processes of “early” perception are quite informationally porous, albeit in ways not clearly understood.25
Experiments on humans using neuroimaging show that visual processing areas (including V1) can be recruited for purposes of decoding language (Sadato et al., 1996), as well as for problem-solving tasks and when engaging in mental simulation (Kosslyn & Thompson, 2003; Kosslyn, Ganis, & Thompson, 2001; Mellet et al., 2000), violating not only the requirement of informational encapsulation but also that of domain specificity. Recent brain imaging studies in blind subjects have provided additional support to these results, demonstrating visual cortex activation during semantic tasks (Burton, Diamond, & McDermott, 2003; Noppeney, Friston, & Price, 2003), speech processing (Röder, Stock, Bien, Neville, & Rösler, 2002), and Braille reading (Sadato et al., 1996). Back-projections in the sensory processing pathways, as well as bidirectional intracortical and cortico-subcortical connections more generally, further undermine the case for informational encapsulation (for discussion, see Callaway, 2005; Churchland, Ramachandran, & Sejnowski, 1994). Given the routine violation of informational encapsulation even in simple, low-level perceptual cases, claims for the modularity of higher cognitive processes—including morality—are slimmer still.
Neurobiology and Domain Specificity
Language was the great hope for a complex cognitive ability sustained by a domain-specific, innate learning module (see, e.g., Pinker, 1994). Nonetheless, over the years, this hope has not garnered substantial support from the relevant neurobiology (Evans & Levinson, 2009; Panksepp & Panksepp, 2000), or from genetics. As for modular organization in the brain regions subserving the everyday function of language (as opposed to its acquisition)—what Haidt might call a second-order module (for language)—the case for a “grammar box” in Broca's area and a “semantic box” in Wernicke's area seems to have become weaker rather than stronger over time. Imaging studies of language comprehension and production, for example, show activity in a wide range of brain areas, including the right hemisphere and limbic structures, and careful lesion studies show results consistent with the imaging results (see Proverbio, Crotti, Zani, & Adorni, 2009; Prat, Keller, & Just, 2007; Dronkers, Wilkins, van Valin, Redfern, & Jaeger, 2004). Many questions do, of course, remain open concerning language learning and use, but at this stage, the hypothesis of a domain-specific, informationally encapsulated language module looks less promising than approaches investigating the various contributions of diverse regions.26
What about moral learning, in the form of MFT's learning modules and the second-order modules they produce? There is, of course, plenty of learning, of moral norms and much else; humans are prodigious learners. But what do we know about the brain that would suggest that any of this learning is subserved by domain-specific modules? Although it is well-known that there is regional specialization in the mature brain, the precise ways in which specialization and massive convergence work together is not well-understood (Meyer & Damasio, 2009; Logothetis, 2008). Furthermore, exactly how specialization emerges and the nature of the interacting roles of genes and experience-dependent plasticity are also still unclear. A leading hypothesis (Ringach, 2009; see also Dragoi, Turcu, & Sur, 2001 and Ben-Yishai, Bar-Or, & Sompolinsky, 1995) is that the statistics of the input, together with the general principle that neurons that “fire together wire together” (i.e., Hebbian learning), may account for quite a lot of observed regional specialization. This does not take us very far in the direction Haidt is going, however, as regional specialization is consistent with cross-talk and convergence at large spatial scales, and says nothing in support of modularity in the sense Haidt favors.
Likewise, small-world architecture suggests that some form of modularity may exist in densely connected micronetworks (Bullmore & Sporns, 2009; Sporns, Chialvo, Kaiser, & Hilgetag, 2004), but there is nothing to imply that modularity in this sense connects in any meaningful way with Haidt's domain-specific, informationally encapsulated moral modules. As noted above, the brain's small-world architecture actually weighs against the idea of encapsulation for complex functions owing, among other things, to the short synaptic distances between any two areas.
A further, and even more telling, point is this: Social skills, capacities, learning, and behavior can be very productively approached in terms of learning processes subserved by neurobiological systems that cut across a variety of domains (see Squire et al., 2008, Chapters 14–22). These circuits may give rise to certain patterns of representation and behavior in the social domain, but they are not modules in any sense that will give aid and comfort to Haidt; they are not dedicated to a single job, and they are not encapsulated from top–down, lateral, or subcortical input. Instead of modules, there seem to be a number of more general functions that are recruited in social learning and behavior, including pattern recognition, reward- and affect-based learning, attachment/bonding, fear, defensive behavior, impulse control, and planning (Churchland, 2011; Suhler & Churchland, forthcoming). For example, early damage to ventromedial prefrontal cortex can cause profound deficits in an individual's ability to acquire normal moral patterns of behavior (Anderson, Bechara, Damasio, Tranel, & Damasio, 1999). However, this does not mean that ventromedial prefrontal cortex is a module for moral learning (or for any other particular domain); rather, it is crucial to affect-based learning, motivation, planning, and choice across a variety of domains (see Damasio, 1994, 1996).
The availability of nonmodular neurobiological explanations for the range of abilities supposedly underwritten by second-order modules extends the challenge. Haidt characterizes these modules as “little bits of input–output programming” (Haidt & Joseph, 2007, p. 379). However, computationally speaking, having a dedicated bit of neural “programming” for each arbitrarily specific situation–response pattern would be a highly inefficient use of neurobiological resources, and it is probably not what happens in the brain. Consider the example of much-practiced skills, a category which plausibly includes moral judgment and decision-making (Casebeer & Churchland, 2003). Here, although neurobiological changes are observed in the network involved in the skill (e.g., increases in efficiency; Wu, Kansaku, & Hallett, 2004), the components of the network do not become specifically dedicated to the task in question, instead participating in networks involved in other tasks. The pervasiveness of such neural overlap and connectivity casts doubt on the hyperspecialized modules that Haidt appears to favor with his second-order modules.
Another example of a neurobiological system that cuts across domains—including the domains picked out by the five foundations—is the mammalian system for attachment and bonding, in which the neuropeptides oxytocin and vasopressin are known to play a crucial role (for reviews, see Carter, Grippo, Pournajafi-Nazarloo, Ruscio, & Porges, 2008; Donaldson & Young, 2008). This system is deeply involved in care for preferred conspecifics such as offspring and mates (Carter et al., 2008; Carter, 2003; Carter, DeVries, & Getz, 1995), but it also plays an important role in group-oriented behaviors such as defensive responses to threats (Carter et al., 2008; Landgraf & Neumann, 2004; Carter, 1998). As such, it plausibly cuts across the harm/care and ingroup/loyalty foundations. And as indicated by recent work by experimental economists on the effects of oxytocin on exchange behavior in humans (e.g., Zak, Stanton, & Ahmadi, 2007; Kosfeld, Heinrichs, Zak, Fischbacher, & Fehr, 2005; Zak, Kurzban, & Matzner, 2005), the oxytocin/vasopressin system may also make important contributions to behaviors associated with the fairness/reciprocity foundation.27 Moreover, there is now evidence associating variation in a gene for the receptor for oxytocin with differences in brain structure and responsivity in the amygdala and hypothalamus and thence to temperamental variations in sociability (Tost et al., 2010). The gene should not be characterized as a gene “for” sociability as it is only part of a gene–brain–behavior network (Kendler & Greenspan, 2006). Nevertheless, the research is an example of what needs to be done to begin to link social behavior to neurobiology and genetics (for further discussion, see Churchland, 2011). Finally, consider that, in mammals, the neural systems regulating negative affect, including pain, appear to have been modified to support separation distress, thereby providing a platform for social emotions that are important—but certainly not specific—to morality (Tucker, Luu, & Derryberry, 2005; Panksepp, 1998). One emotion of this sort is the empathic experience of another individual's pain (Tucker et al., 2005). But the neural systems for pain may also be important to the psychological pain experienced when one is cheated or contemplates cheating another individual (fairness/reciprocity), and perhaps even to negative social experiences associated with violations in all of the alleged five foundational domains (see, e.g., Eisenberger, Lieberman, & Williams, 2003). Interestingly, individuals exhibit greater empathic responses when observing harms to members of their own group than when observing identical harms to outgroup individuals (Xu, Zuo, Wang, & Han, 2009), and emotions experienced upon cheating or being cheated during an economic exchange exhibit a similar ingroup intensification effect (Burton-Chellew, Ross-Gillespie, & West, 2010). How these examples may be accounted for within the evolutionary, foundational, and modularity stories told by MFT is far from apparent.
A common theme running through these examples is that evolution, as well as brain development, tends to be very conservative, typically extending, reappropriating, or modifying extant neurobiological mechanisms rather than engineering wholly new, dedicated mechanisms (e.g., modules) for each new challenge that the environment throws an organism's way (see Flint et al., 2010). These nonmodular mechanisms do, of course, help to explain the concerns that the five foundations collectively capture, but they also suggest that claims of domain specificity or of strict divisions between the foundations are not honored by the neurobiology. And although neurobiology certainly does constrain and guide the development of morality as well as the various abilities and domains of concern that comprise it, this guidance appears neither to involve modules nor to amount to innateness in any explanatorily useful sense.
Advances in genetics have encouraged research on the links between genes and behavior, while at the same time warning ever more loudly that the linkages wind their way through dynamical, intertwining networks of gene–gene, gene–brain, and gene–environment interactions. If this complexity is daunting in the case of aggression in the fruit fly—and it is (Flint et al., 2010; Dierick & Greenspan, 2006)—then it is all the more so in the case of the social behavior of large-brained mammals. To make matters worse, the remarkable plasticity of the human nervous system must also be factored in; our brains did not evolve to read books or ride bicycles, but read and ride we do. Although the general idea that our evolved biology contributes to “our nature” is unsullied, the constraints on tenable hypotheses regarding specific links between genes, brains, and behavior have become vastly more demanding in the last decade. Claims to the effect that a given behavior is “innate,” “prepared for,” or “organized in advance of experience” are much more difficult to substantiate, now that we have a clearer idea what, evidentially, we are up against.
Haidt's MFT seeks to explain morality by connecting evolution, psychology, and development. Although we applaud the project's ambition, the overall execution is disappointingly insensitive to the height of the evidence bar. No detailed factual support from neuroscience, molecular biology, or evolutionary biology (save for very general adaptationist speculations) is marshaled for the theory's substantive claims, and although some speculations are consistent with what is known, others are not. Mere consistency, in any case, is a far cry from the confirmation or disconfirmation yielded in a tough test of a theory, a point long emphasized by philosophers of science (e.g., Popper, 1963).
We thank Lucia Foglia for insightful discussion of these issues. We also thank Jonathan Haidt and two anonymous referees for their helpful comments on an earlier draft.
Reprint requests should be sent to Christopher L. Suhler, Department of Philosophy, University of California, San Diego, 9500 Gilman Drive, La Jolla, CA 92093-0119, or via e-mail: email@example.com.
For a review of scientific research on morality, with a focus on neuroscience, see Suhler and Churchland (forthcoming).
For short, we shall henceforth say “Haidt” rather than “Haidt and colleagues,” but this should not be taken to diminish the contributions that these collaborators have made to the development of MFT. Craig Joseph, in particular, has been the coauthor of all of Haidt's works focusing on innateness and modularity, two concepts that will be central to our discussion in what follows.
Haidt and Joseph (2004) actually acknowledge only four “official” foundations while mentioning a strong candidate for a fifth foundation in a footnote (p. 63, note 15). That candidate—ingroup/loyalty—has, in subsequent works (e.g., Haidt & Graham, 2009; Haidt & Joseph, 2007), been elevated to the status of a full-fledged foundation.
This list is adapted from Haidt and Joseph (2007, p. 382, Table 19.1).
The quote Haidt and Joseph attribute to Sperber is, in fact, slightly inaccurate. Sperber, himself borrowing a phrase from Fodor (1983, p. 37), actually says “modular to some interesting extent” (1994, pp. 39, 48, emphasis added). We note this for the sake of accuracy, but of course, it does not affect the substance of the idea being appealed to.
This role for learning is a major way in which MFT incorporates constructivist as well as nativist elements. Our focus here will be predominantly on the nativist elements, for a number of reasons. First, the nativist portion of the theory is the one which has been most extensively developed, for instance, though arguments concerning selection for certain domains of concern and the positing of specific mechanisms such as modules. Second, it is the part of the theory that we believe to be most questionable in its details. Third, and relatedly, without an adequate understanding of what sorts of evolutionary and cognitive processes are constraining moral development, efforts to conduct developmental studies that flesh out the theory's constructivist elements are likely to founder upon a lack of specificity in the theoretical edifice underlying them. For example, without a clear understanding of what the purported “learning modules” are and what computational features they possess, the design and interpretation of developmental studies will be seriously underdetermined.
Notably, the modules that develop in this way need not always be concerned with social patterns, events, and so forth which directly correspond to the evolutionary challenges listed earlier. Consequently, one output of the learning module for fairness/reciprocity might be a “cutting-in-line detector” (Haidt & Joseph, 2007, p. 379), even though forming queues was not, in anything like its modern form, a part of the social environment in which the bulk of hominid evolution took place. In allowing for this possibility, Haidt and Joseph (2007, p. 381) invoke the distinction, common in the modularity literature (e.g., Sperber, 1994), between a module's proper domain and its actual domain. The proper domain is the set of triggers that natural selection shaped the module to respond to, whereas the actual domain is the set of all events that actually do trigger the module. For fairness/reciprocity, then, an example of a trigger falling within the module's proper (and actual) domain would be someone cheating in an exchange of goods or favors, whereas an example of a trigger falling in its actual (but not proper) domain would be someone cutting in front of you in line at the store. This distinction allows MFT to account for the fact that our current moral concerns extend well beyond situations that have direct counterparts in our evolutionary past.
We do note that in psychology and neuroscience—but not, generally speaking, in philosophy (although there have long been important dissenters—e.g., Hume, 1739/1740/2008)—researchers are moving away from this reasoning-focused picture as evidence accumulates that affect is central to moral judgment (for reviews, see Suhler & Churchland, forthcoming; Haidt, 2007; Moll, Zahn, de Oliveira-Souza, Krueger, & Grafman, 2005).
Although it has fallen out of favor among many contemporary moral philosophers, as well as many scientists studying morality, this feature of the theory, too, has a long history. It dates at least to ancient Greece, where Aristotle, in his Nicomachean Ethics (see the 2000 translation by Roger Crisp), devotes much attention to describing not only the virtues one must possess but also the sort of social and political environment in which the youth must be raised if they are to develop those virtues. For more recent examples of work that incorporates evolutionary as well as developmental elements, see Baumeister (2005), Hrdy (1999), and Ridley (1997).
Haidt professes to be sensitive to this, noting, in the course of describing the notion of innateness at work in his theory, the importance of being clear about how one is using this term (see Haidt & Joseph, 2007, pp. 373–374).
See also Richardson (2007) for a critical discussion of the concept of preparedness and the supposedly paradigmatic examples (e.g., fear of snakes and spiders) around which the concept has been built.
Perhaps the most famous example of this in social psychology is the vast body of research on cognitive dissonance initiated by Festinger's (1957) seminal work. It is worth nothing that mainstream psychological researchers studying cognitive dissonance and numerous other phenomena exhibiting some degree of (what Haidt would regard as) partial informational encapsulation have rarely, if ever, been tempted to make the move from this partial encapsulation to conclusions about modularity.
Note also that adding the requirement that some supposedly modular trait has been shaped by natural selection would do little to save the account of modularity on offer. Although certain evolutionary psychologists (e.g., Cosmides & Tooby, 1994, 1997) have tried to argue a priori that natural selection “must” lead to mental modules, there is little or no actual evidence that it really does lead to modular cognitive organization (see the Neurobiology and MFT section). As such, one cannot move from claims about adaptiveness to claims about modularity. Indeed, it is entirely possible to construct a scientific account of morality that sees an important role for evolution but makes no mention of modular processes (see Churchland, 2011; Suhler & Churchland, forthcoming).
Haidt is careful to leave open this possibility, at times saying that morality consists of “at least” five foundations (see, e.g., Haidt & Graham, 2009, p. 381).
We are not, to be clear, arguing that industry and modesty are foundational values, only that they have as good a claim to be foundations as those in Haidt's current list. Other concerns may have strong claims as well; patience and humor, for example, have figured in other lists.
Haidt and Joseph may tacitly acknowledge this when they list as a characteristic emotion of the ingroup/loyalty foundation “rage at traitors” (2007, p. 382, Table 19.1); the use of “rage” seems like little more than linguistic sleight of hand to avoiding mentioning anger as a characteristic emotion for two different foundations.
Similar points about content can be made about other foundations. For example, different groups may have radically different beliefs about what is “harmful/fair.” American libertarians, for instance, may equate fairness with whatever results from social and economic activity that is free, to the greatest extent possible, from regulation or other forms of government involvement, and regard departures from this as intolerably unfair and harmful. By contrast, American liberals, as well as many western Europeans, may see the libertarian's ideal of fairness as grossly unfair, instead endorsing a conception of fairness in which progressive redistribution of wealth and state involvement in certain services is a requirement of any fair social arrangement.
As in our discussion of additional foundations in the Additional Foundations section, we adopt the framework and language of MFT in this section merely as a device for highlighting phenomena we believe MFT's current foundational taxonomy does not adequately account for. As such, talk of (say) how areas of concern might be the product of one or another foundation should not be taken as an endorsement of MFT or of the particular foundations it posits.
In a recent paper, Haidt, Graham, and Joseph (2009) argue for this very conclusion. Using cluster analysis, Haidt and his coauthors find four principal moral subtypes characterized by different combinations of foundational concerns. In addition to the familiar “secular liberal” and “social conservative” subtypes, the analysis yields two other subtypes, which the researchers term “libertarian” and “the religious left.” As we describe in this section and the next, libertarians are one group we have concerns about MFT's present ability to account for, and so we look forward to seeing the results of Haidt et al.'s (2009) analysis further incorporated into the theory in their future work.
Even this broader survey of “modern” groups would leave out a good deal of complexity, as it neglects historical groups that do not fit neatly into the contemporary categories of “modern” and “traditional”. How, for example, would the moralities of the ancient Greeks or of members of the European Enlightenment be situated within a foundations-style framework?
Haidt's learning modules, in particular, cry out for a more detailed neurobiological explanation, because from what Haidt has said about them, they appear to be far less the products of experience than second-order modules. They seem, in other words, to be the clearest example in MFT of something constituting “organization in advance of experience.”
Incidentally, a dedicated face recognition area in the fusiform gyrus is perhaps a more probable candidate than language for modularity. Although Nancy Kanwisher (2010) has marshaled evidence from neuroimaging and lesions in support of this hypothesis, even here, the case remains controversial (see Hanson & Halchenko, 2008, as well as the collection by Hanson & Bunzl, 2010).