|The Original Sin of Cognition:
Race, Prejudice and Generalization*
To appear in The Journal of Philosophy
Penultimate Draft – Please check final version before citing
Long before we learn to talk, our expectations concerning novel members of a category are shaped by our experience with already encountered members. We expect, for example, that objects that share obvious perceptible qualities will also share dispositional properties. If a given item rattles when shaken, nine-month olds expect other items which share the same perceptible profile will rattle when shaken.1 By our first birthday, these inductive inferences are guided by language; we expect that even superficially dissimilar objects will share their hidden properties if they are identified by the same common noun; if, for example, each is introduced as ‘a blickett’.2 From the very beginning, we are inclined to generalize from experience with a given item to other items that we perceive as belonging to a common category.
There is, presumably, some innate cognitive mechanism that is responsible for these early inductive generalizations. In earlier papers, I argue that generics – sentences such as ‘ravens are black’ and ‘tigers are striped’ – express the generalizations that are delivered by this basic mechanism of generalization.3 If this is so, then generics provide us with a window onto the workings of this mechanism. In this paper, I am concerned with a particular aspect of this mechanism, namely the route by which we reach general conclusions regarding dangerous or harmful features. What follows in no way purports to be an exhaustive analysis of all the factors that have formed and sustained prejudiced attitudes. Rather, my aim is to identify and discuss one particular cognitive bias that has given birth to many a prejudice.
While I adopt a cognitive perspective here, this is not to imply that economic, political and cultural perspectives are not of equal value and importance. These various perspectives are not in competition with each other; rather they complement each other by providing different levels of explanation. Moreover, even within the domain of cognitive explanations of prejudice, I do not purport to offer anything close to a full psychological account of prejudiced attitudes. The focus is on a particular subset of negative stereotypes: ones that involve generalizing extreme and horrific behavior from a few individuals to a group, for example Muslims are terrorists or Blacks are rapists.
Two things, though obvious, are worth noting at the beginning. First, cognitive bias explanations do not excuse racial or cultural prejudice, any more than noting than we are hardwired to seek out and accumulate resources serves as an excuse for extreme covetousness or for theft. Secondly, offering a psychological explanation for prejudice does not entail that prejudice is inevitable. Quite the contrary – the closing sections of this paper will discuss some ways in which we might combat prejudice, with a particular focus on how we might prevent the formation of these attitudes in the course of childhood development. These suggestions are based on recent psychological research, and so – far from implying that prejudice is an inevitable feature of human psychology –the cognitive perspective on prejudice may point to some novel means of combating it.
Striking Property Generalizations
In previous work, I argue that a variety of philosophical, linguistic and psychological considerations suggest that generic sentences may be language’s way of letting us give voice to cognitively primitive generalizations.4 This hypothesis has subsequently received further support by new psychological data.5 There are now a variety of convergent reasons for supposing that the generalizations we articulate using generics reflect deep-seated aspects of our psychology.
In theorizing about generic generalizations it is helpful to identify various sub-classes of these generalizations, one of which I term ‘striking property’ generalizations. This class includes claims such as:
Mosquitoes carry the West Nile virus
Sharks attack bathers
Deer ticks carry Lime Disease
Pit-bulls maul children
Tigers eat people
These claims are intuitively true, even though very few members of the kind in question possess the predicated property. As it happens, less than one percent of mosquitoes carry the West Nile virus, and yet we are quick to assent to ‘mosquitoes carry the West Nile virus,’ even after learning this statistical fact. (Conversely, ‘mosquitoes don’t carry the West Nile virus’ remains patently false, even though 99% of mosquitoes don’t carry the virus.)
It may appear that these generics require for their truth only that some of the kind possess the property in question.6 This is not true for generics in general; for example, some cats are female, but ‘cats are female’ is false, and some (in fact, most) mosquitoes don’t carry the West Nile virus, but the corresponding generic is false. Such examples abound. I suggest that the generics above are special in that their predicates express properties that we have a strong interest in avoiding. 7 If even just a few members of a kind possess a property that is harmful or dangerous, then a generic that attributes that property to the kind is likely to be judged true.
Since we are working under the hypothesis that generics give voice to psychologically primitive generalizations, this observation implies that our basic way of dealing with dangerous or harmful information involves the rapid generalization of this information to the appropriate kind or category. We do not wait around to see what percentage of tigers eat people before drawing a general conclusion – even a single instance may be enough for us to conclude that tigers eat people. It is not hard to see the evolutionary benefits of such a disposition, since the costs of under-generalizing such information are potentially huge. Our ancestors were far better off jumping to conclusions, as it were, rather than taking the time to judiciously determine the precise likelihood of their being eaten.
The tendency to rapidly generalize such striking information manifests itself elsewhere in our thinking. Consider, for example, how many murders one much commit to be a murderer, versus how many times one must worry to be a worrier. The latter case requires one to worry with considerable regularity, whereas a single murder suffices to make one a murderer.
The disposition to generalize strikingly negative information on the basis of even a single event thus appears to be a pervasive aspect of our thinking. For generalizations concerning neutral or positive information, we require the instances or events to occur with a significant regularity; this is not so with negative information. There is a fundamental asymmetry between the impact of very negative information and the impact of neutral or more positive information on our intuitive generalizations.8
The ‘introduction’ conditions, as it were, of striking property generalizations – how the world must be for us to form or accept these generalizations – are very undemanding when it comes to how prevalent the property has to be in the relevant population. What, though, of the ‘elimination’ conditions of these generalizations – how does acceptance or rejection of these generalizations impact the inferences we are willing to draw?9 We are content to accept “ticks carry Lyme disease” despite knowing that very few ticks are actually infected with the disease. One might suppose that an ideally rational agent would be very hesitant to suppose that an arbitrary tick carries Lyme disease, in light of these statistical facts.
However, recent psychological results suggest that acceptance of a generic strongly influences our judgments concerning whether an arbitrary member of a kind has a property over and above our beliefs about the prevalence of the property. This is especially so for striking property generalizations. Sangeet Khemlani, Sam Glucksberg and I found that people were as likely to agree that Jumpy the tick carries Lyme disease as they were to agree that Joe the Canadian is right-handed – despite the very large discrepancy between the subjects own (roughly correct) judgments of the prevalence of the respective properties in the respective populations. Sixty-five percent of our participants who accepted the striking property generalizations judged – with varying degrees of confidence – that an arbitrary member of the kind would have the striking property.10 Andrei Cimpian, Amanda Brandone, and Susan Gelman found similar results using a very different experimental design. While their participants frequently accepted novel striking property generics at low prevalence levels, if they were presented with a novel striking property generic and asked to estimate how prevalent the property might be among the kind, they gave extremely high estimates – in many cases, 100%.11
These findings suggest that these generalizations play a powerful role in guiding our inferences concerning property possession, despite their relatively weak acceptance conditions. These primitive generalizations are not psychologically inert – rather they play a powerful role in guiding our judgments about members of a kind. In an early paper on generics, Robert Abelson and David Kanouse noted that some generics require very little evidence for acceptance, and yet “once accepted psychologically they appear to be commonly taken in a rather strong sense, as though the quantifier always had implicitly crept into their interpretation”. 12 Our most primitive method of generalization seems to encourage us in reasoning from ‘some’ to “many” or “most”, or even to “all”, at least when striking properties are in play.
Fear, Race, and Generalization
The cognitive disposition to widely generalize strikingly negative information may serve a useful purpose in the non-social realm. When we turn to generalizations about groups of people, however, it leads to disastrous consequences. I should reiterate that what follows is in no way intended as an exhaustive account of the cognitive factors underlying racism, and it does not touch on the social and cultural factors that have bred and sustained racism. Rather, I seek only to identify a particular and pervasive aspect of our thinking which itself leads us down the dark road to prejudice. Our most primitive method of generalization has the potential to enshrine a pervasive form of bigoted thinking.13
The basic idea is quite simple: just as it takes but a few instances of sharks attacking bathers, or of mosquitoes carrying the West Nile virus, for us to make the corresponding category-wide generalization, so also a strikingly negative action on behalf of a few members of a racial, ethnic, or religious minority (or “out-group”) may lead us to a general belief concerning their entire group. I discuss below why it may be that these particular social groups, as opposed to others, tend to be the targets of such generalizations, and I then identify the fundamental error behind these generalizations in the case of social groups. (That is, I explain how it may be true that mosquitoes carry the West Nile virus, and yet decidedly false that, say, Muslims are terrorists.) Once we identify the enabling error behind such generalizations over social groups, the opportunity arises to consider novel ways of combating this way of thinking.
As a result of a profound and pervasive cognitive bias built into our most primitive method of generalization, a few appalling acts on behalf of some members of a given group can lead us to draw conclusions about the group in general. As the available experimental evidence suggests, acceptance of a striking property generalization can lead one to draw the corresponding conclusion about the arbitrary member of the group – conclusions which go beyond even the perceived statistical facts. Extreme and aberrant actions on behalf of the few can thus lead to conclusions concerning the group at large, and these conclusions will influence our judgments concerning a newly encountered member of the group.
A rather pristine version of this pattern of reasoning is presented in Laura Ingalls Wilder’s Little House on the Prairie. The Ingalls’ rather prejudiced neighbor, Mrs. Scott claims that “The only good Indian is a dead Indian” – certainly a very sweeping and inclusive claim. She immediately justifies the claim by citing the Minnesota Massacre, going as far as to say “To anyone who disagrees, I say, ‘remember the Minnesota Massacre’!” Mrs. Scott reasons from the single horrific incident of the Minnesota Massacre to the conclusion that there are no good (living) Indians. Considering that the Scotts and the Ingalls were living in Kansas at the time, it is unlikely that she believed any of the Indians they actually encountered had any personal involvement in the Minnesota Massacre. Nonetheless, she took the single incident to justify the claim that the only good Indians were dead Indians. She was also apparently confident that only someone who had forgotten about the massacre would disagree with her on the point.14
Mrs. Scott’s reasoning is a perfect illustration of how human beings can move from a horrific particular to a sweepingly prejudiced generalization. If reasoning of this sort really is a pervasive cognitive disposition then we should find many examples of it, in whatever historical period we happen to examine. We should not be surprised if this mechanism of generalization has hovered perpetually in the background wherever human beings were formulating prejudiced attitudes towards social groups. A detailed historical analysis is beyond the scope of this paper, but let us briefly consider a recent and vivid example.
Nothing has done more to harm the plight of Muslims in America than 9/11. In the aftermath of 9/11, hate crimes against Muslims rose more than 1,600%, according to FBI statistics.15 Hate crimes are, by definition, crimes motivated by the mere fact that the victim is a member of a particular group; the hate crimes following 9/11 were motivated by the fact that the victims in question were Muslims. Many of these crimes were committed against Muslim women and children; the perpetrators surely were not under the impression that their victims were themselves involved in or personally responsible for the 9/11 bombings. It was sufficient that the victims were Muslims. We might characterize the reasoning of the hate crime perpetrators as moving from the horrific events of 9/11 – events which involved a rather small number of extreme individuals – to the conclusion that the arbitrary Muslim deserved to be victimized in virtue of being Muslim. The conclusions drawn from the 9/11 attacks did not concern just the bombers and their supporters, but concerned Muslims in general.
Such generalizations were made even by members of Congress. Shortly after 9/11, Representative John Cooksey told a Louisiana radio station, “If I see someone [who] comes in that's got a diaper on his head and a fan belt wrapped around the diaper on his head, that guy needs to be pulled over.” In Georgia, Representative, now Senator, C. Saxby Chambliss told law enforcement official to “just turn [the sheriff] loose and have him arrest every Muslim that crosses the state line.”16 These statements again reflect conclusions pertaining to Muslims quite generally. They do not reflect the more moderate conclusion that only some Muslims had any involvement in 9/11 whatsoever.
Obviously the aftermath of 9/11 is not an isolated historical example. Consider, for example, the origins of anti-Algerian prejudice in France. The relationship between the French and the Algerians is a complicated one, as would be expected given their history of war, colonization and occupation, and indeed anti-Algerian racism is still virulent in France today. The cognitive bias under investigation here is, of course, too simple to account for all the subtleties of a racism with such a complex history. However, if we trace anti-Algerian prejudice to its early days, we find again the dead hand of striking property generic reasoning.
Early relations between Algerian workers and France’s indigenous population were quite amicable.17 In 1923, however, French-Algerian relations took a sharp turn for the worse. A wave of anti-Algerian violence began, in which North Africans were attacked at random. The attacks included a public lynching in the rue Fremicourt in Paris, and it was unsafe for North Africans to venture into the surrounding area. The media denigrated North Africans, and petitions were circulated that called for “the undesirables to be driven from the area”.18
According to Neil MacMaster, this rapid swell in hostility towards North Africans can be traced to a single catalytic incident. On November 7 1923, Khemile Ousliman, an unemployed North African man, knifed a woman in the Rue Fondary. Ousliman, who was likely mentally ill, had been obsessed with the woman, and had repeatedly made sexual advances towards her. When she refused, he slit her throat, then turned in a frenzy on some passers by, killing another woman and wounding two others.
Immediately following this incident, there began a surge of anti-Arab violence, hatred, and discrimination throughout France. Seven years later, Paul Catrice, a Catholic priest and immigration expert, remarked that “If the Sidi, in general, inspires a certain repulsive fear, it is because of the memories of certain sensational crimes from which Parisians have drawn unconsidered generalizations.”19 These “unconsidered generalizations” are exactly those considered in this paper. There is no a priori reason to think that human beings would be disposed to reason from a single sensational event to a category-wide generalization; certainly there is no logical demand for such thinking. We are, however, possessed of a particular cognitive bias – a style of generalization – that makes such reasoning not only possible, but pervasive.
More speculatively, since the veil of years here is thicker, the early origins of Anglo-American prejudice towards Africans and Native Americans may have been fueled by sensational reports of horrific acts relayed in “travel logs”, which were extremely popular among the newly literate population of Britain. Very few people could afford to travel abroad themselves, so the reports of a small number of explorers were the source of public knowledge of foreign lands and their inhabitants. The initial impressions of the English population vis-à-vis Africans and Native Americans derived almost wholly from these travel logs.20 Winthrop Jordan, in his discussion of travel logs on Africa, writes
To judge from the comments of voyagers, Englishmen had an unquenchable thirst for the details of savage life … It is scarcely surprising that civilized Englishmen should have taken an interest in reports about cosmetic mutilation, polygamy, infanticide, ritual murder and the like – of course English men did not really do any of these things themselves … It would be a mistake to slight the importance of the Negro’s savagery, since it fascinated Englishmen from the very first. English observers in West Africa were sometimes so profoundly impressed by the Negro’s deviant behavior that they resorted to a powerful metaphor with which to express their own sense of difference from him. They knew perfectly well that Negroes were men, yet they frequently described the Africans as “brutish” or “bestial” or “beastly.” The hideous tortures, the cannibalism, the rapacious warfare, the revolting diet (and so forth page after page) seemed somehow to place the Negro among the beasts.21
These travel logs, which did so much to shape England’s early image of Africa, contained endless gory accounts of shocking behavior on behalf of the Africans. It should be noted that the travelers themselves often reported only specific incidents of cannibalism, or other specific instances of horrific violence. That is, it would be overly simplistic to place the blame for the formation of early negative stereotypes squarely on the explorers. Many of them were quite responsible in their reporting, and did not indulge themselves in broad generalizations. Given the nature of our default system of generalization, they did not have to. The reporting of specific instances would suffice to encourage very general beliefs in the mind of the reader.
Generalizations, Dispositions and Predictors
If the foregoing is correct, then the same pattern of generalization is in play for both claims like mosquitoes carry the West Nile virus and claims like Muslims are terrorists. Surely though, there must be some dissimilarities between them: in particular, is it not the case that the former claim is true, while the latter is false? Even if the same unreflective mechanism is responsible for both judgments, it is surely making an error in judging that Muslims are terrorists – an error which is not (necessarily) involved in judging that mosquitoes carry the West Nile virus.
Let us then consider striking property generics in more detail. Their truth conditions are not quite as straightforward as the earlier discussion suggests. We have been speaking as if a generic ‘Ks are F’ is true iff some Ks are F, given that being F is a dangerous or harmful property. But this would suggest that ‘insects carry the West Nile virus’, or even ‘animals carry the West Nile virus’ would also be true – certainly there are some insects, and therefore some animals that carry the virus, namely those few unfortunate mosquitoes. Similarly, the truth of ‘tigers eat people’ would entail the truth of ‘mammals eat people’, and from the truth of ‘sharks attack bathers’ we should conclude that fish attack bathers. People do not tend to find these inferences acceptable, so the truth conditions of these generics must involve some further complexity.
In earlier work, I suggested that the mechanism of generalization in question seeks a good predictor of the property in question.22 It is easy enough to see an evolutionary rationale behind generalizing striking properties only so far up the taxonomic hierarchy. If our ancestors had undertaken to avoid all mammals after seeing a tiger eating one of their companions, the costs of doing so may well have outweighed the benefits. (One could waste a lot of time running from small harmless creatures.) Someone who avoided all animals, big or small, after witnessing a lion maul his companion would be at a significant disadvantage relative to a more sophisticated competitor who limited his conclusions to lions alone.
An efficient generalizing mechanism, we might suppose, should seek a good predictor of the striking property – a kind that is inclusive enough to aid us in avoiding the property, but not so inclusive as to needlessly hamper our activities.
I further suggest that what makes a kind a good predictor of a striking property is that the members of the kind that do not possess the property are typically disposed to possess it.23 It matters, then, for the truth of ‘mosquitoes carry the West Nile Virus’ that the virus-free mosquitoes will carry the virus if circumstances allow. ‘Sharks attack bathers’ is true only if the sharks that never in fact cause harm to humans would typically do so given half a chance, and so on. Statements such as ‘animals carry the West Nile Virus’ and ‘sea creatures attack bathers’ are false, because the members of the kinds in question do not share the relevant dispositions. A generic statement in which a striking property is predicated is, I claim, true if and only if some members of the kind in question possess the relevant property, and the others are typically disposed to possess it.24 To determine which striking property generics are strictly true and strictly false, then, would require some rather detailed knowledge of dispositions and capacities. Dispositions are not directly observable in the way their manifestations are, and so we do not normally possess such knowledge. We thus often operate under uncertainty when it comes to attributing dispositions, and so must adopt certain heuristics to guide our judgments. To probe this strategy further, let us set aside questions of whether sentences are true or false, and consider how our basic mechanism of generalization must work, if my theses here are correct.