Perceptions of Race

Abstract

UNTIL RECENTLY, EXPERIMENTS ON PERSON PERCEPTION HAD LED TO TWO UNWELCOME CONCLUSIONS: (1) people encode the race of each individual they encounter, and (2) race encoding is caused by computational mechanisms whose operation is automatic and mandatory. Evolutionary analyses rule out the hypothesis that the brain mechanisms that cause race encoding evolved for that purpose. Consequently, race encoding must be a byproduct of mechanisms that evolved for some alternative function. But which one? Race is not encoded as a byproduct of domain-general perceptual processes. Two families of byproduct hypotheses remain: one invokes inferential machinery designed for tracking coalitional alliances, the other machinery designed for reasoning about natural kinds. Recent experiments show that manipulating coalitional variables can dramatically decrease the extent to which race is noticed and remembered.

Full text

Perceptions of race
Leda Cosmides
1
, John Tooby
2
and Robert Kurzban
3
1
Department of Psychology, University of California, Santa Barbara, CA 93106, USA
2
Department of Anthropology, University of California, Santa Barbara, CA 93106, USA
3
Department of Psychology, University of Pennsylvania, 405 Psychology Office Building, 3815 Walnut Street, Philadelphia, PA, USA
Until recently, experiments on person perception had
led to two unwelcome conclusions: (1) people encode
the race of each individual they encounter, and (2) race
encoding is caused by computational mechanisms
whose operation is automatic and mandatory. Evol-
utionary analyses rule out the hypothesis that the brain
mechanisms that cause race encoding evolved for that
purpose. Consequently, race encoding must be a bypro-
duct of mechanisms that evolved for some alternative
function. But which one? Race is not encoded as a
byproduct of domain-general perceptual processes.
Two families of byproduct hypotheses remain: one
invokes inferential machinery designed for tracking coa-
litional alliances, the other machinery designed for
reasoning about natural kinds. Recent experiments
show that manipulating coalitional variables can dra-
matically decrease the extent to which race is noticed
and remembered.
Race-based inferences – stereotypes – are easy to activate
and inactivate, given the appropriate context [1–2]. But
what about race encoding? The race of an individual must
be noticed and remembered before a racial stereotype can
be activated or racially motivated behavior can occur. Is it
possible not to notice a person’s race?
Race exists in the minds of human beings. But
geneticists have failed to discover objective patterns in
the world that could easily explain the racial categories
that seem so perceptually obvious to adults (see Fig. 1).
When, beginning in the mid-1960s, the technology
emerged to sequence genes and the proteins they code
for, the distribution of alleles could at last be objectively
mapped in the human species. What geneticists discovered
was an underlying reality that bore no resemblance to
existing hereditarian and folk theories of race. The first
idea to be falsified was that most genetic variation in the
human species served to differentiate races. Within-
population genetic variance was found to be ,10 times
greater than between-race genetic variance (i.e. two
neighbors of the same ‘race’ differ many times more,
genetically speaking, than a mathematically average
member of one ‘race’ differs from an average member of
another [3–5]). Second, compared with other similar
species, there is little genetic diversity among humans.
For example, the diversity of gene sequences among
chimpanzees is almost four times higher than for humans
[6], despite the fact that their population sizes are far
smaller. In fact, the genetic distances for protein loci
between European, Asian and sub-Saharan African
populations ‘are of the same order of magnitude as those
for local populations in other organisms and considerably
smaller than those for subspecies.’ (Ref. [4], p. 11). That is,
by the criteria that biologists typically use to apply the
concept of ‘subspecies’ or ‘races’, humans do not qualify.
Most telling of all, virtually no expressed genes have been
identified that are shared by all normal members of one
race (and hence could explain a common racial appear-
ance) that are not also present at substantial levels in
other races (thereby failing to sort individuals into
races) (see Fig. 1).
Fig. 1. Despite perceptions to the contrary, geneticists have shown that humanity
is not divided into distinct racial types (for discussion, see Refs [16,40– 42]). Geneti-
cists have investigated the distribution of common protein-building genes – that
is, those that can cause detectable phenotypic differences, and so have the poten-
tial for differentiating groups by appearance (shared racial appearance cannot be
caused by genes that are unexpressed or rare). Most genes influencing appear-
ance are not known, but by using the hundreds of common proteins that are
known, human population structure could be mapped. The weak dimensions of
observable geographical variation are typically clines whose scale, directionality,
distribution and slope do not reflect each other (as they should if racial typologies
were reflected in the real world) [43]. As an example, sorting human populations
by the O allele of the ABO blood group gives a characteristically counterintuitive
result, with Icelanders clustering with Japanese, Ethiopians clustered with
Swedes, and so forth. Each block contains populations having a similar percentage
of the O allele [40,5]. Different colors denote continent of origin. Sorting by other
protein-building alleles gives different but equally counterintuitive results. In gen-
eral, the frequency of a common, protein-building allele will cluster populations
into groups that typically violate rather than support traditional racial categories
[40,44]. The human species simply cannot be reliably sorted into types based on
sets of genes that are shared by most members of the type but different from the
genes shared within other types. Significantly, the patterns of genetic and pheno-
typic variation are sufficiently rich that by choosing alternative criteria, the human
mind could be trained to cluster humans into a large number of alternative,
mutually contradictory groupings.
TRENDS in Cognitive Sciences
>60%
Navahos, S.
Amer.Toba
55-59%
Icelanders,
Japanese
50-54%
Scots
45-50%
Australian
Aborigines,
Sicilians
40-44%
English,
Spaniards,
Eskimos,
Norwegians
35-39%
Swedes,
Ethiopians
30-34%
Finns, Buriats,
Kirghz, Russians,
Chinese, Pygmies
<29%
Tartars,
Egyptians,
Blackfoot
Corresponding author: Leda Cosmides (cosmides@psych.ucsb.edu).
Review TRENDS in Cognitive Sciences Vol.7 No.4 April 2003 173
http://tics.trends.com 1364-6613/03/$ - see front matter q2003 Elsevier Science Ltd. All rights reserved. doi:10.1016/S1364-6613(03)00057-3

Nevertheless, the claim that humanity is not divided into
distinct racial types is often met with incredulity – who am I
going to believe, scientists or theclear evidence of my senses?
Incredulity reigns because we do perceive race. Using
an unobtrusive measure – a memory-confusion protocol
developed by Taylor et al. [7] (Box 1) – social psychologists
found that when adults encounter a new individual, they
encode that individual’s race, sex and age [7–9] (see
[10–13] for review and discussion). These dimensions can
be encoded without other individuating information; for
example, one might recall that one’s new neighbor is a
young, white woman, without remembering anything else
about her [7–11] – her name, her hair color, her job. Until
recently, it appeared that race – along with sex and age –
was encoded in an automatic and mandatory fashion.
The encoding of race was thought to be spontaneous and
automatic because the pattern of recall errors that
indicates race encoding (Box 1) occurred in the absence
of instructions to attend to the race of targets, and across a
wide variety of experimental situations. It was thought to
be mandatory – encoded with equal strength across all
situations – because every attempt to increase or decrease
the extent to which subjects encode the race of targets had
failed [7–9].(Box 2) Such results led some to propose that
race, sex, and age are ‘primary’ or ‘primitive’ dimensions of
person perception [12,13], built into our cognitive archi-
tecture. Until recently, no context manipulation – whether
social, instructional, or attentional – had been able to
budge this race effect.
Automatic race encoding is a puzzle
Natural selection would plausibly have favored neuro-
computational machinery that automatically encodes an
individual’s sex and age. For millions of years, our
ancestors inhabited a social world in which registering
the sex and life-history stage of an individual would have
enabled a large variety of useful probabilistic inferences
about that individual (e.g. adolescent girl; toddler boy). By
contrast, ‘race’ is a very implausible candidate for a
conceptual primitive to have been built into our evolved
cognitive machinery. Ancestral hunter-gatherers traveled
primarily by foot, making social contact geographically
local [14]. Given the breeding structure inherent in such a
world, the typical individual would almost never have
encountered people drawn from populations genetically
distant enough to qualify as belonging to a different ‘race’
(Fig. 1). If individuals typically would not have encoun-
tered individuals of other races, then there could have
been no selection for cognitive adaptations designed to
preferentially encode such a dimension, much less encode
it in an automatic and mandatory fashion. Race encoding
may be a robust and reliable phenomenon, but it cannot be
Box 1. The memory confusion protocol (‘who said what?’)
Developed in the 1970s by Taylor et al. [7], the memory confusion
protocol uses errors in recall to unobtrusively reveal whether
subjects are categorizing target individuals along a dimension of
interest, such as race or sex. During phase 1, subjects are told that
they will see some individuals engaged in a conversation, and that
they should try to form an impression of each individual. They are
then shown a series of photos of individuals, each of which is paired
with a sentence that was supposedly uttered by that individual. The
individuals in the photos differ along one or more dimensions of
interest –race, sex, age, and so on. In a recent study by Kurzban et al.
[21], each photo was shown for 8.5 s, and phase 1 lasted ,4 min.
(Fig. I). Phase 2 is a surprise recall test (after a distracter task) in which
the subject is shown each sentence (in random order), and asked
which individual uttered it (all photos are simultaneously present).
This ‘who said what?’task is difficult, and subjects make many
mistakes. By analyzing the pattern of errors, the experimenter can tell
whether the subject had encoded a categorical dimension of interest
during phase 1. For example, suppose half the targets were women
and half were men. If a subject encodes the sex of targets during
phase 1, that subject’s errors in phase 2 will not be random: they will
be more likely to misattribute a sentence uttered by a man to another
man rather than to a woman (and vice versa). Indeed, for any
category of interest –race, sex, age, and so on –the experimenter
will find that pðwithin category errorÞ.pðbetween category errorÞ
for any category that the subject did, in fact, encode. By contrast, a
subject who did not encode sex during phase 1, will produce errors
that are random with respect to that category during phase 2, that is,
pðwithin category errorÞ¼pðbetween category errorÞ:Past experi-
ments using the memory confusion protocol showed that when
the targets differed in race (black vs. white), subjects made
significantly more within-race errors than between-race errors
[7–9]. This means that subjects were encoding the race of targets
in those experiments.
Fig. I. An example of two photos used in a memory confusion protocol exper-
iment [21]. The pattern of errors made in these experiments can show
whether subjects have encoded race.
Box 2. Attempts to reduce race encoding
Over the last two decades, considerable effort has been expended on
the search to find conditions under which race is not encoded,
without success [7–9]. Studies carried out in the US and Britain,
using targets that were ‘black’and ‘white’, showed that the size of the
race-encoding effect remained roughly the same over the following
experimental manipulations:
1. Targets are discussing race relations versus a race-neutral
topic [8].
2. Subject is ‘black’or ‘white’[7 –8].
3. Subject is led to believe that s/he would soon be interacting
with the targets [8] (a context that is believed to promote the
encoding of individuating information over category-based
information [10–11]).
4. Subject operates under cognitive load or not [8].
5. Subject is warned that there would be a recall test [7].
6. A competing dimension was included (e.g. targets differed in
both sex and race) [9].
7. Attention is drawn to targets’race (‘pay attention to people’s
race’), or away from it (‘pay attention to people’s sex’)[9].
The evolutionarily derived prediction that race is encoded as a
proxy for coalition led to the discovery of the only context known so
far to reduce race encoding: a context of conflict in which coalitional
membership is uncorrelated with race [21].
Review TRENDS in Cognitive Sciences Vol.7 No.4 April 2003
174
http://tics.trends.com

caused by computational machinery that was designed by
natural selection for that purpose.
This means that race encoding must be a side-effect of
machinery that was designed by selection for some
alternative function. Three proposals have been advanced:
(1) Race encoding is a byproduct of domain-general
perceptual/correlational systems [7].
(2) Race encoding is a byproduct of an essentialist
inference system that evolved for reasoning about
natural kind categories [15–20].
(3) Race encoding is a byproduct of computational
machinery that evolved for tracking coalitions and
alliances [21].
Is race encoding a byproduct of perceptual/correlational
systems?
Many species, including humans, appear to have compu-
tational machinery that is well-designed to pick up
correlations between perceived features and events
[22,23] – indeed, classical conditioning is possible only
because such machinery exists [24]. Given that humans
rely on vision more than most animals, and that we have
good color vision, could it be that race is automatically
encoded merely as a byproduct of the ordinary operation of
our visual and correlation-detecting systems?
Hirschfeld [16] reviews evidence from genetics and
anthropology that undermines the notion that races are
biological kinds, merely ‘out there’ to be discovered by our
perceptual systems. Real human variation is continuous,
and substantially uncorrelated (Fig. 1). It is true, however,
that if individuals who are descended from populations
that inhabited mutually distant parts of the globe for many
thousands of years – central Africa, northern Europe, east
Asia – are juxtaposed, one will see a few phenotypic
features that are found more commonly in one group than
another, including some that are intercorrelated in a
Roschian, family resemblance fashion. In contrast to the
ancestral world, the recent advent of long distance
transportation leads to individuals descended from dis-
parate populations sometimes living side by side. Perhaps
our perceptual/correlational systems are merely detecting
these perceptual clusters. On this view, the automatic,
mandatory encoding of race is merely a byproduct of the
automatic, mandatory encoding of perceptual attributes
such as color and shape, which operate equally on all
stimuli, whether people or objects. This perceptual
byproduct hypothesis is plausible, and was the first
hypothesis proposed to explain race encoding in the
memory confusion protocol [7].
Several predictions follow directly from the percep-
tual byproduct hypothesis; evidence from Stangor et al.
[9] and Hirschfeld [16] weigh strongly against them
(see Box 3). In addition to failing these empirical tests,
the perceptual byproduct hypothesis has a major
limitation: It cannot explain why membership in a
racial category – as determined by perceivable,
phenotypic surface features – is the basis for making
inferences about a person’s behavior. Yet one of the
most obvious (and disturbing) features of racial thinking is
racial stereotypes: inferences about a person’s traits,
personality, goals, moral dispositions, affiliations, and
behavior based on racial category membership. By
contrast, the essentialist and coalitional byproduct
hypotheses easily explain why category membership –
whether racial or not – will support inferences about
behavior and traits.
Is race encoding a byproduct of essentialist reasoning?
Although there are important differences between
their views, Rothbart and Taylor [15],Hirschfeld[16],
Box 3. Evidence against the perceptual byproduct
hypothesis
Four predictions follow directly from the perceptual byproduct
hypothesis. Data from Stangor et al. [9] and from Hirschfeld [16]
speak against them. Briefly:
Prediction 1. If race is encoded merely as a byproduct of domain-
general categorization processes that cannot help but use color, then
subjects will encode the color of objects in an automatic and
mandatory manner – even when color differences have no social
significance (i.e. they do not indicate race, team membership,
personality, status, politics or any other social dimension).
Using the memory confusion protocol, Stangor et al. (Exp. 5) showed
that the encoding of color is not automatic and mandatory. They
examined whether people automatically encode clothing color
(black versus white shirts), when it has no social significance
(i.e. does not indicate team, status, coalition, etc.). Targets were
black and white women; shirt color was uncorrelated with race.
Subjects encoded race, but did not encode shirt color at all
(limitations on attention cannot explain this result; when the two
dimensions are race and sex, both are strongly encoded [9]). Brewer
et al. [45: Exp 1] found similar results, even in the absence of
competing dimensions (targets all same sex and race).
When judging similarity –especially for human targets –children
do not privilege skin color over other perceptual dimensions
[16: pp. 93–101].
Prediction 2. Task demands that decrease or increase the extent to
which subjects encode perceptual dimensions in tasks with non-
social stimuli should decrease or increase the extent to which
subjects encode race.
Instructions to attend to color (or shape) increase the extent to which
it is encoded with non-social stimuli [46,47]. But Stangor et al. [Exps
1, 2, note 3] showed that the same task demands have no effect on
encoding of race (see Box 2, points 6,7).
Prediction 3. Perceptual similarity should affect how strongly race is
encoded; that is, there should be prototype effects.
If the encoding of race is merely a byproduct of the low-level, domain-
general computational machinery, then one should see prototypi-
cality effects for racial categories, just as one sees for artificial
categories [48 –50]. E.g., perceived similarity of individuals of a given
race should produce powerful effects in categorization experiments,
affecting learning, inference, recall, and recognition. Stangor et al.
(Exp 2) directly tested this prediction and found no prototypicality
effects: race was encoded just as strongly in conditions where the
same race targets were very different in physical appearance as in
those where their physical appearance was very similar.
Prediction 4. Development of racial categories is not bottom up.
If racial categories were built inductively from perceptual features in
a bottom-up fashion, then recognizing which people are members of
each category should be trivial. But it is not: Preschoolers, who know
and use racial category terms, are rather poor at telling which people
in the world fall into which racial category [16]. Hirschfeld shows their
categories are driven by labels, not perceptual features.
Review TRENDS in Cognitive Sciences Vol.7 No.4 April 2003 175
http://tics.trends.com

Gil-White [17] and several others [18 – 20] have all
proposed that inferences about race are a byproduct of
an essentialist inference system that is functionally
specialized for reasoning about natural kinds (Box 4).
The way people reason about natural kinds – tigers,
oysters, gold, oak trees – is different from the way they
reason about arbitrarily defined categories, such as ‘white
things’ (a motley category that includes pearls, doves,
some sinks but not others, etc.) [25 –27]. If two things are
both judged to be members of the same natural kind, we
infer that they share many properties in common,
including nonobvious or even hidden ones. For example,
having seen that one zebra has a heart, eats grass, runs
fast, and fears lions, I might infer that other zebras do too.
Use of a category label can support the judgment that two
entities belong to the same natural kind (e.g. someone tells
me that each of those animals is a ‘zebra’); so can
perceptual similarity (both animals are horsey and
striped). But perceptual similarity is not necessary for
kind membership – an albino zebra is still a zebra.
Children and adults act as although natural kinds have
defining essences, ‘underlying natures that make them the
thing that they are’ (Ref. [26], pp. 1476 –1477).
Many features of racial categorization and reasoning
become comprehensible if one assumes that they are
governed by this inference system. Cross-culturally
recurrent features of racial thinking include the following
notions (see Hirschfeld [16]): (1) there are different kinds of
humans; (2) people of different races are different in kind;
(3) being of a certain race causes many properties, both
physical and non-physical, including nonobvious ones
such as inner traits and dispositions (temperament,
character, shared ‘blood’); (4) possessing the correct
‘underlying nature’ is what makes one of given race,
regardless of perceptual properties; and (5) how you look
may be a good clue to your race, but a person may look like
one race but ‘really’ be of another. If the human mind were
(mis)interpreting race as indicating membership in a
natural kind category, these are precisely the assumptions
and inferences that the essentialist system would produce.
The essentialist system doesn’t care whether human
‘races’ in fact form natural kinds. The system is activated
by particular input conditions, and any stimuli that fit
those conditions will be treated as a natural kind. On this
view, some of the system’s input conditions invite the
inference that people are divided into different races,
constituting different ‘kinds’ of people. Just as being told
that there are ‘dolphins’ and ‘fish’ leads children to encode
otherwise similar-looking fishy creatures as members of
two different natural kinds [25], being told that there are
‘black people’ and ‘white people’ leads children to encode
otherwise similar-looking people as members of two
different natural kinds [16– 18,20]. Perceptual similarity
might also feed the system: The clusters of correlated
phenotypic traits that one sees in modern societies could be
interpreted as distinguishing several different, phenoty-
pically defined ‘kinds’ of people [19]. Indeed, the presence
of kind labels may organize and drive perception, grace-
fully explaining why children who know racial terms do
not sort individuals into racial categories on the basis of
the perceptual features used by adults [16] (Box 3).
Empirical work on this topic has explored inference, not
encoding (e.g. [16,17]). But the application to race
encoding is straightforward. If subsets of people fit the
input conditions for the essentialist system, it will swing
into action. Once it infers (wrongly, in this case) that
people are divided into intrinsically different racial kinds
and has acquired some criteria for sorting individuals into
racial categories [28], it should cause a person’s race to be
automatically encoded. A person’s race would be auto-
matically encoded for the same reason that their sex is –
and, indeed, for the same reason that one automatically
encodes whether the organism one is looking at is a lion, a
bird, or a human. Knowing which natural kind an entity
belongs to supports many inferences, and the system has
mistakenly mapped arbitrary racial categories onto the
conceptual apparatus for reasoning about natural kinds.
Is race encoding a byproduct of coalitional psychology?
According to Kurzban, Tooby, and Cosmides [21], no part of
the human cognitive architecture is designed specifically
to encode race. The (apparently) automatic and mandatory
encoding of race is instead a byproduct of adaptations that
evolved for an alternative function that was a regular part
of the lives of our foraging ancestors: detecting coalitions
Box 4. Natural kinds and essentialism: three theories
(1) Essentialism and social categories
Rothbart and Taylor [15] propose that the essentialist system can
be applied to any social category (including racial ones), and that this
occurs whenever cultural beliefs imply that category membership is
inalterable and carries inductive potential.
(2) Essentialism and folk sociology
According to Hirschfeld [16], ‘humans organize themselves into
collectivities and define themselves into social kinds as a function of
group membership’(p. 119). In consequence, the human cognitive
architecture evolved an intuitive ‘theory of society’, the function of
which is to produce expectations about the skeletal structure of
society, including what kinds of people there are. The system uses an
essentialist mode of construal. Essentialist reasoning enables (but
does not require) the notion that humans are biologically clustered,
because this same essentialist mode functions across many
domains, and plays a role in generating a folk biology. By pitting
race against alternative social categories and perceptual cues,
Hirschfeld showed that the inductive inferences of children and
adults follow racial category for many traits.
(3) Essentialism and ethnic groups
Gil-White [17] argues that the human cognitive architecture has a
module designed for reasoning about living kinds (species), which
includes an essentialist inference system, and is activated by cues of
category-based endogamy (within-group breeding) and descent-
based membership. Because ethnic groups manifest both cues, they
would have activated this module under ancestral conditions,
entirely as a byproduct of the module’s design. Nevertheless,
because different ethnic groups have different cultural norms and
they (as opposed to clans and other groups) represent norm
boundaries, the essentialist inference that category membership
predicts clusters of non-obvious shared properties conferred a net
selective advantage. As a result, the living kinds module was co-
opted by natural selection, which produced a new module designed
for reasoning about ethnic groups. It is especially prone to produce
inferences about ethnic differences in status, norms, morals, and
customs (rather than phenotypic differences). In studies of Kazakhs
and Mongols, Gil-White showed that trait induction is often based on
the ethnicity of the birth parents, rather than on the ethnicity of the
parents who raised a child.
Review TRENDS in Cognitive Sciences Vol.7 No.4 April 2003
176
http://tics.trends.com

and alliances. Hunter-gatherers lived in bands, and
neighboring bands often came into conflict with one
another [29–31]. Moreover, there were also coalitions
and alliances within bands [32] (a pattern also found in
related primate species [33]). To foresee the likely social
consequences of alternative courses of action, and to
navigate their social world successfully, our ancestors
would have benefitted by being equipped with neuro-
cognitive machinery that tracked these shifting alliances.
Computational machinery that is well-designed for
detecting coalitions and alliances in the ancestral world
should be sensitive to two factors: (1) patterns of
coordinated action, cooperation, and competition, and (2)
cues that predict – whether purposefully or incidentally –
each individual’s political allegiances [34–36].
Alliance cues
Like other behaviors, actions that reveal coalitional
dispositions are usually transitory. Alliance tracking
machinery should therefore be designed to note these
rare revelatory behaviors when they occur, and then use
them to isolate further cues that happen to correlate with
coalition but that are more continuously present and
perceptually easier to assay. Because this circuitry detects
correspondences between allegiance and appearance, stable
dimensions of shared appearance – which might be
otherwise meaningless (e.g. dress, dialect, etc.) – emerge
in the cognitive system as markers of social categories.
Coalitional computation increases their subsequent percep-
tual salience, and encodes them at higher rates.
Dynamic revision
Patterns of alliance usually change whenever fresh issues
arise whose possible resolutions differentially affect new
subsets of the local social world. Consequently, coalitions
shifted over time, varying in composition, surface cues,
duration and internal cohesion. To track these changes,
cue validities would need to be computed and revised
dynamically: No single coalitional cue (including cues to
race) should be uniformly encoded across all contexts.
Furthermore, arbitrary cues – such as skin color – should
pick up significance only insofar as they acquire predictive
validity for coalitional membership [36].
In societies that are not completely racially integrated,
shared appearance – a highly visible and always present
cue – can be correlated with patterns of association,
cooperation, and competition [36]. Under these conditions,
coalition detectors may perceive (or misperceive) race-
based social alliances, and the mind will map ‘race’ onto
the cognitive variable coalition. According to this hypoth-
esis, race encoding is not automatic and mandatory. It
appeared that way only because the relevant research was
conducted in certain social environments where the
construct of ‘race’ happened, for historical reasons [16],
to be one valid probabilistic cue to a different underlying
variable, one that the mind was designed to automatically
seek out: coalitional affiliation [34– 36].
Is coalition encoded?
Using the memory confusion protocol, Kurzban et al. [21]
first showed that people do automatically encode the
coalitional alliances of targets. The targets were males,
some black, some white; each made statements
suggesting allegiance with one of two antagonistic
coalitions. Crucially, race was not correlated with
coalitional affiliation.
Subjects encoded coalitional alliance even in the
absence of shared appearance cues – merely from patterns
of agreement and disagreement. When a shared appear-
ance cue – jersey color – was added, coalition encoding
was boosted dramatically, to levels higher than any found
for race. [N.B. Jersey color is not encoded at all when it
lacks social meaning (Box 3)].
Race as a proxy for coalition?
The results further showed that, as predicted, race
encoding is not mandatory. When coalition encoding was
boosted by a shared appearance cue, there was an
accompanying decrease in race encoding, which was
diminished in one experiment and eliminated in another.
Other tests showed that the decrease in race encoding
could not be attributed to domain-general constraints on
attention.
Subjects had a lifetime’s experience of race predicting
patterns of cooperation and conflict. The decreases in these
experiments occurred in response to only 4 min of
exposure to an alternative world where race did not
predict coalitional alliance. This is expected if (1) race is
encoded (in real life) because it serves as a rough-and-
ready coalition cue, and (2) coalition cues are revised
dynamically, to reflect newly emerging coalitions. There
are many contexts that decrease racial stereotyping
(inferences); creating alliances uncorrelated with race is
the first social context found that decreases race encoding.
The coalitional byproduct hypothesis fits neatly with
the literature on in group favoritism and outgroup
derogation, readily explaining why stereotypes of racial
outgroups often include derogatory elements. It dovetails
with Sidanius and Pratto’s [36] findings about social
dominance relations between racial groups. It can also
explain Levin’s [28] recent discoveries that a deficit in
cross-race face recognition: (1) is not universal; (2) does not
disappear after exposure per se to faces of another race; (3)
is rare in coalitional sports fans (where race does not
predict team membership); and (4) is caused by excessive
focus on facial features that discriminate races – a case of
racial (coalitional?) categorization driving the perception
of similarity, rather than vice versa.
Is race encoding a byproduct of coalitional psychology or
an essentialist system?
The human cognitive architecture contains many func-
tionally specialized computational systems. It would not
be surprising to find that it contains a system designed for
tracking coalitional alliances coexisting alongside several
different essentialist inference systems: one designed for
‘living kinds’ (species) [18–20], another for social group-
ings [15–16], and a third for ethnic groups [17]. (see Box 4).
But which of these is responsible for our perceptions of
race? It is too early to answer that question, but a few
observations can be made.
Review TRENDS in Cognitive Sciences Vol.7 No.4 April 2003 177
http://tics.trends.com

Encoding?
That coalitional context could decrease race encoding was
specifically predicted by the coalitional byproduct view; it
might also be consistent with Hirschfeld’s folk sociology,
although not required by it. We note, however, that it is not
a natural prediction of the other essentialist views. For
example, one’s ethnicity (and one’s species) remains the
same, regardless of which coalition one temporarily joins.
If race encoding were a byproduct of a ‘living kinds’
template [18–20] or an ethnicity module [17], then four
minutes of exposure to an orthogonal coalitional conflict
should not have decreased race encoding to such low levels.
If appearance [19] or cultural beliefs about the immut-
ability and inductive potential of race cause racial groups
to be interpreted as natural kinds [15], then race encoding
should decrease only as these change. Four minutes of
exposure to temporary coalitions (and no change in
appearance) should have no effect on such beliefs, and
therefore no effect on race encoding.
Folk concepts and cognitive architecture
The use of cognitive science to understand real world
cultural phenomena, such as folk concepts of race, is a
promising and exciting development [16–20,37]. But it is
not without difficulties.
Racial thinking, for example, includes inferences about
coalitional identity and about phenotypic traits; Hirsch-
feld’s experiments suggest that inferences about these two
types of properties do not follow the same logic ([16],
pp. 166 –180). Inferences about bodily traits reflect those of
a ‘living kinds’ template, whereas inferences about
identity appear to follow a more coalitional logic. Thus,
different folk beliefs about race can be generated by
different inferential machinery. Moreover, folk beliefs
about race (or religion or other matters) can be a byproduct
of different evolved inference mechanisms in different
subpopulations within a culture, depending on the history
and local distribution of beliefs [18 –20,37]. A facet of ‘race’
might be understood using a ‘living kinds’ concept in some
places and times; in others, coalitional concepts or
ethnicity templates might guide inference. By contrast,
inferences about the solidity and motion of rocks should be
similar across time and place. This is because the mind
contains inference systems designed for reasoning about
inanimate objects [38,39], whereas reasoning about race is
necessarily a byproduct of systems designed for other
functions.
Highly articulated models of the underlying inference
mechanisms and careful tests will be needed to untangle
these issues, but the payoff will be cultural theories that
are truly explanatory (and perhaps predictive) because
they grow out of an understanding of our evolved human
nature.
References
1 Macrae, C.N. and Bodenhausen, G.V. (2000) Social cognition: Thinking
categorically about others. Annu. Rev. Psychol. 51, 93–120
2 Wheeler, M.E. et al. (2001) fMRI study of three cognitive tasks that
differentially modulate stereotype accessibility and human amygdala
response to racial outgroup faces. Soc. Neurosci. Abstr. 27, 317.13
3 Lewontin, R. (1972) The apportionment of human diversity. Evol. Biol.
6, 381– 398
4 Nei, M. and Roychoudhury, A. (1982) Genetic relationship and
evolution of human races. Evol. Biol. 14, 1–59
5 Nei, M. and Roychoudhury, A. (1993) Evolutionary relationships of
human populations on a global scale. Mol. Biol. Evol. 10, 927 –943
6 Kaessmann, H. et al. (1999) Extensive nuclear DNA sequence diversity
among chimpanzees. Science 286, 1159– 1162
7 Taylor, S. et al. (1978) Categorical bases of person memory and
stereotyping. J. Pers. Soc. Psychol. 36, 778 –793
8 Hewstone, M. et al. (1991) Social categorization and person memory:
The pervasiveness of race as an organizing principle. Eur. J. Soc.
Psychol. 21, 517–528
9 Stangor, C. et al. (1992) Categorization of individuals on the basis of
multiple social features. J. Pers. Soc. Psychol. 62, 207 –218
10 Brewer, M. (1988) A dual process model of impression formation. In
Advances in Social Cognition (Vol. 1) (Srull, T. and Wyer, R., eds)
pp. 1–36, Academic Press
11 Fiske, S. and Neuberg, S. (1990) A continuum of impression formation,
from category-based to individuating processes: Influences of infor-
mation and motivation on attention and interpretation. In Advances in
Experimental Social Psychology (Vol. 23) (Zanna, M., ed.), pp. 1– 74,
Academic Press
12 Hamilton, D. et al. (1994) Social cognition and the study of
stereotyping. In Social Cognition: Impact on Social Psychology
(Devine, P.G. et al., eds), pp. 291 –321, Academic Press
13 Messick, D. and Mackie, D. (1989) Intergroup relations. Annu. Rev.
Psychol. 40, 45–81
14 Kelly, R. (1995) The Foraging Spectrum, Smithsonian Institution
Press
15 Rothbart, M. and Taylor, M. (1990) Category labels and social reality:
Do we view social categories as natural kinds? In Language and Social
Cognition (Semin, G. and Fiedler, K., eds) Sage
16 Hirschfeld, L. (1996) Race in the Making, MIT Press
17 Gil-White, F. (2001) Are ethnic groups biological ‘species’ to the human
brain? Essentialism in our cognition of some social categories. Curr.
Anthropol. 42, 515– 554
18 Sperber, D. (1994) The modularity of thought and the epidemiology of
representations. In Mapping the Mind: Domain Specificity in Cogni-
tion and Culture (Hirschfeld, L. and Gelman, S., eds) pp. 39 –67,
Cambridge University Press
19 Atran, S. (1990) The Cognitive Foundations of Natural History,
Cambridge University Press
20 Atran, S. (1998) Folk biology and the anthropology of science. Behav.
Brain Sci. 21, 547–611
21 Kurzban, R. et al. (2001) Can race be erased? Coalitional computation
and social categorization. Proc. Natl. Acad. Sci. U. S. A. 98,
15387– 15392
22 Lewicki, P. (1986) Nonconscious Social Information Processing,
Academic Press
23 Gigerenzer, G. et al. (1991) Probabilistic mental models: A Brunswi-
kean theory of confidence. Psychol. Rev. 98, 506–528
24 Gallistel, C.R. and Gibbon, J. (2000) Time, rate, and conditioning.
Psychol. Rev. 107, 289– 344
25 Markman, E. (1990) Categorization and Naming in Children,MIT
Press
26 Medin, D. (1989) Concepts and conceptual structure. Am. Psychol. 44,
1469– 1481
27 Gelman, S. and Hirschfeld, L. (1999) How biological is essentialism? In
Folkbiology (Medin, D. and Atran, S., eds) pp. 403 –446, MIT Press
28 Levin, D. (2000) Race as a visual feature: Using visual search and
perceptual discrimination tasks to understand face categories and the
cross-race recognition deficit. J. Exp. Psychol. Gen. 129, 559 –574
29 Keeley, L. (1996) War Before Civilization, Oxford
30 Ember, C.R. (1978) Myths about hunter-gatherers. Ethnology 27,
239– 448
31 Manson, J. and Wrangham, R. (1991) Intergroup aggression in
chimpanzees and humans. Curr. Anthropol. 32, 369 –390
32 Chagnon, N. (1992) Yanomamo, 4th edn, Harcourt
33 Smuts, B. et al. (1987) Primate Societies, University of Chicago Press
34 Kurzban, R. (2001) The social psychophysi cs of cooperation: Nonverbal
communication in a public goods game. J. Nonverbal Behav. 25,
241– 259
35 Tooby, J. and Cosmides, L. (1988) The evolution of war and its cognitiv e
foundations. Institute for Evolutionary Studies Technical Report;
Review TRENDS in Cognitive Sciences Vol.7 No.4 April 2003
178
http://tics.trends.com

Reprinted in Evolutionary Psychology: Foundational Papers (Tooby, J.
and Cosmides, L., eds), MIT Press
36 Sidanius, J. and Pratto, F. (1999) Social Dominance: An Intergroup
Theoryof Social Hierarchy and Oppression, Cambridge UniversityPress
37 Boyer, P. (2001) Religion Explained: The Evolutionary Origins of
Religious Thought, Basic Books
38 Spelke, E. (1990) Priciples of object perception. Cogn. Sci. 14, 29– 56
39 Leslie, A. (1994) ToMM, ToBy, and Agency: Core architecture and
domain specificity. In Mapping the Mind: Domain Specificity in
Cognition and Culture (Hirschfeld, L. and Gelman, S., eds)
pp. 119– 148, Cambridge University Press
40 Graves, J. (2001) The Emperor ’s New Clothes: Biological Theories of
Race at the Millennium, Rutgers University Press
41 Marks, J. (1995) Human Biodiversity: Genes, Races, and History,
Aldine de Gruyter
42 Tooby, J. and Cosmides, L. (1990) On the universality of human nature
and the uniqueness of the individual: The role of genetics and
adaptation. J. Pers. 58, 17–67
43 Brace, C.L. (1964) A non-racial approach toward the understanding of
human diversity. In The Concept of Race (Montagu, A., ed.), University
of Nebraska Press
44 Dunn, L. and Dobzhansky, T. (1946) Heredity, Race, and Society,
Mentor Books
45 Brewer, M. et al. (1995) Person memory in intergroup contexts:
Categorization versus individuation. J. Pers. Soc. Psychol. 69, 29 –40
46 Nosofsky, R. (1986) Attention, similarity, and the identification-
categorization relationship. J. Exp. Psychol. Gen. 115, 39– 57
47 Nosofsky, R. (1987) Attention and learning processes in the identifi-
cation and categorization of integral stimuli. J. Exp. Psychol. Learn.
Mem. Cogn. 13, 87– 108
48 Rosch, E. et al. (1976) Structural bases of typicality effects. J. Exp.
Psychol. Hum. Percept. Perform. 2, 491 –502
49 Anglin, J. (1976) Les premiers termes de reference de l’enfant. In La
Memoire Semantique (Ehrlich, S. and Tulving, E., eds) Bulletin de
Psychologie
50 Homa, D. and Cultice, J. (1984) Role of feedback, category size, and
stimulus distortion on the acquisition and utilization of ill-defined
categories. J. Exp. Psychol. Learn. Mem. Cogn. 10, 83–94
Could you name the most significant papers published in
life sciences this month?
Updated daily, Research Update presents short, easy-to-read commentary on the latest hot papers,
enabling you to keep abreast with advances across the life sciences.
Written by laboratory scientists with a keen understanding of their field, Research Update will
clarify the significance and future impact of this research.
Our experienced in-house team are under the guidance of a panel of experts from across
the life sciences who offer suggestions and advice to ensure that we have high calibre authors
and have spotted the ‘hot’papers.
Visit the Research Update daily at http://update.bmn.com and sign up for email alerts to
make sure you don’t miss a thing.
This is your chance to have your opinion read by the life science community, if you would like to
contribute contact us at research.update@elsevier.com
Review TRENDS in Cognitive Sciences Vol.7 No.4 April 2003 179
http://tics.trends.com