Culture, evolution and the puzzle of human cooperation
Action editors: Luca Tummolini and Cristiano Castelfranchi
Joseph Henrich *, Natalie Henrich
Department of Anthropology, Emory University, Geosciences Building, 1557 Pierce Drive, Atlanta, GA 30322-1720, United States
Received 15 March 2005; accepted 7 November 2005
Available online 9 March 2006
Abstract
Synthesizing existing work from diverse disciplines, this paper introduces a culture-gene coevolutionary approach to human
behavior and psychology, and applies it to the evolution of cooperation. After a general discussion of cooperation in humans,
this paper summarizes Dual Inheritance Theory and shows how cultural transmission can be brought under the Darwinian umbrella
in order to analyze how culture and genes coevolve and jointly influence human behavior and psychology. We then present a
generally applicable mathematical characterization of the problem of cooperation. From a Dual Inheritance perspective, we review
and discuss work on kinship, reciprocity, reputation, social norms, and ethnicity, and their application to solving the problem of
cooperation.
Ó 2006 Published by Elsevier B.V.
Keywords: Cooperation; Culture-gene coevolution; Reciprocity; Reputation; Punishment; Norms; Ethnicity; Cultural transmission; Dual inheritance
theory
It must not be forgotten that although a high standard of
morality gives but a slight or no advantage to each individual
man and his children over the other men of the
same tribe, yet that an increase in the number of wellendowed
men and an advancement in the standard of
morality will certainly give an immense advantage to
one tribe over another. A tribe including many members
who, from possessing in a high degree the spirit of patriotism,
fidelity, obedience, courage, and sympathy, were
always ready to aid one another, and to sacrifice themselves
for the common good, would be victorious over
most other tribes; and this would be natural selection.
At all times throughout the world tribes have supplanted
other tribes; and as morality is one important element in
their success, the standard of morality and the number of
well-endowed men will thus everywhere tend to rise and
increase.
­Charles Darwin (Descent of Man (1871, Chapter 5)).
The tendency to imitate may come into direct conflict with
the prudential teachings of pleasure and pain, and yet may
be acted upon. A child may do, and keep on doing, imitations
which cause him pain.
­James Mark Baldwin (Mental Development in the Child
and the Race (1906, Chapter 10)).
1. The why puzzle1
All around us we see people contributing to the welfare
of others, even when it is not convenient and may be
costly in terms of time or money, or may affect their personal
and professional relationships. In fact, we see so
much of this cooperation in daily life that we usually
don't notice it or stop to question why people are bothering
to help others, or how such seemingly ubiquitous
1389-0417/$ - see front matter Ó 2006 Published by Elsevier B.V.
doi:10.1016/j.cogsys.2005.11.010
*
Corresponding author. Tel.: +1 404 727 5248; fax: +1 404 727 2860.
E-mail address: jhenric@emory.edu (J. Henrich).
1
The paper synthesizes material drawn from our book, Why
People Cooperate (in press).
www.elsevier.com/locate/cogsys
Cognitive Systems Research 7 (2006) 220­245
cooperation might be explained. When asked why someone
cooperates, a common reply is that helping others
is ``the right thing to do'' and that people ``should'' help
each other. Some scholars have merely accepted such
cooperative acts as part of being human (Durkheim,
1933), without endeavoring to question why we cooperate
sometimes, but not others, or why different societies seem
to cooperate to differing degrees. In fact, not only are
there times that we don't cooperate when we know we
could have, there are many times when we don't even perceive
an opportunity for cooperation when one exists.
Taking this commonsense observation as a point of
departure, we will address the question of when and why
people incur personal costs in order to help another person
or group of people.
Even casual observation suggests some robust patterns
in cooperative social behavior. You've probably noticed
that people can be quite particular about who they will
help, when, and how much. First, people act frequently,
and sometimes at great cost to themselves, to help their
families, especially their kids. Why is that? People help
friends, and sometimes acquaintances, but there is
something different about `the rules' for helping these
people vs. helping close family. Friends who break the
`helping rules' often drop from `friend' to `acquaintance'.
Kids, on the other hand (even as adults), are not only
given more latitude, but are evaluated using quite different
`rules'. Why are friends different from family members,
and where do these `rules' come from? Moreover,
what about helping strangers? For instance, have you
ever considered why you would willingly stop to give
directions to a lost visitor on the street rather than
continuing onwards? Such an action wastes your time,
has some risks (the person may be a thief, con-artist
or murderer), and you'll likely never see that person
again. If you don't stop to help this stranger, why would
you feel bad about it? Are you more likely to help some
strangers than others? If so, who are you more likely to
help, and why?
Addressing such why-questions requires distinguishing
at least two levels of explanation. The first, the proximate
psychological level, focuses on understanding the psychological
processes and preferences that propel certain decisions
and behaviors. For example, how does the
psychology `loving your kin' work? Who qualifies as
`kin'? Are there different kinds of `kin' who get different
amounts of love or help? The second level, the ultimate
evolutionary level, explores the evolutionary processes
that produced the proximate psychologies that, in turn,
produce decisions and behaviors. If you are already
familiar with evolutionary theories that seek to explain
the kinds of social behavior mentioned above, you may
still find the following discussion interesting, as in our
view `evolutionary theory' includes both cultural
evolution and culture-gene coevolution. The framework
below integrates what we call the `canonical models' of
cooperation (e.g., kin selection and reciprocal altruism)
within a broader coevolutionary or dual inheritance
framework.2
This paper lays out a set of evolutionary theories,
derived from a single framework, aimed at explaining the
ultimate origins of different aspects of cooperative behavior.
Applying these ultimate theories, we are able to derive
predictions and set up research questions about the details
of the proximate psychological mechanisms for social
behavior and cooperation. Thus, dealing with the ultimate
why-question allows us to better address the proximate psychological
(and behavioral) questions. Simultaneously,
dealing with the proximate psychological issues provides
the only way to fully test our ideas about ultimate causes.
2. What do we mean by `cooperation' and `prosociality'?
Cooperation occurs when an individual incurs a cost in
order to provide a benefit for another person or people.
Costs include things that relate to genetic fitness like
resources (e.g., money, time, labor, and food). Throughout
our discussions we often refer to cooperative acts as
`giving help' ­ but cooperative acts are not limited to giving
help. Cooperative interactions take place within pairs,
small groups or large groups, and can occur among
friends, relatives or strangers. In pairs, cooperation might
involve babysitting, giving a friend a ride to the airport,
loaning sugar to a neighbor, or making dinner for a sick
person. Among large groups, examples of cooperation
include voting, participating in Neighborhood Watch,
recycling, contributing to public radio, sharing food,
and paying taxes. In these cases, a large group of people
benefit from the costly actions of an individual. To fully
understand why these behaviors qualify as cooperation,
let's look more closely at three of them: voting, food sharing
and recycling.
When a person goes to the voting booth on Election
Day, she incurs a cost. There is the time it takes to drive
or walk to the voting place, the time to vote, and the time
to return back to work (in the US, unlike many countries,
voting occurs on a work day instead of a holiday). There
may be a financial cost if the voter has to pay for parking
and/or gas, and there is the opportunity cost in that the
person could have been using this time to do something
else, like finishing a report for the boss or spending time
with her children. Benefits also occur from voting, namely
2
The coevolutionary approach championed here stems directly from the
research program laid out by Darwin in the Descent of Man (1871), and
subsequently developed by Baldwin (Baldwin, 1896a, 1896b). At the
opening of this paragraph, we wrote ``at least two levels of explanation''
because with the emergence of cultural capacities and cultural evolution, it
is sometimes useful to distinguish between levels of explanation: (1)
Ultimate level. Natural selection builds the psychological capacities for
cultural learning; (2) Intermediate level. Culture evolves, accumulates and
adapts non-genetically (intermediate level) to produce local skills, preferences,
and cognitive abilities; (3) Proximate level. Psychological mechanisms
that are the joint product of genetic and cultural evolutionary
history propel decision-making and behavior.
J. Henrich, N. Henrich / Cognitive Systems Research 7 (2006) 220­245 221
the support of the democratic system and the electoral
process ­ if no one voted, the system would collapse
and the government could be deemed illegitimate. All
members of the voting community share this benefit. Contrary
to what many voters believe, however, one of the
benefits of voting is not that their candidate of choice is
more likely to be elected because of the ballot that the
individual voter cast. Have you ever heard of a president
being voted into office by a margin of one vote? While it
is true that the democratic system would collapse if no
one voted, it is also true that any single vote has a negligible
effect. Therefore, when a person votes she performs
a costly act that helps the group (i.e., preserves democracy)
but does not help herself (since her candidate will
not be elected because of her vote), and she incurs the
costs associated with voting. Consequently, voting qualifies
as a large-group cooperative act.
Food sharing, also a cooperative act, can be found
throughout the world. Within our own society, food
sharing includes offering some of your meal to others
when eating at a restaurant, providing food and drink
to guests in your home, and donating non-perishables
to food banks. In hunting-and-gathering societies food
sharing is wide spread and important. Among groups
such as the Ache´ of Paraguay (Hill, 2002) and the Hadza
of Tanzania (Hawkes, O'Connell, & Blurton Jones,
2001), meat sharing is the norm, with hunters routinely
sharing their kill with the group. In sharing, the hunter
incurs a personal cost (his time, effort and the loss of
some of the meat that he and/or his family could have
consumed alone) while benefiting the rest of the group
by providing them with meat. Like voting, the distribution
of costs and benefits qualifies food sharing as a form
of cooperative behavior, which likely has deep evolutionary
importance.
Recycling is an excellent example of large-scale cooperation.
In the 1990s, vigorous campaigns were launched to
promote recycling of paper, glass, metal and plastic products.
In many North American cities, each household
received boxes for each type of recyclable material and
the city provided regular pick-up of these items. When
we recycle we incur several costs, including washing out
cans and bottles rather than tossing the dirty containers
directly into the garbage can, sorting our garbage by product
type, and taking a multitude of containers to the curb
on collection day (assuming we can even remember which
day each type of recyclable is collected!). Although this
process is not a huge burden, it certainly takes more time
and effort than simply throwing all our refuse into the
trash, and often requires special containers, more attention
and memory, and more household space. Now consider the
benefits from these individual-level costs: the planet and all
of its inhabitants get to live in a cleaner, healthier world.
What happens if you (one person) decide not to recycle?
Nothing happens. The contribution of any one person is
insignificant in terms of the planet's health, and you and
your kids still get to live in a cleaner world, as long as many
other people pay the costs of recycling.3
The benefits that
you personally create by recycling do not outweigh your
costs, and consequently any self-interested, rational individual
who weighed the costs and benefits of his actions
should refrain from recycling. Yet, if everyone did this,
the environment would suffer.
Recycling, like voting, is an example of a public goods
problem. In a public goods problem any given individual
has an incentive to refrain from cooperating even though
in the long run everyone will suffer from the loss of the joint
benefit. In public goods situations individuals can free-ride
by not cooperating while still reaping the benefits created
by the contributions of others. If you have ever been in a
group project in school, in which the group receives a single
grade for its efforts, you may have experienced free-riding.
In this context, free-riders avoid doing work but still reap
the benefits (the grade) of the others' work. In the recycling
example, a free-rider would be someone who chooses not
to recycle, but still enjoys breathing the cleaner air that
results from those who do.
Besides cooperative behaviors in which an individual
provides a direct benefit to others at a cost to herself, there
is a larger class of behaviors that we will call prosocial. The
best example of a prosocial behavior is what economists
call altruistic punishment (Fehr & Ga¨chter, 2002). Here,
an individual pays a cost to inflict a cost on another individual
in order to maintain an individually costly behavior
in a group. For example, in the gasoline crisis of the 1970s
there were long lines at the pumps. Occasionally, individuals
would attempt to free-ride by entering the line near the
front. Inevitably, this free-riding would infuriate at least
one person in the line who would often threaten, and if necessary
physically assault, the `line-jumper'. The presence of
the punishers no doubt dissuaded some individuals from
entering the line near the front. This altruistic punishment
is costly for the punishers, as he or she (it was usually a he)
risked getting beat-up, but benefits the group by maintaining
orderly lines.4
Non-punishers in the line reap the benefits
created by punishers without having to pay the cost of
wrestling with frustrated and enraged line-jumpers. As we
will discuss more below, theoretical models have repeatedly
shown that if individuals are willing to punish others at a
cost to themselves, some otherwise puzzling forms of cooperation
can be explained. If there are altruistic punishers
out there, anyone who can learn will fall into line, even if
he is completely selfish. Furthermore, experimental evidence
clearly shows that, at least in some societies, people
are willing to punish anonymous strangers at a cost to
3
We realize that there is some debate about whether recycling, when all
the costs and effects are taken in account, will actually achieve these
beneficial ends. However, as will be come clear below, social norms can
potentially maintain any behavior, even ones that don't yield overall
benefits.
4
Those who doubt the group benefit created by orderly lines have not
spent much time in countries where it is not the custom to make orderly
lines for scarce resources.
222 J. Henrich, N. Henrich / Cognitive Systems Research 7 (2006) 220­245
themselves (Henrich et al., 2004). Thus, our term prosocial
encompasses both cooperation as we described above
(`helping'), and altruistic punishment. We have avoided
extending `cooperation' to cover altruistic punishment
because of the lack of fit with common intuitive understandings
of `cooperation'.
3. Why is it so hard to explain cooperation and prosociality?
As we saw above, cooperation and altruistic punishment
always involve a cost to the cooperator or punisher. This
led many evolutionary and rational choice scholars to ask
the natural question: if cooperation is costly to the individual,
why does anyone do it? Evolutionary biologists like
Dawkins (1976) emphasize the logic through which natural
selection produces ``selfish genes''. The logic suggests that
genes that, on-average, cause their bearers (the individuals)
to pay fewer costs and reap more benefits relative to others,
are the ones more likely to be transmitted into the next gen-
eration.5
If a person has a gene that leads her to incur costs
to help other individuals, (i.e., provides benefits to these
people), then this individual will, on-average, produce
fewer offspring than individuals who do not possess these
genes. Taken at face value, this verbal reasoning indicates
that cooperation will generally be selected against by natural
selection, and that cooperation ought to be rare, both in
humans and throughout nature.6
Let's illustrate this with a more concrete, and ethnographically
relevant, example. Consider a person with a
gene that leads her to engage in sharing food.7
By sharing
her food with the group, she is increasing the fitness of
everyone else in the group by providing them with extra
calories and nutrients. However, at the same time she is
lowering her fitness and the fitness of her offspring by taking
away food from them and giving it to others in the
group. All else being equal, her generosity will result in
her rearing fewer healthy offspring to adulthood than a
person who avoids sharing. The generous food sharer will
likely have some daughters who also share food with the
group, and these daughters, like their mother, will have
fewer offspring than the non-sharers in their group. As
you can see, in each generation, the frequency of the `sharing
gene' will decrease because of the behavior it promotes.
Even if `food sharing genes' were initially very common,
they would gradually disappear from a population over
many generations. Nevertheless, we see lots of food sharing
in the world.
While the broad thrust of theoretical evolutionary biology
suggests that cooperation and prosociality should generally
be rare in nature, there has arisen in the last 40 years
a vibrant and growing set of theoretical models (theoretical
= mathematical) that have demonstrated a variety of
evolutionary pathways to cooperation. These pathways
are not mutually exclusive solutions to the dilemma of
cooperation, and different pathways may cross in ways that
either facilitate more cooperation or debilitate it by creating
conflicts. Any particular organism may have evolved
to make use of one, two, all of them, or none of them.
Some of the pathways or `classes of models' that we discuss
below are applicable to an enormous range of species,
while others are premised on a heavy reliance on high fidelity
cultural learning, and thus may be restricted to humans.
The second half of this paper provides an introduction to
each of the models, and begins to track down their empirical
entailments. Before we turn to the models, we first need
to layout some background information on social learning
and evolutionary psychology, as these concepts underlie
important components of subsequent discussions.
3.1. The puzzle deepens for our species
The puzzle of cooperation is both more interesting and
more enigmatic for our species. This is because the nature
of human cooperation, while similar in some ways to the
patterns observed in other species, is quite different in several
key respects. At a macro level, human cooperation varies
substantially from non-human primates in both its scale
and the nature of its variability. While the scale of cooperation
in other primates rarely exceeds two or three individuals
(e.g., in grooming and coalitions), humans in some
societies, including many hunter-gatherer societies, cooperate
on scales involving hundreds, thousands or even more
(e.g., war, voting, recycling, and exchange networks). However,
it is not merely the scale of cooperation that is different
from all non-human mammalian species, but also the
degree of variability across social groups. The scale of
human cooperation varies dramatically across social
groups, from societies that are economically independent
at the family level ­ showing little cooperation outside
the extended kin circle (e.g., Johnson & Earle, 2000: Machiguenga,
Shoshone) ­ to the vast scales found in chiefdoms
and modern states (Richerson & Boyd, 2000). While ecological
factors are certainly part of the explanation for this
variation, substantial degrees of variation in the scale of
cooperation can be observed among social groups inhabiting
identical environments (e.g., Atran et al., 1999; Itza
Ladinos and Kekchi; Kelly, 1985: the Nuer and Dinka).
Moreover, historical sources show that the scale of cooperation
in many societies has increased by orders of magnitude
in historical time (Diamond, 1997), thereby
indicating the presence of some non-genetic evolutionary
process that has been ratcheting up the scale of cooperation
5
This assumes that these `costs' and `benefits' translate in some manner
(however weakly) into differences in survival and reproduction.
6
Economics has also sought to explain the puzzle of cooperation.
Interestingly, however, the reasons why the puzzle has arisen in Economics
has more to do with the disciplinary tradition of assuming individuals are
self-interested rather than any prima-facie deductive logic. Other than its
heritage in Enlightenment philosophy, there is no reason (that we know
of) why Economics has typically assumed pure self-interest. Smith (2000),
for example, wrote eloquently about the importance of moral sentiments.
7
The assumption of a `single gene' is a rhetorical and modeling
convenience and does not substantially impact the outcomes of these
evolutionary arguments.
J. Henrich, N. Henrich / Cognitive Systems Research 7 (2006) 220­245 223
(a process that has not been observed in other species).8
Finally, while primate species typically show little variation
in the behavioral domains of cooperative behavior, human
social groups vary substantially in their domains of cooperation.
For example, some groups cooperate in fishing, but
not house-building or warfare, while other groups cooperate
in house-building and warfare, but not fishing. A complete
approach to the puzzle of human cooperation needs
to be able to explain these patterns in a manner that links
humans to the rest of the natural world, but at the same
time explains our distinctiveness (Henrich, 2004; Henrich
et al., 2003).
3.1.1. Intelligence is not the answer
Because the scale of human cooperation is so much
greater than that found among other mammals, and particularly
other primates, there is a common intuition that
human cooperation must result from our `superior intelligence'.
We believe this intuition is likely wrong for both
theoretical and empirical reasons. First, a substantial
amount of theoretical work in Economics, Anthropology
and Biology shows that more intelligence usually leads to
less cooperation, not more. Cognitive capacities for strategic
thinking that include planning for the future, storing
data on past interactions and more accurately assessing
potential costs and benefits do not lead to more cooperation,
as many people think (more on this below in the reciprocity
subsection). In a complicated world with imperfect
information, the skills of deception, deceit, trickery and
manipulation, which are improved by some kinds of intelligence,
are more powerful at destroying cooperation than
are capacities for tracking past interactions and preserving
it. There are always many easier ways to break something
fragile than to protect it.
Second, empirically the `intelligence hypothesis' does
not lead to the kind of cooperation that characterizes our
species. As just noted above, the scales of human cooperation
vary dramatically across social groups and domains
(even when groups inhabit the identical environment),
and have changed over historical time. It is difficult to
see how the `intelligence hypothesis' could explain these
fundamental patterns. All humans are smart, but some
human societies actually cooperate less, and live in smaller
groups than non-humans, while other human societies
cooperate on massive scales. Moreover, scales of human
cooperation have increased dramatically over the last
5000 years in many societies, but we have little reason to
believe this dramatic increase resulted from genetic changes
influencing human `intelligence'.
Finally, we get insight on the relationship between `intelligence'
and `cooperation' by looking comparatively at
other cooperative species. Besides humans, the next best
cooperators in the animal kingdom are the eusocial insects
(bees, wasps, ants, etc.). These critters manage to achieve
massive levels of cooperation ­ only achieved by humans
in historical time ­ with very few neurons per individual.
There are many species that have substantially more neurons
(by several orders of magnitudes) than eusocial insects
­ including all primates ­ but all of these species cooperate
less than these insects. There is apparently no necessary
relationship in nature between intelligence and cooperation.
Thus, other than a strong intuition, there is little to
support the `intelligence hypotheses'.
4. Culture is a part of evolutionary theory and human biology
Since the rise of human sociobiology in the 1970s, culture
and biology or cultural explanations and evolutionary
explanations have often been opposed, and the seeming
opposition between the categories has led to a great deal
of unnecessary dispute and debate. This dichotomy, and
the associated arguments, are outmoded and unproductive.
A wide range of human behaviors, which most
would think of as purely cultural (dress, greetings, food
taboos, etc.), are actually 100% cultural and 100% genetic.
Behaviors are cultural in that they are socially learned by
observation and interaction in a social group. All culturally
acquired behaviors, beliefs, preferences, strategies,
practices (hereafter, we refer to all these collectively as
`traits') are also genetic in the sense that their acquisition
requires brain machinery that allows for substantial
amounts of complex, high-fidelity social learning. We
know that there are `human genes' that allow for cultural
behavior, as chimpanzees raised (enculturated) alongside
human children do not acquire anything approaching
adult human behavioral patterns or social norms.9
In general,
our species' social learning capacities far outstrip all
others (we are a hyper-cultural species), and this capacity
can best be understood as a genetically-evolved adaptation
for acquiring adaptive traits in complex, variable
environments (Boyd & Richerson, 1985). For example,
if you were born into a band of hunter-gatherers, it would
likely be more adaptive for you to simply copy how the
other members of your group make bows and arrows
than to individually re-invent all the complex details of
the manufacturing process (Henrich, in press). If you
can take advantage of the experience of other hunters,
and possibly the wisdom of previous generations that
has been passed down through traditional practices, then
natural selection will favor social learning, a.k.a. cultural
capacities. With regard to cooperation, it doesn't take
much ethnographic reading to realize that understanding
social learning is central to figuring out how, when, with
whom and how much people cooperate. Because of its
relevance to cooperation, we briefly explore some of the
8
We can rule out genetic evolution for these changes in social behavior
because the time periods of these historical changes are too short.
9
Interestingly, one experiment of this kind was discontinued because,
while the chimpanzees had ceased learning much human behavior from his
adopted family, the chimp's human brother had started acquiring many
chimp-behaviors via imitation.
224 J. Henrich, N. Henrich / Cognitive Systems Research 7 (2006) 220­245
evolutionary theory underlying social learning before continuing
with theory more directly linked to cooperation. In
doing this, we put culture under the umbrella of Darwinian
Theory.
5. The evolution of cultural capacities and cultural evolution
Dual Inheritance Theory allows culture to be fully incorporated
into evolutionary theory. The approach can be
summarized with three key ideas:
(1) Culture, cultural transmission and cultural evolution
arise from genetically evolved psychological adaptations
for acquiring ideas, beliefs, values, practices,
mental models, and strategies from other individuals
by observation and imitation. Below, we summarize
how evolutionary theory has been used to predict
the psychological details of these cultural learning
cognitive capacities.
(2) These psychological mechanisms for social learning
led to behaviors that were, on-average, adaptive in
the varying ancestral environments that characterized
our human lineage. Any particular individual's
behavior or group's cultural practice may be adaptively
neutral or maladaptive. By specifying some of
the psychological details of these cultural learning
abilities (see #1 above), cultural evolutionary models
enable us to predict the patterns and conditions of
maladaptation, and thus provide theories of both
adaptation and of maladaptation (Boyd & Richerson,
1985, chap. 7). This is an advantage over the
models traditionally used in sociobiology and human
behavioral ecology, in which a behavior is either
`adaptive', or inexplicable.
(3) The emergence of cultural learning capacities in the
human lineage creates population processes that
change the selective environments faced by genes.
For example, suppose the practice of cooking meat
spread by imitative learning in ancestral human populations.
In an environment of `cooked meat', natural
selection may favor genes that shorten our energetically
costly intestines and alter our digestive chemistry.
Such a reduction of digestive tissue may have
freed up energy for more `brain building'. In this
way, human biology is adapting to culturally transmitted
behavior. The interactive effect is called culture-gene
coevolution. As discussed below, this
interaction may be critical for understanding some
aspects of human cooperation, particularly largescale
cooperation among non-relatives (Baldwin,
1896a; Boyd & Richerson, 2002b; Durham, 1991;
Henrich, 2004; Richerson & Boyd, 1998, 2000).
Below we expand on each of these ideas in greater detail,
although for a complete understanding of this approach to
cultural evolution, readers should begin with Henrich and
McElreath (2003) or Richerson and Boyd (2005).
5.1. Evolved psychological mechanisms for learning culture
The approach of understanding culture using evolutionary
theory begins by considering what kinds of cognitive
learning abilities would have allowed individuals, in
the changing environments of our hunter-gatherer ancestors,
to efficiently and effectively extract adaptive ideas,
beliefs, and practices from their social worlds. This
approach diverges from mainstream evolutionary psychology
in its emphasis on the costly information hypothesis
and the evolution of specialized social learning mechanisms.
The costly information hypothesis focuses on the
evolutionary tradeoffs between acquiring accurate behavioral
information at high cost (and less flexibility) and
gleaning less accurate information at low cost (and
greater flexibility). By formally exploring how the costly
information hypothesis generates trade-offs in the evolution
of our social learning capacities, we can generate predictive
theories about the details of human cultural
psychology (Henrich & McElreath, 2003). When information
is costly, natural selection will favor cultural learning
mechanisms that allow individuals to extract adaptive
information, strategies, practices, heuristics and beliefs
from other members of their social group at a lower cost
than through alternative individual mechanisms (like trialand-error
learning). Human cognition probably contains
numerous heuristics, directed attentional mechanisms
and biased inferential tendencies that facilitate the acquisition
of useful traits.
Such cultural learning mechanisms can be categorized
into (1) content biases and (2) context biases. Content
biases, or what Boyd and Richerson (1985) have called
direct biases, cause us to more readily acquire certain
beliefs, ideas or behaviors because some aspect of their content
makes them more appealing. For example, imagine
three practices involving different additives to popcorn:
the first involves putting salt on popcorn, the second favors
adding sugar, and the third involves sprinkling sawdust on
the kernels. Innate content biases that affect cultural transmission
will guarantee that sawdust will not be a popular
popcorn additive in any human societies. An innovative
company may try to market sawdust popcorn, but it is
unlikely to spread as long as salt and sugar are out there
as alternatives. Both salt and sugar have innate content
biases for sensible evolutionary reasons.10
Of course, if
you grew up in a society that only salts its popcorn, you
may steadfastly adhere to your salting preference even once
you find that sugar is the standard popcorn additive in many
places. Many such innate biases may have evolved because
they facilitate the acquisition of fitness-enhancing cultural
traits. However, content biases may also be a byproduct
(not specifically selected for) of other psychological processes,
10
Foods with salty or sugary flavors were likely both important sources
of scarces nutrients and calories in ancestral human environments, and in
short supply. Thus, natural selection should favor a bias to acquire a taste
for salty or sweet foods.
J. Henrich, N. Henrich / Cognitive Systems Research 7 (2006) 220­245 225
or the product of cultural transmission.11
Because content
biases are likely numerous and generally confined to particular
domains of culture, we omit further discussion of them
here.
Context biases, on the other hand, guide social learning
by exploiting cues from the individuals who are being
learned from (we will term these individual `models'),
rather than features of the thing being learned (the cultural
trait). There is a great deal of adaptive information embodied
in both who holds ideas and the commonness of the
ideas or practices (i.e., the number or percent of people
who have the trait). For example, because information is
costly to acquire, individuals will do better if they preferentially
pay attention to, and learn from, people who are
highly successful, particularly skilled and well-respected.
In fact, this is true even if the person's skill or success is
not directly connected to the behavior, belief or practice
in question. A large amount of mathematical modeling
effort has been expended in exploring the conditions under
which different context biases will evolve, how they should
be constructed psychologically, and what population patterns
will emerge from individuals using such learning
mechanisms. Moreover, and perhaps more importantly, a
vast amount of field and laboratory data confirms that
these learning biases are indeed an important part of our
cognition. Our remaining discussion of psychological
mechanisms focuses on two categories of context biases
in cultural learning: (1) success and prestige bias and (2)
conformity bias.
5.1.1. Success and prestige bias
If solutions to the problems of survival are tough to figure
out on your own, but you can imitate others, who
should you imitate? If possible, you should imitate people
who are winning at the game of life ­ that is, successful
people. More precisely, if individuals vary in skills (e.g.,
tool making), strategies (e.g., tracking techniques), and/or
preferences (e.g., for foods) in ways that affect fitness,
and at least some components of those differences can be
acquired via cultural learning, then natural selection will
favor cognitive capacities that cause individuals to preferentially
learn from more successful individuals. The greater
the variation in acquirable skills among individuals, and
the more difficult those skills are to acquire by individual
learning, the greater the pressure to preferentially focus
one's attention on, and imitate, the most skilled individuals.
In general, this prediction has been largely confirmed
with data from across the social sciences: people are powerfully
motivated to (unconsciously) imitate skilled and
successful people (Henrich & Gil-White, 2001).
The problem of figuring out what things about a person
to imitate is trickier than it might appear at first glance.
Knowing that an individual is more successful than others
does not tell the learner which of an individual's many
traits are responsible for the success. Is a businesswoman
successful because of her interactional style, religious
beliefs, professional attire, workout regime, vitamin preferences,
or work ethic? Her success could be attributable to
any, most, or all of these. Because of this ambiguity, the
theory suggests that humans have likely evolved a propensity
to copy a wide range of cultural traits from successful
individuals, only some of which may actually relate to the
individuals' success (Henrich & Gil-White, 2001). One outcome
of such a copying bias is that many neutral and some
slightly maladaptive traits can hitch-hike along with adaptive
cultural traits. For example, many successful business
people, among other things, work long hours at the office
and carry a leather briefcase. A person who wants to be
successful at work may copy both of these practices, as well
as many others. Working long hours is probably related to
the model's success whereas carrying a leather briefcase
may be a neutral trait. Through this imitative process, generations
of business people carry leather briefcases even
though it doesn't increase their success at work. In a world
of costly information, cognitive adaptations don't always
produce adaptive behaviors from the point of view of genes
and the theory allows for predictions about the conditions
under which neutral and maladaptive cultural traits will
spread.
The predictions derived from this approach have already
been confirmed by a variety of evidence from across the
social sciences. Adults and children do preferentially imitate
(often unconsciously) successful and skilled individuals,
and they imitate them in a variety of ways, even in
areas well outside of the person's domain of success. Henrich
and Gil-White (2001) summarize much of the general
evidence in this regard, and experimental economists have
clearly demonstrated that people readily use imitation to
figure out how to behave in social interactions involving
cooperation and competition, especially when payoffs are
on the line (Selten & Apestegula, 2002).
5.1.2. Conformist transmission
What do you do when observable differences in behavior
among individuals do not covary with the observable differences
in success and prestige? For example, suppose
you're a farmer and everyone in your county uses chemical
pesticides, except one farmer who uses natural pesticides
and obtains fairly average yields. Do you adopt chemical
or natural pesticides? One solution for dealing with such
11
In thinking about content biases, it is important to keep in mind a
number of things. First, jury-rigged evolutionary products, like human
minds, are likely to contain accidental by-products and latent structures
that create biases for fitness-neutral behaviors, ideas, beliefs and values.
Boyer (2001) details one kind of by-product content bias in his
explanation for the universality of religious concepts (like ghosts). Second,
even content biases that arose because they led to the adoption of fitnessenhancing
behavior in ancient environments may now promote the
adoption of quite maladaptive practices. Third, content biases may be
either reliably developing products of our species-shared genetic heritage
or they may be culture specific. People may learn valuable content cues via
cultural learning. Then, having acquired this idea or practice via cultural
transmission, they may be more likely to acquire another, because the two
``fit together'' in some cognitive or psychological sense.
226 J. Henrich, N. Henrich / Cognitive Systems Research 7 (2006) 220­245
information-poor dilemmas is conformist transmission:
copy the cultural traits of the majority (Boyd & Richerson,
1985; Henrich & Boyd, 1998; Kameda & Nakanishi, 2002).
Conformist transmission allows individuals to aggregate
information over the behavior of many individuals.
Because a person's traits implicitly contain the effects of
his individual experiences and social learning (including
his prestige-biased transmission), conformist transmission
can be the best route to adaptation in information-poor
environments. To see this, suppose every individual
receives an unreliable (but not useless) piece of information
from the environment about the highest yielding practice
for the current circumstances. This information, for any
one individual, might give that person a 60% chance of
noticing that chemicals yield slightly larger returns than
natural pesticides. Thus, using individual learning alone,
individuals will adopt the more efficient farming practice
with a 60% chance. But, if an individual samples the behavior
of 10 other individuals, and simply adopts the majority
behavior, his chances of adopting the superior chemical
pesticide technology increase to, say, 75%. By aggregating
the partial information of other individuals, conformist
transmissions can improve an individual's chances of making
adaptive decisions.
Consistent with this theoretical work, a substantial
amount of empirical research shows that people do use
conformist transmission in a wide range of circumstances,
particularly when problems are complex or difficult to figure
out on one's own. This work reveals that humans have
two different forms of conformity that operate in different
contexts. The first, often called informational conformity,
matches theoretical expectations from models of conformist
transmission, and is used to figure out difficult problems
and results in people actually altering their private opinions
and beliefs about something. The second, often called normative
conformity, is conformity for the purposes of going
along with the group, and not appearing deviant. Under
this type of conformity, people alter their superficial behavior,
but often don't change their underlying opinions or
beliefs. We argue that the ultimate origins of this second
type of conformity can be explained by the evolutionary
process that we describe under the rubric of social norms,
punishment and prosociality, below.
It is now fairly well established that cultural learning is
one of our primary means of behavioral adaptation
(Alvard, 2003; Tomasello, 1999). Our capacities for cultural
learning appear to be adaptations (products of natural
selection acting on genes) for acquiring useful
behaviors, practices and strategies in complex, information-poor
environments. These adaptations can be understood
as a set of specialized psychological mechanisms
(e.g., prestige and conformist biases) designed to extract a
vast array of useful information from other individuals.
While themselves part of our genetic evolutionary heritage,
they give rise to a second system of inheritance (culture)
that evolves in parallel, and in interaction with, our
genome. However, culture and cultural learning is only
part of the story. As will be clear in the following section,
we also believe that human psychology contains a number
of important non-cultural psychological adaptations that
deeply influence human social behavior and decisionmaking,
while interacting with ongoing cultural evolutionary
processes.
6. Evolutionary theories of cooperation and social
psychology
Here we discuss five evolutionary theories that provide
potential ultimate solutions to the dilemma of cooperation.
Our objectives are to provide the reader with an intuitive
understanding of how evolutionary theories can lead to
cooperation in humans, and to derive a set of proximate
psychological mechanisms and observable behaviors from
each theory. Our five classes of evolutionary models are
(1) kinship, (2) reciprocity, (3) reputation, (4) social norms
and punishment, and (5) ethnicity. In laying these out, we
will discuss both some of the fine nuances of how they
work (or fail to work) and the role played by culture in
both the phylogenetic and ontogenetic (developmental)
emergence of these forms of cooperation. Unlike other
introductory works on cooperation, one of our main
take-home theoretical messages is that all pathways to
cooperation require ­ in humans ­ some understanding
of cultural transmission, either in their operation, or in
both their evolutionary origins and their operation. For
example, all humans societies have both kinship (psychological
and behavioral biases to help genetic relatives),
and a culturally-transmitted kinship system that has important
effects on individual behavior, but cannot be fully
explained by kin selection. Because much of this theoretical
material is based on mathematical models that are beyond
the scope of this introductory paper, we hope to inspire our
readers to learn mathematics and game theory, and explore
the primary literature.12
6.1. The core dilemma in the evolution of cooperation
There is a simple core principle that underlies nearly all
solutions to the puzzle of cooperation: cooperation can
evolve under circumstances in which natural selection can
take advantage of a stable regularity that allows cooperators
to preferentially bestow their benefits on other cooperators.
In other words, cooperation can evolve when
cooperators tend to cooperate with other cooperators.
Eq. (1) expresses this condition in its most general sense
(see Henrich, 2004):
bb > c. 1
Here c is the cost paid by the cooperator in order to deliver
an amount of benefits, b, to another individual or group of
individuals. These costs and benefits are measured in units
12
Henrich (2004) and associated commentaries provides an entre´e into
the primary theoretical literature.
J. Henrich, N. Henrich / Cognitive Systems Research 7 (2006) 220­245 227
of fitness (e.g., number of offspring). b is a statistical relationship
called a regression coefficient. It measures the degree
to which `being a cooperator' predicts `bestowing
benefits on other cooperators'. In the simplest case, it is
the probability that a cooperator is bestowing benefits on
another cooperator. When this inequality is satisfied, natural
selection can favor the spread of genes that build proximate
psychologies for cooperative behavior. All solutions
(and their psychological incarnations) must be able to create,
or at least sustain, the statistical regularity that allows
cooperators to benefit other cooperators. Interestingly, in
the history of the study of the evolution of cooperation
(Frank, 1995, 1998), this simple equation was first derived
for specific cases (e.g., kinship and reciprocity) before this
more general, abstract condition was derived.
6.1.1. The rise and fall of green-bearded cooperators13
Once this general problem is understood, a simple solution
suggests itself. Imagine a gene that causes its bearer to
both `have a green beard' and to only help other greenbearded
individuals. In our equation above, b would be
at its maximum value of one (greenbeards only deliver benefits
to other greenbeards) and cooperation would spread
rapidly. Soon the entire world would be green-bearded
cooperators, and everyone would be merrily cooperating
with everyone else. However, the statistical relationship
(represented by b) between bestowing help and being a
cooperator must be reliable and durable. Now imagine
we are in a jolly world of green-bearded cooperators and
a mutant green-beard emerges. This fellow has the requisite
green beard, but he's not a cooperator. He never helps anyone,
but everyone always help him because of his lush,
green beard. Consequently, the mutant, and his mutant
gene, will be very successful, and will eventually drive the
green-bearded cooperators to extinction. The world now
consists entirely of green-bearded defectors. The trick to
solving the dilemma of cooperation is not so much in producing
a positive b value (i.e., in creating a way for cooperators
to find other cooperators), but in figuring out
how to maintain a reliable, stable, b value.14
Kinship provides
one set of avenues to cooperation.
6.2. Kinship15
Let's begin with one of the most prevalent forms of
cooperation in nature. Consider a mother with a gene (or
genes) that cause her to experience deep positive emotions
toward her offspring (e.g., baby, egg), and these emotions
cause the mother to give help to her offspring. Why would
this gene spread? Following the general logic outlined
above, it could only spread if the recipient of the `help'
was also a `helper'. That is, if the offspring is also a carrier
of the gene the leads to a psychology that evokes the rule
`help your offspring'. This leads to the question: What
are the chances that her offspring has the helping gene
given that the mother has the helping gene? The answer
can be rather complicated, but if the organism has a genetic
system like humans (diploid), then the chance should be at
least 50%. The main reason is simple. Mother's offspring
will on-average share 50% of her genes, which means that
there is a 50% chance that the offspring has a copy of the
mother's `help your offspring' gene. Thus, the characteristic
of `being one's offspring' is a reliable predictor (has
b = 0.5) of sharing the relevant cooperation gene(s). This
kind of evolutionary process is called `Kin Selection'
because mother and offspring (or any blood relatives) share
copies of the same gene by descent from a recent common
ancestor. In a sense, by causing the mother to help her offspring,
the gene is helping itself to survive and reproduce
down the lineage.
To illustrate the crux of the dilemma of cooperation,
suppose a variant of the `helper' gene just described
mutated into existence and caused its bearer to feel equally
affectionate towards all juveniles, not just her own offspring,
and thus to direct help toward whomever most
needed it. Could this mutant gene spread? No, because offspring
of the original gene, `help your own offspring',
would not only get help from its own mother, but also help
from other mothers carrying the mutant gene, while offspring
of mothers with the mutant gene would get only
minimal help from their own mom (who would be spreading
their help around), and no help from mom's carrying
the `help your own offspring' gene. This would, on-average,
lead the `help your own offspring' gene to outcompete the
`help everyone's offspring' gene.
Returning to the Core Principle, natural selection will
favor the evolution of psychological mechanisms that allow
cooperators to focus their benefits on individuals who are
likely to be cooperators. From this perspective, kinship
represents a class of different proximate mechanisms that
take advantage of the fact that some individuals in any
population will tend to share the same genes by descent
from a recent common ancestor (like a parent). This fact
of biology creates numerous evolutionary opportunities
for natural selection to find reliable statistical regularities
to exploit. `Blood relatives' share many other characteristics,
besides the all-important cooperation gene(s), that
natural selection can take advantage of in building psychologies
that help in directing benefits at other individuals
likely to share cooperation genes. For example, close kin,
such as siblings, may share a similar appearance or smell
that natural selection may use as cues to build a psychology
like `help those who look and smell like you'. These similarities
may be related to sharing some of the same genes (ones
not related to cooperation), or they may be related to
13
Dawkins (1976) coined the colorful ``greenbeard'' example.
14
Interestingly, `greenbeard solutions' constantly re-emerge in the
literature on the evolution of cooperation. Unknowingly, researchers
continually re-produce `solutions' to the evolution of cooperation that are
actually green-beards in disguise. Having now slandered the green-beard
solution, it must be noted that there is at least some evidence for greenbeards
in nature (Keller & Ross, 1998).
15
Key references include Hamilton (1964, 1972), Grafen (1985), Frank
(1997) and Queller (1992).
228 J. Henrich, N. Henrich / Cognitive Systems Research 7 (2006) 220­245
having been reared by the same female in the same environment
(e.g., nest). Even proximity can provide an evolutionary
opportunity to construct a psychological mechanism
capable of maintaining stable cooperation (e.g., `help those
who tend to hang around mom'). It is important to realize
that natural selection does not `care' about how a particular
cue arises (i.e., it does not care about kinship per se),
only that the cue is a stable and reliable predictor of who
is likely to be a cooperator. It so happens that in a variety
of species, many aspects of reproduction create reliable statistical
patterns on which natural selection can build. However,
the details always depend on the specifics of the
particular species. Natural selection might, for example,
lead to a system that causes newborns to smell those
around them during the first few days after birth, and subsequently
direct benefit towards those individuals for life.
This species must have a social structure that reliably
places newborns among close kin, a physiology such that
either individuals or close kin produce distinct (distinguishable)
scents, and some way to exclude non-kin from sneaking
into the nest during those first few days or faking the
scent of other family members ­ the `exclusion' need not
be fool proof for some cooperation to evolve, but the better
it is, the more cooperation can evolve.
In studying how kinship can solve the dilemma of cooperation,
Hamilton (1964) derived the simple rule that now
bears his name (Hamilton's rule):
rb > c. 2
The reader should note the similarity between Eqs. (1) and
(2). As in (1), b is the benefits bestowed and c is the cost to
the bestower. Now, however, r replaces b and represents
the `coefficient of relatedness', which specifies both the
average proportion of genes shared by the two individuals
as a consequence of recent common descent, and more
importantly, the probability that the receiver of help shares
a specific gene (i.e., cooperative gene). The parameter r is a
special case of b that occurs when the foundation of cooperation
is built on kinship and an organism can, by some
means, direct benefits preferentially toward their kin.16
If we assume that by using a variety of these cues,
humans (and other animals) can assess their degree of relatedness,
then we should be able to predict who they will be
most likely to help. This suggests that individuals should
cooperate with relatives according to their `coefficient of
relatedness' (r), which is the probability that they share
the same gene by descent. In our species, siblings and parent-offspring
have the highest r at 0.50. Assuming only one
line of descent (no interbreeding) grandparents and their
grandchildren and half-siblings have the next highest at
r = 0.25. First cousins are related at r = 0.125. Using the
evolutionary logic of kinship, parents and their children,
and siblings, should cooperate a fair amount, and first
cousins may cooperate a tiny bit, but more distant relatives
have too little relatedness to cooperate with each other.
Among siblings, for example, the fitness cost (in terms of
survival and reproduction) must be less than half of the fitness
benefits delivered before cooperation would be
favored. To get an intuitive sense of this, consider to whom
you would give a kidney. In the US, 68% of kidney donations
come from kin, while less than one-half of 1% come
from anonymous strangers (and thousands die every year
waiting for kidneys).
This body of theoretical work predicts that humans will
likely possess a ``kin psychology'' that is designed to (1)
identify kin and (2) direct benefits toward close kin. This
should be particularly true when the costs of cooperation
are relatively high compared to the expected benefits to
the person helped. The available evidence suggests that
humans may use a variety of cues to assess kinship, including
physical similarity to themselves and other family members,
scent, proximity during youth (Wolf, 1995), food
sharing, and social learning (by watching and listening to
others) to `figure out' to whom they are related, and thus
who is more likely to be a `cooperator'. These cues affect
affective states or feelings that, in turn, influence (along
with many other factors) cooperative behavior and trust
with cued individuals. For example, in experiments measuring
trust using allocations of real money, researchers
have shown that people are more trusting of individuals
who resemble themselves. The researchers varied `resemblance'
by using an image-morphing technology to combine
an image of the subject with another person
(DeBruine, 2002). This suggests that physical resemblance
to self may cue affective states built by natural selection
to benefit kin. Using the same kind of technology, other
researchers have shown that men are more kindly disposed
towards babies that resemble themselves (Platek, Burch,
Panyavin, Wassserman, & Gallup, 2002; Platek et al.,
2003). This is important because men, unlike women, can
rarely be completely positive that they are a child's genetic
father. While not focused on cooperation, a variety of
other studies show that our evolved psychology uses scent
and early-life proximity to calibrate affect and target
behavior.
In our book on cooperation (in press) we show that
Chaldean immigrants in Detroit restrict their most costly
forms of cooperation to close relatives, thereby confirming
a standard finding from small-scale societies (Chagnon &
Irons, 1979; Cronk, Chagnon, & Irons, 2000). However,
this work yields two additional insights. First, while Chal-
16
It is a common misconception that kinship depends on sharing a
certain percent of the same genes, and that r gives the fraction of genes
shared by two relatives. This is false for two reasons. First, natural
selection will favor genes that direct benefits at identical copies of
themselves, not other genes, so percent of shared genes is theoretically
irrelevant. Thus, r should be thought of as the probability that another
individual has a copy of the `helping gene' given that the first individual
has it. Second, humans already share most of the same genes because we
are the same species. Sharing genes for building finger and blood vessels is
not important for understanding human cooperation. Thus, r does not
give the percent of shared genes; it does happen to correspond to the
percent of genes that are identical by descent from a recent common
ancestor. But, as we said, this is only relevant in that it may create reliable
patterns that natural selection can exploit.
J. Henrich, N. Henrich / Cognitive Systems Research 7 (2006) 220­245 229
deans have an explicit cultural ideology that all Chaldeans
are ``one big family'' and that everyone is related, no one is
fooled behaviorally and the lines of costly cooperation are
clearly drawn at the outskirts of the immediate family. This
finding bears on the claim made by some evolutionary psychologists
that large-scale human cooperation in contemporary
societies results from the faulty (maladaptive)
operation of a psychology designed for cooperating in
the kin-based societies that many believe characterized
human ancestral societies.17
At least among the Chaldeans
of metro-Detroit, there is no confusion about kin and no
evidence for misdirected cooperation, despite the fact that
they live in a large urban center. Thus, while kin psychology
can explain cooperation among closely related kin, it
does not explain cooperation in other contexts.
The Chaldeans also illuminate the issue of kinship and
cooperation in a second way. Our findings show kinship
and the cultural expectations about behavior (carried from
Iraq) combine to allow many Chaldean storeowners to
out compete non-Chaldeans economically. Moreover, it
appears that Chaldean cultural beliefs are currently shifting
toward models typical of other middle-class Americans.
The case material suggests that while kin psychology is
constant, adaptive cultural learning processes are gradually
shifting cultural beliefs, which is altering the details of `who
helps whom'. This shift has important economic, social
and educational impacts for Chaldeans. That is, culture
matters.
6.3. Reciprocity
Reciprocity represents another well-studied class of
potential solutions to the evolutionary puzzle of cooperation
(Trivers, 1971). Reciprocity works on the commonsense
logic of ``if you scratch my back or someone else's,
I'll scratch yours''. Individuals, by applying such tit-for-tat
reciprocal strategies, can preferentially associate themselves
with other cooperators, and thereby increase both their
chances of bestowing their benefits on other individuals
with the same genes, and by increasing their likelihood of
receiving benefits from others by putting themselves in
the company of other reciprocators.
Progress on reciprocity as the answer to the dilemma of
cooperation has differed from kinship in two important
(and often unrecognized) ways. First, despite literally hundreds
of papers on the topic (Axelrod & D'Ambrosio,
1994: shows 209 publications from 1987 to 1993), the theoretical
conclusions derived for the mathematical models
and computer simulations are substantially more ambiguous,
nuanced and qualified than those for kinship. Second,
the empirical evidence for reciprocity-based cooperation in
non-human species is scant (Hammerstein, 2003), especially
when compared to the evidence for kin-based cooperation.
Nevertheless, in our species, reciprocity consistently reemerges
from both ethnographic and experimental studies.
Below, we deal with these two aspects of reciprocity by
synthesizing the qualitative findings from a substantial
body of theoretical work and linking these to our broader
empirical efforts. Unlike most other general treatments, we
will set this work within the general framework of the Core
Dilemma, and integrate it with humanity's reliance on cultural
learning. This last connection allows us to explain
why reciprocity-based solutions are rare in nature while
being plentiful and diverse in human societies. Theoretically,
there are two take-home messages here: (1) reciprocity-based
solutions to cooperation are much less robust
than many scholars think, and (2) these `solutions' are so
prominent in humans because of ­ not despite ­ our
evolved cultural learning capacities. Empirically, we use
the qualitative insights developed below to generate a series
of predictions about how our evolved `reciprocity psychology'
works. In our book on cooperation, we bring a range
of experimental and ethnographic data to bear on many of
these predictions.
Our general theoretical framework shows that the evolution
of cooperation requires that benefits be preferentially
bestowed on cooperators. Above, we explained how kinship
can facilitate this by providing reliable `cues' for identifying
cooperators. Reciprocity-based solutions, however,
use different mechanisms to support cooperation among
non-kin. In this class of solutions, natural selection favors
individuals who can use the past behavior of other individuals
as an indicator of whether they are a cooperator or
not. If a person's past behavior suggests that he may be a
cooperator (regardless of what genes he carries!), then natural
selection should favor a psychology that promotes
bestowing benefits on that person. There are two reasons
for this selection pressure. First, an individual's past history
may act as a costly cue that he/she carries a `cooperative
gene'. By bestowing benefits on those with a history of
cooperation, cooperators may be able to preferentially
direct benefits toward others with the same gene. Second,
bestowing benefits on certain individuals can cause a return
flow of benefits back to the bestower. In this case the specific
genes carried by the other individual are irrelevant.
What matters is that bestowing benefits causes a return
flow of benefits back to the one with the `cooperation gene'.
There are two sources of information about past behavior
that might diagnose on whom one should bestow benefits.
These sources lead to the two commonly discussed forms of
reciprocity: (1) Direct reciprocity, based on direct, personal
experience with the other individual(s), and (2) Indirect reciprocity,
which involves getting information about potential
interactants by observing them with others, or by
gathering reputational information (culturally transmitted
information) about their past behavior with other
individuals.
17
This hypothesis for explaining large-scale human cooperation is the
kinship version of the ``Big Mistake Hypothesis'' (Boyd & Richerson,
2002b). Despite being repeatedly criticized on both empirical (ethnographic
and experimental) and theoretical grounds (Boyd & Richerson,
2002b; Fehr & Henrich, 2003), the Big Mistake Hypothesis remains widely
believed in many corners of evolutionary scholarship.
230 J. Henrich, N. Henrich / Cognitive Systems Research 7 (2006) 220­245
6.3.1. Direct reciprocity
The intuition behind direct reciprocity is simple: If you
help me, I will help you. If you stop helping me, I will stop
helping you. Direct reciprocity (DR) depends on direct,
ongoing experience between interacting individuals. To
illustrate, suppose Joe and Natalie are each small business
owners. Each business goes through an annual period of a
cash shortfall at different times of the year. If each business
can get through the period of cash shortfall then each can
make an overall profit, but if it cannot, the business makes
no profit. Getting through the period of scarcity requires
$100. If both Joe and Natalie give each other money when
the other needs it, both make $300. If neither gives money
to the other, both make zero profit. If Natalie gives Joe
money during his scarce period, but Joe decides not to give
Natalie any money during her scarce period, Joe will make
$400 ($300 plus the $100 he did not give out = $400) and
Natalie will lose $100 (zero profit minus $100 given to
Joe) (see Payoff Matrix, below). By `defecting' (not giving
the money), Joe does relatively better than Natalie (by
$500) in that year, and Natalie may go out of business,
leaving Joe's `business practices' to proliferate as people
see his financial success and copy his strategy. Over time
as more and more people act like Joe, no one will give
any money and everyone will go out of business. If Joe
and Natalie had sustained cooperation then the pair would
have done much better overall than they did when Joe
defected. Had they sustained cooperation, after 10 years
each would have $3000 profit. Instead, Joe made $500
the year he defected but never made anymore in subsequent
years because his trading partner, Natalie, went out of
business and lost $100. Over 10 years, their combined profit
was only $400.
Payoff Matrix
Prisoner's
dilemma
Cooperate
(give money)
Defect
(don't give money)
Cooperate 300, 300 100, 400
Defect 400, 100 0,0
To study the kinds of strategies or behavioral rules that
will succeed in maintaining cooperation through reciprocity,
theorists have formalized the above situations in an
abstract format called the Iterated Prisoner's Dilemma
(IPD). This mathematical formalization allows researchers
to systematically study the properties of various strategies,
and other variables, in solving the dilemma of cooperation.
In the IPD, individuals are paired at random, interact
repeatedly for some expected number of rounds (like Joe
and Natalie giving money to each other each year), and
receive payoffs based on their decisions and those of their
partner. Individuals using strategies that give them higher
payoffs relative to other strategies will increase in relative
frequency compared to those using strategies that yield
lower payoffs. The game matrix illustrates the payoffs,
and captures Joe and Natalie's business situation.
Some of the earliest work by Axelrod and Hamilton on
direct reciprocity using this approach suggested that simple
reciprocating strategies could generate long-term, stable,
cooperation (Axelrod & Hamilton, 1981). One of the central
analytical findings in Axelrod's book (1984), The Evolution
of Cooperation, shows us how repeated interaction
with the same player is merely another way to address
the Core Dilemma in the evolution of cooperation (also
see Maynard Smith, 1982). If we take a simple reciprocating
strategy that cooperates in the first round and then in
subsequent rounds does whatever the other player did in
the previous round (TFT = tit-for-tat) and assume that this
strategy (TFT) is common in a population, the condition
for the reciprocating strategy to remain common against
low frequency invading defectors is
xb > c. 3
Here, as above, b and c are the costs and benefits of cooperation,
but now x (replacing b from (1)) is the probability
that the interaction with the same individual will continue
to the next round. Essentially, x is a measure of how long
the cooperation can continue if both individuals keep
cooperating. In this way, prior interactions with the same
person in a repeated sequence of interactions provide a
`cue' about who to preferentially interact with, so as to receive
a flow of benefits. The longer the interaction continues
with the same reciprocator, the greater the amount of
cooperation that can be sustained, because the individuals
receive more total benefits.
While several of the general findings put forth by Axelrod
have held up well (we discuss those shortly), many of
these findings have not stood the test of time, and have
been over interpreted by others. Perhaps the starkest example
of this is the reciprocating strategy ``tit-for-tat'' (TFT),
which was thought to be the most robust and simplest
solution to the problem of cooperation. All the hoopla surrounding
TFT resulted in a number of misunderstandings,
and led a well-know evolutionary biologist to write in a
textbook that TFT ``is superior to all other [strategies] in
playing repeated games of prisoner's dilemma'' (Trivers,
1985, 391). Not only was such a statement not supported
by the existing work at the time, but soon after 1985 a long
parade of papers showed that TFT wasn't even `pretty
good' in many situations.18
For example, in environments
in which the information about the payoffs or behavior
of one's partner is less than perfect, or simply in larger
groups, TFT does not lead to very much cooperation and
there are much better strategies. In these and other situations,
there is a multiplicity of other ways to preserve cooperation,
but this diversity in itself creates an evolutionary
challenge that only culture can solve.
18
Readers interested in exploring all the ways in which TFT fails should
begin with papers by Bendor, Kramer, and Stout (1991), Bendor, Kramer,
and Swistak (1996); Boyd (1989, 1992); Boyd and Lorderbaum (1987);
Hirshleifer and Martinez-Coll (1988).
J. Henrich, N. Henrich / Cognitive Systems Research 7 (2006) 220­245 231
Overall, the important intuitive understandings about
reciprocity that have arisen since the mid-1980s suggest
that (1) reciprocity-based solutions are substantially more
complicated than many thought and (2) while these complicated,
contingent patterns lack the desired simple elegance
sought by many, these appear more consistent with empirical
patterns among human societies and across species.
Below, we synthesize what is known about reciprocity in
an effort to paint a coherent image about the evolution
of cooperation via direct reciprocity. Using this image,
we sketch the entailments for our evolved psychology and
human behavior.
On the positive side, while no simple strategy exists that
yield anything approaching a robust reciprocity-based solution
to the dilemma of cooperation, much of this theoretical
work suggests that some types of reciprocating
strategies can lead to substantial amounts of cooperation
in a wide range of circumstances, as long as group size
remains small (cooperation occurs in dyads, triads, etc.)
and interaction endures for a sufficiently long time (large
x). Yet, while some kinds of reciprocating strategies can
be successful in particular circumstances, the details of such
strategies vary tremendously. To give a sense of which
aspects of the cooperative environment influence the success
of different kinds of reciprocating strategies, we review
four key factors: group size, noise, the ecology of other
strategies, and variation in the length of interaction across
partners.
6.3.2. Direct reciprocity: does group size matter?
The success of TFT and other reciprocity-based strategies
in pairwise interactions (interaction between only 2
individuals at a time) in much of the early work led many
to assume that these 2-person findings would extend to
explain cooperation in larger groups ­ what are called nperson
cooperative dilemmas or public goods problems.
Rather than two individuals helping one another, the n
players in public goods games contribute to the benefit of
everyone in the group. If one assumes that individuals
occasionally make mistakes (defecting when they meant
to cooperate), then reciprocity-based strategies (like TFT)
do not generate cooperation in larger groups to nearly
the degree that they do in 2-person cooperative situa-
tions.19
The capacity for reciprocity to maintain cooperation
decreases geometrically with increasing group size.
At the sociological level, this line of theoretical work suggests
that we should not expect direct reciprocity to be
the primary factor in maintaining cooperation in large
groups. At the psychological level, this finding predicts that
human psychology should be motivated by direct reciprocity
only when the cooperative unit is small ­ dyads should
be the preferred group for direct reciprocity. From the perspective
of institutional design, our direct reciprocity psychology
will be most effectively employed when the
cooperative group in small and enduring.
6.3.3. Direct reciprocity: what if information about payoffs
or the other people's behavior is noisy?
Reciprocity-based strategies of all kinds are entirely reliant
on the information that people receive about the past
behavior of their partners. Given this, what if individuals
receive inaccurate information about what their partners
did, or what payoffs they received in previous interactions?
This is certainly a potential problem in the real world, as
information is often ambiguous or uncertain.20
Continuing
with the example from above, what if Natalie figured that
Joe had decided not to give her the money, but actually
Joe's check got lost in the mail? Joe tried to cooperate
but Natalie perceived him as a defector. Or imagine a case
in which your partner helps you move furniture for the full
6 hours that you wanted him, but due to some confusion,
you thought he was only there for 4 hours (and you helped
him 6 hours last time)? What do you do when he asks you
for 6 hours of help next time? Do you stop reciprocating
entirely (perhaps you think he's taking advantage of
you), or do you reciprocate only 4 hours (playing TFT)?
Or, do you help him for the full 6 hours?
Noise in this kind of information creates a tradeoff
between a strategy's PROVAKABILITY and a strategy's GENEROSITY.
PROVOKABLE strategies show a hair-trigger willingness
to stop cooperating as soon as a person has any indication
that her partner is not cooperating. GENEROUS strategies
reciprocate with more than they are given ­ so if you give
me 4 hours of your time (or at least I think that you do),
and I give you 6 next time, then I'm GENEROUS. TFT, for
example, is quite PROVOKABLE but not GENEROUS at all. This
is because a person using TFT immediately stops cooperating
when her partner appears to have defected. Being PROVOKABLE
is important to avoid being exploited by defecting
strategies that take advantage of noisy environments. TFT
is not GENEROUS because TFT gives back whatever its partner
appears to have given. Not being GENEROUS in a noisy
environment can lead to substantially reduced amounts
of cooperation, as misunderstandings lead to irredeemable
losses in cooperation.
To see this, suppose two TFT were interacting ­ let's call
them TFT1 and TFT2. On round one TFT1 gave 10 (the
maximum) to TFT2, but due to the noisy environment,
TFT2 thought he was given only 8. Then, in the next round
TFT2 gave 8 to TFT1, but TFT1 thought he had received
7, and he returned 7. Now, while some misunderstanding
can drive exchanges back up, the average exchange would
not be near 10 ­ actually, this setup leads to an average
exchange of 5 in the long run. In contrast, a GENEROUS
strategy can deal with this. Suppose two TFT + 2 strategies
19
The relevant theoretical findings were independently arrived at by
different researchers around the same time, using somewhat different
models (Bendor & Mookherjee, 1987; Boyd & Richerson, 1988; Joshi,
1987).
20
All of Axelrod's early work was in a ``noiseless environment'' in which
actors always had perfect information about their partners' past behavior.
232 J. Henrich, N. Henrich / Cognitive Systems Research 7 (2006) 220­245
(TFT + 2 gives back two more than it was given) are playing
in a noisy environment. If TFT + 2[1] gives 10 to
TFT + 2[2], but TFT + 2[2] believes that only 8 was sent,
he will still give 10 back to TFT + 2[1]. Careful analysis
has shown that this generous approach leads to more cooperation
in a noisy environment than a PROVOKABLE strategy.
However, the effectiveness of being GENEROUS depends critically
on what other strategies are lurking out there because
generous strategies that can maintain effective cooperation
in noisy environments are susceptible to exploitation by
crafty strategies that hide under the cover of noise and
exploit the GENEROUS strategies. If you are in a nice environment
with mostly reciprocating strategies, GENEROUS
makes the most of cooperation. But, if exploiter strategies
emerge in sufficient numbers, being PROVOKABLE is the only
defense. Interestingly, GENEROUS strategies can thrive if
there are a sufficient number of PROVOKABLE strategies to
keep the crafty defecting strategies at bay. GENEROUS strategies
like TFT + 2 bring out the best in PROVOKABLE strategies
like TFT. In noisy environments, TFT cooperates at
much higher levels with TFT + 2 than with another TFT
and TFT keeps lurking exploiter strategies at bay.21
Strategic
diversification can help immunize a population against
invading defectors and thereby promote cooperation by
reducing the inherent tradeoffs between being PROVOKABLE
vs. GENEROUS.
Besides favoring generosity, noisy environments can
also favor CONTRITION. Contrite or remorseful strategies
accept punishment (in the form of a defection), after they
have defected. CTFT (Contrite TFT) cooperates in the first
round and then reciprocates cooperation and defection in
subsequent rounds. But, after a mistaken defection by
CTFT, if CTFT's partner defects then CTFT will cooperate
in the next round despite its partner's defection. After
this, CTFT will return to playing TFT. This allows CTFT
to repair cooperative relationships that would otherwise be
destroyed by some kinds of errors (Boyd, 1989). For example,
if CTFT is interacting with TFT and CTFT accidentally
defects when he meant to cooperate, he will `accept'
TFT's defection on the next interaction and cooperate
one time after this. This cooperation will bring out cooperation
from TFT and the pair will return to cooperative
interaction. If two TFTs were interacting, all cooperation
would have ceased.
However, contrite strategies are susceptible to `error in
perception' (Boerlijst, Nowak, & Sigmund, 1997). With
these errors, there arises a mismatch between what the individuals
in the interaction believe happened. For example,
Natalie might believe Joe defected when, in fact, Joe cooperated
and believes he cooperated. Even if Joe is using a
contrite strategy, the pair won't be able to repair their relationship
because Natalie will defect in the next round, and
Joe won't understand why (because he thinks he cooperated).
Contrition cannot always save you from errors.
One final aspect of noisy environments is the potential
importance of remembering more than just the previous
interaction ­ having a longer-term memory for previous
interactions. While some kinds of memory can be an effective
means to sustain cooperation in noisy environments,
longer memories are not necessarily better and can actually
lead to less cooperation. Longer memory can lead individuals
to adapt themselves to the noise, rather than the
opportunities for cooperation available from the array of
other strategies in the ecology. Additionally, more memory
is cognitively costly, especially as a person's number of
partners increases (Bendor, 1987; Bendor, 1993). The
importance of memory in sustaining cooperation via reciprocity
is often greatly overrated. It has been frequently
assumed that humans cooperate more than other animals
because they can remember a longer history of interactions.
This is a case in which common intuitions fail in the face of
rigorous formal models.
6.3.4. Direct reciprocity: what if the length of interaction
varies across different individuals?
Under a wide range of conditions, reciprocating strategies
that lead to long-term cooperation are generally NICE.
That is, successful strategies generally cooperate in the first
round of an interaction. Non-NICE reciprocators, like suspicious
tit-for-tat (STFT) who defects in the first round and
then plays TFT, can often spread and drive out NICE reciprocators.
But, usually, STFT does not create much longterm
cooperation when it's common. Consequently, being
NICE should depend on cues about how long interactions
might go on, and whether there are other reciprocating
strategies out there to cooperate with ­ if everyone is a
defector, being NICE does nothing for cooperation. One
implication is that in a population that has individuals
who vary in how long they stay around (different x's),
and this difference cannot be easily detected, then non-NICE
strategies who wait until they are sure they are with a longterm
interactant before initiating cooperation can be the
most successful. By waiting, non-NICE strategies can effectively
target their cooperation at other long-term interactants.
This suggests that natural selection should favor
individuals who are (1) NICE to long-term interactants, (2)
non-NICE and non-cooperative to short-term interactants
when such individuals can be distinguished from longtermers,
and (3) initially wary (non-NICE) but eventually
cooperative when the population is a mix of long-term
and short-term interactants and the types cannot easily
be distinguished (Boyd, 1992).
It has been argued that humans, as a consequence of our
long history in small-scale societies, would not have experienced
enough ephemeral interactions that had important
21
These insights were gleaned from a variety of sources (Bendor, 1993;
Bendor et al., 1991; Bendor & Swistak, 1997; Boyd, 1989; Hirshleifer &
Martinez-Coll, 1988; Wu & Axelrod, 1995). Another interesting aspect of
generosity in noisy environments is that the most successful strategies
overall may actually lose to every other strategy in direct pairwise
competition. Similarly, a strategy that defeats every other strategy in
pairwise competition will often place dead last when a variety of strategies
are let loose together.
J. Henrich, N. Henrich / Cognitive Systems Research 7 (2006) 220­245 233
fitness consequences for our psychology to be calibrated to
differentiate short-term interactants from long-term interactants.
This argument leads to the claim that people
(unconsciously) assume that everyone is a long-term interactant.
That is, even though we can consciously recognize
that we aren't likely to meet someone again, we cannot
really think that way. This version of the ``Big Mistake
Hypothesis'' has numerous problems, but first among these
problems is that the ethnographic, ethnohistorical and
archaeological evidence indicates that people in small-scale
societies routinely have important (i.e., fitness relevant)
interactions that are short-term or one-shot. The Big Mistake
Hypothesis is rooted in a false, but widely believed,
anthropological myth about the nature of life in small-scale
societies. There is, in fact, every reason to believe that
ancestral human life involved fitness-relevant interaction
with short-term interactants (Fehr & Henrich, 2003).
6.3.5. Direct reciprocity: changing ecologies of strategies and
the real challenge of reciprocity
Above we have hinted that the success of a strategy
depends on the other strategies that are in the mix (exist
in the population). This turns out to be a remarkably general
property of reciprocity-based cooperation. For any
strategy one can devise, there is a combination of other
strategies that will destroy it! In a variety of situations it
has even been possible to prove (mathematically) that no
reciprocating strategy is safe and robust against other combinations
of strategies (Bendor, 1993; Bendor & Swistak,
1997; Boyd & Lorderbaum, 1987; Farrell & Ware, 1989;
Lorberbaum, 1994; Lorberbaum, Bohning, Shastri, & Sine,
2002). This means that successful individuals need to be
able to rapidly adapt their strategies to the changing balance
of other strategies in the population. This presents a
particularly prickly problem for genetic evolution and natural
selection to solve, since shifts in the social­ecological
balance may occur rapidly, even within the lifetime of an
individual. Because there are so many possible social ecologies,
individuals in many species (especially humans) are
likely to encounter numerous social ecologies that their
species has never encountered in the evolutionary past.
As we'll explain, we believe that the human adaptation
for cultural learning provides one of the few means through
which natural selection can meet this ­ the real challenge of
the evolution of cooperation via direct reciprocity.
6.3.6. Direct reciprocity: why is reciprocity-based
cooperation rare in non-humans?
The impact of small shifts in the ecology of other strategies
in the environment, as well as the importance of variables
like uncertainty and group size, have been
underappreciated by many evolutionary scholars. We think
that the right take home message from all this theoretical
work is that the genetic evolution of cooperation via direct
reciprocity is not a robust solution for most animals in
most complex exchange contexts. In a sense, the theoretical
work shows that there is no way to build an `all-purpose
reciprocity machine'. Even if one could somehow encode
the rules for several different reciprocity strategies into
the genes of an organism, there is a virtually infinite number
of strategies one would need. This creates a combinatorial
explosion of potential contingencies that would have to
be built into the genes.22
The situation is especially acute in
humans, given the variety of cultural and physical environments
in which humans operate.
Consistent with the above logic, the empirical record
shows little evidence for reciprocity in non-humans (Hammerstein,
2003), and the non-human examples that do exist
are relegated to special case situations.23
In nature, and in
general, reciprocity is simply not a wide-spread and robust
solution to the dilemma of cooperation in the way that the
kinship solution is robust.
6.3.7. Direct reciprocity: enter cultural transmission
However, while this line of theoretical argument is consistent
with non-human data, we are left with a puzzle in
the human case. In contrast to non-humans, the available
ethnographic and empirical evidence shows that direct reciprocity
is a strong and recurrent pattern among our species,
and across cultures. Yet, interestingly, the details
and circumstances of direct reciprocity are highly variable
across contexts and human social groups (Fiske, 1991).
We argue that the combinatorial explosion is solved in
humans through cultural transmission. The adaptive nature
of cultural learning works with a psychology for reciprocity
(that coevolved with culture) to create `custom-fit'
strategies (solutions) that are dynamic and adaptable to
the challenges of reciprocity described above.
To understand how culture solves the combinatorial
explosion of reciprocity, it is important to recognize that
(1) cultural evolution occurs at much faster rates than
genetic evolution (e.g., novel practices can spread through
a population in a single generation: Rogers, 1995), (2) such
cultural learning processes allow populations to rapidly
adapt to novel situations without any genetic change,24
and (3) cultural transmission can, to a degree, construct
its own environment (e.g., written records reduce the
`noise' in payoffs between interactants, or institutional
structures can turn n-person public goods problems into
dyadic interactions). Genetic evolution will continually
22
Those familiar with evolutionary psychology might note that this is the
same `combinatorial explosion' logic used to defend the theoretical claim
that our cognition is composed of many special-purpose modules. Thus,
the same logic that leads one to believe the mind is modular also leads one
to be rather suspicious of arguments that humans have `reciprocity
psychology' that operates without adaptive cultural input.
23
We think the intuitive appeal of TFT clouded what should have been
an evident puzzle: if TFT is so simple and robust, why don't we see more
cooperation of this type in nature? Answer: TFT isn't that robust, so we
shouldn't see much reciprocity-based cooperation.
24
Most of human adaptation to physical environments is cultural. Blow
guns, kayaks, boomerangs, bone tools, and poison arrows are all
adaptations in the classic sense, and almost entirely culturally learned
(Henrich, in press).
234 J. Henrich, N. Henrich / Cognitive Systems Research 7 (2006) 220­245
introduce novel mutations (that lead to new strategies), but
cultural evolution will be able to counter such mutations by
building new culturally-evolved strategies, or altering the
social environment by building new forms of organization
(e.g., ID cards, police forces, neighborhood watch), at a
rate much faster than genetic evolution. Cultural learning
processes like prestige-biased transmission will allow individuals,
within their lifetime, to adapt their reciprocitybased
strategies to shifting social and physical ecologies.
The effect of cultural evolution on reciprocity-based cooperation
accounts for the empirical fact that humans in different
societies use reciprocity to differing degrees and in
different contexts. Our approach can account for both the
presence of local variation and the universal aspects of
human reciprocity.
While genetic evolution alone cannot provide a solution
to the real challenge of direct reciprocity, it can facilitate
the success of an individual by providing learning
mechanisms, biases, and default settings that allow individuals
to rapidly and effectively acquire locally useful
strategies. We suspect that our `reciprocity psychology'
is a product of the coevolution between genes and culture,
and rather than supplying rules like TFT, our psychology
provides the machinery for rapidly learning the locallysuccessful
rules of reciprocity from other people and from
experience. Following from our prior discussion of cultural
learning, by imitating successful individuals, people
can acquire the locally adaptive strategic nuances that
fit the local ecology of strategies, noise, etc. Projecting
back into evolutionary time, we might imagine that after
our capacities for cultural learning evolved genetically
then cultural evolution produced the first simple rules of
reciprocity (via imitation25
), in the same fashion as cultural
evolution produced early stone tool traditions. Following
on the heels of cultural evolution, genetic
evolution via natural selection might have favored `learning
genes' that allow nai¨ve individuals to more rapidly
acquire reciprocity-based `rules' (strategies). With these
new learning biases in place, cultural evolution may have
been able to solve an increasing number of cooperative
problems using a variety of different reciprocity-based
strategies in different contexts and places. The outcome
of this interactional process would be an evolved reciprocity-psychology
that operates through (at least partially),
and is dependent on, cultural learning to acquire the
adaptive strategies for social behavior. This evolved psychology
should assist individuals in acquiring effective
strategies, avoiding exploitation, and identifying good
long-term partners. In a sense, genetic and cultural evolution
have teamed-up to solve a variety of cooperative
dilemmas through the logic of direct reciprocity.
6.3.8. Patterns of direct reciprocity
Based on intuitions that we have built up from reviewing
these many theoretical findings on reciprocity, we now
have an idea of the general shape of this cognitive and
emotional adaptation. Because our evolved `reciprocity
psychology' coevolved with culture, most of what's below
can be understood as learning biases that assist individuals
in acquiring successful strategies in various ways.
Many aspects of these psychologies can be thought of
as learning biases that favor the acquisition of certain
strategies in some circumstances or as inferential rules
that assist individuals in figuring out what underlying
strategies a successful person might be applying ­ remember,
people only see the overt behavior of others, and they
must infer the underlying strategy in order to acquire it.
Such biases, in combination with other cultural learning
mechanisms like prestige-biased transmission and conformist
transmission, allow individuals to adapt and calibrate
their strategy to the local situation. While these
biases exist in the minds of individuals, they have a direct
impact on the kinds of reciprocity we should expect to
observe in groups. Here, we enumerate some of the psychological
predictions derived from the available theoretical
work.
(1) NICENESS. How nice should an individual be in the
beginning of a potentially longer-term cooperative
interaction? When groups are large (n > 5), the
expected length of the interaction is short or highly
variable, or the ratio of benefits to costs (b/c) is small,
individuals should tend to adopt SUSPICIOUS strategies
(strategies that aren't NICE in the first round, but can
be persuaded to cooperate by some NICE strategies).
Elsewhere (Henrich & Henrich, in press), we have discussed
how cultural evolution can facilitate NICENESS,
and thus promote sustained cooperation, by (1) turning
an n-person cooperation dilemma into a dyadic
situation, (2) guaranteeing that individuals will stick
around for the long run, and (3) using kinship to
transcend the life of an individual and extend the time
horizon into the future (using kinship to make x
large: families can cooperate over generations). We
will also explain how culture can solve other aspects
of NICENESS by setting standards for `how nice to be
on a first interaction' ­ without culture this is completely
unspecified, and there are no standards or
expectations.
(2) GENEROUS/CONTRITE VS. PROVOKABLE. How GENEROUS,
CONTRITE and PROVOKABLE should a strategy be in
the local environment and ecology? Generosity
should be favored in environments (contexts) and
ecologies that are both (1) noisy and (2) have a significant
proportion of reciprocating strategies and few
defectors. When little noise clouds the payments of
one's partners, or when defectors appear plentiful,
individuals should prefer strategies that are more
PROVOKABLE.
25
TFT is easily derived from imitation. It is simply `cooperate initially'
and imitate the other guy. Young children readily engage in the kinds of
imitative exchanges and behaviors of all kinds (Meltzoff & Prinz, 2002).
J. Henrich, N. Henrich / Cognitive Systems Research 7 (2006) 220­245 235
(3) MEMORY. How much of a partner's past behavior does
one remember and incorporate into her decisions
about reciprocation? Overall, individuals in noisy
environments should use selective forms of longerterm
memory (e.g., remember only big defections)
for only a small cadre of valued partners. For a large
set of partners, memory capacities will be stressed so
individuals should prefer strategies that require little
memory, but then restrict their interactions to low
noise environments and deploy more PROVOKABLE
strategies. Ethnographically, we have shown how culture
can establish different domains of cooperation in
the same social group. In those domains that involve
a large number of potential dyadic interactions and a
means of reducing the noise in payoffs, cultural evolution
favors fairly strict 1-for-1 in-kind exchanges,
which make interactions easily trackable and less
noisy, and allows short-term accounting for past
behavior. In other domains, such as those involving
a small number of individuals who make a wide range
of different kinds of exchanges (creating lots of noise),
favored strategies will involve generosity, contrition,
and some kind of longer-term accounting.
Table 1 summarizes the patterns described above. While
it seems more likely that the different parameters of reciprocity
strategies (e.g., NICENESS) are more accurately
thought of as continuous, Table 1's trichotomy provides
a condensed, heuristic presentation of the ideas just
discussed.
6.3.9. Indirect reciprocity and reputation
As mentioned above, reciprocity-based approaches to
cooperation can take two forms: direct and indirect reciprocity.
In direct reciprocity favors are exchanged directly
and repeatedly between individuals: A helps B and, in
return, B helps A. Under indirect reciprocity, individuals
interact with each other only occasionally (sometimes only
once), but now ­ before interacting ­ individuals receive
information about the past behavior of the individual with
whom they are about to interact. For example, A and B
interact and A defects on B (B cooperates); then A and C
meet, but because C knows (somehow) about A's past
behavior with B, C defects instead of cooperating (C is normally
a cooperator). If C had met B instead of A, C would
have cooperated because C knows that B cooperated. C's
knowledge of the behavior of A or B from previous interactions
corresponds to what we commonly refer to as reputation.
The incorporation of reputation, which provides
information about past behavior, generates a plausible
mechanism to explain cases of cooperation that involve
neither kin nor long-term interactions (Leimar & Hammerstein,
2001; Sugden, 1986; Wedekind & Milinski, 2000). As
you will see, the importance of reputation and the practicalities
of disseminating reputational information creates
a close relationship between two different pathways to
cooperation: indirect reciprocity (IR) and social norms
and punishment (SNP). In fact, SNP lays a foundation that
dramatically strengthens the power of IR.
6.3.10. Indirect reciprocity: solving the core dilemma, again
As discussed above, cooperation can evolve according to
the degree by which individuals can preferentially bestow
their benefits on cooperators. Theoretical work on indirect
reciprocity shows the Core Dilemma once again. This
research shows that reciprocating strategies can be maintained
against invading defectors when (Nowak & Sigmund,
1998a, 1998b):
/b > c. 4
As above, b is the benefit of the help contributed, c is the
cost of helping to the helper, and / is the probability that
the helper accurately knows the reputation of the helpee.
This simple equation, along with extensive theoretical
and computer simulation research, shows that the availability
of accurate reputational information is the key to
indirect reciprocity's ability to solve the puzzle of cooperation
(Panchanathan & Boyd, 2003). This means that, ceteris
paribus, variables such as the size of the cooperative
group (the number of individuals in any given interaction),
the population size (the number of individuals in the pool
of potential interactants), the density of social connections
Table 1
Strategies by type of partners and context
Categories of partners Context and ecology Psychology and behavior
``Close friends'' Substantial noise ­ exchanges across many domains NICE GENEROUS CONTRITE
High b/c
Longer memories of important interactions
Small number of individuals (memory constraints)
Distant friends, and other acquaintances Low noise ­ in-kind, 1-for-1 exchanges NICE PROVOKABLE
Medium b/c
Short memories of interactions
Potentially large # individuals
Others n-person dilemma (public goods situation) SUSPICIOUS PROVOKABLE
Short time horizon (low x)
Low b/c
236 J. Henrich, N. Henrich / Cognitive Systems Research 7 (2006) 220­245
between individuals in the population, and people's beliefs
about gossip will strongly influence the effectiveness of
indirect reciprocity. More specifically, the larger the cooperative
group, the less a cooperator will be able to direct his
help preferentially to individuals with good reputations.26
Likewise, the larger the population the less likely it will
be that an individual will have the necessary reputation
information about a particular interactant, the more bits
of reputational information each individual will have to
store in her memory, and the less accurate that information
will be. Less dense social networks imply that a cooperator
is less likely to have (or be able to get) accurate information
about a particular individual. Finally, culturally acquired
beliefs about how much to gossip, and how accurate that
gossip must be, can greatly influence the effectiveness of
indirect reciprocity.27
Individuals, for example, may be
punished by others for spreading inappropriate or false
gossip. Such punishment is a kind of altruistic punishment
in an n-person public goods situation, and is quite difficult
to explain without the social norms and punishment (and
their supporting psychological mechanisms) discussed
below. However, once this punishment of those breaking
the gossip norms is in place, much greater amounts of
cooperation can be explained by IR.
6.3.11. Cultural capacities vastly increase the potential for
indirect reciprocity
Our cultural capacities provide the means to increase
both the volume of reputational information available in
a population and the accuracy of that information. First,
part of our endowment of cultural capacities involves language,
and linguistic communication opens up the potential
for vast amounts of reputational information to
spread around (Smith, 2003). This is only part of the story,
as merely increasing the volume of information won't facilitate
the evolution of cooperation via indirect reciprocity,
unless that information remains highly accurate. In this
regard, it is important to realize two things: (1) individuals
will have incentives to inject false information into the system
to either hurt competitors (e.g., spread false rumors
about someone), or help themselves (e.g., spread good
rumors about themselves), and (2) as information flows
from one person to another accuracy will decreases as people
misremember, misunderstand and miscommunicate.
Given all this, how can IR persist at all? Cultural transmission
provides a partial rescue.
While not usually thought of in this way, acquiring reputational
information from other people about other people
is a kind of cultural learning. Consequently, cultural
learning mechanisms like prestige-biased transmission
and conformist transmission can improve the accuracy of
acquiring reputational information for the same reasons
that it helps in acquiring adaptive cultural behaviors. This
cultural transmission likely involves observing how other
individuals behave towards, and what they say about,
potential interactants. This includes both what people
might intentionally say about someone as well as how they
unconsciously behave toward, and speak about, them. For
example, prestige-biased transmission allows learners to
culturally acquire reputational information about potential
interactants from those individuals who are most likely to
have good information. For instance, it may involve giving
more weight to the opinions individuals who are particularly
successful in assessing the reputations of others. Alternatively,
it may involve learning from someone who has
spent the most time with the individual in question, or from
someone who is judged (by the learner) as best able to
accurately evaluate the individual in question.
Similarly, conformist transmission allows learners to
integrate reputational information received from many
people about a particular individual and acquire a more
accurate sense of the individual by `throwing-out' the outliers
(this helps remove the noise). For example, suppose
people judge the reputation of others on a scale from 0
to 100, and five people have mentally assigned scores of
72, 79, 82, 75 and 0 to ``Tom''. If a learner samples the
words and deeds of these five people with regard to Tom's
reputation, our learner might estimate the beliefs of the
observed judges as 75, 73, 79, 85, and 0. The learner estimated
these scores from observations of the five others.
Further suppose that the `true' reputational score of the
individual is 77, and the individual who provided the score
of 0 completely misunderstood something Tom did. A conformist
learner would throw out the 0 and the 85, and
assign the individual a reputation score of 76 (by averaging
the remaining 3 estimates). Under a wide range of conditions
this allows a better estimate of an individual's `true
reputation' than averaging across all observations equally.
Combinations of prestige and conformist mechanisms are a
potent means to improve the accuracy of the reputational
information that is the lifeblood of Indirect Reciprocity.28
This means that IR will likely only be a powerful force in
creating cooperation in highly cultural species (of which
humans are the only one known).
26
When a cooperative unit has more than two people (i.e., more than two
people are contributing to the provision of a shared good, like taxes or a
restaurant bill), it is impossible to withhold cooperation from a defector
without also withholding it from cooperators, and this causes all the
cooperation in the group to unravel. Here's why: When a cooperator
decides not to cooperate in response to a defection, he withholds
cooperation not only from the person who previously defected but also
from all the people who cooperated. The remaining cooperators then
perceive the punisher as a free-rider so they respond by withholding
cooperation from him. This process repeats itself until all the reciprocators
stop cooperating.
27
See Conte and Paolucci (2002) for an in-depth simulation-based
analysis of this issue.
28
Biases like `self-similarity', sex, age, and ethnicity can further hone the
accuracy of one's reputational information (on-average!). For detailed
models showing how conformist transmission can improve noisy information,
see Henrich and Boyd (2002).
J. Henrich, N. Henrich / Cognitive Systems Research 7 (2006) 220­245 237
6.3.12. Ethnicity and indirect reciprocity
When individuals encounter someone they don't know
and have never heard of (i.e., someone that they cannot
easily get reputational information on), what should they
do? Theoretical work on IR shows that they should be
SUSPICIOUS: that is, they should defect and see if the other
person cooperates. If the other does cooperate, they
should switch to cooperating with that person. This is
applicable to a variety of circumstances that might lead
some social networks to be more densely interconnected
than others. However, vis-a`-vis our empirical work and
the world more generally, these circumstances often occur
when members of an ethnic group tend to preferentially
interact mostly with each other, and less with outsiders.
Under such conditions, we would expect to see suspicious
behavior with non-co-ethnics (because they are outside
the reputational network), and nice behavior with co-ethnics.
While this observation is important, we do not
think it explains the `ethnic bias' in behavior and interaction.
Rather, it's a sociological pattern that arises from
individuals interacting with others who carry with them
both a psychology for indirect reciprocity and a bias
for interaction with co-ethnics (McElreath, Boyd, &
Richerson, 2003; Panchanathan & Boyd, 2003). Below
we lay out the evolutionary theory that predicts an `ethnic
bias'.
6.3.13. Costly signaling and indirect reciprocity
Are there any circumstances when an evolutionarily
successful indirect reciprocator should deliver help to (1)
a receiver who is not a known `cooperator', or (2) a group
in which the benefits will be diffused though the group
(some of whom may be defectors; this is the n-person
case)? The answer is `yes' (Engelmann & Fischbacher,
2002; Panchanathan & Boyd, 2003). This can arise if
cooperating (and making a big show of it) can have a sufficiently
positive impact on one's reputation in future
interaction so as to counteract the immediate costs of
helping others (even defectors). This `reputation building'
form of indirect reciprocity corresponds to another class
of evolutionary solutions to cooperation termed costly sig-
naling.29
In this context, individuals are signaling their
quality as cooperators to future interactants using reputational
effects. Such cooperative acts are especially useful
in situations where many members of a person's social
network are present and paying attention. Even if the initial
recipient doesn't reciprocate, the giver can have his
costs offset by the benefits he receives from the other people
who observed his giving and now give to him.
Humans should have a psychology that is geared up to
look for opportunities of `high broadcast value' in which
they can obtain big reputational benefits. As we have
shown elsewhere (Henrich & Henrich, in press), this effect
explains a substantial portion of helping behavior in
experimental settings and it appears to emerge, intact, in
the real world. Among the Chaldeans, for example, men
seize the opportunity to pay for everyone's dinner at a
restaurant and argue over who will have the chance to
pay. The important thing is that everyone knows that
they paid. This does not appear to be done with the conscious
intent of improving one's reputation; the person
truly believes he is being generous solely for the purpose
of doing something nice for others ­ but this does not
change the underlying evolutionary basis.
Another implication of this is that individuals should
`cheat the system' when they can by seeking out activities
and interactions that enhance their reputation at low cost
to themselves. For example, being part of a prosocial charity
enhances the reputation of its members. Inside the
group, however, free-riding in n-person public goods problems
reigns because (1) IR is not powerful enough to solve
n-person dilemmas within the group, and (2) outsiders to
the charity don't have access to information about the relative
contributions of individual members. Chaldean social
life provides a clear case study of this phenomenon.
6.3.14. Psychological and sociological implications
The available theoretical work on indirect reciprocity
allows for a series of general predictions about the nature
of human psychology and sociology. Psychologically, IR
predicts the following:
1. People should care about their reputation, especially
when behaving in the presence of members linked to
their own social networks.
2. When interacting with individuals not linked to their
social network and for whom they have no reputational
information, individuals should be SUSPICIOUS.
3. Individuals will prefer to be NICE to others for whom
they lack reputational information if those others are
part of their social network. For example, Chaldeans
should be NICE to other Chaldeans because they are
tightly connected in dense social networks.
4. Individuals should prefer cooperative interactions in
dyads or small groups (and will likely not cooperate in
larger groups), unless an opportunity presents itself for
`reputation-building'. This will occur mostly in public
contexts because they have the most broadcast value.
Sociologically, IR predicts the following:
5. Dense, bounded social networks that are stable through
time lead to the highest levels of IR-based cooperation.
Such networks decrease the chance of meeting someone
for whom an individual lacks reputational information
and increases the opportunity to `cross-check' information
and allow conformist transmission and prestigebias
to sort out the accurate reputational information.
Finally, they reduce the proportion of new `immigrants'
who lack sufficient reputational information.
29
We believe that this form of indirect reciprocity is equivalent to forms
of costly signaling in which individuals are signaling their value as future
cooperative interactants. There are other forms of costly signaling that
could lead to cooperative acts that are not part of IR (Bliege Bird, Smith,
& Bird, 2001).
238 J. Henrich, N. Henrich / Cognitive Systems Research 7 (2006) 220­245
6. The most effective forms of cooperation via IR will
involve dyads. Public goods problems won't be solved
by IR, except under certain conditions when broadcast
value is high.
7. Culturally-transmitted beliefs that extend a person's reputation
to her kin will facilitate cooperation and increase
conformity. Such beliefs have the effect of increasing
(sometimes drastically) the impact of having a bad reputation
by extending those effects to an individual's close
kin, which effectively extends them through time ­ e.g., a
son gets the reputation of his father. Thus, if a father is
considering damaging his reputation in some irresistible
defection or norm violation, he faces damaging not only
himself, but also his son and perhaps his son's son. Here
`cultural evolution' has taken advantage of our kin psychology
to enhance the power of IR.
6.4. Social norms, punishment and prosociality
Life in human social groups is regulated by social norms
that go beyond cooperation. Norms can be identified by
three characteristics (Henrich et al., 2003): (1) they prescribe
`proper' behavior for individuals within a population,
or some subset of the population (e.g., women must
wear veils), (2) these prescriptions are widely shared by at
least some important portion of the population, and (3)
failure to adhere sufficiently closely to these prescriptions
will anger other individuals in the population (even if the
action does not otherwise affect them), and these angry
individuals may take actions that are costly to both themselves
and the norm violator(s). Some examples from
anthropology and common experience include eating tabooed
food (e.g., pork), going nude at formal weddings,
having sex with one's parents, not cutting your lawn frequently
enough, telling jokes during a funeral services,
etc. None of the above theories can explain these panhuman
patterns. Why would uninvolved third parties care
about what others do, and why would they care enough to
take an action that is costly to themselves? And what does
this `norm stuff' have to do with human cooperation?
In thinking about norms, it is important to start with
individual minds and then aggregate up (using mathematical
models of social interaction and learning) to population-level
phenomena. At the individual level, norms
begin as sets of mental representations, stored in individual's
brains, which we commonly refer to as ideas, preferences,
beliefs, values and practices. These representations
prescribe both what their possessor should do in certain situations,
and what others should do. As prescriptions,
norms are often attached to powerful emotions (anger,
guilt and shame) and motivations that lead to strong reactions
when the individual, or others, violate a norm. These
reactions may lead to a variety of forms of punishment,
ranging from gossip to banishment to homicide.
At the population-level, these mental representations
(norms) are shared to some degree, for a variety of reasons,
by many members of the same social group. One important
reason is that cultural learning mechanisms (like conformist
transmission) will cause members of a social group to
adopt similar mental representations. Second, the punishment
evoked by norms will further lead individuals to
adopt practices (if not similar mental representations) to
avoid punishment. There are other reasons, but these two
classes of mechanisms alone will cause groups to share similar
mental representations (norms) about various kinds of
behavior. Some of these representations and behaviors
have to do with cooperation.
Social norms, despite being one of the most discussed
concepts in the social sciences (Bendor & Swistak, 2001),
have lacked a serious evolutionary explanation that can
account for their character and diversity. We think a firm
evolutionary explanation for social norms lies in the fact
that humans ­ unlike other animals ­ rely heavily on their
evolved cognitive adaptations for cultural learning to
acquire a large portion of their behavioral repertoire,
including their social behavior. When both adherence to
a norm and a willingness to punish norm violations are
influenced by cultural learning, the mechanisms of prestige-biased
and conformist transmission can lead to stable
situations in which most people acquire and follow the
rules, prescriptions and punishments associated with a
social norm. This applies to any norm, be it adaptive, neutral
or maladaptive, and includes norms for costly cooperation.
This body of formal theoretical work further
suggests that, while neutral and somewhat maladaptive
norms could be maintained within any particular group,
group beneficial norms can spread by competition and
selection across social groups that have different culturally
evolved norms that vary in their group-beneficial properties
­ a process termed cultural group selection. Furthermore,
if these competitions among groups with different
norms have been occurring for a long time (tens of thousands
of years), the theory shows that the punishment of
norm violators within a group will cause natural selection
to favor `prosocial genes' (genes that would favor Darwin's
``high standards of morality'', see the quotation opening
this paper).30
As a term of reference, we will call the
evolved aspects of human psychology derived from this
kind of culture-gene coevolutionary process our Prosocial
Psychology (Henrich, 2004).
6.4.1. Cultural learning and punishment leads to social norms
The interaction of cultural learning and punishment
leads to norms that are locally stable in social groups.
Although the mathematics showing that a combination
of prestige-biased transmission and conformist transmission
will lead to stable cooperative norms is somewhat
30
Some students of evolution and human behavior have a visceral
reaction to anything that uses the term ``group selection''. Fear not.
Cultural group selection models do NOT carry the problems typically
associated with models of genetic group selection (Boyd, Gintis, Bowles, &
Richerson, 2003; Henrich, 2004; Smith & Bliege Bird, 2000).
J. Henrich, N. Henrich / Cognitive Systems Research 7 (2006) 220­245 239
more complicated than that used for models of kinship and
reciprocity (Henrich & Boyd, 2001), the logic is fairly
straightforward. Prestige-bias transmission generally acts
to favor the imitation of behaviors, values and practices
that lead to the most successful (and often the most fitness
enhancing) behaviors. Conformist transmission acts to
favor the spread of the majority's behavior/practices, and
principally acts when the difference in the success between
various behaviors or strategies is small or difficult to figure
out. That is, conformist transmission can only maintain
behaviors in a group if they are neutral, not too costly,
and/or sufficiently ambiguous. Thus, for most important
cases of cooperation, or other costly norms, we expect that
the costs of doing a cooperative behavior (the `norm') will
be too costly for conformist transmission to have any
important effects.31
Adding culturally-transmitted tendencies or `tastes' for
punishing norm violators can turn the tables on this logic.
Imagine a group in which strategies for both punishing
norm violators and adhering to a costly norm-behavior
(e.g., cooperation) are common. Punishing norm violators
is costly compared to not punishing norm violators, so
many learning and decision-making mechanisms will still
favor not-punishing. However, punishing norm violators
can be substantially less costly than adhering to a norm
itself, so punishment of norms is easier to stabilize than
many costly norms. Here's why: If our group consists
mostly of individuals who adhere to the norm and punish
those who do not, then most people will stick to the norm
(e.g., recycle tin cans) in order to avoid the costs of being
punished. If punishers are common, the cost of being punished
can easily exceed the costs of sticking to a costly
norm. But, if everyone sticks to the norm (because of punishment),
punishers don't have to do anything, and being a
punisher is not particularly costly. If some few individuals
occasionally violate the norm (say, by mistake), punishing
costs can still remain pretty small. Because these punishing
costs are substantially smaller than those associated with
the cost of sticking to the norm itself, conformist transmission
can often maintain the punishment of norm violators
in the population, against forces like prestige-bias transmission
and experiential learning, even when conformist
transmission is not strong enough to stabilize the costly
norm itself (without any punishment). How well this trick
works depends on the ratio of the costs of punishing to
the costs of being punished. The more effective punishment
is (i.e., the smaller the ratio of the costs of punishing to the
costs of being punished), the more costly can be the norms
that are maintained. In summary, this can work because (1)
punishers don't have to pay the costs of punishing very
often if being punished is more costly than the costs associated
with sticking to the norm ­ everyone generally sticks
to the norm, and punishers need only punish occasional
deviants; (2) the cost of punishing is small (and probably
ambiguous), so conformist transmission can overcome it,
and keep a strategy of punishing stable in the social group;
and (3) when punishing norm violators is common, everyone
tends to adhere to the norm because the costs of being
punished for violating the norm exceed the costs of sticking
with the norm.32
However, suppose that conformist transmission is not
strong enough to stabilize the punishment of norm violators.
Can costly norms still be maintained? Under such circumstances
it is possible that conformist transmission can
act on a tendency to punish individuals who fail to punish
norm violators. We call this the `punishment of non-punishers'.
By the same logic as above, if a group contains mostly
individuals who adhere to a norm, punish violators of the
norm, and punish those who fail to punish norm violators,
conformist transmission can stabilize all of these behaviors
by maintaining the least costly behavior ­ the punishment
of those who fail to punish norm violators. This behavior
is the `cheapest' way to maintain the norm because those
who punish those who fail to punish norm violators will
only have to pay a cost when a norm violation occurs
(which are rare to begin with) and someone fails to punish
that violation (another rare event). This means that these
costs of punishing need only be paid after the conjunction
of two rare events ­ making it an order of magnitude less
costly than punishing norm violators. By favoring the punishment
of non-punishers, the strategy of punishing norm
violators is maintained because the costs of being punished
for not punishing norm violators exceeds the cost of occasionally
administering punishment, and in turn, the presence
of those who punish norm violators means the costs
of being punished for violating the norm can exceed the
costs of sticking to the norm. In this way, conformist transmission
can indirectly stabilize quite costly norms.
Actually, the mathematical analysis behind the above
idea is even more nuanced. It shows that `b' (the benefits
from the cooperative act­see above), does not influence
the creation of a stable norm! That is, the math shows that
a combination of prestige and conformist biases and punishing
behaviors can stabilize any costly norm (`c', the cost
of the behavior, does matter) independent of whether it
benefits anyone. If conformist transmission can favor the
punishment of a behavior then punishment will cause the
behavior, practice or belief to stay common in the group.
31
In analyzing the evolution of cooperation, the real challenge is to
explain the maintenance or stability of cooperative and prosocial
(punishment of non-cooperators) behavior or tendencies, not their spread
from zero. All kinds of random effects, population shocks, stochastic
migrations patterns, etc. to the evolving system can make cooperators
common at a particular place and time, but the key is to explain why these
cooperators won't simply vanish back into evolutionary history as the
system returns to equilibrium. The same situation holds true for
reciprocity solutions; TFT, for example, is not favored until it is
sufficiently common.
32
Boyd et al. (2003) take advantage of this logic in a computer
simulation to show that, even without conformist transmission, the
introduction of punishing strategies substantially enhances the effect of
competition between social groups, and thereby favors the evolution of
substantial amounts of pro-group cooperation.
240 J. Henrich, N. Henrich / Cognitive Systems Research 7 (2006) 220­245
The benefit, b, of the behavior could be zero, or negative,
and the group could still `lock in' on the norm. This was
an unexpected prediction that fell out of theoretical work
on cooperation (Boyd & Richerson, 1992), and just happens
to explain the massive database of anthropological
findings showing that many idiosyncratic social norms
are either neutral or even maladaptive (Edgerton, 1992).
While this aspect of the theory explains social norms in
general (and cooperative norms are one kind of costly
norm), it does not explain why cooperative social norms
should be more common, or more likely to spread, than
other kinds of norms.
The answer to ``why are there so many cooperative social
norms?'' is explained by cultural group selection (CGS). As
different social groups arrive at and `lock in' on different
social norms, CGS provides a process for selection among
alternative social norms. Some groups will develop norms
about constructing community buildings, not eating snakes
and fishing in cooperative units, while other groups may culturally
evolve norms about cooperatively raiding other
groups, sending children to school, and givingyoung girls cliterectomies.
With these different norms in place, social
groups can compete in a variety of ways. First, some groups
have cooperative norms that yield greater success in warfare,
and they may spread their norms by conquering other groups
(Soltis, Boyd, & Richerson, 1995). Second, some groups
have norms (probably cooperative) that increase their economic
(making more food or tools) or demographic (more
kids) production, so they may spread their norms by generating
more carriers of the norm than other groups. For
example, if a group has economic practices that enable them
to better feed their children than neighboring groups, then
they will likely have more children that survive to spread
their norms. This effect would be magnified if the group also
had sexual norms that led to higher reproductive rates than
other groups. Or third, if members of different social groups
have sufficient interaction (which they often do), prestigebiased
transmission can lead individuals to preferentially
imitate people from more successful groups, such as groups
with higher qualities of life in terms of health, housing and
material possessions. Thus, the norms of a successful group
can preferentially spread from group to group relatively rapidly
(Boyd & Richerson, 2002a). Each of the above processes
can lead to the preferential proliferation of cooperative
norms, and each has been observed in the ethnographic,
archaeological and historical record (Atran et al., 2002; Diamond,
1997; Flannery & Marcus, 2000; Kelly, 1985; Shennan,
2003; Soltis et al., 1995; Stark, 1997; Wilson, 2002).
One of the best-documented cases of cultural evolution
through intergroup competition occurred during the 18th
century among the anthropologically famous ethnic groups
of the Nuer and Dinka. Before 1820, the Nuer and Dinka
(Kelly, 1985) occupied adjacent regions in the southern
Sudan. Although the groups inhabited similar environments
and possessed identical technology, they differed in
significant ways. Economically, both the Dinka and Nuer
raised cattle, but the Dinka maintained smaller herds of
approximately nine cows per bull, while the Nuer maintained
larger herds with two cows per bull. The Nuer ate
mostly milk, corn, and millet and rarely slaughtered cows,
while the Dinka frequently ate beef. Politically, the Dinka
lived in small groups, the largest of which corresponded to
their wet season encampment. In contrast, the Nuer organized
according to a patrilineal kin system that structured
tribal membership across much larger geographic areas.
Consequently, the size of a Dinka social group was limited
by geography, whereas the Nuer system could organize
much larger numbers of people over greater expanses of
territory. Despite the similarity of their environments, these
two groups showed substantial economic and political differences.
Over about 100 years, starting around 1820, the
Nuer dramatically expanded their territory at the expense
of the Dinka, who were driven off, killed, or captured
and assimilated. As a result, Nuer beliefs and practices
spread fairly rapidly across the landscape relative to Dinka
beliefs and practices ­ even though the Nuer were soon living
in the once ``Dinka environment'' and the fact that
many Nuer were formerly Dinka who had adopted Nuer
customs (so the cause of the differential success can't be
attributed to environmental or genetic differences).
6.4.2. Genes respond in the wake of cultural group selection
Imagine yourself back in human prehistory, say 55,000
years ago. Social groups are competing and cooperative
norms of various types are spreading through the human
species via cultural learning. Groups have different sets of
norms, including those that promote cooperative hunting,
the sharing of tool-making techniques, community house
building, trade (which requires norms for interaction), raiding,
and warfare. The spreading of these different norms
effectively changes the `selective environment' faced by
genes, because now successful genes have to adapt an individual
to a world in which one is punished for norm violations
­ and norms are often cooperative or prosocial. This
culturally-evolved selection pressure will favor genes that
allow individuals to rapidly acquire the local norms
(thereby avoiding punishment), and avoid the temptation
of norm violation (i.e., defection). Once the newly selected
for genes have spread, cultural learning mechanisms will
favor even larger, more costly cooperative norms, and
CGS will continue to spread those norms that allow social
groups to compete more effectively with other groups
(Henrich & Boyd, 2001). Gradually this interaction of cultural
and genetic transmission ratcheted up our sociality,
honed our preferences for helping others, led to punishing
violators and avoiding punishment, hardened our ability to
inhibit quick defections, and refined our learning capacities
to promote the efficient acquisition of norms until we
became the only ultra-social primate (Richerson & Boyd,
1998). The Prosocial Psychology that underpins much of
our contemporary social life had emerged.
The theory just described above, leads to several general
predictions about the nature of human sociality and our
social psychology:
J. Henrich, N. Henrich / Cognitive Systems Research 7 (2006) 220­245 241
1. Different human groups will be characterized by different
social norms ­ some cooperative, some not. Some
norms will be maladaptive. Non-cultural species will
not show this kind of variability.
2. Negative reactions by third parties to norm violations
will be a human universal (third parties are individuals
who suffer little or no cost from the norm violation).
Non-cultural species will not show third-party punishment
of this kind.
3. More costly norms, such as cooperative norms, will
involve costly punishment of norm violators by third
parties, and possibly the punishment of non-punishers.
This also implies that very low cost cooperative behaviors
can be favored by conformist transmission alone,
without punishment.
4. Human social groups will vary in their `willingness to punish',
and this willingness to engage in costly punishment
will be context specific. When applied to cooperative
interactions this `willingness to punish' accounts for the
empirical phenomena of altruistic punishment and thirdparty
punishment that has been rigorously documented
in experimental work (Fehr & Fischbacher, 2003; Fehr,
Fischbacher, & Gaechter, 2002; Fehr & Ga¨chter, 2002).
5. Norms will usually be context specific. Using experimental
data, we and others have shown how context matters
(Henrich & Henrich, in press).
6. The strength of norms depends on the ratio of the cost of
punishinganotherindividualtothecost ofbeingpunished.
Consequently, social situations that allow for effective
punishment to be dealt out at low cost will favor the maintenance
of group norms, including cooperative norms.
7. In groups with dense social networks, punishment can
operate through damaging of reputation. This provides
a cheap and effective means to punish norm violators.
8. Cultural beliefs that extend the impact of punishment
(such as by effecting the reputation of close kin of the
punished person) increase the effectiveness of punishment
for norm maintenance. This parallels our earlier
discussion of reputation and indirect reciprocity.
9. Social norms that facilitate the spreading of accurate reputational
information, in part by punishing those who
spread false reputational information, can lay a foundation
for substantially increasing the degree of cooperation
that can be supported by Indirect Reciprocity.
6.5. Ethnicity, norms and cooperation
The evolution of norms lays the ground work for the culture-gene
coevolution of an `ethnic psychology' and the
sociological emergence of `ethnic groups'. Our theorizing
begins by considering how the cultural evolution of norms
yields new selection pressures on genes. Two problems present
themselves. First, individuals have to figure out what the
`right norm' is for getting along in their social groups, keeping
in mind that different social groups evolve different
norms. By `right' we mean the norm that allows the individual
to avoid punishment for norm violations and coordinate
their behavior with other members of their social group.
Because norms can have an arbitrary character, natural
selection will favor genes that direct individuals to learn
from those people who are most likely to have the `right
norms'. The second problem arises because, once an individual
has adopted some norms, he would be best off to
interact with other individuals who also share his norms otherwise
he may get punished and/or miscoordinate with
those with whom he is interacting. For example, if a guy
believes that his wife should come with a dowry, he needs
to find women whose families believe that they should pay
a dowry. The root of the puzzle is that people's norms are
not stamped on their foreheads. If people knew the underlying
norms of all people, they could be sure to interact with
those sharing their norms. The tricky part is that underlying
norms are more typically hidden properties of individuals
that rarely surface, and only in specific situations.
So far we have focused on norms enforced by punishment,
especially `cooperative norms'. However, the evolution
of our ethnic psychology and ethnic groups can be
equally well explained by `coordination norms'. Unlike
cooperative norms, situations involving coordination
problems do not have a free rider problem. Individuals
achieve the highest payoff or fitness by doing what everyone
else is doing. For example, the decision to drive on
the right or left side of the road is a coordination problem.
If you are in England, you want to drive on the left,
and if you are in Germany it's best to drive on the right.
Human societies are full of coordination problems, but
many of them are substantially more moralized than
`which side to drive on'. Consider the marriage customs
of some social groups, which demand that a dowry be
sent along with their daughter to the groom, or to the
family of the groom. In others groups, the groom or
the groom's family (or both) are expected to pay for the
bride in, for example, cows, service (labor to the bride's
family) or precious metals. Miscoordinations occur when
both bride and groom's families are expecting payment
from the other, and usually everyone loses as no wedding
occurs. In a sense, conformist transmission and punishment
turn cooperative dilemmas into these coordination
situations, so both can contribute to the emergence of
an ethnic psychology and ethnic groups.
Now back to the two problems: How does natural selection:
(1) figure out who to learn norms from so as to avoid
getting the `wrong' norms, and (2) make sure individuals
preferentially interact with others most likely to share their
norms? Recent theoretical work on the question (McElreath
et al., 2003) suggests that selection `looks' for statistically
reliable correlations between `observables' and these
underlying norms.33
Natural selection is fortunate in this
33
Note the parallel with our discussion of kin selection, where we
explained how `smell' or `hanging around mom' can provide reliable
predictors of relatedness, and thus reliable predictors on `having the same
gene'.
242 J. Henrich, N. Henrich / Cognitive Systems Research 7 (2006) 220­245
case because the same cultural learning processes that
transmit the behaviors and beliefs related to social norms
also often transmit things like dress, language, accent, dialect,
behavioral mannerisms and food preferences. Therefore,
this theoretical work predicts that human
psychology evolved to seek out `indicator traits' (language,
dress, etc.) that match its own because such markers tend
to reliably co-occur with the `right norms' ­ `right' meaning
both the norms a learner wants to learn, and the norms
that match one's own. Using such markers, individuals
can bias both their learning and interaction towards those
individuals who share their same culturally-transmitted
`indicator traits'. As above, the proximate psychological
mechanism of this `ethnic psychology' will involve attentional
biases (who's interesting?) and affective motives
(who does one tend to `like' being around?).
Building on this theory, new empirical work suggests
that people tend to think about certain human groups
(those that we'd typically call `ethnic groups') in the same
way that humans think about different biological species:
we understand the characteristics and attributes of these
groups and biological species in both essentialist (all members
carry the same unchangeable essence that gives rise
to their shared characteristics) and primordialist (this
`essence' is transmitted down blood lines) terms (GilWhite,
1999, 2001). In the case of ethnic groups, this
empirical work suggests that people see ethnic markers
or ethnic membership as a cue of hidden, underlying qualities
or fundamental (unchangeable) attributes. The three
characteristics of a group that tend to spark this type of
`species thinking' are: (1) shared markings (shared language,
dress, etc.), (2) similarity between parents and offspring,
and (3) group endogamy (marriage and mating
occurs within the `ethnic' group) and common descent.
The greater the degree to which these characteristics are
present in a human group ­ or the greater the degree to
which individuals believe or perceive these as present,
the more likely individuals are to use `species thinking'
vis-a`-vis members of that group. The more individuals
use these modes of essentialist and primordialist thinking,
the stronger the ethnic bias in interaction and learning,
and the greater the importance of ethnic membership on
social behavior and cooperation.34
Understanding human ethnic-psychology, and the
coevolutionary processes that lie behind it, illuminates a
large number of puzzling patterns in the world. Perhaps
the most general puzzle is why ethnically-marked groups
seem so important in the world in comparison to other
kinds of human groups: Why are so-called ethnic boundaries
so important in the world compared to other possible
boundaries? Why are political parties so ethnically biased
(instead of class or `common interest' biased)? Why do people
often support political candidates that share their ethnic
markers? Why does violence, oppression and warfare
so often fall along ethnic lines? Why is ethnicity so important
in marriage and sex? The theory summarized above
leads to a series of predictions that help address some of
these puzzles:
1. People (children and adults) use ethnic cues to figure out
from whom to learn. This kind of learning involves
biases in both attention and memory.
2. People prefer to interact with individuals who share
their ethnic markers.
3. At a sociological level, these psychological learning
biases cause individuals who share ethnic markers to
share lots of other norms, beliefs and values.
4. This psychological preference also creates all kinds of
`ethnic clumping', as people seek out members of their
own ethnic groups in marriage, clubs, religion, politics,
etc.
5. Ethnic markers tend to be `hard-to-fake' (honest signals),
as these provide the most reliable cues of the
underlying norms. Language and dialect are particularly
important, as these cannot be easily learned and can
rarely be perfectly faked.
Elsewhere we summarize some of the field and experimental
evidence for these empirical entailments (Henrich
& Henrich, in press, chap. 8).
7. Conclusion
Our goal here has been to integrate an understanding
of cultural evolution into a broader culture-gene coevolutionary
framework. As we see it, one of the main problems
with standard evolutionary approaches is that they
fail to account for the fact that humans, to a degree not
observed in other animals, have evolved to rely heavily
on observational learning to acquire vast swaths of their
behavioral repertoire ­ a process that can, in turn, influence
the direction of genetic evolution. On the other
hand, traditional `cultural approaches' tend to ignore or
deny any influence of evolution on human behavior,
and seem blind to the fact the cultural learning itself is
best understood as a kind of adaptation, whose origins
and operation can be rigorously examined with a combination
of formal models, ethnographic inquiry, and
focused experimentation.
34
We've not the space to deal with this at any length, but this theory of
ethnicity explains the essentialist and primordial nature of human thinking
vis-a`-vis race as a byproduct of the coevolutionary processes that
produced it for ethnicity (Gil-White, 1999, 2001). In human ancestral
history, individuals probably rarely, if ever, encountered individuals who
looked very different from themselves (i.e., biologically: skin color,
morphology, hair, etc.). When people did finally meet other such human
groups late in human history, these phenotypic differences, by chance, fit
easily with the cues for different ethnic groups ­ racial difference may
super-stimulate this tendency to `species-thinking'. To be perfectly clear,
we are NOT saying that there are essential or primordial differences
between either racial or ethnic groups. We are addressing why humans so
frequently (and incorrectly) think about such groups in these terms.
J. Henrich, N. Henrich / Cognitive Systems Research 7 (2006) 220­245 243
References
Alvard, M. (2003). The adaptive nature of culture. Evolutionary Anthropology,
12(3), 136­147.
Atran, S., Medin, D., Ross, N., Lynch, E., Vapnarsky, V., Ek', E. U.,
et al. (2002). Folkecology, cultural epidemiology, and the spirit of the
commons: a garden experiment in the Maya lowlands, 1991­2001.
Current Anthropology, 43(3), 421­450.
Atran, S., Medin, D., Ross, N., Lynch, E., Coley, J., Ek', E. U., &
Vapnarsky, V. (1999). Folkecology and commons management in the
Maya lowlands. Proceedings of the National Academy of Sciences, 96,
7598­7603.
Axelrod, R. (1984). The evolution of cooperation. New York: Basic Books.
Axelrod, R., & D'Ambrosio, L. (1994). Annotated bibiolography on the
evolution of cooperation. Unpublished manuscript, Ann Arbor.
Axelrod, R., & Hamilton, W. D. (1981). The evolution of cooperation.
Science, 211, 1390­1396.
Baldwin, J. M. (1896a). A new factor in evolution. American Naturalist,
30(441­457), 536­554.
Baldwin, J. M. (1896b). Physical and social heredity. American Naturalist,
30, 422­428.
Bendor, J. (1987). In good times and bad: reciprocity in an uncertain
world. American Journal of Political Science, 31, 531­558.
Bendor, J. (1993). Uncertainty and the evolution of cooperation. Journal
of Conflict Resolution, 37(4), 709­734.
Bendor, J., Kramer, R. M., & Stout, S. (1991). When in doubt . . .
Cooperation in a noisy prisoners-dilemma. Journal of Conflict Resolution,
35(4), 691­719.
Bendor, J., Kramer, R., & Swistak, P. (1996). Cooperation under
uncertainty: what is new, what is true, and what is important?
American Sociological Review, 61(2), 333­338.
Bendor, J., & Mookherjee, D. (1987). Institutional structure and the logic
of ongoing collective action. The American Political Science Review,
81(1), 129­154.
Bendor, J., & Swistak, P. (1997). The evolutionary stability of cooperation.
American Political Science Review, 91(2), 290­307.
Bendor, J., & Swistak, P. (2001). The evolution of norms. American
Journal of Sociology, 107(2), 1493­1545.
Bliege Bird, R. L., Smith, E. A., & Bird, D. W. (2001). Bliege the hunting
handicap: costly signaling in human foraging societies. Behavioral
Ecology and Sociobiology, 50, 9­19.
Boerlijst, M. C., Nowak, M. A., & Sigmund, K. (1997). The logic of
contrition. Journal of Theoretical Biology, 185, 281­293.
Boyd, R. (1989). Mistakes allow evolutionary stability in the repeated
prisoner's dilemma game. Journal of Theoretical Biology, 136, 47­
56.
Boyd, R. (1992). The evolution of reciprocity when conditions vary. In F.
Dewaal & A. H. Harcourt (Eds.), Coalitions in humans and other
animals (pp. 473­492). New York: Oxford University Press.
Boyd, R., Gintis, H., Bowles, S., & Richerson, P. J. (2003). The evolution
of altruistic punishment. PNAS, 100(6), 3531­3535.
Boyd, R., & Lorderbaum, J. P. (1987). No pure strategy is evolutionarily
stable in the repeated prisoner's dilemma game. Nature, 32(6117),
58­59.
Boyd, R., & Richerson, P. (1988). The evolution of reciprocity in sizable
groups. Journal of Theoretical Biology, 132, 337­356.
Boyd, R., & Richerson, P. (1992). Punishment allows the evolution of
cooperation (or anything else) in sizable groups. Ethology & Sociobiology,
13(3), 171­195.
Boyd, R., & Richerson, P. (2002a). Group beneficial norms can spread
rapidly in a structured population. Journal of Theoretical Biology, 215,
287­296.
Boyd, R., & Richerson, P. (2002b). Solving the puzzle of human
cooperation. In S. Levinson (Ed.), Evolution and culture. Cambridge
MA: MIT Press.
Boyd, R., & Richerson, P. J. (1985). Culture and the evolutionary process.
Chicago, IL: University of Chicago Press.
Boyer, P. (2001). Religion explained: The evolutionary origins of religious
thought. New York: Basic Books.
Chagnon, N., & Irons, W. (1979). Evolutionary biology and human social
behavior: An anthropological perspective. North Scituate: Duxbury
Press.
Conte, R., & Paolucci, M. (2002). Reputation in artificial societies: Social
beliefs for social order. Boston: Kluwer Academic Publishers.
Cronk, L., Chagnon, N., & Irons, W. (Eds.). (2000). Adaptation and
human behavior. New York: Aldine De Gruyter.
Dawkins, R. (1976). The selfish gene. Oxford: Oxford Unversity Press.
DeBruine, L. (2002). Facial resemblance enhances trust. Proceedings of the
Royal Society of London Series B ­ Biological Sciences, 269,
1307­1312.
Diamond, J. M. (1997). Guns, germs, and steel: The fates of human
societies. New York: W.W. Norton & Co.
Durham, W. H. (1991). Coevolution: Genes, culture, and human diversity.
Stanford, CA: Stanford University Press.
Durkheim, E. (1933). The division of labor in society. New York: Free
Press.
Edgerton, R. B. (1992). Sick societies: Challenging the myth of primitive
harmony. New York: Free Press.
Engelmann, D., & Fischbacher, U. (2002). Indirect reciprocity and
strategic reputation building in an experimental helping game.
Unpublished manuscript, University of Zurich.
Farrell, J., & Ware, R. (1989). Evolutionary stability in the repeated
prisoner's dilemma. Theoretical Population Biology, 36(161­166).
Fehr, E., & Fischbacher, U. (2003). Third party sanction and social
norms. Evolution and Human Behavior, 25(2004), 63­87.
Fehr, E., Fischbacher, U., & Gaechter, S. (2002). Strong reciprocity,
human cooperation and the enforcement of social norms. Human
Nature, 13, 1­25.
Fehr, E., & Ga¨chter, S. (2002). Altruistic punishment in humans. Nature,
415, 137­140.
Fehr, E., & Henrich, J. (2003). Is strong reciprocity a maladaption? In P.
Hammerstein (Ed.), Genetic and cultural evolution of cooperation.
Cambridge: MIT Press.
Fiske, A. (1991). Structures of social life. New York: Free Press.
Flannery, K. V., & Marcus, J. (2000). Formative Mexican chiefdoms and
the myth of the 'mother culture'. Journal of Anthropological Archaeology,
19(1­37).
Frank, S. (1995). George price's contributions to evolutionary genetics.
Journal of Theoretical Biology, 175, 373­388.
Frank, S. (1997). The price equation, Fisher's fundamental theorem, kin
selection, and causal analysis. The Society for the Study of Evolution,
51(6), 1712­1729.
Frank, S. (1998). Foundations of social evolution. Princeton: Princeton
University Press.
Gil-White, F. (1999). How thick is blood? The plot thickens. Ethnic and
Racial Studies, 22(5), 789­820.
Gil-White, F. (2001). Are ethnic groupes biological `species' to the human
brain? Essentialism in our cognition of some social categories. Cultural
Anthropology, 42(4), 515­554.
Grafen, A. (1985). Hamilton's rule OK. Nature, 318, 310­311.
Hamilton, W. D. (1964). The genetical evolution of social behavior.
Journal of Theoretical Biology, 7, 1­52.
Hamilton, W. D. (1972). Altruism and related phenomena, mainly in the
social insects. Annual Review of Ecology and Systematics, 3, 192­232.
Hammerstein, P. (2003). Why is reciprocity so rare in social animals? In P.
Hammerstein (Ed.), Genetic and cultural evolution of cooperation.
Cambridge: MIT Press.
Hawkes, K., O'Connell, J. F., & Blurton Jones, N. (2001). Hadza meat
sharing. Evolution and Human Behavior, 22, 113­142.
Henrich, J. (2004). Cultural group selection, coevolutionary processes and
large-scale cooperation. Journal of Economic Behavior & Organization,
53, 3­35.
Henrich, J. (in press). A cultural species. In M. Brown (Ed.), Explaining
culture scientifically. Seattle: University of Washington Press.
244 J. Henrich, N. Henrich / Cognitive Systems Research 7 (2006) 220­245
Henrich, J., Bowles, S., Smith, E. A., Young, P., Boyd, R., Sigmund, K.,
et al. (2003). The cultural and genetic evolution of human cooperation.
In P. Hammerstein (Ed.), Genetic and cultural evolution of
cooperation. Cambridge: MIT Press.
Henrich, J., & Boyd, R. (1998). The evolution of conformist transmission
and the emergence of between-group differences. Evolution and Human
Behavior, 19, 215­242.
Henrich, J., & Boyd, R. (2001). Why people punish defectors: weak
conformist transmission can stabilize costly enforcement of norms in
cooperative dilemmas. Journal of Theoretical Biology, 208, 79­89.
Henrich, J., & Boyd, R. (2002). On modeling cultural evolution: why
replicators are not necessary for cultural evolution? Journal of
Cognition and Culture, 2(2), 87­112.
Henrich, J., Boyd, R., Bowles, S., Camerer, C., Fehr, E., & Gintis, H.
(Eds.). (2004). Foundations of human sociality: Economic experiments
and ethnographic evidence from fifteen small-scale societies. Oxford:
Oxford University Press.
Henrich, J., & Gil-White, F. (2001). The evolution of prestige: freely
conferred deference as a mechanism for enhancing the benefits of
cultural transmission. Evolution and Human Behavior, 22(3), 165­196.
Henrich, J., & McElreath, R. (2003). The evolution of cultural evolution.
Evolutionary Anthropology, 12(3), 123­135.
Henrich, N. S., & Henrich, J. (in press). Why humans cooperate: A cultural
and evolutionary explanation. Oxford: Oxford University Press.
Hill, K. (2002). Altruistic cooperation during foraging by the ache and the
evolved human predispostion to cooperate. Human Nature, 13(1),
105­128.
Hirshleifer, J., & Martinez-Coll, J. C. (1988). What strategies can support
the evolutionary emergence of cooperation? Journal of Conflict
Resolution, 32(2), 367­398.
Johnson, A., & Earle, T. (2000). The evolution of human societies (2nd ed.).
Stanford: Stanford University Press.
Joshi, N. V. (1987). Evolution of cooperation by reciprocation within
structured demes. Journal of Genetics, 66(1), 69­84.
Kameda, T., & Nakanishi, D. (2002). Cost-benefit analysis of social/
cultural learning in a non-stationary uncertain environment: an
evolutionary simulation and an experiment with human subjects.
Evolution and Human Behavior, 23, 373­393.
Keller, L., & Ross, K. G. (1998). Selfish genes, a green beard in the red fire
ant. Nature, 394, 573­575.
Kelly, R. C. (1985). The Nuer conquest. Ann Arbor: University of
Michigan Press.
Leimar, O., & Hammerstein, P. (2001). Evolution of cooperation through
indirect reciprocity. Proceedings: Biological Sciences, 268(1468),
745­753.
Lorberbaum, J. (1994). No strategy is evolutionarily stable in the repeated
prisoner's dilemma. Journal of Theoretical Biology, 168, 117­130.
Lorberbaum, J. P., Bohning, D. E., Shastri, A., & Sine, L. (2002). Are
there really no evolutionary stable strategies in the iterated prisoner's
dilemma? Journal of Theoretical Biology, 214, 155­169.
Maynard Smith, J. (1982). Evolution and the theory of games. Cambridge:
Cambridge University Press.
McElreath, R., Boyd, R., & Richerson, P. (2003). Shared norms can lead
to the evolution of ethnic markers. Current Anthropology, 44(1),
123­129.
Meltzoff, A. N., & Prinz, W. (Eds.). (2002). The imitative mind. New York:
Cambridge University Press.
Nowak, M. A., & Sigmund, K. (1998a). The dynamics of indirect
reciprocity. Journal of Theoretical Biology, 194, 561­574.
Nowak, M. A., & Sigmund, K. (1998b). Evolution of indirect reciprocity
by image scoring. Nature, 393(6685), 573­577.
Panchanathan, K., & Boyd, R. (2003). A tale of two defectors: the
importance of standing for the evolution of indirect reciprocity.
Journal of Theoretical Biology, 224, 115­126.
Platek, S. M., Burch, R. L., Panyavin, I. S., Wassserman, B. H., &
Gallup, G. G. (2002). Reactions to children's faces: resemblance
affects males more than females. Evolution and Human Behavior, 23,
159­166.
Platek, S. M., Critton, S. R., Burch, R. L., Frederick, D. A., Myers, T.
E., & Gallup, G. G. (2003). How much paternal resmblance is
enough? Sex differences in hypothetical investment decisions but not
in the detection of resemblance. Evolution and Human Behavior, 24,
81­87.
Queller, D. C. (1992). A general model for kin selection. Evolution, 46(2),
376­380.
Richerson, P., & Boyd, R. (1998). The evolution of ultrasociality. In I.
Eibl-Eibesfeldt & F. K. Salter (Eds.), Indoctrinability, ideology and
warfare (pp. 71­96). New York: Berghahn Books.
Richerson, P., & Boyd, R. (2000). Complex societies: the evolutionary
dynamics of a crude superorganism. Human Nature, 10, 253­289.
Richerson, P., & Boyd, R. (2005). Not by genes alone: How culture
transformed human evolution. Chicago: University of Chicago Press.
Rogers, E. M. (1995). Diffusion of innovations. New York: Free Press.
Selten, R., & Apestegula, J. (2002). Experimentally observed imitation and
cooperation in price competition on the circle. Unpublished
manuscript.
Shennan, S. (2003). Genes, memes, and human history: Darwinian
archaeology and cultural evolution. London: Thames & Hudson.
Smith, A. (2000). The theory of moral sentiments. New York: Prometheus
Books.
Smith, E. A. (2003). Human cooperation: perspectives from behavioral
ecology. In P. Hammerstein (Ed.), Genetic and cultural evolution of
cooperation. Cambridge: MIT Press.
Smith, E. A., & Bliege Bird, R. L. (2000). Turtle hunting and tombstone
opening: public generosity as costly signaling. Evolution and Human
Behavior, 21(4), 245­261.
Soltis, J., Boyd, R., & Richerson, P. J. (1995). Can group-functional
behaviors evolve by cultural group selection? An empirical test.
Current Anthropology, 36(3), 473­494.
Stark, R. (1997). The rise of christianity: How the obscure, marginal Jesus
movement became the dominant religious force in the Western World in a
few centuries. New York: Harper Collins.
Sugden, R. (1986). The economics of rights, co-operation, and welfare.
Oxford: Basil and Blackwell.
Tomasello, M. (1999). The human adaptation for culture. Annual Reviews
of Anthropology, 28, 509­529.
Trivers, R. L. (1971). The evolution of reciprocal altruism. Quarterly
Review of Biology, 46, 35­57.
Trivers, R. (1985). Social evolution. Menlo Park, CA: Benjamin/
Cummings.
Wedekind, C., & Milinski, M. (2000). Cooperation through image scoring
in humans. Science, 288, 850­852.
Wilson, D. S. (2002). Darwin's Cathedral: Evolution, religion, and the
nature of society. Chicago: University of Chicago.
Wolf, A. P. (1995). Sexual attraction and childhood association: A
Chinese brief for Edward Westermarck. Stanford University Press.
Wu, J., & Axelrod, R. (1995). How to cope with noise in the iterated
prisoner's dilemma. Journal of Conflict Resolution, 39(1), 183­189.
J. Henrich, N. Henrich / Cognitive Systems Research 7 (2006) 220­245 245