To appear in Biology
and Philosophy (2007)
Defining and explaining
culture
(comments on Richerson and Boyd, Not by
genes alone)
DAN SPERBER and NICOLAS CLAIDIÈRE
Institut Jean Nicod
(CNRS/EHESS/ENS)
Correspondence
to: dan@sperber.com
There is much to admire in the work of
Robert Boyd and Peter Richerson, and much that we agree with. In particular we
share the goal of developing a “population thinking” approach to cultural
evolution that sees it neither as a mere extension of biological evolution (as
in pop sociobiology), nor as a mere analog of biological evolution (as in pop
memetics). Not by genes alone (Richerson
and Boyd 2005) provides
a good overview of their contribution, and an appropriate target for
discussion. Here we focus on general issues linked to their very definition of
culture.[1]
What is at issue is not a matter of
conceptual analysis, let alone of terminology, it is a matter of explanatory
adequacy. True, cultural anthropology gets by without any clear and agreed upon
definition of culture, but then the goals of most anthropologists—and their more
obvious achievements—are ethnographical and interpretive rather than
theoretical. When the goal is to develop a naturalistic and theoretical
approach, one’s definition, or at least one’s characterization of culture
matters.
Richerson and Boyd write:
Culture
is information capable of affecting individuals’ behavior that they acquire
from other members of their species through teaching, imitation, and other
forms of social transmission.
By information, we mean any kind of mental state, conscious or not, that
is acquired or modified by social learning, and affects behavior. (p.5 – their
italics)
Information is an abstract relational
property. It is not something that, in and of itself, has causes or effects.
Rather, it is a property that material items may possess in virtue of their
causal connections. For instance, tree rings contain information about the age
of a tree in virtue of being caused by seasonal changes in tree growth. The
brain state that realizes my perception of a computer screen in front of me
contains the information that there is a computer screen in front of me in
virtue of being caused, through appropriate perceptual processes, by the
computer screen in front of me. The picture of Madonna on the computer screen
contains information about Madonna’s face in virtue of having been caused (via
a complex causal route before reaching this screen) by light reflected from
Madonna’s face. Unlike tree rings, perceptions and pictures not only contain
information, but have the function of making the information they
contain available to information-using systems (to other brain mechanisms in
the case of perceptions, and to people in the case of pictures). The
information contained in items that have such a function is indicative not only
of their past history but also of their likely future effects. Whereas, with
very few exceptions, tree rings have no causal effects, pictures of Madonna
cause perceptions that cause recognitions that cause various actions such as
imitating her hairstyle or buying her CDs.
So, in spite of its abstract character, information
can be relevant to identifying the past and future causal relationships of
items—e.g. genes, brain states, or pictures—that contain it. Still, the causal
powers of these items depend on their material properties, not just on the
information these material properties implement. The same information, say about
Madonna’s face, displayed on a computer screen, stored in an electronic file at
www.madonna.com, or printed on a CD jacket, has, in each of its implementations,
different effects, in particular a different cultural impact. To understand how
information is distributed, one must understand how it is implemented.
Cultural information spreads across members
of a population through their interactions, that is, through their producing,
in their common environment, events and objects that carry information that
others can pick up. So, is cultural information located in people’s mind/brain
or in their behaviors and artifacts, or in both? On this question, Richerson
and Boyd might seem to waver. In the citation above, they define culture as
information, and information as mental states. Later on they qualify this
mentalistic approach, stating that “culture is (mostly) information in brains”
and noting that “undoubtedly, some cultural information is stored in artifacts”
(p.61). Later still, discussing the example of the bowline knot, they note “If
we could look inside people’s heads, we might find out that different
individuals have different mental representations of a bowline, even when they
tie it exactly the same way” (p.64). This might suggest, contrary to what Richerson
and Boyd want, that the cultural items in this case are the knots themselves, rather
than their variable mental representations.
Mental representations are, we
agree, of special importance to culture, since the very existence of culture
presupposes a population capable of mental representations—no mind, no
culture—, while there is no well-defined type of behavior or of artifact that
is a necessary ingredient of culture. Still, we see no good reason to deny that
behavior and artifacts through which cultural information is transmitted are cultural
too. Against such a point of view, Dawkins and other memeticists have argued
that, in fact, mental aspects of culture are to behavioral and artifactual
aspects what genes are to their phenotypical expressions (Dawkins
1976, 1982).
However, in the absence of anything resembling a cultural germline, and in the
presence, rather, of a systematic back-and-forth, in the causal chains of
culture, between mind-internal and mind-external episodes, there is no more
reason to consider that, say, tokens of the competence involved in tying a
bowline knot beget other tokens of the same competence by producing actual
bowline knots for other to see than to consider that token bowline knots beget
other token bowline knots by recruiting people’s cognitive and motor
capacities.
Richerson and Boyd do not argue for their
definition of culture as located (mostly) in mental representations by invoking
a spurious analogy with the genotype/phenotype distinction. Their approach to
cultural evolution, particularly manifest in their more formal models and
simulations (that are evoked but not discussed in this book), gives them, we surmise,
a different rationale to consider that culture consists in mental
representations. For them, the most basic type of micro-event in cultural
evolution is the adoption by an individual of some cultural variant. The state
of a culture at a given moment corresponds to the distribution of variants resulting
from these micro-events, and the evolution of culture is that of this
distribution. Richerson and Boyd are particularly interested in the way in
which the adoption of a variant by some individuals may cause others to adopt
it too. With such a focus, the role played by behaviors and artifacts in
cultural transmission, even if indispensable, is less central than the role
played by individual decisions. Richerson and Boyd take the collective
phenomenon of culture to be the evolving outcome of the aggregation of these decisions
(with random and biological factors interfering in various ways). They are more
interested in the relatively simple psychology of decision—even if, unlike many
methodological individualists, they don’t see it as a mere implementation of
rational choice theory—than in the necessarily more complex psychology of the
formation of mental contents.
Richerson and Boyd would, no doubt, be the
first to acknowledge that such an approach is based on a simplification of the
cultural process. Moreover, their book is full of insightful qualifications and
nuances correcting this simplification. Nevertheless, they would argue—and
rightfully so—, without extreme simplification, no useful modeling or
simulation is possible.
The kind of simplification of the cultural
process that Richerson and Boyd opt for has proved quite fruitful in their work.
Still, we want to suggest, it would be a mistake to assume that this particular
simplification zeroes in on the defining properties of culture and abstracts
away only from relatively peripheral or less important properties.
In their discussion of the forces of
cultural evolution, Richerson and Boyd distinguish three major types: random
forces, natural selection, and “decision-making forces.” This third type,
specific to cultural evolution, is itself divided in two sub-types: “guided
variation” and “biased transmission.” Biased transmission in turn has three sub
categories: “content-based bias,” “frequency-based bias,” and “model-based
bias” Much of Richerson and Boyd’s most valuable work has been devoted to
exploring the consequences of these two last types of biases. Frequency-based
and model-based biases have to do with the choice of one cultural variant over
others, and can reasonably be described as “decision-making forces.” “Guided
variation” and “content-based bias” on the other hand do not belong to the
psychology of decision, even broadly understood.
Here is a simple example of “guided
variation.” Imagine a foreign stew recipe being introduced with some success in
a population. Cooks however make mostly unconscious and idiosyncratic decisions
regarding the proportion of the ingredients and hardly ever reproduce the
model. These modified stews don’t depart at random from the original recipe;
rather, they gravitate towards a new recipe more in the style of the local
cuisine. Over time, this new recipe becomes the one people have in mind and on
their plate. To model such a plausible evolution one should take into account
not only frequency of adoption but also rate and directionality of the
variations that occur in the actions of cooks.
Here is a simple example of “content-based bias.”
Imagine a comedian telling two new jokes one evening on a television show. Both
jokes are much appreciated and adopted by the same number of viewers for future
retellings. However joke 2 is harder to remember than joke 1, so that, say, 80%
of the people who adopt it forget in less than a month, whereas only 20% forget
joke 1 in the same period. Quite plausibly, joke 1 will spread and become a
standard joke in the culture, and joke 2 won’t. To model such a plausible
evolution one should take into account not only frequency of adoption but also
frequency of forgetting.
Our disagreement with Richerson and Boyd is
about the character and role of processes such as content-based bias and guided
transmission. Their picture, as we understand it, is that most of cultural transmission
consists in the choice, by individuals, of some cultural variant among those on
offer. Chosen cultural variants are acquired through learning processes
(imitation in particular) that are essentially preservative. Even if they need not
result in strict replication, they preserve across episodes of transmission the informational
content of a variant sufficiently well for it to be socially shared and hence
cultural (with frequency and model-based biases contributing to the effectiveness
of the process). In the case of “content-based bias” (e.g. some jokes being
better remembered than others), further psychological processes alter the
probability of a specific variant being effectively implemented in people’s
mind as a function of its content. Such content-based biases don’t modify the contents
of cultural variants: they affect only their frequency. So, even if they are
not strictly speaking decision processes, Richerson and Boyd can see their
effect as similar to that of decision proper: in the end some variants are more
often retained than others. The only constructive process involved in cultural
evolution, that is the only process capable of introducing new contents in a
non-random way (as in the example of the evolving stew recipe) is, according to
Richerson and Boyd, guided variation.
We, on the other hand, believe (and have argued at length elsewhere, e.g. Sperber 1996, Sperber & Hirschfeld 2004, Sperber and Claidière forthcoming) that psychological mechanisms involved in social learning always involves a combination of preservative and constructive processes. All learning (with the possible exception of rote learning of nonsense material) is biased by content. What this means however is not just that some input-content is more easily and therefore more often learned than some other. It also means that, when an input-content is neither too hard nor maximally easy to learn, it is likely to be transformed in the direction of greater ease. For instance a foreign word is remembered with a normalized phonology; a story is remembered without its irrelevant details; a novel idea is remembered as just a version of an already familiar one; the recipe of an original stew is remembered so as better to fit the cook’s mental habits, and so forth. If so, there is no clear distinction between content-based bias and guided variation.
Content-based biases often result in
non-random variations, and guided variation is often guided by content-biases. It
might seem that guided variation, being most important as a source of
innovation, cannot, in this, be guided by content-based biases. In fact,
applying a content-based bias to novel material (e.g. a local cuisine bias to a
foreign recipe), or approaching familiar material with a content-based bias other
than the most obvious one (e.g. approaching an algebraic problem with a
geometric bias) can result in innovation.
Given this, we would suggest a different
classification of what Richerson and Boyd call “decision-making forces.” To
begin, we would rename the whole category psychological forces since we
believe the exclusive focus on decision is misleading. Among psychological
forces, we would distinguish source-based biases and content-based
biases. Source-based biases affect the probability that a given content be
adopted depending on the source(s) from which it is received. Both Richerson
and Boyd’s frequency-based and model-based biases are examples of source-based
biases. Content-based biases are effects of the cognitive mechanisms that
construct a mental representation on the basis of informational input. We
believe that most if not all of these cognitive mechanisms are domain-specific
and treat different contents differently (Hirschfeld & Gelman 1994). The
construction of a mental representation involves greater or lesser
transformation of the input information, with two limiting case, that of total
loss of information or complete forgetting when cognitive mechanisms just ignore
or filter out the input information, and that of the construction of a mental
representation containing exactly the same information as the input, as when
you correctly remember a phone number. Most processing of input information
results neither in total loss nor in exact copy; it is, as we insisted, both
preservative and constructive.
Incidentally (since we won’t pursue the
matter here), beside random forces, natural selection, and what we prefer to
call psychological forces, we would suggest adding a fourth category: ecological
forces that act on the behaviors and artifacts involved in the causal
chains of culture. What may cause a stew recipe to evolve is the local
availability of ingredients and possible substitutes. Higher population density
with the increased availability of the expertise of others buttresses
folk-knowledge, protects it from the risk of drift, and allows it to
complexify. Hard-to-remember narratives nevertheless reach a cultural level of
distribution when writing provides an external memory. Complex calculus is much
more commonly performed and has a greater cultural impact when it can be handled
by computers, and so on. Just as psychological forces involve mental mechanisms
that are in part genetically determined and in part the output of culturally
informed cognitive development, ecological forces involve aspects of the
environment that are themselves the result of human action, and therefore of
human culture (a point interestingly discussed under the label
“niche-construction” by Odling-Smee,
Laland et al. 2003). Richerson
and Boyd are of course well aware of these ecological factors and give many
examples, but they don’t give them a place among the different kinds of force
they identify.
The information contained in the behaviors
and artifacts through which culture is transmitted is quite generally
insufficient to determine by itself the contents of the corresponding mental
representations. In order to exploit this information, learners must bring to
bear on it not only general learning or imitation skills, but also
domain-specific information and procedures already present in their minds. In
other terms, we believe that Chomsky’s poverty of the stimulus argument
generalizes, mutadis mutandis, from language acquisition to all forms of cultural
learning. The learning process involves not just extraction but also
interpretation of input information, and interpretation typically involves
enrichment of the information interpreted.
One might object: if preservation of
information were not secured to a high degree by general preservative
mechanisms such as imitation or communication, how could any informational
content ever end up being shared in a population to a degree sufficient to determine
an identifiable cultural item? Isn’t the very existence of culture proof that
there are mechanisms of information preservation effective enough to secure its
relative stability? We reply: the burden of securing population-scale content
stability does not have to be wholly carried by preservative processes.
Richerson and Boyd themselves show how frequency and model-based biases can, at
a propulation scale, help compensate for the
insufficiencies of individual preservation processes. We agree, and suggest
that constructive processes are also a major factor of population-scale
stability, since these processes tend, across individuals, to interpret input
information in a common direction. In the case of language acquisition, for
instance, assuming that there is an evolved language acquisition device helps
explain how children of the same language community end up having very similar mental
grammars when the linguistic evidence on the basis of which these grammars are
acquired consist in the quite different set of utterances heard by each child
over the learning years. In cultural transmission, the limits of preservative
processes are, we claim, to a crucial extent compensated by the convergence of
constructive processes.
What Richerson and Boyd say about bowline
knots is likely to be quite commonly the case: people’s mental representation
of cultural information is likely to possess an important degree of individual
variation. Provided that these variation gravitate towards the same
“attractors” (Sperber
1996) in the space of
possibilities (for instance towards the same phonological regularities in word
learning or towards the same balance of tastes in a stew recipe), these
variations need not compromise cultural stability. Of course, not all mental
representations of cultural contents exhibit the same level of individual
idiosyncrasy. The mentally represented phonology of words for instance is
likely to exhibit much less individual variation than their mentally
represented semantics. However, even when learners produce in each of their
individual minds a quasi-exact counterpart of a cultural model, it would be a
mistake to assume that they do so by actually copying the model in all its relevant
details. Learners can achieve what looks like strict reproduction when in fact the
input information is incomplete and ambiguous, provided that their constructive
abilities converge towards the same specific outcomes.
More importantly, thanks both to
source-base biases and to converging constructive processes, there can be
variations at every step—mental or environmental—in the causal chains that
distribute cultural information without compromising the population-scale
stability of this information. Let us underscore, especially for those who accept
a gene-meme analogy, that these variations are quite different from phenotypic
variations that play no role in the mechanics of genes replication. In the
transmission process we are describing, learners do not acquire true cultural
information by ignoring idiosyncratic aspects of the input and extracting and copying
only its cultural core, but by interpreting the information as provided by
means of constructive mechanisms they share with one another.
If we are right, cultural contents owe much
of their stability to the directionality of constructive psychological processes.
These processes are typically complex, domain-specific, and therefore much
better at stabilizing some contents than others. Richerson and Boyd themselves
give excellent example of the type of processes we have in mind in discussing,
for instance, the evolution of word phonology. But then, simplifying cultural
evolution by focusing almost exclusively on decision processes, while it has,
in their modelling work, proved remarkably insightful, may also suggest a distorted
view of the general character of culture.
So let us suggest another way of defining
or at least characterizing what is cultural—“what is cultural” rather than
“what is culture” because, as we shall argue, culture is better viewed as a
property rather than as a thing.
In non-human animals, relatively
little information if any is acquired by social learning. Humans on the other hand owe much of their information to others.
Many criss-crossing causal chains distribute a great amount of information
throughout any human population, and, as rightly stressed by Richerson and
Boyd, this information accumulates. Still, even among humans, most mental
representations play an ephemeral role in just one individual’s mind and are
not transmitted at all. Many causal chains distributing information are short: the
information is about local and transient situations, e.g. children’s need,
where to find food, gossip about relatives, and does not flow beyond its narrow
perimeter of relevance. Some information travels further and last longer, for
instance a rumor in a village, a restaurant becoming fashionable in a neighborhood.
Other information still, e.g. general knowledge, technical skills or religious
myths, does propagate over wider social space and for a longer time and may even
become prevalent in a whole population for generations.
When anthropologists and others talk of
culture—independently of the way they might define it—, they refer to this
widely distributed information and to the mental representations, behaviors,
artifacts, and institutions that, one way or another,
implement this information. Richerson and Boyd’s definition of culture as
“information capable of affecting individuals’ behavior that they acquire from
other members of their species” does not mention the scale of the distribution
and would be satisfied, for instance by the micro-local information John
acquires from Helen when she says, “Careful, the coffee is hot!”. Still, it is
clear that they mean by “culture” widely distributed beliefs, norms and skills,
and not such ephemeral trivia. What we want to stress, however, is that there
is a continuum of cases between these and widely distributed information.
Throughout this continuum, most mental representations and behaviors are shaped
by a mix of individual and social inputs, so that there is no way to pry apart
cultural information from all the information found in a human population.
At the most individual end of the spectrum,
we find mental representations that are not communicated and behaviors that are
not addressed to others. Still, an individual uncommunicated thoughts, plans,
or even dreams are typically built with ingredients—concepts, pieces of
knowledge, or of know-how—that were socially acquired. Moreover, even if not
communicated to others, these idiosyncratic mental representations do
contribute to shaping behavior and, as a result, something of their tenor seeps
through into the causal chains of social communication. I may not tell my
dreams, but they may inspire my choice of metaphors. I may neither explain nor even
demonstrate my way of preparing a stew, but my stews themselves carry information
about my methods.
At the other, more cultural end of the continuum,
there is still often much idiosyncrasy in mind or in behavior. It is easy to
find culturally stable behaviors or artifacts, ritual behaviors and their paraphernalia
for instance, but their very cultural success has much to do with the fact that
they can be mentally interpreted with a high degree of idiosyncrasy (Sperber
1975). It is also easy to find
deeply entrenched cultural attitudes and ideas, such as the notion that some
other ethnic group is inferior to one’s own. Such prejudices often survive even
when their explicit expression is discouraged. In such cases, quite diverse implicit
public manifestations, each adjusted to some micro-local interaction, may
suffice to secure the wide and lasting distribution of these attitudes.
We are not denying that there are many cases
where quasi-identical mental representations and quasi-identical behaviors
propagate by causing each other in turns, for instance children counting
rhymes, or calligraphic skills. Still, it would be not just a simplification,
but also a serious distortion to take such “memish” cases as paradigm examples
of the cultural process.
If we are right and there is a continuum of
cases without any demarcation among humans between more individual and more cultural
information, then “culture” is better viewed as a property that human mental
representations and practices exhibit to a varying degree than as a type or a
subclass of these representations and practices (or of “information”). To explain
culture so understood, the object of study must be the overall flow of
information among humans, through its mental and public implementations; the
question that must be answered is what causes some of causal chains to extend
more than others in time and space and to stabilize better than others the
contents they vehiculate. For this, the study of culture must be embedded in a
more general epidemiology of representations and practices that attends —as does medical epidemiology—to the complexities of both individual
and ecological mechanisms. Boyd and Richerson’s work over the years is, of
course, a major contribution to such an epidemiology.
References
Claidière, N and Sperber, D. (submitted). The
role of attraction in cultural evolution (Reply to J. Henrich and R. Boyd, “On
modeling cognition and culture”, Culture
and Cognition 2:65-112, 2002).
Dawkins, R. (1976). The selfish gene.
New York, Oxford University Press.
Dawkins, R. (1982). The extended
phenotype. Oxford: Oxford University Press
Hirschfeld, L. & Gelman, S. eds.
(1994). Mapping the Mind:
Domain specificity in cognition and culture. New York: Cambridge
University Press.
Odling-Smee, F. J., K. N. Laland, et
al. (2003). Niche
construction : the neglected process in evolution.
Princeton, N.J., Princeton University Press.
Richerson, P. J. and R. Boyd (2005). Not by genes alone
: how culture transformed human evolution. Chicago
; London, University of Chicago Press.
Sperber, D. (1975). Rethinking Symbolism. Cambridge: Cambridge University
Press.
Sperber, D. (1996). Explaining culture : a naturalistic approach. Cambridge,
Mass., Blackwell.
Sperber, D. and Claidière, N. (forthcoming) Why modeling cultural evolution is still such a challenge. In Biological Theory (1) 1.
Sperber, D. and Hirschfeld, L.
(2004). The cognitive
foundations of cultural stability and diversity. Trends
in Cognitive Sciences. 8 (1) 40-46.
[1] Richerson
and Boyd discuss some of the ideas of Sperber 1996 and underscore points of
agreement and disagreement. We believe that much of the disagreement is only
apparent and due to a misconstrual of Sperber’s exact view (for which, given
his own past miscontruals of Boyd and Richerson’s views, Sperber bears a good
part of the responsibility). In Claidière and Sperber (submitted), we discuss a
genuine point of disagreement reagarding the role of attraction in cultural
evolution.