Biol Philos (2008) 23:301–315 DOI 10.1007/s10539-007-9084-8

Response to our critics Peter J. Richerson Æ Robert Boyd

Received: 21 August 2006 / Accepted: 11 August 2007 / Published online: 21 November 2007 Ó Springer Science+Business Media B.V. 2007

Introduction We are grateful to all the authors for their thoughtful critiques. In no case have we had to deal with plain errors or rhetorical arguments. The issues raised are important and worth careful attention.

Comments of Stephen Shennan We thank Professor Shennan for his supportive appreciation of our approach to cultural evolution in Not By Genes Alone which, as he notes, summarizes a now considerable body of work tracing back to Donald T. Campbell’s pioneering essays in the 1960s and Luca Cavalli-Sforza and Marc Feldman’s first mathematical models in the 1970s. We want to underscore his point about the lack of dogmatism inherent in Darwinism and which we hope our work reflects. The Darwinian method is not committed to any particular explanation of any particular genetic or cultural trajectory. Darwinians do appreciate that the world is very complex, that many processes interact to explain any given example of history, and that some important processes are poorly understood. On the issue of niche construction, we are admirers of the work of Odling-Smee and his colleagues. They themselves have shown how genetic, cultural, and environmental factors can all be combined into a unified interacting evolutionary system. We can appreciate the frustrations that archaeologists experience in endeavoring to apply mathematical models of cultural evolution directly to their data. The behavior of the models typically turns on the details of how culture is transmitted, how costly it is for individuals to evaluate different alternative technologies or forms of social organization, P. J. Richerson (&) Department of Environmental Science and Policy, University of California Davis, Davis, CA 95616, USA e-mail: [email protected] R. Boyd Department of Anthropology, University of California, Los Angeles, CA 90095, USA

123

302

P. J. Richerson, R. Boyd

how large populations are, what migration rates are, how much diffusion of knowledge is inhibited by ethnic or other cultural boundaries, and the like. The archaeologist’s data speaks softly, usually very softly, on such topics. Sometimes it may be possible to fit models directly to archaeological data, but even the closest attempts to do this (null models aside) are heuristic rather than direct mathematical fits (Bettinger and Eerkens 1999; Eerkens and Lipo 2005). Direct fits of the models are more likely to be feasible when the data are more forthcoming, as in laboratory experiments (McElreath 2004) or contemporary field data (Henrich 2001). Some archaeological sequences with abundant data from stratified and datable deposits may support direct modeling (Eerkens and McElreath, personal communication). The archaeologist, like the paleontologist, would like to rely upon a body of quantitative empirical microevolutionary studies that at least guides the intuition in applying the model to the data-limited cases of the past. For the moment, such studies are it their infancy (McElreath 2004), though much of use can be gleaned from many classic bodies of data in the social sciences as we hope Not By Genes Alone shows. Much of our formal modeling work over the years has been inspired by bodies of social science data that are much in the spirit of the ideal cultural microevolutionary analysis, such as sociolinguistics and the study of the diffusion of innovations. However, practically no studies are exactly parallel to the theory, leading to merely qualitative tests of the theory (but see the discussion of the study by Hout et al. (2001) below). We think that archaeology ought to play to its strength and not (normally) try to fit microevolutionary models directly to its data. The comparative advantage of archaeology is control of the long time scale dynamics of cultural evolution and gene-culture coevolution. For example, many observers think that rates of cultural evolution have increased toward the present. Anyone can think up some good reasons why this might be true, but to our knowledge no one has ever estimated actual rates of cultural change as a function of time. Paleontologists have uncovered many fascinating patterns of change in the organic record. For example, organic diversity of readily fossilizable marine invertebrates has generally increased over the last 500 million years. Perhaps diversity has increased because the amount of shallow bottom near continents has increased as the continents have been fractured into smaller chunks with longer with longer coastlines and because over the last 65 million years, the polar seas have become cool, creating more temperature related niches than ancient seas with less temperature variation (Valentine 1973). Human evolution is rife with unexplained progressive trends on time scales ranging from millions of years (brain size increase) to thousands (the trend toward intensive plant use by huntergatherers in the Holocene, sometimes but not always resulting in agriculture). One of the most interesting macroevolutionary phenomena is that estimated evolutionary rates vary inversely with the time scale over which the rates are measured (Gingerich 1983). Rates of evolution measured by field biologists on the generation to generation time scale are orders of magnitude faster than the rates measured by paleontologists over millions of years. It is as if short-term evolution is mostly a random noodling about very slowly changing long-term evolution. Such a pattern is consistent with the sorts of selection pressures that field biologists measure. Peter and Rosemary Grant’s famous study of the evolution of beak depth of Galapagos finches found that beak depth increased during droughts as progressively harder seeds were all that were left in the seed bank for finches to eat. Presumably, the beak depth of finches increases and decrease as droughts and rainy periods come and go. Only over the long haul as mean rainfall increases or decreases will the fossils record any significant change on the geological time scale. Thus, the sense we have that current rates of cultural evolution are much more rapid than they

123

Response to our critics

303

were in ancient times could be an illusion. In the past, we can seldom measure rates of cultural evolution at the same scale we measure them in the present, so evolution in the past seems slower simply because time resolution in the past is lower than in the present. The Gingerich curve and the parable of the finches lead us to worry about Shennan’s idea of using random ‘‘null’’ models to separate selective processes from those that are due to random innovation and stochastic variation in small populations. We put ‘‘null’’ in quotes to signify our objection to privileging any particular model. Much existing empirical research and theory tells us that natural selection and bias forces are as real and as important as random variation and drift. If innovation rates were low, and population size constrained, it might be possible verify that only random variation and drift were important. But if not and if selection in the short run results mainly in random noodling, and if various decision-making forces more or less shadow selection, but operate on a faster time scale, the problem becomes altogether more difficult. This is often called the ‘‘inverse problem.’’ Given parameter values, models can be used to predict the aggregate behavior of the system modeled. For example climatologists attempt to predict future temperatures using climate models. This is the ‘‘forward problem.’’ However, sometimes we want to infer the parameter values of the model from data on aggregate behavior. For example, decide the relative importance of adaptation and drift from measured evolutionary trajectories. Sometimes the inverse problem is easy to solve, sometimes not. Indeed, in biology—cultural evolution is a species of biological problem in our view— both forward and inverse problems are often very hard. The dimensions of biological problems are enormous relative to the data we typically have to resolve parameter estimates. Such difficult problems as chaotic dynamics afflict prediction for good theories and the parameter estimates from good natural data. When archaeologists have relatively fine-grained data on the evolution of pottery styles or on the shapes of projective points, fitting an evolutionary model in order to try to solve the inverse problem is certainly worth a try. But randomness due to the random components of selection might be well, but spuriously, fit by a model that allows only random errors and stochastic drift, particularly if error rates and population sizes are poorly known. If we privilege a null model we will often accept it as true when it is not and reject an adaptive explanation when it is true. The less risky procedure is to fit all the plausible models we have to the data and see which, if any particular one, fits best. Each model is a hypothesis if you like and we want to pit them against each other. Lately, ecologists have made some useful conceptual and methodological advances in model fitting and model selection using ideas derived from application of information theory to statistics (Efferson and Richerson 2007). The basic idea is to start with the science we know. For example, students of stylistic evolution suggest that a common process that people use to modify their stylistic productions is in response to limited neophilia (Deutscher 2005; Martindale 1975). Audiences like something mainly familiar, but just a little edgy for the sake of novelty. Hence, old styles become boring and new ones gradually become incorporated into a population’s repertoire. The actual choice of new style may be random but is also likely to have selective components. Pigments are likely to be incorporated into glazes if they are neither too common (boring) nor too expensive (no customers for excessively expensive pots). The idea is to make a ‘‘full model’’ with all the terms that, based on existing knowledge, are reasonable potential explanations of the data. In the case of stylistic evolution, the full model might have four or five parameters to fit, including a term for random errors and one for stochastic drift. Typically maximum likelihood methods are used to fit models.

123

304

P. J. Richerson, R. Boyd

The fit of the full model and any reduced models that might also be plausible candidate explanations are compared using information theoretic goodness of fit statistics. The information theoretic approach aims to minimize the information loss in explaining data with models. Privileging any very simple model runs the risk of ‘‘underfitting’’ the data, of not efficiently extracting information actually in the data. Privileging a complex model like the full model runs the risk of ‘‘overfitting,’’ of treating the inevitable noise in the data as if it were information and drawing false conclusions for that reason. The information theoretic approach aims to minimize the risks of both types of error, to extract as much information as possible about the relative truth of the models on the table. Information theoretic statistics penalize models with larger numbers of parameters (lack of parsimony) in principled way; the full model is the least parsimonious and therefore carries the highest burden of proof. Simpler ‘‘null’’ models, with fewer parameters, have a parsimony advantage in the comparison, but a random variation and drift model is on an equal footing with a reduced two parameter model with terms for limited neophilia and selection based on cost. The entire family of models is scored by the information theoretic goodness of fit criterion. If the inverse problem is severe, the full model will surely fail due to its lack of parsimony, but more than one reduced model may fit equally well. One of the well-fitting models may be the best fit statistically, but have implausible parameter values, so the analyst has sound non-statistical grounds for eliminating it. But if we do face an inverse problem, we can discover that fact because no one model is clearly superior.

Sperber and Claidie`re Definitions of culture are always troublesome, but in the end we don’t think we disagree much with Sperber and Claidie`re. Culture as information in the head does have a sort of genotype to phenotype relationship to behavior and artifacts. The difference between genes and culture is that cultural information is transmitted via behavior and artifacts. We would insist that behavior and artifacts are essential ingredients of culture. They are necessary to transmit culture from mind to mind and they are necessary to make any real-world use of culture. In a way, this makes a science of cultural evolution easier than a science of genetic evolution. Behavior and artifacts are easy to observe whereas states of brains are hard to observe. Genes, until gene sequencing became cheap, were hidden from easy view and most of what we knew about evolution had to be based on phenotypic traits. Even now, both sciences are hard because mappings from brain states to behavior and artifacts and back again is very complex, as is the mapping of genes to phenotypes. Thus, we deal with artifacts, anatomy, physiology, and behavior rather than genes or states of mind because the former are observable and the latter are not. While much philosophical water has been hauled to try to float one set of conceptions and sink the other, the practice of science differs little if we adopt mentalism or behaviorism. As the mapping between brain states and behavior and artifacts is refined, definitions of culture are liable to have to be revised. In the meantime, since cultural transmission depends upon phenotypic performances, we can measure variation at this observable phase of the life cycle without going too far astray. Certainly the definition of the gene has changed with each revolution in genetics starting with the original Mendelian experiments that caused its initial coinage and continuing today with recent the recent discovery that DNA sequences coding for microRNA’s are possibly as numerous as protein coding sequences. Sperber and Claidie`re have misunderstood our conceptions of content based bias and guided variation. We would likely model the modification of an exotic recipe to suit local

123

Response to our critics

305

tastes as a content bias in favor of familiar local ingredients not guided variation, depending on details not in their characterization. The key distinction is how the novel recipe spreads, particularly when it is rare. Guided variation is driven by individual innovation. Suppose many local cooks try out the exotic recipe, find it strange-tasting and decide substitute a particular local ingredient for one of the exotic ones. Then, the spread of the hybrid recipe will not depend upon exposure to the new hybrid recipe, and therefore its rate of increase will be greatest when it is rare. As more and more cooks discover the new recipe, the number left to discover it dwindles and the rate of discovery drops. This is the signature of guided variation. Biased transmission is driven by differential rates of spread. Assume that only a few cooks hit upon a key substitution of an exotic ingredient by a local one, perhaps by chance, and then the hybrid recipe spreads because cooks are exposed to the new recipe and prefer it to the wholly exotic and to wholly local ones. When the hybrid is rare, few cooks will have the opportunity to observe it and the rate of increase of the preferred rare hybrid variant will be slow. As the hybrid becomes more common, more cooks will be exposed, and the rate of evolution will accelerate as more cooks get the opportunity to experience the hybrid and convert to it. Of course the two processes are not mutually exclusive; they can act together to shape the spread of a novel variant. At the psychological level, the same process can underlie both a bias and guided variation—the fact that the hybrid recipe tastes better than the local and exotic variants. The relative importance of the two processes will depend upon how likely are naı¨ve cooks to invent the hybrid on their own. Thus, we agree completely with Sperber and Claidie`re about preservation and construction being involved in all episodes of social learning. From the mathematical point of view, the parameters in a model with terms for guided variation and content bias (to keep the discussion simple) can take on any values we want, in principle. If the strength of the sum of all forces is small, accurate transmission can dominate. If the sum of all forces is large enough, transmission will not exist at all. The trait will be determined by genes or a learning rule guided by genes. Culture in the sense of socially learned information need not play a part at all. The relative strength of preservative and constructive processes is, of course, an empirical matter. Some human behaviors may indeed have little social influence and be almost entirely constructed by each individual. Sexual behavior in a puritanical society in which any discussion of sexual practices is taboo may be an example. Other traits, like complex technologies may be transmitted with only modest modification by any individual. The culture of other animals is interesting in this regard (Heyes and Galef 1996). Many other species have simple systems of social learning, but few if any other animals can imitate rapidly and accurately. Construction seems to be more important than preservation in such cases. Contrary to what Sperber and Claidie`re argue, we think that much of human culture is accurately preserved variation very modestly subject to construction at the level of individual psychology. Constructed culture species lack complex adaptations that are the cumulative product of many small constructive inventions piled one on top of another for, sometimes, millennia. Complex culture seems to be a human specialty and it seems to depend upon accurate preservation of a series of small, incremental, usually adaptive, constructions. Our set-piece example in NBGA is complex technology such as the kayaks of Inuit and their arctic relatives. No one individual invents such complex technology. Craftspeople imitate or are taught the techniques of knowledgeable elders, although of course individuals always introduce slight variations and modifications by accident and design. Kayaks varied from place to place in the arctic for both functional and non-functional reasons. No doubt, constructive processes play a role in boat-building traditions, but the

123

306

P. J. Richerson, R. Boyd

preservative aspects are strong enough that (1) complex boats evolve slowly, and (2) different boatbuilding traditions preserve fundamentally different designs. The Chinese junk is very different in almost every detail from the European wooden sailing ship, despite their being functional equivalents and the existence contact between the traditions (Needham 1979). So it is with language, social organization, art, and all the various domains of culture. They all contain many examples in which the primary role of individuals is accurate conservation. Experiments show that accurate, imitation is a peculiarly human skill (Tomasello 1996; Whiten and Custance 1996). The constructive elements of the equation in these cases are relatively weak forces that gradually shape traditions over many generations in populations of social learners. The mathematical models tell us that constructive decision-making forces can act a lot like natural selection. Feeble forces that a psychologist focused on individual behavior might not notice become a mighty evolutionary engine if they act consistently over many individuals and many generations. And ‘‘many’’ is surprisingly few to those who think that evolution is something for the geological time scale. Cultural evolution has great ‘‘reserve capacity.’’ If something leads to persistent decision-making in a consistent direction, populations can change smartly. The relatively small wooden ships of Medieval times, with a simple square rig and steering oars, evolved into the nineteenth century clipper ship over a period of half a millennium. The Plains Indians built a novel adaptation centered on the horse and the fur trade in a few generations. We take it for granted that no completely natural taxonomy of forces of cultural evolution will ever be found. Different authors will feel free to slice and dice such phenomena to suit their convenience. So we have no particular quarrel with the definitions of psychological and ecological forces of Sperber and Claidie`re. We do not make a particular distinction between psychological and ecological forces because we assume that biased transmission and guided variation are typically innately or culturally coded psychological rules open to environmental reinforcement. The decision to adopt a new recipe may sometimes depend only on our innate psychology. But, most likely it also depends upon our training and experiences. Most of the time the evolution of cuisines will be shaped by environment-dependent considerations of what nutritious items can be collected, hunted, or raised in a particular physical environment and the fads and fashions circulating in the social environment. In our models of guided variation and content based bias, we often assume that individuals have weak, relatively general-purpose learning devices. Kayak makers are not marine architects. Nevertheless they likely have a good practical sense of how a given kayak design slips through the water. To the extent that this is so, accurate conservation across generations greatly amplifies the power of constructions that are rare and small at the individual level. Fancy cultural adaptations may sometimes build up much like selection creates fancy adaptations using random variation as the only fundamental constructive process. However, that biased transmission can act selectively when innovations are guided by constructive systems means that the resulting combined process will generate adaptations more rapidly than if constructions were random. We think that swifter evolution is the basic adaptive secret of human culture and that weak relatively general purpose decision-making is often all that is required for a decisive advantage. The Inuit evolved their high-latitude maritime adaptation over just the past few thousand years. The range of environments in which humans live and the diverse means we have evolved for living in them have no comparison in the animal kingdom. The reason why adaptations like kayaks could not evolve by the means of strongly constructive domain specific modules is easy to understand, again with the help of a little

123

Response to our critics

307

math (Boyd and Richerson 1983). Domain specific, highly constructive cognitive systems are just ordinary genetic adaptations. They can adapt to new environments only by the relatively slow process of natural selection acting on genetic variation. The Inuit would have to evolve a module or modules specifically adapted to kayak building while the Polynesians would need a module or modules for outrigger canoe building. Domain specific modules are fine for the kinds of adaptive problems that pretty much the same everywhere, like copulating, or those that adapt specialists to relatively narrow but stable niches. Homo sapiens became, at the genetic level, the consummate generalist by using cumulative culture to leverage weak general purpose learning. Larger societies and literacy certainly enlarge the scope of such mechanisms, and extremely isolated populations such as the Tasmanians can lose complex adaptations (Henrich 2004), but contrary to what Sperber and Claidie`re argue, complex cultural adaptations were the basis of hunter-gather adaptation. Lumsden and Wilson (1981, 2006) focus on models in which cultural adaptation is strongly influenced by individual psychology and little by transmission from previous generations. This leads to rapid genetic evolution of learning mechanisms, what Lumsden and Wilson called the ‘‘thousand year rule,’’ because when little information is transmitted culturally, the work of adaptation must fall on genes. Indeed, in the introduction to the 25th Anniversary Edition of Genes, Mind, and Culture, Charles Lumsden recommends Sperber’s (1996) Understanding Culture. He endorses the massive modularity hypothesis, citing John Tooby and Leda Cosmides. Unlike most proponents of massive modularity, Lumsden also cites favorably data on individual genetic differences in behavior. Others argue that there are major between-group differences in mental abilities (Rushton 2000). The less information culture transmits and the more constructive cognitive systems are, the stronger will be selection on genes related to behavior and the more likely we are to observe genetic variation in different human populations for behavioral traits. Since human behavior has been revolutionized in just the last 250,000 years, and since we have spread out of Africa to the rest of the world in the last 50,000 years, many genes coding for constructive cognitive systems ought to have evolved over these time periods on the massive modularity account. On the other hand, if culture is mostly about preserving traditions and building upon them using weakly constructive general purpose learning, we need imagine few changes in the genes underlying constructive systems. Rather, recent selective changes in the genes active in brains should be modest. The changes we do see should be associated with added mental capacities to acquire and accurately preserve culture. We might also expect to see recently coevolved genes that came under selection due, for example, to the rapid increase in human population size and heavy contact with domestic animals and plants after the origins of agriculture. Agricultural population densities led to the emergence of new epidemic diseases and diets based on agricultural staples required alteration of human physiology to cope with novel nutrition problems. Sabeti et al. (2006) review the evidence on recent selection for about 90 genes and gene families. While the methods employed to detect selection are very recent, the available sample provides a tantalizing glimpse of what we will know with much greater certainty over the next decade or two. Many of the genes having undergone selection very recently appear to be coevolved adaptations to the dietary and disease changes associated with the switch to agricultural subsistence beginning about 10,000 years ago. An example is the LCT gene that codes for the continued production into adulthood of the enzyme that makes milk sugar digestible. Genes that code for disease resistance, particularly to malaria, are numerous. One gene related to brain size is in this very recent category as is a gene related

123

308

P. J. Richerson, R. Boyd

to sensation-seeking personality and attention deficit disorder. Other genes related to brain size and nervous system development apparently evolved in the late Pleistocene. The FOXP2 gene a pathological variant of which affects speech, is an example. Genes related to olfaction and taste also fall into this age range. Rather older genes that evolved over the last few million years are related to reproductive function, taste, and pheromone reception. The functional significance of most of these examples of selection is so far only hinted at by evidence of the effects of pathological variants of the alleles. One can imagine that the recently evolved genes affecting brain development, personality, and sensory modalities signal the evolution of new or modified modules that play a heavily constructive role in culture. Or one could imagine them to be enlargements of the brain to permit accurate imitation and the storage of large amounts of social transmitted information. Genes related to language could be highly constructive cognitive modules, or mainly sensory, motor, and information storage areas related to the need to decode speech rapidly, control the vocal apparatus with precision, and store a large vocabulary. The changes in olfaction and taste could be feeding information into modules that play a strong constructive role in diet, mating strategies and the like. Or they might represent the abandonment of precision sensory information collecting as the enlarging culture processing areas of the brain cannibalized expensive and partly redundant sensory capacities. For example the development of cooking and culturally transmitted cuisines would have reduced the need for innate sophistication about plant toxins and meat spoilage. Like Sperber and Claidie`re, we see no reason to be dogmatic about these issues. Molecular biologists are training the big guns of brute force empiricism on these functional questions. We look forward to seeing whether culture-driven coevolution or the gene-driven evolution of strongly constructive modules, if either, turns out to dominate recent genetic evolution. We have elsewhere discussed the problem of distinguishing between simple communications and cultural transmission (Richerson and Boyd 1997). We agree with Sperber and Claidie`re that a continuum exists between the communication of ephemeral interest like the price of bread in a given store on a given day and the durable concepts, skills, and attitudes that we normally think of as culture. The evolutionary models and concepts do not necessarily become useless as the things communicated become more ephemeral, as we can shrink the time scale of the models arbitrarily if we want to study the spread of rumor or even panic in crowd. The relationship between ephemeral communication and cultural transmission may be quite complex. Ephemeral communications do have durable cultural content, if only the meanings of the words used in the communication. The meaning of a word or a ritual may be built up by aggregating the effects of many ephemeral communications. Similarly, the relationship between highly idiosyncratic bits of information and widely distributed and largely shared items of culture is complex. Idiosyncratic mental representations are irrelevant to cultural evolution to the extent that they are not manifest in behavior that forms the basis of social learning. For example, two people might have qualitatively distinct grammatical rules, but if these differences lead to similar sentences, the qualitative differences might lead to only minor quantitative differences or none at all on the part of social learners who acquire the rule or model. Private variation is a little like the silent changes in genes deriving from the redundancy of the triplet code for amino acids and public variation is analogous to the differences in the triplet code that lead to amino acid substitutions in proteins. Much more important are those that affect public performance and hence can be preserved as a cultural variant. Presumably, cultural variants that are now common began as some individual’s brainstorm or random mistake.

123

Response to our critics

309

Simpson and Beckes Like Simpson and Beckes, we regret the absence of mathematical models in Not By Genes Alone. In an ideal world every college educated adult would be able to handle a little math. However, the fact is that extensive mathematical discussions intimidate and distract many interested readers, including many who are doing useful scientific work. Even the thoroughly numerate find the mathematical exposition of cultural evolution hard slogging until they are familiar with conventions of population biology. We did reference the mathematical models on which so much of our intuition about cultural evolution is grounded. Mathematical readers can treat the book as an annotated bibliography. Simpson and Beckes provide a good review of the currently viable hypotheses regarding the evolution of human ultra-sociality. We have rehearsed our arguments on this topic in more detail elsewhere than in Not By Genes Alone (Richerson et al. 2003). We agree that the issue of the role of self interest versus group interest to be badly entangled. Selection at the individual level on genes probably remains quite important in humans. In an imperfectly group selected system, altruistic individuals will want to be careful of their reputations in order to avoid altruistic punishment and garner whatever rewards are due to good behavior. The hardened individualists among us will deceptively cultivate the best reputations they can to avoid punishment and garner undeserved rewards. All but the saintliest of us take holidays from public virtue to take care of our families and indulge ourselves. Reluctant cooperators will be induced by rewards and coerced by the threat of punishment. The complex mixture of individualism and altruism in human behavior makes human social life endlessly fascinating—most fiction and much history hold our attention by trading on the conflicts the mix engenders. In Richerson et al. (2003) we argue that circumstantial evidence from several consilient domains implicate a measure of cultural group selection for strong reciprocity as the foundation of human ultra-sociality (see also Fehr and Fischbacher 2003). In economics and evolutionary biology an individualist dogma long held sway while in the rest of the social sciences a sort of naı¨vely group selectionist functionalism was common. We sense a more sophisticated and productive debate is becoming possible as more people begin to realize that the questions are complex and hard to settle and that both dogmatism and naı¨vete´ have slowed the search for a solution to the puzzle of human cooperation. The archaeological and paleoanthropological records give up information about the evolution of human social instincts very reluctantly. However, we are not entirely dependent upon ethnographic examples. As we argued in Not By Genes Alone pp. 225 ff, the Upper Paleolithic peoples of Europe made stone and bone tools on the same order of complexity as living hunter-gatherers. Their cave art and ivory carvings bear comparison with that made by the socially most complex hunter-gatherers, such as the people of the Northwest Coast of North America. Raw materials were moved much greater distances than in previous times and some art styles are very widely distributed, as in ethnographically known systems. The Upper Paleolithic people either traded with one another across any social boundaries that existed or were able to move freely over large areas. Either requires protection by institutions that mitigate predatory violence on travelers and traders. Either would be evidence of social sophistication on a considerable scale. We have strong reason to think that artifact complexity will reflect the size of populations in which people are in regular contact. The Tasmanians (perhaps *4,000 people) lost most complex artifacts, including boats and tailored clothing, after they were isolated from Australia by Holocene sea level rise (Henrich 2004). The social simplest ethnographic hunter-gatherers—the Great Basin Shoshone are a commonly cited example—are both very late in time

123

310

P. J. Richerson, R. Boyd

and have considerably more sophisticated institutions than is allowed by those who want to make them do service as institution poor models of Upper Paleolithic peoples (Murphy and Murphy 1986). We find it hard to imagine how language can be the only or original foundation of complex social life, necessary as it is for that life. Ask the origin of language question in the negative form: not why do we have language, but why do no other species have it? All sorts of animal communication systems exist that could serve an origin point for language. Chimpanzees can achieve a respectable symbolic vocabulary—when taught by humans. We think that the answer is that most animals have no use for language. Within the family and other very small cooperative groups, individuals and their routines are too familiar to require sophisticated language. Outside kin groups, most animals could not trust their conspecifics to say anything that would not be self-serving. Like Ulysses and his crew confronting the Sirens, most species are better off stopping their ears to anything their conspecifics have to say. Only when you can have some reasonable confidence that another person will tell you something good for you not just good for them could language evolve. If complex languages are like other complex cultural systems, they will require a good many speakers to maintain their complexity. Language depends upon a goodly dose of altruism on the part of speakers distantly related or unrelated to hearers. Somehow, the evolution of human cooperation generally, and of language specifically, worked around this problem. It is uncertain how this happened. Simpson and Beckes’ idea that it started in small groups and somehow got extended to large ones makes sense. We’ve speculated among ourselves that cultural group selection would have begun by marginally expanding kin altruism. Perhaps a population with the most rudimentary culturally transmitted social institutions could maintain a rule that tricked our evolved psychology so that close cousins were treated as if they were brothers, and that actual brothers acted as if they were more related than ½. This little clan would tend to out-compete clans hewing to more conservative nepotistic rules. Such social units might punish those who deviate from the institution frequently enough to induce some genetic coevolution, which in turn would permit slightly larger scale clans to arise by more inclusive culturally transmitted rules. Tribes composed of a 100 or more families, together with the necessary social instincts to organize cooperation among quite distantly related folks, perhaps resulted from hundreds of thousands of years of such coevolution. This argument requires that cultural evolution can occur without language, a possibility discussed by Merlin Donald (1991). The capabilities of nineteenth century deaf-mutes are fairly well known. Their social and technical handicaps were much less than one might suppose, supporting Donald’s argument that humans were able to advance quite a way toward a modern mind using memetic capacities but no spoken language at all. Given that scholars are highly advanced language users, we perhaps are too quick to assume it essential even at the very beginnings of the evolution of complex sociality. We need to think harder about how much social complexity might have preceded language as we know it. Looked at this way, language itself is one of the best bits of evidence for the tribal social instincts hypothesis. Note that languages in simpler societies are tribal scale affairs. Different tribes typically spoke different languages, or at least different dialects, as if language has evolved to stop our ears from understanding those duplicitous foreigners who don’t have our interests reasonably at heart. The African evidence suggests that humans there evolved the Upper Paleolithic tool kit over a period of perhaps 200,000 years between 250,000 and 50,000 years ago (McBrearty and Brooks 2000). Stylistic variation in tools

123

Response to our critics

311

became apparent in Africa perhaps 100,000 years ago. Highly symbolic artifacts like statuettes apparently appeared only shortly before Africans left Africa to become global citizens. Regardless of how social units of a few 100 to a few 1,000 people demarcated by language and other symbolic markers arose, once the existed they would have the effect of protecting considerable between group variation. How could group selection on such entities not be a significant force? Given that making hypotheses about exactly how human capacities arose in the late Pleistocene is easy and testing them is hard, how can we make any headway at all? No one sort of data constrains speculation very much. In the last chapter of Not By Genes Alone we advocate a strategy that takes advantage of consilience and subjects evolutionary hypotheses to tests in as many domains as possible, probing their logical consistency with mathematical models, looking for the proximal mechanisms that are implied, checking the microevolutionary and macroevolutionary evidence, and assessing the pattern of adaptations and maladaptations that ought to flow from the hypothesis. For example, we can produce fairly good microevolutionary data suggesting that cultural group selection acted in the ethnographically known situations (Soltis et al. 1995; Kelly 1985). We submit that the tribal social instincts hypothesis passes a lot of challenges in all these domains. When alternate hypotheses are better developed, this debate will get more interesting. Our response to Simpson and Beckes arguments under the heading of strong evidence for the ‘‘special design’’ of psychological mechanisms is much the same as to Sperber and Claidie`re above on the issue of construction versus transmission of culture. We don’t have any disagreement with Simpson and Beckes regarding such adaptations as snake fears. Dangerous snakes have been an adaptive problem for most mammals since the their evolution. The fossil record of snakes is poor but they are thought to have radiated along with the mammals beginning 65 million years ago. Venomous snakes are ambush predators specializing on fast-moving mammalian prey. They also use the threat of envenomation as a defense against predators like humans. Our homeland, Africa, has plenty of nasty serpents today. So it seems likely that fear of snakes must have been under selection in our quite remote ancestors. This is exactly the opposite of the much more ephemeral, localized adaptive problems that we think human culture evolved to solve.

Mameli We are pleased that Mameli finds so much food for philosophical thought in the theory of cultural evolution. Mameli’s characterization of our view of how cumulative cultural adaptations arise is slightly different from our own. We certainly think that prestige/success based and conformity biases are often important, that much new individual level contains a strong random element, and that cultural transmission is often noisier than genetic transmission. However, we also think that individual learning and choices among cultural variants based directly on the properties of the variants are probably important as well. Natural selection on cultural variation may sometimes be appreciable. We also think that on these points our ignorance is vast. We can point to much literature in the social and behavioral sciences indicating that all of these effects are real. However, quantitative studies of the relative importance of the various forces in the field are very few. Exceptions include work by for example by sociolinguists (Labov 2001), and students of the diffusion of innovations (Rogers 2003), and sociologists of religion (Hout et al. 2001) We certainly agree with Mameli that considerable work will be required before we

123

312

P. J. Richerson, R. Boyd

can make confident generalizations about the evolutionary forces acting in different domains of culture. We certainly agree that biases have costs. The accuracy with which biases find adaptive cultural variants likely covary with time and cognitive effort expended in assessing them. Cheaper to apply biases like conformity and prestige can indeed behave quite pathologically under plausible assumptions, perhaps leading to such things as runaway status competitions, such as those that afflict modern consumers. The chapter Culture Is Maladaptive in Not By Genes Alone outlines this and several other processes that are prone to generate maladaptations. Invention of one’s own variants (guided variation) or direct evaluation of the functionality of alternate cultural practices (direct bias) are costly to exercise and, unless they are accurate, run the risk of misjudgment. We also agree with Mameli and Sterelny that group selection may pay part of the cost of biases. Individual inventors, authors, scientists and artists expend considerable effort producing their creations. The whole society enjoys benefits from them. Absent some reward, their creative effort will be suboptimal from society’s point of view; an inventor whose innovation is easily copied may earn only the tiniest fraction of its total value to humanity in general. Group selection is the crudest form of reward for creative work. An optimal patent and copyright policy is a more sophisticated process to the same end. The prestige system is even more sophisticated. Prestige often confers some material benefits on the prestigious but it also exploits people’s sociability if not their vanity (Henrich and Gil-White 2001). Praise and admiration are potent rewards. We apparently have left the misimpression that cultural transmission is typically costly and inaccurate. While we do think that cumulative culture evolution can occur when social learning falls short of the low-cost, high accuracy standards of genetic transmission (NBGA, pp. 80–86), we agree that the fidelity of cultural transmission is probably highly variable and that kin selection (including that due to diffuse relatedness in larger groups) will favor accurate teaching. Word learning is an example of fairly accurate cultural transmission at fairly low cost. Normal language learners acquire tens of thousands of words with enough accuracy to permit reasonably efficient communication. Along the way a good deal of culture is acquired along with word meanings. Word learning is costly compared to genetic transmission. Mammals the size of humans can become sexually mature in a couple of years rather than the 20 that we require. Our adult brains consume something like 15% of our total energy intake versus something like 5% for the average modern mammal. These are crude measures of the costs imposed by the cultural system. Interestingly, the evidence suggests that in ancestral societies adults invest little effort in teaching. In traditional societies task groups will generally be kin-based, so much cultural transmission is liable to be ‘‘extended’’ vertical, as Mameli suggests. However, children in ethnographically known traditional societies imitate adults in their play, and as soon as their play becomes sufficiently sophisticated, they are incorporated into adult task groups, doing whatever part of the task they are up to contributing to (Rogoff et al. 2003). Their learning in these task groups seems to be passively tolerated as much as actively encouraged. The formal, costly teaching of the young characteristic of modern societies seems to be the exception not the rule. We know of no evidence that oblique and horizontal transmission are systematically noisier that vertical. Abundant evidence suggests that horizontal and oblique transmission were important in traditional societies. Many cultural traits in traditional societies are characteristic of ‘‘culture areas’’ that are sub-continental (California) or even continental in scale (Australia). Adaptive subsistence techniques, attractive art styles and forms of worship, and words and grammatical constructions apparently readily diffuse among

123

Response to our critics

313

societies with modest regard for linguistic boundaries. Speakers of Dene languages in North America were fishing specialists on the Northwest Coast, boreal forest big game hunters inland in Alaska and Canada, salmon fishers and acorn processors in California, and maize farmers in the Southwest. Of course, intermarriage combined with largely vertical transmission could diffuse traits widely if more slowly than is the case if horizontal and oblique transmission are involved, especially if selection or strong biases favor the same trait in many different cultures. Hewlett and Cavalli-Sforza’s (1986) pioneering study of the learning of hunting techniques among the Aka Pigmies mentioned by Mameli is misleading in this regard. Young Aka boys learned hunting techniques from their fathers and other extended vertical kin because these were the first task groups to incorporate them. But for most techniques, the population had no variation and so older youths had nothing new to learn by horizontal or oblique transmission. An important exception was recently introduced crossbows. Crossbow use was generally learned by horizontal and oblique transmission because most young men’s close male relatives were unfamiliar with them. Language is another interesting case. Although primary caretakers are undoubtedly very important in language learning, kids end up adopting the language of peers not parents. Students of the diffusion of innovations tell us that new techniques with obvious adaptive advantages spread rapidly between societies by horizontal and oblique transmission. In the historical period, stone age peoples avidly adopted steel tools and many other European manufactures. In the American West, horse riding spread from northern Mexico to the Canadian Great Plains in perhaps a century. Mameli concludes that a combination of accurate extended vertical transmission and natural selection are perhaps the main motors of cumulative cultural evolution, at least during the Pleistocene when our capacity for culture arose. Such a system would bear a burden of opportunity costs, in the jargon of economists. In the limit, such a system becomes as conservative as genes. It would not be able to take advantage of the kind ephemeral niches that are available to a system that takes more advantage of horizontal and oblique transmission combined with cheap biases like prestige and conformity. However, it would explain the apparent conservatism of even Upper Paleolithic cultures. Our own view is that high amplitude millennial and sub-millennial scale climate variation in the late Pleistocene put a premium on adaptation to a dynamic environment using a relatively bold use decision-making forces and non-vertical transmission. Of course, structure of the cultural evolutionary system is itself culturally variable, and so it may have been in the last glacial. In quieter periods cultural evolution should tend at least relatively to more vertical transmission and weaker effects of decision-making forces. In more dynamic periods it should depart in the opposite direction. But we admit that the apparent conservatism of Pleistocene lifeways may represent niche chasing, in which relatively specialized big game hunters exploited a quite limited subset of Pleistocene environments. The wide distribution of the Mousterians, Aurignacians, and Gravettians might have been produced by conservative people following a few favorable niches as climate change drove them about on the dynamic landscape. As we discussed in response to Shennan’s comments, we are still far from being able to fit models to archaeological data. Dating problems even in the Upper Paleolithic preclude placing sites in time with enough resolution to resolve even multi-millennial scale changes. Perhaps the more dynamic sort of cultural system has operated only in the Holocene and especially modern times. Certainly, we seem poorly adapted to the explosion of horizontal and oblique transmission brought about by mass literacy and mass media (Newson et al. 2005). Our intuitions and our interpretations of the data may be badly calibrated as a result of viewing the past through the lens of the present.

123

314

P. J. Richerson, R. Boyd

Conclusions Most of the issues we and our commentators have debated boil down to quantitative empirical issues. To date, most of the empirical data available are qualitative. This is especially so for the deeper past and the ethnographic sample of contemporary human variation. For want of high quality data, we modelers have been beguiled, and no doubt in some part misled, by a few neat theoretical results. We need to move forward the agenda of what we call The New Quantitative Ethnography. We need good quantitative descriptions of how cultural evolution proceeds in a goodly diversity of field studies. We need better descriptive instruments. Recently, experimental games have been deployed in field settings to make theory-driven quantitative observations (Henrich et al. 2004). We need much more data from experimental models of cultural transmission and evolution. The neurobiological basis of human culture is sure to be interesting (Rizzolatti 2005). Many macroevolutionary questions remain open for historians, archaeologists and paleoanthropologists. Paleoecology is not yet strong enough to translate the very strong climate fluctuations of the last ice age into even a roughly reliable estimate of variation in human carrying capacity as a function of climate forcing (Huntley and Allen 2003). When substantially more data is available, some of us will turn out to be prophets and others goats. Who will be which will be mostly a matter of luck. We know so little, yet progress is visible on so many fronts!

References Bettinger RL, Eerkens J (1999) Point typologies, cultural transmission, and the spread of bow-and-arrow technology in the prehistoric Great Basin. Am Antiq 64(2):231–242 Boyd R, Richerson PJ (1983) Why is culture adaptive? Q Rev Biol 58:209–214 Deutscher G (2005) The unfolding of language: an evolutionary tour of mankind’s greatest invention. Henry Holt, New York Donald M (1991) Origins of the modern mind: three stages in the evolution of culture and cognition. Harvard University Press, Cambridge, MA Eerkens JW, Lipo CP (2005) Cultural transmission, copying errors, and the generation of variation in material culture and the archaeological record. J Anthropol Archaeol 24:316–334 Efferson C, Richerson PJ (2007) A prolegomenon to non-linear empiricism in the human sciences. Biol Philos 22:1–33 Fehr E, Fischbacher U (2003) The nature of human altruism. Nature 425:785–791 Gingerich PD (1983) Rates of evolution: Effects of time and temporal scaling. Science 222:159–161 Henrich J (2001) Cultural transmission and the diffusion of innovations: adoption dynamics indicate that biased cultural transmission is the predominate force in behavioral change. Am Anthropol 103(4): 992–1013 Henrich J (2004) Demography and cultural evolution: Why adaptive cultural processes produced maladaptive losses in Tasmania. Am Antiq 69(2):197–221 Henrich J, Gil-White FJ (2001) The evolution of prestige—freely conferred deference as a mechanism for enhancing the benefits of cultural transmission. Evol Hum Behav 22(3):165–196 Henrich J, Boyd R, Bowles S, Camerer C, Fehr E, Gintis H (2004). Foundations of human sociality: economic experiments and ethnographic evidence from fifteen small-scale societies, xix + 451 ed. Oxford university Press, Oxford Hewlett BS, Cavalli-Sforza LL (1986) Cultural transmission among Aka Pygmies. Am Anthropol 88: 922–934 Heyes CM, Galef BG (1996) Social learning in animals: the roots of culture. Academic Press, San Diego Hout M, Greeley AM, Wilde MJ (2001) The demographic imperative in religious change in the United States. Am J Sociol 107(2):468–486 Huntley B, Allen JRM (2003) Glacial environments III: paleo-vegetation patterns in last glacial Europe. In: van Andel TH, Davies W (eds) Neanderthals and modern humans during the last glaciation: archaeological results of the stage 3 project. McDonald Institute for Archaeological Research, Cambridge

123

Response to our critics

315

Kelly RC (1985) The Nuer conquest: the structure and development of an expansionist system. University of Michigan Press, Ann Arbor Labov W (2001) Principles of linguistic change: social factors. In: Trudgill P (ed) Language in Society, vol. 29. Blackwell, Malden, MA Lumsden C, Wilson EO (2006) Genes, mind, and culture: the coevolutionary process. 25th anniversary edition. World Scientific, New Jersey Martindale C (1975) Romantic progression: the psychology of literary history. Hemisphere, Washington, DC McBrearty S, Brooks AS (2000) The revolution that wasn’t: a new interpretation of the origin of modern human behavior. J Hum Evol 39(5):453–563 McElreath R (2004) Social learning and the maintenance of cultural variation: an evolutionary model and data from East Africa. Am Anthropol 106(2):308–321 Murphy RF, Murphy Y (1986) Northern Shoshone and Bannock. In: d’Azevedo WL (ed) Handbook of North American Indians: Great Basin. Smithsonian Institution Press, Washington, DC Needham J (1979) Science in traditional China: a comparative perspective. The Chinese University Press, Hong Kong Newson L, Postmes T, Lea SEG, Webley P (2005) Why are modern families small? Toward an evolutionary and cultural explanation for the demographic transition. Pers Soc Psychol Rev 9:360–375 Richerson PJ, Boyd R (1997) Types of transmission: a taxonomy of cultural inheritance systems. In: Weingart P, Mitchell SD, Richerson PJ, Maasen S (eds) Human by nature: between biology and the social sciences. Lawrence Erlbaum, Mahwah NJ Richerson PJ, Boyd R, Henrich J (2003) Cultural evolution of human cooperation. In: Hammerstein P (ed) Genetic and cultural evolution of cooperation. MIT Press, Berlin Rizzolatti G (2005) The mirror neuron system and imitation. In: Hurley S, Chater N (eds) Perspectives on imitation: from neuroscience to social science. MIT Press, Cambridge, MA Rogers E (2003) The diffusion of innovations. The Free Press, NY Rogoff B, Paradise R, Arauz RM, Correa-Chavez M, Angelillo C (2003) Firsthand learning through intent participation. Annu Rev Psychol 54:175–203 Rushton JP (2000) Race, evolution, and behavior: a life history perspective, 3rd edn. Charles Darwin Research Institute, Port Huron, MI Sabeti PC, Schaffner SF, Fry B, Lohmueller J, Varilly P, Shamovsky O, Palma A, Mikkelsen TS, Altshuler D, Lander ES (2006) Positive natural selection in the human lineage. Science 312:1615–1620 + Supplementary Material online Soltis J, Boyd R, Richerson PJ (1995) Can group-functional behaviors evolve by cultural group election? An empirical test. Current Anthropol 36(3):473–494 Sperber D (1996) Explaining culture. Blackwell, Oxford Tomasello M (1996) Do apes ape? In: Heyes CM, Galef BG Jr (eds) Social learning in animals: the roots of culture. Academic Press, New York Valentine JW (1973) Evolutionary Paleoecology of the marine biosphere. Prentice-Hall, Englewood Cliffs, New Jersey Whiten A, Custance D (1996) Studies of imitation in chimpanzees and children. In: Galef BG, Heyes CM (eds) Social learning in animals: the roots of culture. Academic Press, New York

123