Many scientific disciplines have a ‘paradigmatic experiment’, a simple example of how the discipline is applied to a problem. Usually, the paradigmatic experiment is like a light that, if switched on, suddenly illuminates the rest of the field. Mendel’s pea crosses are the paradigmatic experiment for classical genetics. Sometimes, for disciplines where direct experimentation is a little harder, there is not so much a paradigmatic experiment as a paradigmatic example. For instance, the body of data relating to the distribution of sickle cell anaemia is a paradigmatic example of the process of molecular evolution in human populations. Such paradigmatic instances are often found in textbooks, and with good reason, given the purpose of textbooks. Sometimes they also constitute the founding experiment in the field, but they need not do so; Mendel’s pea crosses were the founding work in genetics, but the sickle cell anaemia story only gradually emerged from a large body of research on biochemical variation in human populations.
So what is the paradigmatic example in memetics? My own favourite is the ‘homebodies vs. the hellraisers’ model used as an introductory vehicle by Boyd & Richerson (1985). Homebodies raise large families of more homebodies. Hellraisers by contrast, tend to have fewer progeny, themselves also of a hellraising tendency. The hellraisers would be heading straight for the evolutionary graveyard, if it were not for one factor: put a homebody and a hellraiser together for any length of time, and the homebody can convert into a hellraiser. Thus, depending on the frequency and strength of social interaction, and the relative reproductive differential of the two groups, hellraising can persist in an often fragile equilibrium.
This simple model is attractive because it encapsulates so many of the themes of the memetic approach to cultural evolution. Notice that the nature of the tendency to homeliness or hellraising is left open. It is possible to conceive at least part of the model in genetic terms: each group has progeny of its own kind because of an innate disposition. Alternatively, one might see homeliness or hellraising as purely cultural attributes, in which case the tendency of progeny to acquire the same phenotype as their parents is merely due to prolonged contact. What is certain is that the conversion of homebodies into hellraisers must be purely cultural. Thus the model contains what Cavalli-Sforza & Feldman (1981) defined as ‘horizontal’ and ‘vertical’ transmission of traits. The unidirectional nature of the horizontal transmission also illustrates ‘cultural selection’. The greater fecundity of homebodies illustrates ‘natural selection’. By subtly varying the parameters of each of these processes, Boyd & Richerson (1985) display several equilibrium states.
I call this a paradigmatic example, because if one can fully understand this simple model, and appreciate its sensitivity to parameter changes, the essence of most current work in cultural evolution and gene-culture co-evolution can be grasped. The problems also become apparent. Just what is the relevance of such a model to the real world? For those who take a co-evolutionary approach, ie. those who try to see such problems as involving genes as well as memes, there remains the problem of how, at a mechanistic level, genes influence behaviour. Evolutionary biology and sociobiology are not yet fully integrated into molecular genetics. To express it in Aristotelian terms, genes may, for some, provide a convincing ‘material cause’ for behaviour, but they have not yet been directly demonstrated to provide an ‘efficient cause’. Even for those who are happy to see the example as one that involves exclusively cultural traits, there remains the problem of how to ascertain the types for the individuals. In the model, the phenotypes are defined solely in terms of the tendency to have a certain number of progeny. In real life, the number of progeny is a normal distribution, and discrete categorization may be difficult. A further problem is why certain traits are culturally selected over others. Even if one abandons that concern to the psychologists, and simply accepts cultural selection as a given, what grounds are there for assuming that it will remain at a constant level, or not even go into reverse if other conditions change?
Because of these kinds of difficulties, the main stage for memetics research has been on the computer rather than in the wild. Software objects, or ‘agents’, can be programmed to behave in ways that make it possible to investigate the basic propositions of memetics. They can be given, in the jargon of object-oriented programming, attributes which are discrete values, and methods of manipulating those attributes, and the attributes of their fellows, in ways that correspond to strict rules set by the programmer. The homebodies and hellraisers of the model can be exactly reconstituted as software objects, with their parameters of reproduction and interaction precisely defined. The programmer can then easily expand the model in various ways; perhaps the external environment in which the agents exist can be partitioned into regions, each agent allowed a geographical location, and the parameters of reproduction and interaction varied in different regions. The agents might then be allowed to move, to seek out others of their own type as neighbours, or to culturally interact over long distances. One can even introduce elements of game theory into their actions, to allow them to negotiate and to deceive each other. When compared with the messy reality of human social life - difficulties of accurate categorization of individuals, uncertainty about what is cultural and what is genetic, even a lack of agreement about what culture is (eg. see Kuper 2000) - it is easy to see the appeal of artificial life as a laboratory for memeticists.
However, not everything is orderly in cyber-societies. Each small modification to the basic model adds another layer of complexity, and it is often easy to find oneself with a model that exhibits behaviour that one does not understand. Artificial societies can be just as perplexing as real ones. In the real world, we don’t completely understand individuals, so we are not particularly surprised when we don’t understand the societies they form. But in the artificial world, we do understand the individuals. We understand them perfectly, since we designed them. Yet, put them into social groups and inexplicable things can start to happen.
These caveats are not just peculiar to the creation of memetic models in artificial societies, but are well recognised in the larger social simulation community. They do not constitute an objection to the method in principle, but merely a warning against placing too much faith in it. So, in summary, what methodology can be used for memetics in the 21st century? I would recommend the following.
Since positive results only give indications that the real world could be as you have modeled it, whereas negative results give much stronger indications that the real world is not the same as the model, it seems that this methodology is actually quite Popperian. The function of the memeticist is to cast doubt on theories rather than to confirm them.
Is this not however, just an argument in favour of casting off the label of memeticists, and plunging into the social simulation community? Not quite; I propose that memeticists can form a recognizable sub-community by our adherence to the following requirements:
What are those areas which have reasonable bodies of quantitative data? One obvious one is politics or any situation where human choices are recorded by ballots (eg. Gatherer 2004). Another one is anthropology where there are extensive ethnographic databases (eg. Gatherer 2002b). A third is demography where censuses of population movement are available. There are many more. The best way to find one is to ask a social scientist, geographer or historian.
As well as my two published efforts in this direction (Gatherer 2002b, 2004), let me give one other preliminary example before finishing. There has been a lot of interest within the memetics community in the issue of terrorism (eg. Marsden 2001; Lynch 2002). It was even briefly alluded to by Dawkins (1976). Memeticists often seem convinced that this is an area where the approach can make some kind of contribution. There is also a wealth of literature on the subject from other academic disciplines, including social simulation (Raczynski 2004; Eidelson & Lustick 2004), and terrorist behaviour has been variously assessed as a form of reactive warfare, organised crime, or psychopathology. There is, in at least one case, an impressive body of quantitative data: the CAIN database on the conflict in Northern Ireland from 1969-2001, available from the University of Ulster (http://cain.ulst.ac.uk). This database contains details on every politically motivated killing (over 3600 of them) during the 32 year period of its coverage, as well as data on the political and economic background. Figure 1 shows the number of murders attributed per year to Loyalist and Republican paramilitaries.
Figure 1: Number of killings per year attributed to Loyalist and Republican paramilitary organizations from 1969 to 2001
How do existing memetic models of terrorism stand up against such a data pattern? Is it possible to design a set of agents who would attack each other in such a way? Alternatively, is it possible to design sets of agents based on existing models who then fail to attack each other in such a way, thus casting doubt upon the model? Lynch’s model (2002) only really goes as far as describing a variety of reasons, of varying degrees of cogency, why terrorism is likely to spread. One might say that Lynch is simply positing that terrorism is, in the standard terminology, culturally selected. Unfortunately, under simulation conditions, a model that simplistic merely leads to terrorism running to fixation within the population under the dynamics of a sigmoid curve (the “contagionist paradigm”, see Gatherer 2002a). In effect, the agents simply kill each other off completely. The data in Figure 1 require dynamics that would explain the waxing and waning of the Republican kill rate at approximately 2-year intervals, at least during the period 1972-1989, while the opposing Loyalist kill rate was relatively unchanged year-on-year from 1979-1991. Looking not just at the overall distribution of murders, but also at their individual timing relative to each other, using a Run Test, suggests that a tit-for-tat mechanism may have been in place at some phases (data not shown). Thus a good memetic model for terrorism would not merely concentrate on trying to explain cultural selection, but would also require some game theory to be incorporated.
In summary, memetics can survive as a data-oriented branch of the social simulation field specializing in the evaluation of co-evolutionary or cultural models. Most of that evaluation will be negative, as the nature of the real world can never be confirmed using a computer model. Nevertheless, it may help us to identify cases where our a priori thinking about a cultural phenomenon is inadequate. Karl Popper would have been pleased with this.
Boyd R & Richerson PJ (1985) Culture and the Evolutionary Process, The University of Chicago Press.
Cavalli-Sforza LL & Feldman MW (1981) Cultural Transmission and Evolution. Princeton University Press, Princeton.
Dawkins, R. (1976) The Selfish Gene. Paperback edition 1978. Granada Paladin: London.
Eidelson BM & Lustick I (2004) VIR-POX: An Agent-Based Analysis of Smallpox Preparedness and Response Policy. Journal of Artificial Societies and Social Simulation 7, no. 3 <http://jasss.soc.surrey.ac.uk/7/3/6.html>
Gatherer D (2002a) The Spread of
Irrational Behaviours by Contagion: An Agent Micro-Simulation.
of Memetics - Evolutionary Models of Information Transmission, 6.
Gatherer D (2002b) Identifying cases of
contagion using memetic isolation: comparison of the dynamics of a
simulation with an ethnographic data set. Journal of Artificial
Societies and Social Simulation 5, no. 4 <http://jasss.soc.surrey.ac.uk/5/4/5.html>
Gatherer D (2004) Birth of a
meme: the origin and evolution of collusive voting patterns in the
Song Contest. Journal of Memetics - Evolutionary Models of
Transmission, 8. <http://cfpm.org/jom-emit/2004/vol8/gatherer_d_letter.html>
Kuper A (2000). If memes are the answer, what is the question? In: Aunger R (ed) Darwinizing Culture: The Status of Memetics as a Science. Oxford: Oxford University Press
Lynch A (2002) Thought contagion in the dynamics of mass conflict. Swedish Defence Research Agency base data report FOI-R—0745—SE. Edited by Dahlberg M.
Marsden, P. (2001). Copycat Terrorism: Fanning the Fire. Journal of Memetics - Evolutionary Models of Information Transmission 5. <http://cfpm.org/jom-emit/2001/vol5/marsden_p_let.html>
Raczynski S (2004) Simulation of The Dynamic Interactions Between Terror and Anti-Terror Organizational Structures Journal of Artificial Societies and Social Simulation 7, no. 2 <http://jasss.soc.surrey.ac.uk/7/2/8.html>
to Issue 1 Volume 9