Cultural evolution

Charles Darwin’s theory of evolution is deceptively simple. Things vary. This variation persists over time. And some variants are often more likely to persist than others. Darwin applied this theory to the evolution of biological organisms. But many scholars since have noticed that the same principles characterise human cultural change. Cultural traits - ideas, beliefs, attitudes, norms etc. - vary. They persist by being copied from individual to individual (what is called social learning or cultural transmission). And some of these cultural variants are more likely to persist (be copied) than others. This is the idea of cultural evolution.

Much of my research has involved exploring and developing the theory of cultural evolution. Cultural evolution is Darwinian, but still differs in many ways from genetic evolution. The task is to figure out how exactly it works, just like early biologists did for genetic evolution: where cultural variation comes from, how it is transmitted from person to person, and why some cultural variants are more likely to persist than others.

The theory of cultural evolution provides a set of methods and concepts that can explain cultural change scientifically, in my view more rigorously than in the mainstream social sciences and humanities. It also suggests a reason why humans have been so successful as a species: while other species rely on genetic evolution to produce complex adaptations such as wings or eyes over long time periods, we also have cumulative cultural evolution, which can rapidly generate cultural adaptations such as complex technologies and social institutions. The theory of cultural evolution provides a guiding framework within which all of my other research is carried out.

Relevant publications

Virtual arrowhead experiments

A screenshot of the Virtual Arrowhead Task. Participants either directly manipulate the dimensions along the top (asocial/individual learning) or copy the design of one of the other players listed in the left-hand box.

I developed the computer-based Virtual Arrowhead Task to explore how people use social and asocial learning to create a technological artifact - an arrowhead - and how these learning decisions shape technological change and variation. Players design their arrowhead and, over a series of hunts, receive a return in calories depending on how close their arrowhead is to hidden optimal designs. Players can improve their arrowheads either by directly modifying the design using trial-and-error (asocial or individual learning), or by copying the design of one or more other participants (social learning). Various things can be manipulated, such as the shape of the underlying design space (or fitness landscape), the cost of learning, the presence and type of people to copy, and the stability of the environment.

Experiments using this task have been used to recreate actual patterns of prehistoric arrowhead variation in the archaeological record. Archaeologists suggested that unusually low variation in prehistoric arrowhead designs in one region of the Great Basin (central Nevada) resulted from prestige or payoff bias, where all group members copy the single most successful or prestigious hunter. Prehistoric arrowhead designs in another region (eastern California) show much higher diversity, consistent with guided variation, where people copy a design but then modify it using trial-and-error. But without a time machine we cannot go back and actually test these scenarios concerning who copied what from whom, and when. Mike O'Brien and I therefore simulated these scenarios in the lab, showing that the resulting virtual arrowhead variation matches well with the prehistoric arrowhead variation.

Further experiments using this task have explored more general aspects of human social learning of a complex technological trait, including the effect of manipulating the fitness landscape (multi-modal vs unimodal; wide-peaked vs narrow-peaked), people's choice of social learning strategy (payoff bias, conformity, random copying etc.), whether people 'protect' their good designs from being copied by others via informational access costs, whether people use minimal cues of prestige to select people from whom to copy, and the extent of cross-cultural variation in social learning use.

Relevant publications

Model handaxe experiments

Replica foam handaxes carved by successive participants in Schillinger et al. (2016), later subjected to phylogenetic analysis to determine whether this lineage can be accurately reconstructed

This set of studies, all in collaboration with Stephen Lycett (SUNY Buffalo) and our former PhD students Marius Kempe and Kerstin Schillinger, uses lab experiments to explore how artifacts evolve as they are passed from person to person. We focus on Acheulean handaxes as a model system, using either virtual handaxes or replica physical handaxes made of plasticine or foam. Acheulean handaxes are archaeologically important, having been the dominant mode of human technology for over a million years, and well studied, providing a rich real-world dataset against which to compare our experimental data.

A virtual handaxe used in Kempe et al. (2012) to study how perceptual error affects artifact evolution

We have found that (i) imperceptible changes are introduced when people attempt to copy artifacts due to inherent inaccuracies in human perception, resulting in significant long-term change in artifact size over multiple copying episodes (Kempe et al. 2012), (ii) copying error increases as people have less time to manufacture an artifact (Schillinger et al. 2014), (iii) there is less faithful artifact copying for 'reductive' technologies where material cannot be added back on to the artifact, such as handaxes, compared to 'additive' technology where material can be replaced, such as baskets or pots (Schillinger et al. 2014), (iv) imitation (copying the actions of others) results in higher copying fidelity than emulation (copying the end product) (Schillinger et al. 2015) and (v) phylogenetic reconstructions are more accurate when artifact copying fidelity is high (Schillinger et al. 2016).

Relevant publications

Cultural variation / Migration

A Chinese-language version of the Virtual Arrowhead Task, used to test social learning use in both Hong Kong and mainland China (Mesoudi et al. 2015)

Most lab experiments are performed with Western undergraduates, who are a very narrow sample of humanity. Recently I have been extending my experiments to non-Western participants, and migrants in the West with non-Western heritage.

In collaboration with Prof. Lei Chang (Macau University) and my former postdoc Keelin Murray, I used the Virtual Arrowhead Task to show that people from mainland China copy more frequently than people from the UK, as well as people from Hong Kong and Chinese migrants studying in the UK (Mesoudi et al. 2015). This suggests that there is non-trivial cross-cultural variation in social learning. We subsequently wrote a review, with Alex Thornton and Sasha Dall, synthesising findings across species relating to individual and cultural variation in social learning (Mesoudi et al. 2016).

Another project has used migration as a semi-natural experiment to explore the maintenance of cross-cultural variation in thinking styles. Cultural psychologists have found extensive cross-cultural variation in thinking styles, along dimensions such as analytic-holistic cognition and individualism-collectivism. Yet it is unclear how this variation is maintained over time. In one study (Mesoudi et al. 2016) we found that British Bangladeshi migrants in London show rapid and substantial shifts in just a generation or two towards the typical thinking styles of the local non-migrant population. This suggests that culturally variable psychological characteristics are substantially influenced by horizontal cultural transmission, such as the mass media or educational systems, rather than vertical cultural (or genetic) transmission from parents. A subsequent study has modelled migrant acculturation (i.e. the shift towards local norms) as conformity (Mesoudi 2018). The Thinking Styles project website contains more information.

Relevant publications

Transmission chain experiments

The transmission chain method, developed by Bartlett (1932), is where written material is passed along chains of participants similar to the children's game "Chinese Whispers" or "Telephone". This allows us to test for hypothesised content biases in cultural transmission that distort information in a systematic manner, or favour the transmission of some types of information over others.

A schematic of the transmission chain method from Mesoudi & Whiten (2008). Here there are four parallel, independent chains A-D all starting with the same material, which is passed along four generations 1-4.

Data from Mesoudi et al. (2006) showing that social information (either juicy gossip or trivial social interactions like asking for directions) is transmitted better along transmission chains than non-social information (e.g. about global warming).

Data from Mesoudi et al. (2006) showing that social information (either juicy gossip or trivial social interactions like asking for directions) is transmitted better along transmission chains than non-social information (e.g. about global warming).

One study (Mesoudi, Whiten & Dunbar, 2006) found that information concerning social relationships is transmitted better than equivalent non-social information, consistent with the social brain hypothesis, which argues that primate (including human) intelligence evolved in order to solve complex social problems. In another study (Mesoudi & Whiten, 2004) we found that a hierarchical structure is spontaneously imposed on descriptions of everyday events (e.g. visiting a restaurant) as those descriptions are transmitted along chains of participants, consistent with schema/script theories from cognitive psychology.

Relevant publications:

Cumulative culture

Human culture, compared to the culture of other species, appears to be uniquely cumulative, in the sense that we build on what previous generations have achieved. This results in knowledge and technology – from spacecraft to smartphones, quantum physics to germ theory – that could never have been produced by a single individual alone.

An image from Pitt-Rivers' 1875 book On The Evolution Of Culture, showing the gradual accumulation, modification and diversification of cultural artifacts over time

The results of Kempe et al. (2014), showing the mean cultural complexity (indicated by the numbered contours) reached at different values of social learning accuracy and numbers of cultural models.

A recent modelling study (Kempe, Lycett & Mesoudi 2014) explored the factors that might be responsible for this unique human capacity, suggesting that cumulative culture only evolves when the accuracy of social learning and the number of individuals from whom one can copy both exceed certain thresholds. A follow-up lab experiment (Kempe & Mesoudi 2014) showed that the latter factor - the number of demonstrators - does indeed constrain cumulative culture using a jigsaw task. In another project (Mesoudi 2011) I modelled the costs of acquiring increasingly complex knowledge. These models suggest that, all else being equal, there must come a time when accumulation stops, at the point where the amount of knowledge accumulated by society is too great for an individual to acquire in a single lifetime. Mesoudi & Thornton (2018) explored these and various other properties of cumulative cultural evolution, in an attempt to clarify often conflicting definitions of this term.

Relevant publications:

Other projects

Copycat suicide

An agent-based simulation of copycat suicide (Mesoudi 2008). In A there is no social learning of suicide behaviour, and hence no clusters in space or time. In B there is social learning, resulting in point clusters of suicides in space and time.

This project (Mesoudi 2009) involved the agent-based modelling of copycat suicide, a phenomenon documented by sociologists where suicide risk increases in response to exposure to the suicide of another individual. The simulations support sociologists' claims that documented clusters of suicides in time and space are indicative of social influence on suicide risk, and linked certain kinds of clusters with specific social learning mechanisms. For example, localised point clusters may be generated by peer-to-peer influence within social networks, while mass clusters, which occur across a large region in response to celebrity suicides, are likely generated by a combination of one-to-many transmission (indicative of the mass media) and prestige bias. I have also written about mass shootings as potentially being influenced by similar media-driven copycat effects (Mesoudi 2013).

Partible paternity and human mating behaviour

In collaboration with Kevin Laland (University of St Andrews), I modelled the coevolution of cultural beliefs regarding partible paternity, the idea that children can have more than one father, with genetic predispositions regarding human mating behaviour (e.g. monogamy or polygamy), finding that partible paternity beliefs allow more polygamous mating behaviour to evolve relative to singular paternity or non-cultural populations (Mesoudi & Laland 2007).

Random copying models of cultural change

This project, in collaboration with Stephen Lycett, looked at how random copying (the cultural analogue of neutral drift in genetics) generates a distinct power-law frequency distribution of cultural traits. We extended previous random copying models to explore the consequences of various forms of biased cultural transmission - conformity, anti-conformity and frequency-dependent trimming - on the power-law distribution (Mesoudi & Lycett, 2009).

Hierarchical organisation of cultural traits

In collaboration with Michael O'Brien (University of Missouri), I used an agent-based model to explore the conditions under which hierarchically-organised recipe-like cultural traits are most likely to emerge in cultural evolution, as opposed to non-hierarchical holistic or diffuse traits (Mesoudi & O'Brien 2008). We found that hierarchically-organised traits emerge when learning is costly and when technological artifacts are composed of repeating sub-units.