ProgramFriday 23. October, room S37 of the SSC building
Challenges for integrating 'culture' in "Out of Africa" - agent-based models
According to the “Out of Africa” theory genus Homo originated in Africa and from there dispersed all over Eurasia. At least two distinct dispersals out of the African continent can be identified from fossil and archeological record. We refer to “Out of Africa 1” as the earliest dispersals out of Africa which started approximately 2 million years ago and involved early hominin species such as Homo erectus/ergaster and to “Out of Africa 2” as the dispersal of Homo sapiens which started approximately 130 thousand years ago.
The underlying mechanisms of the dispersals are unknown. On the one hand there are various hypotheses that include environmental factors such as climate, vegetation and resources. On the other hand there are hypotheses that focus on the biological as well as the cognitive/cultural evolution of the early hominins. Cultural innovations certainly have facilitated hominin dispersals out of Africa. New technologies have extended early hominins resource space and supported them in coping or overcoming various barriers. More sophisticated communication via language and symbols have improved the transfer and duration of information. Tradition assured that the information is transferred cross-generational.
Nonetheless, we do not know to what extent cultural evolution had impact on the early hominin dispersals out of the African continent. We want explore the potential of agent-based modeling in modeling culture in agent-based models for the early hominin expansion out of the African continent. Therefore, we defined the major challenges for integrating culture in “Out of Africa” agent-based models.
The Variability Selection and the Evolution of Human Cultural Behaviour
Many uniquely human characteristics such as the strong dependence on material culture and social environment, extensive social networks, learning and cultural transmission can be regarded as examples of behavioural plasticity. However, despite their significance, only a limited number of theoretical models have been developed to explain how they came to be in the first place.
The Variability Selection Hypothesis proposed by Potts postulates the evolution of such behaviours among early hominins arising during periods of strong environmental fluctuations in the last six million years. It argues that the inconsistency in selection regimes caused by the rapid environmental fluctuations produced particularly strong selection pressure on adapting to change rather than any particular set of conditions (termed 'adaptive complexity', 'adaptive versatility', or simply 'versatilists organisms'). This ABM implementation aims to assess other factors that might have influenced the evolution of behaviours described as adaptive plasticity. In particular, the focus is on dispersal, a process directly visible in the archaeological record and, therefore, providing a good link between the theoretical model and the empirical evidence.
The agent-based simulation investigates the dynamics between individuals with different positions and range on the adaptative spectrum (including the ‘versatilist’ individuals) within a non-homogenous population and under a range of environmental regimes. The initial results shows that using heterogeneous multi-agent simulation can successfully replicate a previous formal implementation of the Variability Selection Model but also shed light on how the adaptations necessary to engage in cultural behaviour might have emerged.
The impact of
social learning on subsistence
Several social learning strategies guide the selection of cultural variants through an evaluation of the expected benefits associated with each observed alternatives. The assumption of the social learner is that the payoff received by the transmitter, while displaying a given variant, is a proxy for inferring the qualitative features of the variant itself. Here we explore the theoretical implications when: 1) the payoff attributed to a given cultural variant directly affects the reproductive success of its bearer; and 2) payoff is density dependent, that is its expected amount is a function of the number of individuals possessing the same variant. We consider situations where such density dependence is both direct and inverse, with minimum and maximum critical thresholds bounding the range of positive payoff and a peak value at midpoint.
This non-linear relationship, known as Allee effect, portrays a large number of behavioural traits that are enhanced by cooperation or mutual facilitation, but are constrained by finite amount of resources. Here we present a computer simulation exploring how different degrees of reliance on social learning can exhibit a variety of equilibria, including episodes of reversion, stable co-existence, and fixation. The insights of the model suggest that social learning alone could explain some of the evolutionary trajectories recently identified in the archaeological record.
The Role of Symbolic Interaction and Learning in the Process of Cultural Evolution: An Agent-Based Modeling Approach
I present a computational agent-based model of cultural evolution. Culture is a complex emergent phenomenon which arises through individual-level social interactions iterated on large scales. Some of the hypothesized causal processes which bring about stable cultural structures are still largely untested. In my work I take an evolutionary approach, implementing the conceptual mechanisms of dual inheritance processes into a genetic algorithm acting on agents’ cultural traits. The goal of the model is to elucidate the micro-level dynamics which allow the establishment of culture from a pristine state and to study the roles of learning and symbolic interaction in the process of cultural evolution. Finally, another objective is to observe whether it is possible to generate conditions in the model which lead to the occurrence of specialized culture maintaining structures within the society.
To this end I developed an agent-based simulation model which (1) imposes only low-level intentionality and knowledge on the part of the agents, (2) is as general as possible in the representation of the society, while placing minimal assumptions on its initial state, (3) allows the representation of symbols and the study of their role in cultural evolution, and (4) implements an agent learning mechanism.
The model consists of a large number of agents who posses indicator traits, preference traits and biased cultural traits. Indicator traits are directly observable by other agents. Preference traits store information about which indicator traits the agent prefers in other agents. Biased traits represent agents patterns of behaviors, beliefs, values etc. and they are not directly observable by other agents. Agents then interact with each other and they must coordinate in their biased traits. Ability to coordinate is rewarded, while failure to do so is punished by a penalty to the agents. Interaction is more likely with individuals who possess indicators preferred by the active agent. Agents also have the ability to “learn” from other agents at certain times during the simulation. Learning is costly and it consists of copying the “teacher's” biased traits and strategically choosing to adopt one from among these and other traits learned previously.
The single most important force driving the model is the indirect bias with respect to the hidden traits which is driven by changes in indicator and preference traits. I chose this conceptual mechanism because it allows the selection of interaction partners without imposing direct knowledge of culturally significant traits of others pertaining to values, beliefs or behaviors.
The motivation for rewarding coordination and punishing dissonance in biased traits stems from the apparent practicality of synergy in actions and thoughts. To give an example, if there are two drivers on the road, and both of them drive on the same side of the road, the probability of an accident is much higher than if they drove on opposite sides. Similarly, the success of a particular trade may hinge heavily on the fact whether both of the parties accept the idea that small green pieces of paper will allow them to purchase other goods in the future. In the same fashion, the process of learning comes at a cost to an agent’s fitness because it is assumed that learning requires time and capacities which could otherwise be spent on other activities which may potentially be more immediately rewarding to the agent.
The results of the simulation runs have shown that the model of indirectly biased cultural evolution based on immediately observable indicator traits alone (i.e. WITHOUT learning) can in certain cases lead to equilibrium states of institutionalization of social interactions. The selected indicator traits begin to act as arbitrarily defined symbols of socialization and a symbolic culture emerges. The indirect bias mechanism has shown signs of better performance in “primitive” societies, marked by smaller populations and a low degree of outward diversity, represented by narrower ranges of possible indicator traits in the simulations.
Cultural learning has shown an effect on the emergence of culture in the model as well. However these effects are significantly varying in societies with different initial states. In smaller, less diverse populations the learning mechanism is a negligible factor when contrasted with the evolutionary mechanism of indirect bias. Only in much larger populations does learning show its true contribution. Without the learning mechanism, institutionalization occurs on average much later and less often.
However, the effect of learning becomes completely opposite in more individualistic societies, with a greater diversity of directly observable traits. Such populations who employed learning as a form of cultural evolution have resulted more often into culturally fragmented states, than those who relied solely on indirect bias.
The simulations of the model with learning have also shown interesting emergent characteristics of the dominant channels of socialization. In most cases, after an initial period agent populations arrive at a state of near-total indicator uniformity, with the exception of a handful of agents who are exceptional in their indicators and in that they themselves educate a vast majority of the remaining population. This is representative of the primitive stages of human societies, which are characterized by their external uniformity and a presence of charismatic leaders – heroes, chiefs, shamans, strongmen, etc. The model thus gives explanatory power to Weber’s concept of charismatic authority.
After this period, once full institutionalization occurs, the teaching caste responsible for the maintenance of culture is unified in the sense that it adopts a single indicator which acts as a signifier of its members’ affiliation. Moreover this signifier is usually different from the indicator which at that point is still dominant in the common population. Thus the caste is not only united but also stratified from the rest of society. This is somewhat resemblant of the specialized culture maintenance organizations in the real world, such as religious churches or nationwide public education systems.
Sociocultural Dimension of hunter-gatherer mobility
Playing Games, not asking Questions: using ABM as elicitation tools in rural Cameroon
Agent Based Culture?
An easy way to model culture is by looking at different behavioral patterns occurring in different cultures. E.g. in western Europe unknown people will greet each other with a handshake, while in Japan people will bow. However, these behavioral patterns are more like visible consequences of culture than actually part of the defining characteristics of culture. The most fundamental starting point of cultural expression is a difference in priority between values. We assume a kind of universal set of values as given by Schwartz (Schwartz, 2006) in the fundamental value systems of a population. A common characterization of cultural differences as given by Hofstede (Hofstede 2005) can be modeled through value priorities. E.g. collectivism can be characterized as consequence of prioritizing universalism over self enhancement. Taking values as starting point of a model of culture means that we also have to include values as part of the agent deliberation model. In (Dechesne 2012) we have shown how this can be done and leads to realistic simulations for a particular domain. However, cultures are also formed by shared experiences, which can be seen in the social practices in a society. Social practices are a kind of standard way of interacting that are shared and used in common situations. More importantly, the actions within these social practices also have common social interpretations that form part of their success (or failure). E.g. a man shaking the hand of a woman can be seen as a token of respect in western culture, while it can be seen as a token of disrespect in Arabic countries. Thus, social practices seem an important link between values and behavior in societies. In (Dignum 2015) we have shown how social practices can be modeled and how they can be incorporated in the agent deliberation. We have made some first steps in modeling culture in agent-based simulations. We are currently investigating fundamentally new agent architectures (see (Dignum 2014)) and their theories in which all these elements have a proper place.
(Dechesne 2012) Francien Dechesne, Gennaro di Tosto, Frank Dignum, Virginia Dignum No smoking here: Values, norms and culture in multi-agent systems. Special issue on Simulation, Norms and Law, of the Journal of Artificial Intelligence and Law, Springer, 1-29, 2012. (Dignum 2014) Frank Dignum, Rui Prada, Gert Jan Hofstede. From autistic to social agents. In Alessio Lomuscio, Paul Scerri, Ana Bazzan and Michael Huhns (eds.), Proceedings of the 13th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2014), May 5-9, 2014, Paris, France. (Dignum 2015) Virginia Dignum and Frank Dignum. Contextualized Planning Using Social Practices, Coordination, Organizations, Institutions, and Norms in Agent Systems X. (to be published by Springer in 2015) (Hofstede 2005) Geert Hofstede and Gert Jan Hofstede. Cultures and Organizations: Software of the Mind. New York, McGraw-Hill USA, 2005. (Schwartz 2006) Schwartz, S. H. (2006). Basic human values: Theory, measurement, and applications. Revue française de sociologie.
|Ethnography of a Criminal Network|