AnthroGrid and Distributed Computing

Larry Kuznar and Dwight Read

One of the most significant developments in late 20th century science was complexity theory. Complex systems are characterized by:

Individual agents interact in non-linear ways with each other Agents are influenced by their environment (including other agents), and their behaviors in turn influence their environments Agents typically have access to local but not global information Path dependence - agent behaviors or other events can cut off former possibilities and create new opportunities; history is not reversable Emergence - global level properties emerge out of the local level behavior of agents Agents often governed by simple rules of behavior Agents interact through networks

Culture is a constructed conceptual system that enables forms of social organization that could not arise, from ancestral non-human primate forms of social organization through biological evolution alone.

Cultures are complex in the following ways: Individuals interact in complex ways People influence their social environment (agency), and are also influenced by the cultural domain in which they are embedded Individual lives and careers are path dependent Norms, ideologies, and some structures can be the consequence of, or arise from complex interactions People do not typically have access or can use global information People interact through networks that both arise through behavior and through cultural instantiation of cultural constructs

It is important to note that human complex systems may have systems of organization that are expressed through culture that are more the product of intent and are not necessarily emergent phenomena.

The notion that cultures are complex systems is not new. Herbet Spencer spoke of complexity in society in ways not dissimilar from authors 150 years later. E.B. Tylor's concept of "That complex whole..." identifies incisively culture as a complex system. Franz Boas admonished anthropologists to unravel the connection between the individual and culture (micro-macro distinction). Radcliffe-Brown's structural lfunctionalism was based on the notion of interdependent interactions between components of culture. Julian steward's multilinear evolution was an attempt to capture the path dependence of cultural development, and Marvin Harris struggled with how to focus on individual behavior in a larger cultural context. Recent anthropologists have examined how optimal foraging and costly signaling may be a way that cultural conventions and norms emerge. New attention to complexity in anthropology and other disciplines (economics, physics, political science, and others), and the development of new methods, are enabling anthropologists to work out concepts and means to better represent and analyze ideas about cultural complexity expressed by earlier anthropologists.

Agent-based modelling provides a methodology for representing and anlyzing the working of complex systems, like culture and its attendent social system. Agent based models are based on instantating many individual agents with their own rules of behavior and properties, who interact with one another in some environmental (physical, social and cultural) context. Their interaction arises from their rules of behavior through evaluation of local conditions as perceived by individual agents. From these interactions, we can see the influence of pre-existing social structures, as well as the emergence at the local level of evolving preferences and actions and social structures at the global level.

Example: Population Dynamics andd Marriage among the !king San Simulation of population dynamics and marriage behavior among the !Kung san. Two sets of behavior: (1) birth spacing by women based on their views regarding having children and having healthy familiies and (2) their marriage restrictions based on appropriate and inappropriate marriages according to the (cultural) kin relation of the persons involved. Goals: (1) work out the dyanmics between changing population size and birth spacing by females, taking into account the resorce richness of the group's environment and (2) determine whether or not their kinship based marrriage restrctions account for the ethnographic observation of no general rule of exogamy (the !Kung san say that one may marry within the residence group living together around a waterhole) yet the !Kung san never marry within a camp.

Rules of behavior for birth spacing are based on their statements that they want to have as many children as possible, but they also want to have healthy families. In the simulation a woman takes into consideration her current cost of foragingand her current cost of parenting (which varies with the number and age of her offspring), then decides to have an offspring or defer having an offspring based on whether having another child is compatible with her time/energy budget for foraging and parenting. Unlike modeling that aassigns average characteristics to each agent (hence enabling treating the agents in the aggregate), the agent based modeling allows for each woman to make a decision regarding birth spacing based on her current sitation (cost of parenting, cost of foraging) and some women will decide to have a child and other women, at the same time, will defer having a pregnancy.

Rules of behavior for marriage and residence group membershp are based on what they say about marriage restrictions and who can be a member of a residence group and the residence group a couple will live in after marriage. The agent based simulation keeps track of the close relatives of a person, where relatives are determined solely through appropriate kin terms that may be used to refer a person and for a new born are determermined from the cultural relatives of one's parents calculated through kin term compuations. Genealogy is tracked, but is irrelevant to determining kin relationships. Marriage is possible with any person of the appropriate age range for whom one uses a kin term that signifies the otehr person is a potential spouse, independent of residence group. Newly married individuals live in the residence gropu of the spouse due to obligation by the spouse to provide bride service and bride service ends when the couple have a child of marriageable age. At this point the couple stays in that residence group or moves to his natal residence group (if his natal residence group is a different group) based on the relative size of the two residence groups on the assumption that the smaller residence group has access to more resources.

Conclusions from the simulation. (1) The decision model leads to the emergent property of a stabilized population size, but with the stablilizes population size being futher below absolute carrying capacity in resource poorer environments than resource rich environments. Further, with very rich environments, the population will not be stabilized. Thus we would expect more intergroup conflict to occur between hunter-gatherer groups in resource rich than resouce poor environments. Data from Australis mathes the predicted pattern regarding the relationship between the stabilized population size and the absolute carring capacity. Very little data are available on conflict, but these data are also consistent with the predctions regaring conflict. (2) In the simulation only 2% of the marriages actually occurred between individuals within a residence group, which is consistent with ethnogaphic observation. The simulated camp sizes match observed camp sizes in a camp size versus camp rank plot.

Example: Modeling the Political Histories of Palestinian Political and Radical Groups Agent-based modeling was used to instantate teories of risk sensitive decision making and small group social psychology to produce virtual histories of Palestinian political groups. Based on earlier research that demonstrated that individuals take risks to increase their social status when they are near a class boundary, a model of risk sensitive decision making was instantiated to determine how a particular agent would join or defect on a partner. As an individual's relative social status changed, his perferences evolved. The fact that relative status near multiple class boundaries is operant explains both the involvement of middle-class individuals in political radicalism and near zero-correlations between poverty and terrorist behavior. In the model, as individuals joined with one another, they formed new networks of allies, thereby building new, emergent social structures. These structures inturn would influence agent behavior. Small group social psychological research deonstrates that highly risk prone groups tend to be insular, promoting hyper cooperation within the group, and hyper competition with out-group individuals. Therefore, the emergent groups' evolving risk sensitivity would recurse onto group members, shifting the individual group members toward more risk prone or averse behavior accordingly. In this way, not only would new structures evolve, creating new environments that were both influencing and influenced by agents, but group histories were created that tracked the dynamically shifting memberships of groups. Data on socioeconomic status from the Palestinian Authority were used to initialize the model, and emmpirical data on the membership Hamas and Palestinian Islamic Jihad (PIJ) were used to validate the simulation finddings. The primary findings included:

1. Extremely radical, risk prone groups tend to drive themselves extinict. 2. Very open, ecumenical groups go extinct as their membership is drawn off into other groups. 3. The radicalism of most groups oscillates through time, alternating between more open and less radical groups and more risk prone and insular groups. 4. The large, radical groups generated by the model were socio-economically different from the general Palestinian population and strikingly similar to the known membership of Hamas and PIJ.

These examples demonstrate the potential that agent-based modeling has for addressing fundamental issues in complexity theory and in the anthropology of social complexity. However, these simulations were relatively simple and small, so they could successfully be implemented on standard desk-top personal computers. Of course, many more demographic, economic, social, political, and psychological processes operate on individuals and through cultures. Furthermore, realistic simulations would ideally involve 1000s if not millions of agents. As the cognitive complexity and number of agents increases, exponential increases in working memory and computational power can be needed. Distributed computing (also called parallel computing) addresses this need in a manner that mimics social structures themselves.

The standard operating procedure for personal computers is sequential processing, in which instructions are carried out in a linear order. Of course, that is not what happens in a distributed system. The world's 6 billion people do not queue up and execute their daily tasks in turn. Instead, people, villages, tribes, and nations are all doing their things asynchronously. A distributed computational environment divides its constituent tasks among a number of processors (called nodes, and usually represented by individual computers), allowing asynchronous, not sequential, processing. Conceptually, if one is modeling a complex society with different components operating across a network, then a distributed framework is more analogous to the social reality one is simulating. Practically, a distributed network of old, slow machines working asynchronously can be far more efficient (faster) than a very fast single processor operating sequentially. From the viewpoint of agent based modeling, each node in a grid system can be a single agent, each operating with its own clock. Agents can interact with each other through an evironment that each agent can access and affect.

There is no discipline that more thoroughly intetgrates all aspects of the human condition than anthroplogy; it is at the root of our holistic tradition. However, if we are to take this tradition seriously and integrate multiple complex theories and instantiate them in complex social structures containing many agents, the computational capabilities of standard sequential processors will be rapidly overwhelmed and can be dependent upon the sequential operations. Therefore, as the AnthroGrid develops, it should be developed in such a way that the infrastructure for conjoining many computers around the world into a large distributed network that can be used for this type of computational modeling. In addition, the social networking capabilities of the AnthroGrid will become essential in providing researchers with the software, training, and collaboration necessary to develop large, distributed agent-based models.