Agent-based Modeling for the Past, Present, and Future
by Stefani Crabtree, archaeologist at the Santa Fe Institute
Imagine you come home to a house in disarray. The drawers are opened, their contents spread over the floor, and all your electronics have disappeared. Would you have a hypothesis as to what happened? Burglars, the CIA, raccoons, a lone typhoon? Each could produce a similar result.
Archaeologists are frequently challenged to explain “endpoint” scenarios, such as the messy house, with limited material evidence for the process. (For example, how would you prove that a society that lived thousands of years ago was hierarchical if there was no written language and very limited evidence that there were “haves” and “have nots”?) While your hypotheses may be encompassing, without being able to run experiments and subject them to rigorous testing your hypothesis couldn’t necessarily be proven. Nor could you definitely remove alternative hypotheses from the mix.
This is a challenge that confronts many scientists; we might want to figure out how something happened in the past from a myriad of beginning points to learn how to avoid that outcome again. Alternatively, we may want to predict how something happens before it occurs to figure out the best course of action, such as predicting how the economy will respond to external perturbations. But how can we predict, or how can we retrodict, when we have only portions of data?
One tool that many scientists increasingly rely on for these exact scenarios is agent-based modeling. From uses in public health, economics, ecology, to archaeology these computer simulations provide ways to examine many possible outcomes, refine hypotheses, and build theories about the past, present, and future. Agent-based modeling allows you to build experiments in silico, imbuing individuals (the agents) with characteristics specific to the question: metabolism, infection rate, age, selfishness. The agents interact with one another and their digital environment, and the simulations are run hundreds to thousands of times to make predictions about the study system.
The example of hierarchy above was tested in the American Southwest with an agent-based model to show that hierarchical societies could develop from egalitarian antecedents. In this way, computer models can help to make invisible processes visible through careful implementation and experimentation.
The true benefit to agent-based modeling lies in the ability to experiment. Often times as scientists we are confronted with situations where experiments are either impossible (on the archaeological past, for example) or unethical (examining how a disease spreads through a population, for example). Yet we still want to know possible ranges of outcomes for the process. Agent-based models provide that platform to test these theories; you can build a model on well-known principles (how humans move through a landscape, e.g.) and examine how the model plays out over time and space. Models have been useful in the field of archaeology for understanding migrations of human populations out of Africa, for understanding domestication events, for examining the growth of hierarchies, and many other pressing questions.
Readers who would like to experiment with Agent-Based Models, but don’t know where to begin, can check out a recent three-part series published in Advances in Archaeological Practice. In the series, my coauthors and I walk through how to build a simulation from the ground up, how to incorporate realistic landscapes into the simulation, and how to run the simulation to extract data for analysis. While written for an archaeological audience, this three-part series can be used for anyone interested in agent-based modeling as a tool for science, outreach, or policy. It can also be used to complement to Complexity Explorer’s online course, Introduction to Agent-Based Modeling with Bill Rand and Anamaria Berea.
Whether you’d like to predict how a global region will respond to drought, or compare destruction patterns between raccoons, typhoons, and burglars, the tools introduced in these papers and in the Complexity Explorer allow for the systematic exploration of possible chains of events. All without bothering a single human (or raccoon).
Read the how-to series in Advances in Archaeological Practice:
Part I: Agent-based Modeling for Archaeologists
Part II: Combining Geographical Information Systems and Agent-Based Models
