2  ABM in archaeology

ABM has conceptual roots and applications in other fields dating back to the early days of computer science. However, this methodology took its current form and was introduced more broadly into the social sciences only in the 1990s.

ABM has been exceptionally well accepted by archaeologists, given its pre-adaptation to distributed and stochastic processes. Most phenomena of interest to archaeology can be represented with ABM, to archaeologists’ satisfaction. Other modelling and simulation approaches (e.g., Dynamic Systems) were, and still are, used but often encounter strong resistance because of their higher abstraction and simplification.

2.1 A transdisciplinary approach

ABM flexibility is also highly valued by archaeology researchers, given that the discipline is historically and thematically positioned between other disciplines at the overlap of the so-called natural and artificial worlds. An ABM model can handle multiple layers of entities and relationships, allowing it to integrate entire models under the same hood.

Disciplines such as ecology, environmental science, and geography have exploited the same advantage. This approach has driven model development in archaeology for more than two decades. Some authors have called this transdisciplinary framework the research of socio-ecological systems (SES), defined in close relation to the broader complexity science approach. One of the first public successes of ABM in archaeology, the Artificial Anasazi model (Axtell et al. 2002), emerged from this approach as a collaboration among researchers orbiting the Santa Fe Institute.

Despite its positive influence in pushing the field forward, the SES approach has also limited the diversity of scope and theory used with ABM in archaeology. ABM under SES aligns particularly well, for example, with research questions related to landscape and environmental archaeology and formulated from a processualist perspective. This tutorial is no exception. To the potential “new blood” in this field, I recommend keeping your mind open to all questions and theoretical frameworks, especially those you are already invested in.

ABM in archaeology has been prolific, though it remains a small community. Here, we will only cover a small part of the field, specifically from my perspective. For a broader introduction to the many contributions in this field, I refer you to any of the introductions in the References (particularly Angourakis 2026; also Daems 2026). For a more practical, in-depth introduction, I recommend the textbook by Romanowska, Wren & Crabtree (2021), which includes many practical exercises in NetLogo, though it uses a programming style and philosophy that differ from those used in this course.

2.2 Domains of application and examples

Here is a non-exhaustive list of examples of simulation and ABM in archaeology organised by topics:

2.3 Focus examples

Artificial Anasazi model in NetLogo

The Artificial Anasazi model was developed to explore population dynamics in Long House Valley, Arizona. The model represents a population of households, with a simplified food economy based on maize cultivation. By simulating this system, researchers tested the hypothesis that climate change was the main cause of the valley’s abandonment.

Artificial Anasazi

Griffin et al. 2010, Fig. 1

Griffin et al. 2010, Fig. 6b

Barton et al. 2012, Fig. 2

Barton et al. 2012, Fig. 12

Rogers et al. 2012, Fig. 2

Rogers et al. 2012, Fig. 12

Angourakis et al. 2022, graphical abstract

Andros-Spica/diagrams/RoadMapSoFar_2022-06.png

repository: https://github.com/Andros-Spica/indus-village-model

2.4 Unfinished business: representation and validation

Despite the relative success and proliferation of agent-based modelling in archaeology, an unsolved debate remains over the epistemological nature of simulation models as explanatory.

To be explanatory, models must connect a process to a phenomenon, as defined to the best of our knowledge (representation), and to the evidence we raise and select as relevant (validation). Both archaeologists and non-simulation modellers sometimes dismiss agent-based models in archaeology as straying too far from archaeological evidence. Why should we dare (or even bother) to simulate past processes that cannot be observed through material remains?

Unfortunately, it is too easy to ignore the difference between descriptive and explicative models and leave archaeological interpretations imposed on descriptive models unchecked by formalisation. When a descriptive mathematical model is used, a good validation result (fit given an evidence set) does not guarantee that the interpretative model represents the process to which it is attributed. More importantly, good validation can be mistaken for good representation.

For example, imagine we, as archaeologists, are convinced that humans could feed on a certain kind of rock at some remote point in the past. When we apply the most robust methods in cartography and inferential statistics to evidence of human presence and findings of such rocks, we might find that a positive correlation confirms our belief. However, the representational link between our interpretation and the reality of the past process is weak. Our belief would hold only if it were impermeable to formalisation and a broader knowledge base. Alternatively, perhaps, we would quickly realise the metabolic constraints of stone digestion.

Modelling structures your thoughts about the world, and mathematical modelling does so with extra discipline. Explanatory and, particularly, simulation modelling, however, is doing that specifically for reconstructing processes, stories, or, in other words, the mechanisms or webs of causality that we presume to be behind the semi-static reality we observe (e.g., archaeological materials).

Non-simulation computational modelling in archaeological research types of math models

Simulation modelling in archaeological research types of math models

We will practice distinguishing between phenomena, evidence, and mechanism when designing our own conceptual model.