3. Agent-based modelling for radicalisation and terrorism: a scoping review
3.1 Agent-based modelling
Agent-based modelling (ABM) has been exploited for many years in the biological and physical sciences (Railsback and Grimm 2012), but has only really been used as a tool for social research in the past two decades (Gilbert and Troitzsch 2010). It involves the creation of a computer programme in which autonomous agents interact with their environment, allowing various situations to be simulated, manipulated and analysed. Agents can be homogenous or heterogeneous, and endowed with multiple attributes specified by the researcher. The environment can be equally varied, with the ability to affect, or be affected by, interactions with the agents, time, or a multitude of other variables. Agent-based modelling can either be based upon a theory, which it sets out to test, or it can be generative, allowing theories to be developed from behaviour that emerges from the model (Gilbert 2008). ABM is based upon four essential principles: the autonomy of agents; the interdependence of agents; that agents follow simple rules (which together may generate complex behaviours); and that agents are adaptive and backward looking (having the ability to learn) (Macy and Willer 2002:146).
There are different types of agent-based models depending upon what the researcher hopes to achieve. Abstract models are very simplistic, aiming to demonstrate a basic process in an abstract manner rather than replicate a specific environment or situation. These are in direct contrast to facsimile models, which use specific case studies and empirical data to replicate a phenomenon as accurately as possible, often with the purpose of being able to make
predictions from these models. Middle-range models lie between the two, aiming to describe the characteristics of a phenomenon in a way that it is still generalizable, giving qualitative resemblances and not solely based on a single case study (Gilbert 2008). One of the most attractive functions of agent-based modelling is that it allows hypothesis testing in areas for which data collection may be difficult, for example for physical, ethical, or financial reasons. It is equally useful for the refinement of theory in newer areas of research for which the
empirical base is yet to be developed (Epstein 2008, Gilbert and Troitzsch 2010). However, it is important to note that for a model to be useful, it must be based on robust theorisation and correct coding – if the program is written incorrectly, or based on unsound premises, it cannot hope to be valid.
3.1.1 Agent-based modelling in social research
Bearing this in mind, we may agree with Watkins et al (2008:1) when they state that it is easier to create models in the physical sciences, where ‘well understood, commonly accepted, and validated models of physical phenomena’ exist. The unfortunate truth is that such a set of agreed-upon concepts and theories is rare in the realm of the social sciences. Those theories which do exist may not be easily converted to an agent-based model, due to a lack of in-depth formalisation or to conceptual ambiguity. The process required to encode social science theory into an agent-based model requires formalisation and clear articulation of concepts, which is a strength of the approach. Bruch and Atwell (2013:2) note that ‘agent-based models are very useful for sharpening one’s thinking about an empirical problem and identifying key
explanatory mechanisms’, which is exactly what is needed in the field of radicalisation studies. The problem remains, however, that ‘the major difficulty we face in building a model of a complex socioeconomic system is in quantifying social situations’ (MacKerrow 2003:186). It has been argued that the qualitative methods which are used within social research can be complemented by agent-based modelling in various ways: ABM ‘can be used as a tool to perform ‘thought experiments’ to test the consistency of social theories’, as well as suggesting ‘new questions for the fieldwork’, which scientists can then set out to answer (Tubaro and Casilli 2010:66). Indeed, it is suggested that ‘qualitatively-informed ABM achieves a clearer, more relevant and more understandable description of social structures and processes’ (ibid: 67). It may also remove some of the apprehension felt by social scientists when dealing with
computer simulation and the fear of the ‘black box’, whereby phenomena are supposedly modelled but mechanisms and processes are not understood. Another strength of agent-based modelling is its potential use as a policy guide: by allowing policy makers to test the effect of their proposed interventions in silico, these policies can be refined to achieve maximum impact with the resources available to them, while revealing possible unintended
consequences. A two-way feedback can then occur, with the data informing the model, the results of which then inform the situation being modelled.
The ideal-type agent-based model would use an interwoven set of micro-theories to create an environment where macro-behaviour, such as terrorism or radicalisation, can emerge and begin to be tested and understood. The concept of emergence has a uniquely specific meaning in the context of this methodology, being defined as ‘system dynamics that arise from how the system’s individual components interact with and respond to each other and their
environment’ (Railsback and Grimm 2012:10). Gilbert and Troitzsch (2010:11) describe it in slightly simpler terms as ‘when interactions among objects at one level give rise to different types of objects at another level.’ It is emergent behaviour which a model endeavours to generate, yet the more complex the behaviour, the more difficult it is to model. In recent years a number of social science agent-based models have been created and successfully validated, from Epstein’s seminal work on modelling civil violence (2002), which is the basis of some of the works included in this scoping review, to the recent testing of criminological theories regarding burglary patterns (Birks, Townsley and Stewart 2012). Individual, group, and even large-scale societal behaviour has been modelled and tested, with agent-based modelling proving a flexible enough technique to encompass disciplines as diverse as particle physics and warfare. Indeed it is extremely useful for those seeking a multidisciplinary approach to their work, since ‘agent-based models can integrate data and theories from many different sources and at many levels of analysis’ (Bruch and Atwell, 2013:4). The best model, however, is that which can be validated by real-world data, which unfortunately is not always readily available. 3.1.2 Agent-based modelling for radicalisation and terrorism
The current state of research in the areas of radicalisation and terrorism is such that high quality data, in sufficient amounts to inform and validate a theory, are exceptionally difficult to obtain. Genkin and Gutfraind (2011:6) are correct in stating that ‘there is a shortage of
empirical sources, as well [as] difficulties in generalizing beyond the cases examined’. The main methodologies used are either interviews with proponents of radical or terrorist views and actions, or larger scale quantitative works based on the number and location of attacks (LaFree et al 2008), alongside the affiliation or basic socio-economic characteristics of the attackers.
and this is not helped by issues which underpin the foundation of the field: while figures such as Sageman (2004) and Hoffman (2006) are oft quoted, there is no field-wide, national, or international recognition of a definition of radicalisation or terrorism, and certainly no agreed- upon theory of the causes of such phenomena.
This necessarily makes creating an agent-based model in this area difficult from the very beginning: the lack of foundation for a model makes building one problematic. The number of theories which are sufficiently detailed to allow the coding process to be implemented are minimal, necessitating the development of one within this research, and we see that the theories utilised and tested in our studies below are often from parallel or divergent
disciplines. Nonetheless, agent-based modelling as a methodology has much to offer the fields of radicalisation and terrorism, and it is beginning to be explored. By employing and combining theories from sociology and psychology, such as group identity and grievance theory, alongside those such as opinion dynamics and epidemiology, which help to explain the transmission of ideas and behaviours, we may be able to model the emergence of such complex behaviours as radicalisation and recruitment to terrorist groups. Indeed, ‘by creating artificial societies one can systematically manipulate the parameters of interest to discern meaningful relationships and isolate factors that will be influential over the long term’ (Genkin and Gutfraind 2011:6). Due to the lack of empirical data within the field of radicalisation, the initial aim of agent- based modelling creation may be testing for the coherency and completeness of the theory rather than validation in a traditional, statistical, sense. By programming in and adjusting existing and novel theories, it may be possible to discover and isolate those variables which appear to have the most profound impact upon the process of radicalisation, in order to inform empirical research at a later date. Rather than agent-based modelling being an
experimental or predictive tool as it has been in other disciplines, it is possible here for it to be used for the development and refinement of theory and the explanation of past and present behaviours. Along these lines, Johnson and Groff (2014) have discussed the use of agent-based models for testing the fitness of criminological theories, as well as assessing how well specified they are, and therefore, how good a foundation they provide for empirical research. The use of agent-based modelling in this field is still in its infancy, yet its successful utilisation in areas such as economics and natural resource management (Gilbert 2008), and the tentative steps that are being taken by those working in areas such as computer science and homeland security, make a review of its progress to date a worthwhile endeavour.