Agent-Based Modeling with Social Networks for Terrorist Recruitment

Teresa H. Ko, Nina M. BerryAAAI 2004

Abstract

The Seldon model, developed at Sandia National Laboratories, provides a computational framework for exploring the social mechanics of radicalization. By integrating Agent-Based Modeling (ABM) with Social Network Analysis (SNA), the model simulates the transition of isolated individuals into terrorist organizations through the influence of social settings and group dynamics. The research highlights the roles of "homophily" and "bridge" agents in the recruitment process, offering a tool to test mitigation strategies for social mobilization.

Introduction

Predicting large-scale human social behavior is a complex challenge, often likened to the "psycho-history" concept in Isaac Asimov’s Foundation series. The Seldon model addresses this by modeling "social weather" rather than specific individual actions. It focuses on the recruitment pipeline for "expatriates"—foreign nationals who may experience social isolation. Unlike static models, Seldon captures the dynamic evolution of relationships within specific social environments, such as mosques and cliques, to understand the conditions necessary for radicalization.

Methodology: A Hybrid Architecture

The Seldon model utilizes a unique hybrid architecture that distinguishes between two primary types of entities:

  1. Simple Agents: These represent individuals, categorized as "expatriates" (isolated individuals) or "bridges" (individuals with existing ties to terrorist organizations).
  2. Abstract Agents: These are conceptual entities with their own agency, including Society (representing overarching culture), Mosques (gathering places), and Cliques (tight-knit friend groups).

The simulation tracks five overlapping, daily-evolving social networks: World, Mosque, Acquaintance, Strong Bonds, and Clique networks. The primary driver of connection is homophily—the tendency for similar agents to bond. This leads to "stickiness," where agents become socially anchored to institutions that reflect their own attributes.

Key Insights: Recruitment as a Group Phenomenon

A central finding of the Seldon model is that recruitment is rarely an individual journey. Instead, it is a group event facilitated by two conditions:

  • Clique Disgruntlement: An entire clique must reach a high level of "disgruntlement."
  • Bridge Availability: A bridge agent must be present to connect the disgruntled clique to the extremist organization.

Abstract agents like mosques play a critical role in this process. A "jihadi" mosque acts as an environment that increases attendee disgruntlement, while "neutral" mosques tend to decrease it. Because of homophily, disgruntled individuals naturally gravitate toward extremist environments, creating a reinforcing feedback loop.

Simulation Results and Mitigation

The simulation successfully replicated a realistic path to recruitment. As shown in the visualization, a sharp social division emerges between expatriates in neutral settings and those in jihadi settings.

Visualization of the Seldon model simulation showing agent distribution across different mosques and the formation of enlisted cells.

In test runs, the model demonstrated how a single bridge agent could successfully recruit entire cliques originating from a jihadi mosque. Beyond observation, the researchers used the model to identify effective mitigation strategies:

  • Societal Tolerance: Increasing tolerance reduces the initial disgruntlement of expatriates.
  • Interaction Restrictions: Limiting "weak tie" acquaintance interactions slows the spread of influence.
  • Friendship Thresholds: Raising the requirements for forming "strong bonds" significantly hinders the formation of recruitable cliques.

Conclusion

While the Seldon model is not a predictive "crystal ball," it serves as a powerful tool for illuminating the dynamics of group radicalization. By simplifying complex emotions into measurable metrics like "disgruntlement" and "bond strength," it allows researchers to test the efficacy of various social interventions. The modular Java-based design suggests future applications in modeling other group phenomena, such as urban gangs or political movements.

Summary updated

BibTeX
@inproceedings{ko2004agent,
  author = {Teresa H. Ko and Nina M. Berry},
  title = {{Agent-Based Modeling with Social Networks for Terrorist Recruitment}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2004},
}
Figure