What is agent-based modeling?

Whilst traditional ‘data science’ involves the analysis of historical datasets and existing information, Agent-Based Modeling (ABM) takes a quite different approach.

ABM is inherently forward looking, and does not require large volumes of high-quality historical data: only a rough idea of how a system or its component parts might work. Starting from this conception, a microscale digital ‘model’ is built that attempts to approximate the results which might be expected in the real-world.

What is a model?

Models attempt to represent and simulate real or imagined scenarios, which we call the ‘target’.

Models are great because they force us to consider the many features of a target, and express the relationships between them. This in turn allows for discovery of potential emergent phenomenon.

Nigel Gilbert’s little green book on Agent-Based Models considers three primary types of models:

  1. Scale models are those which attempt to replicate targets at a smaller-scale. System detail or complexity is reduced accordingly. For example, a scale model of an apartment complex might show room shapes and doorways, but might not include such intricate detail as locks on the doors, or light switches on the walls.
  2. Ideal-type models simplify problems by exaggerating some characteristic of a target (e.g. assuming an economy is perfectly competitive, or that information flows instantly between all participants in a market, without any misunderstandings).
  3. Analogical models, most often applied to systems not yet well understood, draw on well-understood real-world phenomena to speculate how a system might work. These analogical models are rarely perfect, but often an constitute important first step in tackling a problem (e.g. Dalton’s 1808 Billiard Ball Model was replaced by the Cubic Model, the Plum Pudding Model, the Saturnian model, the Rutherford model, the Bohr Model, and the Atomic Orbital model, each in turn).

Agent-based models, or ABM, are sometimes referred to as ABMS. This term denotes ‘Agent-Based Modeling and Simulation’ more fully.

So what is an agent-based model?

An agent-based model consists of 4 core components.

Key components of an Agent-Based Model:

  1. Agents: social things in a model that can interact with each other and an environment, and pass information between each other. Agents might represent animals, individuals, households, organizations, or even entire countries.
  2. Properties: agents have properties. A property might be memory; a state, characteristic, or attribute, such as hunger, speed, or health. Properties are discrete, and can be binary (yes/no), or numerical (e.g. on a 1–100 scale).
  3. Environment: this is the virtual world in which agents act and interact. It can be 2D, or 3D. A neutral medium with no effect on agents whatsoever, or a prime determinant of their ability to act. It can be abstract and imagined, or a replication of real-world buildings, cities, or vast geographies.
  4. Rules: these govern what happens when agents interact (or come into contact) with each other, or their environments. They may also govern how learning and adaptation occur within an environment. These rules may be pre-programmed, or automatically inferred in the case of some smart ABM platforms.

The complexity of a model may be constrained by either:

  1. Resources: the tools, time and manpower available to build in all known rules, information, agents, and their properties; or
  2. Understanding: the limits of our certainty and knowledge about the rules that govern a system, or the agents that make it up.

What is ABM good for?

Agent-based models excel at identifying rare and emergent complex phenomenon beyond the scope of human imagination.

Things like economic crisis modeling have been pioneered by the likes of Doyne Farmer at the Institute for New Economic Thinking at the University of Oxford.

Agent-based simulations of war-games can yield counter-intuitive or surprising insights and suggest highly-successful alternative strategies to those employed by human planners.

ABM applied to retail-spaces can suggest more optimal store layouts when combined with information about customers, and traditional data-science.

huge array of experimental medical use-cases hint at what is possible in a new era of computerized drug research and development, as well as patient treatment.

Providing a different means of analysis to that afforded by traditional data-science methods, whilst retaining an ability to build on and utilize data collected over many years, ABM harbors immense potential — without requiring quality or quantity of top-down data to start generating results.

ABMS allows for almost anything to be modelled intuitively, from the ground-up, and emergence in complex systems to be captured accurately. The only requirement is a detailed understanding of how a systems’ individual features work.

What is the future of ABM?

The best way to build accurate models of the world at scale is going to involve the creation of lots of small models of small things, composed and combined together in larger ‘meta-models’.

No existing ABM platform truly allows for this. At HASH, we’re building a new agent-based-modeling platform to allow for real-time modeling and simulation of complex systems and emergent phenomena.