Part V. Agent-Based Modeling and Simulation

The term "agent-based modeling" is an umbrella term that subsumes many different approaches to simulation, typically focused on modeling (collections of) entities/objects/individuals/agents and their interactions with each other and with their environment. In any case, since the interactions of agents are based on discrete perception and action events, it is natural to define an agent-based modeling and simulation approach as an extension of a DES approach, such that it is an option to use the concept of agents along with the more basic concepts of objects and events. Along these lines, OEM&S can be extended by adding the concept of agents, together with concepts of perception and action events as well as communication, resulting in Agent/Object Event Modeling and Simulation (A/OEM&S).

In academic research, the term "agent-based" M&S is used ambiguously both for individual-based and cognitive agent M&S. The former, which is also called "microscopic" simulation (or microsimulation), is focused on modeling (collections of) individuals and their interactions with each other and with their environment for modeling complex systems, whereas the latter is more concerned with modeling the cognitive state and cognitive operations of an agent.

Consequently, it seems natural to distinguish between a weak concept of agents, which we call basic agents, where agents are entities that interact with their environment and with each other, and a strong concept, called cognitive agents, that is based on modeling the cognitive (or mental) state and operations of agents.

The cognitive state of basic agents has only one component: their information state containing propositional information about their environment and about themselves resulting both from perception and from communication. A propositional information item of an information state can be expressed in the form of a triple statement (or, simply, triple), which is an atomic predicate logic sentence that consists of (1) an object name, (2) a property name, and (3) a property value. Such an information item can be viewed as a belief of an agent, or as a knowledge item (where knowledge means correct information or true belief).

Beliefs represent the typically partial and sometimes incorrect subjective information of agents about their environment and about themselves. They are the most basic component of the cognitive state of an agent. The simplest model of a cognitive state only consists of beliefs, while more advanced models may also include commitments, goals, intentions, emotions, etc.

Many agent-based (in particular, individual-based) M&S approaches do not attempt to support the incompleteness of beliefs/information or the possibility of incorrect beliefs/information. They make the (tacit) assumption that agents have perfect information and focus on modeling interactive behavior: (1) the interaction of an agent with its inanimate environment via a perception-action cycle, and (2) the communication between agents.

A general model of the interactive behavior of agents depends, at least, on their information state, which is

  1. queried for decision making, and from which information items are retrieved for informing other agents;
  2. updated when new information is obtained by means of perception or communication.

In the general case of a cognitive agent with possibly false beliefs, its belief state (or subjective information state) has to be represented as a kind of restricted and modified duplicate of the objective information state managed by the simulator, requiring a more complex simulator architecture. Within such an architecture, the information state of a perfect information agent has to be short-circuited with the objective information state managed by the simulator.