Agents in agent-based modeling (ABM)

Abstract

Agent-based modeling (ABM) is a dynamically developing method of modeling broadly used in various areas of science and in everyday life. Integration of ABM with geographic information systems provides advanced and comprehensive instruments for geomodeling.
Agent-based models are digital representation of such systems as eco-systems, societies and economies composed of elements and objects located in common environment (action environment). Unique nature of agent-based modeling consists in the possibility to define the rules of decision-making of individual agents, to determine conditions of their functioning and to implement these rules in any number of iterations in order to analyze the results of the system operation.
Agents in the model may be highly diversified. They may be alive (e.g. farmers, inhabitants, landlords) or inanimate (e.g. companies, cars). They may be also grouped in bigger units (e.g. buildings, households, cities, road networks) and they may be mobile (e.g. companies changing their seat, inhabitants moving to other places). Because of the internal structure, we divide agents into strong and weak. Weak agents have simplified internal structure and simple decision-making rules, while decision-making rules of strong agents draw from the knowledge of artificial intelligence and these agents can learn, solve problems and make plans. Agents have attributes allowing them to describe their present state. They also have defined decision-making rules allowing them to take decisions about time and place and actions taken by the agents after the decision is made.
Multiple application of agent-based models entails extremely varied characteristic features of agents and this, in turn makes difficult assigning to them universal and common features. Nevertheless, agents usually have a few features which do not change depending on the model applied, namely: autonomy, variety, activeness, goal, interactivity, limited rationality, mobility and ability to learn.
In mathematical models most often all elements and objects of a given type are identical. Possibility to differentiate agents and to randomize their behaviours is assumed in agent-based modeling even when they have similar structure. Agents may have identical attributes but extremely different decision-making rules, which allows to introduce to the model the element of randomness so important in natural sciences.