Page 35 -

28 P. Davidsson and H. Verhagen

differentiate between different types of models. We will first look at how individuals
are being modelled, then on the interaction between the individuals, and finally how
the environment is being modelled.

3.4.1 Individuals

A model of an individual can range from being very simple, such a one binary
variable (e.g. alive or dead) that is changed using only a single rule, to being very
complex. The complexity of the model for a given simulation should be determined
by the complexity of the individuals being simulated. Note, however, that very
complex collective behaviour could be achieved from very simple individual
models, if the number is sufficiently large.
We can distinguish between modelling the state of an individual and the
behaviour of the individual, i.e. the decisions and actions it takes. The state of
an individual, in turn, can be divided into the physical and the mental state. The
description of the physical state may include the position of the individual and
features such as age, sex, and health status. The physical state is typically modelled
as a feature vector, i.e. a list of attribute/value pairs. However, this is not always the
case as in some domain the physical state of individual is not modelled at all. An
example is the PSI agent mentioned earlier that was used to give students theoretical
insights in the area of psychological theory.
Whereas the physical state is often simple to model, representing the mental
state is typically much more complex, especially if the individuals modelled are
human beings. A common approach is to model the beliefs, desires, and intentions
of the individual, for instance, by using the BDI model (Bratman 1987; Georgeff
et al. 1998). Such a model may include the social state of the individual, i.e. which
norms it adheres to, which coalitions it belongs to, etc. Although the BDI model
is not based on any experimental evidence of human cognition, it has proven to be
quite useful in many applications. There has also been some work on incorporating
emotions in models of the mental state of individuals (cf. Bazzan and Bordini 2001)
as well as obligations, like the BOID model (Broersen et al. 2001), which extends
the BDI with obligations.
Modelling the behaviours (and decisions) of the individuals can be done in a
variety of ways, from simple probabilities to sophisticated reasoning and planning
mechanisms. As an example of the former, we should mention dynamic micro-
simulation (Gilbert and Troitzsch 2005), which was one of the first ways of
performing individual-based simulation and is still frequently used. The purpose is
to simulate the effect the passing of time has on individuals. Data (feature vectors)
from a random sample from the population is used to initially characterize the
simulated individuals. A set of transition probabilities are then used to describe
how these features will change over a time period, e.g. there is a probability that an
employed person becomes unemployed during a year. The transition probabilities
are applied to the population for each individual in turn and then repeatedly