Information Technology Reference
In-Depth Information
It is stressed that agents can only control their effectors but not
necessarily the outcome of their actions (cp. Section 2.3.7). Therefore,
agents have to determine which effector they want to use, but a
simulation engine has to take care of the computation of the actual
outcome and possible effects. In order for actions to fail (i.e., produce
effects different from the intended ones) decision rules are necessary.
These rules are defined within a simulation model and enforced by
a simulation engine. As they limit the autonomy of agents they are
also denoted as
laws
or
constraints
[14, 116].
5.3.6 Abstraction from specific agent architectures
The internal structure of agents and their behavior model are not
of primary interest in context of this thesis. Nevertheless, as 'nearly
all systems ... constitute a definite intra-agent architecture' [24]a
reference model must be able to represent different agent architectures.
This requirement is in agreement with assumptions made by Modi
et al. for the agent layer of the Agent Systems Reference Model: 'We
make no assumptions about the internal processing structures of an
agent. An agent could be built with a complex cognitive model or it
could be a simple rulebased system' [91].
5.3.7 Formal specification
The development of a reference model for agent-based modeling and
simulation that is precise in its definitions as well as unambiguous
in its interpretation requires, in general, a formal or semi-formal
specification. Besides guaranteeing the exactness of definitions, a
formal specification could also provide the necessary well-defined basis
for the intended future use and investigations. On the other hand, a
reference model has to be flexible, adoptable to actual problems and
(more or less) easily comprehensible.
Therefore, these two contrasting requirements need to be balanced.
Although precision and clarity are predominant requirements, it is
also important that a formal specification is flexible enough to cope
