Artificial Agents
23
ing many multiple agents. These have heterogeneous structures and are
decentralized in nature [207].
2.2.2 Characteristics of Agent-Based Models
Mimicking human societies is a challenge as human behavior varies from
person to person, in character and personality. These use various interaction
rules that are either defined earlier or introduced during simulations.
Techniques such as genetic algorithms or neural networks can be used to
produce agents randomly born with learning or evolutionary capabilities. This
means that the agent would act differently if on its own, than when it grows
in a group. As Ilachinski [94] argues “emergent properties are properties of
‘whole’ that are not possessed by any of individual parts making up that
whole: an air molecule is not a tornado and a neuron is not conscious”.
Figure 2.4 shows a complicated structure of a human agent, modeled by
[184]. Different levels of complexity exist, such as sensors, alarms, long-term
memory and even an action hierarchy based on particular situations with action priority. Such a model would be increasingly complex in a computational
perspective. Making assumptions and specifying model objectives can help abstract some of this complexity, making it easy to model humans in controlled
environments.
In terms of intelligent agents, Wooldridge and Jennings [205] have reviewed
various techniques for constructing and understanding these. The authors
point out that while building intelligent agents one should consider,
Agent theory. Implies that human behavior can be specified as a set of
attributes. These attributes can be beliefs, desires or intentions (BDI).
A system which has beliefs and desires is a first-order intentional system,
whereas a system having beliefs and desires about beliefs and desires is
a second-order intentional system. Beliefs are represented as norms in a
system like rules in Prolog.
Believe(Mary, world is flat) → Mary believes the world is flat.
These rules are defined as a collection in possible world semantics as
syntactic representation of languages. It consists of a modal-language
(modal operators) and a meta-language (possible world). The latter
refers to rule beliefs about goals and correspondence theories [204]. Figure 2.5 describes the various components of strong and weak actions in
agents. These use rules to achieve goals and desires. Agent communication languages use KQML (Knowledge query and manipulation language) and KIF (Knowledge interchange format) for message representation.
Agent architecture. These can belong to three strands as follows:
23
ing many multiple agents. These have heterogeneous structures and are
decentralized in nature [207].
2.2.2 Characteristics of Agent-Based Models
Mimicking human societies is a challenge as human behavior varies from
person to person, in character and personality. These use various interaction
rules that are either defined earlier or introduced during simulations.
Techniques such as genetic algorithms or neural networks can be used to
produce agents randomly born with learning or evolutionary capabilities. This
means that the agent would act differently if on its own, than when it grows
in a group. As Ilachinski [94] argues “emergent properties are properties of
‘whole’ that are not possessed by any of individual parts making up that
whole: an air molecule is not a tornado and a neuron is not conscious”.
Figure 2.4 shows a complicated structure of a human agent, modeled by
[184]. Different levels of complexity exist, such as sensors, alarms, long-term
memory and even an action hierarchy based on particular situations with action priority. Such a model would be increasingly complex in a computational
perspective. Making assumptions and specifying model objectives can help abstract some of this complexity, making it easy to model humans in controlled
environments.
In terms of intelligent agents, Wooldridge and Jennings [205] have reviewed
various techniques for constructing and understanding these. The authors
point out that while building intelligent agents one should consider,
Agent theory. Implies that human behavior can be specified as a set of
attributes. These attributes can be beliefs, desires or intentions (BDI).
A system which has beliefs and desires is a first-order intentional system,
whereas a system having beliefs and desires about beliefs and desires is
a second-order intentional system. Beliefs are represented as norms in a
system like rules in Prolog.
Believe(Mary, world is flat) → Mary believes the world is flat.
These rules are defined as a collection in possible world semantics as
syntactic representation of languages. It consists of a modal-language
(modal operators) and a meta-language (possible world). The latter
refers to rule beliefs about goals and correspondence theories [204]. Figure 2.5 describes the various components of strong and weak actions in
agents. These use rules to achieve goals and desires. Agent communication languages use KQML (Knowledge query and manipulation language) and KIF (Knowledge interchange format) for message representation.
Agent architecture. These can belong to three strands as follows: