The basic premise behind the affective agent paradigm is that the agent behaves
with an emotional frame of reference with which to weigh its decisions. Besides
providing the agent with emotional states such as anger, contentment or fear,
we also made it to mimic the basic lifecycle of humans: agents have a childhood,
maturity and old age, with their corresponding goals and priorities.
The affective model plays two roles in the behavior of the agent: action
selection (e.g., what to do next based on the current emotional state) and
adaptation (e.g., short or long-term changes in behavior due to the
emotional states).
The short term variables which control the behavior of the agent are the
action tendency and the conflict tendency. The dynamic action tendency
is the probability whether an agent will fight or flee in a given situation.
To adapt the action tendency to the outcome of the agent's interactions, the
action tendency differs whether the agent is experiencing loss or success. This
leads to the definition of an Adaptation Rule, AR. We let r be the basic
action tendency of agent A in the absence of any adaptation and m be the
current action tendency (r = m prior to the initial application of the
Adaptation Rule). For losses,
|
|
The calculations for an agent experiencing success are,
This dynamic action tendency is used in calculating another dynamic parameter
of the adaptive agents, namely, conflict tendency. This parameter determines
whether an agent seeks conflicts or avoids them. A range is defined such that
an agent can be classified as angry or fearful. Regardless of its conflict
tendency, agents will not seek conflicts with agents of the same type.
Besides the mood of agent, its behavior is determined by its age. The agent
remains content, without adaptation, until the agent comes of age, which is
set, with apologies to Tolkien, to 33 cycles. After the age of maturity, the
agent’s conflict tendency is adapted every 10 cycles. To avoid agents becoming
bogged down in an emotional quagmire, a catalyst was installed in the way of a
mood swing. At 50 cycles and every 25 subsequent cycles, the agent’s current
action tendency is randomly reset to a new value and then action tendency and
conflict tendency are recalculated. This provides a potentially dramatic change
in mood. An agent could easily shift from an action tendency of 0.8 angry to
0.3 fearful.
The age of maturity was also employed to delay the agent’s mating.
Moreover, the agents’ mating is also limited by mood and by energy level.
An agent that is angry cannot mate. Only an emotional state of fearful or
content will allow the agent to mate.
|
