Information Technology Reference
In-Depth Information
More generally, meta-level controls make the relationships between behav-
iors explicit, as much a part of the agent design as behaviors themselves. They
allow behaviors to affect one another directly when necessary, rather than mak-
ing interbehavioral effects subtle side-effects of the agent design. Meta-level
controls give agent builders more power to expose the inner workings of agents
by letting them access and then express aspects of behavior processing that
other systems leave implicit.
Putting it all together
Narrative psychology suggests that narrative comprehension is context-sensi-
tive, focuses on agent motivation, and seeks connections between events over
time. The Expressivator supports comprehensibility by expressing the agent's
actions with signs and signifiers, the reasons for agent activity with transitions,
and the coherent threads through activities with meta-level controls.
These architectural mechanisms are described separately, but used together
in the agent design process, changing the way in which agents are designed. In
a typical behavior-based system, an agent is defined in 3 major steps: (1) de-
ciding on the high-level behaviors in which the agent will engage; (2) imple-
menting each high-level behavior, generally in terms of a number of low-level
behaviors and some miscellaneous behavior to knit them together; (3) using
environmental triggers, conflicts, and other design strategies to know when
each behavior is appropriate for the creature to engage in. With the Expres-
sivator, the choice and expression of these structural 'units' for the agent is
not enough; in order to support the user's comprehension, the designer must
also give careful consideration to expressing the reasons for and connections
between those units. These connections are designed and implemented with
transitions, which alter the signifiers they connect into a narrative sequence. In
practice, transitions are the keystone of the architecture, combining signifiers
in meaningful ways through the use of meta-level controls.
Results
The best way to see how the Expressivator changes the quality of agent behav-
ior is to look at how its transitions work in detail. Here, I will go over one
point where the agent switches behaviors, and explain how transitions make
this switch more narratively comprehensible. One example does not proof
Search WWH ::




Custom Search