Information Technology Reference
In-Depth Information
Generally, the Expressivator supports narrative comprehension using the
following heuristic:
Behaviors should be
as simple as possible
. The agent's life comes from thinking
out the
connections
between behaviors and
displaying
them to the user.
Simpler behaviors are essential because
complex processing is lost on the user
.
Most of the time, the user has a hard time picking up on the subtle differences
in behavior which bring such pleasure to the heart of the computer program-
mer. But the properties of narrative interpretation mean that simpler behav-
iors are also
enough
. Because the user is very good at interpretation,
minimal
behavioral cues suffice
.
More specifically, the Expressivator provides systematic support for narra-
tive comprehensibility through the following mechanisms:
-
context-sensitivity and negotiability
: Rather than building an agent from
conventional context- and communication-independent actions and be-
haviors, a designer builds agents from context-dependent
signs
and
signi-
fiers
which are to be communicated to the user.
-
intentional state entailment
:
Trans i t i ons
are added between signifiers to
explain why the agent's observed behavior is changing.
-
diachronicity
:Signifierscanuse
meta-level controls
to influence one an-
other, presenting a coherent behavioral picture over time.
Signs, signifiers, and sign management
Typically, behavior-based agents are designed for correctness, not for user com-
prehensibility. The first step the Expressivator takes in creating narratively un-
derstandable agents is to open the architecture up for communication. Agent
design is based, not on the functions the agent must fulfill, but on its intended,
context-dependent interpretation by the user. In the Expressivator, signs and
signifiers support the construction of clearly communicated behavior; sign
management allows the agent itself to keep track of what has been commu-
nicated, so it can tailor subsequent behavioral communication to the user's
current interpretation.
Signs and signifiers
Current behavior-based approaches are based on an internal, problem-solving
approach, and generally divide an agent into activities in which the agent likes
to or needs to engage. Typical behavior-based systems divide an agent into
