Information Technology Reference
In-Depth Information
people simply attribute agency to the computer itself. “I did this, and then
the computer did that.” They also attribute agency to application programs,
e.g., “My word processor trashed my fi le.” They often distinguish between
the agency of system software and applications (“My new operating sys-
tem crashed my app”). They attribute agency to smaller program elements
and/or their representations, e.g., “The spelling checker in my word pro-
cessor found an error.”
In social and legal terms, an agent is one who is empowered to act on
behalf of another. In the mimetic world of human-computer interaction,
this defi nition implies that, beyond simply performing actions, computer-
based agents perform a special kind of actions; namely, actions under-
taken “on behalf of” people. It also therefore implies that some sort of
implicit or explicit communication must occur between person and sys-
tem in order for the person's needs and goals to be inferred. I think that
this defi nition is both too narrow and too altruistic. There may be con-
texts in which it is useful to create a computer-based agent whose “goals”
are orthogonal or even inimical to those of human agents; for instance, in
simulations of combat or other situations that involve confl icting forces.
Agents may also work in an utterly self-directed manner, offering the
results of their work up to people after the fact. For now, we will use
the broader defi nition of agents to apply to human-computer interaction:
“Characters” can initiate and perform actions
based upon input from the pro-
gram or the interactor. Like dramatic characters, they consist of
bundles of
traits or predispositions to act in certain ways
.
Traits circumscribe the actions (or kinds of actions) that an agent has
the capability to perform, thereby defi ning the agent's potential. There are
two kinds of traits:
Internal traits
determine how an agent can act, and
ex-
ternal traits
that represent those internal predispositions. People take cues
from the external representation of an agent to infer its internal traits. Why?
Because traits function as a kind of
cognitive shorthand
that allows people to
predict and comprehend agents' actions. Inferred internal traits are a com-
ponent of both dramatic probability (an element of plot, as described in the
next chapter) and “ease of use” (especially in terms of the minimization
of human errors) in human-computer systems. Part of the art of creating
both dramatic characters and computer-based agents is the art of selecting
and representing external traits that accurately refl ect the agent's potential
for action.
Aristotle outlined four criteria for dramatic characters that can also
be applied to computer-based agents (
Poetics
1454a, 15-40). The fi rst cri-
