Page 80 - Socially Intelligent Agents Creating Relationships with Computers and Robots

Cooperative Interface Agents 63

culty in employing the cited models is that traits are defined through natural
language descriptions and are not easily formalised into the “mental state” of
an agent. The first and most relevant contribution to a cognitive theory of per-
sonalities was due to Carbonell [4], who saw them as combinations of degrees
of importance assigned to goals. A second example, to which we will refer in
particular in this chapter, is Castelfranchi and Falcone’s theory of cooperation
in multi-agent systems [5].
Although affective expressions may contribute to increase interface agents’
friendliness, its acceptability is driven by the level of help provided to the user,
that is by its “cooperation attitude”. This level of help should not be equal for
all users but should be tailored to their attitudes towards computers in general,
and towards the specific software to which the agent is applied in particular.
These attitudes may be synthesised in a level of delegation of tasks that the
user adopts towards the agent. To select the helping attitude that best suits
the user needs, the agent has to be endowed with a reasoning capacity that
enables it to observe the user, to model her expected abilities and needs and to
plan the “best” response in every context. We had already applied the theory
of Castelfranchi and Falcone to formalise the mental state of agents and their
reasoning capacities in our a Project GOLEM [6]. With the project described
in this chapter, we extend that research in the direction of embodied animated
agents.
XDM-Agent is an embodied animated character that helps the user in per-
forming the tasks of a given application; its cooperation attitude changes ac-
cording to the user and the context. Although the agent is domain-independent,
we will take electronic mail as a case study, to show some examples of how it
behaves in helping to use Eudora. In a software of wide use like this, all proce-
dures should be very natural and easy to perform. The first goal of XDM-Agent
is then “to make sure that the user performs the main tasks without too much effort”.
At the same time, the agent should avoid providing too much help when this is
not needed; a second goal is therefore “to make sure that the user does not see the
agent as too intrusive or annoying”. These general goals may specialise into more
specific ones, according to the “cooperation attitude” of the agent. In deciding
the level and the type of help to provide, XDM-Agent should consider, at the
same time, the user experience and her “delegation attitude”. The agent’s de-
cision of whether and how to help the user relies on the following knowledge
sources:

Own Mental State. This is the representation of the agent’s goals (Goal
XDM (Tg)) and abilities (Bel XDM (CanDo XDM a)) and the actions it intends to
perform (Bel XDM (IntToDo XDM a)).