Information Technology Reference
In-Depth Information
Motivating Agents in Unreliable Environments:
A Computational Model
Patrick Krümpelmann, Matthias Thimm, Gabriele Kern-Isberner, and Regina Fritsch
Information Engineering Group, Department of Computer Science,
Technische Universität Dortmund, Germany
Abstract.
The development of formal models for rational agents is mainly driven
by the well-established
BDI
approach which divides an agent's mental state into
beliefs, desires, and intentions. In this paper, we argue that motivation as well has
to be taken into account in order to allow for a flexible and proactive behavior of
intelligent agents in unreliable environments. In our approach,
motives
take the
role of describing an agent's personality and are the driving force for creating de-
sires and abandoning previously selected goals. We investigate the relationships
between motives and their associated desires as well as the impact brought about
by the uncertainty and unreliability of the environment.
1
Introduction
Today, formal approaches for representing the mental state of an intelligent agent mostly
employ the
BDI
model, a framework that originated in psychology and describes in-
telligent behavior such as decision making, deliberation, and means-end reasoning in
rational beings. This model divides a mental state into
beliefs
,
desires
,and
intentions
and gives a formal account for their interactions. Beginning with the work [1] many
researchers in the field of artificial intelligence and intelligent agents applied this (in-
formal) framework to formalize autonomous intelligent behavior [10].
However, the
BDI
model is an abstraction of a mental model of an intelligent be-
ing and given some sufficiently complex scenario it is not enough to represent proper
decision making adequately. In the area of intelligent agents specifically the
desires
of
an agent are oversimplified and often assumed to be initially given. A desire represents
some state of the world such as “
I want to be rich
”or“
I want to get a professorship
”that
the agent wishes to achieve. In contrast to
intentions
which represent a currently pur-
sued course of action the set of desires does not need to be consistent, i. e. completely
achievable, as the previously mentioned statements have shown. Furthermore, a single
desire might be unachievable in general or unachievable in some specific context, e. g.
due to some physical (“
I want to fly
”) or mental (“
I want to learn all languages of the
world
”) limiting factors. Formal models employing the
BDI
model assume the desires
of an agent to be given as agents are typically situated in some constrained environ-
ment with limited capabilities and tasks. Looking at the well-known cleaner robot the
two desires “
I want to clean the room
”and“
I want to have a high battery level
”are
completely sufficient to describe its world. Generally, however, assuming desires to be
given inhibits real autonomous behavior. Bringing agents into more complex environ-
ments demands for mechanisms that allow an agent to set its desires by itself. Looking
