Machine cognition: Approaches to autonomy of software agents

Philip Larkin

Abstract

Humans have long been fascinated by the idea that a machine may be able
to think and act on its own behalf. Whether it be Marvin the paranoid
android from Douglas Adams' Hitchhikers Guide ?trilogy?, Hal the
disembodied psychotic computer from Stanley Kubrick and Arthur C.
Clarks' 2001 a Space Odyssey, or the myriad of robots in the George
Lucas' Starwars movies. Fiction writers have been fascinated by the idea
of  machines with the ability to plan, reason and show emotion similar
to those in humans. It is not however just the imagination of authors
which has been captured by the idea of intelligent artificial creatures,
but also the imagination of computer scientists. Even though the working
of the human brain is still largely a mystery, there have been many
enlightening discoveries over the last two decades in the field of
cognitive psychology which can be utilised in the field of artificial
agents.
This work currently looks at the state of research into cognitive
psychology, to be applied as the basis for cognitive models in
artificial agents, within virtual environments (particularly computer
games as they are heavy users of intelligent artificial agents).
Ultimately the aim is to produce an artificial character, which can,
when placed in "situations", act in a way that is more convincingly
human than those currently available. This means that the agent must
show some diversity of reaction to the situations in which it is placed
and not just operate from a decision tree (with a predictable outcome).