Abstract

The aim of this paper is to establish the logically necessary preconditions for the existence of self-awareness in an artificial or a natural agent. We examine the terms, agent, situated, embodied, embedded, and representation, as employed ubiquitously in cognitive science, attempting to clarify their meaning and the limits of their use. We discuss the minimal conditions for an agent’s environment constituting a ‘world’ and reject most, though not all, types of virtual world. We argue that to qualify as genuinely situated an agent should function in real time within the dynamic world we inhabit, or some close simulacrum of it. We show that embodied agents will possess or evolve local co-ordinate systems, or points of view, locating, identifying and interacting with objects relative to their current position in space-time, and we discuss various types of embodiment, arguing that most current situated and embodied systems are too limited to be candidates for even the most minimal claim to self-identity. We argue that a truly autonomous agent has to be active in its participation with the world, able to synthesise and order its internal representations from its own point of view, and to do this effectively the agent will have to be embedded. To this end we propose a six point definition of embeddedness. Ultimately we argue for a philosophical-cum-cognitive science model of the self that satisfies essential elements of both sets of definitions of the term.