Abstract

Cognitive science and artificial intelligence currently lack a robust account of the emergence and change of structured representation. This is a result of limiting assumptions about the nature of representation: what makes a representation about something else. These limiting assumptions are reflected in methodological approaches to the modelling of cognitive agents that require any representational components of the agent to be placed in the model--handcoded--by the creator of the model. This handcoding precludes the possibility of an explanation of representation emergence and change. I provide an outline of how to characterize handcoding and argue for why we must take the issue of handcoding seriously or risk loosing the explanatory power of our models. ;I use the Structure-Mapping Theory and High-level Perception models of analogical cognition as a case study to demonstrate the successes and utility of structured representation-based explanation, and to highlight current limits with respect to accounting for fundamental representation emergence and change. I demonstrate how each model relies upon an antecedently fixed representational grammar that has to be handcoded by the creator of the model. This grammar constitutes the fundamental representational building-blocks available for any representation construction or manipulation in the model. These stable, content-identity-bearing atomic units cannot themselves emerge and change. I present evidence, however, that strongly suggests that such emergence and change must be possible, and is directly implicated in analogy-making. The current dependence on handcoding of representational grammars therefore precludes the possibility of these models accounting for the central role of fundamental representation emergence and change in analogical cognition. ;I utilize Mark H. Bickhard's insights into the nature of representation and his interactive model to escape the non-emergence impasse faced by current models. I develop and augment his model to propose an initial account of the possibility for the emergence and change of structured representation in an artificial life model. The result is the situated representation framework: a proposal for how structured representational content can emerge and develop in autonomous agents