Opinion is still divided over the role that internal world models can play in autonomous behaviour. Researchers who dispute the necessity of such models often have a restricted view of how they are constituted and may associate the whole enterprise of modelling with the dubious practices of GOFAI. However, this paper pursues Roitblat's approach [Roitblat, 1994] in developing a more general and less assumption-laden interpretation of what `representationalism' really means. It presents a no-strings theory of representation which shows why we should expect autonomous agents to use internal models, and what these models will look like.
Download compressed postscript file