Redescription, Information and Access

Over the past two decades Annette Karmiloff-Smith has developed a theory of cognition that depicts the mind as dynamic and changing rather than static and fixed (e. g. this volume). It maintains that cognitive development, in both adults and children, goes through a cycle of modularisation and explicitation, in which modules form and their contents are then explicitated to become available to other modules. Explicitation proceeds through a process of Representational Redescription, which redescribes the knowledge in the system in increasingly abstract and accessible forms, all of which are retained and may be accessed when necessary. The Representational Redescription Hypothesis (RRH) is based on considerable experimental data and is clearly of great importance to psychology, philosophy and cognitive science. But it is not without its problems. For one thing we have no idea how redescription takes place, that is, we have no idea about the underlying mechanism that makes it possible, and there are no plausible candidates in sight. For another, redescription is supposed to re-represent the same knowledge and yet give us new knowledge, and these claims certainly appear to be inconsistent. Finally, the concept of a representation is as problematic today as the idea of an idea was in the seventeenth century. In this paper I propose a radical reinterpretation of the RRH that is not subject to these problems, and that can be implemented. My reinterpretation dispenses with representations and redescription (and you can't get more radical than that). Rather than talking about redescribing representations I talk about accessing the information that underlies abilities—more specifically, accessing this information and then recursively accessing the information that makes this access possible. Now this may sound like the Cheshire Cat, which vanishes but leaves its smile behind, for how can we have representational redescription without representations or redescription? Let me explain. My theory is radically different to the RRH (since it replaces representations and redescription with information, access and ability), but it actually provides a more conservative interpretation of Karmiloff-Smith's data. There are broadly two ways of construing that data. Karmiloff-Smith's way The proposed implementation in this paper is due to my students Corey Venour and Alex Young. We arrived at our positions on access and redescription at about the same time, but from different directions. I came in from an Information Theoretic viewpoint. They came to their views by studying unsupervised neural networks. We do not agree about representations. …