Representations: The Hard Problem for Artificial Life

Representationalism, the original driving theory behind artificial intelligence and cognitive science, has fallen upon hard times. Representationalism identifies cognitive processes with the manipulation of explicit symbolic structures according to well-defined rules. Rocha and Hordijk put forward three requirements for something to qualify as a representation: dynamically incoherent memory, construction code, and self-organization; they also forsake the traditional idea of representation as a “standing-in” relationship between a representation and its content. While we note that Rocha and Hordijk’s three requirements are an improvement over the definition of representation as a style of explanation, we argue that this “standing-in” relationship is the defining characteristic of representations. We define representations as physically implemented structures that can create local effects that are effective by virtue of their correspondence with non-local parts of the world, and argue that this definition, while being perfectly natural, does present a difficult, although not insurmountable, challenge for artificial life.