How to Build a Machine with Emergent Representational Content

Encodingism as a fundamental model of the nature of representation rests on a logical incoherence. This incoherence is manifested, among many other ways, in the empty symbol problem — the inability to provide any representational content to the symbols upon which cognitive science depends. Interactivism is model of representation that avoids the incoherences and aporias of encodingism — in fact, it explains them. It provides a model of the emergence of functional representation out of non-representational phenomena — in fact, out of relatively simple principles of interactive system organization. It also provides a model for the emergence of encodings out of an interactivist representational foundation. It thereby provides a perspective within which human representation can be modeled and understood. And it thereby provides an approach within which machines with emergent representational content can be built. It is by now a commonplace that cognitive science does not know how to provide its representations with representational content. One generic term for this embarrassment is “the empty symbol problem” (Block, 1980, 1981; Harnad, 1987; Haugeland, 1981) — the sense in which the so-called symbols of cognitive science are not really symbolic of anything — they are empty of representational content — except, of course, for the designer or user or observer. In other words, the problem is to provide representational content for the system or machine itself. The first task is a diagnosis: Why is it so difficult to provide representational content? The answer, simply, is that the field labors under a false and ultimately incoherent conception of the nature of representation, one that makes any such nonempty, genuine representation impossible. The second task is to offer a solution: a model