Abstract

A theory of structural content is developed and applied to various systems which learn, including neural networks and human agents. This content is carried by internal structural states of the system; states which are formed through causal encounters with the environment. The content of the structural state is that a certain type of history formed, or installed, that state. This history is a time-ordered sequence of events linked causally. The state is structural because after successful learning the state becomes a permanent part of the network or agent. This state doesn't cause output all the time, but acts as a background condition for output given the right external conditions; hence the state plays a different causal role from other belief states. Further, its content is not generally available for reasoning or evidential assessment. ;This state gets installed as a consequence of learning, and so allows other regularly occurring belief states, such as perceptual states, to get their hands on the steering wheel. This accounts for the notion of promoting an indicator state to a representational state. Before learning, the indicator state did not result in beneficial motion. After learning, it did. But this was because the connecting state was installed. Hence, it is shown that learning promotes an indicator state to a representational state via formations of connections which have content