Article,
An Adaptive Network That Constructs and Uses an Internal Model of its World
Cognition and Brain Theory, (1981)
Abstract
Many theorists have emphasized the role of an "internal model of the
world" in directing intelligent adaptive behavior. An internal model
can be used to internally simulate the consequences of possible actions
in order to choose among them without the necessity of overtly performing
them. Animal learning theorists have taken latent learning experiments
as demonstrations that animals can learn and use such internal models.
In this paper, we describe an adaptive network of neuronlike components
that constructs and uses an internal model, and we demonstrate this
ability by describing a computer simulation of its behavior in a
simplified analog of a latent learning task. The task has been made
as simple as possible while still retaining those features that make
behavior in latent learning tasks difficult to account for by connectionist
models. The network illustrates a principle by which connectionist-like
learning rules can give rise to behavior apparently requiring the
formation and use of internal models. As such, it may help form a
bridge between brain theory and connectionist models on the one hand,
and cognitive and information processing models on the other.
