Artificial intelligence and brain theory: Unities and diversities

This review article synthesizes studies ofartificial intelligence (AI) andbrain theory (BT). In the control ofmovement, AI offers insight into overallplanning of behavior; while control theory enables BT to modelfeedback andfeedforward adjustments by the spinal cord, brainstem, and cerebellum. We stressaction-oriented perception—analyzing perception in terms of preparation for interaction with the world, and offer a new concept of aschema as the internal representation of an “object” in the sense of a domain of interaction. A schema comprises input-matching routines, action routines, and competition and cooperation routines. The internal representation of the world is then given by a “collage” of tuned and activated schemas.Segmentation of input andregion labeling are offered as two mechanisms in the activation of a suitable “collage.”We see a number of studies that offer hope of a unified theory ofcompetition and cooperation within a single subsystem: brain theory models of the reticular formation, of the frog midbrain visual system, and of segmentation on prewired features; and AI models of segmentation on ad hoc features, and of region labeling. We then turn to the modeling of a set of brain regions as acooperative computation system—a distributed structure in which each system has its own “goal structure” for selecting information to act on from its environment, and for transmitting the results to suitable receivers. We use this to describe a few findings ofneurology. We then sample AI studies of computerunderstanding of natural language, ascribing particular significance to a speech understanding system configured as a cooperative computation system.The literatures of AI and BT hardly overlap at all, and differ widely in choice of both problem and method. The aim of this article is to overcome these diversities by extracting contributions from extant AI and BT that can be melded into the creation of atop-down brain theory: the building of a coherent model of cooperative computation within which the computational roles of brain regions, and of neurons within those regions, can be analyzed.