Although the brain modeler can gain many useful insights from such concepts of orthodox automata theory as finite automata, network complexity theory, and Turing machines, we here stress that the best neural modelling will bear little resemblance to a straight application of such techniques.
This general perspective is complemented by a survey of eight levels of neural modelling, coupled with an extensive bibliography. The eight levels are: formfunction relations in single neurons; lateral inhibition; mode selection; statistical mechanics; adaptive neural networks; holography; control theory; and cognitive modelling.
An Extended View of Automata Theory
At its origin, all automata theory was biologically or psychologically motivated. delimited the behavior of any computer, in the then current sense of a human calculating according to some well-specified rules. McCulloch & Pitts (1943) sought to discern the logical calculus immanent in mental activity by formalizing certain basic properties of neurons. Von Neumann (1951) sought to combine the work of Turing and McCulloch & Pitts with the emerging study of computers to analyze both brain function and processes of genetics and reproduction. In the landmark collection of papers Automata Studies (Shannon & McCarthy, 1956) the majority of authors were still concerned to model biological and psychological processes, with an especially large concern for neural models. One might expect, then, that almost 20 years later, this review article would chart a massive development from this basis, and that the two special issues on brain theory of the International Journal of Man-Machine Studies [(1975) 7, (3) and (4)] which it introduces would combine deep insight into the brain with the fruits of *An earlier version of this paper was presented at