Page 54 - Deep Learning

The Nature of the Enterprise 37

The term “control structure” is borrowed from the computational sci-
ences. Some psychologists prefer the term “the central executive” while others
prefer “executive functions” or “the global workspace”; a closely related term
30
is “cognitive architecture.” This cognitive entity is a conceptual cousin to the
psychoanalytic concept of the Ego. Given the absence of any Ego Module in
31
the brain, exactly how, by which processes, the mind implements a coordinat-
ing function in its distributed, center-less neural substrate is a fundamental
but as yet unsolved problem of neuroscience. While awaiting the solution, we
can make progress by describing the control structure in terms of how it works
instead of how it is made.

The Turing-Newell Vision

The insight that an intelligent agent – be it animal, human, robot or space
alien – can be modeled in a precise way by specifying its representations,
its basic processes and its control structure independently of their material
embodiment was first formulated explicitly by the British mathematician and
32
World War II code breaker Alan Turing in the 1930s. It is one of the founda-
tional insights of the cognitive sciences and the basis for the design of general
purpose, programmable computers like the laptop on which this book was
written.
The transfer of this type of theory into psychology was the collective
achievement of the founding generation of cognitive psychologists, includ-
ing, among others, Donald Broadbent, Jerome Bruner, Noam Chomsky, Allan
Collins, George A. Miller, Ulric Neisser, Allen Newell, Donald A. Norman,
Zenon Pylyshyn, Roger C. Schank and Herbert Simon. In a pioneering 1958
paper, quoted at the beginning of this chapter, Allen Newell, J. C. Shaw and
Herbert A. Simon turned Turing’s insight into a radical and novel concept of
psychological explanation: To explain a behavior (or a regularity therein) is to
specify a program, that is, a control structure, a set of processes and a stock of
representations, that generates this behavior (or regularity). To verify that the
33
program does indeed generate the explanatory target, implement the program,
run it on a computer and observe whether the behavior or the regularity is,
in fact, produced. Explanation is reenactment. This intellectual move created
modern cognitive psychology as a distinct discipline.
In two papers published in 1972 and 1973, Newell developed this idea into
a vision for cognitive psychology. The end goal of cognitive psychology is not
34
a long list of representation-process-control triplets, one for each phenome-
non that we want to explain, but – and this is Newell’s radical idea – a single