Page 41 - Deep Learning
P. 41
2
The Nature of the Enterprise
In psychology, we are overwhelmed with things to explain, and somewhat
underwh elmed by things to explain them with.
Robert Cummins 1
At [the information processing] level of theorizing, an explanation of an observed
behavior … is provided by a program of primitive information processes that
generates this behavior.
Allen Newell, J. S. Shaw and Herbert A. Simon 2
A theory of change must be rooted in some initial conception of the thing
that is changing, some working assumption, however incomplete and prelim-
inary. This has been so in other sciences. Chemists had to describe material
3
substances in terms of atoms and molecules before they could explain chem-
ical reactions in terms of the rearrangement of atoms. The description of the
double helix structure of DNA paved the way for a deeper understanding of
genetic mutations and biological evolution, and the discovery that the crust of
the Earth consists of continental plates floating on a molten core was a prereq-
uisite for the plate tectonics theory of geological change. Likewise, the prin-
ciples about the nature of mind set forth in this chapter serve as a backdrop
for the study of cognitive change. They are not new, and they are not mine
but represent a collective achievement of the cognitive sciences. Although the
principles are unlikely to be entirely accurate, they constitute the best available
framework for the study of deep learning. In a mature science, a chapter like
this would be unnecessary, but as things stand, disagreements go deep, so a
theorist needs to declare his intellectual commitments.
The second task of this chapter is to address the question of what does and
does not constitute a satisfactory explanation of cognitive change. This ques-
tion has not been a target of much discussion in the psychology of learning
24