Page 63 - Deep Learning

46 Introduction

accumulate and how the multiple mechanisms interact in shaping action, dis-
course and the stream of consciousness. That is, a theory of change is neces-
sarily concerned with scale, both across time and across system levels.

Theory Articulation
Abstract specifications of learning mechanisms do not by themselves explain
anything. To explain is to articulate general principles vis-à-vis a case at hand –
that is, to work out how the abstract principles apply to the event, pattern or
phenomenon of interest.
Articulation typically requires chains of reasoning that rely on extensive
factual knowledge. Examples from the domain of biology include the explana-
tion for the evolution of flight, the large egg of the Kiwi bird and altruism in
53
social insects. The principle of natural selection applies in each case, but it is
articulated differently. The explanation for flight proposes that wings evolved
in two distinct stages. They grew as temperature regulators until they had the
size required to support flight, at which point a new set of selective pressures
were brought to bear. The large egg of the Kiwi bird might have come about by a
shrinking of the body of a larger bird. Because the slope of the regression of egg
size onto body size is lower within a bird species than between species, evolving
a smaller body tends to leave the bird with a larger than average ratio of egg size
to body size. Altruism in social insects can only be understood in terms of natu-
ral selection once we know that the conspecifics for which soldier ants sacrifice
themselves carry many of the same genes as the soldier ants themselves. The
general lesson of these examples is that explanations for particular speciation
events require not merely the general principle of natural selection but also a
database of facts specific to each case, and the particular way in which the gen-
eral principle of natural selection is applied to each case depends on those facts.
Similarly, to explain cognitive change requires that the relevant learning
mechanisms and their triggering conditions are spelled out with respect to the
specific case or pattern that the explanation is supposed to explain. The articu-
lation of a general principle vis-à-vis a particular case is typically complex and
depends on multiple auxiliary assumptions and often requires considerable
creativity on the part of the theorist.

Challenges for Componential Explanations
The construction of componential explanations for cognitive change encoun-
ters a variety of issues and challenges. These include the distinction between