Page 65 - Deep Learning

48 Introduction

if a learning theory postulates separate mechanisms for generalization and spe-
cialization, what is the long-term outcome of their interaction?
The issue of scale across system levels can be pursued to the social level.
Groups, teams, organizations and communities are cognitive systems, and
they change over time. Do the properties of cognitive change impact change at
the level of such collectives, or is the latter an independent system level, where
change is decoupled, or nearly so, from the nature of change at the level of
the components, the individual persons? Psychologists tend to treat cognitive
psychology as one discipline and social psychology as another, but given that
social systems consist of individuals, this is not productive.
Finally, we can ask how a learning theory handles the transition from
a descriptive mode (What is true of learning?) to a prescriptive and applied
mode (How can we support learning in real contexts?). We would expect an
accurate theory to generate workable practical applications.
In short, an attempt to explain cognitive change has to address the issues of
existence versus prevalence, interactions among multiple mechanisms, scaling
across time, scaling over system levels and practical usefulness. These issues
are intrinsic to componential explanations of any type of change.

Criteria of Adequacy
To summarize, a satisfactory explanation for cognitive change consists of at
least the following components:
1. A description of the explanatory target itself. The latter can be a unique
learning event, a particular change that happened for a person at some
point in time and place, a type of change (belief revision) or a recurring
pattern of change (the learning curve).
2. A background theory of the relevant aspect or aspects of the cognitive archi-
tecture. It will include a specification of the types of mental representations
assumed; the repertoire of basic cognitive processes that create, manipulate
and utilize these representations; and the mechanism that passes control
among those processes. This background theory serves as a processing context
within which the postulated change mechanisms are assumed to operate.
3. A repertoire of learning mechanisms. The change produced by a learning
mechanism is typically small in scope compared to the explanatory tar-
get. The micro-theories proposed in this book distinguish mechanisms for
monotonic learning from mechanisms for non-monotonic learning.
4. A specification of the triggering conditions under which each learning
mechanism tends to occur.