Page 137 - Deep Learning

120 Creativity

other moment during problem solving. The difference in outcome is entirely a
function of differences in the amount of activation, the amount and content
of the feedback, the activation and strength levels associated with the various
knowledge elements and their relations to the activation threshold. For some
values on those parameters, the interactions among the basic processes will
generate an insight; for other values, those same processes will produce an
analytical solution or a terminal impasse.
As did the Gestalters, the present theory claims that insight during prob-
lem solving is rooted in perception, even though the relation is conceptualized
differently. According to the present theory, a change in problem perception
affects problem solving indirectly, by affecting the problem representation.
This starts a cascade of changes: New working memory elements alter the dis-
tribution of activation over long-term memory, which leads to the retrieval of
unheeded knowledge elements, which in turn opens up new options, so the
heuristic search mechanism faces different choices. The contemporary under-
standing of the cognitive system provides precise causal links that replace the
inexplicable goodness gradient of the Gestalt theory.

Experimental Grounding

The redistribution theory provides an in-principle explanation of the insight
sequence. If it is accurate, it should help us understand results, regularities
and patterns in the data from particular insight experiments. Articulating an
abstract theory vis-à-vis a specific instance often requires auxiliary principles
that mediate between the general mechanism and the specific situation in
which its operation is observed, similar to the way in which the effects of air
resistance mediate between the Newtonian principle of constant gravitational
acceleration and the observable fact that feathers fall slower than marbles. Two
such auxiliary principles have emerged in the work of myself and my associ-
ates and collaborators. Both principles address the most striking property of
the classical insight problems: the gap between their objective simplicity and
their subjective difficulty.
The first principle addresses the question of why two problems with very
similar solutions can vary drastically in their difficulty. The Principle of Scope
claims that the larger the scope of the representational change that is required
to solve a problem, the lower the probability that the change will be achieved
in a given time period. The term “scope” refers to the proportion of the mental
representation that needs to be revised to bring the productive options to mind.
The larger the required change, the lower the probability that it will occur.