Page 133 - Deep Learning

116 Creativity

solve a certain problem, we would have to know the exact biases that will shape
his initial representation of the problem. We would have to know the exact
content and structure of his memory network, including the strength of each
link along which activation might spread. To predict the point in the heuristic
search at which he will encounter an impasse, we would need to know which
heuristics he is likely to apply. It is impossible to know the content of another
person’s mind at this level of detail, so the occurrence or nonoccurrence of
individual insight events is in practice unpredictable. Mind is as massively
contingent as other complex systems.

ANSWERS TO THE CREATIVITY QUESTIONS

Alterations in mode and tempo of creative problem solving are caused by com-
plex interactions among multiple cognitive processes. The initial perception
of a problem serves as a retrieval probe and the retrieved knowledge elements
shape the initial solution space. For analytical thinking to proceed without
being blown up in a combinatorial explosion, knowledge has to be brought to
bear on the current problem to constrain the set of possibilities considered.
Certain options appear more likely to lead to the solution than others and
hence are tried first. If prior experience – the biases embedded in the relative
strengths and activation levels of perceptual processing units and the links
that connect them – does not extrapolate to the problem at hand, the initial
solution space might not contain the desired solution and an impasse results.
The only cure for this is to draw back to leap: The constraints imposed by
prior knowledge must be relaxed, so as to open up a wider space of possi-
bilities. Continued attention to the problem generates feedback that even-
tually shifts the balance among one or more links. If a previously dormant
link rises above threshold, the change might propagate upward through the
processing layers and alter the contents of working memory. This might lead
to the retrieval of previously unheeded but potentially useful knowledge ele-
ments, including actions and inference rules. Heuristic search resumes in the
revised search space. Once again, the trick to progress is to constrain the set of
options in the new search space so as not to get lost in the maze of possibilities.
But every constraint threatens to exclude the desired solution from the set of
paths considered, and every time an impasse is encountered, some constraint
or constraints have to be relaxed. Creative problem solving oscillates between
imposing and relaxing constraints.
How does this theory answer the four creativity questions posed in
Chapter 3? First, how are novel ideas possible? Human cognition operates with