Page 153 - Deep Learning

136 Creativity

possibility space with thousands of nodes, this process is too laborious, costly
and time-consuming – not to forget dangerous – for rapid progress. The prob-
lem of flight remained unsolved for two centuries after it was first formulated
in terms of its three main components.
According to Bradshaw’s analysis, the Wright Brothers proceeded dif-
ferently. They separated the overall design into the three subproblems of lift,
propulsion and steering, and worked on each problem separately. They did
not attempt to build an airplane until they had developed solutions to each
subproblem, and they adopted their methods to the demands of each subgoal.
They investigated lift by building a wind tunnel in which they could measure
the lift provided by different wing shapes, and also by building and testing
large kites. They investigated propulsion by, among other methods, making
a large number of model propellers and measuring the drag achieved with
each one. In this way, they could search the space of possible propeller shapes
faster than if they had tried to build full-scale airplanes and fly them. When
they went to Kitty Hawk in 1903, they understood their three subproblems and
had some confidence in their solutions. Although much work was needed to
make the design practical, there was little doubt that their machine would fly.
Subgoaling, the partitioning of the overall problem into nearly independent
parts, won the race for mastery of the air.
The list of examples could be extended, but these instances already demon-
strate that the theory of analytical thinking that underpins the theory of insight is
applicable to complex creative projects. It is not surprising that search reappears
at higher levels of complexity. The structure of heuristic search is rooted in the
basic facts of uncertainty and fallibility; in an unfamiliar possibility space, steps at
any level of complexity are necessarily tentative, so the generate-evaluate-backup-
vary pattern is bound to be level-invariant along the dimension of complexity.
Hierarchical organization is a technique for managing complexity, so it is even
less surprising that subgoaling plays a role in complex projects.
Instances of a theoretical principle provide an existence proof but say
nothing about frequency of occurrence. Data on prevalence – what propor-
tion of significant creative projects exhibit search and subgoaling – are not
available. For present purposes, the existence proof is sufficient.

Complexity: Impasse and Insight

If inventors search and subgoal, do they proceed at a steady pace or do they
experience successive alterations in mode and tempo? Accounts by historians
of technology and autobiographical accounts by inventors suggest the latter.