Page 159 - Deep Learning

142 Creativity

1945, to construct the electron microscope. Andrew Wiles spent eight years,
25
from 1986 to 1994, inventing the first valid proof of Fermat’s Last Theorem. 26
These examples support the statement by Gruber, “It seems safe to say that
all examples of creative insights occur within protracted creative processes.”
27
Specifically, the examples suggest that significant creative projects stretch over
approximately two orders of magnitude on the time scale, from 0.10 to 10 years.
It is plausible that they cluster toward the middle of that range. Estimation is
complicated by the fact that a person sometimes puts a project aside, perhaps
for years, before completing it. In such cases, the sum of the periods of active
engagement might fall within the .10 – 10 range. The range is a rough approx-
imation, but it will serve.
If significant projects only require a dozen or so insights, and if an insight
can occur in a few seconds or minutes, why do creative projects take .10 to 10
years? The principles of Accumulation (see Chapter 3) and Raisins in the Dough
(see Chapter 4) provide the answer. A complex project generates multiple sub-
problems, each of which might require a creative response. From the point of
view of the overall project, the solution to one subproblem is only a partial
insight, a step forward but not the entire solution. In addition, there is the ana-
lytical or noncreative work. Inventing a device, formulating a scientific theory
or completing a work of art involves many processes and activities other than
generating novel ideas. A product has to be made, not merely thought of, and
making is a complicated affair. Creative products require the accumulation of
work of both the creative and the noncreative sort, and accumulation takes
time.
If we observe the rate of progress across months or even years, the slow-
downs and speed-ups associated with single insights might disappear in the
aggregating process. Adjectives like slow and sudden are relative to a time
scale. A squall might seem like a sudden change in the weather during a
walk in the park, but that same squall will not appear as a distinct event in
a chart of rainfall per year; the daily ups and downs of the stock market dis-
appear in a graph of the increase in economic output over the last century;
the death of a single mosquito makes a big difference to the mosquito but
little difference to a hiker; and so on. But if events of small magnitude or
scope disappear from view as we zoom out to a longer time scale, then the
properties associated with those events are not necessarily replicated at the
higher level.
However, there are two scaling mechanisms that make alterations in
the rate of progress almost unavoidable at the significant time band. The
first mechanism is that although a single insight usually does not suffice to